xref: /openbmc/qemu/tcg/sparc64/tcg-target.c.inc (revision b6d69fcefbd45ca33b896abfbc8e27e0f713bdf0)
1/*
2 * Tiny Code Generator for QEMU
3 *
4 * Copyright (c) 2008 Fabrice Bellard
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/* We only support generating code for 64-bit mode.  */
26#ifndef __arch64__
27#error "unsupported code generation mode"
28#endif
29
30/* Used for function call generation. */
31#define TCG_REG_CALL_STACK              TCG_REG_O6
32#define TCG_TARGET_STACK_BIAS           2047
33#define TCG_TARGET_STACK_ALIGN          16
34#define TCG_TARGET_CALL_STACK_OFFSET    (128 + 6 * 8 + TCG_TARGET_STACK_BIAS)
35#define TCG_TARGET_CALL_ARG_I32         TCG_CALL_ARG_EXTEND
36#define TCG_TARGET_CALL_ARG_I64         TCG_CALL_ARG_NORMAL
37#define TCG_TARGET_CALL_ARG_I128        TCG_CALL_ARG_NORMAL
38#define TCG_TARGET_CALL_RET_I128        TCG_CALL_RET_NORMAL
39
40#ifdef CONFIG_DEBUG_TCG
41static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
42    "%g0",
43    "%g1",
44    "%g2",
45    "%g3",
46    "%g4",
47    "%g5",
48    "%g6",
49    "%g7",
50    "%o0",
51    "%o1",
52    "%o2",
53    "%o3",
54    "%o4",
55    "%o5",
56    "%o6",
57    "%o7",
58    "%l0",
59    "%l1",
60    "%l2",
61    "%l3",
62    "%l4",
63    "%l5",
64    "%l6",
65    "%l7",
66    "%i0",
67    "%i1",
68    "%i2",
69    "%i3",
70    "%i4",
71    "%i5",
72    "%i6",
73    "%i7",
74};
75#endif
76
77#define TCG_CT_CONST_S11  0x100
78#define TCG_CT_CONST_S13  0x200
79
80#define ALL_GENERAL_REGS  MAKE_64BIT_MASK(0, 32)
81
82/* Define some temporary registers.  T3 is used for constant generation.  */
83#define TCG_REG_T1  TCG_REG_G1
84#define TCG_REG_T2  TCG_REG_G2
85#define TCG_REG_T3  TCG_REG_O7
86
87#ifndef CONFIG_SOFTMMU
88# define TCG_GUEST_BASE_REG TCG_REG_I5
89#endif
90
91#define TCG_REG_TB  TCG_REG_I1
92
93static const int tcg_target_reg_alloc_order[] = {
94    TCG_REG_L0,
95    TCG_REG_L1,
96    TCG_REG_L2,
97    TCG_REG_L3,
98    TCG_REG_L4,
99    TCG_REG_L5,
100    TCG_REG_L6,
101    TCG_REG_L7,
102
103    TCG_REG_I0,
104    TCG_REG_I1,
105    TCG_REG_I2,
106    TCG_REG_I3,
107    TCG_REG_I4,
108    TCG_REG_I5,
109
110    TCG_REG_G3,
111    TCG_REG_G4,
112    TCG_REG_G5,
113
114    TCG_REG_O0,
115    TCG_REG_O1,
116    TCG_REG_O2,
117    TCG_REG_O3,
118    TCG_REG_O4,
119    TCG_REG_O5,
120};
121
122static const int tcg_target_call_iarg_regs[6] = {
123    TCG_REG_O0,
124    TCG_REG_O1,
125    TCG_REG_O2,
126    TCG_REG_O3,
127    TCG_REG_O4,
128    TCG_REG_O5,
129};
130
131static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
132{
133    tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
134    tcg_debug_assert(slot >= 0 && slot <= 3);
135    return TCG_REG_O0 + slot;
136}
137
138#define INSN_OP(x)  ((x) << 30)
139#define INSN_OP2(x) ((x) << 22)
140#define INSN_OP3(x) ((x) << 19)
141#define INSN_OPF(x) ((x) << 5)
142#define INSN_RD(x)  ((x) << 25)
143#define INSN_RS1(x) ((x) << 14)
144#define INSN_RS2(x) (x)
145#define INSN_ASI(x) ((x) << 5)
146
147#define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff))
148#define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff))
149#define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff))
150#define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20))
151#define INSN_OFF19(x) (((x) >> 2) & 0x07ffff)
152#define INSN_COND(x) ((x) << 25)
153
154#define COND_N     0x0
155#define COND_E     0x1
156#define COND_LE    0x2
157#define COND_L     0x3
158#define COND_LEU   0x4
159#define COND_CS    0x5
160#define COND_NEG   0x6
161#define COND_VS    0x7
162#define COND_A     0x8
163#define COND_NE    0x9
164#define COND_G     0xa
165#define COND_GE    0xb
166#define COND_GU    0xc
167#define COND_CC    0xd
168#define COND_POS   0xe
169#define COND_VC    0xf
170#define BA         (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2))
171
172#define RCOND_Z    1
173#define RCOND_LEZ  2
174#define RCOND_LZ   3
175#define RCOND_NZ   5
176#define RCOND_GZ   6
177#define RCOND_GEZ  7
178
179#define MOVCC_ICC  (1 << 18)
180#define MOVCC_XCC  (1 << 18 | 1 << 12)
181
182#define BPCC_ICC   0
183#define BPCC_XCC   (2 << 20)
184#define BPCC_PT    (1 << 19)
185#define BPCC_PN    0
186#define BPCC_A     (1 << 29)
187
188#define BPR_PT     BPCC_PT
189
190#define ARITH_ADD  (INSN_OP(2) | INSN_OP3(0x00))
191#define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10))
192#define ARITH_AND  (INSN_OP(2) | INSN_OP3(0x01))
193#define ARITH_ANDCC (INSN_OP(2) | INSN_OP3(0x11))
194#define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05))
195#define ARITH_OR   (INSN_OP(2) | INSN_OP3(0x02))
196#define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12))
197#define ARITH_ORN  (INSN_OP(2) | INSN_OP3(0x06))
198#define ARITH_XOR  (INSN_OP(2) | INSN_OP3(0x03))
199#define ARITH_SUB  (INSN_OP(2) | INSN_OP3(0x04))
200#define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14))
201#define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08))
202#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
203#define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
204#define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
205#define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
206#define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f))
207#define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09))
208#define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d))
209#define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d))
210#define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
211#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
212
213#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
214#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
215
216#define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
217#define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
218#define SHIFT_SRA  (INSN_OP(2) | INSN_OP3(0x27))
219
220#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
221#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
222#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
223
224#define RDY        (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
225#define WRY        (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
226#define JMPL       (INSN_OP(2) | INSN_OP3(0x38))
227#define RETURN     (INSN_OP(2) | INSN_OP3(0x39))
228#define SAVE       (INSN_OP(2) | INSN_OP3(0x3c))
229#define RESTORE    (INSN_OP(2) | INSN_OP3(0x3d))
230#define SETHI      (INSN_OP(0) | INSN_OP2(0x4))
231#define CALL       INSN_OP(1)
232#define LDUB       (INSN_OP(3) | INSN_OP3(0x01))
233#define LDSB       (INSN_OP(3) | INSN_OP3(0x09))
234#define LDUH       (INSN_OP(3) | INSN_OP3(0x02))
235#define LDSH       (INSN_OP(3) | INSN_OP3(0x0a))
236#define LDUW       (INSN_OP(3) | INSN_OP3(0x00))
237#define LDSW       (INSN_OP(3) | INSN_OP3(0x08))
238#define LDX        (INSN_OP(3) | INSN_OP3(0x0b))
239#define STB        (INSN_OP(3) | INSN_OP3(0x05))
240#define STH        (INSN_OP(3) | INSN_OP3(0x06))
241#define STW        (INSN_OP(3) | INSN_OP3(0x04))
242#define STX        (INSN_OP(3) | INSN_OP3(0x0e))
243#define LDUBA      (INSN_OP(3) | INSN_OP3(0x11))
244#define LDSBA      (INSN_OP(3) | INSN_OP3(0x19))
245#define LDUHA      (INSN_OP(3) | INSN_OP3(0x12))
246#define LDSHA      (INSN_OP(3) | INSN_OP3(0x1a))
247#define LDUWA      (INSN_OP(3) | INSN_OP3(0x10))
248#define LDSWA      (INSN_OP(3) | INSN_OP3(0x18))
249#define LDXA       (INSN_OP(3) | INSN_OP3(0x1b))
250#define STBA       (INSN_OP(3) | INSN_OP3(0x15))
251#define STHA       (INSN_OP(3) | INSN_OP3(0x16))
252#define STWA       (INSN_OP(3) | INSN_OP3(0x14))
253#define STXA       (INSN_OP(3) | INSN_OP3(0x1e))
254
255#define MEMBAR     (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
256
257#define NOP        (SETHI | INSN_RD(TCG_REG_G0) | 0)
258
259#ifndef ASI_PRIMARY_LITTLE
260#define ASI_PRIMARY_LITTLE 0x88
261#endif
262
263#define LDUH_LE    (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
264#define LDSH_LE    (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
265#define LDUW_LE    (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
266#define LDSW_LE    (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
267#define LDX_LE     (LDXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
268
269#define STH_LE     (STHA  | INSN_ASI(ASI_PRIMARY_LITTLE))
