2
* Tiny Code Generator for QEMU
4
* Copyright (c) 2008 Fabrice Bellard
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:
13
* The above copyright notice and this permission notice shall be included in
14
* all copies or substantial portions of the Software.
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
25
/* We only support generating code for 64-bit mode. */
27
#error "unsupported code generation mode"
30
#include "../tcg-ldst.c.inc"
31
#include "../tcg-pool.c.inc"
33
#ifdef CONFIG_DEBUG_TCG
34
static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
70
#define TCG_CT_CONST_S11 0x100
71
#define TCG_CT_CONST_S13 0x200
72
#define TCG_CT_CONST_ZERO 0x400
74
#define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 32)
76
/* Define some temporary registers. T3 is used for constant generation. */
77
#define TCG_REG_T1 TCG_REG_G1
78
#define TCG_REG_T2 TCG_REG_G2
79
#define TCG_REG_T3 TCG_REG_O7
82
# define TCG_GUEST_BASE_REG TCG_REG_I5
85
#define TCG_REG_TB TCG_REG_I1
87
static const int tcg_target_reg_alloc_order[] = {
116
static const int tcg_target_call_iarg_regs[6] = {
125
static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot)
127
tcg_debug_assert(kind == TCG_CALL_RET_NORMAL);
128
tcg_debug_assert(slot >= 0 && slot <= 3);
129
return TCG_REG_O0 + slot;
132
#define INSN_OP(x) ((x) << 30)
133
#define INSN_OP2(x) ((x) << 22)
134
#define INSN_OP3(x) ((x) << 19)
135
#define INSN_OPF(x) ((x) << 5)
136
#define INSN_RD(x) ((x) << 25)
137
#define INSN_RS1(x) ((x) << 14)
138
#define INSN_RS2(x) (x)
139
#define INSN_ASI(x) ((x) << 5)
141
#define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff))
142
#define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff))
143
#define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff))
144
#define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20))
145
#define INSN_OFF19(x) (((x) >> 2) & 0x07ffff)
146
#define INSN_COND(x) ((x) << 25)
164
#define BA (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2))
173
#define MOVCC_ICC (1 << 18)
174
#define MOVCC_XCC (1 << 18 | 1 << 12)
177
#define BPCC_XCC (2 << 20)
178
#define BPCC_PT (1 << 19)
180
#define BPCC_A (1 << 29)
182
#define BPR_PT BPCC_PT
184
#define ARITH_ADD (INSN_OP(2) | INSN_OP3(0x00))
185
#define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10))
186
#define ARITH_AND (INSN_OP(2) | INSN_OP3(0x01))
187
#define ARITH_ANDCC (INSN_OP(2) | INSN_OP3(0x11))
188
#define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05))
189
#define ARITH_OR (INSN_OP(2) | INSN_OP3(0x02))
190
#define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12))
191
#define ARITH_ORN (INSN_OP(2) | INSN_OP3(0x06))
192
#define ARITH_XOR (INSN_OP(2) | INSN_OP3(0x03))
193
#define ARITH_SUB (INSN_OP(2) | INSN_OP3(0x04))
194
#define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14))
195
#define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08))
196
#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
197
#define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
198
#define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
199
#define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
200
#define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f))
201
#define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09))
202
#define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d))
203
#define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d))
204
#define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
205
#define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
207
#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
208
#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
210
#define SHIFT_SLL (INSN_OP(2) | INSN_OP3(0x25))
211
#define SHIFT_SRL (INSN_OP(2) | INSN_OP3(0x26))
212
#define SHIFT_SRA (INSN_OP(2) | INSN_OP3(0x27))
214
#define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
215
#define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
216
#define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
218
#define RDY (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
219
#define WRY (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
220
#define JMPL (INSN_OP(2) | INSN_OP3(0x38))
221
#define RETURN (INSN_OP(2) | INSN_OP3(0x39))
222
#define SAVE (INSN_OP(2) | INSN_OP3(0x3c))
223
#define RESTORE (INSN_OP(2) | INSN_OP3(0x3d))
224
#define SETHI (INSN_OP(0) | INSN_OP2(0x4))
225
#define CALL INSN_OP(1)
226
#define LDUB (INSN_OP(3) | INSN_OP3(0x01))
227
#define LDSB (INSN_OP(3) | INSN_OP3(0x09))
228
#define LDUH (INSN_OP(3) | INSN_OP3(0x02))
229
#define LDSH (INSN_OP(3) | INSN_OP3(0x0a))
230
#define LDUW (INSN_OP(3) | INSN_OP3(0x00))
231
#define LDSW (INSN_OP(3) | INSN_OP3(0x08))
232
#define LDX (INSN_OP(3) | INSN_OP3(0x0b))
233
#define STB (INSN_OP(3) | INSN_OP3(0x05))
234
#define STH (INSN_OP(3) | INSN_OP3(0x06))
235
#define STW (INSN_OP(3) | INSN_OP3(0x04))
236
#define STX (INSN_OP(3) | INSN_OP3(0x0e))
237
#define LDUBA (INSN_OP(3) | INSN_OP3(0x11))
238
#define LDSBA (INSN_OP(3) | INSN_OP3(0x19))
239
#define LDUHA (INSN_OP(3) | INSN_OP3(0x12))
240
#define LDSHA (INSN_OP(3) | INSN_OP3(0x1a))
241
#define LDUWA (INSN_OP(3) | INSN_OP3(0x10))
242
#define LDSWA (INSN_OP(3) | INSN_OP3(0x18))
243
#define LDXA (INSN_OP(3) | INSN_OP3(0x1b))
244
#define STBA (INSN_OP(3) | INSN_OP3(0x15))
245
#define STHA (INSN_OP(3) | INSN_OP3(0x16))
246
#define STWA (INSN_OP(3) | INSN_OP3(0x14))
247
#define STXA (INSN_OP(3) | INSN_OP3(0x1e))
249
#define MEMBAR (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
251
#define NOP (SETHI | INSN_RD(TCG_REG_G0) | 0)
