xref: /openbmc/qemu/tcg/arm/tcg-target.c.inc (revision 6604e475)
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
95typedef enum {
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} ARMCond;
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_LDMIA     = 0x08b00000,
145    INSN_STMDB     = 0x09200000,
146
147    INSN_LDR_IMM   = 0x04100000,
148    INSN_LDR_REG   = 0x06100000,
149    INSN_STR_IMM   = 0x04000000,
150    INSN_STR_REG   = 0x06000000,
151
152    INSN_LDRH_IMM  = 0x005000b0,
153    INSN_LDRH_REG  = 0x001000b0,
154    INSN_LDRSH_IMM = 0x005000f0,
155    INSN_LDRSH_REG = 0x001000f0,
156    INSN_STRH_IMM  = 0x004000b0,
157    INSN_STRH_REG  = 0x000000b0,
158
159    INSN_LDRB_IMM  = 0x04500000,
160    INSN_LDRB_REG  = 0x06500000,
161    INSN_LDRSB_IMM = 0x005000d0,
162    INSN_LDRSB_REG = 0x001000d0,
163    INSN_STRB_IMM  = 0x04400000,
164    INSN_STRB_REG  = 0x06400000,
165
166    INSN_LDRD_IMM  = 0x004000d0,
167    INSN_LDRD_REG  = 0x000000d0,
168    INSN_STRD_IMM  = 0x004000f0,
169    INSN_STRD_REG  = 0x000000f0,
170
171    INSN_DMB_ISH   = 0xf57ff05b,
172    INSN_DMB_MCR   = 0xee070fba,
173
174    /* Architected nop introduced in v6k.  */
175    /* ??? This is an MSR (imm) 0,0,0 insn.  Anyone know if this
176       also Just So Happened to do nothing on pre-v6k so that we
177       don't need to conditionalize it?  */
178    INSN_NOP_v6k   = 0xe320f000,
179    /* Otherwise the assembler uses mov r0,r0 */
180    INSN_NOP_v4    = (COND_AL << 28) | ARITH_MOV,
181
182    INSN_VADD      = 0xf2000800,
183    INSN_VAND      = 0xf2000110,
184    INSN_VBIC      = 0xf2100110,
185    INSN_VEOR      = 0xf3000110,
186    INSN_VORN      = 0xf2300110,
187    INSN_VORR      = 0xf2200110,
188    INSN_VSUB      = 0xf3000800,
189    INSN_VMUL      = 0xf2000910,
190    INSN_VQADD     = 0xf2000010,
191    INSN_VQADD_U   = 0xf3000010,
192    INSN_VQSUB     = 0xf2000210,
193    INSN_VQSUB_U   = 0xf3000210,
194    INSN_VMAX      = 0xf2000600,
195    INSN_VMAX_U    = 0xf3000600,
196    INSN_VMIN      = 0xf2000610,
197    INSN_VMIN_U    = 0xf3000610,
198
199    INSN_VABS      = 0xf3b10300,
200    INSN_VMVN      = 0xf3b00580,
201    INSN_VNEG      = 0xf3b10380,
202
203    INSN_VCEQ0     = 0xf3b10100,
204    INSN_VCGT0     = 0xf3b10000,
205    INSN_VCGE0     = 0xf3b10080,
206    INSN_VCLE0     = 0xf3b10180,
207    INSN_VCLT0     = 0xf3b10200,
208
209    INSN_VCEQ      = 0xf3000810,
210    INSN_VCGE      = 0xf2000310,
211    INSN_VCGT      = 0xf2000300,
212    INSN_VCGE_U    = 0xf3000310,
213    INSN_VCGT_U    = 0xf3000300,
214
215    INSN_VSHLI     = 0xf2800510,  /* VSHL (immediate) */
216    INSN_VSARI     = 0xf2800010,  /* VSHR.S */
217    INSN_VSHRI     = 0xf3800010,  /* VSHR.U */
218    INSN_VSLI      = 0xf3800510,
219    INSN_VSHL_S    = 0xf2000400,  /* VSHL.S (register) */
220    INSN_VSHL_U    = 0xf3000400,  /* VSHL.U (register) */
221
222    INSN_VBSL      = 0xf3100110,
223    INSN_VBIT      = 0xf3200110,
224    INSN_VBIF      = 0xf3300110,
225
226    INSN_VTST      = 0xf2000810,
227
228    INSN_VDUP_G    = 0xee800b10,  /* VDUP (ARM core register) */
229    INSN_VDUP_S    = 0xf3b00c00,  /* VDUP (scalar) */
230    INSN_VLDR_D    = 0xed100b00,  /* VLDR.64 */
231    INSN_VLD1      = 0xf4200000,  /* VLD1 (multiple single elements) */
232    INSN_VLD1R     = 0xf4a00c00,  /* VLD1 (single element to all lanes) */
233    INSN_VST1      = 0xf4000000,  /* VST1 (multiple single elements) */
234    INSN_VMOVI     = 0xf2800010,  /* VMOV (immediate) */
235} ARMInsn;
236
237#define INSN_NOP   (use_armv7_instructions ? INSN_NOP_v6k : INSN_NOP_v4)
238
239static const uint8_t tcg_cond_to_arm_cond[] = {
240    [TCG_COND_EQ] = COND_EQ,
241    [TCG_COND_NE] = COND_NE,
242    [TCG_COND_LT] = COND_LT,
243    [TCG_COND_GE] = COND_GE,
244    [TCG_COND_LE] = COND_LE,
245    [TCG_COND_GT] = COND_GT,
246    /* unsigned */
247    [TCG_COND_LTU] = COND_CC,
248    [TCG_COND_GEU] = COND_CS,
249    [TCG_COND_LEU] = COND_LS,
250    [TCG_COND_GTU] = COND_HI,
251};
252
253static int encode_imm(uint32_t imm);
254
255/* TCG private relocation type: add with pc+imm8 */
256#define R_ARM_PC8  11
257
258/* TCG private relocation type: vldr with imm8 << 2 */
259#define R_ARM_PC11 12
260
261static bool reloc_pc24(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
262{
263    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
264    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) >> 2;
265
266    if (offset == sextract32(offset, 0, 24)) {
267        *src_rw = deposit32(*src_rw, 0, 24, offset);
268        return true;
269    }
270    return false;
271}
272
273static bool reloc_pc13(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
274{
275    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
276    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
277
278    if (offset >= -0xfff && offset <= 0xfff) {
279        tcg_insn_unit insn = *src_rw;
280        bool u = (offset >= 0);
281        if (!u) {
282            offset = -offset;
283        }
284        insn = deposit32(insn, 23, 1, u);
285        insn = deposit32(insn, 0, 12, offset);
286        *src_rw = insn;
287        return true;
288    }
289    return false;
290}
291
292static bool reloc_pc11(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
293{
294    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
295    ptrdiff_t offset = (tcg_ptr_byte_diff(target, src_rx) - 8) / 4;
296
297    if (offset >= -0xff && offset <= 0xff) {
298        tcg_insn_unit insn = *src_rw;
299        bool u = (offset >= 0);
300        if (!u) {
301            offset = -offset;
302        }
303        insn = deposit32(insn, 23, 1, u);
304        insn = deposit32(insn, 0, 8, offset);
305        *src_rw = insn;
306        return true;
307    }
308    return false;
309}
310
311static bool reloc_pc8(tcg_insn_unit *src_rw, const tcg_insn_unit *target)
312{
313    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
314    ptrdiff_t offset = tcg_ptr_byte_diff(target, src_rx) - 8;
315    int imm12 = encode_imm(offset);
316
317    if (imm12 >= 0) {
318        *src_rw = deposit32(*src_rw, 0, 12, imm12);
319        return true;
320    }
321    return false;
322}
323
324static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
325                        intptr_t value, intptr_t addend)
326{
327    tcg_debug_assert(addend == 0);
328    switch (type) {
329    case R_ARM_PC24:
330        return reloc_pc24(code_ptr, (const tcg_insn_unit *)value);
331    case R_ARM_PC13:
332        return reloc_pc13(code_ptr, (const tcg_insn_unit *)value);
333    case R_ARM_PC11:
334        return reloc_pc11(code_ptr, (const tcg_insn_unit *)value);
335    case R_ARM_PC8:
336        return reloc_pc8(code_ptr, (const tcg_insn_unit *)value);
337    default:
338        g_assert_not_reached();
339    }
340}
341
342#define TCG_CT_CONST_ARM  0x100
343#define TCG_CT_CONST_INV  0x200
344#define TCG_CT_CONST_NEG  0x400
345#define TCG_CT_CONST_ZERO 0x800
346#define TCG_CT_CONST_ORRI 0x1000
347#define TCG_CT_CONST_ANDI 0x2000
348
349#define ALL_GENERAL_REGS  0xffffu
350#define ALL_VECTOR_REGS   0xffff0000u
351
352/*
353 * r0-r2 will be overwritten when reading the tlb entry (softmmu only)
354 * and r0-r1 doing the byte swapping, so don't use these.
355 * r3 is removed for softmmu to avoid clashes with helper arguments.
356 */
357#ifdef CONFIG_SOFTMMU
358#define ALL_QLOAD_REGS \
359    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1) | \
360                          (1 << TCG_REG_R2) | (1 << TCG_REG_R3) | \
361                          (1 << TCG_REG_R14)))
362#define ALL_QSTORE_REGS \
363    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1) | \
364                          (1 << TCG_REG_R2) | (1 << TCG_REG_R14) | \
365                          ((TARGET_LONG_BITS == 64) << TCG_REG_R3)))
366#else
367#define ALL_QLOAD_REGS   ALL_GENERAL_REGS
368#define ALL_QSTORE_REGS \
369    (ALL_GENERAL_REGS & ~((1 << TCG_REG_R0) | (1 << TCG_REG_R1)))
370#endif
371
372/*
373 * ARM immediates for ALU instructions are made of an unsigned 8-bit
374 * right-rotated by an even amount between 0 and 30.
375 *
376 * Return < 0 if @imm cannot be encoded, else the entire imm12 field.
377 */
378static int encode_imm(uint32_t imm)
379{
380    uint32_t rot, imm8;
381
382    /* Simple case, no rotation required. */
383    if ((imm & ~0xff) == 0) {
384        return imm;
385    }
386
387    /* Next, try a simple even shift.  */
388    rot = ctz32(imm) & ~1;
389    imm8 = imm >> rot;
390    rot = 32 - rot;
391    if ((imm8 & ~0xff) == 0) {
392        goto found;
393    }
394
395    /*
396     * Finally, try harder with rotations.
397     * The ctz test above will have taken care of rotates >= 8.
398     */
399    for (rot = 2; rot < 8; rot += 2) {
400        imm8 = rol32(imm, rot);
401        if ((imm8 & ~0xff) == 0) {
402            goto found;
403        }
404    }
405    /* Fail: imm cannot be encoded. */
406    return -1;
407
408 found:
409    /* Note that rot is even, and we discard bit 0 by shifting by 7. */
410    return rot << 7 | imm8;
411}
412
413static int encode_imm_nofail(uint32_t imm)
414{
415    int ret = encode_imm(imm);
416    tcg_debug_assert(ret >= 0);
417    return ret;
418}
419
420static bool check_fit_imm(uint32_t imm)
421{
422    return encode_imm(imm) >= 0;
423}
424
425/* Return true if v16 is a valid 16-bit shifted immediate.  */
426static bool is_shimm16(uint16_t v16, int *cmode, int *imm8)
427{
428    if (v16 == (v16 & 0xff)) {
429        *cmode = 0x8;
430        *imm8 = v16 & 0xff;
431        return true;
432    } else if (v16 == (v16 & 0xff00)) {
433        *cmode = 0xa;
434        *imm8 = v16 >> 8;
435        return true;
436    }
437    return false;
438}
439
440/* Return true if v32 is a valid 32-bit shifted immediate.  */
441static bool is_shimm32(uint32_t v32, int *cmode, int *imm8)
442{
443    if (v32 == (v32 & 0xff)) {
444        *cmode = 0x0;
445        *imm8 = v32 & 0xff;
446        return true;
447    } else if (v32 == (v32 & 0xff00)) {
448        *cmode = 0x2;
449        *imm8 = (v32 >> 8) & 0xff;
450        return true;
451    } else if (v32 == (v32 & 0xff0000)) {
452        *cmode = 0x4;
453        *imm8 = (v32 >> 16) & 0xff;
454        return true;
455    } else if (v32 == (v32 & 0xff000000)) {
456        *cmode = 0x6;
457        *imm8 = v32 >> 24;
458        return true;
459    }
460    return false;
461}
462
463/* Return true if v32 is a valid 32-bit shifting ones immediate.  */
464static bool is_soimm32(uint32_t v32, int *cmode, int *imm8)
465{
466    if ((v32 & 0xffff00ff) == 0xff) {
467        *cmode = 0xc;
468        *imm8 = (v32 >> 8) & 0xff;
469        return true;
470    } else if ((v32 & 0xff00ffff) == 0xffff) {
471        *cmode = 0xd;
472        *imm8 = (v32 >> 16) & 0xff;
473        return true;
474    }
475    return false;
476}
477
478/*
479 * Return non-zero if v32 can be formed by MOVI+ORR.
480 * Place the parameters for MOVI in (cmode, imm8).
481 * Return the cmode for ORR; the imm8 can be had via extraction from v32.
482 */
483static int is_shimm32_pair(uint32_t v32, int *cmode, int *imm8)
484{
485    int i;
486
487    for (i = 6; i > 0; i -= 2) {
488        /* Mask out one byte we can add with ORR.  */
489        uint32_t tmp = v32 & ~(0xffu << (i * 4));
490        if (is_shimm32(tmp, cmode, imm8) ||
491            is_soimm32(tmp, cmode, imm8)) {
492            break;
493        }
494    }
495    return i;
496}
497
498/* Return true if V is a valid 16-bit or 32-bit shifted immediate.  */
499static bool is_shimm1632(uint32_t v32, int *cmode, int *imm8)
500{
501    if (v32 == deposit32(v32, 16, 16, v32)) {
502        return is_shimm16(v32, cmode, imm8);
503    } else {
504        return is_shimm32(v32, cmode, imm8);
505    }
506}
507
508/* Test if a constant matches the constraint.
509 * TODO: define constraints for:
510 *
511 * ldr/str offset:   between -0xfff and 0xfff
512 * ldrh/strh offset: between -0xff and 0xff
513 * mov operand2:     values represented with x << (2 * y), x < 0x100
514 * add, sub, eor...: ditto
515 */
516static bool tcg_target_const_match(int64_t val, TCGType type, int ct)
517{
518    if (ct & TCG_CT_CONST) {
519        return 1;
520    } else if ((ct & TCG_CT_CONST_ARM) && check_fit_imm(val)) {
521        return 1;
522    } else if ((ct & TCG_CT_CONST_INV) && check_fit_imm(~val)) {
523        return 1;
524    } else if ((ct & TCG_CT_CONST_NEG) && check_fit_imm(-val)) {
525        return 1;
526    } else if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
527        return 1;
528    }
529
530    switch (ct & (TCG_CT_CONST_ORRI | TCG_CT_CONST_ANDI)) {
531    case 0:
532        break;
533    case TCG_CT_CONST_ANDI:
534        val = ~val;
535        /* fallthru */
536    case TCG_CT_CONST_ORRI:
537        if (val == deposit64(val, 32, 32, val)) {
538            int cmode, imm8;
539            return is_shimm1632(val, &cmode, &imm8);
540        }
541        break;
542    default:
543        /* Both bits should not be set for the same insn.  */
544        g_assert_not_reached();
545    }
546
547    return 0;
548}
549
550static void tcg_out_b_imm(TCGContext *s, ARMCond cond, int32_t offset)
551{
552    tcg_out32(s, (cond << 28) | 0x0a000000 |
553                    (((offset - 8) >> 2) & 0x00ffffff));
554}
555
556static void tcg_out_bl_imm(TCGContext *s, ARMCond cond, int32_t offset)
557{
558    tcg_out32(s, (cond << 28) | 0x0b000000 |
559                    (((offset - 8) >> 2) & 0x00ffffff));
560}
561
562static void tcg_out_blx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
563{
564    tcg_out32(s, (cond << 28) | 0x012fff30 | rn);
565}
566
567static void tcg_out_blx_imm(TCGContext *s, int32_t offset)
568{
569    tcg_out32(s, 0xfa000000 | ((offset & 2) << 23) |
570                (((offset - 8) >> 2) & 0x00ffffff));
571}
572
573static void tcg_out_dat_reg(TCGContext *s, ARMCond cond, ARMInsn opc,
574                            TCGReg rd, TCGReg rn, TCGReg rm, int shift)
575{
576    tcg_out32(s, (cond << 28) | (0 << 25) | opc |
577                    (rn << 16) | (rd << 12) | shift | rm);
578}
579
580static void tcg_out_mov_reg(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rm)
581{
582    /* Simple reg-reg move, optimising out the 'do nothing' case */
583    if (rd != rm) {
584        tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, rm, SHIFT_IMM_LSL(0));
585    }
586}
587
588static void tcg_out_bx_reg(TCGContext *s, ARMCond cond, TCGReg rn)
589{
590    tcg_out32(s, (cond << 28) | 0x012fff10 | rn);
591}
592
593static void tcg_out_b_reg(TCGContext *s, ARMCond cond, TCGReg rn)
594{
595    /*
596     * Unless the C portion of QEMU is compiled as thumb, we don't need
597     * true BX semantics; merely a branch to an address held in a register.