270#define STW_LE     (STWA  | INSN_ASI(ASI_PRIMARY_LITTLE))
271#define STX_LE     (STXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
272
273#ifndef use_vis3_instructions
274bool use_vis3_instructions;
275#endif
276
277static bool check_fit_i64(int64_t val, unsigned int bits)
278{
279    return val == sextract64(val, 0, bits);
280}
281
282static bool check_fit_i32(int32_t val, unsigned int bits)
283{
284    return val == sextract32(val, 0, bits);
285}
286
287#define check_fit_tl    check_fit_i64
288#define check_fit_ptr   check_fit_i64
289
290static bool patch_reloc(tcg_insn_unit *src_rw, int type,
291                        intptr_t value, intptr_t addend)
292{
293    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
294    uint32_t insn = *src_rw;
295    intptr_t pcrel;
296
297    value += addend;
298    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx);
299
300    switch (type) {
301    case R_SPARC_WDISP16:
302        if (!check_fit_ptr(pcrel >> 2, 16)) {
303            return false;
304        }
305        insn &= ~INSN_OFF16(-1);
306        insn |= INSN_OFF16(pcrel);
307        break;
308    case R_SPARC_WDISP19:
309        if (!check_fit_ptr(pcrel >> 2, 19)) {
310            return false;
311        }
312        insn &= ~INSN_OFF19(-1);
313        insn |= INSN_OFF19(pcrel);
314        break;
315    case R_SPARC_13:
316        if (!check_fit_ptr(value, 13)) {
317            return false;
318        }
319        insn &= ~INSN_IMM13(-1);
320        insn |= INSN_IMM13(value);
321        break;
322    default:
323        g_assert_not_reached();
324    }
325
326    *src_rw = insn;
327    return true;
328}
329
330/* test if a constant matches the constraint */
331static bool tcg_target_const_match(int64_t val, int ct,
332                                   TCGType type, TCGCond cond, int vece)
333{
334    if (ct & TCG_CT_CONST) {
335        return 1;
336    }
337
338    if (type == TCG_TYPE_I32) {
339        val = (int32_t)val;
340    }
341
342    if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
343        return 1;
344    } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
345        return 1;
346    } else {
347        return 0;
348    }
349}
350
351static void tcg_out_nop(TCGContext *s)
352{
353    tcg_out32(s, NOP);
354}
355
356static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
357                          TCGReg rs2, int op)
358{
359    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
360}
361
362static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
363                           int32_t offset, int op)
364{
365    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
366}
367
368static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
369			   int32_t val2, int val2const, int op)
370{
371    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
372              | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
373}
374
375static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
376{
377    if (ret != arg) {
378        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
379    }
380    return true;
381}
382
383static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg)
384{
385    if (ret != arg) {
386        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
387    } else {
388        tcg_out_nop(s);
389    }
390}
391
392static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
393{
394    tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
395}
396
397/* A 13-bit constant sign-extended to 64 bits.  */
398static void tcg_out_movi_s13(TCGContext *s, TCGReg ret, int32_t arg)
399{
400    tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
401}
402
403/* A 32-bit constant sign-extended to 64 bits.  */
404static void tcg_out_movi_s32(TCGContext *s, TCGReg ret, int32_t arg)
405{
406    tcg_out_sethi(s, ret, ~arg);
407    tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
408}
409
410/* A 32-bit constant zero-extended to 64 bits.  */
411static void tcg_out_movi_u32(TCGContext *s, TCGReg ret, uint32_t arg)
412{
413    tcg_out_sethi(s, ret, arg);
414    if (arg & 0x3ff) {
415        tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
416    }
417}
418
419static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
420                             tcg_target_long arg, bool in_prologue,
421                             TCGReg scratch)
422{
423    tcg_target_long hi, lo = (int32_t)arg;
424    tcg_target_long test, lsb;
425
426    /* A 13-bit constant sign-extended to 64-bits.  */
427    if (check_fit_tl(arg, 13)) {
428        tcg_out_movi_s13(s, ret, arg);
429        return;
430    }
431
432    /* A 32-bit constant, or 32-bit zero-extended to 64-bits.  */
433    if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
434        tcg_out_movi_u32(s, ret, arg);
435        return;
436    }
437
438    /* A 13-bit constant relative to the TB.  */
439    if (!in_prologue) {
440        test = tcg_tbrel_diff(s, (void *)arg);
441        if (check_fit_ptr(test, 13)) {
442            tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
443            return;
444        }
445    }
446
447    /* A 32-bit constant sign-extended to 64-bits.  */
448    if (arg == lo) {
449        tcg_out_movi_s32(s, ret, arg);
450        return;
451    }
452
453    /* A 32-bit constant, shifted.  */
454    lsb = ctz64(arg);
455    test = (tcg_target_long)arg >> lsb;
456    if (lsb > 10 && test == extract64(test, 0, 21)) {
457        tcg_out_sethi(s, ret, test << 10);
458        tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
459        return;
460    } else if (test == (uint32_t)test || test == (int32_t)test) {
461        tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch);
462        tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
463        return;
464    }
465
466    /* Use the constant pool, if possible. */
467    if (!in_prologue) {
468        new_pool_label(s, arg, R_SPARC_13, s->code_ptr,
469                       tcg_tbrel_diff(s, NULL));
470        tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB));
471        return;
472    }
473
474    /* A 64-bit constant decomposed into 2 32-bit pieces.  */
475    if (check_fit_i32(lo, 13)) {
476        hi = (arg - lo) >> 32;
477        tcg_out_movi_u32(s, ret, hi);
478        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
479        tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
480    } else {
481        hi = arg >> 32;
482        tcg_out_movi_u32(s, ret, hi);
483        tcg_out_movi_u32(s, scratch, lo);
484        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
485        tcg_out_arith(s, ret, ret, scratch, ARITH_OR);
486    }
487}
488
489static void tcg_out_movi(TCGContext *s, TCGType type,
490                         TCGReg ret, tcg_target_long arg)
491{
492    tcg_debug_assert(ret != TCG_REG_T3);
493    tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T3);
494}
495
496static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
497{
498    g_assert_not_reached();
499}
500
501static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
502{
503    g_assert_not_reached();
504}
505
506static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs)
507{
508    tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND);
509}
510
511static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs)
512{
513    tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL);
514    tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL);
515}
516
517static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs)
518{
519    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA);
520}
521
522static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs)
523{
524    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL);
525}
526
527static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
528{
529    tcg_out_ext32s(s, rd, rs);
530}
531
532static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
533{
534    tcg_out_ext32u(s, rd, rs);
535}
536
537static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs)
538{
539    tcg_out_ext32u(s, rd, rs);
540}
541
542static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
543{
544    return false;
545}
546
547static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
548                             tcg_target_long imm)
549{
