xref: /openbmc/qemu/tcg/sparc64/tcg-target.c.inc (revision 641f1c53862aec64810c0b93b5b1de49d55fda92)
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_POPC (INSN_OP(2) | INSN_OP3(0x2e))
214#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
215
216#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
217#define ARITH_ADDXCCC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x13))
218#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
219
220#define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
221#define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
222#define SHIFT_SRA  (INSN_OP(2) | INSN_OP3(0x27))
223
224#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
225#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
226#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
227
228#define RDY        (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
229#define WRY        (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
230#define WRCCR      (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(2))
231#define JMPL       (INSN_OP(2) | INSN_OP3(0x38))
232#define RETURN     (INSN_OP(2) | INSN_OP3(0x39))
233#define SAVE       (INSN_OP(2) | INSN_OP3(0x3c))
234#define RESTORE    (INSN_OP(2) | INSN_OP3(0x3d))
235#define SETHI      (INSN_OP(0) | INSN_OP2(0x4))
236#define CALL       INSN_OP(1)
237#define LDUB       (INSN_OP(3) | INSN_OP3(0x01))
238#define LDSB       (INSN_OP(3) | INSN_OP3(0x09))
239#define LDUH       (INSN_OP(3) | INSN_OP3(0x02))
240#define LDSH       (INSN_OP(3) | INSN_OP3(0x0a))
241#define LDUW       (INSN_OP(3) | INSN_OP3(0x00))
242#define LDSW       (INSN_OP(3) | INSN_OP3(0x08))
243#define LDX        (INSN_OP(3) | INSN_OP3(0x0b))
244#define STB        (INSN_OP(3) | INSN_OP3(0x05))
245#define STH        (INSN_OP(3) | INSN_OP3(0x06))
246#define STW        (INSN_OP(3) | INSN_OP3(0x04))
247#define STX        (INSN_OP(3) | INSN_OP3(0x0e))
248#define LDUBA      (INSN_OP(3) | INSN_OP3(0x11))
249#define LDSBA      (INSN_OP(3) | INSN_OP3(0x19))
250#define LDUHA      (INSN_OP(3) | INSN_OP3(0x12))
251#define LDSHA      (INSN_OP(3) | INSN_OP3(0x1a))
252#define LDUWA      (INSN_OP(3) | INSN_OP3(0x10))
253#define LDSWA      (INSN_OP(3) | INSN_OP3(0x18))
254#define LDXA       (INSN_OP(3) | INSN_OP3(0x1b))
255#define STBA       (INSN_OP(3) | INSN_OP3(0x15))
256#define STHA       (INSN_OP(3) | INSN_OP3(0x16))
257#define STWA       (INSN_OP(3) | INSN_OP3(0x14))
258#define STXA       (INSN_OP(3) | INSN_OP3(0x1e))
259
260#define MEMBAR     (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
261
262#define NOP        (SETHI | INSN_RD(TCG_REG_G0) | 0)
263
264#ifndef ASI_PRIMARY_LITTLE
265#define ASI_PRIMARY_LITTLE 0x88
266#endif
267
268#define LDUH_LE    (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
269#define LDSH_LE    (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
270#define LDUW_LE    (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
271#define LDSW_LE    (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
272#define LDX_LE     (LDXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
273
274#define STH_LE     (STHA  | INSN_ASI(ASI_PRIMARY_LITTLE))
275#define STW_LE     (STWA  | INSN_ASI(ASI_PRIMARY_LITTLE))
276#define STX_LE     (STXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
277
278static bool use_popc_instructions;
279#if defined(__VIS__) && __VIS__ >= 0x300
280#define use_vis3_instructions  1
281#else
282static bool use_vis3_instructions;
283#endif
284
285static bool check_fit_i64(int64_t val, unsigned int bits)
286{
287    return val == sextract64(val, 0, bits);
288}
289
290static bool check_fit_i32(int32_t val, unsigned int bits)
291{
292    return val == sextract32(val, 0, bits);
293}
294
295#define check_fit_tl    check_fit_i64
296#define check_fit_ptr   check_fit_i64
297
298static bool patch_reloc(tcg_insn_unit *src_rw, int type,
299                        intptr_t value, intptr_t addend)
300{
301    const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
302    uint32_t insn = *src_rw;
303    intptr_t pcrel;
304
305    value += addend;
306    pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx);
307
308    switch (type) {
309    case R_SPARC_WDISP16:
310        if (!check_fit_ptr(pcrel >> 2, 16)) {
311            return false;
312        }
313        insn &= ~INSN_OFF16(-1);
314        insn |= INSN_OFF16(pcrel);
315        break;
316    case R_SPARC_WDISP19:
317        if (!check_fit_ptr(pcrel >> 2, 19)) {
318            return false;
319        }
320        insn &= ~INSN_OFF19(-1);
321        insn |= INSN_OFF19(pcrel);
322        break;
323    case R_SPARC_13:
324        if (!check_fit_ptr(value, 13)) {
325            return false;
326        }
327        insn &= ~INSN_IMM13(-1);
328        insn |= INSN_IMM13(value);
329        break;
330    default:
331        g_assert_not_reached();
332    }
333
334    *src_rw = insn;
335    return true;
336}
337
338/* test if a constant matches the constraint */
339static bool tcg_target_const_match(int64_t val, int ct,
340                                   TCGType type, TCGCond cond, int vece)
341{
342    if (ct & TCG_CT_CONST) {
343        return 1;
344    }
345
346    if (type == TCG_TYPE_I32) {
347        val = (int32_t)val;
348    }
349
350    if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
351        return 1;
352    } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
353        return 1;
354    } else {
355        return 0;
356    }
357}
358
359static void tcg_out_nop(TCGContext *s)
360{
361    tcg_out32(s, NOP);
362}
363
364static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
365                          TCGReg rs2, int op)
366{
367    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
368}
369
370static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
371                           int32_t offset, int op)
372{
373    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
374}
375
376static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
377                           int32_t val2, int val2const, int op)
378{
379    tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
380              | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
381}
382
383static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
384{
385    if (ret != arg) {
386        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
387    }
388    return true;
389}
390
391static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg)
392{
393    if (ret != arg) {
394        tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
395    } else {
396        tcg_out_nop(s);
397    }
398}
399
400static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
401{
402    tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
403}
404
405/* A 13-bit constant sign-extended to 64 bits.  */
406static void tcg_out_movi_s13(TCGContext *s, TCGReg ret, int32_t arg)
407{
408    tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
409}
410
411/* A 32-bit constant sign-extended to 64 bits.  */
412static void tcg_out_movi_s32(TCGContext *s, TCGReg ret, int32_t arg)
413{
414    tcg_out_sethi(s, ret, ~arg);
415    tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
416}
417
418/* A 32-bit constant zero-extended to 64 bits.  */
419static void tcg_out_movi_u32(TCGContext *s, TCGReg ret, uint32_t arg)
420{
421    tcg_out_sethi(s, ret, arg);
422    if (arg & 0x3ff) {
423        tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
424    }
425}
426
427static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
428                             tcg_target_long arg, bool in_prologue,
429                             TCGReg scratch)
430{
431    tcg_target_long hi, lo = (int32_t)arg;
432    tcg_target_long test, lsb;
433
434    /* A 13-bit constant sign-extended to 64-bits.  */
435    if (check_fit_tl(arg, 13)) {
436        tcg_out_movi_s13(s, ret, arg);
437        return;
438    }
439
440    /* A 32-bit constant, or 32-bit zero-extended to 64-bits.  */
441    if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
442        tcg_out_movi_u32(s, ret, arg);
443        return;
444    }
445
446    /* A 13-bit constant relative to the TB.  */
447    if (!in_prologue) {
448        test = tcg_tbrel_diff(s, (void *)arg);
449        if (check_fit_ptr(test, 13)) {
450            tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
451            return;
452        }
453    }
454
455    /* A 32-bit constant sign-extended to 64-bits.  */
456    if (arg == lo) {
457        tcg_out_movi_s32(s, ret, arg);