598     */
599    if (use_armv5t_instructions) {
600        tcg_out_bx_reg(s, cond, rn);
601    } else {
602        tcg_out_mov_reg(s, cond, TCG_REG_PC, rn);
603    }
604}
605
606static void tcg_out_dat_imm(TCGContext *s, ARMCond cond, ARMInsn opc,
607                            TCGReg rd, TCGReg rn, int im)
608{
609    tcg_out32(s, (cond << 28) | (1 << 25) | opc |
610                    (rn << 16) | (rd << 12) | im);
611}
612
613static void tcg_out_ldstm(TCGContext *s, ARMCond cond, ARMInsn opc,
614                          TCGReg rn, uint16_t mask)
615{
616    tcg_out32(s, (cond << 28) | opc | (rn << 16) | mask);
617}
618
619/* Note that this routine is used for both LDR and LDRH formats, so we do
620   not wish to include an immediate shift at this point.  */
621static void tcg_out_memop_r(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
622                            TCGReg rn, TCGReg rm, bool u, bool p, bool w)
623{
624    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24)
625              | (w << 21) | (rn << 16) | (rt << 12) | rm);
626}
627
628static void tcg_out_memop_8(TCGContext *s, ARMCond cond, ARMInsn opc, TCGReg rt,
629                            TCGReg rn, int imm8, bool p, bool w)
630{
631    bool u = 1;
632    if (imm8 < 0) {
633        imm8 = -imm8;
634        u = 0;
635    }
636    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
637              (rn << 16) | (rt << 12) | ((imm8 & 0xf0) << 4) | (imm8 & 0xf));
638}
639
640static void tcg_out_memop_12(TCGContext *s, ARMCond cond, ARMInsn opc,
641                             TCGReg rt, TCGReg rn, int imm12, bool p, bool w)
642{
643    bool u = 1;
644    if (imm12 < 0) {
645        imm12 = -imm12;
646        u = 0;
647    }
648    tcg_out32(s, (cond << 28) | opc | (u << 23) | (p << 24) | (w << 21) |
649              (rn << 16) | (rt << 12) | imm12);
650}
651
652static void tcg_out_ld32_12(TCGContext *s, ARMCond cond, TCGReg rt,
653                            TCGReg rn, int imm12)
654{
655    tcg_out_memop_12(s, cond, INSN_LDR_IMM, rt, rn, imm12, 1, 0);
656}
657
658static void tcg_out_st32_12(TCGContext *s, ARMCond cond, TCGReg rt,
659                            TCGReg rn, int imm12)
660{
661    tcg_out_memop_12(s, cond, INSN_STR_IMM, rt, rn, imm12, 1, 0);
662}
663
664static void tcg_out_ld32_r(TCGContext *s, ARMCond cond, TCGReg rt,
665                           TCGReg rn, TCGReg rm)
666{
667    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 0);
668}
669
670static void tcg_out_st32_r(TCGContext *s, ARMCond cond, TCGReg rt,
671                           TCGReg rn, TCGReg rm)
672{
673    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 0);
674}
675
676static void tcg_out_ldrd_8(TCGContext *s, ARMCond cond, TCGReg rt,
677                           TCGReg rn, int imm8)
678{
679    tcg_out_memop_8(s, cond, INSN_LDRD_IMM, rt, rn, imm8, 1, 0);
680}
681
682static void tcg_out_ldrd_r(TCGContext *s, ARMCond cond, TCGReg rt,
683                           TCGReg rn, TCGReg rm)
684{
685    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
686}
687
688static void __attribute__((unused))
689tcg_out_ldrd_rwb(TCGContext *s, ARMCond cond, TCGReg rt, TCGReg rn, TCGReg rm)
690{
691    tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 1);
692}
693
694static void tcg_out_strd_8(TCGContext *s, ARMCond cond, TCGReg rt,
695                           TCGReg rn, int imm8)
696{
697    tcg_out_memop_8(s, cond, INSN_STRD_IMM, rt, rn, imm8, 1, 0);
698}
699
700static void tcg_out_strd_r(TCGContext *s, ARMCond cond, TCGReg rt,
701                           TCGReg rn, TCGReg rm)
702{
703    tcg_out_memop_r(s, cond, INSN_STRD_REG, rt, rn, rm, 1, 1, 0);
704}
705
706/* Register pre-increment with base writeback.  */
707static void tcg_out_ld32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
708                             TCGReg rn, TCGReg rm)
709{
710    tcg_out_memop_r(s, cond, INSN_LDR_REG, rt, rn, rm, 1, 1, 1);
711}
712
713static void tcg_out_st32_rwb(TCGContext *s, ARMCond cond, TCGReg rt,
714                             TCGReg rn, TCGReg rm)
715{
716    tcg_out_memop_r(s, cond, INSN_STR_REG, rt, rn, rm, 1, 1, 1);
717}
718
719static void tcg_out_ld16u_8(TCGContext *s, ARMCond cond, TCGReg rt,
720                            TCGReg rn, int imm8)
721{
722    tcg_out_memop_8(s, cond, INSN_LDRH_IMM, rt, rn, imm8, 1, 0);
723}
724
725static void tcg_out_st16_8(TCGContext *s, ARMCond cond, TCGReg rt,
726                           TCGReg rn, int imm8)
727{
728    tcg_out_memop_8(s, cond, INSN_STRH_IMM, rt, rn, imm8, 1, 0);
729}
730
731static void tcg_out_ld16u_r(TCGContext *s, ARMCond cond, TCGReg rt,
732                            TCGReg rn, TCGReg rm)
733{
734    tcg_out_memop_r(s, cond, INSN_LDRH_REG, rt, rn, rm, 1, 1, 0);
735}
736
737static void tcg_out_st16_r(TCGContext *s, ARMCond cond, TCGReg rt,
738                           TCGReg rn, TCGReg rm)
739{
740    tcg_out_memop_r(s, cond, INSN_STRH_REG, rt, rn, rm, 1, 1, 0);
741}
742
743static void tcg_out_ld16s_8(TCGContext *s, ARMCond cond, TCGReg rt,
744                            TCGReg rn, int imm8)
745{
746    tcg_out_memop_8(s, cond, INSN_LDRSH_IMM, rt, rn, imm8, 1, 0);
747}
748
749static void tcg_out_ld16s_r(TCGContext *s, ARMCond cond, TCGReg rt,
750                            TCGReg rn, TCGReg rm)
751{
752    tcg_out_memop_r(s, cond, INSN_LDRSH_REG, rt, rn, rm, 1, 1, 0);
753}
754
755static void tcg_out_ld8_12(TCGContext *s, ARMCond cond, TCGReg rt,
756                           TCGReg rn, int imm12)
757{
758    tcg_out_memop_12(s, cond, INSN_LDRB_IMM, rt, rn, imm12, 1, 0);
759}
760
761static void tcg_out_st8_12(TCGContext *s, ARMCond cond, TCGReg rt,
762                           TCGReg rn, int imm12)
763{
764    tcg_out_memop_12(s, cond, INSN_STRB_IMM, rt, rn, imm12, 1, 0);
765}
766
767static void tcg_out_ld8_r(TCGContext *s, ARMCond cond, TCGReg rt,
768                          TCGReg rn, TCGReg rm)
769{
770    tcg_out_memop_r(s, cond, INSN_LDRB_REG, rt, rn, rm, 1, 1, 0);
771}
772
773static void tcg_out_st8_r(TCGContext *s, ARMCond cond, TCGReg rt,
774                          TCGReg rn, TCGReg rm)
775{
776    tcg_out_memop_r(s, cond, INSN_STRB_REG, rt, rn, rm, 1, 1, 0);
777}
778
779static void tcg_out_ld8s_8(TCGContext *s, ARMCond cond, TCGReg rt,
780                           TCGReg rn, int imm8)
781{
782    tcg_out_memop_8(s, cond, INSN_LDRSB_IMM, rt, rn, imm8, 1, 0);
783}
784
785static void tcg_out_ld8s_r(TCGContext *s, ARMCond cond, TCGReg rt,
786                           TCGReg rn, TCGReg rm)
787{
788    tcg_out_memop_r(s, cond, INSN_LDRSB_REG, rt, rn, rm, 1, 1, 0);
789}
790
791static void tcg_out_movi_pool(TCGContext *s, ARMCond cond,
792                              TCGReg rd, uint32_t arg)
793{
794    new_pool_label(s, arg, R_ARM_PC13, s->code_ptr, 0);
795    tcg_out_ld32_12(s, cond, rd, TCG_REG_PC, 0);
796}
797
798static void tcg_out_movi32(TCGContext *s, ARMCond cond,
799                           TCGReg rd, uint32_t arg)
800{
801    int imm12, diff, opc, sh1, sh2;
802    uint32_t tt0, tt1, tt2;
803
804    /* Check a single MOV/MVN before anything else.  */
805    imm12 = encode_imm(arg);
806    if (imm12 >= 0) {
807        tcg_out_dat_imm(s, cond, ARITH_MOV, rd, 0, imm12);
808        return;
809    }
810    imm12 = encode_imm(~arg);
811    if (imm12 >= 0) {
812        tcg_out_dat_imm(s, cond, ARITH_MVN, rd, 0, imm12);
813        return;
814    }
815
816    /* Check for a pc-relative address.  This will usually be the TB,
817       or within the TB, which is immediately before the code block.  */
818    diff = tcg_pcrel_diff(s, (void *)arg) - 8;
819    if (diff >= 0) {
820        imm12 = encode_imm(diff);
821        if (imm12 >= 0) {
822            tcg_out_dat_imm(s, cond, ARITH_ADD, rd, TCG_REG_PC, imm12);
823            return;
824        }
825    } else {
826        imm12 = encode_imm(-diff);
827        if (imm12 >= 0) {
828            tcg_out_dat_imm(s, cond, ARITH_SUB, rd, TCG_REG_PC, imm12);
829            return;
830        }
831    }
832
833    /* Use movw + movt.  */
834    if (use_armv7_instructions) {
835        /* movw */
836        tcg_out32(s, (cond << 28) | 0x03000000 | (rd << 12)
837                  | ((arg << 4) & 0x000f0000) | (arg & 0xfff));
838        if (arg & 0xffff0000) {
839            /* movt */
840            tcg_out32(s, (cond << 28) | 0x03400000 | (rd << 12)
841                      | ((arg >> 12) & 0x000f0000) | ((arg >> 16) & 0xfff));
842        }
843        return;
844    }
845
846    /* Look for sequences of two insns.  If we have lots of 1's, we can
847       shorten the sequence by beginning with mvn and then clearing
848       higher bits with eor.  */
849    tt0 = arg;
850    opc = ARITH_MOV;
851    if (ctpop32(arg) > 16) {
852        tt0 = ~arg;
853        opc = ARITH_MVN;
854    }
855    sh1 = ctz32(tt0) & ~1;
856    tt1 = tt0 & ~(0xff << sh1);
857    sh2 = ctz32(tt1) & ~1;
858    tt2 = tt1 & ~(0xff << sh2);
859    if (tt2 == 0) {
860        int rot;
861
862        rot = ((32 - sh1) << 7) & 0xf00;
863        tcg_out_dat_imm(s, cond, opc, rd,  0, ((tt0 >> sh1) & 0xff) | rot);
864        rot = ((32 - sh2) << 7) & 0xf00;
865        tcg_out_dat_imm(s, cond, ARITH_EOR, rd, rd,
866                        ((tt0 >> sh2) & 0xff) | rot);
867        return;
868    }
869
870    /* Otherwise, drop it into the constant pool.  */
871    tcg_out_movi_pool(s, cond, rd, arg);
872}
873
874/*
875 * Emit either the reg,imm or reg,reg form of a data-processing insn.
876 * rhs must satisfy the "rI" constraint.
877 */
878static void tcg_out_dat_rI(TCGContext *s, ARMCond cond, ARMInsn opc,
879                           TCGReg dst, TCGReg lhs, TCGArg rhs, int rhs_is_const)
880{
881    if (rhs_is_const) {
882        tcg_out_dat_imm(s, cond, opc, dst, lhs, encode_imm_nofail(rhs));
883    } else {
884        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
885    }
886}
887
888/*
889 * Emit either the reg,imm or reg,reg form of a data-processing insn.
890 * rhs must satisfy the "rIK" constraint.
891 */
892static void tcg_out_dat_rIK(TCGContext *s, ARMCond cond, ARMInsn opc,
893                            ARMInsn opinv, TCGReg dst, TCGReg lhs, TCGArg rhs,
894                            bool rhs_is_const)
895{
896    if (rhs_is_const) {
897        int imm12 = encode_imm(rhs);
898        if (imm12 < 0) {
899            imm12 = encode_imm_nofail(~rhs);
900            opc = opinv;
901        }
902        tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
903    } else {
904        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
905    }
906}
907
908static void tcg_out_dat_rIN(TCGContext *s, ARMCond cond, ARMInsn opc,
909                            ARMInsn opneg, TCGReg dst, TCGReg lhs, TCGArg rhs,
910                            bool rhs_is_const)
911{
912    /* Emit either the reg,imm or reg,reg form of a data-processing insn.
913     * rhs must satisfy the "rIN" constraint.