550    /* This function is only used for passing structs by reference. */
551    g_assert_not_reached();
552}
553
554static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
555                            TCGReg a2, int op)
556{
557    tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
558}
559
560static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
561                         intptr_t offset, int op)
562{
563    if (check_fit_ptr(offset, 13)) {
564        tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
565                  INSN_IMM13(offset));
566    } else {
567        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
568        tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
569    }
570}
571
572static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
573                       TCGReg arg1, intptr_t arg2)
574{
575    tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
576}
577
578static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
579                       TCGReg arg1, intptr_t arg2)
580{
581    tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
582}
583
584static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
585                        TCGReg base, intptr_t ofs)
586{
587    if (val == 0) {
588        tcg_out_st(s, type, TCG_REG_G0, base, ofs);
589        return true;
590    }
591    return false;
592}
593
594static void tcg_out_sety(TCGContext *s, TCGReg rs)
595{
596    tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
597}
598
599static const uint8_t tcg_cond_to_bcond[16] = {
600    [TCG_COND_EQ] = COND_E,
601    [TCG_COND_NE] = COND_NE,
602    [TCG_COND_TSTEQ] = COND_E,
603    [TCG_COND_TSTNE] = COND_NE,
604    [TCG_COND_LT] = COND_L,
605    [TCG_COND_GE] = COND_GE,
606    [TCG_COND_LE] = COND_LE,
607    [TCG_COND_GT] = COND_G,
608    [TCG_COND_LTU] = COND_CS,
609    [TCG_COND_GEU] = COND_CC,
610    [TCG_COND_LEU] = COND_LEU,
611    [TCG_COND_GTU] = COND_GU,
612};
613
614static const uint8_t tcg_cond_to_rcond[16] = {
615    [TCG_COND_EQ] = RCOND_Z,
616    [TCG_COND_NE] = RCOND_NZ,
617    [TCG_COND_LT] = RCOND_LZ,
618    [TCG_COND_GT] = RCOND_GZ,
619    [TCG_COND_LE] = RCOND_LEZ,
620    [TCG_COND_GE] = RCOND_GEZ
621};
622
623static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
624{
625    tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
626}
627
628static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
629{
630    int off19 = 0;
631
632    if (l->has_value) {
633        off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
634    } else {
635        tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
636    }
637    tcg_out_bpcc0(s, scond, flags, off19);
638}
639
640static void tcg_out_cmp(TCGContext *s, TCGCond cond,
641                        TCGReg c1, int32_t c2, int c2const)
642{
643    tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const,
644                   is_tst_cond(cond) ? ARITH_ANDCC : ARITH_SUBCC);
645}
646
647static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
648                               int32_t arg2, int const_arg2, TCGLabel *l)
649{
650    tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
651    tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
652    tcg_out_nop(s);
653}
654
655static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret,
656                          int32_t v1, int v1const)
657{
658    tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret)
659              | INSN_RS1(tcg_cond_to_bcond[cond])
660              | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
661}
662
663static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
664                                TCGReg c1, int32_t c2, int c2const,
665                                int32_t v1, int v1const)
666{
667    tcg_out_cmp(s, cond, c1, c2, c2const);
668    tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const);
669}
670
671static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
672                               int32_t arg2, int const_arg2, TCGLabel *l)
673{
674    /* For 64-bit signed comparisons vs zero, we can avoid the compare.  */
675    int rcond = tcg_cond_to_rcond[cond];
676    if (arg2 == 0 && rcond) {
677        int off16 = 0;
678
679        if (l->has_value) {
680            off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
681        } else {
682            tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
683        }
684        tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
685                  | INSN_COND(rcond) | off16);
686    } else {
687        tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
688        tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
689    }
690    tcg_out_nop(s);
691}
692
693static void tcg_out_movr(TCGContext *s, int rcond, TCGReg ret, TCGReg c1,
694                         int32_t v1, int v1const)
695{
696    tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | (rcond << 10)
697              | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
698}
699
700static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
701                                TCGReg c1, int32_t c2, int c2const,
702                                int32_t v1, int v1const)
703{
704    /* For 64-bit signed comparisons vs zero, we can avoid the compare.
705       Note that the immediate range is one bit smaller, so we must check
706       for that as well.  */
707    int rcond = tcg_cond_to_rcond[cond];
708    if (c2 == 0 && rcond && (!v1const || check_fit_i32(v1, 10))) {
709        tcg_out_movr(s, rcond, ret, c1, v1, v1const);
710    } else {
711        tcg_out_cmp(s, cond, c1, c2, c2const);
712        tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const);
713    }
714}
715
716static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
717                                TCGReg c1, int32_t c2, bool c2const, bool neg)
718{
719    /* For 32-bit comparisons, we can play games with ADDC/SUBC.  */
720    switch (cond) {
721    case TCG_COND_LTU:
722    case TCG_COND_GEU:
723        /* The result of the comparison is in the carry bit.  */
724        break;
725
726    case TCG_COND_EQ:
727    case TCG_COND_NE:
728        /* For equality, we can transform to inequality vs zero.  */
729        if (c2 != 0) {
730            tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
731            c2 = TCG_REG_T1;
732        } else {
733            c2 = c1;
734        }
735        c1 = TCG_REG_G0, c2const = 0;
736        cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
737	break;
738
739    case TCG_COND_TSTEQ:
740    case TCG_COND_TSTNE:
741        /* Transform to inequality vs zero.  */
742        tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_AND);
743        c1 = TCG_REG_G0;
744        c2 = TCG_REG_T1, c2const = 0;
745        cond = (cond == TCG_COND_TSTEQ ? TCG_COND_GEU : TCG_COND_LTU);
746	break;
747
748    case TCG_COND_GTU:
749    case TCG_COND_LEU:
750        /* If we don't need to load a constant into a register, we can
751           swap the operands on GTU/LEU.  There's no benefit to loading
752           the constant into a temporary register.  */
753        if (!c2const || c2 == 0) {
754            TCGReg t = c1;
755            c1 = c2;
756            c2 = t;
757            c2const = 0;
758            cond = tcg_swap_cond(cond);
759            break;
760        }
761        /* FALLTHRU */
762
763    default:
764        tcg_out_cmp(s, cond, c1, c2, c2const);
765        tcg_out_movi_s13(s, ret, 0);
766        tcg_out_movcc(s, cond, MOVCC_ICC, ret, neg ? -1 : 1, 1);
767        return;
768    }
769
770    tcg_out_cmp(s, cond, c1, c2, c2const);
771    if (cond == TCG_COND_LTU) {
772        if (neg) {
773            /* 0 - 0 - C = -C = (C ? -1 : 0) */
774            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_SUBC);
775        } else {
776            /* 0 + 0 + C =  C = (C ? 1 : 0) */
777            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
778        }
779    } else {
780        if (neg) {
781            /* 0 + -1 + C = C - 1 = (C ? 0 : -1) */
782            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_ADDC);
783        } else {
784            /* 0 - -1 - C = 1 - C = (C ? 0 : 1) */
785            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
786        }
787    }
788}
789
790static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