458        return;
459    }
460
461    /* A 32-bit constant, shifted.  */
462    lsb = ctz64(arg);
463    test = (tcg_target_long)arg >> lsb;
464    if (lsb > 10 && test == extract64(test, 0, 21)) {
465        tcg_out_sethi(s, ret, test << 10);
466        tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
467        return;
468    } else if (test == (uint32_t)test || test == (int32_t)test) {
469        tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch);
470        tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
471        return;
472    }
473
474    /* Use the constant pool, if possible. */
475    if (!in_prologue) {
476        new_pool_label(s, arg, R_SPARC_13, s->code_ptr,
477                       tcg_tbrel_diff(s, NULL));
478        tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB));
479        return;
480    }
481
482    /* A 64-bit constant decomposed into 2 32-bit pieces.  */
483    if (check_fit_i32(lo, 13)) {
484        hi = (arg - lo) >> 32;
485        tcg_out_movi_u32(s, ret, hi);
486        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
487        tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
488    } else {
489        hi = arg >> 32;
490        tcg_out_movi_u32(s, ret, hi);
491        tcg_out_movi_u32(s, scratch, lo);
492        tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
493        tcg_out_arith(s, ret, ret, scratch, ARITH_OR);
494    }
495}
496
497static void tcg_out_movi(TCGContext *s, TCGType type,
498                         TCGReg ret, tcg_target_long arg)
499{
500    tcg_debug_assert(ret != TCG_REG_T3);
501    tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T3);
502}
503
504static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
505{
506    g_assert_not_reached();
507}
508
509static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
510{
511    g_assert_not_reached();
512}
513
514static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs)
515{
516    tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND);
517}
518
519static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs)
520{
521    tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL);
522    tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL);
523}
524
525static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs)
526{
527    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA);
528}
529
530static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs)
531{
532    tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL);
533}
534
535static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
536{
537    tcg_out_ext32s(s, rd, rs);
538}
539
540static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
541{
542    tcg_out_ext32u(s, rd, rs);
543}
544
545static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs)
546{
547    tcg_out_ext32u(s, rd, rs);
548}
549
550static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
551{
552    return false;
553}
554
555static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
556                             tcg_target_long imm)
557{
558    /* This function is only used for passing structs by reference. */
559    g_assert_not_reached();
560}
561
562static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
563                            TCGReg a2, int op)
564{
565    tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
566}
567
568static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
569                         intptr_t offset, int op)
570{
571    if (check_fit_ptr(offset, 13)) {
572        tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
573                  INSN_IMM13(offset));
574    } else {
575        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
576        tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
577    }
578}
579
580static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
581                       TCGReg arg1, intptr_t arg2)
582{
583    tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
584}
585
586static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
587                       TCGReg arg1, intptr_t arg2)
588{
589    tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
590}
591
592static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
593                        TCGReg base, intptr_t ofs)
594{
595    if (val == 0) {
596        tcg_out_st(s, type, TCG_REG_G0, base, ofs);
597        return true;
598    }
599    return false;
600}
601
602static void tcg_out_sety(TCGContext *s, TCGReg rs)
603{
604    tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
605}
606
607static const uint8_t tcg_cond_to_bcond[16] = {
608    [TCG_COND_EQ] = COND_E,
609    [TCG_COND_NE] = COND_NE,
610    [TCG_COND_TSTEQ] = COND_E,
611    [TCG_COND_TSTNE] = COND_NE,
612    [TCG_COND_LT] = COND_L,
613    [TCG_COND_GE] = COND_GE,
614    [TCG_COND_LE] = COND_LE,
615    [TCG_COND_GT] = COND_G,
616    [TCG_COND_LTU] = COND_CS,
617    [TCG_COND_GEU] = COND_CC,
618    [TCG_COND_LEU] = COND_LEU,
619    [TCG_COND_GTU] = COND_GU,
620};
621
622static const uint8_t tcg_cond_to_rcond[16] = {
623    [TCG_COND_EQ] = RCOND_Z,
624    [TCG_COND_NE] = RCOND_NZ,
625    [TCG_COND_LT] = RCOND_LZ,
626    [TCG_COND_GT] = RCOND_GZ,
627    [TCG_COND_LE] = RCOND_LEZ,
628    [TCG_COND_GE] = RCOND_GEZ
629};
630
631static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
632{
633    tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
634}
635
636static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
637{
638    int off19 = 0;
639
640    if (l->has_value) {
641        off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
642    } else {
643        tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
644    }
645    tcg_out_bpcc0(s, scond, flags, off19);
646}
647
648static void tcg_out_br(TCGContext *s, TCGLabel *l)
649{
650    tcg_out_bpcc(s, COND_A, BPCC_PT, l);
651    tcg_out_nop(s);
652}
653
654static void tcg_out_cmp(TCGContext *s, TCGCond cond,
655                        TCGReg c1, int32_t c2, int c2const)
656{
657    tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const,
658                   is_tst_cond(cond) ? ARITH_ANDCC : ARITH_SUBCC);
659}
660
661static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
662                               int32_t arg2, int const_arg2, TCGLabel *l)
663{
664    tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
665    tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
666    tcg_out_nop(s);
667}
668
669static void tcg_out_movcc(TCGContext *s, int scond, int cc, TCGReg ret,
670                          int32_t v1, int v1const)
671{
672    tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret) | INSN_RS1(scond)
673              | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
674}
675
676static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
677                                TCGReg c1, int32_t c2, int c2const,
678                                int32_t v1, int v1const)
679{
680    tcg_out_cmp(s, cond, c1, c2, c2const);
681    tcg_out_movcc(s, tcg_cond_to_bcond[cond], MOVCC_ICC, ret, v1, v1const);
682}
683
684static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
685                               int32_t arg2, int const_arg2, TCGLabel *l)
686{
687    /* For 64-bit signed comparisons vs zero, we can avoid the compare.  */
688    int rcond = tcg_cond_to_rcond[cond];
689    if (arg2 == 0 && rcond) {
690        int off16 = 0;
691
692        if (l->has_value) {
693            off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
694        } else {
695            tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
696        }
697        tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
698                  | INSN_COND(rcond) | off16);
699    } else {
700        tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
701        tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
702    }
703    tcg_out_nop(s);
704}
705
706static void tcg_out_movr(TCGContext *s, int rcond, TCGReg ret, TCGReg c1,
707                         int32_t v1, int v1const)
708{
709    tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | (rcond << 10)
710              | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
711}
712
713static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
714                                TCGReg c1, int32_t c2, int c2const,
715                                int32_t v1, int v1const)
716{
717    /* For 64-bit signed comparisons vs zero, we can avoid the compare.
