xref: /openbmc/qemu/tcg/sparc64/tcg-target.c.inc (revision 33aba058c8fcc9b1581b03a1fbac45d8d91baac6)
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_ADDCCC (INSN_OP(2) | INSN_OP3(0x18))
203#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
204#define ARITH_SUBCCC (INSN_OP(2) | INSN_OP3(0x1c))
205#define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
206#define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
207#define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
208#define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f))
209#define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09))
210#define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d))
211#define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d))
212#define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
213#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
214
215#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
216#define ARITH_ADDXCCC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x13))
217#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
218
219#define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
220#define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
221#define SHIFT_SRA  (INSN_OP(2) | INSN_OP3(0x27))
222
223#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
224#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
225#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
226
227#define RDY        (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
228#define WRY        (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
229#define WRCCR      (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(2))
230#define JMPL       (INSN_OP(2) | INSN_OP3(0x38))
231#define RETURN     (INSN_OP(2) | INSN_OP3(0x39))
232#define SAVE       (INSN_OP(2) | INSN_OP3(0x3c))
233#define RESTORE    (INSN_OP(2) | INSN_OP3(0x3d))
234#define SETHI      (INSN_OP(0) | INSN_OP2(0x4))
235#define CALL       INSN_OP(1)
236#define LDUB       (INSN_OP(3) | INSN_OP3(0x01))
237#define LDSB       (INSN_OP(3) | INSN_OP3(0x09))
238#define LDUH       (INSN_OP(3) | INSN_OP3(0x02))
239#define LDSH       (INSN_OP(3) | INSN_OP3(0x0a))
240#define LDUW       (INSN_OP(3) | INSN_OP3(0x00))
241#define LDSW       (INSN_OP(3) | INSN_OP3(0x08))
242#define LDX        (INSN_OP(3) | INSN_OP3(0x0b))
243#define STB        (INSN_OP(3) | INSN_OP3(0x05))
244#define STH        (INSN_OP(3) | INSN_OP3(0x06))
245#define STW        (INSN_OP(3) | INSN_OP3(0x04))
246#define STX        (INSN_OP(3) | INSN_OP3(0x0e))
247#define LDUBA      (INSN_OP(3) | INSN_OP3(0x11))
248#define LDSBA      (INSN_OP(3) | INSN_OP3(0x19))
249#define LDUHA      (INSN_OP(3) | INSN_OP3(0x12))
250#define LDSHA      (INSN_OP(3) | INSN_OP3(0x1a))
251#define LDUWA      (INSN_OP(3) | INSN_OP3(0x10))
252#define LDSWA      (INSN_OP(3) | INSN_OP3(0x18))
253#define LDXA       (INSN_OP(3) | INSN_OP3(0x1b))
254#define STBA       (INSN_OP(3) | INSN_OP3(0x15))
255#define STHA       (INSN_OP(3) | INSN_OP3(0x16))
256#define STWA       (INSN_OP(3) | INSN_OP3(0x14))
257#define STXA       (INSN_OP(3) | INSN_OP3(0x1e))
258
259#define MEMBAR     (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
260
261#define NOP        (SETHI | INSN_RD(TCG_REG_G0) | 0)
262
263#ifndef ASI_PRIMARY_LITTLE
264#define ASI_PRIMARY_LITTLE 0x88
265#endif
266
267#define LDUH_LE    (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
268#define LDSH_LE    (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
269#define LDUW_LE    (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
270#define LDSW_LE    (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
271#define LDX_LE     (LDXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
272
273#define STH_LE     (STHA  | INSN_ASI(ASI_PRIMARY_LITTLE))
274#define STW_LE     (STWA  | INSN_ASI(ASI_PRIMARY_LITTLE))
275#define STX_LE     (STXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
276
277#ifndef use_vis3_instructions
278bool use_vis3_instructions;
279#endif
280
281static bool check_fit_i64(int64_t val, unsigned int bits)
282{
283    return val == sextract64(val, 0, bits);
284}
285
286static bool check_fit_i32(int32_t val, unsigned int bits)
287{
288    return val == sextract32(val, 0, bits);
289}
290
291#define check_fit_tl    check_fit_i64
292#define check_fit_ptr   check_fit_i64
293
294static bool patch_reloc(tcg_insn_unit *src_rw, int type,
295                        intptr_t value, intptr_t addend)
296{
297    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
298    uint32_t insn = *src_rw;
299    intptr_t pcrel;
300
301    value += addend;
302    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx);
303
304    switch (type) {
305    case R_SPARC_WDISP16:
306        if (!check_fit_ptr(pcrel >> 2, 16)) {
307            return false;
308        }
309        insn &= ~INSN_OFF16(-1);
310        insn |= INSN_OFF16(pcrel);
311        break;
312    case R_SPARC_WDISP19:
313        if (!check_fit_ptr(pcrel >> 2, 19)) {
314            return false;
315        }
316        insn &= ~INSN_OFF19(-1);
317        insn |= INSN_OFF19(pcrel);
318        break;
319    case R_SPARC_13:
320        if (!check_fit_ptr(value, 13)) {
321            return false;
322        }
323        insn &= ~INSN_IMM13(-1);
324        insn |= INSN_IMM13(value);
325        break;
326    default:
327        g_assert_not_reached();
328    }
329
330    *src_rw = insn;
331    return true;
332}
333
334/* test if a constant matches the constraint */
335static bool tcg_target_const_match(int64_t val, int ct,
336                                   TCGType type, TCGCond cond, int vece)
337{
338    if (ct & TCG_CT_CONST) {
339        return 1;
340    }
341
342    if (type == TCG_TYPE_I32) {
343        val = (int32_t)val;
344    }
345
346    if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
347        return 1;
348    } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
349        return 1;
350    } else {
351        return 0;
352    }
353}
354
355static void tcg_out_nop(TCGContext *s)
356{
357    tcg_out32(s, NOP);
358}
359
360static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
361                          TCGReg rs2, int op)
362{
363    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
364}
365
366static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
367                           int32_t offset, int op)
368{
369    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
370}
371
372static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
373                           int32_t val2, int val2const, int op)
374{
375    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
376              | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
377}
378
379static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
380{
381    if (ret != arg) {
382        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
383    }
384    return true;
385}
386
387static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg)
388{
389    if (ret != arg) {
390        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
391    } else {
392        tcg_out_nop(s);
393    }
394}
395
396static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
397{
398    tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
399}
400
401/* A 13-bit constant sign-extended to 64 bits.  */
402static void tcg_out_movi_s13(TCGContext *s, TCGReg ret, int32_t arg)
403{
404    tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
405}
406
407/* A 32-bit constant sign-extended to 64 bits.  */
408static void tcg_out_movi_s32(TCGContext *s, TCGReg ret, int32_t arg)
409{
410    tcg_out_sethi(s, ret, ~arg);
411    tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
412}
413
414/* A 32-bit constant zero-extended to 64 bits.  */
415static void tcg_out_movi_u32(TCGContext *s, TCGReg ret, uint32_t arg)
416{
417    tcg_out_sethi(s, ret, arg);
418    if (arg & 0x3ff) {
419        tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
420    }
421}
422
423static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
424                             tcg_target_long arg, bool in_prologue,
425                             TCGReg scratch)
426{
427    tcg_target_long hi, lo = (int32_t)arg;
428    tcg_target_long test, lsb;
429
430    /* A 13-bit constant sign-extended to 64-bits.  */
431    if (check_fit_tl(arg, 13)) {
432        tcg_out_movi_s13(s, ret, arg);
433        return;
434    }
435
436    /* A 32-bit constant, or 32-bit zero-extended to 64-bits.  */
437    if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
438        tcg_out_movi_u32(s, ret, arg);
439        return;
440    }
441
442    /* A 13-bit constant relative to the TB.  */
443    if (!in_prologue) {
444        test = tcg_tbrel_diff(s, (void *)arg);