914     */
915    if (rhs_is_const) {
916        int imm12 = encode_imm(rhs);
917        if (imm12 < 0) {
918            imm12 = encode_imm_nofail(-rhs);
919            opc = opneg;
920        }
921        tcg_out_dat_imm(s, cond, opc, dst, lhs, imm12);
922    } else {
923        tcg_out_dat_reg(s, cond, opc, dst, lhs, rhs, SHIFT_IMM_LSL(0));
924    }
925}
926
927static void tcg_out_mul32(TCGContext *s, ARMCond cond, TCGReg rd,
928                          TCGReg rn, TCGReg rm)
929{
930    /* if ArchVersion() < 6 && d == n then UNPREDICTABLE;  */
931    if (!use_armv6_instructions && rd == rn) {
932        if (rd == rm) {
933            /* rd == rn == rm; copy an input to tmp first.  */
934            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
935            rm = rn = TCG_REG_TMP;
936        } else {
937            rn = rm;
938            rm = rd;
939        }
940    }
941    /* mul */
942    tcg_out32(s, (cond << 28) | 0x90 | (rd << 16) | (rm << 8) | rn);
943}
944
945static void tcg_out_umull32(TCGContext *s, ARMCond cond, TCGReg rd0,
946                            TCGReg rd1, TCGReg rn, TCGReg rm)
947{
948    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
949    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
950        if (rd0 == rm || rd1 == rm) {
951            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
952            rn = TCG_REG_TMP;
953        } else {
954            TCGReg t = rn;
955            rn = rm;
956            rm = t;
957        }
958    }
959    /* umull */
960    tcg_out32(s, (cond << 28) | 0x00800090 |
961              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
962}
963
964static void tcg_out_smull32(TCGContext *s, ARMCond cond, TCGReg rd0,
965                            TCGReg rd1, TCGReg rn, TCGReg rm)
966{
967    /* if ArchVersion() < 6 && (dHi == n || dLo == n) then UNPREDICTABLE;  */
968    if (!use_armv6_instructions && (rd0 == rn || rd1 == rn)) {
969        if (rd0 == rm || rd1 == rm) {
970            tcg_out_mov_reg(s, cond, TCG_REG_TMP, rn);
971            rn = TCG_REG_TMP;
972        } else {
973            TCGReg t = rn;
974            rn = rm;
975            rm = t;
976        }
977    }
978    /* smull */
979    tcg_out32(s, (cond << 28) | 0x00c00090 |
980              (rd1 << 16) | (rd0 << 12) | (rm << 8) | rn);
981}
982
983static void tcg_out_sdiv(TCGContext *s, ARMCond cond,
984                         TCGReg rd, TCGReg rn, TCGReg rm)
985{
986    tcg_out32(s, 0x0710f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
987}
988
989static void tcg_out_udiv(TCGContext *s, ARMCond cond,
990                         TCGReg rd, TCGReg rn, TCGReg rm)
991{
992    tcg_out32(s, 0x0730f010 | (cond << 28) | (rd << 16) | rn | (rm << 8));
993}
994
995static void tcg_out_ext8s(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rn)
996{
997    if (use_armv6_instructions) {
998        /* sxtb */
999        tcg_out32(s, 0x06af0070 | (cond << 28) | (rd << 12) | rn);
1000    } else {
1001        tcg_out_dat_reg(s, cond, ARITH_MOV,
1002                        rd, 0, rn, SHIFT_IMM_LSL(24));
1003        tcg_out_dat_reg(s, cond, ARITH_MOV,
1004                        rd, 0, rd, SHIFT_IMM_ASR(24));
1005    }
1006}
1007
1008static void __attribute__((unused))
1009tcg_out_ext8u(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rn)
1010{
1011    tcg_out_dat_imm(s, cond, ARITH_AND, rd, rn, 0xff);
1012}
1013
1014static void tcg_out_ext16s(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rn)
1015{
1016    if (use_armv6_instructions) {
1017        /* sxth */
1018        tcg_out32(s, 0x06bf0070 | (cond << 28) | (rd << 12) | rn);
1019    } else {
1020        tcg_out_dat_reg(s, cond, ARITH_MOV,
1021                        rd, 0, rn, SHIFT_IMM_LSL(16));
1022        tcg_out_dat_reg(s, cond, ARITH_MOV,
1023                        rd, 0, rd, SHIFT_IMM_ASR(16));
1024    }
1025}
1026
1027static void tcg_out_ext16u(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rn)
1028{
1029    if (use_armv6_instructions) {
1030        /* uxth */
1031        tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rn);
1032    } else {
1033        tcg_out_dat_reg(s, cond, ARITH_MOV,
1034                        rd, 0, rn, SHIFT_IMM_LSL(16));
1035        tcg_out_dat_reg(s, cond, ARITH_MOV,
1036                        rd, 0, rd, SHIFT_IMM_LSR(16));
1037    }
1038}
1039
1040static void tcg_out_bswap16(TCGContext *s, ARMCond cond,
1041                            TCGReg rd, TCGReg rn, int flags)
1042{
1043    if (use_armv6_instructions) {
1044        if (flags & TCG_BSWAP_OS) {
1045            /* revsh */
1046            tcg_out32(s, 0x06ff0fb0 | (cond << 28) | (rd << 12) | rn);
1047            return;
1048        }
1049
1050        /* rev16 */
1051        tcg_out32(s, 0x06bf0fb0 | (cond << 28) | (rd << 12) | rn);
1052        if ((flags & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) {
1053            /* uxth */
1054            tcg_out32(s, 0x06ff0070 | (cond << 28) | (rd << 12) | rd);
1055        }
1056        return;
1057    }
1058
1059    if (flags == 0) {
1060        /*
1061         * For stores, no input or output extension:
1062         *                              rn  = xxAB
1063         * lsr tmp, rn, #8              tmp = 0xxA
1064         * and tmp, tmp, #0xff          tmp = 000A
1065         * orr rd, tmp, rn, lsl #8      rd  = xABA
1066         */
1067        tcg_out_dat_reg(s, cond, ARITH_MOV,
1068                        TCG_REG_TMP, 0, rn, SHIFT_IMM_LSR(8));
1069        tcg_out_dat_imm(s, cond, ARITH_AND, TCG_REG_TMP, TCG_REG_TMP, 0xff);
1070        tcg_out_dat_reg(s, cond, ARITH_ORR,
1071                        rd, TCG_REG_TMP, rn, SHIFT_IMM_LSL(8));
1072        return;
1073    }
1074
1075    /*
1076     * Byte swap, leaving the result at the top of the register.
1077     * We will then shift down, zero or sign-extending.
1078     */
1079    if (flags & TCG_BSWAP_IZ) {
1080        /*
1081         *                              rn  = 00AB
1082         * ror tmp, rn, #8              tmp = B00A
1083         * orr tmp, tmp, tmp, lsl #16   tmp = BA00
1084         */
1085        tcg_out_dat_reg(s, cond, ARITH_MOV,
1086                        TCG_REG_TMP, 0, rn, SHIFT_IMM_ROR(8));
1087        tcg_out_dat_reg(s, cond, ARITH_ORR,
1088                        TCG_REG_TMP, TCG_REG_TMP, TCG_REG_TMP,
1089                        SHIFT_IMM_LSL(16));
1090    } else {
1091        /*
1092         *                              rn  = xxAB
1093         * and tmp, rn, #0xff00         tmp = 00A0
1094         * lsl tmp, tmp, #8             tmp = 0A00
1095         * orr tmp, tmp, rn, lsl #24    tmp = BA00
1096         */
1097        tcg_out_dat_rI(s, cond, ARITH_AND, TCG_REG_TMP, rn, 0xff00, 1);
1098        tcg_out_dat_reg(s, cond, ARITH_MOV,
1099                        TCG_REG_TMP, 0, TCG_REG_TMP, SHIFT_IMM_LSL(8));
1100        tcg_out_dat_reg(s, cond, ARITH_ORR,
1101                        TCG_REG_TMP, TCG_REG_TMP, rn, SHIFT_IMM_LSL(24));
1102    }
1103    tcg_out_dat_reg(s, cond, ARITH_MOV, rd, 0, TCG_REG_TMP,
1104                    (flags & TCG_BSWAP_OS
1105                     ? SHIFT_IMM_ASR(8) : SHIFT_IMM_LSR(8)));
1106}
1107
1108static void tcg_out_bswap32(TCGContext *s, ARMCond cond, TCGReg rd, TCGReg rn)
1109{
1110    if (use_armv6_instructions) {
1111        /* rev */
1112        tcg_out32(s, 0x06bf0f30 | (cond << 28) | (rd << 12) | rn);
1113    } else {
1114        tcg_out_dat_reg(s, cond, ARITH_EOR,
1115                        TCG_REG_TMP, rn, rn, SHIFT_IMM_ROR(16));
1116        tcg_out_dat_imm(s, cond, ARITH_BIC,
1117                        TCG_REG_TMP, TCG_REG_TMP, 0xff | 0x800);
1118        tcg_out_dat_reg(s, cond, ARITH_MOV,
1119                        rd, 0, rn, SHIFT_IMM_ROR(8));
1120        tcg_out_dat_reg(s, cond, ARITH_EOR,
1121                        rd, rd, TCG_REG_TMP, SHIFT_IMM_LSR(8));
1122    }
1123}
1124
1125static void tcg_out_deposit(TCGContext *s, ARMCond cond, TCGReg rd,
1126                            TCGArg a1, int ofs, int len, bool const_a1)
1127{
1128    if (const_a1) {
1129        /* bfi becomes bfc with rn == 15.  */
1130        a1 = 15;
1131    }
1132    /* bfi/bfc */
1133    tcg_out32(s, 0x07c00010 | (cond << 28) | (rd << 12) | a1
1134              | (ofs << 7) | ((ofs + len - 1) << 16));
1135}
1136
1137static void tcg_out_extract(TCGContext *s, ARMCond cond, TCGReg rd,
1138                            TCGReg rn, int ofs, int len)
1139{
1140    /* ubfx */
1141    tcg_out32(s, 0x07e00050 | (cond << 28) | (rd << 12) | rn
1142              | (ofs << 7) | ((len - 1) << 16));
1143}
1144
1145static void tcg_out_sextract(TCGContext *s, ARMCond cond, TCGReg rd,
1146                             TCGReg rn, int ofs, int len)
1147{
1148    /* sbfx */
1149    tcg_out32(s, 0x07a00050 | (cond << 28) | (rd << 12) | rn
1150              | (ofs << 7) | ((len - 1) << 16));
1151}
1152
1153static void tcg_out_ld32u(TCGContext *s, ARMCond cond,
1154                          TCGReg rd, TCGReg rn, int32_t offset)
1155{
1156    if (offset > 0xfff || offset < -0xfff) {
1157        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1158        tcg_out_ld32_r(s, cond, rd, rn, TCG_REG_TMP);
1159    } else
1160        tcg_out_ld32_12(s, cond, rd, rn, offset);
1161}
1162
1163static void tcg_out_st32(TCGContext *s, ARMCond cond,
1164                         TCGReg rd, TCGReg rn, int32_t offset)
1165{
1166    if (offset > 0xfff || offset < -0xfff) {
1167        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1168        tcg_out_st32_r(s, cond, rd, rn, TCG_REG_TMP);
1169    } else
1170        tcg_out_st32_12(s, cond, rd, rn, offset);
1171}
1172
1173static void tcg_out_ld16u(TCGContext *s, ARMCond cond,
1174                          TCGReg rd, TCGReg rn, int32_t offset)
1175{
1176    if (offset > 0xff || offset < -0xff) {
1177        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1178        tcg_out_ld16u_r(s, cond, rd, rn, TCG_REG_TMP);
1179    } else
1180        tcg_out_ld16u_8(s, cond, rd, rn, offset);
1181}
1182
1183static void tcg_out_ld16s(TCGContext *s, ARMCond cond,
1184                          TCGReg rd, TCGReg rn, int32_t offset)
1185{
1186    if (offset > 0xff || offset < -0xff) {
1187        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1188        tcg_out_ld16s_r(s, cond, rd, rn, TCG_REG_TMP);
1189    } else
1190        tcg_out_ld16s_8(s, cond, rd, rn, offset);
1191}
1192
1193static void tcg_out_st16(TCGContext *s, ARMCond cond,
1194                         TCGReg rd, TCGReg rn, int32_t offset)
1195{
1196    if (offset > 0xff || offset < -0xff) {
1197        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1198        tcg_out_st16_r(s, cond, rd, rn, TCG_REG_TMP);
1199    } else
1200        tcg_out_st16_8(s, cond, rd, rn, offset);
1201}
1202
1203static void tcg_out_ld8u(TCGContext *s, ARMCond cond,
1204                         TCGReg rd, TCGReg rn, int32_t offset)
1205{
1206    if (offset > 0xfff || offset < -0xfff) {
1207        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1208        tcg_out_ld8_r(s, cond, rd, rn, TCG_REG_TMP);
1209    } else
1210        tcg_out_ld8_12(s, cond, rd, rn, offset);
1211}
1212
1213static void tcg_out_ld8s(TCGContext *s, ARMCond cond,
1214                         TCGReg rd, TCGReg rn, int32_t offset)
1215{
1216    if (offset > 0xff || offset < -0xff) {
1217        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1218        tcg_out_ld8s_r(s, cond, rd, rn, TCG_REG_TMP);
1219    } else
1220        tcg_out_ld8s_8(s, cond, rd, rn, offset);
1221}
1222
1223static void tcg_out_st8(TCGContext *s, ARMCond cond,
1224                        TCGReg rd, TCGReg rn, int32_t offset)
1225{
1226    if (offset > 0xfff || offset < -0xfff) {
1227        tcg_out_movi32(s, cond, TCG_REG_TMP, offset);
1228        tcg_out_st8_r(s, cond, rd, rn, TCG_REG_TMP);
1229    } else
1230        tcg_out_st8_12(s, cond, rd, rn, offset);
1231}
1232
1233/*
1234 * The _goto case is normally between TBs within the same code buffer, and
1235 * with the code buffer limited to 16MB we wouldn't need the long case.
1236 * But we also use it for the tail-call to the qemu_ld/st helpers, which does.
1237 */
1238static void tcg_out_goto(TCGContext *s, ARMCond cond, const tcg_insn_unit *addr)
1239{
1240    intptr_t addri = (intptr_t)addr;
1241    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1242    bool arm_mode = !(addri & 1);
1243
1244    if (arm_mode && disp - 8 < 0x01fffffd && disp - 8 > -0x01fffffd) {
1245        tcg_out_b_imm(s, cond, disp);
1246        return;
1247    }
1248
1249    /* LDR is interworking from v5t. */
1250    if (arm_mode || use_armv5t_instructions) {
1251        tcg_out_movi_pool(s, cond, TCG_REG_PC, addri);
1252        return;
1253    }
1254
1255    /* else v4t */
1256    tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1257    tcg_out_bx_reg(s, COND_AL, TCG_REG_TMP);
1258}
1259
1260/*
1261 * The call case is mostly used for helpers - so it's not unreasonable
1262 * for them to be beyond branch range.
1263 */
1264static void tcg_out_call(TCGContext *s, const tcg_insn_unit *addr)
1265{
1266    intptr_t addri = (intptr_t)addr;
1267    ptrdiff_t disp = tcg_pcrel_diff(s, addr);
1268    bool arm_mode = !(addri & 1);
1269
1270    if (disp - 8 < 0x02000000 && disp - 8 >= -0x02000000) {
1271        if (arm_mode) {
1272            tcg_out_bl_imm(s, COND_AL, disp);
1273            return;
1274        }
1275        if (use_armv5t_instructions) {
1276            tcg_out_blx_imm(s, disp);
1277            return;
1278        }
1279    }
1280
1281    if (use_armv5t_instructions) {
1282        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1283        tcg_out_blx_reg(s, COND_AL, TCG_REG_TMP);
1284    } else if (arm_mode) {
1285        /* ??? Know that movi_pool emits exactly 1 insn.  */
1286        tcg_out_mov_reg(s, COND_AL, TCG_REG_R14, TCG_REG_PC);
1287        tcg_out_movi_pool(s, COND_AL, TCG_REG_PC, addri);
1288    } else {
1289        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, addri);
1290        tcg_out_mov_reg(s, COND_AL, TCG_REG_R14, TCG_REG_PC);
1291        tcg_out_bx_reg(s, COND_AL, TCG_REG_TMP);
1292    }
1293}
1294
1295static void tcg_out_goto_label(TCGContext *s, ARMCond cond, TCGLabel *l)
1296{
1297    if (l->has_value) {
1298        tcg_out_goto(s, cond, l->u.value_ptr);
1299    } else {
1300        tcg_out_reloc(s, s->code_ptr, R_ARM_PC24, l, 0);
1301        tcg_out_b_imm(s, cond, 0);
1302    }
1303}
1304
1305static void tcg_out_mb(TCGContext *s, TCGArg a0)
1306{
1307    if (use_armv7_instructions) {
1308        tcg_out32(s, INSN_DMB_ISH);
1309    } else if (use_armv6_instructions) {
1310        tcg_out32(s, INSN_DMB_MCR);
1311    }
1312}
1313
1314static TCGCond tcg_out_cmp2(TCGContext *s, const TCGArg *args,
1315                            const int *const_args)
1316{
1317    TCGReg al = args[0];
1318    TCGReg ah = args[1];
1319    TCGArg bl = args[2];
1320    TCGArg bh = args[3];
1321    TCGCond cond = args[4];
1322    int const_bl = const_args[2];
1323    int const_bh = const_args[3];
1324
1325    switch (cond) {
1326    case TCG_COND_EQ:
1327    case TCG_COND_NE:
1328    case TCG_COND_LTU:
1329    case TCG_COND_LEU:
1330    case TCG_COND_GTU:
1331    case TCG_COND_GEU:
1332        /* We perform a conditional comparision.  If the high half is
1333           equal, then overwrite the flags with the comparison of the
1334           low half.  The resulting flags cover the whole.  */
1335        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, ah, bh, const_bh);
1336        tcg_out_dat_rI(s, COND_EQ, ARITH_CMP, 0, al, bl, const_bl);
1337        return cond;
1338
1339    case TCG_COND_LT:
1340    case TCG_COND_GE:
1341        /* We perform a double-word subtraction and examine the result.