791                                TCGReg c1, int32_t c2, bool c2const, bool neg)
792{
793    int rcond;
794
795    if (use_vis3_instructions && !neg) {
796        switch (cond) {
797        case TCG_COND_NE:
798            if (c2 != 0) {
799                break;
800            }
801            c2 = c1, c2const = 0, c1 = TCG_REG_G0;
802            /* FALLTHRU */
803        case TCG_COND_LTU:
804            tcg_out_cmp(s, cond, c1, c2, c2const);
805            tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
806            return;
807        default:
808            break;
809        }
810    }
811
812    /* For 64-bit signed comparisons vs zero, we can avoid the compare
813       if the input does not overlap the output.  */
814    rcond = tcg_cond_to_rcond[cond];
815    if (c2 == 0 && rcond && c1 != ret) {
816        tcg_out_movi_s13(s, ret, 0);
817        tcg_out_movr(s, rcond, ret, c1, neg ? -1 : 1, 1);
818    } else {
819        tcg_out_cmp(s, cond, c1, c2, c2const);
820        tcg_out_movi_s13(s, ret, 0);
821        tcg_out_movcc(s, cond, MOVCC_XCC, ret, neg ? -1 : 1, 1);
822    }
823}
824
825static void tcg_out_brcond(TCGContext *s, TCGType type, TCGCond cond,
826                           TCGReg arg1, TCGArg arg2, bool const_arg2,
827                           TCGLabel *l)
828{
829    if (type == TCG_TYPE_I32) {
830        tcg_out_brcond_i32(s, cond, arg1, arg2, const_arg2, l);
831    } else {
832        tcg_out_brcond_i64(s, cond, arg1, arg2, const_arg2, l);
833    }
834}
835
836static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
837                        TCGReg arg1, TCGReg arg2, TCGLabel *l)
838{
839    tcg_out_brcond(s, type, cond, arg1, arg2, false, l);
840}
841
842static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond cond,
843                         TCGReg arg1, tcg_target_long arg2, TCGLabel *l)
844{
845    tcg_out_brcond(s, type, cond, arg1, arg2, true, l);
846}
847
848static const TCGOutOpBrcond outop_brcond = {
849    .base.static_constraint = C_O0_I2(r, rJ),
850    .out_rr = tgen_brcond,
851    .out_ri = tgen_brcondi,
852};
853
854static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
855                            TCGReg ret, TCGReg c1,
856                            TCGArg c2, bool c2const, bool neg)
857{
858    if (type == TCG_TYPE_I32) {
859        tcg_out_setcond_i32(s, cond, ret, c1, c2, c2const, neg);
860    } else {
861        tcg_out_setcond_i64(s, cond, ret, c1, c2, c2const, neg);
862    }
863}
864
865static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
866                         TCGReg dest, TCGReg arg1, TCGReg arg2)
867{
868    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, false);
869}
870
871static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
872                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
873{
874    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, false);
875}
876
877static const TCGOutOpSetcond outop_setcond = {
878    .base.static_constraint = C_O1_I2(r, r, rJ),
879    .out_rrr = tgen_setcond,
880    .out_rri = tgen_setcondi,
881};
882
883static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
884                            TCGReg dest, TCGReg arg1, TCGReg arg2)
885{
886    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, true);
887}
888
889static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
890                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
891{
892    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, true);
893}
894
895static const TCGOutOpSetcond outop_negsetcond = {
896    .base.static_constraint = C_O1_I2(r, r, rJ),
897    .out_rrr = tgen_negsetcond,
898    .out_rri = tgen_negsetcondi,
899};
900
901static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh,
902                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
903                                int32_t bh, int bhconst, int opl, int oph)
904{
905    TCGReg tmp = TCG_REG_T1;
906
907    /* Note that the low parts are fully consumed before tmp is set.  */
908    if (rl != ah && (bhconst || rl != bh)) {
909        tmp = rl;
910    }
911
912    tcg_out_arithc(s, tmp, al, bl, blconst, opl);
913    tcg_out_arithc(s, rh, ah, bh, bhconst, oph);
914    tcg_out_mov(s, TCG_TYPE_I32, rl, tmp);
915}
916
917static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
918                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
919                                int32_t bh, int bhconst, bool is_sub)
920{
921    TCGReg tmp = TCG_REG_T1;
922
923    /* Note that the low parts are fully consumed before tmp is set.  */
924    if (rl != ah && (bhconst || rl != bh)) {
925        tmp = rl;
926    }
927
928    tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
929
930    if (use_vis3_instructions && !is_sub) {
931        /* Note that ADDXC doesn't accept immediates.  */
932        if (bhconst && bh != 0) {
933           tcg_out_movi_s13(s, TCG_REG_T2, bh);
934           bh = TCG_REG_T2;
935        }
936        tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC);
937    } else if (bh == TCG_REG_G0) {
938	/* If we have a zero, we can perform the operation in two insns,
939           with the arithmetic first, and a conditional move into place.  */
940	if (rh == ah) {
941            tcg_out_arithi(s, TCG_REG_T2, ah, 1,
942			   is_sub ? ARITH_SUB : ARITH_ADD);
943            tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0);
944	} else {
945            tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD);
946	    tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0);
947	}
948    } else {
949        /*
950         * Otherwise adjust BH as if there is carry into T2.
951         * Note that constant BH is constrained to 11 bits for the MOVCC,
952         * so the adjustment fits 12 bits.
953         */
954        if (bhconst) {
955            tcg_out_movi_s13(s, TCG_REG_T2, bh + (is_sub ? -1 : 1));
956        } else {
957            tcg_out_arithi(s, TCG_REG_T2, bh, 1,
958                           is_sub ? ARITH_SUB : ARITH_ADD);
959        }
960        /* ... smoosh T2 back to original BH if carry is clear ... */
961        tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst);
962	/* ... and finally perform the arithmetic with the new operand.  */
963        tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD);
964    }
965
966    tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
967}
968
969static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest,
970                               bool in_prologue, bool tail_call)
971{
972    uintptr_t desti = (uintptr_t)dest;
973
974    tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
975                     desti & ~0xfff, in_prologue, TCG_REG_T2);
976    tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7,
977                   TCG_REG_T1, desti & 0xfff, JMPL);
978}
979
980static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
981                                 bool in_prologue)
982{
983    ptrdiff_t disp = tcg_pcrel_diff(s, dest);
984
985    if (disp == (int32_t)disp) {
986        tcg_out32(s, CALL | (uint32_t)disp >> 2);
987    } else {
988        tcg_out_jmpl_const(s, dest, in_prologue, false);
989    }
990}
991
992static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
993                         const TCGHelperInfo *info)
994{
995    tcg_out_call_nodelay(s, dest, false);
996    tcg_out_nop(s);
997}
998
999static void tcg_out_mb(TCGContext *s, TCGArg a0)
1000{
1001    /* Note that the TCG memory order constants mirror the Sparc MEMBAR.  */
1002    tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
1003}
1004
1005/* Generate global QEMU prologue and epilogue code */
1006static void tcg_target_qemu_prologue(TCGContext *s)
1007{
1008    int tmp_buf_size, frame_size;
1009
1010    /*
1011     * The TCG temp buffer is at the top of the frame, immediately
1012     * below the frame pointer.  Use the logical (aligned) offset here;
1013     * the stack bias is applied in temp_allocate_frame().
1014     */
1015    tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
1016    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
1017
1018    /*
1019     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
1020     * otherwise the minimal frame usable by callees.