718       Note that the immediate range is one bit smaller, so we must check
719       for that as well.  */
720    int rcond = tcg_cond_to_rcond[cond];
721    if (c2 == 0 && rcond && (!v1const || check_fit_i32(v1, 10))) {
722        tcg_out_movr(s, rcond, ret, c1, v1, v1const);
723    } else {
724        tcg_out_cmp(s, cond, c1, c2, c2const);
725        tcg_out_movcc(s, tcg_cond_to_bcond[cond], MOVCC_XCC, ret, v1, v1const);
726    }
727}
728
729static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
730                                TCGReg c1, int32_t c2, bool c2const, bool neg)
731{
732    /* For 32-bit comparisons, we can play games with ADDC/SUBC.  */
733    switch (cond) {
734    case TCG_COND_LTU:
735    case TCG_COND_GEU:
736        /* The result of the comparison is in the carry bit.  */
737        break;
738
739    case TCG_COND_EQ:
740    case TCG_COND_NE:
741        /* For equality, we can transform to inequality vs zero.  */
742        if (c2 != 0) {
743            tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
744            c2 = TCG_REG_T1;
745        } else {
746            c2 = c1;
747        }
748        c1 = TCG_REG_G0, c2const = 0;
749        cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
750        break;
751
752    case TCG_COND_TSTEQ:
753    case TCG_COND_TSTNE:
754        /* Transform to inequality vs zero.  */
755        tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_AND);
756        c1 = TCG_REG_G0;
757        c2 = TCG_REG_T1, c2const = 0;
758        cond = (cond == TCG_COND_TSTEQ ? TCG_COND_GEU : TCG_COND_LTU);
759        break;
760
761    case TCG_COND_GTU:
762    case TCG_COND_LEU:
763        /* If we don't need to load a constant into a register, we can
764           swap the operands on GTU/LEU.  There's no benefit to loading
765           the constant into a temporary register.  */
766        if (!c2const || c2 == 0) {
767            TCGReg t = c1;
768            c1 = c2;
769            c2 = t;
770            c2const = 0;
771            cond = tcg_swap_cond(cond);
772            break;
773        }
774        /* FALLTHRU */
775
776    default:
777        tcg_out_cmp(s, cond, c1, c2, c2const);
778        tcg_out_movi_s13(s, ret, 0);
779        tcg_out_movcc(s, tcg_cond_to_bcond[cond],
780                      MOVCC_ICC, ret, neg ? -1 : 1, 1);
781        return;
782    }
783
784    tcg_out_cmp(s, cond, c1, c2, c2const);
785    if (cond == TCG_COND_LTU) {
786        if (neg) {
787            /* 0 - 0 - C = -C = (C ? -1 : 0) */
788            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_SUBC);
789        } else {
790            /* 0 + 0 + C =  C = (C ? 1 : 0) */
791            tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
792        }
793    } else {
794        if (neg) {
795            /* 0 + -1 + C = C - 1 = (C ? 0 : -1) */
796            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_ADDC);
797        } else {
798            /* 0 - -1 - C = 1 - C = (C ? 0 : 1) */
799            tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
800        }
801    }
802}
803
804static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
805                                TCGReg c1, int32_t c2, bool c2const, bool neg)
806{
807    int rcond;
808
809    if (use_vis3_instructions && !neg) {
810        switch (cond) {
811        case TCG_COND_NE:
812            if (c2 != 0) {
813                break;
814            }
815            c2 = c1, c2const = 0, c1 = TCG_REG_G0;
816            /* FALLTHRU */
817        case TCG_COND_LTU:
818            tcg_out_cmp(s, cond, c1, c2, c2const);
819            tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
820            return;
821        default:
822            break;
823        }
824    }
825
826    /* For 64-bit signed comparisons vs zero, we can avoid the compare
827       if the input does not overlap the output.  */
828    rcond = tcg_cond_to_rcond[cond];
829    if (c2 == 0 && rcond && c1 != ret) {
830        tcg_out_movi_s13(s, ret, 0);
831        tcg_out_movr(s, rcond, ret, c1, neg ? -1 : 1, 1);
832    } else {
833        tcg_out_cmp(s, cond, c1, c2, c2const);
834        tcg_out_movi_s13(s, ret, 0);
835        tcg_out_movcc(s, tcg_cond_to_bcond[cond],
836                      MOVCC_XCC, ret, neg ? -1 : 1, 1);
837    }
838}
839
840static void tcg_out_brcond(TCGContext *s, TCGType type, TCGCond cond,
841                           TCGReg arg1, TCGArg arg2, bool const_arg2,
842                           TCGLabel *l)
843{
844    if (type == TCG_TYPE_I32) {
845        tcg_out_brcond_i32(s, cond, arg1, arg2, const_arg2, l);
846    } else {
847        tcg_out_brcond_i64(s, cond, arg1, arg2, const_arg2, l);
848    }
849}
850
851static void tgen_brcond(TCGContext *s, TCGType type, TCGCond cond,
852                        TCGReg arg1, TCGReg arg2, TCGLabel *l)
853{
854    tcg_out_brcond(s, type, cond, arg1, arg2, false, l);
855}
856
857static void tgen_brcondi(TCGContext *s, TCGType type, TCGCond cond,
858                         TCGReg arg1, tcg_target_long arg2, TCGLabel *l)
859{
860    tcg_out_brcond(s, type, cond, arg1, arg2, true, l);
861}
862
863static const TCGOutOpBrcond outop_brcond = {
864    .base.static_constraint = C_O0_I2(r, rJ),
865    .out_rr = tgen_brcond,
866    .out_ri = tgen_brcondi,
867};
868
869static void tcg_out_setcond(TCGContext *s, TCGType type, TCGCond cond,
870                            TCGReg ret, TCGReg c1,
871                            TCGArg c2, bool c2const, bool neg)
872{
873    if (type == TCG_TYPE_I32) {
874        tcg_out_setcond_i32(s, cond, ret, c1, c2, c2const, neg);
875    } else {
876        tcg_out_setcond_i64(s, cond, ret, c1, c2, c2const, neg);
877    }
878}
879
880static void tgen_setcond(TCGContext *s, TCGType type, TCGCond cond,
881                         TCGReg dest, TCGReg arg1, TCGReg arg2)
882{
883    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, false);
884}
885
886static void tgen_setcondi(TCGContext *s, TCGType type, TCGCond cond,
887                          TCGReg dest, TCGReg arg1, tcg_target_long arg2)
888{
889    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, false);
890}
891
892static const TCGOutOpSetcond outop_setcond = {
893    .base.static_constraint = C_O1_I2(r, r, rJ),
894    .out_rrr = tgen_setcond,
895    .out_rri = tgen_setcondi,
896};
897
898static void tgen_negsetcond(TCGContext *s, TCGType type, TCGCond cond,
899                            TCGReg dest, TCGReg arg1, TCGReg arg2)
900{
901    tcg_out_setcond(s, type, cond, dest, arg1, arg2, false, true);
902}
903
904static void tgen_negsetcondi(TCGContext *s, TCGType type, TCGCond cond,
905                             TCGReg dest, TCGReg arg1, tcg_target_long arg2)
906{
907    tcg_out_setcond(s, type, cond, dest, arg1, arg2, true, true);
908}
909
910static const TCGOutOpSetcond outop_negsetcond = {
911    .base.static_constraint = C_O1_I2(r, r, rJ),
912    .out_rrr = tgen_negsetcond,
913    .out_rri = tgen_negsetcondi,
914};
915
916static void tgen_movcond(TCGContext *s, TCGType type, TCGCond cond,
917                         TCGReg ret, TCGReg c1, TCGArg c2, bool c2const,
918                         TCGArg v1, bool v1const, TCGArg v2, bool v2consf)
919{
920    if (type == TCG_TYPE_I32) {
921        tcg_out_movcond_i32(s, cond, ret, c1, c2, c2const, v1, v1const);
922    } else {
923        tcg_out_movcond_i64(s, cond, ret, c1, c2, c2const, v1, v1const);
924    }
925}
926
927static const TCGOutOpMovcond outop_movcond = {
928    .base.static_constraint = C_O1_I4(r, r, rJ, rI, 0),
929    .out = tgen_movcond,
930};
931
932static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest,
933                               bool in_prologue, bool tail_call)
934{
935    uintptr_t desti = (uintptr_t)dest;
936
937    tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
938                     desti & ~0xfff, in_prologue, TCG_REG_T2);
939    tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7,
940                   TCG_REG_T1, desti & 0xfff, JMPL);
941}
942
943static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
944                                 bool in_prologue)
945{
946    ptrdiff_t disp = tcg_pcrel_diff(s, dest);
947
948    if (disp == (int32_t)disp) {
949        tcg_out32(s, CALL | (uint32_t)disp >> 2);
950    } else {
951        tcg_out_jmpl_const(s, dest, in_prologue, false);
952    }
953}
954
955static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
956                         const TCGHelperInfo *info)
957{
958    tcg_out_call_nodelay(s, dest, false);
959    tcg_out_nop(s);
960}
961
962static void tcg_out_mb(TCGContext *s, unsigned a0)
963{
964    /* Note that the TCG memory order constants mirror the Sparc MEMBAR.  */
965    tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
966}
967
968/* Generate global QEMU prologue and epilogue code */
969static void tcg_target_qemu_prologue(TCGContext *s)
970{
971    int tmp_buf_size, frame_size;
972
973    /*
974     * The TCG temp buffer is at the top of the frame, immediately
975     * below the frame pointer.  Use the logical (aligned) offset here;
976     * the stack bias is applied in temp_allocate_frame().
977     */
978    tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
979    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
980
981    /*
982     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
983     * otherwise the minimal frame usable by callees.
