?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
target/ppc: convert xxsel to vector operations
[qemu/kevin.git] / tcg / sparc / tcg-target.inc.c
blob7a61839dc10dc5ea26cafcad8c4bcef8a483165b
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 #include "tcg-pool.inc.c"
26
27 #ifdef CONFIG_DEBUG_TCG
28 static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = {
29 "%g0",
30 "%g1",
31 "%g2",
32 "%g3",
33 "%g4",
34 "%g5",
35 "%g6",
36 "%g7",
37 "%o0",
38 "%o1",
39 "%o2",
40 "%o3",
41 "%o4",
42 "%o5",
43 "%o6",
44 "%o7",
45 "%l0",
46 "%l1",
47 "%l2",
48 "%l3",
49 "%l4",
50 "%l5",
51 "%l6",
52 "%l7",
53 "%i0",
54 "%i1",
55 "%i2",
56 "%i3",
57 "%i4",
58 "%i5",
59 "%i6",
60 "%i7",
61 };
62 #endif
63
64 #ifdef __arch64__
65 # define SPARC64 1
66 #else
67 # define SPARC64 0
68 #endif
69
70 /* Note that sparcv8plus can only hold 64 bit quantities in %g and %o
71 registers. These are saved manually by the kernel in full 64-bit
72 slots. The %i and %l registers are saved by the register window
73 mechanism, which only allocates space for 32 bits. Given that this
74 window spill/fill can happen on any signal, we must consider the
75 high bits of the %i and %l registers garbage at all times. */
76 #if SPARC64
77 # define ALL_64 0xffffffffu
78 #else
79 # define ALL_64 0xffffu
80 #endif
81
82 /* Define some temporary registers. T2 is used for constant generation. */
83 #define TCG_REG_T1 TCG_REG_G1
84 #define TCG_REG_T2 TCG_REG_O7
85
86 #ifndef CONFIG_SOFTMMU
87 # define TCG_GUEST_BASE_REG TCG_REG_I5
88 #endif
89
90 #define TCG_REG_TB TCG_REG_I1
91 #define USE_REG_TB (sizeof(void *) > 4)
92
93 static 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_G2,
111 TCG_REG_G3,
112 TCG_REG_G4,
113 TCG_REG_G5,
114
115 TCG_REG_O0,
116 TCG_REG_O1,
117 TCG_REG_O2,
118 TCG_REG_O3,
119 TCG_REG_O4,
120 TCG_REG_O5,
121 };
122
123 static const int tcg_target_call_iarg_regs[6] = {
124 TCG_REG_O0,
125 TCG_REG_O1,
126 TCG_REG_O2,
127 TCG_REG_O3,
128 TCG_REG_O4,
129 TCG_REG_O5,
130 };
131
132 static const int tcg_target_call_oarg_regs[] = {
133 TCG_REG_O0,
134 TCG_REG_O1,
135 TCG_REG_O2,
136 TCG_REG_O3,
137 };
138
139 #define INSN_OP(x) ((x) << 30)
140 #define INSN_OP2(x) ((x) << 22)
141 #define INSN_OP3(x) ((x) << 19)
142 #define INSN_OPF(x) ((x) << 5)
143 #define INSN_RD(x) ((x) << 25)
144 #define INSN_RS1(x) ((x) << 14)
145 #define INSN_RS2(x) (x)
146 #define INSN_ASI(x) ((x) << 5)
147
148 #define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff))
149 #define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff))
150 #define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff))
151 #define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20))
152 #define INSN_OFF19(x) (((x) >> 2) & 0x07ffff)
153 #define INSN_COND(x) ((x) << 25)
154
155 #define COND_N 0x0
156 #define COND_E 0x1
157 #define COND_LE 0x2
158 #define COND_L 0x3
159 #define COND_LEU 0x4
160 #define COND_CS 0x5
161 #define COND_NEG 0x6
162 #define COND_VS 0x7
163 #define COND_A 0x8
164 #define COND_NE 0x9
165 #define COND_G 0xa
166 #define COND_GE 0xb
167 #define COND_GU 0xc
168 #define COND_CC 0xd
169 #define COND_POS 0xe
170 #define COND_VC 0xf
171 #define BA (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2))
172
173 #define RCOND_Z 1
174 #define RCOND_LEZ 2
175 #define RCOND_LZ 3
176 #define RCOND_NZ 5
177 #define RCOND_GZ 6
178 #define RCOND_GEZ 7
179
180 #define MOVCC_ICC (1 << 18)
181 #define MOVCC_XCC (1 << 18 | 1 << 12)
182
183 #define BPCC_ICC 0
184 #define BPCC_XCC (2 << 20)
185 #define BPCC_PT (1 << 19)
186 #define BPCC_PN 0
187 #define BPCC_A (1 << 29)
188
189 #define BPR_PT BPCC_PT
190
191 #define ARITH_ADD (INSN_OP(2) | INSN_OP3(0x00))
192 #define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10))
193 #define ARITH_AND (INSN_OP(2) | INSN_OP3(0x01))
194 #define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05))
195 #define ARITH_OR (INSN_OP(2) | INSN_OP3(0x02))
196 #define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12))
197 #define ARITH_ORN (INSN_OP(2) | INSN_OP3(0x06))
198 #define ARITH_XOR (INSN_OP(2) | INSN_OP3(0x03))
199 #define ARITH_SUB (INSN_OP(2) | INSN_OP3(0x04))