253
#ifndef ASI_PRIMARY_LITTLE
254
#define ASI_PRIMARY_LITTLE 0x88
257
#define LDUH_LE (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
258
#define LDSH_LE (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
259
#define LDUW_LE (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
260
#define LDSW_LE (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
261
#define LDX_LE (LDXA | INSN_ASI(ASI_PRIMARY_LITTLE))
263
#define STH_LE (STHA | INSN_ASI(ASI_PRIMARY_LITTLE))
264
#define STW_LE (STWA | INSN_ASI(ASI_PRIMARY_LITTLE))
265
#define STX_LE (STXA | INSN_ASI(ASI_PRIMARY_LITTLE))
267
#ifndef use_vis3_instructions
268
bool use_vis3_instructions;
271
static bool check_fit_i64(int64_t val, unsigned int bits)
273
return val == sextract64(val, 0, bits);
276
static bool check_fit_i32(int32_t val, unsigned int bits)
278
return val == sextract32(val, 0, bits);
281
#define check_fit_tl check_fit_i64
282
#define check_fit_ptr check_fit_i64
284
static bool patch_reloc(tcg_insn_unit *src_rw, int type,
285
intptr_t value, intptr_t addend)
287
const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
288
uint32_t insn = *src_rw;
292
pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, src_rx);
295
case R_SPARC_WDISP16:
296
if (!check_fit_ptr(pcrel >> 2, 16)) {
299
insn &= ~INSN_OFF16(-1);
300
insn |= INSN_OFF16(pcrel);
302
case R_SPARC_WDISP19:
303
if (!check_fit_ptr(pcrel >> 2, 19)) {
306
insn &= ~INSN_OFF19(-1);
307
insn |= INSN_OFF19(pcrel);
310
if (!check_fit_ptr(value, 13)) {
313
insn &= ~INSN_IMM13(-1);
314
insn |= INSN_IMM13(value);
317
g_assert_not_reached();
324
/* test if a constant matches the constraint */
325
static bool tcg_target_const_match(int64_t val, int ct,
326
TCGType type, TCGCond cond, int vece)
328
if (ct & TCG_CT_CONST) {
332
if (type == TCG_TYPE_I32) {
336
if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
338
} else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
340
} else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
347
static void tcg_out_nop(TCGContext *s)
352
static void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
355
tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
358
static void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
359
int32_t offset, int op)
361
tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
364
static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
365
int32_t val2, int val2const, int op)
367
tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
368
| (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
371
static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg)
374
tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
379
static void tcg_out_mov_delay(TCGContext *s, TCGReg ret, TCGReg arg)
382
tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
388
static void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
390
tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
393
/* A 13-bit constant sign-extended to 64 bits. */
394
static void tcg_out_movi_s13(TCGContext *s, TCGReg ret, int32_t arg)
396
tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
399
/* A 32-bit constant sign-extended to 64 bits. */
400
static void tcg_out_movi_s32(TCGContext *s, TCGReg ret, int32_t arg)
402
tcg_out_sethi(s, ret, ~arg);
403
tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
406
/* A 32-bit constant zero-extended to 64 bits. */
407
static void tcg_out_movi_u32(TCGContext *s, TCGReg ret, uint32_t arg)
409
tcg_out_sethi(s, ret, arg);
411
tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
415
static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
416
tcg_target_long arg, bool in_prologue,
419
tcg_target_long hi, lo = (int32_t)arg;
420
tcg_target_long test, lsb;
422
/* A 13-bit constant sign-extended to 64-bits. */
423
if (check_fit_tl(arg, 13)) {
424
tcg_out_movi_s13(s, ret, arg);
428
/* A 32-bit constant, or 32-bit zero-extended to 64-bits. */
429
if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
430
tcg_out_movi_u32(s, ret, arg);
434
/* A 13-bit constant relative to the TB. */
436
test = tcg_tbrel_diff(s, (void *)arg);
437
if (check_fit_ptr(test, 13)) {
438
tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
443
/* A 32-bit constant sign-extended to 64-bits. */
445
tcg_out_movi_s32(s, ret, arg);
449
/* A 32-bit constant, shifted. */
451
test = (tcg_target_long)arg >> lsb;
452
if (lsb > 10 && test == extract64(test, 0, 21)) {
453
tcg_out_sethi(s, ret, test << 10);
454
tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
456
} else if (test == (uint32_t)test || test == (int32_t)test) {
457
tcg_out_movi_int(s, TCG_TYPE_I64, ret, test, in_prologue, scratch);
458
tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
462
/* Use the constant pool, if possible. */
464
new_pool_label(s, arg, R_SPARC_13, s->code_ptr,
465
tcg_tbrel_diff(s, NULL));
466
tcg_out32(s, LDX | INSN_RD(ret) | INSN_RS1(TCG_REG_TB));
470
/* A 64-bit constant decomposed into 2 32-bit pieces. */
471
if (check_fit_i32(lo, 13)) {
472
hi = (arg - lo) >> 32;
473
tcg_out_movi_u32(s, ret, hi);
474
tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
475
tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
478
tcg_out_movi_u32(s, ret, hi);
479
tcg_out_movi_u32(s, scratch, lo);
480
tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
481
tcg_out_arith(s, ret, ret, scratch, ARITH_OR);
485
static void tcg_out_movi(TCGContext *s, TCGType type,
486
TCGReg ret, tcg_target_long arg)
488
tcg_debug_assert(ret != TCG_REG_T3);
489
tcg_out_movi_int(s, type, ret, arg, false, TCG_REG_T3);
492