984     */
985    frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
986    frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
987    frame_size += TCG_TARGET_STACK_ALIGN - 1;
988    frame_size &= -TCG_TARGET_STACK_ALIGN;
989    tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
990              INSN_IMM13(-frame_size));
991
992#ifndef CONFIG_SOFTMMU
993    if (guest_base != 0) {
994        tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG,
995                         guest_base, true, TCG_REG_T1);
996        tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
997    }
998#endif
999
1000    /* We choose TCG_REG_TB such that no move is required.  */
1001    QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
1002    tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
1003
1004    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
1005    /* delay slot */
1006    tcg_out_nop(s);
1007
1008    /* Epilogue for goto_ptr.  */
1009    tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
1010    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1011    /* delay slot */
1012    tcg_out_movi_s13(s, TCG_REG_O0, 0);
1013}
1014
1015static void tcg_out_tb_start(TCGContext *s)
1016{
1017    /* nothing to do */
1018}
1019
1020static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1021{
1022    int i;
1023    for (i = 0; i < count; ++i) {
1024        p[i] = NOP;
1025    }
1026}
1027
1028static const TCGLdstHelperParam ldst_helper_param = {
1029    .ntmp = 1, .tmp = { TCG_REG_T1 }
1030};
1031
1032static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1033{
1034    MemOp opc = get_memop(lb->oi);
1035    MemOp sgn;
1036
1037    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1038                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1039        return false;
1040    }
1041
1042    /* Use inline tcg_out_ext32s; otherwise let the helper sign-extend. */
1043    sgn = (opc & MO_SIZE) < MO_32 ? MO_SIGN : 0;
1044
1045    tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1046    tcg_out_call(s, qemu_ld_helpers[opc & (MO_SIZE | sgn)], NULL);
1047    tcg_out_ld_helper_ret(s, lb, sgn, &ldst_helper_param);
1048
1049    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1050    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1051                       (intptr_t)lb->raddr, 0);
1052}
1053
1054static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1055{
1056    MemOp opc = get_memop(lb->oi);
1057
1058    if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1059                     (intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1060        return false;
1061    }
1062
1063    tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1064    tcg_out_call(s, qemu_st_helpers[opc & MO_SIZE], NULL);
1065
1066    tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1067    return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1068                       (intptr_t)lb->raddr, 0);
1069}
1070
1071typedef struct {
1072    TCGReg base;
1073    TCGReg index;
1074    TCGAtomAlign aa;
1075} HostAddress;
1076
1077bool tcg_target_has_memory_bswap(MemOp memop)
1078{
1079    return true;
1080}
1081
1082/* We expect to use a 13-bit negative offset from ENV.  */
1083#define MIN_TLB_MASK_TABLE_OFS  -(1 << 12)
1084
1085/*
1086 * For system-mode, perform the TLB load and compare.
1087 * For user-mode, perform any required alignment tests.
1088 * In both cases, return a TCGLabelQemuLdst structure if the slow path
1089 * is required and fill in @h with the host address for the fast path.
1090 */
1091static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1092                                           TCGReg addr_reg, MemOpIdx oi,
1093                                           bool is_ld)
1094{
1095    TCGType addr_type = s->addr_type;
1096    TCGLabelQemuLdst *ldst = NULL;
1097    MemOp opc = get_memop(oi);
1098    MemOp s_bits = opc & MO_SIZE;
1099    unsigned a_mask;
1100
1101    /* We don't support unaligned accesses. */
1102    h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1103    h->aa.align = MAX(h->aa.align, s_bits);
1104    a_mask = (1u << h->aa.align) - 1;
1105
1106#ifdef CONFIG_SOFTMMU
1107    int mem_index = get_mmuidx(oi);
1108    int fast_off = tlb_mask_table_ofs(s, mem_index);
1109    int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1110    int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1111    int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1112                        : offsetof(CPUTLBEntry, addr_write);
1113    int add_off = offsetof(CPUTLBEntry, addend);
1114    int compare_mask;
1115    int cc;
1116
1117    /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx].  */
1118    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T2, TCG_AREG0, mask_off);
1119    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T3, TCG_AREG0, table_off);
1120
1121    /* Extract the page index, shifted into place for tlb index.  */
1122    tcg_out_arithi(s, TCG_REG_T1, addr_reg,
1123                   s->page_bits - CPU_TLB_ENTRY_BITS, SHIFT_SRL);
1124    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T2, ARITH_AND);
1125
1126    /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2.  */
1127    tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T3, ARITH_ADD);
1128
1129    /*
1130     * Load the tlb comparator and the addend.
1131     * Always load the entire 64-bit comparator for simplicity.
1132     * We will ignore the high bits via BPCC_ICC below.
1133     */
1134    tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_T2, TCG_REG_T1, cmp_off);
1135    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T1, TCG_REG_T1, add_off);
1136    h->base = TCG_REG_T1;
1137
1138    /* Mask out the page offset, except for the required alignment. */
1139    compare_mask = s->page_mask | a_mask;
1140    if (check_fit_tl(compare_mask, 13)) {
1141        tcg_out_arithi(s, TCG_REG_T3, addr_reg, compare_mask, ARITH_AND);
1142    } else {
1143        tcg_out_movi_s32(s, TCG_REG_T3, compare_mask);
1144        tcg_out_arith(s, TCG_REG_T3, addr_reg, TCG_REG_T3, ARITH_AND);
1145    }
1146    tcg_out_cmp(s, TCG_COND_NE, TCG_REG_T2, TCG_REG_T3, 0);
1147
1148    ldst = new_ldst_label(s);
1149    ldst->is_ld = is_ld;
1150    ldst->oi = oi;
1151    ldst->addr_reg = addr_reg;
1152    ldst->label_ptr[0] = s->code_ptr;
1153
1154    /* bne,pn %[xi]cc, label0 */
1155    cc = addr_type == TCG_TYPE_I32 ? BPCC_ICC : BPCC_XCC;
1156    tcg_out_bpcc0(s, COND_NE, BPCC_PN | cc, 0);
1157#else
1158    /*
1159     * If the size equals the required alignment, we can skip the test
1160     * and allow host SIGBUS to deliver SIGBUS to the guest.
1161     * Otherwise, test for at least natural alignment and defer
1162     * everything else to the helper functions.
1163     */
1164    if (s_bits != memop_alignment_bits(opc)) {
1165        tcg_debug_assert(check_fit_tl(a_mask, 13));
1166        tcg_out_arithi(s, TCG_REG_G0, addr_reg, a_mask, ARITH_ANDCC);
1167
1168        ldst = new_ldst_label(s);
1169        ldst->is_ld = is_ld;
1170        ldst->oi = oi;
1171        ldst->addr_reg = addr_reg;
1172        ldst->label_ptr[0] = s->code_ptr;
1173
1174        /* bne,pn %icc, label0 */
1175        tcg_out_bpcc0(s, COND_NE, BPCC_PN | BPCC_ICC, 0);
1176    }
1177    h->base = guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0;
1178#endif
1179
1180    /* If the guest address must be zero-extended, do in the delay slot.  */
1181    if (addr_type == TCG_TYPE_I32) {
1182        tcg_out_ext32u(s, TCG_REG_T2, addr_reg);
1183        h->index = TCG_REG_T2;
1184    } else {
1185        if (ldst) {
1186            tcg_out_nop(s);
1187        }
1188        h->index = addr_reg;
1189    }
1190    return ldst;
1191}
1192
1193static void tgen_qemu_ld(TCGContext *s, TCGType type, TCGReg data,
1194                         TCGReg addr, MemOpIdx oi)
1195{
1196    static const int ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = {
1197        [MO_UB]   = LDUB,
1198        [MO_SB]   = LDSB,
1199        [MO_UB | MO_LE] = LDUB,
1200        [MO_SB | MO_LE] = LDSB,
1201
1202        [MO_BEUW] = LDUH,
1203        [MO_BESW] = LDSH,
1204        [MO_BEUL] = LDUW,
1205        [MO_BESL] = LDSW,
1206        [MO_BEUQ] = LDX,
1207        [MO_BESQ] = LDX,
1208
1209        [MO_LEUW] = LDUH_LE,
1210        [MO_LESW] = LDSH_LE,
1211        [MO_LEUL] = LDUW_LE,
1212        [MO_LESL] = LDSW_LE,
1213        [MO_LEUQ] = LDX_LE,
1214        [MO_LESQ] = LDX_LE,
1215    };
1216
1217    TCGLabelQemuLdst *ldst;
1218    HostAddress h;
1219
1220    ldst = prepare_host_addr(s, &h, addr, oi, true);
1221
1222    tcg_out_ldst_rr(s, data, h.base, h.index,
1223                    ld_opc[get_memop(oi) & (MO_BSWAP | MO_SSIZE)]);
1224
1225    if (ldst) {
1226        ldst->type = type;
1227        ldst->datalo_reg = data;
1228        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1229    }
1230}
1231
1232static const TCGOutOpQemuLdSt outop_qemu_ld = {
1233    .base.static_constraint = C_O1_I1(r, r),
1234    .out = tgen_qemu_ld,
1235};
1236
1237static const TCGOutOpQemuLdSt2 outop_qemu_ld2 = {
1238    .base.static_constraint = C_NotImplemented,
1239};
1240
1241static void tgen_qemu_st(TCGContext *s, TCGType type, TCGReg data,
1242                         TCGReg addr, MemOpIdx oi)
1243{
1244    static const int st_opc[(MO_SIZE | MO_BSWAP) + 1] = {
1245        [MO_UB]   = STB,
1246
1247        [MO_BEUW] = STH,
1248        [MO_BEUL] = STW,
1249        [MO_BEUQ] = STX,
1250
1251        [MO_LEUW] = STH_LE,
1252        [MO_LEUL] = STW_LE,
1253        [MO_LEUQ] = STX_LE,
1254    };
1255
1256    TCGLabelQemuLdst *ldst;
1257    HostAddress h;
1258
1259    ldst = prepare_host_addr(s, &h, addr, oi, false);
1260
1261    tcg_out_ldst_rr(s, data, h.base, h.index,
1262                    st_opc[get_memop(oi) & (MO_BSWAP | MO_SIZE)]);
1263
1264    if (ldst) {
1265        ldst->type = type;
1266        ldst->datalo_reg = data;
1267        ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1268    }
1269}
1270
1271static const TCGOutOpQemuLdSt outop_qemu_st = {
1272    .base.static_constraint = C_O0_I2(rz, r),
1273    .out = tgen_qemu_st,
1274};
1275
1276static const TCGOutOpQemuLdSt2 outop_qemu_st2 = {
1277    .base.static_constraint = C_NotImplemented,
1278};
1279
1280static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1281{
1282    if (check_fit_ptr(a0, 13)) {
1283        tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1284        tcg_out_movi_s13(s, TCG_REG_O0, a0);
1285        return;
1286    } else {
1287        intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0);
1288        if (check_fit_ptr(tb_diff, 13)) {
1289            tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1290            /* Note that TCG_REG_TB has been unwound to O1.  */
1291            tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1292            return;
1293        }
1294    }
1295    tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1296    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1297    tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1298}
1299
1300static void tcg_out_goto_tb(TCGContext *s, int which)
1301{
1302    ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which));
1303
1304    /* Load link and indirect branch. */
1305    set_jmp_insn_offset(s, which);
1306    tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off);
1307    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1308    /* delay slot */
1309    tcg_out_nop(s);
1310    set_jmp_reset_offset(s, which);
1311
1312    /*
1313     * For the unlinked path of goto_tb, we need to reset TCG_REG_TB
1314     * to the beginning of this TB.