1021     */
1022    frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
1023    frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
1024    frame_size += TCG_TARGET_STACK_ALIGN - 1;
1025    frame_size &= -TCG_TARGET_STACK_ALIGN;
1026    tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
1027              INSN_IMM13(-frame_size));
1028
1029#ifndef CONFIG_SOFTMMU
1030    if (guest_base != 0) {
1031        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG,
1032                         guest_base, true, TCG_REG_T1);
1033        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1034    }
1035#endif
1036
1037    /* We choose TCG_REG_TB such that no move is required.  */
1038    QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
1039    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
1040
1041    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
1042    /* delay slot */
1043    tcg_out_nop(s);
1044
1045    /* Epilogue for goto_ptr.  */
1046    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1047    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1048    /* delay slot */
1049    tcg_out_movi_s13(s, TCG_REG_O0, 0);
1050}
1051
1052static void tcg_out_tb_start(TCGContext *s)
1053{
1054    /* nothing to do */
1055}
1056
1057static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1058{
1059    int i;
1060    for (i = 0; i < count; ++i) {
1061        p[i] = NOP;
1062    }
1063}
1064
1065static const TCGLdstHelperParam ldst_helper_param = {
1066    .ntmp = 1, .tmp = { TCG_REG_T1 }
1067};
1068
1069static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1070{
1071    MemOp opc = get_memop(lb->oi);
1072    MemOp sgn;
1073
1074    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1075                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1076        return false;
1077    }
1078
1079    /* Use inline tcg_out_ext32s; otherwise let the helper sign-extend. */
1080    sgn = (opc & MO_SIZE) < MO_32 ? MO_SIGN : 0;
1081
1082    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1083    tcg_out_call(s, qemu_ld_helpers[opc & (MO_SIZE | sgn)], NULL);
1084    tcg_out_ld_helper_ret(s, lb, sgn, &ldst_helper_param);
1085
1086    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1087    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1088                       (intptr_t)lb->raddr, 0);
1089}
1090
1091static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1092{
1093    MemOp opc = get_memop(lb->oi);
1094
1095    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1096                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1097        return false;
1098    }
1099
1100    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1101    tcg_out_call(s, qemu_st_helpers[opc & MO_SIZE], NULL);
1102
1103    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1104    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1105                       (intptr_t)lb->raddr, 0);
1106}
1107
1108typedef struct {
1109    TCGReg base;
1110    TCGReg index;
1111    TCGAtomAlign aa;
1112} HostAddress;
1113
1114bool tcg_target_has_memory_bswap(MemOp memop)
1115{
1116    return true;
1117}
1118
1119/* We expect to use a 13-bit negative offset from ENV.  */
1120#define MIN_TLB_MASK_TABLE_OFS  -(1 << 12)
1121
1122/*
1123 * For system-mode, perform the TLB load and compare.
1124 * For user-mode, perform any required alignment tests.
1125 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1126 * is required and fill in @h with the host address for the fast path.
1127 */
1128static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1129                                           TCGReg addr_reg, MemOpIdx oi,
1130                                           bool is_ld)
1131{
1132    TCGType addr_type = s->addr_type;
1133    TCGLabelQemuLdst *ldst = NULL;
1134    MemOp opc = get_memop(oi);
1135    MemOp s_bits = opc & MO_SIZE;
1136    unsigned a_mask;
1137
1138    /* We don't support unaligned accesses. */
1139    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1140    h->aa.align = MAX(h->aa.align, s_bits);
1141    a_mask = (1u << h->aa.align) - 1;
1142
1143#ifdef CONFIG_SOFTMMU
1144    int mem_index = get_mmuidx(oi);
1145    int fast_off = tlb_mask_table_ofs(s, mem_index);
1146    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1147    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1148    int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1149                        : offsetof(CPUTLBEntry, addr_write);
1150    int add_off = offsetof(CPUTLBEntry, addend);
1151    int compare_mask;
1152    int cc;
1153
1154    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
1155    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T2, TCG_AREG0, mask_off);
1156    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T3, TCG_AREG0, table_off);
1157
1158    /* Extract the page index, shifted into place for tlb index.  */
1159    tcg_out_arithi(s, TCG_REG_T1, addr_reg,
1160                   s->page_bits - CPU_TLB_ENTRY_BITS, SHIFT_SRL);
1161    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T2, ARITH_AND);
1162
1163    /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2.  */
1164    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T3, ARITH_ADD);
1165
1166    /*
1167     * Load the tlb comparator and the addend.
1168     * Always load the entire 64-bit comparator for simplicity.
1169     * We will ignore the high bits via BPCC_ICC below.
1170     */
1171    tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_T2, TCG_REG_T1, cmp_off);
1172    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T1, TCG_REG_T1, add_off);
1173    h->base = TCG_REG_T1;
1174
1175    /* Mask out the page offset, except for the required alignment. */
1176    compare_mask = s->page_mask | a_mask;
1177    if (check_fit_tl(compare_mask, 13)) {
1178        tcg_out_arithi(s, TCG_REG_T3, addr_reg, compare_mask, ARITH_AND);
1179    } else {
1180        tcg_out_movi_s32(s, TCG_REG_T3, compare_mask);
1181        tcg_out_arith(s, TCG_REG_T3, addr_reg, TCG_REG_T3, ARITH_AND);
1182    }
1183    tcg_out_cmp(s, TCG_COND_NE, TCG_REG_T2, TCG_REG_T3, 0);
1184
1185    ldst = new_ldst_label(s);
1186    ldst->is_ld = is_ld;
1187    ldst->oi = oi;
1188    ldst->addr_reg = addr_reg;
1189    ldst->label_ptr[0] = s->code_ptr;
1190
1191    /* bne,pn %[xi]cc, label0 */
1192    cc = addr_type == TCG_TYPE_I32 ? BPCC_ICC : BPCC_XCC;
1193    tcg_out_bpcc0(s, COND_NE, BPCC_PN | cc, 0);
1194#else
1195    /*
1196     * If the size equals the required alignment, we can skip the test
1197     * and allow host SIGBUS to deliver SIGBUS to the guest.
1198     * Otherwise, test for at least natural alignment and defer
1199     * everything else to the helper functions.
1200     */
1201    if (s_bits != memop_alignment_bits(opc)) {
1202        tcg_debug_assert(check_fit_tl(a_mask, 13));
1203        tcg_out_arithi(s, TCG_REG_G0, addr_reg, a_mask, ARITH_ANDCC);
1204
1205        ldst = new_ldst_label(s);
1206        ldst->is_ld = is_ld;
1207        ldst->oi = oi;
1208        ldst->addr_reg = addr_reg;
1209        ldst->label_ptr[0] = s->code_ptr;
1210
1211        /* bne,pn %icc, label0 */
1212        tcg_out_bpcc0(s, COND_NE, BPCC_PN | BPCC_ICC, 0);
1213    }
1214    h->base = guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0;
1215#endif
1216
1217    /* If the guest address must be zero-extended, do in the delay slot.  */
1218    if (addr_type == TCG_TYPE_I32) {
1219        tcg_out_ext32u(s, TCG_REG_T2, addr_reg);
1220        h->index = TCG_REG_T2;
1221    } else {
1222        if (ldst) {
1223            tcg_out_nop(s);
1224        }
1225        h->index = addr_reg;
1226    }
1227    return ldst;
1228}
1229
1230static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
1231                            MemOpIdx oi, TCGType data_type)
1232{
1233    static const int ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = {
1234        [MO_UB]   = LDUB,
1235        [MO_SB]   = LDSB,
1236        [MO_UB | MO_LE] = LDUB,
1237        [MO_SB | MO_LE] = LDSB,
1238
1239        [MO_BEUW] = LDUH,
1240        [MO_BESW] = LDSH,
1241        [MO_BEUL] = LDUW,
1242        [MO_BESL] = LDSW,
1243        [MO_BEUQ] = LDX,
1244        [MO_BESQ] = LDX,
1245
1246        [MO_LEUW] = LDUH_LE,
1247        [MO_LESW] = LDSH_LE,
1248        [MO_LEUL] = LDUW_LE,
1249        [MO_LESL] = LDSW_LE,
1250        [MO_LEUQ] = LDX_LE,
1251        [MO_LESQ] = LDX_LE,
1252    };
1253
1254    TCGLabelQemuLdst *ldst;
1255    HostAddress h;
1256
1257    ldst = prepare_host_addr(s, &h, addr, oi, true);
1258
1259    tcg_out_ldst_rr(s, data, h.base, h.index,
1260                    ld_opc[get_memop(oi) & (MO_BSWAP | MO_SSIZE)]);
1261
1262    if (ldst) {
1263        ldst->type = data_type;
1264        ldst->datalo_reg = data;
1265        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1266    }
1267}
1268
1269static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr,
1270                            MemOpIdx oi, TCGType data_type)
1271{
1272    static const int st_opc[(MO_SIZE | MO_BSWAP) + 1] = {
1273        [MO_UB]   = STB,
1274
1275        [MO_BEUW] = STH,
1276        [MO_BEUL] = STW,
1277        [MO_BEUQ] = STX,
1278
1279        [MO_LEUW] = STH_LE,
1280        [MO_LEUL] = STW_LE,
1281        [MO_LEUQ] = STX_LE,
1282    };
1283
1284    TCGLabelQemuLdst *ldst;
1285    HostAddress h;
1286
1287    ldst = prepare_host_addr(s, &h, addr, oi, false);
1288
1289    tcg_out_ldst_rr(s, data, h.base, h.index,
1290                    st_opc[get_memop(oi) & (MO_BSWAP | MO_SIZE)]);
1291
1292    if (ldst) {
1293        ldst->type = data_type;
1294        ldst->datalo_reg = data;
1295        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1296    }
1297}
1298
1299static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1300{
1301    if (check_fit_ptr(a0, 13)) {
1302        tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1303        tcg_out_movi_s13(s, TCG_REG_O0, a0);
1304        return;
1305    } else {
1306        intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0);
1307        if (check_fit_ptr(tb_diff, 13)) {
1308            tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1309            /* Note that TCG_REG_TB has been unwound to O1.  */
1310            tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1311            return;
1312        }
1313    }
1314    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1315    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1316    tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1317}
1318
1319static void tcg_out_goto_tb(TCGContext *s, int which)
1320{
1321    ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which));
1322
1323    /* Load link and indirect branch. */
1324    set_jmp_insn_offset(s, which);
1325    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off);
1326    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1327    /* delay slot */
1328    tcg_out_nop(s);
1329    set_jmp_reset_offset(s, which);
1330
1331    /*
1332     * For the unlinked path of goto_tb, we need to reset TCG_REG_TB
1333     * to the beginning of this TB.