static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
494
g_assert_not_reached();
497
static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg rd, TCGReg rs)
499
g_assert_not_reached();
502
static void tcg_out_ext8u(TCGContext *s, TCGReg rd, TCGReg rs)
504
tcg_out_arithi(s, rd, rs, 0xff, ARITH_AND);
507
static void tcg_out_ext16u(TCGContext *s, TCGReg rd, TCGReg rs)
509
tcg_out_arithi(s, rd, rs, 16, SHIFT_SLL);
510
tcg_out_arithi(s, rd, rd, 16, SHIFT_SRL);
513
static void tcg_out_ext32s(TCGContext *s, TCGReg rd, TCGReg rs)
515
tcg_out_arithi(s, rd, rs, 0, SHIFT_SRA);
518
static void tcg_out_ext32u(TCGContext *s, TCGReg rd, TCGReg rs)
520
tcg_out_arithi(s, rd, rs, 0, SHIFT_SRL);
523
static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
525
tcg_out_ext32s(s, rd, rs);
528
static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg rd, TCGReg rs)
530
tcg_out_ext32u(s, rd, rs);
533
static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg rd, TCGReg rs)
535
tcg_out_ext32u(s, rd, rs);
538
static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2)
543
static void tcg_out_addi_ptr(TCGContext *s, TCGReg rd, TCGReg rs,
546
/* This function is only used for passing structs by reference. */
547
g_assert_not_reached();
550
static void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
553
tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
556
static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
557
intptr_t offset, int op)
559
if (check_fit_ptr(offset, 13)) {
560
tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
563
tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
564
tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
568
static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
569
TCGReg arg1, intptr_t arg2)
571
tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
574
static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
575
TCGReg arg1, intptr_t arg2)
577
tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
580
static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
581
TCGReg base, intptr_t ofs)
584
tcg_out_st(s, type, TCG_REG_G0, base, ofs);
590
static void tcg_out_sety(TCGContext *s, TCGReg rs)
592
tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
595
static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1,
596
int32_t val2, int val2const, int uns)
598
/* Load Y with the sign/zero extension of RS1 to 64-bits. */
600
tcg_out_sety(s, TCG_REG_G0);
602
tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA);
603
tcg_out_sety(s, TCG_REG_T1);
606
tcg_out_arithc(s, rd, rs1, val2, val2const,
607
uns ? ARITH_UDIV : ARITH_SDIV);
610
static const uint8_t tcg_cond_to_bcond[16] = {
611
[TCG_COND_EQ] = COND_E,
612
[TCG_COND_NE] = COND_NE,
613
[TCG_COND_TSTEQ] = COND_E,
614
[TCG_COND_TSTNE] = COND_NE,
615
[TCG_COND_LT] = COND_L,
616
[TCG_COND_GE] = COND_GE,
617
[TCG_COND_LE] = COND_LE,
618
[TCG_COND_GT] = COND_G,
619
[TCG_COND_LTU] = COND_CS,
620
[TCG_COND_GEU] = COND_CC,
621
[TCG_COND_LEU] = COND_LEU,
622
[TCG_COND_GTU] = COND_GU,
625
static const uint8_t tcg_cond_to_rcond[16] = {
626
[TCG_COND_EQ] = RCOND_Z,
627
[TCG_COND_NE] = RCOND_NZ,
628
[TCG_COND_LT] = RCOND_LZ,
629
[TCG_COND_GT] = RCOND_GZ,
630
[TCG_COND_LE] = RCOND_LEZ,
631
[TCG_COND_GE] = RCOND_GEZ
634
static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
636
tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
639
static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
644
off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
646
tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
648
tcg_out_bpcc0(s, scond, flags, off19);
651
static void tcg_out_cmp(TCGContext *s, TCGCond cond,
652
TCGReg c1, int32_t c2, int c2const)
654
tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const,
655
is_tst_cond(cond) ? ARITH_ANDCC : ARITH_SUBCC);
658
static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
659
int32_t arg2, int const_arg2, TCGLabel *l)
661
tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
662
tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
666
static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret,
667
int32_t v1, int v1const)
669
tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret)
670
| INSN_RS1(tcg_cond_to_bcond[cond])
671
| (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
674
static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
675
TCGReg c1, int32_t c2, int c2const,
676
int32_t v1, int v1const)
678
tcg_out_cmp(s, cond, c1, c2, c2const);
679
tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const);
682
static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
683
int32_t arg2, int const_arg2, TCGLabel *l)
685
/* For 64-bit signed comparisons vs zero, we can avoid the compare. */
686
int rcond = tcg_cond_to_rcond[cond];
687
if (arg2 == 0 && rcond) {
691
off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
693
tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
695
tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
696
| INSN_COND(rcond) | off16);
698
tcg_out_cmp(s, cond, arg1, arg2, const_arg2);
699
tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
704
static void tcg_out_movr(TCGContext *s, int rcond, TCGReg ret, TCGReg c1,
705
int32_t v1, int v1const)
707
tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | (rcond << 10)
708
| (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
711
static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
712
TCGReg c1, int32_t c2, int c2const,
713
int32_t v1, int v1const)
715
/* For 64-bit signed comparisons vs zero, we can avoid the compare.