445        if (check_fit_ptr(test, 13)) {
446            tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
447            return;
448        }
449    }
450
451    /* A 32-bit constant sign-extended to 64-bits.  */
452    if (arg == lo) {
453        tcg_out_movi_s32(s, ret, arg);
454        return;
455    }
456
457    /* A 32-bit constant, shifted.  */
458    lsb = ctz64(arg);
459    test = (tcg_target_long)arg >> lsb;
460    if (lsb > 10 && test == extract64(test, 0, 21)) {
461        tcg_out_sethi(s, ret, test << 10);
462        tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
463        return;
464    } else if (test == (uint32_t)test || test == (int32_t)test) {
465        tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch);
466        tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
467        return;
468    }
469
470    /* Use the constant pool, if possible. */
471    if (!in_prologue) {
472        new_pool_label(s, arg, R_SPARC_13, s->code_ptr,
473                       tcg_tbrel_diff(s, NULL));
474        tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB));
475        return;
476    }
477
478    /* A 64-bit constant decomposed into 2 32-bit pieces.  */
479    if (check_fit_i32(lo, 13)) {
480        hi = (arg - lo) >> 32;
481        tcg_out_movi_u32(s, ret, hi);
482        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
483        tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
484    } else {
485        hi = arg >> 32;
486        tcg_out_movi_u32(s, ret, hi);
487        tcg_out_movi_u32(s, scratch, lo);
488        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
489        tcg_out_arith(s, ret, ret, scratch, ARITH_OR);
490    }
491}
492
493static void tcg_out_movi(TCGContext *s, TCGType type,
494                         TCGReg ret, tcg_target_long arg)
495{
496    tcg_debug_assert(ret != TCG_REG_T3);
497    tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T3);
498}
499
500static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
501{
502    g_assert_not_reached();
503}
504
505static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
506{
507    g_assert_not_reached();
508}
509
510static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs)
511{
512    tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND);
513}
514
515static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs)
516{
517    tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL);
518    tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL);
519}
520
521static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs)
522{
523    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA);
524}
525
526static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs)
527{
528    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL);
529}
530
531static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
532{
533    tcg_out_ext32s(s, rd, rs);
534}
535
536static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
537{
538    tcg_out_ext32u(s, rd, rs);
539}
540
541static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs)
542{
543    tcg_out_ext32u(s, rd, rs);
544}
545
546static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
547{
548    return false;
549}
550
551static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
552                             tcg_target_long imm)
553{
554    /* This function is only used for passing structs by reference. */
555    g_assert_not_reached();
556}
557
558static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
559                            TCGReg a2, int op)
560{
561    tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
562}
563
564static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
565                         intptr_t offset, int op)
566{
567    if (check_fit_ptr(offset, 13)) {
568        tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
569                  INSN_IMM13(offset));
570    } else {
571        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
572        tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
573    }
574}
575
576static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
577                       TCGReg arg1, intptr_t arg2)
578{
579    tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
580}
581
582static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
583                       TCGReg arg1, intptr_t arg2)
584{
585    tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
586}
587
588static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
589                        TCGReg base, intptr_t ofs)
590{
591    if (val == 0) {
592        tcg_out_st(s, type, TCG_REG_G0, base, ofs);
593        return true;
594    }
595    return false;
596}
597
598static void tcg_out_sety(TCGContext *s, TCGReg rs)
599{
600    tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
601}
602
603static const uint8_t tcg_cond_to_bcond[16] = {
604    [TCG_COND_EQ] = COND_E,
605    [TCG_COND_NE] = COND_NE,
606    [TCG_COND_TSTEQ] = COND_E,
607    [TCG_COND_TSTNE] = COND_NE,
608    [TCG_COND_LT] = COND_L,
609    [TCG_COND_GE] = COND_GE,
610    [TCG_COND_LE] = COND_LE,
611    [TCG_COND_GT] = COND_G,
612    [TCG_COND_LTU] = COND_CS,
613    [TCG_COND_GEU] = COND_CC,
614    [TCG_COND_LEU] = COND_LEU,
615    [TCG_COND_GTU] = COND_GU,
616};
617
618static const uint8_t tcg_cond_to_rcond[16] = {
619    [TCG_COND_EQ] = RCOND_Z,
620    [TCG_COND_NE] = RCOND_NZ,
621    [TCG_COND_LT] = RCOND_LZ,
622    [TCG_COND_GT] = RCOND_GZ,
623    [TCG_COND_LE] = RCOND_LEZ,
624    [TCG_COND_GE] = RCOND_GEZ
625};
626
627static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
628{
629    tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
630}
631
632static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
633{
634    int off19 = 0;
635
636    if (l->has_value) {
637        off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
638    } else {
639        tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
640    }
641    tcg_out_bpcc0(s, scond, flags, off19);
642}
643
644static void tcg_out_br(TCGContext *s, TCGLabel *l)
645{
646    tcg_out_bpcc(s, COND_A, BPCC_PT, l);
647    tcg_out_nop(s);
648}
649
650static void tcg_out_cmp(TCGContext *s, TCGCond cond,
651                        TCGReg c1, int32_t c2, int c2const)
652{
653    tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const,
654                   is_tst_cond(cond) ? ARITH_ANDCC : ARITH_SUBCC);
655}
656
657static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
658                               int32_t arg2, int const_arg2, TCGLabel *l)
659{
660    tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
661    tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
662    tcg_out_nop(s);
663}
664
665static void tcg_out_movcc(TCGContext *s, int scond, int cc, TCGReg ret,
666                          int32_t v1, int v1const)
667{
668    tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret) | INSN_RS1(scond)
669              | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
670}
671
672static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
673                                TCGReg c1, int32_t c2, int c2const,
674                                int32_t v1, int v1const)
675{
676    tcg_out_cmp(s, cond, c1, c2, c2const);
677    tcg_out_movcc(s, tcg_cond_to_bcond[cond], MOVCC_ICC, ret, v1, v1const);
678}
679
680static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
681                               int32_t arg2, int const_arg2, TCGLabel *l)
682{
683    /* For 64-bit signed comparisons vs zero, we can avoid the compare.  */
684    int rcond = tcg_cond_to_rcond[cond];
685    if (arg2 == 0 && rcond) {
686        int off16 = 0;
687
688        if (l->has_value) {
689            off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
690        } else {
691            tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
692        }
693        tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
694                  | INSN_COND(rcond) | off16);
695    } else {
696        tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
697        tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
698    }
699    tcg_out_nop(s);
700}
701
702static void tcg_out_movr(TCGContext *s, int rcond, TCGReg ret, TCGReg c1,
703                         int32_t v1, int v1const)
704{
705    tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | (rcond << 10)
706              | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
707}
708
709static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
710                                TCGReg c1, int32_t c2, int c2const,
711                                int32_t v1, int v1const)
712{
713    /* For 64-bit signed comparisons vs zero, we can avoid the compare.