1315     */
1316    off = -tcg_current_code_size(s);
1317    if (check_fit_i32(off, 13)) {
1318        tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD);
1319    } else {
1320        tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off);
1321        tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1322    }
1323}
1324
1325static void tcg_out_goto_ptr(TCGContext *s, TCGReg a0)
1326{
1327    tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1328    tcg_out_mov_delay(s, TCG_REG_TB, a0);
1329}
1330
1331void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1332                              uintptr_t jmp_rx, uintptr_t jmp_rw)
1333{
1334}
1335
1336
1337static void tgen_add(TCGContext *s, TCGType type,
1338                     TCGReg a0, TCGReg a1, TCGReg a2)
1339{
1340    tcg_out_arith(s, a0, a1, a2, ARITH_ADD);
1341}
1342
1343static void tgen_addi(TCGContext *s, TCGType type,
1344                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1345{
1346    tcg_out_arithi(s, a0, a1, a2, ARITH_ADD);
1347}
1348
1349static const TCGOutOpBinary outop_add = {
1350    .base.static_constraint = C_O1_I2(r, r, rJ),
1351    .out_rrr = tgen_add,
1352    .out_rri = tgen_addi,
1353};
1354
1355static void tgen_addco_rrr(TCGContext *s, TCGType type,
1356                           TCGReg a0, TCGReg a1, TCGReg a2)
1357{
1358    tcg_out_arith(s, a0, a1, a2, ARITH_ADDCC);
1359}
1360
1361static void tgen_addco_rri(TCGContext *s, TCGType type,
1362                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1363{
1364    tcg_out_arithi(s, a0, a1, a2, ARITH_ADDCC);
1365}
1366
1367static const TCGOutOpBinary outop_addco = {
1368    .base.static_constraint = C_O1_I2(r, r, rJ),
1369    .out_rrr = tgen_addco_rrr,
1370    .out_rri = tgen_addco_rri,
1371};
1372
1373static void tgen_addci_rrr(TCGContext *s, TCGType type,
1374                           TCGReg a0, TCGReg a1, TCGReg a2)
1375{
1376    if (type == TCG_TYPE_I32) {
1377        tcg_out_arith(s, a0, a1, a2, ARITH_ADDC);
1378    } else if (use_vis3_instructions) {
1379        tcg_out_arith(s, a0, a1, a2, ARITH_ADDXC);
1380    } else {
1381        tcg_out_arith(s, TCG_REG_T1, a1, a2, ARITH_ADD);  /* for CC */
1382        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_ADD);  /* for CS */
1383        /* Select the correct result based on actual carry value. */
1384        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1385    }
1386}
1387
1388static void tgen_addci_rri(TCGContext *s, TCGType type,
1389                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1390{
1391    if (type == TCG_TYPE_I32) {
1392        tcg_out_arithi(s, a0, a1, a2, ARITH_ADDC);
1393        return;
1394    }
1395    /* !use_vis3_instructions */
1396    if (a2 != 0) {
1397        tcg_out_arithi(s, TCG_REG_T1, a1, a2, ARITH_ADD); /* for CC */
1398        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_ADD);  /* for CS */
1399        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1400    } else if (a0 == a1) {
1401        tcg_out_arithi(s, TCG_REG_T1, a1, 1, ARITH_ADD);
1402        tcg_out_movcc(s, COND_CS, MOVCC_XCC, a0, TCG_REG_T1, false);
1403    } else {
1404        tcg_out_arithi(s, a0, a1, 1, ARITH_ADD);
1405        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, a1, false);
1406    }
1407}
1408
1409static TCGConstraintSetIndex cset_addci(TCGType type, unsigned flags)
1410{
1411    if (use_vis3_instructions && type == TCG_TYPE_I64) {
1412        /* Note that ADDXC doesn't accept immediates.  */
1413        return C_O1_I2(r, rz, rz);
1414    }
1415    return C_O1_I2(r, rz, rJ);
1416}
1417
1418static const TCGOutOpAddSubCarry outop_addci = {
1419    .base.static_constraint = C_Dynamic,
1420    .base.dynamic_constraint = cset_addci,
1421    .out_rrr = tgen_addci_rrr,
1422    .out_rri = tgen_addci_rri,
1423};
1424
1425/* Copy %xcc.c to %icc.c */
1426static void tcg_out_dup_xcc_c(TCGContext *s)
1427{
1428    if (use_vis3_instructions) {
1429        tcg_out_arith(s, TCG_REG_T1, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
1430    } else {
1431        tcg_out_movi_s13(s, TCG_REG_T1, 0);
1432        tcg_out_movcc(s, COND_CS, MOVCC_XCC, TCG_REG_T1, 1, true);
1433    }
1434    /* Write carry-in into %icc via {0,1} + -1. */
1435    tcg_out_arithi(s, TCG_REG_G0, TCG_REG_T1, -1, ARITH_ADDCC);
1436}
1437
1438static void tgen_addcio_rrr(TCGContext *s, TCGType type,
1439                            TCGReg a0, TCGReg a1, TCGReg a2)
1440{
1441    if (type != TCG_TYPE_I32) {
1442        if (use_vis3_instructions) {
1443            tcg_out_arith(s, a0, a1, a2, ARITH_ADDXCCC);
1444            return;
1445        }
1446        tcg_out_dup_xcc_c(s);
1447    }
1448    tcg_out_arith(s, a0, a1, a2, ARITH_ADDCCC);
1449}
1450
1451static void tgen_addcio_rri(TCGContext *s, TCGType type,
1452                            TCGReg a0, TCGReg a1, tcg_target_long a2)
1453{
1454    if (type != TCG_TYPE_I32) {
1455        /* !use_vis3_instructions */
1456        tcg_out_dup_xcc_c(s);
1457    }
1458    tcg_out_arithi(s, a0, a1, a2, ARITH_ADDCCC);
1459}
1460
1461static TCGConstraintSetIndex cset_addcio(TCGType type, unsigned flags)
1462{
1463    if (use_vis3_instructions && type == TCG_TYPE_I64) {
1464        /* Note that ADDXCCC doesn't accept immediates.  */
1465        return C_O1_I2(r, rz, rz);
1466    }
1467    return C_O1_I2(r, rz, rJ);
1468}
1469
1470static const TCGOutOpBinary outop_addcio = {
1471    .base.static_constraint = C_Dynamic,
1472    .base.dynamic_constraint = cset_addcio,