716
Note that the immediate range is one bit smaller, so we must check
718
int rcond = tcg_cond_to_rcond[cond];
719
if (c2 == 0 && rcond && (!v1const || check_fit_i32(v1, 10))) {
720
tcg_out_movr(s, rcond, ret, c1, v1, v1const);
722
tcg_out_cmp(s, cond, c1, c2, c2const);
723
tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const);
727
static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
728
TCGReg c1, int32_t c2, int c2const, bool neg)
730
/* For 32-bit comparisons, we can play games with ADDC/SUBC. */
734
/* The result of the comparison is in the carry bit. */
739
/* For equality, we can transform to inequality vs zero. */
741
tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
746
c1 = TCG_REG_G0, c2const = 0;
747
cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
752
/* Transform to inequality vs zero. */
753
tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_AND);
755
c2 = TCG_REG_T1, c2const = 0;
756
cond = (cond == TCG_COND_TSTEQ ? TCG_COND_GEU : TCG_COND_LTU);
761
/* If we don't need to load a constant into a register, we can
762
swap the operands on GTU/LEU. There's no benefit to loading
763
the constant into a temporary register. */
764
if (!c2const || c2 == 0) {
769
cond = tcg_swap_cond(cond);
775
tcg_out_cmp(s, cond, c1, c2, c2const);
776
tcg_out_movi_s13(s, ret, 0);
777
tcg_out_movcc(s, cond, MOVCC_ICC, ret, neg ? -1 : 1, 1);
781
tcg_out_cmp(s, cond, c1, c2, c2const);
782
if (cond == TCG_COND_LTU) {
784
/* 0 - 0 - C = -C = (C ? -1 : 0) */
785
tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_SUBC);
787
/* 0 + 0 + C = C = (C ? 1 : 0) */
788
tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
792
/* 0 + -1 + C = C - 1 = (C ? 0 : -1) */
793
tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_ADDC);
795
/* 0 - -1 - C = 1 - C = (C ? 0 : 1) */
796
tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
801
static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
802
TCGReg c1, int32_t c2, int c2const, bool neg)
806
if (use_vis3_instructions && !neg) {
812
c2 = c1, c2const = 0, c1 = TCG_REG_G0;
815
tcg_out_cmp(s, cond, c1, c2, c2const);
816
tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
823
/* For 64-bit signed comparisons vs zero, we can avoid the compare
824
if the input does not overlap the output. */
825
rcond = tcg_cond_to_rcond[cond];
826
if (c2 == 0 && rcond && c1 != ret) {
827
tcg_out_movi_s13(s, ret, 0);
828
tcg_out_movr(s, rcond, ret, c1, neg ? -1 : 1, 1);
830
tcg_out_cmp(s, cond, c1, c2, c2const);
831
tcg_out_movi_s13(s, ret, 0);
832
tcg_out_movcc(s, cond, MOVCC_XCC, ret, neg ? -1 : 1, 1);
836
static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh,
837
TCGReg al, TCGReg ah, int32_t bl, int blconst,
838
int32_t bh, int bhconst, int opl, int oph)
840
TCGReg tmp = TCG_REG_T1;
842
/* Note that the low parts are fully consumed before tmp is set. */
843
if (rl != ah && (bhconst || rl != bh)) {
847
tcg_out_arithc(s, tmp, al, bl, blconst, opl);
848
tcg_out_arithc(s, rh, ah, bh, bhconst, oph);
849
tcg_out_mov(s, TCG_TYPE_I32, rl, tmp);
852
static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
853
TCGReg al, TCGReg ah, int32_t bl, int blconst,
854
int32_t bh, int bhconst, bool is_sub)
856
TCGReg tmp = TCG_REG_T1;
858
/* Note that the low parts are fully consumed before tmp is set. */
859
if (rl != ah && (bhconst || rl != bh)) {
863
tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
865
if (use_vis3_instructions && !is_sub) {
866
/* Note that ADDXC doesn't accept immediates. */
867
if (bhconst && bh != 0) {
868
tcg_out_movi_s13(s, TCG_REG_T2, bh);
871
tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC);
872
} else if (bh == TCG_REG_G0) {
873
/* If we have a zero, we can perform the operation in two insns,
874
with the arithmetic first, and a conditional move into place. */
876
tcg_out_arithi(s, TCG_REG_T2, ah, 1,
877
is_sub ? ARITH_SUB : ARITH_ADD);
878
tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0);
880
tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD);
881
tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0);
885
* Otherwise adjust BH as if there is carry into T2.
886
* Note that constant BH is constrained to 11 bits for the MOVCC,
887
* so the adjustment fits 12 bits.
890
tcg_out_movi_s13(s, TCG_REG_T2, bh + (is_sub ? -1 : 1));
892
tcg_out_arithi(s, TCG_REG_T2, bh, 1,
893
is_sub ? ARITH_SUB : ARITH_ADD);
895
/* ... smoosh T2 back to original BH if carry is clear ... */
896
tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst);
897
/* ... and finally perform the arithmetic with the new operand. */
898
tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD);
901
tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
904
static void tcg_out_jmpl_const(TCGContext *s, const tcg_insn_unit *dest,
905
bool in_prologue, bool tail_call)
907
uintptr_t desti = (uintptr_t)dest;
909
tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
910
desti & ~0xfff, in_prologue, TCG_REG_T2);
911
tcg_out_arithi(s, tail_call ? TCG_REG_G0 : TCG_REG_O7,
912
TCG_REG_T1, desti & 0xfff, JMPL);
915
static void tcg_out_call_nodelay(TCGContext *s, const tcg_insn_unit *dest,
918
ptrdiff_t disp = tcg_pcrel_diff(s, dest);
920
if (disp == (int32_t)disp) {
921
tcg_out32(s, CALL | (uint32_t)disp >> 2);
923
tcg_out_jmpl_const(s, dest, in_prologue, false);
927
static void tcg_out_call(TCGContext *s, const tcg_insn_unit *dest,
928
const TCGHelperInfo *info)
930
tcg_out_call_nodelay(s, dest, false);
934
static void tcg_out_mb(TCGContext *s, TCGArg a0)
936
/* Note that the TCG memory order constants mirror the Sparc MEMBAR. */
937
tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
940
/* Generate global QEMU prologue and epilogue code */
941
static void tcg_target_qemu_prologue(TCGContext *s)
943
int tmp_buf_size, frame_size;
946
* The TCG temp buffer is at the top of the frame, immediately
947
* below the frame pointer. Use the logical (aligned) offset here;
948
* the stack bias is applied in temp_allocate_frame().