714       Note that the immediate range is one bit smaller, so we must check
715       for that as well.  */
716    int rcond = tcg_cond_to_rcond[cond];
717    if (c2 == 0 && rcond && (!v1const || check_fit_i32(v1, 10))) {
718        tcg_out_movr(s, rcond, ret, c1, v1, v1const);
719    } else {
720        tcg_out_cmp(s, cond, c1, c2, c2const);
721        tcg_out_movcc(s, tcg_cond_to_bcond[cond], MOVCC_XCC, ret, v1, v1const);
722    }
723}
724
725static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
726                                TCGReg c1, int32_t c2, bool c2const, bool neg)
727{
728    /* For 32-bit comparisons, we can play games with ADDC/SUBC.  */
729    switch (cond) {
730    case TCG_COND_LTU:
731    case TCG_COND_GEU:
732        /* The result of the comparison is in the carry bit.  */
733        break;
734
735    case TCG_COND_EQ:
736    case TCG_COND_NE:
737        /* For equality, we can transform to inequality vs zero.  */
738        if (c2 != 0) {
739            tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
740            c2 = TCG_REG_T1;
741        } else {
742            c2 = c1;
743        }
744        c1 = TCG_REG_G0, c2const = 0;
745        cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
746        break;
747
748    case TCG_COND_TSTEQ:
749    case TCG_COND_TSTNE:
750        /* Transform to inequality vs zero.  */
751        tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_AND);
752        c1 = TCG_REG_G0;
753        c2 = TCG_REG_T1, c2const = 0;
754        cond = (cond == TCG_COND_TSTEQ ? TCG_COND_GEU : TCG_COND_LTU);
755        break;
756
757    case TCG_COND_GTU:
758    case TCG_COND_LEU:
759        /* If we don't need to load a constant into a register, we can
760           swap the operands on GTU/LEU.  There's no benefit to loading
761           the constant into a temporary register.  */
762        if (!c2const || c2 == 0) {
763            TCGReg t = c1;
764            c1 = c2;
765            c2 = t;
766            c2const = 0;
767            cond = tcg_swap_cond(cond);
768            break;
769        }
770        /* FALLTHRU */
771
772    default:
773        tcg_out_cmp(s, cond, c1, c2, c2const);
774        tcg_out_movi_s13(s, ret, 0);
775        tcg_out_movcc(s, tcg_cond_to_bcond[cond],
776                      MOVCC_ICC, ret, neg ? -1 : 1, 1);
777        return;
778    }
779
780    tcg_out_cmp(s, cond, c1, c2, c2const);
781    if (cond == TCG_COND_LTU) {
782        if (neg) {
783            /* 0 - 0 - C = -C = (C ? -1 : 0) */
784            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_SUBC);
785        } else {
786            /* 0 + 0 + C =  C = (C ? 1 : 0) */
787            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
788        }
789    } else {
790        if (neg) {
791            /* 0 + -1 + C = C - 1 = (C ? 0 : -1) */
792            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_ADDC);
793        } else {
794            /* 0 - -1 - C = 1 - C = (C ? 0 : 1) */
795            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
796        }
797    }
798}
799
800static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
801                                TCGReg c1, int32_t c2, bool c2const, bool neg)
802{
803    int rcond;
804
805    if (use_vis3_instructions && !neg) {
806        switch (cond) {
807        case TCG_COND_NE:
808            if (c2 != 0) {
809                break;
810            }
811            c2 = c1, c2const = 0, c1 = TCG_REG_G0;
812            /* FALLTHRU */
813        case TCG_COND_LTU:
814            tcg_out_cmp(s, cond, c1, c2, c2const);
815            tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
816            return;
817        default:
818            break;
819        }
820    }
821
822    /* For 64-bit signed comparisons vs zero, we can avoid the compare
823       if the input does not overlap the output.  */
824    rcond = tcg_cond_to_rcond[cond];
825    if (c2 == 0 && rcond && c1 != ret) {
826        tcg_out_movi_s13(s, ret, 0);
827        tcg_out_movr(s, rcond, ret, c1, neg ? -1 : 1, 1);
828    } else {
829        tcg_out_cmp(s, cond, c1, c2, c2const);
830        tcg_out_movi_s13(s, ret, 0);
831        tcg_out_movcc(s, tcg_cond_to_bcond[cond],
832                      MOVCC_XCC, ret, neg ? -1 : 1, 1);
833    }
834}
835
836static void tcg_out_brcond(TCGContext *s, TCGType type, TCGCond cond,
837                           TCGReg arg1, TCGArg arg2, bool const_arg2,
838                           TCGLabel *l)
839{
840    if (type == TCG_TYPE_I32) {
841        tcg_out_brcond_i32(s, cond, arg1, arg2, const_arg2, l);
842    } else {
843        tcg_out_brcond_i64(s, cond, arg1, arg2, const_arg2, l);
844    }
845}
846
847static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
848                        TCGReg arg1, TCGReg arg2, TCGLabel *l)
849{
850    tcg_out_brcond(s, type, cond, arg1, arg2, false, l);
851}
852
853static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond cond,
854                         TCGReg arg1, tcg_target_long arg2, TCGLabel *l)
855{
856    tcg_out_brcond(s, type, cond, arg1, arg2, true, l);
857}
858
859static const TCGOutOpBrcond outop_brcond = {
860    .base.static_constraint = C_O0_I2(r, rJ),
861    .out_rr = tgen_brcond,
862    .out_ri = tgen_brcondi,
863};
864
865static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
866                            TCGReg ret, TCGReg c1,
867                            TCGArg c2, bool c2const, bool neg)
868{
869    if (type == TCG_TYPE_I32) {
870        tcg_out_setcond_i32(s, cond, ret, c1, c2, c2const, neg);
871    } else {
872        tcg_out_setcond_i64(s, cond, ret, c1, c2, c2const, neg);
873    }
874}
875
876static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
877                         TCGReg dest, TCGReg arg1, TCGReg arg2)
878{
879    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, false);
880}
881
882static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
883                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
884{
885    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, false);
886}
887
888static const TCGOutOpSetcond outop_setcond = {
889    .base.static_constraint = C_O1_I2(r, r, rJ),
890    .out_rrr = tgen_setcond,
891    .out_rri = tgen_setcondi,
892};
893
894static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
895                            TCGReg dest, TCGReg arg1, TCGReg arg2)
896{
897    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, true);
898}
899
900static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
901                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
902{
903    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, true);
904}
905
906static const TCGOutOpSetcond outop_negsetcond = {
907    .base.static_constraint = C_O1_I2(r, r, rJ),
908    .out_rrr = tgen_negsetcond,
909    .out_rri = tgen_negsetcondi,
910};
911
912static void tgen_movcond(TCGContext *s, TCGType type, TCGCond cond,
913                         TCGReg ret, TCGReg c1, TCGArg c2, bool c2const,
914                         TCGArg v1, bool v1const, TCGArg v2, bool v2consf)
915{
916    if (type == TCG_TYPE_I32) {
917        tcg_out_movcond_i32(s, cond, ret, c1, c2, c2const, v1, v1const);
918    } else {
919        tcg_out_movcond_i64(s, cond, ret, c1, c2, c2const, v1, v1const);
920    }
921}
922
923static const TCGOutOpMovcond outop_movcond = {
924    .base.static_constraint = C_O1_I4(r, r, rJ, rI, 0),
925    .out = tgen_movcond,
926};
927
928static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest,
929                               bool in_prologue, bool tail_call)
930{
931    uintptr_t desti = (uintptr_t)dest;
932
933    tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
934                     desti & ~0xfff, in_prologue, TCG_REG_T2);
935    tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7,
936                   TCG_REG_T1, desti & 0xfff, JMPL);
937}
938
939static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
940                                 bool in_prologue)
941{
942    ptrdiff_t disp = tcg_pcrel_diff(s, dest);
943
944    if (disp == (int32_t)disp) {
945        tcg_out32(s, CALL | (uint32_t)disp >> 2);
946    } else {
947        tcg_out_jmpl_const(s, dest, in_prologue, false);
948    }
949}
950
951static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
952                         const TCGHelperInfo *info)
953{
954    tcg_out_call_nodelay(s, dest, false);
955    tcg_out_nop(s);
956}
957
958static void tcg_out_mb(TCGContext *s, unsigned a0)
959{
960    /* Note that the TCG memory order constants mirror the Sparc MEMBAR.  */
961    tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
962}
963
964/* Generate global QEMU prologue and epilogue code */
965static void tcg_target_qemu_prologue(TCGContext *s)
966{
967    int tmp_buf_size, frame_size;
968
969    /*
970     * The TCG temp buffer is at the top of the frame, immediately
971     * below the frame pointer.  Use the logical (aligned) offset here;
972     * the stack bias is applied in temp_allocate_frame().
973     */
974    tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
975    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
976
977    /*
978     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
979     * otherwise the minimal frame usable by callees.