1334     */
1335    off = -tcg_current_code_size(s);
1336    if (check_fit_i32(off, 13)) {
1337        tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD);
1338    } else {
1339        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off);
1340        tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1341    }
1342}
1343
1344void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1345                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1346{
1347}
1348
1349
1350static void tgen_add(TCGContext *s, TCGType type,
1351                     TCGReg a0, TCGReg a1, TCGReg a2)
1352{
1353    tcg_out_arith(s, a0, a1, a2, ARITH_ADD);
1354}
1355
1356static void tgen_addi(TCGContext *s, TCGType type,
1357                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1358{
1359    tcg_out_arithi(s, a0, a1, a2, ARITH_ADD);
1360}
1361
1362static const TCGOutOpBinary outop_add = {
1363    .base.static_constraint = C_O1_I2(r, r, rJ),
1364    .out_rrr = tgen_add,
1365    .out_rri = tgen_addi,
1366};
1367
1368static void tgen_and(TCGContext *s, TCGType type,
1369                     TCGReg a0, TCGReg a1, TCGReg a2)
1370{
1371    tcg_out_arith(s, a0, a1, a2, ARITH_AND);
1372}
1373
1374static void tgen_andi(TCGContext *s, TCGType type,
1375                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1376{
1377    tcg_out_arithi(s, a0, a1, a2, ARITH_AND);
1378}
1379
1380static const TCGOutOpBinary outop_and = {
1381    .base.static_constraint = C_O1_I2(r, r, rJ),
1382    .out_rrr = tgen_and,
1383    .out_rri = tgen_andi,
1384};
1385
1386static void tgen_andc(TCGContext *s, TCGType type,
1387                      TCGReg a0, TCGReg a1, TCGReg a2)
1388{
1389    tcg_out_arith(s, a0, a1, a2, ARITH_ANDN);
1390}
1391
1392static const TCGOutOpBinary outop_andc = {
1393    .base.static_constraint = C_O1_I2(r, r, r),
1394    .out_rrr = tgen_andc,
1395};
1396
1397static const TCGOutOpBinary outop_clz = {
1398    .base.static_constraint = C_NotImplemented,
1399};
1400
1401static const TCGOutOpUnary outop_ctpop = {
1402    .base.static_constraint = C_NotImplemented,
1403};
1404
1405static const TCGOutOpBinary outop_ctz = {
1406    .base.static_constraint = C_NotImplemented,
1407};
1408
1409static void tgen_divs_rJ(TCGContext *s, TCGType type,
1410                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1411{
1412    uint32_t insn;
1413
1414    if (type == TCG_TYPE_I32) {
1415        /* Load Y with the sign extension of a1 to 64-bits.  */
1416        tcg_out_arithi(s, TCG_REG_T1, a1, 31, SHIFT_SRA);
1417        tcg_out_sety(s, TCG_REG_T1);
1418        insn = ARITH_SDIV;
1419    } else {
1420        insn = ARITH_SDIVX;
1421    }
1422    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1423}
1424
1425static void tgen_divs(TCGContext *s, TCGType type,
1426                      TCGReg a0, TCGReg a1, TCGReg a2)
1427{
1428    tgen_divs_rJ(s, type, a0, a1, a2, false);
1429}
1430
1431static void tgen_divsi(TCGContext *s, TCGType type,
1432                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1433{
1434    tgen_divs_rJ(s, type, a0, a1, a2, true);
1435}
1436
1437static const TCGOutOpBinary outop_divs = {
1438    .base.static_constraint = C_O1_I2(r, r, rJ),
1439    .out_rrr = tgen_divs,
1440    .out_rri = tgen_divsi,
1441};
1442
1443static const TCGOutOpDivRem outop_divs2 = {
1444    .base.static_constraint = C_NotImplemented,
1445};
1446
1447static void tgen_divu_rJ(TCGContext *s, TCGType type,
1448                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1449{
1450    uint32_t insn;
1451
1452    if (type == TCG_TYPE_I32) {
1453        /* Load Y with the zero extension to 64-bits.  */
1454        tcg_out_sety(s, TCG_REG_G0);
1455        insn = ARITH_UDIV;
1456    } else {
1457        insn = ARITH_UDIVX;
1458    }
1459    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1460}
1461
1462static void tgen_divu(TCGContext *s, TCGType type,
1463                      TCGReg a0, TCGReg a1, TCGReg a2)
1464{
1465    tgen_divu_rJ(s, type, a0, a1, a2, false);
1466}
1467
1468static void tgen_divui(TCGContext *s, TCGType type,
1469                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1470{
1471    tgen_divu_rJ(s, type, a0, a1, a2, true);
1472}
1473
1474static const TCGOutOpBinary outop_divu = {
1475    .base.static_constraint = C_O1_I2(r, r, rJ),
1476    .out_rrr = tgen_divu,
1477    .out_rri = tgen_divui,
1478};
1479
1480static const TCGOutOpDivRem outop_divu2 = {
1481    .base.static_constraint = C_NotImplemented,
1482};
1483
1484static const TCGOutOpBinary outop_eqv = {
1485    .base.static_constraint = C_NotImplemented,
1486};
1487
1488static void tgen_mul(TCGContext *s, TCGType type,
1489                     TCGReg a0, TCGReg a1, TCGReg a2)
1490{
1491    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1492    tcg_out_arith(s, a0, a1, a2, insn);
1493}
1494
1495static void tgen_muli(TCGContext *s, TCGType type,
1496                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1497{
1498    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1499    tcg_out_arithi(s, a0, a1, a2, insn);
1500}
1501
1502static const TCGOutOpBinary outop_mul = {
1503    .base.static_constraint = C_O1_I2(r, r, rJ),
1504    .out_rrr = tgen_mul,
1505    .out_rri = tgen_muli,
1506};
1507
1508/*
1509 * The 32-bit multiply insns produce a full 64-bit result.
1510 * Supporting 32-bit mul[us]2 opcodes avoids sign/zero-extensions
1511 * before the actual multiply; we only need extract the high part
1512 * into the separate operand.