950
tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
951
tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
954
* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
955
* otherwise the minimal frame usable by callees.
957
frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
958
frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
959
frame_size += TCG_TARGET_STACK_ALIGN - 1;
960
frame_size &= -TCG_TARGET_STACK_ALIGN;
961
tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
962
INSN_IMM13(-frame_size));
964
#ifndef CONFIG_SOFTMMU
965
if (guest_base != 0) {
966
tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG,
967
guest_base, true, TCG_REG_T1);
968
tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
972
/* We choose TCG_REG_TB such that no move is required. */
973
QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
974
tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
976
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
980
/* Epilogue for goto_ptr. */
981
tcg_code_gen_epilogue = tcg_splitwx_to_rx(s->code_ptr);
982
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
984
tcg_out_movi_s13(s, TCG_REG_O0, 0);
987
static void tcg_out_tb_start(TCGContext *s)
992
static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
995
for (i = 0; i < count; ++i) {
1000
static const TCGLdstHelperParam ldst_helper_param = {
1001
.ntmp = 1, .tmp = { TCG_REG_T1 }
1004
static bool tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1006
MemOp opc = get_memop(lb->oi);
1009
if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1010
(intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1014
/* Use inline tcg_out_ext32s; otherwise let the helper sign-extend. */
1015
sgn = (opc & MO_SIZE) < MO_32 ? MO_SIGN : 0;
1017
tcg_out_ld_helper_args(s, lb, &ldst_helper_param);
1018
tcg_out_call(s, qemu_ld_helpers[opc & (MO_SIZE | sgn)], NULL);
1019
tcg_out_ld_helper_ret(s, lb, sgn, &ldst_helper_param);
1021
tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1022
return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1023
(intptr_t)lb->raddr, 0);
1026
static bool tcg_out_qemu_st_slow_path(TCGContext *s, TCGLabelQemuLdst *lb)
1028
MemOp opc = get_memop(lb->oi);
1030
if (!patch_reloc(lb->label_ptr[0], R_SPARC_WDISP19,
1031
(intptr_t)tcg_splitwx_to_rx(s->code_ptr), 0)) {
1035
tcg_out_st_helper_args(s, lb, &ldst_helper_param);
1036
tcg_out_call(s, qemu_st_helpers[opc & MO_SIZE], NULL);
1038
tcg_out_bpcc0(s, COND_A, BPCC_A | BPCC_PT, 0);
1039
return patch_reloc(s->code_ptr - 1, R_SPARC_WDISP19,
1040
(intptr_t)lb->raddr, 0);
1049
bool tcg_target_has_memory_bswap(MemOp memop)
1054
/* We expect to use a 13-bit negative offset from ENV. */
1055
#define MIN_TLB_MASK_TABLE_OFS -(1 << 12)
1058
* For system-mode, perform the TLB load and compare.
1059
* For user-mode, perform any required alignment tests.
1060
* In both cases, return a TCGLabelQemuLdst structure if the slow path
1061
* is required and fill in @h with the host address for the fast path.
1063
static TCGLabelQemuLdst *prepare_host_addr(TCGContext *s, HostAddress *h,
1064
TCGReg addr_reg, MemOpIdx oi,
1067
TCGType addr_type = s->addr_type;
1068
TCGLabelQemuLdst *ldst = NULL;
1069
MemOp opc = get_memop(oi);
1070
MemOp s_bits = opc & MO_SIZE;
1073
/* We don't support unaligned accesses. */
1074
h->aa = atom_and_align_for_opc(s, opc, MO_ATOM_IFALIGN, false);
1075
h->aa.align = MAX(h->aa.align, s_bits);
1076
a_mask = (1u << h->aa.align) - 1;
1078
#ifdef CONFIG_SOFTMMU
1079
int mem_index = get_mmuidx(oi);
1080
int fast_off = tlb_mask_table_ofs(s, mem_index);
1081
int mask_off = fast_off + offsetof(CPUTLBDescFast, mask);
1082
int table_off = fast_off + offsetof(CPUTLBDescFast, table);
1083
int cmp_off = is_ld ? offsetof(CPUTLBEntry, addr_read)
1084
: offsetof(CPUTLBEntry, addr_write);
1085
int add_off = offsetof(CPUTLBEntry, addend);
1089
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
1090
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T2, TCG_AREG0, mask_off);
1091
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T3, TCG_AREG0, table_off);
1093
/* Extract the page index, shifted into place for tlb index. */
1094
tcg_out_arithi(s, TCG_REG_T1, addr_reg,
1095
s->page_bits - CPU_TLB_ENTRY_BITS, SHIFT_SRL);
1096
tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T2, ARITH_AND);
1098
/* Add the tlb_table pointer, creating the CPUTLBEntry address into R2. */
1099
tcg_out_arith(s, TCG_REG_T1, TCG_REG_T1, TCG_REG_T3, ARITH_ADD);
1102
* Load the tlb comparator and the addend.
1103
* Always load the entire 64-bit comparator for simplicity.
1104
* We will ignore the high bits via BPCC_ICC below.