1473    .out_rrr = tgen_addcio_rrr,
1474    .out_rri = tgen_addcio_rri,
1475};
1476
1477static void tcg_out_set_carry(TCGContext *s)
1478{
1479    /* 0x11 -> xcc = nzvC, icc = nzvC */
1480    tcg_out_arithi(s, 0, TCG_REG_G0, 0x11, WRCCR);
1481}
1482
1483static void tgen_and(TCGContext *s, TCGType type,
1484                     TCGReg a0, TCGReg a1, TCGReg a2)
1485{
1486    tcg_out_arith(s, a0, a1, a2, ARITH_AND);
1487}
1488
1489static void tgen_andi(TCGContext *s, TCGType type,
1490                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1491{
1492    tcg_out_arithi(s, a0, a1, a2, ARITH_AND);
1493}
1494
1495static const TCGOutOpBinary outop_and = {
1496    .base.static_constraint = C_O1_I2(r, r, rJ),
1497    .out_rrr = tgen_and,
1498    .out_rri = tgen_andi,
1499};
1500
1501static void tgen_andc(TCGContext *s, TCGType type,
1502                      TCGReg a0, TCGReg a1, TCGReg a2)
1503{
1504    tcg_out_arith(s, a0, a1, a2, ARITH_ANDN);
1505}
1506
1507static const TCGOutOpBinary outop_andc = {
1508    .base.static_constraint = C_O1_I2(r, r, r),
1509    .out_rrr = tgen_andc,
1510};
1511
1512static const TCGOutOpBinary outop_clz = {
1513    .base.static_constraint = C_NotImplemented,
1514};
1515
1516static void tgen_ctpop(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1517{
1518    tcg_out_arith(s, a0, TCG_REG_G0, a1, ARITH_POPC);
1519}
1520
1521static TCGConstraintSetIndex cset_ctpop(TCGType type, unsigned flags)
1522{
1523    if (use_popc_instructions && type == TCG_TYPE_I64) {
1524        return C_O1_I1(r, r);
1525    }
1526    return C_NotImplemented;
1527}
1528
1529static const TCGOutOpUnary outop_ctpop = {
1530    .base.static_constraint = C_Dynamic,
1531    .base.dynamic_constraint = cset_ctpop,
1532    .out_rr = tgen_ctpop,
1533};
1534
1535static const TCGOutOpBinary outop_ctz = {
1536    .base.static_constraint = C_NotImplemented,
1537};
1538
1539static void tgen_divs_rJ(TCGContext *s, TCGType type,
1540                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1541{
1542    uint32_t insn;
1543
1544    if (type == TCG_TYPE_I32) {
1545        /* Load Y with the sign extension of a1 to 64-bits.  */
1546        tcg_out_arithi(s, TCG_REG_T1, a1, 31, SHIFT_SRA);
1547        tcg_out_sety(s, TCG_REG_T1);
1548        insn = ARITH_SDIV;
1549    } else {
1550        insn = ARITH_SDIVX;
1551    }
1552    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1553}
1554
1555static void tgen_divs(TCGContext *s, TCGType type,
1556                      TCGReg a0, TCGReg a1, TCGReg a2)
1557{
1558    tgen_divs_rJ(s, type, a0, a1, a2, false);
1559}
1560
1561static void tgen_divsi(TCGContext *s, TCGType type,
1562                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1563{
1564    tgen_divs_rJ(s, type, a0, a1, a2, true);
1565}
1566
1567static const TCGOutOpBinary outop_divs = {
1568    .base.static_constraint = C_O1_I2(r, r, rJ),
1569    .out_rrr = tgen_divs,
1570    .out_rri = tgen_divsi,
1571};
1572
1573static const TCGOutOpDivRem outop_divs2 = {
1574    .base.static_constraint = C_NotImplemented,
1575};
1576
1577static void tgen_divu_rJ(TCGContext *s, TCGType type,
1578                         TCGReg a0, TCGReg a1, TCGArg a2, bool c2)
1579{
1580    uint32_t insn;
1581
1582    if (type == TCG_TYPE_I32) {
1583        /* Load Y with the zero extension to 64-bits.  */
1584        tcg_out_sety(s, TCG_REG_G0);
1585        insn = ARITH_UDIV;
1586    } else {
1587        insn = ARITH_UDIVX;
1588    }
1589    tcg_out_arithc(s, a0, a1, a2, c2, insn);
1590}
1591
1592static void tgen_divu(TCGContext *s, TCGType type,
1593                      TCGReg a0, TCGReg a1, TCGReg a2)
1594{
1595    tgen_divu_rJ(s, type, a0, a1, a2, false);
1596}
1597
1598static void tgen_divui(TCGContext *s, TCGType type,
1599                       TCGReg a0, TCGReg a1, tcg_target_long a2)
1600{
1601    tgen_divu_rJ(s, type, a0, a1, a2, true);
1602}
1603
1604static const TCGOutOpBinary outop_divu = {
1605    .base.static_constraint = C_O1_I2(r, r, rJ),
1606    .out_rrr = tgen_divu,
1607    .out_rri = tgen_divui,
1608};
1609
1610static const TCGOutOpDivRem outop_divu2 = {
1611    .base.static_constraint = C_NotImplemented,
1612};
1613
1614static const TCGOutOpBinary outop_eqv = {
1615    .base.static_constraint = C_NotImplemented,
1616};
1617
1618static void tgen_extrh_i64_i32(TCGContext *s, TCGType t, TCGReg a0, TCGReg a1)
1619{
1620    tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX);
1621}
1622
1623static const TCGOutOpUnary outop_extrh_i64_i32 = {
1624    .base.static_constraint = C_O1_I1(r, r),
1625    .out_rr = tgen_extrh_i64_i32,
1626};
1627
1628static void tgen_mul(TCGContext *s, TCGType type,
1629                     TCGReg a0, TCGReg a1, TCGReg a2)
1630{
1631    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1632    tcg_out_arith(s, a0, a1, a2, insn);
1633}
1634
1635static void tgen_muli(TCGContext *s, TCGType type,
1636                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1637{
1638    uint32_t insn = type == TCG_TYPE_I32 ? ARITH_UMUL : ARITH_MULX;
1639    tcg_out_arithi(s, a0, a1, a2, insn);
1640}
1641
1642static const TCGOutOpBinary outop_mul = {
1643    .base.static_constraint = C_O1_I2(r, r, rJ),
1644    .out_rrr = tgen_mul,
1645    .out_rri = tgen_muli,
1646};
1647
1648/*
1649 * The 32-bit multiply insns produce a full 64-bit result.
1650 * Supporting 32-bit mul[us]2 opcodes avoids sign/zero-extensions
1651 * before the actual multiply; we only need extract the high part
1652 * into the separate operand.