200 #define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14))
201 #define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08))
202 #define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
203 #define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
204 #define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
205 #define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
206 #define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f))
207 #define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09))
208 #define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d))
209 #define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d))
210 #define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
211 #define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
212
213 #define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
214 #define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
215
216 #define SHIFT_SLL (INSN_OP(2) | INSN_OP3(0x25))
217 #define SHIFT_SRL (INSN_OP(2) | INSN_OP3(0x26))
218 #define SHIFT_SRA (INSN_OP(2) | INSN_OP3(0x27))
219
220 #define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12))
221 #define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12))
222 #define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12))
223
224 #define RDY (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0))
225 #define WRY (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0))
226 #define JMPL (INSN_OP(2) | INSN_OP3(0x38))
227 #define RETURN (INSN_OP(2) | INSN_OP3(0x39))
228 #define SAVE (INSN_OP(2) | INSN_OP3(0x3c))
229 #define RESTORE (INSN_OP(2) | INSN_OP3(0x3d))
230 #define SETHI (INSN_OP(0) | INSN_OP2(0x4))
231 #define CALL INSN_OP(1)
232 #define LDUB (INSN_OP(3) | INSN_OP3(0x01))
233 #define LDSB (INSN_OP(3) | INSN_OP3(0x09))
234 #define LDUH (INSN_OP(3) | INSN_OP3(0x02))
235 #define LDSH (INSN_OP(3) | INSN_OP3(0x0a))
236 #define LDUW (INSN_OP(3) | INSN_OP3(0x00))
237 #define LDSW (INSN_OP(3) | INSN_OP3(0x08))
238 #define LDX (INSN_OP(3) | INSN_OP3(0x0b))
239 #define STB (INSN_OP(3) | INSN_OP3(0x05))
240 #define STH (INSN_OP(3) | INSN_OP3(0x06))
241 #define STW (INSN_OP(3) | INSN_OP3(0x04))
242 #define STX (INSN_OP(3) | INSN_OP3(0x0e))
243 #define LDUBA (INSN_OP(3) | INSN_OP3(0x11))
244 #define LDSBA (INSN_OP(3) | INSN_OP3(0x19))
245 #define LDUHA (INSN_OP(3) | INSN_OP3(0x12))
246 #define LDSHA (INSN_OP(3) | INSN_OP3(0x1a))
247 #define LDUWA (INSN_OP(3) | INSN_OP3(0x10))
248 #define LDSWA (INSN_OP(3) | INSN_OP3(0x18))
249 #define LDXA (INSN_OP(3) | INSN_OP3(0x1b))
250 #define STBA (INSN_OP(3) | INSN_OP3(0x15))
251 #define STHA (INSN_OP(3) | INSN_OP3(0x16))
252 #define STWA (INSN_OP(3) | INSN_OP3(0x14))
253 #define STXA (INSN_OP(3) | INSN_OP3(0x1e))
254
255 #define MEMBAR (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13))
256
257 #define NOP (SETHI | INSN_RD(TCG_REG_G0) | 0)
258
259 #ifndef ASI_PRIMARY_LITTLE
260 #define ASI_PRIMARY_LITTLE 0x88
261 #endif
262
263 #define LDUH_LE (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE))
264 #define LDSH_LE (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE))
265 #define LDUW_LE (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE))
266 #define LDSW_LE (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE))
267 #define LDX_LE (LDXA | INSN_ASI(ASI_PRIMARY_LITTLE))
268
269 #define STH_LE (STHA | INSN_ASI(ASI_PRIMARY_LITTLE))
270 #define STW_LE (STWA | INSN_ASI(ASI_PRIMARY_LITTLE))
271 #define STX_LE (STXA | INSN_ASI(ASI_PRIMARY_LITTLE))
272
273 #ifndef use_vis3_instructions
274 bool use_vis3_instructions;
275 #endif
276
277 static inline int check_fit_i64(int64_t val, unsigned int bits)
278 {
279 return val == sextract64(val, 0, bits);
280 }
281
282 static inline int check_fit_i32(int32_t val, unsigned int bits)
283 {
284 return val == sextract32(val, 0, bits);
285 }
286
287 #define check_fit_tl check_fit_i64
288 #if SPARC64
289 # define check_fit_ptr check_fit_i64
290 #else
291 # define check_fit_ptr check_fit_i32
292 #endif
293
294 static bool patch_reloc(tcg_insn_unit *code_ptr, int type,
295 intptr_t value, intptr_t addend)
296 {
297 uint32_t insn = *code_ptr;