1106
tcg_out_ld(s, TCG_TYPE_I64, TCG_REG_T2, TCG_REG_T1, cmp_off);
1107
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_T1, TCG_REG_T1, add_off);
1108
h->base = TCG_REG_T1;
1110
/* Mask out the page offset, except for the required alignment. */
1111
compare_mask = s->page_mask | a_mask;
1112
if (check_fit_tl(compare_mask, 13)) {
1113
tcg_out_arithi(s, TCG_REG_T3, addr_reg, compare_mask, ARITH_AND);
1115
tcg_out_movi_s32(s, TCG_REG_T3, compare_mask);
1116
tcg_out_arith(s, TCG_REG_T3, addr_reg, TCG_REG_T3, ARITH_AND);
1118
tcg_out_cmp(s, TCG_COND_NE, TCG_REG_T2, TCG_REG_T3, 0);
1120
ldst = new_ldst_label(s);
1121
ldst->is_ld = is_ld;
1123
ldst->addrlo_reg = addr_reg;
1124
ldst->label_ptr[0] = s->code_ptr;
1126
/* bne,pn %[xi]cc, label0 */
1127
cc = addr_type == TCG_TYPE_I32 ? BPCC_ICC : BPCC_XCC;
1128
tcg_out_bpcc0(s, COND_NE, BPCC_PN | cc, 0);
1131
* If the size equals the required alignment, we can skip the test
1132
* and allow host SIGBUS to deliver SIGBUS to the guest.
1133
* Otherwise, test for at least natural alignment and defer
1134
* everything else to the helper functions.
1136
if (s_bits != get_alignment_bits(opc)) {
1137
tcg_debug_assert(check_fit_tl(a_mask, 13));
1138
tcg_out_arithi(s, TCG_REG_G0, addr_reg, a_mask, ARITH_ANDCC);
1140
ldst = new_ldst_label(s);
1141
ldst->is_ld = is_ld;
1143
ldst->addrlo_reg = addr_reg;
1144
ldst->label_ptr[0] = s->code_ptr;
1146
/* bne,pn %icc, label0 */
1147
tcg_out_bpcc0(s, COND_NE, BPCC_PN | BPCC_ICC, 0);
1149
h->base = guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0;
1152
/* If the guest address must be zero-extended, do in the delay slot. */
1153
if (addr_type == TCG_TYPE_I32) {
1154
tcg_out_ext32u(s, TCG_REG_T2, addr_reg);
1155
h->index = TCG_REG_T2;
1160
h->index = addr_reg;
1165
static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
1166
MemOpIdx oi, TCGType data_type)
1168
static const int ld_opc[(MO_SSIZE | MO_BSWAP) + 1] = {
1171
[MO_UB | MO_LE] = LDUB,
1172
[MO_SB | MO_LE] = LDSB,
1181
[MO_LEUW] = LDUH_LE,
1182
[MO_LESW] = LDSH_LE,
1183
[MO_LEUL] = LDUW_LE,
1184
[MO_LESL] = LDSW_LE,
1189
TCGLabelQemuLdst *ldst;
1192
ldst = prepare_host_addr(s, &h, addr, oi, true);
1194
tcg_out_ldst_rr(s, data, h.base, h.index,
1195
ld_opc[get_memop(oi) & (MO_BSWAP | MO_SSIZE)]);
1198
ldst->type = data_type;
1199
ldst->datalo_reg = data;
1200
ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1204
static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr,
1205
MemOpIdx oi, TCGType data_type)
1207
static const int st_opc[(MO_SIZE | MO_BSWAP) + 1] = {
1219
TCGLabelQemuLdst *ldst;
1222
ldst = prepare_host_addr(s, &h, addr, oi, false);
1224
tcg_out_ldst_rr(s, data, h.base, h.index,
1225
st_opc[get_memop(oi) & (MO_BSWAP | MO_SIZE)]);
1228
ldst->type = data_type;
1229
ldst->datalo_reg = data;
1230
ldst->raddr = tcg_splitwx_to_rx(s->code_ptr);
1234
static void tcg_out_exit_tb(TCGContext *s, uintptr_t a0)
1236
if (check_fit_ptr(a0, 13)) {
1237
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1238
tcg_out_movi_s13(s, TCG_REG_O0, a0);
1241
intptr_t tb_diff = tcg_tbrel_diff(s, (void *)a0);
1242
if (check_fit_ptr(tb_diff, 13)) {
1243
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1244
/* Note that TCG_REG_TB has been unwound to O1. */
1245
tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1249
tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1250
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1251
tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1254
static void tcg_out_goto_tb(TCGContext *s, int which)
1256
ptrdiff_t off = tcg_tbrel_diff(s, (void *)get_jmp_target_addr(s, which));
1258
/* Load link and indirect branch. */
1259
set_jmp_insn_offset(s, which);
1260
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TB, TCG_REG_TB, off);
1261
tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1264
set_jmp_reset_offset(s, which);
1267
* For the unlinked path of goto_tb, we need to reset TCG_REG_TB
1268
* to the beginning of this TB.