1342           We do not actually need the result of the subtract, so the
1343           low part "subtract" is a compare.  For the high half we have
1344           no choice but to compute into a temporary.  */
1345        tcg_out_dat_rI(s, COND_AL, ARITH_CMP, 0, al, bl, const_bl);
1346        tcg_out_dat_rI(s, COND_AL, ARITH_SBC | TO_CPSR,
1347                       TCG_REG_TMP, ah, bh, const_bh);
1348        return cond;
1349
1350    case TCG_COND_LE:
1351    case TCG_COND_GT:
1352        /* Similar, but with swapped arguments, via reversed subtract.  */
1353        tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR,
1354                       TCG_REG_TMP, al, bl, const_bl);
1355        tcg_out_dat_rI(s, COND_AL, ARITH_RSC | TO_CPSR,
1356                       TCG_REG_TMP, ah, bh, const_bh);
1357        return tcg_swap_cond(cond);
1358
1359    default:
1360        g_assert_not_reached();
1361    }
1362}
1363
1364/*
1365 * Note that TCGReg references Q-registers.
1366 * Q-regno = 2 * D-regno, so shift left by 1 whlie inserting.
1367 */
1368static uint32_t encode_vd(TCGReg rd)
1369{
1370    tcg_debug_assert(rd >= TCG_REG_Q0);
1371    return (extract32(rd, 3, 1) << 22) | (extract32(rd, 0, 3) << 13);
1372}
1373
1374static uint32_t encode_vn(TCGReg rn)
1375{
1376    tcg_debug_assert(rn >= TCG_REG_Q0);
1377    return (extract32(rn, 3, 1) << 7) | (extract32(rn, 0, 3) << 17);
1378}
1379
1380static uint32_t encode_vm(TCGReg rm)
1381{
1382    tcg_debug_assert(rm >= TCG_REG_Q0);
1383    return (extract32(rm, 3, 1) << 5) | (extract32(rm, 0, 3) << 1);
1384}
1385
1386static void tcg_out_vreg2(TCGContext *s, ARMInsn insn, int q, int vece,
1387                          TCGReg d, TCGReg m)
1388{
1389    tcg_out32(s, insn | (vece << 18) | (q << 6) |
1390              encode_vd(d) | encode_vm(m));
1391}
1392
1393static void tcg_out_vreg3(TCGContext *s, ARMInsn insn, int q, int vece,
1394                          TCGReg d, TCGReg n, TCGReg m)
1395{
1396    tcg_out32(s, insn | (vece << 20) | (q << 6) |
1397              encode_vd(d) | encode_vn(n) | encode_vm(m));
1398}
1399
1400static void tcg_out_vmovi(TCGContext *s, TCGReg rd,
1401                          int q, int op, int cmode, uint8_t imm8)
1402{
1403    tcg_out32(s, INSN_VMOVI | encode_vd(rd) | (q << 6) | (op << 5)
1404              | (cmode << 8) | extract32(imm8, 0, 4)
1405              | (extract32(imm8, 4, 3) << 16)
1406              | (extract32(imm8, 7, 1) << 24));
1407}
1408
1409static void tcg_out_vshifti(TCGContext *s, ARMInsn insn, int q,
1410                            TCGReg rd, TCGReg rm, int l_imm6)
1411{
1412    tcg_out32(s, insn | (q << 6) | encode_vd(rd) | encode_vm(rm) |
1413              (extract32(l_imm6, 6, 1) << 7) |
1414              (extract32(l_imm6, 0, 6) << 16));
1415}
1416
1417static void tcg_out_vldst(TCGContext *s, ARMInsn insn,
1418                          TCGReg rd, TCGReg rn, int offset)
1419{
1420    if (offset != 0) {
1421        if (check_fit_imm(offset) || check_fit_imm(-offset)) {
1422            tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
1423                            TCG_REG_TMP, rn, offset, true);
1424        } else {
1425            tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, offset);
1426            tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1427                            TCG_REG_TMP, TCG_REG_TMP, rn, 0);
1428        }
1429        rn = TCG_REG_TMP;
1430    }
1431    tcg_out32(s, insn | (rn << 16) | encode_vd(rd) | 0xf);
1432}
1433
1434#ifdef CONFIG_SOFTMMU
1435#include "../tcg-ldst.c.inc"
1436
1437/* helper signature: helper_ret_ld_mmu(CPUState *env, target_ulong addr,
1438 *                                     int mmu_idx, uintptr_t ra)
1439 */
1440static void * const qemu_ld_helpers[MO_SSIZE + 1] = {
1441    [MO_UB]   = helper_ret_ldub_mmu,
1442    [MO_SB]   = helper_ret_ldsb_mmu,
1443#ifdef HOST_WORDS_BIGENDIAN
1444    [MO_UW] = helper_be_lduw_mmu,
1445    [MO_UL] = helper_be_ldul_mmu,
1446    [MO_Q]  = helper_be_ldq_mmu,
1447    [MO_SW] = helper_be_ldsw_mmu,
1448    [MO_SL] = helper_be_ldul_mmu,
1449#else
1450    [MO_UW] = helper_le_lduw_mmu,
1451    [MO_UL] = helper_le_ldul_mmu,
1452    [MO_Q]  = helper_le_ldq_mmu,
1453    [MO_SW] = helper_le_ldsw_mmu,
1454    [MO_SL] = helper_le_ldul_mmu,
1455#endif
1456};
1457
1458/* helper signature: helper_ret_st_mmu(CPUState *env, target_ulong addr,
1459 *                                     uintxx_t val, int mmu_idx, uintptr_t ra)
1460 */
1461static void * const qemu_st_helpers[MO_SIZE + 1] = {
1462    [MO_8]   = helper_ret_stb_mmu,
1463#ifdef HOST_WORDS_BIGENDIAN
1464    [MO_16] = helper_be_stw_mmu,
1465    [MO_32] = helper_be_stl_mmu,
1466    [MO_64] = helper_be_stq_mmu,
1467#else
1468    [MO_16] = helper_le_stw_mmu,
1469    [MO_32] = helper_le_stl_mmu,
1470    [MO_64] = helper_le_stq_mmu,
1471#endif
1472};
1473
1474/* Helper routines for marshalling helper function arguments into
1475 * the correct registers and stack.
1476 * argreg is where we want to put this argument, arg is the argument itself.
1477 * Return value is the updated argreg ready for the next call.
1478 * Note that argreg 0..3 is real registers, 4+ on stack.
1479 *
1480 * We provide routines for arguments which are: immediate, 32 bit
1481 * value in register, 16 and 8 bit values in register (which must be zero
1482 * extended before use) and 64 bit value in a lo:hi register pair.
1483 */
1484#define DEFINE_TCG_OUT_ARG(NAME, ARGTYPE, MOV_ARG, EXT_ARG)                \
1485static TCGReg NAME(TCGContext *s, TCGReg argreg, ARGTYPE arg)              \
1486{                                                                          \
1487    if (argreg < 4) {                                                      \
1488        MOV_ARG(s, COND_AL, argreg, arg);                                  \
1489    } else {                                                               \
1490        int ofs = (argreg - 4) * 4;                                        \
1491        EXT_ARG;                                                           \
1492        tcg_debug_assert(ofs + 4 <= TCG_STATIC_CALL_ARGS_SIZE);            \
1493        tcg_out_st32_12(s, COND_AL, arg, TCG_REG_CALL_STACK, ofs);         \
1494    }                                                                      \
1495    return argreg + 1;                                                     \
1496}
1497
1498DEFINE_TCG_OUT_ARG(tcg_out_arg_imm32, uint32_t, tcg_out_movi32,
1499    (tcg_out_movi32(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1500DEFINE_TCG_OUT_ARG(tcg_out_arg_reg8, TCGReg, tcg_out_ext8u,
1501    (tcg_out_ext8u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1502DEFINE_TCG_OUT_ARG(tcg_out_arg_reg16, TCGReg, tcg_out_ext16u,
1503    (tcg_out_ext16u(s, COND_AL, TCG_REG_TMP, arg), arg = TCG_REG_TMP))
1504DEFINE_TCG_OUT_ARG(tcg_out_arg_reg32, TCGReg, tcg_out_mov_reg, )
1505
1506static TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
1507                                TCGReg arglo, TCGReg arghi)
1508{
1509    /* 64 bit arguments must go in even/odd register pairs
1510     * and in 8-aligned stack slots.
1511     */
1512    if (argreg & 1) {
1513        argreg++;
1514    }
1515    if (use_armv6_instructions && argreg >= 4
1516        && (arglo & 1) == 0 && arghi == arglo + 1) {
1517        tcg_out_strd_8(s, COND_AL, arglo,
1518                       TCG_REG_CALL_STACK, (argreg - 4) * 4);
1519        return argreg + 2;
1520    } else {
1521        argreg = tcg_out_arg_reg32(s, argreg, arglo);
1522        argreg = tcg_out_arg_reg32(s, argreg, arghi);
1523        return argreg;
1524    }
1525}
1526
1527#define TLB_SHIFT	(CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
1528
1529/* We expect to use an 9-bit sign-magnitude negative offset from ENV.  */
1530QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) > 0);
1531QEMU_BUILD_BUG_ON(TLB_MASK_TABLE_OFS(0) < -256);
1532
1533/* These offsets are built into the LDRD below.  */
1534QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, mask) != 0);
1535QEMU_BUILD_BUG_ON(offsetof(CPUTLBDescFast, table) != 4);
1536
1537/* Load and compare a TLB entry, leaving the flags set.  Returns the register
1538   containing the addend of the tlb entry.  Clobbers R0, R1, R2, TMP.  */
1539
1540static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
1541                               MemOp opc, int mem_index, bool is_load)
1542{
1543    int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read)
1544                   : offsetof(CPUTLBEntry, addr_write));
1545    int fast_off = TLB_MASK_TABLE_OFS(mem_index);
1546    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1547    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1548    unsigned s_bits = opc & MO_SIZE;
1549    unsigned a_bits = get_alignment_bits(opc);
1550
1551    /*
1552     * We don't support inline unaligned acceses, but we can easily
1553     * support overalignment checks.
1554     */
1555    if (a_bits < s_bits) {
1556        a_bits = s_bits;
1557    }
1558
1559    /* Load env_tlb(env)->f[mmu_idx].{mask,table} into {r0,r1}.  */
1560    if (use_armv6_instructions) {
1561        tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_AREG0, fast_off);
1562    } else {
1563        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R0, TCG_AREG0, mask_off);
1564        tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R1, TCG_AREG0, table_off);
1565    }
1566
1567    /* Extract the tlb index from the address into R0.  */
1568    tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_R0, TCG_REG_R0, addrlo,
1569                    SHIFT_IMM_LSR(TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS));
1570
1571    /*
1572     * Add the tlb_table pointer, creating the CPUTLBEntry address in R1.
1573     * Load the tlb comparator into R2/R3 and the fast path addend into R1.
1574     */
1575    if (cmp_off == 0) {
1576        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1577            tcg_out_ldrd_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1578        } else {
1579            tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R2, TCG_REG_R1, TCG_REG_R0);
1580        }
1581    } else {
1582        tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
1583                        TCG_REG_R1, TCG_REG_R1, TCG_REG_R0, 0);
1584        if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
1585            tcg_out_ldrd_8(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1586        } else {
1587            tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R1, cmp_off);
1588        }
1589    }
1590    if (!use_armv6_instructions && TARGET_LONG_BITS == 64) {
1591        tcg_out_ld32_12(s, COND_AL, TCG_REG_R3, TCG_REG_R1, cmp_off + 4);
1592    }
1593
1594    /* Load the tlb addend.  */
1595    tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R1,
1596                    offsetof(CPUTLBEntry, addend));
1597
1598    /*
1599     * Check alignment, check comparators.
1600     * Do this in no more than 3 insns.  Use MOVW for v7, if possible,
1601     * to reduce the number of sequential conditional instructions.
1602     * Almost all guests have at least 4k pages, which means that we need
1603     * to clear at least 9 bits even for an 8-byte memory, which means it
1604     * isn't worth checking for an immediate operand for BIC.