980     */
981    frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
982    frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
983    frame_size += TCG_TARGET_STACK_ALIGN - 1;
984    frame_size &= -TCG_TARGET_STACK_ALIGN;
985    tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
986              INSN_IMM13(-frame_size));
987
988#ifndef CONFIG_SOFTMMU
989    if (guest_base != 0) {
990        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG,
991                         guest_base, true, TCG_REG_T1);
992        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
993    }
994#endif
995
996    /* We choose TCG_REG_TB such that no move is required.  */
997    QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
998    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
999
1000    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
1001    /* delay slot */
1002    tcg_out_nop(s);
1003
1004    /* Epilogue for goto_ptr.  */
1005    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1006    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1007    /* delay slot */
1008    tcg_out_movi_s13(s, TCG_REG_O0, 0);
1009}
1010
1011static void tcg_out_tb_start(TCGContext *s)
1012{
1013    /* nothing to do */
1014}
1015
1016static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1017{
1018    int i;
1019    for (i = 0; i < count; ++i) {
1020        p[i] = NOP;
1021    }
1022}
1023
1024static const TCGLdstHelperParam ldst_helper_param = {
1025    .ntmp = 1, .tmp = { TCG_REG_T1 }
1026};
1027
1028static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1029{
1030    MemOp opc = get_memop(lb->oi);
1031    MemOp sgn;
1032
1033    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1034                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1035        return false;
1036    }
1037
1038    /* Use inline tcg_out_ext32s; otherwise let the helper sign-extend. */
1039    sgn = (opc & MO_SIZE) < MO_32 ? MO_SIGN : 0;
1040
1041    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1042    tcg_out_call(s, qemu_ld_helpers[opc & (MO_SIZE | sgn)], NULL);
1043    tcg_out_ld_helper_ret(s, lb, sgn, &ldst_helper_param);
1044
1045    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1046    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1047                       (intptr_t)lb->raddr, 0);
1048}
1049
1050static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1051{
1052    MemOp opc = get_memop(lb->oi);
1053
1054    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1055                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1056        return false;
1057    }
1058
1059    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1060    tcg_out_call(s, qemu_st_helpers[opc & MO_SIZE], NULL);
1061
1062    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1063    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1064                       (intptr_t)lb->raddr, 0);
1065}
1066
1067typedef struct {
1068    TCGReg base;
1069    TCGReg index;
1070    TCGAtomAlign aa;
1071} HostAddress;
1072
1073bool tcg_target_has_memory_bswap(MemOp memop)
1074{
1075    return true;
1076}
1077
1078/* We expect to use a 13-bit negative offset from ENV.  */
1079#define MIN_TLB_MASK_TABLE_OFS  -(1 << 12)
1080
1081/*
1082 * For system-mode, perform the TLB load and compare.
1083 * For user-mode, perform any required alignment tests.
1084 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1085 * is required and fill in @h with the host address for the fast path.
1086 */
1087static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1088                                           TCGReg addr_reg, MemOpIdx oi,
1089                                           bool is_ld)
1090{
1091    TCGType addr_type = s->addr_type;
1092    TCGLabelQemuLdst *ldst = NULL;
1093    MemOp opc = get_memop(oi);
1094    MemOp s_bits = opc & MO_SIZE;
1095    unsigned a_mask;
1096
1097    /* We don't support unaligned accesses. */
1098    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1099    h->aa.align = MAX(h->aa.align, s_bits);
1100    a_mask = (1u << h->aa.align) - 1;
1101
1102#ifdef CONFIG_SOFTMMU
1103    int mem_index = get_mmuidx(oi);
1104    int fast_off = tlb_mask_table_ofs(s, mem_index);
1105    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1106    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1107    int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1108                        : offsetof(CPUTLBEntry, addr_write);
1109    int add_off = offsetof(CPUTLBEntry, addend);
1110    int compare_mask;
1111    int cc;
1112
1113    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
1114    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T2, TCG_AREG0, mask_off);
1115    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T3, TCG_AREG0, table_off);
1116
1117    /* Extract the page index, shifted into place for tlb index.  */
1118    tcg_out_arithi(s, TCG_REG_T1, addr_reg,
1119                   s->page_bits - CPU_TLB_ENTRY_BITS, SHIFT_SRL);
1120    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T2, ARITH_AND);
1121
1122    /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2.  */
1123    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T3, ARITH_ADD);
1124
1125    /*
1126     * Load the tlb comparator and the addend.
1127     * Always load the entire 64-bit comparator for simplicity.
1128     * We will ignore the high bits via BPCC_ICC below.
1129     */
1130    tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_T2, TCG_REG_T1, cmp_off);
1131    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T1, TCG_REG_T1, add_off);
1132    h->base = TCG_REG_T1;
1133
1134    /* Mask out the page offset, except for the required alignment. */
1135    compare_mask = s->page_mask | a_mask;
1136    if (check_fit_tl(compare_mask, 13)) {
1137        tcg_out_arithi(s, TCG_REG_T3, addr_reg, compare_mask, ARITH_AND);
1138    } else {
1139        tcg_out_movi_s32(s, TCG_REG_T3, compare_mask);
1140        tcg_out_arith(s, TCG_REG_T3, addr_reg, TCG_REG_T3, ARITH_AND);
1141    }
1142    tcg_out_cmp(s, TCG_COND_NE, TCG_REG_T2, TCG_REG_T3, 0);
1143
1144    ldst = new_ldst_label(s);
1145    ldst->is_ld = is_ld;
1146    ldst->oi = oi;
1147    ldst->addr_reg = addr_reg;
1148    ldst->label_ptr[0] = s->code_ptr;
1149
1150    /* bne,pn %[xi]cc, label0 */
1151    cc = addr_type == TCG_TYPE_I32 ? BPCC_ICC : BPCC_XCC;
1152    tcg_out_bpcc0(s, COND_NE, BPCC_PN | cc, 0);
1153#else
1154    /*
1155     * If the size equals the required alignment, we can skip the test
1156     * and allow host SIGBUS to deliver SIGBUS to the guest.
1157     * Otherwise, test for at least natural alignment and defer
1158     * everything else to the helper functions.
1159     */
1160    if (s_bits != memop_alignment_bits(opc)) {
1161        tcg_debug_assert(check_fit_tl(a_mask, 13));
1162        tcg_out_arithi(s, TCG_REG_G0, addr_reg, a_mask, ARITH_ANDCC);
1163
1164        ldst = new_ldst_label(s);
1165        ldst->is_ld = is_ld;
1166        ldst->oi = oi;
1167        ldst->addr_reg = addr_reg;
1168        ldst->label_ptr[0] = s->code_ptr;
1169
1170        /* bne,pn %icc, label0 */
1171        tcg_out_bpcc0(s, COND_NE, BPCC_PN | BPCC_ICC, 0);
1172    }
1173    h->base = guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0;
1174#endif
1175
1176    /* If the guest address must be zero-extended, do in the delay slot.  */
1177    if (addr_type == TCG_TYPE_I32) {
1178        tcg_out_ext32u(s, TCG_REG_T2, addr_reg);
1179        h->index = TCG_REG_T2;
1180    } else {
1181        if (ldst) {
1182            tcg_out_nop(s);
1183        }
1184        h->index = addr_reg;
1185    }
1186    return ldst;
1187}
1188
1189static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
1190                            MemOpIdx oi, TCGType data_type)
1191{
1192    static const int ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = {
1193        [MO_UB]   = LDUB,
1194        [MO_SB]   = LDSB,
1195        [MO_UB | MO_LE] = LDUB,
1196        [MO_SB | MO_LE] = LDSB,
1197
1198        [MO_BEUW] = LDUH,
1199        [MO_BESW] = LDSH,
1200        [MO_BEUL] = LDUW,
1201        [MO_BESL] = LDSW,
1202        [MO_BEUQ] = LDX,
1203        [MO_BESQ] = LDX,
1204
1205        [MO_LEUW] = LDUH_LE,
1206        [MO_LESW] = LDSH_LE,
1207        [MO_LEUL] = LDUW_LE,
1208        [MO_LESL] = LDSW_LE,
1209        [MO_LEUQ] = LDX_LE,
1210        [MO_LESQ] = LDX_LE,
1211    };
1212
1213    TCGLabelQemuLdst *ldst;
1214    HostAddress h;
1215
1216    ldst = prepare_host_addr(s, &h, addr, oi, true);
1217
1218    tcg_out_ldst_rr(s, data, h.base, h.index,
1219                    ld_opc[get_memop(oi) & (MO_BSWAP | MO_SSIZE)]);
1220
1221    if (ldst) {
1222        ldst->type = data_type;
1223        ldst->datalo_reg = data;
1224        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1225    }
1226}
1227
1228static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr,
1229                            MemOpIdx oi, TCGType data_type)
1230{
1231    static const int st_opc[(MO_SIZE | MO_BSWAP) + 1] = {
1232        [MO_UB]   = STB,
1233
1234        [MO_BEUW] = STH,
1235        [MO_BEUL] = STW,
1236        [MO_BEUQ] = STX,
1237
1238        [MO_LEUW] = STH_LE,
1239        [MO_LEUL] = STW_LE,
1240        [MO_LEUQ] = STX_LE,
1241    };
1242
1243    TCGLabelQemuLdst *ldst;
1244    HostAddress h;
1245
1246    ldst = prepare_host_addr(s, &h, addr, oi, false);
1247
1248    tcg_out_ldst_rr(s, data, h.base, h.index,
1249                    st_opc[get_memop(oi) & (MO_BSWAP | MO_SIZE)]);
1250
1251    if (ldst) {
1252        ldst->type = data_type;
1253        ldst->datalo_reg = data;
1254        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1255    }
1256}
1257
1258static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1259{
1260    if (check_fit_ptr(a0, 13)) {
1261        tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1262        tcg_out_movi_s13(s, TCG_REG_O0, a0);
1263        return;
1264    } else {
1265        intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0);
1266        if (check_fit_ptr(tb_diff, 13)) {
1267            tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1268            /* Note that TCG_REG_TB has been unwound to O1.  */
1269            tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1270            return;
1271        }
1272    }
1273    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1274    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1275    tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1276}
1277
1278static void tcg_out_goto_tb(TCGContext *s, int which)
1279{
1280    ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which));
1281
1282    /* Load link and indirect branch. */
1283    set_jmp_insn_offset(s, which);
1284    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off);
1285    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1286    /* delay slot */
1287    tcg_out_nop(s);
1288    set_jmp_reset_offset(s, which);
1289
1290    /*
1291     * For the unlinked path of goto_tb, we need to reset TCG_REG_TB
1292     * to the beginning of this TB.