1270
off = -tcg_current_code_size(s);
1271
if (check_fit_i32(off, 13)) {
1272
tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, off, ARITH_ADD);
1274
tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, off);
1275
tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1279
void tb_target_set_jmp_target(const TranslationBlock *tb, int n,
1280
uintptr_t jmp_rx, uintptr_t jmp_rw)
1284
static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1285
const TCGArg args[TCG_MAX_OP_ARGS],
1286
const int const_args[TCG_MAX_OP_ARGS])
1291
/* Hoist the loads of the most common arguments. */
1298
case INDEX_op_goto_ptr:
1299
tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1300
tcg_out_mov_delay(s, TCG_REG_TB, a0);
1303
tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0));
1307
#define OP_32_64(x) \
1308
glue(glue(case INDEX_op_, x), _i32): \
1309
glue(glue(case INDEX_op_, x), _i64)
1312
tcg_out_ldst(s, a0, a1, a2, LDUB);
1315
tcg_out_ldst(s, a0, a1, a2, LDSB);
1318
tcg_out_ldst(s, a0, a1, a2, LDUH);
1321
tcg_out_ldst(s, a0, a1, a2, LDSH);
1323
case INDEX_op_ld_i32:
1324
case INDEX_op_ld32u_i64:
1325
tcg_out_ldst(s, a0, a1, a2, LDUW);
1328
tcg_out_ldst(s, a0, a1, a2, STB);
1331
tcg_out_ldst(s, a0, a1, a2, STH);
1333
case INDEX_op_st_i32:
1334
case INDEX_op_st32_i64:
1335
tcg_out_ldst(s, a0, a1, a2, STW);
1358
case INDEX_op_shl_i32:
1361
/* Limit immediate shift count lest we create an illegal insn. */
1362
tcg_out_arithc(s, a0, a1, a2 & 31, c2, c);
1364
case INDEX_op_shr_i32:
1367
case INDEX_op_sar_i32:
1370
case INDEX_op_mul_i32:
1381
case INDEX_op_div_i32:
1382
tcg_out_div32(s, a0, a1, a2, c2, 0);
1384
case INDEX_op_divu_i32:
1385
tcg_out_div32(s, a0, a1, a2, c2, 1);
1388
case INDEX_op_brcond_i32:
1389
tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1391
case INDEX_op_setcond_i32:
1392
tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2, false);
1394
case INDEX_op_negsetcond_i32:
1395
tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2, true);
1397
case INDEX_op_movcond_i32:
1398
tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1401
case INDEX_op_add2_i32:
1402
tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1403
args[4], const_args[4], args[5], const_args[5],
1404
ARITH_ADDCC, ARITH_ADDC);
1406
case INDEX_op_sub2_i32:
1407
tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1408
args[4], const_args[4], args[5], const_args[5],
1409
ARITH_SUBCC, ARITH_SUBC);
1411
case INDEX_op_mulu2_i32:
1414
case INDEX_op_muls2_i32:
1417
/* The 32-bit multiply insns produce a full 64-bit result. */
1418
tcg_out_arithc(s, a0, a2, args[3], const_args[3], c);
1419
tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1422
case INDEX_op_qemu_ld_a32_i32:
1423
case INDEX_op_qemu_ld_a64_i32:
1424
tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I32);
1426
case INDEX_op_qemu_ld_a32_i64:
1427
case INDEX_op_qemu_ld_a64_i64:
1428
tcg_out_qemu_ld(s, a0, a1, a2, TCG_TYPE_I64);
1430
case INDEX_op_qemu_st_a32_i32:
1431
case INDEX_op_qemu_st_a64_i32:
1432
tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I32);
1434
case INDEX_op_qemu_st_a32_i64:
1435
case INDEX_op_qemu_st_a64_i64:
1436
tcg_out_qemu_st(s, a0, a1, a2, TCG_TYPE_I64);
1439
case INDEX_op_ld32s_i64:
1440
tcg_out_ldst(s, a0, a1, a2, LDSW);
1442
case INDEX_op_ld_i64:
1443
tcg_out_ldst(s, a0, a1, a2, LDX);
1445
case INDEX_op_st_i64:
1446
tcg_out_ldst(s, a0, a1, a2, STX);
1448
case INDEX_op_shl_i64:
1451
/* Limit immediate shift count lest we create an illegal insn. */
1452
tcg_out_arithc(s, a0, a1, a2 & 63, c2, c);
1454
case INDEX_op_shr_i64:
1457
case INDEX_op_sar_i64:
1460
case INDEX_op_mul_i64:
1463
case INDEX_op_div_i64:
1466
case INDEX_op_divu_i64:
1470
case INDEX_op_brcond_i64:
1471
tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1473
case INDEX_op_setcond_i64:
1474
tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2, false);
1476
case INDEX_op_negsetcond_i64:
1477
tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2, true);
1479
case INDEX_op_movcond_i64:
1480
tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1482
case INDEX_op_add2_i64:
1483
tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1484
const_args[4], args[5], const_args[5], false);
1486
case INDEX_op_sub2_i64:
1487
tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1488
const_args[4], args[5], const_args[5], true);
1490
case INDEX_op_muluh_i64:
1491
tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI);
1495
tcg_out_arithc(s, a0, a1, a2, c2, c);
1499
tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c);
1506
case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
1507
case INDEX_op_mov_i64:
1508
case INDEX_op_call: /* Always emitted via tcg_out_call. */
1509
case INDEX_op_exit_tb: /* Always emitted via tcg_out_exit_tb. */
1510
case INDEX_op_goto_tb: /* Always emitted via tcg_out_goto_tb. */
1511
case INDEX_op_ext8s_i32: /* Always emitted via tcg_reg_alloc_op. */
1512
case INDEX_op_ext8s_i64:
1513
case INDEX_op_ext8u_i32:
1514
case INDEX_op_ext8u_i64:
1515
case INDEX_op_ext16s_i32:
1516
case INDEX_op_ext16s_i64:
1517
case INDEX_op_ext16u_i32:
1518
case INDEX_op_ext16u_i64:
1519
case INDEX_op_ext32s_i64:
1520
case INDEX_op_ext32u_i64:
1521
case INDEX_op_ext_i32_i64:
1522
case INDEX_op_extu_i32_i64:
1524
g_assert_not_reached();
1528
static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