1605     */
1606    if (use_armv7_instructions && TARGET_PAGE_BITS <= 16) {
1607        tcg_target_ulong mask = ~(TARGET_PAGE_MASK | ((1 << a_bits) - 1));
1608
1609        tcg_out_movi32(s, COND_AL, TCG_REG_TMP, mask);
1610        tcg_out_dat_reg(s, COND_AL, ARITH_BIC, TCG_REG_TMP,
1611                        addrlo, TCG_REG_TMP, 0);
1612        tcg_out_dat_reg(s, COND_AL, ARITH_CMP, 0, TCG_REG_R2, TCG_REG_TMP, 0);
1613    } else {
1614        if (a_bits) {
1615            tcg_out_dat_imm(s, COND_AL, ARITH_TST, 0, addrlo,
1616                            (1 << a_bits) - 1);
1617        }
1618        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0, addrlo,
1619                        SHIFT_IMM_LSR(TARGET_PAGE_BITS));
1620        tcg_out_dat_reg(s, (a_bits ? COND_EQ : COND_AL), ARITH_CMP,
1621                        0, TCG_REG_R2, TCG_REG_TMP,
1622                        SHIFT_IMM_LSL(TARGET_PAGE_BITS));
1623    }
1624
1625    if (TARGET_LONG_BITS == 64) {
1626        tcg_out_dat_reg(s, COND_EQ, ARITH_CMP, 0, TCG_REG_R3, addrhi, 0);
1627    }
1628
1629    return TCG_REG_R1;
1630}
1631
1632/* Record the context of a call to the out of line helper code for the slow
1633   path for a load or store, so that we can later generate the correct
1634   helper code.  */
1635static void add_qemu_ldst_label(TCGContext *s, bool is_ld, MemOpIdx oi,
1636                                TCGReg datalo, TCGReg datahi, TCGReg addrlo,
1637                                TCGReg addrhi, tcg_insn_unit *raddr,
1638                                tcg_insn_unit *label_ptr)
1639{
1640    TCGLabelQemuLdst *label = new_ldst_label(s);
1641
1642    label->is_ld = is_ld;
1643    label->oi = oi;
1644    label->datalo_reg = datalo;
1645    label->datahi_reg = datahi;
1646    label->addrlo_reg = addrlo;
1647    label->addrhi_reg = addrhi;
1648    label->raddr = tcg_splitwx_to_rx(raddr);
1649    label->label_ptr[0] = label_ptr;
1650}
1651
1652static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1653{
1654    TCGReg argreg, datalo, datahi;
1655    MemOpIdx oi = lb->oi;
1656    MemOp opc = get_memop(oi);
1657    void *func;
1658
1659    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1660        return false;
1661    }
1662
1663    argreg = tcg_out_arg_reg32(s, TCG_REG_R0, TCG_AREG0);
1664    if (TARGET_LONG_BITS == 64) {
1665        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1666    } else {
1667        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1668    }
1669    argreg = tcg_out_arg_imm32(s, argreg, oi);
1670    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1671
1672    /* For armv6 we can use the canonical unsigned helpers and minimize
1673       icache usage.  For pre-armv6, use the signed helpers since we do
1674       not have a single insn sign-extend.  */
1675    if (use_armv6_instructions) {
1676        func = qemu_ld_helpers[opc & MO_SIZE];
1677    } else {
1678        func = qemu_ld_helpers[opc & MO_SSIZE];
1679        if (opc & MO_SIGN) {
1680            opc = MO_UL;
1681        }
1682    }
1683    tcg_out_call(s, func);
1684
1685    datalo = lb->datalo_reg;
1686    datahi = lb->datahi_reg;
1687    switch (opc & MO_SSIZE) {
1688    case MO_SB:
1689        tcg_out_ext8s(s, COND_AL, datalo, TCG_REG_R0);
1690        break;
1691    case MO_SW:
1692        tcg_out_ext16s(s, COND_AL, datalo, TCG_REG_R0);
1693        break;
1694    default:
1695        tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1696        break;
1697    case MO_Q:
1698        if (datalo != TCG_REG_R1) {
1699            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1700            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1701        } else if (datahi != TCG_REG_R0) {
1702            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1703            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_R0);
1704        } else {
1705            tcg_out_mov_reg(s, COND_AL, TCG_REG_TMP, TCG_REG_R0);
1706            tcg_out_mov_reg(s, COND_AL, datahi, TCG_REG_R1);
1707            tcg_out_mov_reg(s, COND_AL, datalo, TCG_REG_TMP);
1708        }
1709        break;
1710    }
1711
1712    tcg_out_goto(s, COND_AL, lb->raddr);
1713    return true;
1714}
1715
1716static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1717{
1718    TCGReg argreg, datalo, datahi;
1719    MemOpIdx oi = lb->oi;
1720    MemOp opc = get_memop(oi);
1721
1722    if (!reloc_pc24(lb->label_ptr[0], tcg_splitwx_to_rx(s->code_ptr))) {
1723        return false;
1724    }
1725
1726    argreg = TCG_REG_R0;
1727    argreg = tcg_out_arg_reg32(s, argreg, TCG_AREG0);
1728    if (TARGET_LONG_BITS == 64) {
1729        argreg = tcg_out_arg_reg64(s, argreg, lb->addrlo_reg, lb->addrhi_reg);
1730    } else {
1731        argreg = tcg_out_arg_reg32(s, argreg, lb->addrlo_reg);
1732    }
1733
1734    datalo = lb->datalo_reg;
1735    datahi = lb->datahi_reg;
1736    switch (opc & MO_SIZE) {
1737    case MO_8:
1738        argreg = tcg_out_arg_reg8(s, argreg, datalo);
1739        break;
1740    case MO_16:
1741        argreg = tcg_out_arg_reg16(s, argreg, datalo);
1742        break;
1743    case MO_32:
1744    default:
1745        argreg = tcg_out_arg_reg32(s, argreg, datalo);
1746        break;
1747    case MO_64:
1748        argreg = tcg_out_arg_reg64(s, argreg, datalo, datahi);
1749        break;
1750    }
1751
1752    argreg = tcg_out_arg_imm32(s, argreg, oi);
1753    argreg = tcg_out_arg_reg32(s, argreg, TCG_REG_R14);
1754
1755    /* Tail-call to the helper, which will return to the fast path.  */
1756    tcg_out_goto(s, COND_AL, qemu_st_helpers[opc & MO_SIZE]);
1757    return true;
1758}
1759#endif /* SOFTMMU */
1760
1761static void tcg_out_qemu_ld_index(TCGContext *s, MemOp opc,
1762                                  TCGReg datalo, TCGReg datahi,
1763                                  TCGReg addrlo, TCGReg addend)
1764{
1765    /* Byte swapping is left to middle-end expansion. */
1766    tcg_debug_assert((opc & MO_BSWAP) == 0);
1767
1768    switch (opc & MO_SSIZE) {
1769    case MO_UB:
1770        tcg_out_ld8_r(s, COND_AL, datalo, addrlo, addend);
1771        break;
1772    case MO_SB:
1773        tcg_out_ld8s_r(s, COND_AL, datalo, addrlo, addend);
1774        break;
1775    case MO_UW:
1776        tcg_out_ld16u_r(s, COND_AL, datalo, addrlo, addend);
1777        break;
1778    case MO_SW:
1779        tcg_out_ld16s_r(s, COND_AL, datalo, addrlo, addend);
1780        break;
1781    case MO_UL:
1782        tcg_out_ld32_r(s, COND_AL, datalo, addrlo, addend);
1783        break;
1784    case MO_Q:
1785        /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1786        if (USING_SOFTMMU && use_armv6_instructions
1787            && (datalo & 1) == 0 && datahi == datalo + 1) {
1788            tcg_out_ldrd_r(s, COND_AL, datalo, addrlo, addend);
1789        } else if (datalo != addend) {
1790            tcg_out_ld32_rwb(s, COND_AL, datalo, addend, addrlo);
1791            tcg_out_ld32_12(s, COND_AL, datahi, addend, 4);
1792        } else {
1793            tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_TMP,
1794                            addend, addrlo, SHIFT_IMM_LSL(0));
1795            tcg_out_ld32_12(s, COND_AL, datalo, TCG_REG_TMP, 0);
1796            tcg_out_ld32_12(s, COND_AL, datahi, TCG_REG_TMP, 4);
1797        }
1798        break;
1799    default:
1800        g_assert_not_reached();
1801    }
1802}
1803
1804#ifndef CONFIG_SOFTMMU
1805static void tcg_out_qemu_ld_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1806                                   TCGReg datahi, TCGReg addrlo)
1807{
1808    /* Byte swapping is left to middle-end expansion. */
1809    tcg_debug_assert((opc & MO_BSWAP) == 0);
1810
1811    switch (opc & MO_SSIZE) {
1812    case MO_UB:
1813        tcg_out_ld8_12(s, COND_AL, datalo, addrlo, 0);
1814        break;
1815    case MO_SB:
1816        tcg_out_ld8s_8(s, COND_AL, datalo, addrlo, 0);
1817        break;
1818    case MO_UW:
1819        tcg_out_ld16u_8(s, COND_AL, datalo, addrlo, 0);
1820        break;
1821    case MO_SW:
1822        tcg_out_ld16s_8(s, COND_AL, datalo, addrlo, 0);
1823        break;
1824    case MO_UL:
1825        tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1826        break;
1827    case MO_Q:
1828        /* Avoid ldrd for user-only emulation, to handle unaligned.  */
1829        if (USING_SOFTMMU && use_armv6_instructions
1830            && (datalo & 1) == 0 && datahi == datalo + 1) {
1831            tcg_out_ldrd_8(s, COND_AL, datalo, addrlo, 0);
1832        } else if (datalo == addrlo) {
1833            tcg_out_ld32_12(s, COND_AL, datahi, addrlo, 4);
1834            tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1835        } else {
1836            tcg_out_ld32_12(s, COND_AL, datalo, addrlo, 0);
1837            tcg_out_ld32_12(s, COND_AL, datahi, addrlo, 4);
1838        }
1839        break;
1840    default:
1841        g_assert_not_reached();
1842    }
1843}
1844#endif
1845
1846static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is64)
1847{
1848    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1849    MemOpIdx oi;
1850    MemOp opc;
1851#ifdef CONFIG_SOFTMMU
1852    int mem_index;
1853    TCGReg addend;
1854    tcg_insn_unit *label_ptr;
1855#endif
1856
1857    datalo = *args++;
1858    datahi = (is64 ? *args++ : 0);
1859    addrlo = *args++;
1860    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1861    oi = *args++;
1862    opc = get_memop(oi);
1863
1864#ifdef CONFIG_SOFTMMU
1865    mem_index = get_mmuidx(oi);
1866    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 1);
1867
1868    /* This a conditional BL only to load a pointer within this opcode into LR
1869       for the slow path.  We will not be using the value for a tail call.  */
1870    label_ptr = s->code_ptr;
1871    tcg_out_bl_imm(s, COND_NE, 0);
1872
1873    tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, addend);
1874
1875    add_qemu_ldst_label(s, true, oi, datalo, datahi, addrlo, addrhi,
1876                        s->code_ptr, label_ptr);
1877#else /* !CONFIG_SOFTMMU */
1878    if (guest_base) {
1879        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1880        tcg_out_qemu_ld_index(s, opc, datalo, datahi, addrlo, TCG_REG_TMP);
1881    } else {
1882        tcg_out_qemu_ld_direct(s, opc, datalo, datahi, addrlo);
1883    }
1884#endif
1885}
1886
1887static void tcg_out_qemu_st_index(TCGContext *s, ARMCond cond, MemOp opc,
1888                                  TCGReg datalo, TCGReg datahi,
1889                                  TCGReg addrlo, TCGReg addend)
1890{
1891    /* Byte swapping is left to middle-end expansion. */
1892    tcg_debug_assert((opc & MO_BSWAP) == 0);
1893
1894    switch (opc & MO_SIZE) {
1895    case MO_8:
1896        tcg_out_st8_r(s, cond, datalo, addrlo, addend);
1897        break;
1898    case MO_16:
1899        tcg_out_st16_r(s, cond, datalo, addrlo, addend);
1900        break;
1901    case MO_32:
1902        tcg_out_st32_r(s, cond, datalo, addrlo, addend);
1903        break;
1904    case MO_64:
1905        /* Avoid strd for user-only emulation, to handle unaligned.  */
1906        if (USING_SOFTMMU && use_armv6_instructions
1907            && (datalo & 1) == 0 && datahi == datalo + 1) {
1908            tcg_out_strd_r(s, cond, datalo, addrlo, addend);
1909        } else {
1910            tcg_out_st32_rwb(s, cond, datalo, addend, addrlo);
1911            tcg_out_st32_12(s, cond, datahi, addend, 4);
1912        }
1913        break;
1914    default:
1915        g_assert_not_reached();
1916    }
1917}
1918
1919#ifndef CONFIG_SOFTMMU
1920static void tcg_out_qemu_st_direct(TCGContext *s, MemOp opc, TCGReg datalo,
1921                                   TCGReg datahi, TCGReg addrlo)
1922{
1923    /* Byte swapping is left to middle-end expansion. */
1924    tcg_debug_assert((opc & MO_BSWAP) == 0);
1925
1926    switch (opc & MO_SIZE) {
1927    case MO_8:
1928        tcg_out_st8_12(s, COND_AL, datalo, addrlo, 0);
1929        break;
1930    case MO_16:
1931        tcg_out_st16_8(s, COND_AL, datalo, addrlo, 0);
1932        break;
1933    case MO_32:
1934        tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1935        break;
1936    case MO_64:
1937        /* Avoid strd for user-only emulation, to handle unaligned.  */
1938        if (USING_SOFTMMU && use_armv6_instructions
1939            && (datalo & 1) == 0 && datahi == datalo + 1) {
1940            tcg_out_strd_8(s, COND_AL, datalo, addrlo, 0);
1941        } else {
1942            tcg_out_st32_12(s, COND_AL, datalo, addrlo, 0);
1943            tcg_out_st32_12(s, COND_AL, datahi, addrlo, 4);
1944        }
1945        break;
1946    default:
1947        g_assert_not_reached();
1948    }
1949}
1950#endif
1951
1952static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is64)
1953{
1954    TCGReg addrlo, datalo, datahi, addrhi __attribute__((unused));
1955    MemOpIdx oi;
1956    MemOp opc;
1957#ifdef CONFIG_SOFTMMU
1958    int mem_index;
1959    TCGReg addend;
1960    tcg_insn_unit *label_ptr;
1961#endif
1962
1963    datalo = *args++;
1964    datahi = (is64 ? *args++ : 0);
1965    addrlo = *args++;
1966    addrhi = (TARGET_LONG_BITS == 64 ? *args++ : 0);
1967    oi = *args++;
1968    opc = get_memop(oi);
1969
1970#ifdef CONFIG_SOFTMMU
1971    mem_index = get_mmuidx(oi);
1972    addend = tcg_out_tlb_read(s, addrlo, addrhi, opc, mem_index, 0);
1973
1974    tcg_out_qemu_st_index(s, COND_EQ, opc, datalo, datahi, addrlo, addend);
1975
1976    /* The conditional call must come last, as we're going to return here.  */
1977    label_ptr = s->code_ptr;
1978    tcg_out_bl_imm(s, COND_NE, 0);
1979
1980    add_qemu_ldst_label(s, false, oi, datalo, datahi, addrlo, addrhi,
1981                        s->code_ptr, label_ptr);
1982#else /* !CONFIG_SOFTMMU */
1983    if (guest_base) {
1984        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP, guest_base);
1985        tcg_out_qemu_st_index(s, COND_AL, opc, datalo,
1986                              datahi, addrlo, TCG_REG_TMP);
1987    } else {
1988        tcg_out_qemu_st_direct(s, opc, datalo, datahi, addrlo);
1989    }
1990#endif
1991}
1992
1993static void tcg_out_epilogue(TCGContext *s);
1994
1995static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1996                       const TCGArg args[TCG_MAX_OP_ARGS],
1997                       const int const_args[TCG_MAX_OP_ARGS])
1998{
1999    TCGArg a0, a1, a2, a3, a4, a5;
2000    int c;
2001
2002    switch (opc) {
2003    case INDEX_op_exit_tb:
2004        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, args[0]);
2005        tcg_out_epilogue(s);
2006        break;
2007    case INDEX_op_goto_tb:
2008        {
2009            /* Indirect jump method */
2010            intptr_t ptr, dif, dil;
2011            TCGReg base = TCG_REG_PC;
2012
2013            tcg_debug_assert(s->tb_jmp_insn_offset == 0);
2014            ptr = (intptr_t)tcg_splitwx_to_rx(s->tb_jmp_target_addr + args[0]);
2015            dif = tcg_pcrel_diff(s, (void *)ptr) - 8;
2016            dil = sextract32(dif, 0, 12);
2017            if (dif != dil) {
2018                /* The TB is close, but outside the 12 bits addressable by
2019                   the load.  We can extend this to 20 bits with a sub of a
2020                   shifted immediate from pc.  In the vastly unlikely event
2021                   the code requires more than 1MB, we'll use 2 insns and
2022                   be no worse off.  */
2023                base = TCG_REG_R0;
2024                tcg_out_movi32(s, COND_AL, base, ptr - dil);
2025            }
2026            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, base, dil);
2027            set_jmp_reset_offset(s, args[0]);
2028        }
2029        break;
2030    case INDEX_op_goto_ptr:
2031        tcg_out_b_reg(s, COND_AL, args[0]);
2032        break;
2033    case INDEX_op_br:
2034        tcg_out_goto_label(s, COND_AL, arg_label(args[0]));
2035        break;
2036
2037    case INDEX_op_ld8u_i32:
2038        tcg_out_ld8u(s, COND_AL, args[0], args[1], args[2]);
2039        break;
2040    case INDEX_op_ld8s_i32:
2041        tcg_out_ld8s(s, COND_AL, args[0], args[1], args[2]);
2042        break;
2043    case INDEX_op_ld16u_i32:
2044        tcg_out_ld16u(s, COND_AL, args[0], args[1], args[2]);
2045        break;
2046    case INDEX_op_ld16s_i32:
2047        tcg_out_ld16s(s, COND_AL, args[0], args[1], args[2]);
2048        break;
2049    case INDEX_op_ld_i32:
2050        tcg_out_ld32u(s, COND_AL, args[0], args[1], args[2]);
2051        break;
2052    case INDEX_op_st8_i32:
2053        tcg_out_st8(s, COND_AL, args[0], args[1], args[2]);
2054        break;
2055    case INDEX_op_st16_i32:
2056        tcg_out_st16(s, COND_AL, args[0], args[1], args[2]);
2057        break;
2058    case INDEX_op_st_i32:
2059        tcg_out_st32(s, COND_AL, args[0], args[1], args[2]);
2060        break;
2061
2062    case INDEX_op_movcond_i32:
2063        /* Constraints mean that v2 is always in the same register as dest,
2064         * so we only need to do "if condition passed, move v1 to dest".