1293     */
1294    off = -tcg_current_code_size(s);
1295    if (check_fit_i32(off, 13)) {
1296        tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD);
1297    } else {
1298        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off);
1299        tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1300    }
1301}
1302
1303static void tcg_out_goto_ptr(TCGContext *s, TCGReg a0)
1304{
1305    tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1306    tcg_out_mov_delay(s, TCG_REG_TB, a0);
1307}
1308
1309void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1310                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1311{
1312}
1313
1314
1315static void tgen_add(TCGContext *s, TCGType type,
1316                     TCGReg a0, TCGReg a1, TCGReg a2)
1317{
1318    tcg_out_arith(s, a0, a1, a2, ARITH_ADD);
1319}
1320
1321static void tgen_addi(TCGContext *s, TCGType type,
1322                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1323{
1324    tcg_out_arithi(s, a0, a1, a2, ARITH_ADD);
1325}
1326
1327static const TCGOutOpBinary outop_add = {
1328    .base.static_constraint = C_O1_I2(r, r, rJ),
1329    .out_rrr = tgen_add,
1330    .out_rri = tgen_addi,
1331};
1332
1333static void tgen_addco_rrr(TCGContext *s, TCGType type,
1334                           TCGReg a0, TCGReg a1, TCGReg a2)
1335{
1336    tcg_out_arith(s, a0, a1, a2, ARITH_ADDCC);
1337}
1338
1339static void tgen_addco_rri(TCGContext *s, TCGType type,
1340                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1341{
1342    tcg_out_arithi(s, a0, a1, a2, ARITH_ADDCC);
1343}
1344
1345static const TCGOutOpBinary outop_addco = {
1346    .base.static_constraint = C_O1_I2(r, r, rJ),
1347    .out_rrr = tgen_addco_rrr,
1348    .out_rri = tgen_addco_rri,
1349};
1350
1351static void tgen_addci_rrr(TCGContext *s, TCGType type,
1352                           TCGReg a0, TCGReg a1, TCGReg a2)
1353{
1354    if (type == TCG_TYPE_I32) {
1355        tcg_out_arith(s, a0, a1, a2, ARITH_ADDC);
1356    } else if (use_vis3_instructions) {
1357        tcg_out_arith(s, a0, a1, a2, ARITH_ADDXC);
1358    } else {
1359        tcg_out_arith(s, TCG_REG_T1, a1, a2, ARITH_ADD);  /* for CC */
1360        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_ADD);  /* for CS */
1361        /* Select the correct result based on actual carry value. */
1362        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1363    }
1364}
1365
1366static void tgen_addci_rri(TCGContext *s, TCGType type,
1367                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1368{
1369    if (type == TCG_TYPE_I32) {
1370        tcg_out_arithi(s, a0, a1, a2, ARITH_ADDC);
1371        return;
1372    }
1373    /* !use_vis3_instructions */
1374    if (a2 != 0) {
1375        tcg_out_arithi(s, TCG_REG_T1, a1, a2, ARITH_ADD); /* for CC */
1376        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_ADD);  /* for CS */
1377        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1378    } else if (a0 == a1) {
1379        tcg_out_arithi(s, TCG_REG_T1, a1, 1, ARITH_ADD);
1380        tcg_out_movcc(s, COND_CS, MOVCC_XCC, a0, TCG_REG_T1, false);
1381    } else {
1382        tcg_out_arithi(s, a0, a1, 1, ARITH_ADD);
1383        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, a1, false);
1384    }
1385}
1386
1387static TCGConstraintSetIndex cset_addci(TCGType type, unsigned flags)
1388{
1389    if (use_vis3_instructions && type == TCG_TYPE_I64) {
1390        /* Note that ADDXC doesn't accept immediates.  */
1391        return C_O1_I2(r, rz, rz);
1392    }
1393    return C_O1_I2(r, rz, rJ);
1394}
1395
1396static const TCGOutOpAddSubCarry outop_addci = {
1397    .base.static_constraint = C_Dynamic,
1398    .base.dynamic_constraint = cset_addci,
1399    .out_rrr = tgen_addci_rrr,
1400    .out_rri = tgen_addci_rri,
1401};
1402
1403/* Copy %xcc.c to %icc.c */
1404static void tcg_out_dup_xcc_c(TCGContext *s)
1405{
1406    if (use_vis3_instructions) {
1407        tcg_out_arith(s, TCG_REG_T1, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
1408    } else {
1409        tcg_out_movi_s13(s, TCG_REG_T1, 0);
1410        tcg_out_movcc(s, COND_CS, MOVCC_XCC, TCG_REG_T1, 1, true);
1411    }
1412    /* Write carry-in into %icc via {0,1} + -1. */
1413    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_T1, -1, ARITH_ADDCC);
1414}
1415
1416static void tgen_addcio_rrr(TCGContext *s, TCGType type,
1417                            TCGReg a0, TCGReg a1, TCGReg a2)
1418{
1419    if (type != TCG_TYPE_I32) {
1420        if (use_vis3_instructions) {
1421            tcg_out_arith(s, a0, a1, a2, ARITH_ADDXCCC);
1422            return;
1423        }
1424        tcg_out_dup_xcc_c(s);
1425    }
1426    tcg_out_arith(s, a0, a1, a2, ARITH_ADDCCC);
1427}
1428
1429static void tgen_addcio_rri(TCGContext *s, TCGType type,
1430                            TCGReg a0, TCGReg a1, tcg_target_long a2)
1431{
1432    if (type != TCG_TYPE_I32) {
1433        /* !use_vis3_instructions */
1434        tcg_out_dup_xcc_c(s);
1435    }
1436    tcg_out_arithi(s, a0, a1, a2, ARITH_ADDCCC);
1437}
1438
1439static TCGConstraintSetIndex cset_addcio(TCGType type, unsigned flags)
1440{
1441    if (use_vis3_instructions && type == TCG_TYPE_I64) {
1442        /* Note that ADDXCCC doesn't accept immediates.  */
1443        return C_O1_I2(r, rz, rz);
1444    }
1445    return C_O1_I2(r, rz, rJ);
1446}
1447
1448static const TCGOutOpBinary outop_addcio = {
1449    .base.static_constraint = C_Dynamic,
1450    .base.dynamic_constraint = cset_addcio,
1451    .out_rrr = tgen_addcio_rrr,
1452    .out_rri = tgen_addcio_rri,
1453};
1454
1455static void tcg_out_set_carry(TCGContext *s)
1456{
1457    /* 0x11 -> xcc = nzvC, icc = nzvC */
1458    tcg_out_arithi(s, 0, TCG_REG_G0, 0x11, WRCCR);
1459}
1460
1461static void tgen_and(TCGContext *s, TCGType type,
1462                     TCGReg a0, TCGReg a1, TCGReg a2)
1463{
1464    tcg_out_arith(s, a0, a1, a2, ARITH_AND);
1465}
1466
1467static void tgen_andi(TCGContext *s, TCGType type,
1468                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1469{
1470    tcg_out_arithi(s, a0, a1, a2, ARITH_AND);
1471}
1472
1473static const TCGOutOpBinary outop_and = {
1474    .base.static_constraint = C_O1_I2(r, r, rJ),
1475    .out_rrr = tgen_and,
1476    .out_rri = tgen_andi,
1477};
1478
1479static void tgen_andc(TCGContext *s, TCGType type,
1480                      TCGReg a0, TCGReg a1, TCGReg a2)
1481{
1482    tcg_out_arith(s, a0, a1, a2, ARITH_ANDN);
1483}
1484
1485static const TCGOutOpBinary outop_andc = {
1486    .base.static_constraint = C_O1_I2(r, r, r),
1487    .out_rrr = tgen_andc,
1488};
1489
1490static const TCGOutOpBinary outop_clz = {
1491    .base.static_constraint = C_NotImplemented,
1492};
1493
1494static const TCGOutOpUnary outop_ctpop = {
1495    .base.static_constraint = C_NotImplemented,
1496};
1497
1498static const TCGOutOpBinary outop_ctz = {
1499    .base.static_constraint = C_NotImplemented,
1500};
1501
1502static void tgen_divs_rJ(TCGContext *s, TCGType type,
1503                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1504{
1505    uint32_t insn;
1506
1507    if (type == TCG_TYPE_I32) {
1508        /* Load Y with the sign extension of a1 to 64-bits.  */
1509        tcg_out_arithi(s, TCG_REG_T1, a1, 31, SHIFT_SRA);
1510        tcg_out_sety(s, TCG_REG_T1);
1511        insn = ARITH_SDIV;