1653 */
1654static TCGConstraintSetIndex cset_mul2(TCGType type, unsigned flags)
1655{
1656    return type == TCG_TYPE_I32 ? C_O2_I2(r, r, r, r) : C_NotImplemented;
1657}
1658
1659static void tgen_muls2(TCGContext *s, TCGType type,
1660                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1661{
1662    tcg_out_arith(s, a0, a2, a3, ARITH_SMUL);
1663    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1664}
1665
1666static const TCGOutOpMul2 outop_muls2 = {
1667    .base.static_constraint = C_Dynamic,
1668    .base.dynamic_constraint = cset_mul2,
1669    .out_rrrr = tgen_muls2,
1670};
1671
1672static const TCGOutOpBinary outop_mulsh = {
1673    .base.static_constraint = C_NotImplemented,
1674};
1675
1676static void tgen_mulu2(TCGContext *s, TCGType type,
1677                       TCGReg a0, TCGReg a1, TCGReg a2, TCGReg a3)
1678{
1679    tcg_out_arith(s, a0, a2, a3, ARITH_UMUL);
1680    tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1681}
1682
1683static const TCGOutOpMul2 outop_mulu2 = {
1684    .base.static_constraint = C_Dynamic,
1685    .base.dynamic_constraint = cset_mul2,
1686    .out_rrrr = tgen_mulu2,
1687};
1688
1689static void tgen_muluh(TCGContext *s, TCGType type,
1690                       TCGReg a0, TCGReg a1, TCGReg a2)
1691{
1692    tcg_out_arith(s, a0, a1, a2, ARITH_UMULXHI);
1693}
1694
1695static TCGConstraintSetIndex cset_muluh(TCGType type, unsigned flags)
1696{
1697    return (type == TCG_TYPE_I64 && use_vis3_instructions
1698            ? C_O1_I2(r, r, r) : C_NotImplemented);
1699}
1700
1701static const TCGOutOpBinary outop_muluh = {
1702    .base.static_constraint = C_Dynamic,
1703    .base.dynamic_constraint = cset_muluh,
1704    .out_rrr = tgen_muluh,
1705};
1706
1707static const TCGOutOpBinary outop_nand = {
1708    .base.static_constraint = C_NotImplemented,
1709};
1710
1711static const TCGOutOpBinary outop_nor = {
1712    .base.static_constraint = C_NotImplemented,
1713};
1714
1715static void tgen_or(TCGContext *s, TCGType type,
1716                     TCGReg a0, TCGReg a1, TCGReg a2)
1717{
1718    tcg_out_arith(s, a0, a1, a2, ARITH_OR);
1719}
1720
1721static void tgen_ori(TCGContext *s, TCGType type,
1722                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1723{
1724    tcg_out_arithi(s, a0, a1, a2, ARITH_OR);
1725}
1726
1727static const TCGOutOpBinary outop_or = {
1728    .base.static_constraint = C_O1_I2(r, r, rJ),
1729    .out_rrr = tgen_or,
1730    .out_rri = tgen_ori,
1731};
1732
1733static void tgen_orc(TCGContext *s, TCGType type,
1734                     TCGReg a0, TCGReg a1, TCGReg a2)
1735{
1736    tcg_out_arith(s, a0, a1, a2, ARITH_ORN);
1737}
1738
1739static const TCGOutOpBinary outop_orc = {
1740    .base.static_constraint = C_O1_I2(r, r, r),
1741    .out_rrr = tgen_orc,
1742};
1743
1744static const TCGOutOpBinary outop_rems = {
1745    .base.static_constraint = C_NotImplemented,
1746};
1747
1748static const TCGOutOpBinary outop_remu = {
1749    .base.static_constraint = C_NotImplemented,
1750};
1751
1752static const TCGOutOpBinary outop_rotl = {
1753    .base.static_constraint = C_NotImplemented,
1754};
1755
1756static const TCGOutOpBinary outop_rotr = {
1757    .base.static_constraint = C_NotImplemented,
1758};
1759
1760static void tgen_sar(TCGContext *s, TCGType type,
1761                     TCGReg a0, TCGReg a1, TCGReg a2)
1762{
1763    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1764    tcg_out_arith(s, a0, a1, a2, insn);
1765}
1766
1767static void tgen_sari(TCGContext *s, TCGType type,
1768                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1769{
1770    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRA : SHIFT_SRAX;
1771    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1772    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1773}
1774
1775static const TCGOutOpBinary outop_sar = {
1776    .base.static_constraint = C_O1_I2(r, r, rJ),
1777    .out_rrr = tgen_sar,
1778    .out_rri = tgen_sari,
1779};
1780
1781static void tgen_shl(TCGContext *s, TCGType type,
1782                     TCGReg a0, TCGReg a1, TCGReg a2)
1783{
1784    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1785    tcg_out_arith(s, a0, a1, a2, insn);
1786}
1787
1788static void tgen_shli(TCGContext *s, TCGType type,
1789                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1790{
1791    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SLL : SHIFT_SLLX;
1792    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1793    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1794}
1795
1796static const TCGOutOpBinary outop_shl = {
1797    .base.static_constraint = C_O1_I2(r, r, rJ),
1798    .out_rrr = tgen_shl,
1799    .out_rri = tgen_shli,
1800};
1801
1802static void tgen_shr(TCGContext *s, TCGType type,
1803                     TCGReg a0, TCGReg a1, TCGReg a2)
1804{
1805    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1806    tcg_out_arith(s, a0, a1, a2, insn);
1807}
1808
1809static void tgen_shri(TCGContext *s, TCGType type,
1810                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1811{
1812    uint32_t insn = type == TCG_TYPE_I32 ? SHIFT_SRL : SHIFT_SRLX;
1813    uint32_t mask = type == TCG_TYPE_I32 ? 31 : 63;
1814    tcg_out_arithi(s, a0, a1, a2 & mask, insn);
1815}
1816
1817static const TCGOutOpBinary outop_shr = {
1818    .base.static_constraint = C_O1_I2(r, r, rJ),
1819    .out_rrr = tgen_shr,
1820    .out_rri = tgen_shri,
1821};
1822
1823static void tgen_sub(TCGContext *s, TCGType type,
1824                     TCGReg a0, TCGReg a1, TCGReg a2)
1825{
1826    tcg_out_arith(s, a0, a1, a2, ARITH_SUB);
1827}
1828
1829static const TCGOutOpSubtract outop_sub = {
1830    .base.static_constraint = C_O1_I2(r, r, r),
1831    .out_rrr = tgen_sub,
1832};
1833
1834static void tgen_subbo_rrr(TCGContext *s, TCGType type,
1835                           TCGReg a0, TCGReg a1, TCGReg a2)
1836{
1837    tcg_out_arith(s, a0, a1, a2, ARITH_SUBCC);
1838}
1839
1840static void tgen_subbo_rri(TCGContext *s, TCGType type,
1841                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1842{
1843    tcg_out_arithi(s, a0, a1, a2, ARITH_SUBCC);
1844}
1845
1846static const TCGOutOpAddSubCarry outop_subbo = {
1847    .base.static_constraint = C_O1_I2(r, rz, rJ),
1848    .out_rrr = tgen_subbo_rrr,
1849    .out_rri = tgen_subbo_rri,
1850};
1851
1852static void tgen_subbi_rrr(TCGContext *s, TCGType type,
1853                           TCGReg a0, TCGReg a1, TCGReg a2)
1854{
1855    /* TODO: OSA 2015 added SUBXC */
1856    if (type == TCG_TYPE_I32) {
1857        tcg_out_arith(s, a0, a1, a2, ARITH_SUBC);
1858    } else {
1859        tcg_out_arith(s, TCG_REG_T1, a1, a2, ARITH_SUB);  /* for CC */
1860        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_SUB);  /* for CS */
1861        /* Select the correct result based on actual borrow value. */
1862        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1863    }
1864}
1865
1866static void tgen_subbi_rri(TCGContext *s, TCGType type,
1867                           TCGReg a0, TCGReg a1, tcg_target_long a2)
1868{
1869    if (type == TCG_TYPE_I32) {
1870        tcg_out_arithi(s, a0, a1, a2, ARITH_SUBC);
1871    } else if (a2 != 0) {
1872        tcg_out_arithi(s, TCG_REG_T1, a1, a2, ARITH_SUB);  /* for CC */
1873        tcg_out_arithi(s, a0, TCG_REG_T1, 1, ARITH_SUB);   /* for CS */
1874        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, TCG_REG_T1, false);
1875    } else if (a0 == a1) {
1876        tcg_out_arithi(s, TCG_REG_T1, a1, 1, ARITH_SUB);
1877        tcg_out_movcc(s, COND_CS, MOVCC_XCC, a0, TCG_REG_T1, false);
1878    } else {
1879        tcg_out_arithi(s, a0, a1, 1, ARITH_SUB);
1880        tcg_out_movcc(s, COND_CC, MOVCC_XCC, a0, a1, false);
1881    }
1882}
1883
1884static const TCGOutOpAddSubCarry outop_subbi = {
1885    .base.static_constraint = C_O1_I2(r, rz, rJ),
1886    .out_rrr = tgen_subbi_rrr,
1887    .out_rri = tgen_subbi_rri,
1888};
1889
1890static void tgen_subbio_rrr(TCGContext *s, TCGType type,
1891                            TCGReg a0, TCGReg a1, TCGReg a2)
1892{
1893    if (type != TCG_TYPE_I32) {
1894        /* TODO: OSA 2015 added SUBXCCC */
1895        tcg_out_dup_xcc_c(s);
1896    }
1897    tcg_out_arith(s, a0, a1, a2, ARITH_SUBCCC);
1898}
1899
1900static void tgen_subbio_rri(TCGContext *s, TCGType type,
1901                            TCGReg a0, TCGReg a1, tcg_target_long a2)
1902{
1903    if (type != TCG_TYPE_I32) {
1904        tcg_out_dup_xcc_c(s);
1905    }
1906    tcg_out_arithi(s, a0, a1, a2, ARITH_SUBCCC);
1907}
1908
1909static const TCGOutOpAddSubCarry outop_subbio = {
1910    .base.static_constraint = C_O1_I2(r, rz, rJ),
1911    .out_rrr = tgen_subbio_rrr,
1912    .out_rri = tgen_subbio_rri,