2065         */
2066        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2067                        args[1], args[2], const_args[2]);
2068        tcg_out_dat_rIK(s, tcg_cond_to_arm_cond[args[5]], ARITH_MOV,
2069                        ARITH_MVN, args[0], 0, args[3], const_args[3]);
2070        break;
2071    case INDEX_op_add_i32:
2072        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD, ARITH_SUB,
2073                        args[0], args[1], args[2], const_args[2]);
2074        break;
2075    case INDEX_op_sub_i32:
2076        if (const_args[1]) {
2077            if (const_args[2]) {
2078                tcg_out_movi32(s, COND_AL, args[0], args[1] - args[2]);
2079            } else {
2080                tcg_out_dat_rI(s, COND_AL, ARITH_RSB,
2081                               args[0], args[2], args[1], 1);
2082            }
2083        } else {
2084            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB, ARITH_ADD,
2085                            args[0], args[1], args[2], const_args[2]);
2086        }
2087        break;
2088    case INDEX_op_and_i32:
2089        tcg_out_dat_rIK(s, COND_AL, ARITH_AND, ARITH_BIC,
2090                        args[0], args[1], args[2], const_args[2]);
2091        break;
2092    case INDEX_op_andc_i32:
2093        tcg_out_dat_rIK(s, COND_AL, ARITH_BIC, ARITH_AND,
2094                        args[0], args[1], args[2], const_args[2]);
2095        break;
2096    case INDEX_op_or_i32:
2097        c = ARITH_ORR;
2098        goto gen_arith;
2099    case INDEX_op_xor_i32:
2100        c = ARITH_EOR;
2101        /* Fall through.  */
2102    gen_arith:
2103        tcg_out_dat_rI(s, COND_AL, c, args[0], args[1], args[2], const_args[2]);
2104        break;
2105    case INDEX_op_add2_i32:
2106        a0 = args[0], a1 = args[1], a2 = args[2];
2107        a3 = args[3], a4 = args[4], a5 = args[5];
2108        if (a0 == a3 || (a0 == a5 && !const_args[5])) {
2109            a0 = TCG_REG_TMP;
2110        }
2111        tcg_out_dat_rIN(s, COND_AL, ARITH_ADD | TO_CPSR, ARITH_SUB | TO_CPSR,
2112                        a0, a2, a4, const_args[4]);
2113        tcg_out_dat_rIK(s, COND_AL, ARITH_ADC, ARITH_SBC,
2114                        a1, a3, a5, const_args[5]);
2115        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2116        break;
2117    case INDEX_op_sub2_i32:
2118        a0 = args[0], a1 = args[1], a2 = args[2];
2119        a3 = args[3], a4 = args[4], a5 = args[5];
2120        if ((a0 == a3 && !const_args[3]) || (a0 == a5 && !const_args[5])) {
2121            a0 = TCG_REG_TMP;
2122        }
2123        if (const_args[2]) {
2124            if (const_args[4]) {
2125                tcg_out_movi32(s, COND_AL, a0, a4);
2126                a4 = a0;
2127            }
2128            tcg_out_dat_rI(s, COND_AL, ARITH_RSB | TO_CPSR, a0, a4, a2, 1);
2129        } else {
2130            tcg_out_dat_rIN(s, COND_AL, ARITH_SUB | TO_CPSR,
2131                            ARITH_ADD | TO_CPSR, a0, a2, a4, const_args[4]);
2132        }
2133        if (const_args[3]) {
2134            if (const_args[5]) {
2135                tcg_out_movi32(s, COND_AL, a1, a5);
2136                a5 = a1;
2137            }
2138            tcg_out_dat_rI(s, COND_AL, ARITH_RSC, a1, a5, a3, 1);
2139        } else {
2140            tcg_out_dat_rIK(s, COND_AL, ARITH_SBC, ARITH_ADC,
2141                            a1, a3, a5, const_args[5]);
2142        }
2143        tcg_out_mov_reg(s, COND_AL, args[0], a0);
2144        break;
2145    case INDEX_op_neg_i32:
2146        tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
2147        break;
2148    case INDEX_op_not_i32:
2149        tcg_out_dat_reg(s, COND_AL,
2150                        ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
2151        break;
2152    case INDEX_op_mul_i32:
2153        tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
2154        break;
2155    case INDEX_op_mulu2_i32:
2156        tcg_out_umull32(s, COND_AL, args[0], args[1], args[2], args[3]);
2157        break;
2158    case INDEX_op_muls2_i32:
2159        tcg_out_smull32(s, COND_AL, args[0], args[1], args[2], args[3]);
2160        break;
2161    /* XXX: Perhaps args[2] & 0x1f is wrong */
2162    case INDEX_op_shl_i32:
2163        c = const_args[2] ?
2164                SHIFT_IMM_LSL(args[2] & 0x1f) : SHIFT_REG_LSL(args[2]);
2165        goto gen_shift32;
2166    case INDEX_op_shr_i32:
2167        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_LSR(args[2] & 0x1f) :
2168                SHIFT_IMM_LSL(0) : SHIFT_REG_LSR(args[2]);
2169        goto gen_shift32;
2170    case INDEX_op_sar_i32:
2171        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ASR(args[2] & 0x1f) :
2172                SHIFT_IMM_LSL(0) : SHIFT_REG_ASR(args[2]);
2173        goto gen_shift32;
2174    case INDEX_op_rotr_i32:
2175        c = const_args[2] ? (args[2] & 0x1f) ? SHIFT_IMM_ROR(args[2] & 0x1f) :
2176                SHIFT_IMM_LSL(0) : SHIFT_REG_ROR(args[2]);
2177        /* Fall through.  */
2178    gen_shift32:
2179        tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1], c);
2180        break;
2181
2182    case INDEX_op_rotl_i32:
2183        if (const_args[2]) {
2184            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
2185                            ((0x20 - args[2]) & 0x1f) ?
2186                            SHIFT_IMM_ROR((0x20 - args[2]) & 0x1f) :
2187                            SHIFT_IMM_LSL(0));
2188        } else {
2189            tcg_out_dat_imm(s, COND_AL, ARITH_RSB, TCG_REG_TMP, args[2], 0x20);
2190            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0, args[1],
2191                            SHIFT_REG_ROR(TCG_REG_TMP));
2192        }
2193        break;
2194
2195    case INDEX_op_ctz_i32:
2196        tcg_out_dat_reg(s, COND_AL, INSN_RBIT, TCG_REG_TMP, 0, args[1], 0);
2197        a1 = TCG_REG_TMP;
2198        goto do_clz;
2199
2200    case INDEX_op_clz_i32:
2201        a1 = args[1];
2202    do_clz:
2203        a0 = args[0];
2204        a2 = args[2];
2205        c = const_args[2];
2206        if (c && a2 == 32) {
2207            tcg_out_dat_reg(s, COND_AL, INSN_CLZ, a0, 0, a1, 0);
2208            break;
2209        }
2210        tcg_out_dat_imm(s, COND_AL, ARITH_CMP, 0, a1, 0);
2211        tcg_out_dat_reg(s, COND_NE, INSN_CLZ, a0, 0, a1, 0);
2212        if (c || a0 != a2) {
2213            tcg_out_dat_rIK(s, COND_EQ, ARITH_MOV, ARITH_MVN, a0, 0, a2, c);
2214        }
2215        break;
2216
2217    case INDEX_op_brcond_i32:
2218        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2219                       args[0], args[1], const_args[1]);
2220        tcg_out_goto_label(s, tcg_cond_to_arm_cond[args[2]],
2221                           arg_label(args[3]));
2222        break;
2223    case INDEX_op_setcond_i32:
2224        tcg_out_dat_rIN(s, COND_AL, ARITH_CMP, ARITH_CMN, 0,
2225                        args[1], args[2], const_args[2]);
2226        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[args[3]],
2227                        ARITH_MOV, args[0], 0, 1);
2228        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(args[3])],
2229                        ARITH_MOV, args[0], 0, 0);
2230        break;
2231
2232    case INDEX_op_brcond2_i32:
2233        c = tcg_out_cmp2(s, args, const_args);
2234        tcg_out_goto_label(s, tcg_cond_to_arm_cond[c], arg_label(args[5]));
2235        break;
2236    case INDEX_op_setcond2_i32:
2237        c = tcg_out_cmp2(s, args + 1, const_args + 1);
2238        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[c], ARITH_MOV, args[0], 0, 1);
2239        tcg_out_dat_imm(s, tcg_cond_to_arm_cond[tcg_invert_cond(c)],
2240                        ARITH_MOV, args[0], 0, 0);
2241        break;
2242
2243    case INDEX_op_qemu_ld_i32:
2244        tcg_out_qemu_ld(s, args, 0);
2245        break;
2246    case INDEX_op_qemu_ld_i64:
2247        tcg_out_qemu_ld(s, args, 1);
2248        break;
2249    case INDEX_op_qemu_st_i32:
2250        tcg_out_qemu_st(s, args, 0);
2251        break;
2252    case INDEX_op_qemu_st_i64:
2253        tcg_out_qemu_st(s, args, 1);
2254        break;
2255
2256    case INDEX_op_bswap16_i32:
2257        tcg_out_bswap16(s, COND_AL, args[0], args[1], args[2]);
2258        break;
2259    case INDEX_op_bswap32_i32:
2260        tcg_out_bswap32(s, COND_AL, args[0], args[1]);
2261        break;
2262
2263    case INDEX_op_ext8s_i32:
2264        tcg_out_ext8s(s, COND_AL, args[0], args[1]);
2265        break;
2266    case INDEX_op_ext16s_i32:
2267        tcg_out_ext16s(s, COND_AL, args[0], args[1]);
2268        break;
2269    case INDEX_op_ext16u_i32:
2270        tcg_out_ext16u(s, COND_AL, args[0], args[1]);
2271        break;
2272
2273    case INDEX_op_deposit_i32:
2274        tcg_out_deposit(s, COND_AL, args[0], args[2],
2275                        args[3], args[4], const_args[2]);
2276        break;
2277    case INDEX_op_extract_i32:
2278        tcg_out_extract(s, COND_AL, args[0], args[1], args[2], args[3]);
2279        break;
2280    case INDEX_op_sextract_i32:
2281        tcg_out_sextract(s, COND_AL, args[0], args[1], args[2], args[3]);
2282        break;
2283    case INDEX_op_extract2_i32:
2284        /* ??? These optimization vs zero should be generic.  */
2285        /* ??? But we can't substitute 2 for 1 in the opcode stream yet.  */
2286        if (const_args[1]) {
2287            if (const_args[2]) {
2288                tcg_out_movi(s, TCG_TYPE_REG, args[0], 0);
2289            } else {
2290                tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2291                                args[2], SHIFT_IMM_LSL(32 - args[3]));
2292            }
2293        } else if (const_args[2]) {
2294            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, args[0], 0,
2295                            args[1], SHIFT_IMM_LSR(args[3]));
2296        } else {
2297            /* We can do extract2 in 2 insns, vs the 3 required otherwise.  */
2298            tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP, 0,
2299                            args[2], SHIFT_IMM_LSL(32 - args[3]));
2300            tcg_out_dat_reg(s, COND_AL, ARITH_ORR, args[0], TCG_REG_TMP,
2301                            args[1], SHIFT_IMM_LSR(args[3]));
2302        }
2303        break;
2304
2305    case INDEX_op_div_i32:
2306        tcg_out_sdiv(s, COND_AL, args[0], args[1], args[2]);
2307        break;
2308    case INDEX_op_divu_i32:
2309        tcg_out_udiv(s, COND_AL, args[0], args[1], args[2]);
2310        break;
2311
2312    case INDEX_op_mb:
2313        tcg_out_mb(s, args[0]);
2314        break;
2315
2316    case INDEX_op_mov_i32:  /* Always emitted via tcg_out_mov.  */
2317    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2318    default:
2319        tcg_abort();
2320    }
2321}
2322
2323static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
2324{
2325    switch (op) {
2326    case INDEX_op_goto_ptr:
2327        return C_O0_I1(r);
2328
2329    case INDEX_op_ld8u_i32:
2330    case INDEX_op_ld8s_i32:
2331    case INDEX_op_ld16u_i32:
2332    case INDEX_op_ld16s_i32:
2333    case INDEX_op_ld_i32:
2334    case INDEX_op_neg_i32:
2335    case INDEX_op_not_i32:
2336    case INDEX_op_bswap16_i32:
2337    case INDEX_op_bswap32_i32:
2338    case INDEX_op_ext8s_i32:
2339    case INDEX_op_ext16s_i32:
2340    case INDEX_op_ext16u_i32:
2341    case INDEX_op_extract_i32:
2342    case INDEX_op_sextract_i32:
2343        return C_O1_I1(r, r);
2344
2345    case INDEX_op_st8_i32:
2346    case INDEX_op_st16_i32:
2347    case INDEX_op_st_i32:
2348        return C_O0_I2(r, r);
2349
2350    case INDEX_op_add_i32:
2351    case INDEX_op_sub_i32:
2352    case INDEX_op_setcond_i32:
2353        return C_O1_I2(r, r, rIN);
2354
2355    case INDEX_op_and_i32:
2356    case INDEX_op_andc_i32:
2357    case INDEX_op_clz_i32:
2358    case INDEX_op_ctz_i32:
2359        return C_O1_I2(r, r, rIK);
2360
2361    case INDEX_op_mul_i32:
2362    case INDEX_op_div_i32:
2363    case INDEX_op_divu_i32:
2364        return C_O1_I2(r, r, r);
2365
2366    case INDEX_op_mulu2_i32:
2367    case INDEX_op_muls2_i32:
2368        return C_O2_I2(r, r, r, r);
2369
2370    case INDEX_op_or_i32:
2371    case INDEX_op_xor_i32:
2372        return C_O1_I2(r, r, rI);
2373
2374    case INDEX_op_shl_i32:
2375    case INDEX_op_shr_i32:
2376    case INDEX_op_sar_i32:
2377    case INDEX_op_rotl_i32:
2378    case INDEX_op_rotr_i32:
2379        return C_O1_I2(r, r, ri);
2380
2381    case INDEX_op_brcond_i32:
2382        return C_O0_I2(r, rIN);
2383    case INDEX_op_deposit_i32:
2384        return C_O1_I2(r, 0, rZ);
2385    case INDEX_op_extract2_i32:
2386        return C_O1_I2(r, rZ, rZ);
2387    case INDEX_op_movcond_i32:
2388        return C_O1_I4(r, r, rIN, rIK, 0);
2389    case INDEX_op_add2_i32:
2390        return C_O2_I4(r, r, r, r, rIN, rIK);
2391    case INDEX_op_sub2_i32:
2392        return C_O2_I4(r, r, rI, rI, rIN, rIK);
2393    case INDEX_op_brcond2_i32:
2394        return C_O0_I4(r, r, rI, rI);
2395    case INDEX_op_setcond2_i32:
2396        return C_O1_I4(r, r, r, rI, rI);
2397
2398    case INDEX_op_qemu_ld_i32:
2399        return TARGET_LONG_BITS == 32 ? C_O1_I1(r, l) : C_O1_I2(r, l, l);
2400    case INDEX_op_qemu_ld_i64:
2401        return TARGET_LONG_BITS == 32 ? C_O2_I1(r, r, l) : C_O2_I2(r, r, l, l);
2402    case INDEX_op_qemu_st_i32:
2403        return TARGET_LONG_BITS == 32 ? C_O0_I2(s, s) : C_O0_I3(s, s, s);
2404    case INDEX_op_qemu_st_i64:
2405        return TARGET_LONG_BITS == 32 ? C_O0_I3(s, s, s) : C_O0_I4(s, s, s, s);
2406
2407    case INDEX_op_st_vec:
2408        return C_O0_I2(w, r);
2409    case INDEX_op_ld_vec:
2410    case INDEX_op_dupm_vec:
2411        return C_O1_I1(w, r);
2412    case INDEX_op_dup_vec:
2413        return C_O1_I1(w, wr);
2414    case INDEX_op_abs_vec:
2415    case INDEX_op_neg_vec:
2416    case INDEX_op_not_vec:
2417    case INDEX_op_shli_vec:
2418    case INDEX_op_shri_vec:
2419    case INDEX_op_sari_vec:
2420        return C_O1_I1(w, w);
2421    case INDEX_op_dup2_vec:
2422    case INDEX_op_add_vec:
2423    case INDEX_op_mul_vec:
2424    case INDEX_op_smax_vec:
2425    case INDEX_op_smin_vec:
2426    case INDEX_op_ssadd_vec:
2427    case INDEX_op_sssub_vec:
2428    case INDEX_op_sub_vec:
2429    case INDEX_op_umax_vec:
2430    case INDEX_op_umin_vec:
2431    case INDEX_op_usadd_vec:
2432    case INDEX_op_ussub_vec:
2433    case INDEX_op_xor_vec:
2434    case INDEX_op_arm_sshl_vec:
2435    case INDEX_op_arm_ushl_vec:
2436        return C_O1_I2(w, w, w);
2437    case INDEX_op_arm_sli_vec:
2438        return C_O1_I2(w, 0, w);
2439    case INDEX_op_or_vec:
2440    case INDEX_op_andc_vec:
2441        return C_O1_I2(w, w, wO);
2442    case INDEX_op_and_vec:
2443    case INDEX_op_orc_vec:
2444        return C_O1_I2(w, w, wV);
2445    case INDEX_op_cmp_vec:
2446        return C_O1_I2(w, w, wZ);
2447    case INDEX_op_bitsel_vec:
2448        return C_O1_I3(w, w, w, w);
2449    default:
2450        g_assert_not_reached();
2451    }
2452}
2453
2454static void tcg_target_init(TCGContext *s)
2455{
2456    /*
2457     * Only probe for the platform and capabilities if we haven't already
2458     * determined maximum values at compile time.