1513 */
1514static TCGConstraintSetIndex cset_mul2(TCGType type, unsigned flags)
1515{
1516    return type == TCG_TYPE_I32 ? C_O2_I2(r, r, r, r) : C_NotImplemented;
1517}
1518
1519static void tgen_muls2(TCGContext *s, TCGType type,
1520                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1521{
1522    tcg_out_arith(s, a0, a2, a3, ARITH_SMUL);
1523    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1524}
1525
1526static const TCGOutOpMul2 outop_muls2 = {
1527    .base.static_constraint = C_Dynamic,
1528    .base.dynamic_constraint = cset_mul2,
1529    .out_rrrr = tgen_muls2,
1530};
1531
1532static const TCGOutOpBinary outop_mulsh = {
1533    .base.static_constraint = C_NotImplemented,
1534};
1535
1536static void tgen_mulu2(TCGContext *s, TCGType type,
1537                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1538{
1539    tcg_out_arith(s, a0, a2, a3, ARITH_UMUL);
1540    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1541}
1542
1543static const TCGOutOpMul2 outop_mulu2 = {
1544    .base.static_constraint = C_Dynamic,
1545    .base.dynamic_constraint = cset_mul2,
1546    .out_rrrr = tgen_mulu2,
1547};
1548
1549static void tgen_muluh(TCGContext *s, TCGType type,
1550                       TCGReg a0, TCGReg a1, TCGReg a2)
1551{
1552    tcg_out_arith(s, a0, a1, a2, ARITH_UMULXHI);
1553}
1554
1555static TCGConstraintSetIndex cset_muluh(TCGType type, unsigned flags)
1556{
1557    return (type == TCG_TYPE_I64 && use_vis3_instructions
1558            ? C_O1_I2(r, r, r) : C_NotImplemented);
1559}
1560
1561static const TCGOutOpBinary outop_muluh = {
1562    .base.static_constraint = C_Dynamic,
1563    .base.dynamic_constraint = cset_muluh,
1564    .out_rrr = tgen_muluh,
1565};
1566
1567static const TCGOutOpBinary outop_nand = {
1568    .base.static_constraint = C_NotImplemented,
1569};
1570
1571static const TCGOutOpBinary outop_nor = {
1572    .base.static_constraint = C_NotImplemented,
1573};
1574
1575static void tgen_or(TCGContext *s, TCGType type,
1576                     TCGReg a0, TCGReg a1, TCGReg a2)
1577{
1578    tcg_out_arith(s, a0, a1, a2, ARITH_OR);
1579}
1580
1581static void tgen_ori(TCGContext *s, TCGType type,
1582                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1583{
1584    tcg_out_arithi(s, a0, a1, a2, ARITH_OR);
1585}
1586
1587static const TCGOutOpBinary outop_or = {
1588    .base.static_constraint = C_O1_I2(r, r, rJ),
1589    .out_rrr = tgen_or,
1590    .out_rri = tgen_ori,
1591};
1592
1593static void tgen_orc(TCGContext *s, TCGType type,
1594                     TCGReg a0, TCGReg a1, TCGReg a2)
1595{
1596    tcg_out_arith(s, a0, a1, a2, ARITH_ORN);
1597}
1598
1599static const TCGOutOpBinary outop_orc = {
1600    .base.static_constraint = C_O1_I2(r, r, r),
1601    .out_rrr = tgen_orc,
1602};
1603
1604static const TCGOutOpBinary outop_rems = {
1605    .base.static_constraint = C_NotImplemented,
1606};
1607
1608static const TCGOutOpBinary outop_remu = {
1609    .base.static_constraint = C_NotImplemented,
1610};
1611
1612static const TCGOutOpBinary outop_rotl = {
1613    .base.static_constraint = C_NotImplemented,
1614};
1615
1616static const TCGOutOpBinary outop_rotr = {
1617    .base.static_constraint = C_NotImplemented,
1618};
1619
1620static void tgen_sar(TCGContext *s, TCGType type,
1621                     TCGReg a0, TCGReg a1, TCGReg a2)
1622{
1623    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1624    tcg_out_arith(s, a0, a1, a2, insn);
1625}
1626
1627static void tgen_sari(TCGContext *s, TCGType type,
1628                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1629{
1630    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1631    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1632    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1633}
1634
1635static const TCGOutOpBinary outop_sar = {
1636    .base.static_constraint = C_O1_I2(r, r, rJ),
1637    .out_rrr = tgen_sar,
1638    .out_rri = tgen_sari,
1639};
1640
1641static void tgen_shl(TCGContext *s, TCGType type,
1642                     TCGReg a0, TCGReg a1, TCGReg a2)
1643{
1644    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1645    tcg_out_arith(s, a0, a1, a2, insn);
1646}
1647
1648static void tgen_shli(TCGContext *s, TCGType type,
1649                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1650{
1651    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1652    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1653    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1654}
1655
1656static const TCGOutOpBinary outop_shl = {
1657    .base.static_constraint = C_O1_I2(r, r, rJ),
1658    .out_rrr = tgen_shl,
1659    .out_rri = tgen_shli,
1660};
1661
1662static void tgen_shr(TCGContext *s, TCGType type,
1663                     TCGReg a0, TCGReg a1, TCGReg a2)
1664{
1665    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1666    tcg_out_arith(s, a0, a1, a2, insn);
1667}
1668
1669static void tgen_shri(TCGContext *s, TCGType type,
1670                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1671{
1672    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1673    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1674    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1675}
1676
1677static const TCGOutOpBinary outop_shr = {
1678    .base.static_constraint = C_O1_I2(r, r, rJ),
1679    .out_rrr = tgen_shr,
1680    .out_rri = tgen_shri,
1681};
1682
1683static void tgen_sub(TCGContext *s, TCGType type,
1684                     TCGReg a0, TCGReg a1, TCGReg a2)
1685{
1686    tcg_out_arith(s, a0, a1, a2, ARITH_SUB);
1687}
1688
1689static const TCGOutOpSubtract outop_sub = {
1690    .base.static_constraint = C_O1_I2(r, r, r),
1691    .out_rrr = tgen_sub,
1692};
1693
1694static void tgen_xor(TCGContext *s, TCGType type,
1695                     TCGReg a0, TCGReg a1, TCGReg a2)
1696{
1697    tcg_out_arith(s, a0, a1, a2, ARITH_XOR);
1698}
1699
1700static void tgen_xori(TCGContext *s, TCGType type,
1701                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1702{
1703    tcg_out_arithi(s, a0, a1, a2, ARITH_XOR);
1704}
1705
1706static const TCGOutOpBinary outop_xor = {
1707    .base.static_constraint = C_O1_I2(r, r, rJ),
1708    .out_rrr = tgen_xor,
1709    .out_rri = tgen_xori,
1710};
1711
1712static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1713{
1714     tgen_sub(s, type, a0, TCG_REG_G0, a1);
1715}
1716
1717static const TCGOutOpUnary outop_neg = {
1718    .base.static_constraint = C_O1_I1(r, r),
1719    .out_rr = tgen_neg,
1720};
1721
1722static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1723{
1724     tgen_orc(s, type, a0, TCG_REG_G0, a1);
1725}
1726
1727static const TCGOutOpUnary outop_not = {
1728    .base.static_constraint = C_O1_I1(r, r),
1729    .out_rr = tgen_not,
1730};
1731
1732
1733static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
1734                       const TCGArg args[TCG_MAX_OP_ARGS],
1735                       const int const_args[TCG_MAX_OP_ARGS])
1736{
1737    TCGArg a0, a1, a2;
1738    int c2;
1739
1740    /* Hoist the loads of the most common arguments.  */
1741    a0 = args[0];
1742    a1 = args[1];
1743    a2 = args[2];
1744    c2 = const_args[2];
1745
1746    switch (opc) {
1747    case INDEX_op_goto_ptr:
1748        tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1749        tcg_out_mov_delay(s, TCG_REG_TB, a0);
1750        break;
1751    case INDEX_op_br:
1752        tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0));
1753        tcg_out_nop(s);
1754        break;
1755
1756#define OP_32_64(x)                             \
1757        glue(glue(case INDEX_op_, x), _i32):    \
1758        glue(glue(case INDEX_op_, x), _i64)
1759
1760    OP_32_64(ld8u):
1761        tcg_out_ldst(s, a0, a1, a2, LDUB);
1762        break;
1763    OP_32_64(ld8s):
1764        tcg_out_ldst(s, a0, a1, a2, LDSB);
1765        break;
1766    OP_32_64(ld16u):
1767        tcg_out_ldst(s, a0, a1, a2, LDUH);
1768        break;