1913};
1914
1915static void tcg_out_set_borrow(TCGContext *s)
1916{
1917    tcg_out_set_carry(s);  /* borrow == carry */
1918}
1919
1920static void tgen_xor(TCGContext *s, TCGType type,
1921                     TCGReg a0, TCGReg a1, TCGReg a2)
1922{
1923    tcg_out_arith(s, a0, a1, a2, ARITH_XOR);
1924}
1925
1926static void tgen_xori(TCGContext *s, TCGType type,
1927                      TCGReg a0, TCGReg a1, tcg_target_long a2)
1928{
1929    tcg_out_arithi(s, a0, a1, a2, ARITH_XOR);
1930}
1931
1932static const TCGOutOpBinary outop_xor = {
1933    .base.static_constraint = C_O1_I2(r, r, rJ),
1934    .out_rrr = tgen_xor,
1935    .out_rri = tgen_xori,
1936};
1937
1938static const TCGOutOpBswap outop_bswap16 = {
1939    .base.static_constraint = C_NotImplemented,
1940};
1941
1942static const TCGOutOpBswap outop_bswap32 = {
1943    .base.static_constraint = C_NotImplemented,
1944};
1945
1946static const TCGOutOpUnary outop_bswap64 = {
1947    .base.static_constraint = C_NotImplemented,
1948};
1949
1950static void tgen_neg(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1951{
1952     tgen_sub(s, type, a0, TCG_REG_G0, a1);
1953}
1954
1955static const TCGOutOpUnary outop_neg = {
1956    .base.static_constraint = C_O1_I1(r, r),
1957    .out_rr = tgen_neg,
1958};
1959
1960static void tgen_not(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1)
1961{
1962     tgen_orc(s, type, a0, TCG_REG_G0, a1);
1963}
1964
1965static const TCGOutOpUnary outop_not = {
1966    .base.static_constraint = C_O1_I1(r, r),
1967    .out_rr = tgen_not,
1968};
1969
1970static const TCGOutOpDeposit outop_deposit = {
1971    .base.static_constraint = C_NotImplemented,
1972};
1973
1974static void tgen_extract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
1975                         unsigned ofs, unsigned len)
1976{
1977    tcg_debug_assert(ofs + len == 32);
1978    tcg_out_arithi(s, a0, a1, ofs, SHIFT_SRL);
1979}
1980
1981static const TCGOutOpExtract outop_extract = {
1982    .base.static_constraint = C_O1_I1(r, r),
1983    .out_rr = tgen_extract,
1984};
1985
1986static void tgen_sextract(TCGContext *s, TCGType type, TCGReg a0, TCGReg a1,
1987                          unsigned ofs, unsigned len)
1988{
1989    tcg_debug_assert(ofs + len == 32);
1990    tcg_out_arithi(s, a0, a1, ofs, SHIFT_SRA);
1991}
1992
1993static const TCGOutOpExtract outop_sextract = {
1994    .base.static_constraint = C_O1_I1(r, r),
1995    .out_rr = tgen_sextract,
1996};
1997
1998static const TCGOutOpExtract2 outop_extract2 = {
1999    .base.static_constraint = C_NotImplemented,
2000};
2001
2002static void tgen_ld8u(TCGContext *s, TCGType type, TCGReg dest,
2003                      TCGReg base, ptrdiff_t offset)
2004{
2005    tcg_out_ldst(s, dest, base, offset, LDUB);
2006}
2007
2008static const TCGOutOpLoad outop_ld8u = {
2009    .base.static_constraint = C_O1_I1(r, r),
2010    .out = tgen_ld8u,
2011};
2012
2013static void tgen_ld8s(TCGContext *s, TCGType type, TCGReg dest,
2014                      TCGReg base, ptrdiff_t offset)
2015{
2016    tcg_out_ldst(s, dest, base, offset, LDSB);
2017}
2018
2019static const TCGOutOpLoad outop_ld8s = {
2020    .base.static_constraint = C_O1_I1(r, r),
2021    .out = tgen_ld8s,
2022};
2023
2024static void tgen_ld16u(TCGContext *s, TCGType type, TCGReg dest,
2025                       TCGReg base, ptrdiff_t offset)
2026{
2027    tcg_out_ldst(s, dest, base, offset, LDUH);
2028}
2029
2030static const TCGOutOpLoad outop_ld16u = {
2031    .base.static_constraint = C_O1_I1(r, r),
2032    .out = tgen_ld16u,
2033};
2034
2035static void tgen_ld16s(TCGContext *s, TCGType type, TCGReg dest,
2036                       TCGReg base, ptrdiff_t offset)
2037{
2038    tcg_out_ldst(s, dest, base, offset, LDSH);
2039}
2040
2041static const TCGOutOpLoad outop_ld16s = {
2042    .base.static_constraint = C_O1_I1(r, r),
2043    .out = tgen_ld16s,
2044};
2045
2046static void tgen_ld32u(TCGContext *s, TCGType type, TCGReg dest,
2047                       TCGReg base, ptrdiff_t offset)
2048{
2049    tcg_out_ldst(s, dest, base, offset, LDUW);
2050}
2051
2052static const TCGOutOpLoad outop_ld32u = {
2053    .base.static_constraint = C_O1_I1(r, r),
2054    .out = tgen_ld32u,
2055};
2056
2057static void tgen_ld32s(TCGContext *s, TCGType type, TCGReg dest,
2058                       TCGReg base, ptrdiff_t offset)
2059{
2060    tcg_out_ldst(s, dest, base, offset, LDSW);
2061}
2062
2063static const TCGOutOpLoad outop_ld32s = {
2064    .base.static_constraint = C_O1_I1(r, r),
2065    .out = tgen_ld32s,
2066};
2067
2068static void tgen_st8_r(TCGContext *s, TCGType type, TCGReg data,
2069                       TCGReg base, ptrdiff_t offset)
2070{
2071    tcg_out_ldst(s, data, base, offset, STB);
2072}
2073
2074static const TCGOutOpStore outop_st8 = {
2075    .base.static_constraint = C_O0_I2(rz, r),
2076    .out_r = tgen_st8_r,
2077};
2078
2079static void tgen_st16_r(TCGContext *s, TCGType type, TCGReg data,
2080                       TCGReg base, ptrdiff_t offset)
2081{
2082    tcg_out_ldst(s, data, base, offset, STH);
2083}
2084
2085static const TCGOutOpStore outop_st16 = {
2086    .base.static_constraint = C_O0_I2(rz, r),
2087    .out_r = tgen_st16_r,
2088};
2089
2090static const TCGOutOpStore outop_st = {
2091    .base.static_constraint = C_O0_I2(rz, r),
2092    .out_r = tcg_out_st,
2093};
2094
2095
2096static TCGConstraintSetIndex
2097tcg_target_op_def(TCGOpcode op, TCGType type, unsigned flags)
2098{
2099    return C_NotImplemented;
2100}
2101
2102static void tcg_target_init(TCGContext *s)
2103{
2104    unsigned long hwcap = qemu_getauxval(AT_HWCAP);
2105
2106    /*
2107     * Only probe for the platform and capabilities if we haven't already
2108     * determined maximum values at compile time.
2109     */
2110    use_popc_instructions = (hwcap & HWCAP_SPARC_POPC) != 0;
2111#ifndef use_vis3_instructions
2112    use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
2113#endif
2114
2115    tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
2116    tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;
2117
2118    tcg_target_call_clobber_regs = 0;
2119    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
2120    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
2121    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
2122    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
2123    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
2124    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
2125    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
2126    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
2127    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
2128    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
2129    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
2130    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
2131    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
2132    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
2133    tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
2134
2135    s->reserved_regs = 0;
2136    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
2137    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
2138    tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
2139    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
2140    tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
2141    tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
2142    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
2143    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
2144    tcg_regset_set_reg(s->reserved_regs, TCG_REG_T3); /* for internal use */
2145}
2146
2147#define ELF_HOST_MACHINE  EM_SPARCV9
2148
2149typedef struct {
2150    DebugFrameHeader h;
2151    uint8_t fde_def_cfa[4];
2152    uint8_t fde_win_save;
2153    uint8_t fde_ret_save[3];
2154} DebugFrame;
2155
2156static const DebugFrame debug_frame = {
2157    .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
2158    .h.cie.id = -1,
2159    .h.cie.version = 1,
2160    .h.cie.code_align = 1,
2161    .h.cie.data_align = -sizeof(void *) & 0x7f,
2162    .h.cie.return_column = 15,            /* o7 */
2163
2164    /* Total FDE size does not include the "len" member.  */
2165    .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
2166
2167    .fde_def_cfa = {
2168        12, 30,                         /* DW_CFA_def_cfa i6, 2047 */
2169        (2047 & 0x7f) | 0x80, (2047 >> 7)
2170    },
2171    .fde_win_save = 0x2d,               /* DW_CFA_GNU_window_save */
2172    .fde_ret_save = { 9, 15, 31 },      /* DW_CFA_register o7, i7 */
2173};
2174
2175void tcg_register_jit(const void *buf, size_t buf_size)
2176{
2177    tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
2178}
2179