1512    } else {
1513        insn = ARITH_SDIVX;
1514    }
1515    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1516}
1517
1518static void tgen_divs(TCGContext *s, TCGType type,
1519                      TCGReg a0, TCGReg a1, TCGReg a2)
1520{
1521    tgen_divs_rJ(s, type, a0, a1, a2, false);
1522}
1523
1524static void tgen_divsi(TCGContext *s, TCGType type,
1525                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1526{
1527    tgen_divs_rJ(s, type, a0, a1, a2, true);
1528}
1529
1530static const TCGOutOpBinary outop_divs = {
1531    .base.static_constraint = C_O1_I2(r, r, rJ),
1532    .out_rrr = tgen_divs,
1533    .out_rri = tgen_divsi,
1534};
1535
1536static const TCGOutOpDivRem outop_divs2 = {
1537    .base.static_constraint = C_NotImplemented,
1538};
1539
1540static void tgen_divu_rJ(TCGContext *s, TCGType type,
1541                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1542{
1543    uint32_t insn;
1544
1545    if (type == TCG_TYPE_I32) {
1546        /* Load Y with the zero extension to 64-bits.  */
1547        tcg_out_sety(s, TCG_REG_G0);
1548        insn = ARITH_UDIV;
1549    } else {
1550        insn = ARITH_UDIVX;
1551    }
1552    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1553}
1554
1555static void tgen_divu(TCGContext *s, TCGType type,
1556                      TCGReg a0, TCGReg a1, TCGReg a2)
1557{
1558    tgen_divu_rJ(s, type, a0, a1, a2, false);
1559}
1560
1561static void tgen_divui(TCGContext *s, TCGType type,
1562                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1563{
1564    tgen_divu_rJ(s, type, a0, a1, a2, true);
1565}
1566
1567static const TCGOutOpBinary outop_divu = {
1568    .base.static_constraint = C_O1_I2(r, r, rJ),
1569    .out_rrr = tgen_divu,
1570    .out_rri = tgen_divui,
1571};
1572
1573static const TCGOutOpDivRem outop_divu2 = {
1574    .base.static_constraint = C_NotImplemented,
1575};
1576
1577static const TCGOutOpBinary outop_eqv = {
1578    .base.static_constraint = C_NotImplemented,
1579};
1580
1581static void tgen_extrh_i64_i32(TCGContext *s, TCGType t, TCGReg a0, TCGReg a1)
1582{
1583    tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX);
1584}
1585
1586static const TCGOutOpUnary outop_extrh_i64_i32 = {
1587    .base.static_constraint = C_O1_I1(r, r),
1588    .out_rr = tgen_extrh_i64_i32,
1589};
1590
1591static void tgen_mul(TCGContext *s, TCGType type,
1592                     TCGReg a0, TCGReg a1, TCGReg a2)
1593{
1594    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1595    tcg_out_arith(s, a0, a1, a2, insn);
1596}
1597
1598static void tgen_muli(TCGContext *s, TCGType type,
1599                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1600{
1601    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1602    tcg_out_arithi(s, a0, a1, a2, insn);
1603}
1604
1605static const TCGOutOpBinary outop_mul = {
1606    .base.static_constraint = C_O1_I2(r, r, rJ),
1607    .out_rrr = tgen_mul,
1608    .out_rri = tgen_muli,
1609};
1610
1611/*
1612 * The 32-bit multiply insns produce a full 64-bit result.
1613 * Supporting 32-bit mul[us]2 opcodes avoids sign/zero-extensions
1614 * before the actual multiply; we only need extract the high part
1615 * into the separate operand.
1616 */
1617static TCGConstraintSetIndex cset_mul2(TCGType type, unsigned flags)
1618{
1619    return type == TCG_TYPE_I32 ? C_O2_I2(r, r, r, r) : C_NotImplemented;
1620}
1621
1622static void tgen_muls2(TCGContext *s, TCGType type,
1623                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1624{
1625    tcg_out_arith(s, a0, a2, a3, ARITH_SMUL);
1626    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1627}
1628
1629static const TCGOutOpMul2 outop_muls2 = {
1630    .base.static_constraint = C_Dynamic,
1631    .base.dynamic_constraint = cset_mul2,
1632    .out_rrrr = tgen_muls2,
1633};
1634
1635static const TCGOutOpBinary outop_mulsh = {
1636    .base.static_constraint = C_NotImplemented,
1637};
1638
1639static void tgen_mulu2(TCGContext *s, TCGType type,
1640                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1641{
1642    tcg_out_arith(s, a0, a2, a3, ARITH_UMUL);
1643    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1644}
1645
1646static const TCGOutOpMul2 outop_mulu2 = {
1647    .base.static_constraint = C_Dynamic,
1648    .base.dynamic_constraint = cset_mul2,
1649    .out_rrrr = tgen_mulu2,
1650};
1651
1652static void tgen_muluh(TCGContext *s, TCGType type,
1653                       TCGReg a0, TCGReg a1, TCGReg a2)
1654{
1655    tcg_out_arith(s, a0, a1, a2, ARITH_UMULXHI);
1656}
1657
1658static TCGConstraintSetIndex cset_muluh(TCGType type, unsigned flags)
1659{
1660    return (type == TCG_TYPE_I64 && use_vis3_instructions
1661            ? C_O1_I2(r, r, r) : C_NotImplemented);
1662}
1663
1664static const TCGOutOpBinary outop_muluh = {
1665    .base.static_constraint = C_Dynamic,
1666    .base.dynamic_constraint = cset_muluh,
1667    .out_rrr = tgen_muluh,
1668};
1669
1670static const TCGOutOpBinary outop_nand = {
1671    .base.static_constraint = C_NotImplemented,
1672};
1673
1674static const TCGOutOpBinary outop_nor = {
1675    .base.static_constraint = C_NotImplemented,
1676};
1677
1678static void tgen_or(TCGContext *s, TCGType type,
1679                     TCGReg a0, TCGReg a1, TCGReg a2)
1680{
1681    tcg_out_arith(s, a0, a1, a2, ARITH_OR);
1682}
1683
1684static void tgen_ori(TCGContext *s, TCGType type,
1685                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1686{
1687    tcg_out_arithi(s, a0, a1, a2, ARITH_OR);
1688}
1689
1690static const TCGOutOpBinary outop_or = {
1691    .base.static_constraint = C_O1_I2(r, r, rJ),
1692    .out_rrr = tgen_or,
1693    .out_rri = tgen_ori,
1694};
1695
1696static void tgen_orc(TCGContext *s, TCGType type,
1697                     TCGReg a0, TCGReg a1, TCGReg a2)
1698{
1699    tcg_out_arith(s, a0, a1, a2, ARITH_ORN);
1700}
1701
1702static const TCGOutOpBinary outop_orc = {
1703    .base.static_constraint = C_O1_I2(r, r, r),
1704    .out_rrr = tgen_orc,
1705};
1706
1707static const TCGOutOpBinary outop_rems = {
1708    .base.static_constraint = C_NotImplemented,
1709};
1710
1711static const TCGOutOpBinary outop_remu = {
1712    .base.static_constraint = C_NotImplemented,
1713};
1714
1715static const TCGOutOpBinary outop_rotl = {
1716    .base.static_constraint = C_NotImplemented,
1717};
1718
1719static const TCGOutOpBinary outop_rotr = {
1720    .base.static_constraint = C_NotImplemented,
1721};
1722
1723static void tgen_sar(TCGContext *s, TCGType type,
1724                     TCGReg a0, TCGReg a1, TCGReg a2)
1725{
1726    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1727    tcg_out_arith(s, a0, a1, a2, insn);
1728}
1729
1730static void tgen_sari(TCGContext *s, TCGType type,
1731                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1732{
1733    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1734    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1735    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1736}
1737
1738static const TCGOutOpBinary outop_sar = {
1739    .base.static_constraint = C_O1_I2(r, r, rJ),
1740    .out_rrr = tgen_sar,
1741    .out_rri = tgen_sari,
1742};
1743
1744static void tgen_shl(TCGContext *s, TCGType type,
1745                     TCGReg a0, TCGReg a1, TCGReg a2)
1746{