1531
case INDEX_op_goto_ptr:
1534
case INDEX_op_ld8u_i32:
1535
case INDEX_op_ld8u_i64:
1536
case INDEX_op_ld8s_i32:
1537
case INDEX_op_ld8s_i64:
1538
case INDEX_op_ld16u_i32:
1539
case INDEX_op_ld16u_i64:
1540
case INDEX_op_ld16s_i32:
1541
case INDEX_op_ld16s_i64:
1542
case INDEX_op_ld_i32:
1543
case INDEX_op_ld32u_i64:
1544
case INDEX_op_ld32s_i64:
1545
case INDEX_op_ld_i64:
1546
case INDEX_op_neg_i32:
1547
case INDEX_op_neg_i64:
1548
case INDEX_op_not_i32:
1549
case INDEX_op_not_i64:
1550
case INDEX_op_ext32s_i64:
1551
case INDEX_op_ext32u_i64:
1552
case INDEX_op_ext_i32_i64:
1553
case INDEX_op_extu_i32_i64:
1554
case INDEX_op_qemu_ld_a32_i32:
1555
case INDEX_op_qemu_ld_a64_i32:
1556
case INDEX_op_qemu_ld_a32_i64:
1557
case INDEX_op_qemu_ld_a64_i64:
1558
return C_O1_I1(r, r);
1560
case INDEX_op_st8_i32:
1561
case INDEX_op_st8_i64:
1562
case INDEX_op_st16_i32:
1563
case INDEX_op_st16_i64:
1564
case INDEX_op_st_i32:
1565
case INDEX_op_st32_i64:
1566
case INDEX_op_st_i64:
1567
case INDEX_op_qemu_st_a32_i32:
1568
case INDEX_op_qemu_st_a64_i32:
1569
case INDEX_op_qemu_st_a32_i64:
1570
case INDEX_op_qemu_st_a64_i64:
1571
return C_O0_I2(rZ, r);
1573
case INDEX_op_add_i32:
1574
case INDEX_op_add_i64:
1575
case INDEX_op_mul_i32:
1576
case INDEX_op_mul_i64:
1577
case INDEX_op_div_i32:
1578
case INDEX_op_div_i64:
1579
case INDEX_op_divu_i32:
1580
case INDEX_op_divu_i64:
1581
case INDEX_op_sub_i32:
1582
case INDEX_op_sub_i64:
1583
case INDEX_op_and_i32:
1584
case INDEX_op_and_i64:
1585
case INDEX_op_andc_i32:
1586
case INDEX_op_andc_i64:
1587
case INDEX_op_or_i32:
1588
case INDEX_op_or_i64:
1589
case INDEX_op_orc_i32:
1590
case INDEX_op_orc_i64:
1591
case INDEX_op_xor_i32:
1592
case INDEX_op_xor_i64:
1593
case INDEX_op_shl_i32:
1594
case INDEX_op_shl_i64:
1595
case INDEX_op_shr_i32:
1596
case INDEX_op_shr_i64:
1597
case INDEX_op_sar_i32:
1598
case INDEX_op_sar_i64:
1599
case INDEX_op_setcond_i32:
1600
case INDEX_op_setcond_i64:
1601
case INDEX_op_negsetcond_i32:
1602
case INDEX_op_negsetcond_i64:
1603
return C_O1_I2(r, rZ, rJ);
1605
case INDEX_op_brcond_i32:
1606
case INDEX_op_brcond_i64:
1607
return C_O0_I2(rZ, rJ);
1608
case INDEX_op_movcond_i32:
1609
case INDEX_op_movcond_i64:
1610
return C_O1_I4(r, rZ, rJ, rI, 0);
1611
case INDEX_op_add2_i32:
1612
case INDEX_op_add2_i64:
1613
case INDEX_op_sub2_i32:
1614
case INDEX_op_sub2_i64:
1615
return C_O2_I4(r, r, rZ, rZ, rJ, rJ);
1616
case INDEX_op_mulu2_i32:
1617
case INDEX_op_muls2_i32:
1618
return C_O2_I2(r, r, rZ, rJ);
1619
case INDEX_op_muluh_i64:
1620
return C_O1_I2(r, r, r);
1623
g_assert_not_reached();
1627
static void tcg_target_init(TCGContext *s)
1630
* Only probe for the platform and capabilities if we haven't already
1631
* determined maximum values at compile time.
1633
#ifndef use_vis3_instructions
1635
unsigned long hwcap = qemu_getauxval(AT_HWCAP);
1636
use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
1640
tcg_target_available_regs[TCG_TYPE_I32] = ALL_GENERAL_REGS;
1641
tcg_target_available_regs[TCG_TYPE_I64] = ALL_GENERAL_REGS;
1643
tcg_target_call_clobber_regs = 0;
1644
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
1645
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
1646
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
1647
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
1648
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
1649
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
1650
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
1651
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
1652
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
1653
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
1654
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
1655
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
1656
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
1657
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
1658
tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
1660
s->reserved_regs = 0;
1661
tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
1662
tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
1663
tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
1664
tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
1665
tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
1666
tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
1667
tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
1668
tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
1669
tcg_regset_set_reg(s->reserved_regs, TCG_REG_T3); /* for internal use */
1672
#define ELF_HOST_MACHINE EM_SPARCV9
1676
uint8_t fde_def_cfa[4];
1677
uint8_t fde_win_save;
1678
uint8_t fde_ret_save[3];
1681
static const DebugFrame debug_frame = {
1682
.h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
1685
.h.cie.code_align = 1,
1686
.h.cie.data_align = -sizeof(void *) & 0x7f,
1687
.h.cie.return_column = 15, /* o7 */
1689
/* Total FDE size does not include the "len" member. */
1690
.h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1693
12, 30, /* DW_CFA_def_cfa i6, 2047 */
1694
(2047 & 0x7f) | 0x80, (2047 >> 7)
1696
.fde_win_save = 0x2d, /* DW_CFA_GNU_window_save */
1697
.fde_ret_save = { 9, 15, 31 }, /* DW_CFA_register o7, i7 */
1700
void tcg_register_jit(const void *buf, size_t buf_size)
1702
tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));