2459     */
2460#if !defined(use_idiv_instructions) || !defined(use_neon_instructions)
2461    {
2462        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2463#ifndef use_idiv_instructions
2464        use_idiv_instructions = (hwcap & HWCAP_ARM_IDIVA) != 0;
2465#endif
2466#ifndef use_neon_instructions
2467        use_neon_instructions = (hwcap & HWCAP_ARM_NEON) != 0;
2468#endif
2469    }
2470#endif
2471
2472    if (__ARM_ARCH < 7) {
2473        const char *pl = (const char *)qemu_getauxval(AT_PLATFORM);
2474        if (pl != NULL && pl[0] == 'v' && pl[1] >= '4' && pl[1] <= '9') {
2475            arm_arch = pl[1] - '0';
2476        }
2477    }
2478
2479    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2480
2481    tcg_target_call_clobber_regs = 0;
2482    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R0);
2483    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R1);
2484    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R2);
2485    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R3);
2486    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R12);
2487    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_R14);
2488
2489    if (use_neon_instructions) {
2490        tcg_target_available_regs[TCG_TYPE_V64]  = ALL_VECTOR_REGS;
2491        tcg_target_available_regs[TCG_TYPE_V128] = ALL_VECTOR_REGS;
2492
2493        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q0);
2494        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q1);
2495        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q2);
2496        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q3);
2497        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q8);
2498        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q9);
2499        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q10);
2500        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q11);
2501        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q12);
2502        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q13);
2503        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q14);
2504        tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_Q15);
2505    }
2506
2507    s->reserved_regs = 0;
2508    tcg_regset_set_reg(s->reserved_regs, TCG_REG_CALL_STACK);
2509    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TMP);
2510    tcg_regset_set_reg(s->reserved_regs, TCG_REG_PC);
2511    tcg_regset_set_reg(s->reserved_regs, TCG_VEC_TMP);
2512}
2513
2514static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg arg,
2515                       TCGReg arg1, intptr_t arg2)
2516{
2517    switch (type) {
2518    case TCG_TYPE_I32:
2519        tcg_out_ld32u(s, COND_AL, arg, arg1, arg2);
2520        return;
2521    case TCG_TYPE_V64:
2522        /* regs 1; size 8; align 8 */
2523        tcg_out_vldst(s, INSN_VLD1 | 0x7d0, arg, arg1, arg2);
2524        return;
2525    case TCG_TYPE_V128:
2526        /* regs 2; size 8; align 16 */
2527        tcg_out_vldst(s, INSN_VLD1 | 0xae0, arg, arg1, arg2);
2528        return;
2529    default:
2530        g_assert_not_reached();
2531    }
2532}
2533
2534static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
2535                       TCGReg arg1, intptr_t arg2)
2536{
2537    switch (type) {
2538    case TCG_TYPE_I32:
2539        tcg_out_st32(s, COND_AL, arg, arg1, arg2);
2540        return;
2541    case TCG_TYPE_V64:
2542        /* regs 1; size 8; align 8 */
2543        tcg_out_vldst(s, INSN_VST1 | 0x7d0, arg, arg1, arg2);
2544        return;
2545    case TCG_TYPE_V128:
2546        /* regs 2; size 8; align 16 */
2547        tcg_out_vldst(s, INSN_VST1 | 0xae0, arg, arg1, arg2);
2548        return;
2549    default:
2550        g_assert_not_reached();
2551    }
2552}
2553
2554static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
2555                        TCGReg base, intptr_t ofs)
2556{
2557    return false;
2558}
2559
2560static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
2561{
2562    if (ret == arg) {
2563        return true;
2564    }
2565    switch (type) {
2566    case TCG_TYPE_I32:
2567        if (ret < TCG_REG_Q0 && arg < TCG_REG_Q0) {
2568            tcg_out_mov_reg(s, COND_AL, ret, arg);
2569            return true;
2570        }
2571        return false;
2572
2573    case TCG_TYPE_V64:
2574    case TCG_TYPE_V128:
2575        /* "VMOV D,N" is an alias for "VORR D,N,N". */
2576        tcg_out_vreg3(s, INSN_VORR, type - TCG_TYPE_V64, 0, ret, arg, arg);
2577        return true;
2578
2579    default:
2580        g_assert_not_reached();
2581    }
2582}
2583
2584static void tcg_out_movi(TCGContext *s, TCGType type,
2585                         TCGReg ret, tcg_target_long arg)
2586{
2587    tcg_debug_assert(type == TCG_TYPE_I32);
2588    tcg_debug_assert(ret < TCG_REG_Q0);
2589    tcg_out_movi32(s, COND_AL, ret, arg);
2590}
2591
2592/* Type is always V128, with I64 elements.  */
2593static void tcg_out_dup2_vec(TCGContext *s, TCGReg rd, TCGReg rl, TCGReg rh)
2594{
2595    /* Move high element into place first. */
2596    /* VMOV Dd+1, Ds */
2597    tcg_out_vreg3(s, INSN_VORR | (1 << 12), 0, 0, rd, rh, rh);
2598    /* Move low element into place; tcg_out_mov will check for nop. */
2599    tcg_out_mov(s, TCG_TYPE_V64, rd, rl);
2600}
2601
2602static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece,
2603                            TCGReg rd, TCGReg rs)
2604{
2605    int q = type - TCG_TYPE_V64;
2606
2607    if (vece == MO_64) {
2608        if (type == TCG_TYPE_V128) {
2609            tcg_out_dup2_vec(s, rd, rs, rs);
2610        } else {
2611            tcg_out_mov(s, TCG_TYPE_V64, rd, rs);
2612        }
2613    } else if (rs < TCG_REG_Q0) {
2614        int b = (vece == MO_8);
2615        int e = (vece == MO_16);
2616        tcg_out32(s, INSN_VDUP_G | (b << 22) | (q << 21) | (e << 5) |
2617                  encode_vn(rd) | (rs << 12));
2618    } else {
2619        int imm4 = 1 << vece;
2620        tcg_out32(s, INSN_VDUP_S | (imm4 << 16) | (q << 6) |
2621                  encode_vd(rd) | encode_vm(rs));
2622    }
2623    return true;
2624}
2625
2626static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece,
2627                             TCGReg rd, TCGReg base, intptr_t offset)
2628{
2629    if (vece == MO_64) {
2630        tcg_out_ld(s, TCG_TYPE_V64, rd, base, offset);
2631        if (type == TCG_TYPE_V128) {
2632            tcg_out_dup2_vec(s, rd, rd, rd);
2633        }
2634    } else {
2635        int q = type - TCG_TYPE_V64;
2636        tcg_out_vldst(s, INSN_VLD1R | (vece << 6) | (q << 5),
2637                      rd, base, offset);
2638    }
2639    return true;
2640}
2641
2642static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece,
2643                             TCGReg rd, int64_t v64)
2644{
2645    int q = type - TCG_TYPE_V64;
2646    int cmode, imm8, i;
2647
2648    /* Test all bytes equal first.  */
2649    if (vece == MO_8) {
2650        tcg_out_vmovi(s, rd, q, 0, 0xe, v64);
2651        return;
2652    }
2653
2654    /*
2655     * Test all bytes 0x00 or 0xff second.  This can match cases that
2656     * might otherwise take 2 or 3 insns for MO_16 or MO_32 below.
2657     */
2658    for (i = imm8 = 0; i < 8; i++) {
2659        uint8_t byte = v64 >> (i * 8);
2660        if (byte == 0xff) {
2661            imm8 |= 1 << i;
2662        } else if (byte != 0) {
2663            goto fail_bytes;
2664        }
2665    }
2666    tcg_out_vmovi(s, rd, q, 1, 0xe, imm8);
2667    return;
2668 fail_bytes:
2669
2670    /*
2671     * Tests for various replications.  For each element width, if we
2672     * cannot find an expansion there's no point checking a larger
2673     * width because we already know by replication it cannot match.
2674     */
2675    if (vece == MO_16) {
2676        uint16_t v16 = v64;
2677
2678        if (is_shimm16(v16, &cmode, &imm8)) {
2679            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2680            return;
2681        }
2682        if (is_shimm16(~v16, &cmode, &imm8)) {
2683            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2684            return;
2685        }
2686
2687        /*
2688         * Otherwise, all remaining constants can be loaded in two insns:
2689         * rd = v16 & 0xff, rd |= v16 & 0xff00.
2690         */
2691        tcg_out_vmovi(s, rd, q, 0, 0x8, v16 & 0xff);
2692        tcg_out_vmovi(s, rd, q, 0, 0xb, v16 >> 8);   /* VORRI */
2693        return;
2694    }
2695
2696    if (vece == MO_32) {
2697        uint32_t v32 = v64;
2698
2699        if (is_shimm32(v32, &cmode, &imm8) ||
2700            is_soimm32(v32, &cmode, &imm8)) {
2701            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2702            return;
2703        }
2704        if (is_shimm32(~v32, &cmode, &imm8) ||
2705            is_soimm32(~v32, &cmode, &imm8)) {
2706            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2707            return;
2708        }
2709
2710        /*
2711         * Restrict the set of constants to those we can load with
2712         * two instructions.  Others we load from the pool.
2713         */
2714        i = is_shimm32_pair(v32, &cmode, &imm8);
2715        if (i) {
2716            tcg_out_vmovi(s, rd, q, 0, cmode, imm8);
2717            tcg_out_vmovi(s, rd, q, 0, i | 1, extract32(v32, i * 4, 8));
2718            return;
2719        }
2720        i = is_shimm32_pair(~v32, &cmode, &imm8);
2721        if (i) {
2722            tcg_out_vmovi(s, rd, q, 1, cmode, imm8);
2723            tcg_out_vmovi(s, rd, q, 1, i | 1, extract32(~v32, i * 4, 8));
2724            return;
2725        }
2726    }
2727
2728    /*
2729     * As a last resort, load from the constant pool.
2730     */
2731    if (!q || vece == MO_64) {
2732        new_pool_l2(s, R_ARM_PC11, s->code_ptr, 0, v64, v64 >> 32);
2733        /* VLDR Dd, [pc + offset] */
2734        tcg_out32(s, INSN_VLDR_D | encode_vd(rd) | (0xf << 16));
2735        if (q) {
2736            tcg_out_dup2_vec(s, rd, rd, rd);
2737        }
2738    } else {
2739        new_pool_label(s, (uint32_t)v64, R_ARM_PC8, s->code_ptr, 0);
2740        /* add tmp, pc, offset */
2741        tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_TMP, TCG_REG_PC, 0);
2742        tcg_out_dupm_vec(s, type, MO_32, rd, TCG_REG_TMP, 0);
2743    }
2744}
2745
2746static const ARMInsn vec_cmp_insn[16] = {
2747    [TCG_COND_EQ] = INSN_VCEQ,
2748    [TCG_COND_GT] = INSN_VCGT,
2749    [TCG_COND_GE] = INSN_VCGE,
2750    [TCG_COND_GTU] = INSN_VCGT_U,
2751    [TCG_COND_GEU] = INSN_VCGE_U,
2752};
2753
2754static const ARMInsn vec_cmp0_insn[16] = {
2755    [TCG_COND_EQ] = INSN_VCEQ0,
2756    [TCG_COND_GT] = INSN_VCGT0,
2757    [TCG_COND_GE] = INSN_VCGE0,
2758    [TCG_COND_LT] = INSN_VCLT0,
2759    [TCG_COND_LE] = INSN_VCLE0,
2760};
2761
2762static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
2763                           unsigned vecl, unsigned vece,
2764                           const TCGArg args[TCG_MAX_OP_ARGS],
2765                           const int const_args[TCG_MAX_OP_ARGS])
2766{
2767    TCGType type = vecl + TCG_TYPE_V64;
2768    unsigned q = vecl;
2769    TCGArg a0, a1, a2, a3;
2770    int cmode, imm8;
2771
2772    a0 = args[0];
2773    a1 = args[1];
2774    a2 = args[2];
2775
2776    switch (opc) {
2777    case INDEX_op_ld_vec:
2778        tcg_out_ld(s, type, a0, a1, a2);
2779        return;
2780    case INDEX_op_st_vec:
2781        tcg_out_st(s, type, a0, a1, a2);
2782        return;
2783    case INDEX_op_dupm_vec:
2784        tcg_out_dupm_vec(s, type, vece, a0, a1, a2);
2785        return;
2786    case INDEX_op_dup2_vec:
2787        tcg_out_dup2_vec(s, a0, a1, a2);
2788        return;
2789    case INDEX_op_abs_vec:
2790        tcg_out_vreg2(s, INSN_VABS, q, vece, a0, a1);
2791        return;
2792    case INDEX_op_neg_vec:
2793        tcg_out_vreg2(s, INSN_VNEG, q, vece, a0, a1);
2794        return;
2795    case INDEX_op_not_vec:
2796        tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a1);
2797        return;
2798    case INDEX_op_add_vec:
2799        tcg_out_vreg3(s, INSN_VADD, q, vece, a0, a1, a2);
2800        return;
2801    case INDEX_op_mul_vec:
2802        tcg_out_vreg3(s, INSN_VMUL, q, vece, a0, a1, a2);
2803        return;
2804    case INDEX_op_smax_vec:
2805        tcg_out_vreg3(s, INSN_VMAX, q, vece, a0, a1, a2);
2806        return;
2807    case INDEX_op_smin_vec:
2808        tcg_out_vreg3(s, INSN_VMIN, q, vece, a0, a1, a2);
2809        return;
2810    case INDEX_op_sub_vec:
2811        tcg_out_vreg3(s, INSN_VSUB, q, vece, a0, a1, a2);
2812        return;
2813    case INDEX_op_ssadd_vec:
2814        tcg_out_vreg3(s, INSN_VQADD, q, vece, a0, a1, a2);
2815        return;
2816    case INDEX_op_sssub_vec:
2817        tcg_out_vreg3(s, INSN_VQSUB, q, vece, a0, a1, a2);
2818        return;
2819    case INDEX_op_umax_vec:
2820        tcg_out_vreg3(s, INSN_VMAX_U, q, vece, a0, a1, a2);
2821        return;
2822    case INDEX_op_umin_vec:
2823        tcg_out_vreg3(s, INSN_VMIN_U, q, vece, a0, a1, a2);
2824        return;
2825    case INDEX_op_usadd_vec:
2826        tcg_out_vreg3(s, INSN_VQADD_U, q, vece, a0, a1, a2);
2827        return;
2828    case INDEX_op_ussub_vec:
2829        tcg_out_vreg3(s, INSN_VQSUB_U, q, vece, a0, a1, a2);
2830        return;
2831    case INDEX_op_xor_vec:
2832        tcg_out_vreg3(s, INSN_VEOR, q, 0, a0, a1, a2);
2833        return;
2834    case INDEX_op_arm_sshl_vec:
2835        /*
2836         * Note that Vm is the data and Vn is the shift count,
2837         * therefore the arguments appear reversed.