1769    OP_32_64(ld16s):
1770        tcg_out_ldst(s, a0, a1, a2, LDSH);
1771        break;
1772    case INDEX_op_ld_i32:
1773    case INDEX_op_ld32u_i64:
1774        tcg_out_ldst(s, a0, a1, a2, LDUW);
1775        break;
1776    OP_32_64(st8):
1777        tcg_out_ldst(s, a0, a1, a2, STB);
1778        break;
1779    OP_32_64(st16):
1780        tcg_out_ldst(s, a0, a1, a2, STH);
1781        break;
1782    case INDEX_op_st_i32:
1783    case INDEX_op_st32_i64:
1784        tcg_out_ldst(s, a0, a1, a2, STW);
1785        break;
1786
1787    case INDEX_op_movcond_i32:
1788        tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1789        break;
1790
1791    case INDEX_op_add2_i32:
1792        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1793                            args[4], const_args[4], args[5], const_args[5],
1794                            ARITH_ADDCC, ARITH_ADDC);
1795        break;
1796    case INDEX_op_sub2_i32:
1797        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1798                            args[4], const_args[4], args[5], const_args[5],
1799                            ARITH_SUBCC, ARITH_SUBC);
1800        break;
1801
1802    case INDEX_op_qemu_ld_i32:
1803        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32);
1804        break;
1805    case INDEX_op_qemu_ld_i64:
1806        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64);
1807        break;
1808    case INDEX_op_qemu_st_i32:
1809        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32);
1810        break;
1811    case INDEX_op_qemu_st_i64:
1812        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64);
1813        break;
1814
1815    case INDEX_op_ld32s_i64:
1816        tcg_out_ldst(s, a0, a1, a2, LDSW);
1817        break;
1818    case INDEX_op_ld_i64:
1819        tcg_out_ldst(s, a0, a1, a2, LDX);
1820        break;
1821    case INDEX_op_st_i64:
1822        tcg_out_ldst(s, a0, a1, a2, STX);
1823        break;
1824
1825    case INDEX_op_movcond_i64:
1826        tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1827        break;
1828    case INDEX_op_add2_i64:
1829        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1830                            const_args[4], args[5], const_args[5], false);
1831        break;
1832    case INDEX_op_sub2_i64:
1833        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1834                            const_args[4], args[5], const_args[5], true);
1835        break;
1836
1837    case INDEX_op_mb:
1838        tcg_out_mb(s, a0);
1839        break;
1840
1841    case INDEX_op_extract_i64:
1842        tcg_debug_assert(a2 + args[3] == 32);
1843        tcg_out_arithi(s, a0, a1, a2, SHIFT_SRL);
1844        break;
1845    case INDEX_op_sextract_i64:
1846        tcg_debug_assert(a2 + args[3] == 32);
1847        tcg_out_arithi(s, a0, a1, a2, SHIFT_SRA);
1848        break;
1849
1850    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
1851    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
1852    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
1853    case INDEX_op_ext_i32_i64:  /* Always emitted via tcg_reg_alloc_op.  */
1854    case INDEX_op_extu_i32_i64:
1855    default:
1856        g_assert_not_reached();
1857    }
1858}
1859
1860static TCGConstraintSetIndex
1861tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
1862{
1863    switch (op) {
1864    case INDEX_op_goto_ptr:
1865        return C_O0_I1(r);
1866
1867    case INDEX_op_ld8u_i32:
1868    case INDEX_op_ld8u_i64:
1869    case INDEX_op_ld8s_i32:
1870    case INDEX_op_ld8s_i64:
1871    case INDEX_op_ld16u_i32:
1872    case INDEX_op_ld16u_i64:
1873    case INDEX_op_ld16s_i32:
1874    case INDEX_op_ld16s_i64:
1875    case INDEX_op_ld_i32:
1876    case INDEX_op_ld32u_i64:
1877    case INDEX_op_ld32s_i64:
1878    case INDEX_op_ld_i64:
1879    case INDEX_op_ext_i32_i64:
1880    case INDEX_op_extu_i32_i64:
1881    case INDEX_op_extract_i64:
1882    case INDEX_op_sextract_i64:
1883    case INDEX_op_qemu_ld_i32:
1884    case INDEX_op_qemu_ld_i64:
1885        return C_O1_I1(r, r);
1886
1887    case INDEX_op_st8_i32:
1888    case INDEX_op_st8_i64:
1889    case INDEX_op_st16_i32:
1890    case INDEX_op_st16_i64:
1891    case INDEX_op_st_i32:
1892    case INDEX_op_st32_i64:
1893    case INDEX_op_st_i64:
1894    case INDEX_op_qemu_st_i32:
1895    case INDEX_op_qemu_st_i64:
1896        return C_O0_I2(rz, r);
1897
1898    case INDEX_op_movcond_i32:
1899    case INDEX_op_movcond_i64:
1900        return C_O1_I4(r, rz, rJ, rI, 0);
1901    case INDEX_op_add2_i32:
1902    case INDEX_op_add2_i64:
1903    case INDEX_op_sub2_i32:
1904    case INDEX_op_sub2_i64:
1905        return C_O2_I4(r, r, rz, rz, rJ, rJ);
1906
1907    default:
1908        return C_NotImplemented;
1909    }
1910}
1911
1912static void tcg_target_init(TCGContext *s)
1913{
1914    /*
1915     * Only probe for the platform and capabilities if we haven't already
1916     * determined maximum values at compile time.
1917     */
1918#ifndef use_vis3_instructions
1919    {
1920        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
1921        use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
1922    }
1923#endif
1924
1925    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
1926    tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;
1927
1928    tcg_target_call_clobber_regs = 0;
1929    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
1930    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
1931    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
1932    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
1933    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
1934    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
1935    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
1936    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
1937    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
1938    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
1939    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
1940    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
1941    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
1942    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
1943    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
1944
1945    s->reserved_regs = 0;
1946    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
1947    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
1948    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
1949    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
1950    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
1951    tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
1952    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
1953    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
1954    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T3); /* for internal use */
1955}
1956
1957#define ELF_HOST_MACHINE  EM_SPARCV9
1958
1959typedef struct {
1960    DebugFrameHeader h;
1961    uint8_t fde_def_cfa[4];
1962    uint8_t fde_win_save;
1963    uint8_t fde_ret_save[3];
1964} DebugFrame;
1965
1966static const DebugFrame debug_frame = {
1967    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
1968    .h.cie.id = -1,
1969    .h.cie.version = 1,
1970    .h.cie.code_align = 1,
1971    .h.cie.data_align = -sizeof(void *) & 0x7f,
1972    .h.cie.return_column = 15,            /* o7 */
1973
1974    /* Total FDE size does not include the "len" member.  */
1975    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1976
1977    .fde_def_cfa = {
1978        12, 30,                         /* DW_CFA_def_cfa i6, 2047 */
1979        (2047 & 0x7f) | 0x80, (2047 >> 7)
1980    },
1981    .fde_win_save = 0x2d,               /* DW_CFA_GNU_window_save */
1982    .fde_ret_save = { 9, 15, 31 },      /* DW_CFA_register o7, i7 */
1983};
1984
1985void tcg_register_jit(const void *buf, size_t buf_size)
1986{
1987    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
1988}
1989