1747    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1748    tcg_out_arith(s, a0, a1, a2, insn);
1749}
1750
1751static void tgen_shli(TCGContext *s, TCGType type,
1752                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1753{
1754    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1755    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1756    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1757}
1758
1759static const TCGOutOpBinary outop_shl = {
1760    .base.static_constraint = C_O1_I2(r, r, rJ),
1761    .out_rrr = tgen_shl,
1762    .out_rri = tgen_shli,
1763};
1764
1765static void tgen_shr(TCGContext *s, TCGType type,
1766                     TCGReg a0, TCGReg a1, TCGReg a2)
1767{
1768    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1769    tcg_out_arith(s, a0, a1, a2, insn);
1770}
1771
1772static void tgen_shri(TCGContext *s, TCGType type,
1773                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1774{
1775    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1776    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1777    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1778}
1779
1780static const TCGOutOpBinary outop_shr = {
1781    .base.static_constraint = C_O1_I2(r, r, rJ),
1782    .out_rrr = tgen_shr,
1783    .out_rri = tgen_shri,
1784};
1785
1786static void tgen_sub(TCGContext *s, TCGType type,
1787                     TCGReg a0, TCGReg a1, TCGReg a2)
1788{
1789    tcg_out_arith(s, a0, a1, a2, ARITH_SUB);
1790}
1791
1792static const TCGOutOpSubtract outop_sub = {
1793    .base.static_constraint = C_O1_I2(r, r, r),
1794    .out_rrr = tgen_sub,
1795};
1796
1797static void tgen_subbo_rrr(TCGContext *s, TCGType type,
1798                           TCGReg a0, TCGReg a1, TCGReg a2)
1799{
1800    tcg_out_arith(s, a0, a1, a2, ARITH_SUBCC);
1801}
1802
1803static void tgen_subbo_rri(TCGContext *s, TCGType type,
1804                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1805{
1806    tcg_out_arithi(s, a0, a1, a2, ARITH_SUBCC);
1807}
1808
1809static const TCGOutOpAddSubCarry outop_subbo = {
1810    .base.static_constraint = C_O1_I2(r, rz, rJ),
1811    .out_rrr = tgen_subbo_rrr,
1812    .out_rri = tgen_subbo_rri,
1813};
1814
1815static void tgen_subbi_rrr(TCGContext *s, TCGType type,
1816                           TCGReg a0, TCGReg a1, TCGReg a2)
1817{
1818    /* TODO: OSA 2015 added SUBXC */
1819    if (type == TCG_TYPE_I32) {
1820        tcg_out_arith(s, a0, a1, a2, ARITH_SUBC);
1821    } else {
1822        tcg_out_arith(s, TCG_REG_T1, a1, a2, ARITH_SUB);  /* for CC */
1823        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_SUB);  /* for CS */
1824        /* Select the correct result based on actual borrow value. */
1825        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1826    }
1827}
1828
1829static void tgen_subbi_rri(TCGContext *s, TCGType type,
1830                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1831{
1832    if (type == TCG_TYPE_I32) {
1833        tcg_out_arithi(s, a0, a1, a2, ARITH_SUBC);
1834    } else if (a2 != 0) {
1835        tcg_out_arithi(s, TCG_REG_T1, a1, a2, ARITH_SUB);  /* for CC */
1836        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_SUB);   /* for CS */
1837        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1838    } else if (a0 == a1) {
1839        tcg_out_arithi(s, TCG_REG_T1, a1, 1, ARITH_SUB);
1840        tcg_out_movcc(s, COND_CS, MOVCC_XCC, a0, TCG_REG_T1, false);
1841    } else {
1842        tcg_out_arithi(s, a0, a1, 1, ARITH_SUB);
1843        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, a1, false);
1844    }
1845}
1846
1847static const TCGOutOpAddSubCarry outop_subbi = {
1848    .base.static_constraint = C_O1_I2(r, rz, rJ),
1849    .out_rrr = tgen_subbi_rrr,
1850    .out_rri = tgen_subbi_rri,
1851};
1852
1853static void tgen_subbio_rrr(TCGContext *s, TCGType type,
1854                            TCGReg a0, TCGReg a1, TCGReg a2)
1855{
1856    if (type != TCG_TYPE_I32) {
1857        /* TODO: OSA 2015 added SUBXCCC */
1858        tcg_out_dup_xcc_c(s);
1859    }
1860    tcg_out_arith(s, a0, a1, a2, ARITH_SUBCCC);
1861}
1862
1863static void tgen_subbio_rri(TCGContext *s, TCGType type,
1864                            TCGReg a0, TCGReg a1, tcg_target_long a2)
1865{
1866    if (type != TCG_TYPE_I32) {
1867        tcg_out_dup_xcc_c(s);
1868    }
1869    tcg_out_arithi(s, a0, a1, a2, ARITH_SUBCCC);
1870}
1871
1872static const TCGOutOpAddSubCarry outop_subbio = {
1873    .base.static_constraint = C_O1_I2(r, rz, rJ),
1874    .out_rrr = tgen_subbio_rrr,
1875    .out_rri = tgen_subbio_rri,
1876};
1877
1878static void tcg_out_set_borrow(TCGContext *s)
1879{
1880    tcg_out_set_carry(s);  /* borrow == carry */
1881}
1882
1883static void tgen_xor(TCGContext *s, TCGType type,
1884                     TCGReg a0, TCGReg a1, TCGReg a2)
1885{
1886    tcg_out_arith(s, a0, a1, a2, ARITH_XOR);
1887}
1888
1889static void tgen_xori(TCGContext *s, TCGType type,
1890                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1891{
1892    tcg_out_arithi(s, a0, a1, a2, ARITH_XOR);
1893}
1894
1895static const TCGOutOpBinary outop_xor = {
1896    .base.static_constraint = C_O1_I2(r, r, rJ),
1897    .out_rrr = tgen_xor,
1898    .out_rri = tgen_xori,
1899};
1900
1901static const TCGOutOpBswap outop_bswap16 = {
1902    .base.static_constraint = C_NotImplemented,
1903};
1904
1905static const TCGOutOpBswap outop_bswap32 = {
1906    .base.static_constraint = C_NotImplemented,
1907};
1908
1909static const TCGOutOpUnary outop_bswap64 = {
1910    .base.static_constraint = C_NotImplemented,
1911};
1912
1913static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1914{
1915     tgen_sub(s, type, a0, TCG_REG_G0, a1);
1916}
1917
1918static const TCGOutOpUnary outop_neg = {
1919    .base.static_constraint = C_O1_I1(r, r),
1920    .out_rr = tgen_neg,
1921};
1922
1923static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1924{
1925     tgen_orc(s, type, a0, TCG_REG_G0, a1);
1926}
1927
1928static const TCGOutOpUnary outop_not = {
1929    .base.static_constraint = C_O1_I1(r, r),
1930    .out_rr = tgen_not,
1931};
1932
1933static const TCGOutOpDeposit outop_deposit = {
1934    .base.static_constraint = C_NotImplemented,
1935};
1936
1937static void tgen_extract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
1938                         unsigned ofs, unsigned len)
1939{
1940    tcg_debug_assert(ofs + len == 32);
1941    tcg_out_arithi(s, a0, a1, ofs, SHIFT_SRL);
1942}
1943
1944static const TCGOutOpExtract outop_extract = {
1945    .base.static_constraint = C_O1_I1(r, r),
1946    .out_rr = tgen_extract,
1947};
1948
1949static void tgen_sextract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
1950                          unsigned ofs, unsigned len)
1951{
1952    tcg_debug_assert(ofs + len == 32);
1953    tcg_out_arithi(s, a0, a1, ofs, SHIFT_SRA);
1954}
1955
1956static const TCGOutOpExtract outop_sextract = {
1957    .base.static_constraint = C_O1_I1(r, r),
1958    .out_rr = tgen_sextract,
1959};
1960
1961static const TCGOutOpExtract2 outop_extract2 = {
1962    .base.static_constraint = C_NotImplemented,
1963};
1964
1965static void tgen_ld8u(TCGContext *s, TCGType type, TCGReg dest,
1966                      TCGReg base, ptrdiff_t offset)
1967{
1968    tcg_out_ldst(s, dest, base, offset, LDUB);
1969}
1970
1971static const TCGOutOpLoad outop_ld8u = {