2838         */
2839        tcg_out_vreg3(s, INSN_VSHL_S, q, vece, a0, a2, a1);
2840        return;
2841    case INDEX_op_arm_ushl_vec:
2842        /* See above. */
2843        tcg_out_vreg3(s, INSN_VSHL_U, q, vece, a0, a2, a1);
2844        return;
2845    case INDEX_op_shli_vec:
2846        tcg_out_vshifti(s, INSN_VSHLI, q, a0, a1, a2 + (8 << vece));
2847        return;
2848    case INDEX_op_shri_vec:
2849        tcg_out_vshifti(s, INSN_VSHRI, q, a0, a1, (16 << vece) - a2);
2850        return;
2851    case INDEX_op_sari_vec:
2852        tcg_out_vshifti(s, INSN_VSARI, q, a0, a1, (16 << vece) - a2);
2853        return;
2854    case INDEX_op_arm_sli_vec:
2855        tcg_out_vshifti(s, INSN_VSLI, q, a0, a2, args[3] + (8 << vece));
2856        return;
2857
2858    case INDEX_op_andc_vec:
2859        if (!const_args[2]) {
2860            tcg_out_vreg3(s, INSN_VBIC, q, 0, a0, a1, a2);
2861            return;
2862        }
2863        a2 = ~a2;
2864        /* fall through */
2865    case INDEX_op_and_vec:
2866        if (const_args[2]) {
2867            is_shimm1632(~a2, &cmode, &imm8);
2868            if (a0 == a1) {
2869                tcg_out_vmovi(s, a0, q, 1, cmode | 1, imm8); /* VBICI */
2870                return;
2871            }
2872            tcg_out_vmovi(s, a0, q, 1, cmode, imm8); /* VMVNI */
2873            a2 = a0;
2874        }
2875        tcg_out_vreg3(s, INSN_VAND, q, 0, a0, a1, a2);
2876        return;
2877
2878    case INDEX_op_orc_vec:
2879        if (!const_args[2]) {
2880            tcg_out_vreg3(s, INSN_VORN, q, 0, a0, a1, a2);
2881            return;
2882        }
2883        a2 = ~a2;
2884        /* fall through */
2885    case INDEX_op_or_vec:
2886        if (const_args[2]) {
2887            is_shimm1632(a2, &cmode, &imm8);
2888            if (a0 == a1) {
2889                tcg_out_vmovi(s, a0, q, 0, cmode | 1, imm8); /* VORRI */
2890                return;
2891            }
2892            tcg_out_vmovi(s, a0, q, 0, cmode, imm8); /* VMOVI */
2893            a2 = a0;
2894        }
2895        tcg_out_vreg3(s, INSN_VORR, q, 0, a0, a1, a2);
2896        return;
2897
2898    case INDEX_op_cmp_vec:
2899        {
2900            TCGCond cond = args[3];
2901
2902            if (cond == TCG_COND_NE) {
2903                if (const_args[2]) {
2904                    tcg_out_vreg3(s, INSN_VTST, q, vece, a0, a1, a1);
2905                } else {
2906                    tcg_out_vreg3(s, INSN_VCEQ, q, vece, a0, a1, a2);
2907                    tcg_out_vreg2(s, INSN_VMVN, q, 0, a0, a0);
2908                }
2909            } else {
2910                ARMInsn insn;
2911
2912                if (const_args[2]) {
2913                    insn = vec_cmp0_insn[cond];
2914                    if (insn) {
2915                        tcg_out_vreg2(s, insn, q, vece, a0, a1);
2916                        return;
2917                    }
2918                    tcg_out_dupi_vec(s, type, MO_8, TCG_VEC_TMP, 0);
2919                    a2 = TCG_VEC_TMP;
2920                }
2921                insn = vec_cmp_insn[cond];
2922                if (insn == 0) {
2923                    TCGArg t;
2924                    t = a1, a1 = a2, a2 = t;
2925                    cond = tcg_swap_cond(cond);
2926                    insn = vec_cmp_insn[cond];
2927                    tcg_debug_assert(insn != 0);
2928                }
2929                tcg_out_vreg3(s, insn, q, vece, a0, a1, a2);
2930            }
2931        }
2932        return;
2933
2934    case INDEX_op_bitsel_vec:
2935        a3 = args[3];
2936        if (a0 == a3) {
2937            tcg_out_vreg3(s, INSN_VBIT, q, 0, a0, a2, a1);
2938        } else if (a0 == a2) {
2939            tcg_out_vreg3(s, INSN_VBIF, q, 0, a0, a3, a1);
2940        } else {
2941            tcg_out_mov(s, type, a0, a1);
2942            tcg_out_vreg3(s, INSN_VBSL, q, 0, a0, a2, a3);
2943        }
2944        return;
2945
2946    case INDEX_op_mov_vec:  /* Always emitted via tcg_out_mov.  */
2947    case INDEX_op_dup_vec:  /* Always emitted via tcg_out_dup_vec.  */
2948    default:
2949        g_assert_not_reached();
2950    }
2951}
2952
2953int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
2954{
2955    switch (opc) {
2956    case INDEX_op_add_vec:
2957    case INDEX_op_sub_vec:
2958    case INDEX_op_and_vec:
2959    case INDEX_op_andc_vec:
2960    case INDEX_op_or_vec:
2961    case INDEX_op_orc_vec:
2962    case INDEX_op_xor_vec:
2963    case INDEX_op_not_vec:
2964    case INDEX_op_shli_vec:
2965    case INDEX_op_shri_vec:
2966    case INDEX_op_sari_vec:
2967    case INDEX_op_ssadd_vec:
2968    case INDEX_op_sssub_vec:
2969    case INDEX_op_usadd_vec:
2970    case INDEX_op_ussub_vec:
2971    case INDEX_op_bitsel_vec:
2972        return 1;
2973    case INDEX_op_abs_vec:
2974    case INDEX_op_cmp_vec:
2975    case INDEX_op_mul_vec:
2976    case INDEX_op_neg_vec:
2977    case INDEX_op_smax_vec:
2978    case INDEX_op_smin_vec:
2979    case INDEX_op_umax_vec:
2980    case INDEX_op_umin_vec:
2981        return vece < MO_64;
2982    case INDEX_op_shlv_vec:
2983    case INDEX_op_shrv_vec:
2984    case INDEX_op_sarv_vec:
2985    case INDEX_op_rotli_vec:
2986    case INDEX_op_rotlv_vec:
2987    case INDEX_op_rotrv_vec:
2988        return -1;
2989    default:
2990        return 0;
2991    }
2992}
2993
2994void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
2995                       TCGArg a0, ...)
2996{
2997    va_list va;
2998    TCGv_vec v0, v1, v2, t1, t2, c1;
2999    TCGArg a2;
3000
3001    va_start(va, a0);
3002    v0 = temp_tcgv_vec(arg_temp(a0));
3003    v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
3004    a2 = va_arg(va, TCGArg);
3005    va_end(va);
3006
3007    switch (opc) {
3008    case INDEX_op_shlv_vec:
3009        /*
3010         * Merely propagate shlv_vec to arm_ushl_vec.
3011         * In this way we don't set TCG_TARGET_HAS_shv_vec
3012         * because everything is done via expansion.
3013         */
3014        v2 = temp_tcgv_vec(arg_temp(a2));
3015        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3016                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3017        break;
3018
3019    case INDEX_op_shrv_vec:
3020    case INDEX_op_sarv_vec:
3021        /* Right shifts are negative left shifts for NEON.  */
3022        v2 = temp_tcgv_vec(arg_temp(a2));
3023        t1 = tcg_temp_new_vec(type);
3024        tcg_gen_neg_vec(vece, t1, v2);
3025        if (opc == INDEX_op_shrv_vec) {
3026            opc = INDEX_op_arm_ushl_vec;
3027        } else {
3028            opc = INDEX_op_arm_sshl_vec;
3029        }
3030        vec_gen_3(opc, type, vece, tcgv_vec_arg(v0),
3031                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3032        tcg_temp_free_vec(t1);
3033        break;
3034
3035    case INDEX_op_rotli_vec:
3036        t1 = tcg_temp_new_vec(type);
3037        tcg_gen_shri_vec(vece, t1, v1, -a2 & ((8 << vece) - 1));
3038        vec_gen_4(INDEX_op_arm_sli_vec, type, vece,
3039                  tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(v1), a2);
3040        tcg_temp_free_vec(t1);
3041        break;
3042
3043    case INDEX_op_rotlv_vec:
3044        v2 = temp_tcgv_vec(arg_temp(a2));
3045        t1 = tcg_temp_new_vec(type);
3046        c1 = tcg_constant_vec(type, vece, 8 << vece);
3047        tcg_gen_sub_vec(vece, t1, v2, c1);
3048        /* Right shifts are negative left shifts for NEON.  */
3049        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3050                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3051        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(v0),
3052                  tcgv_vec_arg(v1), tcgv_vec_arg(v2));
3053        tcg_gen_or_vec(vece, v0, v0, t1);
3054        tcg_temp_free_vec(t1);
3055        break;
3056
3057    case INDEX_op_rotrv_vec:
3058        v2 = temp_tcgv_vec(arg_temp(a2));
3059        t1 = tcg_temp_new_vec(type);
3060        t2 = tcg_temp_new_vec(type);
3061        c1 = tcg_constant_vec(type, vece, 8 << vece);
3062        tcg_gen_neg_vec(vece, t1, v2);
3063        tcg_gen_sub_vec(vece, t2, c1, v2);
3064        /* Right shifts are negative left shifts for NEON.  */
3065        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t1),
3066                  tcgv_vec_arg(v1), tcgv_vec_arg(t1));
3067        vec_gen_3(INDEX_op_arm_ushl_vec, type, vece, tcgv_vec_arg(t2),
3068                  tcgv_vec_arg(v1), tcgv_vec_arg(t2));
3069        tcg_gen_or_vec(vece, v0, t1, t2);
3070        tcg_temp_free_vec(t1);
3071        tcg_temp_free_vec(t2);
3072        break;
3073
3074    default:
3075        g_assert_not_reached();
3076    }
3077}
3078
3079static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
3080{
3081    int i;
3082    for (i = 0; i < count; ++i) {
3083        p[i] = INSN_NOP;
3084    }
3085}
3086
3087/* Compute frame size via macros, to share between tcg_target_qemu_prologue
3088   and tcg_register_jit.  */
3089
3090#define PUSH_SIZE  ((11 - 4 + 1 + 1) * sizeof(tcg_target_long))
3091
3092#define FRAME_SIZE \
3093    ((PUSH_SIZE \
3094      + TCG_STATIC_CALL_ARGS_SIZE \
3095      + CPU_TEMP_BUF_NLONGS * sizeof(long) \
3096      + TCG_TARGET_STACK_ALIGN - 1) \
3097     & -TCG_TARGET_STACK_ALIGN)
3098
3099#define STACK_ADDEND  (FRAME_SIZE - PUSH_SIZE)
3100
3101static void tcg_target_qemu_prologue(TCGContext *s)
3102{
3103    /* Calling convention requires us to save r4-r11 and lr.  */
3104    /* stmdb sp!, { r4 - r11, lr } */
3105    tcg_out_ldstm(s, COND_AL, INSN_STMDB, TCG_REG_CALL_STACK,
3106                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3107                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3108                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_R14));
3109
3110    /* Reserve callee argument and tcg temp space.  */
3111    tcg_out_dat_rI(s, COND_AL, ARITH_SUB, TCG_REG_CALL_STACK,
3112                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3113    tcg_set_frame(s, TCG_REG_CALL_STACK, TCG_STATIC_CALL_ARGS_SIZE,
3114                  CPU_TEMP_BUF_NLONGS * sizeof(long));
3115
3116    tcg_out_mov(s, TCG_TYPE_PTR, TCG_AREG0, tcg_target_call_iarg_regs[0]);
3117
3118    tcg_out_b_reg(s, COND_AL, tcg_target_call_iarg_regs[1]);
3119
3120    /*
3121     * Return path for goto_ptr. Set return value to 0, a-la exit_tb,
3122     * and fall through to the rest of the epilogue.
3123     */
3124    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
3125    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_R0, 0);
3126    tcg_out_epilogue(s);
3127}
3128
3129static void tcg_out_epilogue(TCGContext *s)
3130{
3131    /* Release local stack frame.  */
3132    tcg_out_dat_rI(s, COND_AL, ARITH_ADD, TCG_REG_CALL_STACK,
3133                   TCG_REG_CALL_STACK, STACK_ADDEND, 1);
3134
3135    /* ldmia sp!, { r4 - r11, pc } */
3136    tcg_out_ldstm(s, COND_AL, INSN_LDMIA, TCG_REG_CALL_STACK,
3137                  (1 << TCG_REG_R4) | (1 << TCG_REG_R5) | (1 << TCG_REG_R6) |
3138                  (1 << TCG_REG_R7) | (1 << TCG_REG_R8) | (1 << TCG_REG_R9) |
3139                  (1 << TCG_REG_R10) | (1 << TCG_REG_R11) | (1 << TCG_REG_PC));
3140}
3141
3142typedef struct {
3143    DebugFrameHeader h;
3144    uint8_t fde_def_cfa[4];
3145    uint8_t fde_reg_ofs[18];
3146} DebugFrame;
3147
3148#define ELF_HOST_MACHINE EM_ARM
3149
3150/* We're expecting a 2 byte uleb128 encoded value.  */
3151QEMU_BUILD_BUG_ON(FRAME_SIZE >= (1 << 14));
3152
3153static const DebugFrame debug_frame = {
3154    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
3155    .h.cie.id = -1,
3156    .h.cie.version = 1,
3157    .h.cie.code_align = 1,
3158    .h.cie.data_align = 0x7c,             /* sleb128 -4 */
3159    .h.cie.return_column = 14,
3160
3161    /* Total FDE size does not include the "len" member.  */
3162    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
3163
3164    .fde_def_cfa = {
3165        12, 13,                         /* DW_CFA_def_cfa sp, ... */
3166        (FRAME_SIZE & 0x7f) | 0x80,     /* ... uleb128 FRAME_SIZE */
3167        (FRAME_SIZE >> 7)
3168    },
3169    .fde_reg_ofs = {
3170        /* The following must match the stmdb in the prologue.  */
3171        0x8e, 1,                        /* DW_CFA_offset, lr, -4 */
3172        0x8b, 2,                        /* DW_CFA_offset, r11, -8 */
3173        0x8a, 3,                        /* DW_CFA_offset, r10, -12 */
3174        0x89, 4,                        /* DW_CFA_offset, r9, -16 */
3175        0x88, 5,                        /* DW_CFA_offset, r8, -20 */
3176        0x87, 6,                        /* DW_CFA_offset, r7, -24 */
3177        0x86, 7,                        /* DW_CFA_offset, r6, -28 */
3178        0x85, 8,                        /* DW_CFA_offset, r5, -32 */
3179        0x84, 9,                        /* DW_CFA_offset, r4, -36 */
3180    }
3181};
3182
3183void tcg_register_jit(const void *buf, size_t buf_size)
3184{
3185    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
3186}
3187