1972    .base.static_constraint = C_O1_I1(r, r),
1973    .out = tgen_ld8u,
1974};
1975
1976static void tgen_ld8s(TCGContext *s, TCGType type, TCGReg dest,
1977                      TCGReg base, ptrdiff_t offset)
1978{
1979    tcg_out_ldst(s, dest, base, offset, LDSB);
1980}
1981
1982static const TCGOutOpLoad outop_ld8s = {
1983    .base.static_constraint = C_O1_I1(r, r),
1984    .out = tgen_ld8s,
1985};
1986
1987static void tgen_ld16u(TCGContext *s, TCGType type, TCGReg dest,
1988                       TCGReg base, ptrdiff_t offset)
1989{
1990    tcg_out_ldst(s, dest, base, offset, LDUH);
1991}
1992
1993static const TCGOutOpLoad outop_ld16u = {
1994    .base.static_constraint = C_O1_I1(r, r),
1995    .out = tgen_ld16u,
1996};
1997
1998static void tgen_ld16s(TCGContext *s, TCGType type, TCGReg dest,
1999                       TCGReg base, ptrdiff_t offset)
2000{
2001    tcg_out_ldst(s, dest, base, offset, LDSH);
2002}
2003
2004static const TCGOutOpLoad outop_ld16s = {
2005    .base.static_constraint = C_O1_I1(r, r),
2006    .out = tgen_ld16s,
2007};
2008
2009static void tgen_ld32u(TCGContext *s, TCGType type, TCGReg dest,
2010                       TCGReg base, ptrdiff_t offset)
2011{
2012    tcg_out_ldst(s, dest, base, offset, LDUW);
2013}
2014
2015static const TCGOutOpLoad outop_ld32u = {
2016    .base.static_constraint = C_O1_I1(r, r),
2017    .out = tgen_ld32u,
2018};
2019
2020static void tgen_ld32s(TCGContext *s, TCGType type, TCGReg dest,
2021                       TCGReg base, ptrdiff_t offset)
2022{
2023    tcg_out_ldst(s, dest, base, offset, LDSW);
2024}
2025
2026static const TCGOutOpLoad outop_ld32s = {
2027    .base.static_constraint = C_O1_I1(r, r),
2028    .out = tgen_ld32s,
2029};
2030
2031static void tgen_st8_r(TCGContext *s, TCGType type, TCGReg data,
2032                       TCGReg base, ptrdiff_t offset)
2033{
2034    tcg_out_ldst(s, data, base, offset, STB);
2035}
2036
2037static const TCGOutOpStore outop_st8 = {
2038    .base.static_constraint = C_O0_I2(rz, r),
2039    .out_r = tgen_st8_r,
2040};
2041
2042static void tgen_st16_r(TCGContext *s, TCGType type, TCGReg data,
2043                       TCGReg base, ptrdiff_t offset)
2044{
2045    tcg_out_ldst(s, data, base, offset, STH);
2046}
2047
2048static const TCGOutOpStore outop_st16 = {
2049    .base.static_constraint = C_O0_I2(rz, r),
2050    .out_r = tgen_st16_r,
2051};
2052
2053static const TCGOutOpStore outop_st = {
2054    .base.static_constraint = C_O0_I2(rz, r),
2055    .out_r = tcg_out_st,
2056};
2057
2058
2059static void tcg_out_op(TCGContext *s, TCGOpcode opc, TCGType type,
2060                       const TCGArg args[TCG_MAX_OP_ARGS],
2061                       const int const_args[TCG_MAX_OP_ARGS])
2062{
2063    TCGArg a0, a1, a2;
2064
2065    /* Hoist the loads of the most common arguments.  */
2066    a0 = args[0];
2067    a1 = args[1];
2068    a2 = args[2];
2069
2070    switch (opc) {
2071    case INDEX_op_qemu_ld_i32:
2072        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32);
2073        break;
2074    case INDEX_op_qemu_ld_i64:
2075        tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64);
2076        break;
2077    case INDEX_op_qemu_st_i32:
2078        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32);
2079        break;
2080    case INDEX_op_qemu_st_i64:
2081        tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64);
2082        break;
2083
2084    case INDEX_op_call:     /* Always emitted via tcg_out_call.  */
2085    case INDEX_op_exit_tb:  /* Always emitted via tcg_out_exit_tb.  */
2086    case INDEX_op_goto_tb:  /* Always emitted via tcg_out_goto_tb.  */
2087    default:
2088        g_assert_not_reached();
2089    }
2090}
2091
2092static TCGConstraintSetIndex
2093tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2094{
2095    switch (op) {
2096    case INDEX_op_qemu_ld_i32:
2097    case INDEX_op_qemu_ld_i64:
2098        return C_O1_I1(r, r);
2099
2100    case INDEX_op_qemu_st_i32:
2101    case INDEX_op_qemu_st_i64:
2102        return C_O0_I2(rz, r);
2103
2104    default:
2105        return C_NotImplemented;
2106    }
2107}
2108
2109static void tcg_target_init(TCGContext *s)
2110{
2111    /*
2112     * Only probe for the platform and capabilities if we haven't already
2113     * determined maximum values at compile time.
2114     */
2115#ifndef use_vis3_instructions
2116    {
2117        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2118        use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
2119    }
2120#endif
2121
2122    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2123    tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;
2124
2125    tcg_target_call_clobber_regs = 0;
2126    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
2127    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
2128    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
2129    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
2130    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
2131    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
2132    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
2133    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
2134    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
2135    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
2136    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
2137    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
2138    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
2139    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
2140    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
2141
2142    s->reserved_regs = 0;
2143    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
2144    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
2145    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
2146    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
2147    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
2148    tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
2149    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
2150    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
2151    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T3); /* for internal use */
2152}
2153
2154#define ELF_HOST_MACHINE  EM_SPARCV9
2155
2156typedef struct {
2157    DebugFrameHeader h;
2158    uint8_t fde_def_cfa[4];
2159    uint8_t fde_win_save;
2160    uint8_t fde_ret_save[3];
2161} DebugFrame;
2162
2163static const DebugFrame debug_frame = {
2164    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2165    .h.cie.id = -1,
2166    .h.cie.version = 1,
2167    .h.cie.code_align = 1,
2168    .h.cie.data_align = -sizeof(void *) & 0x7f,
2169    .h.cie.return_column = 15,            /* o7 */
2170
2171    /* Total FDE size does not include the "len" member.  */
2172    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2173
2174    .fde_def_cfa = {
2175        12, 30,                         /* DW_CFA_def_cfa i6, 2047 */
2176        (2047 & 0x7f) | 0x80, (2047 >> 7)
2177    },
2178    .fde_win_save = 0x2d,               /* DW_CFA_GNU_window_save */
2179    .fde_ret_save = { 9, 15, 31 },      /* DW_CFA_register o7, i7 */
2180};
2181
2182void tcg_register_jit(const void *buf, size_t buf_size)
2183{
2184    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2185}
2186