298 intptr_t pcrel;
299
300 value += addend;
301 pcrel = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr);
302
303 switch (type) {
304 case R_SPARC_WDISP16:
305 assert(check_fit_ptr(pcrel >> 2, 16));
306 insn &= ~INSN_OFF16(-1);
307 insn |= INSN_OFF16(pcrel);
308 break;
309 case R_SPARC_WDISP19:
310 assert(check_fit_ptr(pcrel >> 2, 19));
311 insn &= ~INSN_OFF19(-1);
312 insn |= INSN_OFF19(pcrel);
313 break;
314 default:
315 g_assert_not_reached();
316 }
317
318 *code_ptr = insn;
319 return true;
320 }
321
322 /* parse target specific constraints */
323 static const char *target_parse_constraint(TCGArgConstraint *ct,
324 const char *ct_str, TCGType type)
325 {
326 switch (*ct_str++) {
327 case 'r':
328 ct->ct |= TCG_CT_REG;
329 ct->u.regs = 0xffffffff;
330 break;
331 case 'R':
332 ct->ct |= TCG_CT_REG;
333 ct->u.regs = ALL_64;
334 break;
335 case 'A': /* qemu_ld/st address constraint */
336 ct->ct |= TCG_CT_REG;
337 ct->u.regs = TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff;
338 reserve_helpers:
339 tcg_regset_reset_reg(ct->u.regs, TCG_REG_O0);
340 tcg_regset_reset_reg(ct->u.regs, TCG_REG_O1);
341 tcg_regset_reset_reg(ct->u.regs, TCG_REG_O2);
342 break;
343 case 's': /* qemu_st data 32-bit constraint */
344 ct->ct |= TCG_CT_REG;
345 ct->u.regs = 0xffffffff;
346 goto reserve_helpers;
347 case 'S': /* qemu_st data 64-bit constraint */
348 ct->ct |= TCG_CT_REG;
349 ct->u.regs = ALL_64;
350 goto reserve_helpers;
351 case 'I':
352 ct->ct |= TCG_CT_CONST_S11;
353 break;
354 case 'J':
355 ct->ct |= TCG_CT_CONST_S13;
356 break;
357 case 'Z':
358 ct->ct |= TCG_CT_CONST_ZERO;
359 break;
360 default:
361 return NULL;
362 }
363 return ct_str;
364 }
365
366 /* test if a constant matches the constraint */
367 static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
368 const TCGArgConstraint *arg_ct)
369 {
370 int ct = arg_ct->ct;
371
372 if (ct & TCG_CT_CONST) {
373 return 1;
374 }
375
376 if (type == TCG_TYPE_I32) {
377 val = (int32_t)val;
378 }
379
380 if ((ct & TCG_CT_CONST_ZERO) && val == 0) {
381 return 1;
382 } else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) {
383 return 1;
384 } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) {
385 return 1;
386 } else {
387 return 0;
388 }
389 }
390
391 static inline void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1,
392 TCGReg rs2, int op)
393 {
394 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2));
395 }
396
397 static inline void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1,
398 int32_t offset, int op)
399 {
400 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset));
401 }
402
403 static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1,
404 int32_t val2, int val2const, int op)
405 {
406 tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1)
407 | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2)));
408 }
409
410 static inline void tcg_out_mov(TCGContext *s, TCGType type,
411 TCGReg ret, TCGReg arg)
412 {
413 if (ret != arg) {
414 tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR);
415 }
416 }
417
418 static inline void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg)
419 {
420 tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10));
421 }
422
423 static inline void tcg_out_movi_imm13(TCGContext *s, TCGReg ret, int32_t arg)
424 {
425 tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR);
426 }
427
428 static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret,
429 tcg_target_long arg, bool in_prologue)
430 {
431 tcg_target_long hi, lo = (int32_t)arg;
432 tcg_target_long test, lsb;
433
434 /* Make sure we test 32-bit constants for imm13 properly. */
435 if (type == TCG_TYPE_I32) {
436 arg = lo;
437 }
438
439 /* A 13-bit constant sign-extended to 64-bits. */
440 if (check_fit_tl(arg, 13)) {
441 tcg_out_movi_imm13(s, ret, arg);
442 return;
443 }
444
445 /* A 13-bit constant relative to the TB. */
446 if (!in_prologue && USE_REG_TB) {
447 test = arg - (uintptr_t)s->code_gen_ptr;
448 if (check_fit_ptr(test, 13)) {
449 tcg_out_arithi(s, ret, TCG_REG_TB, test, ARITH_ADD);
450 return;
451 }
452 }
453
454 /* A 32-bit constant, or 32-bit zero-extended to 64-bits. */
455 if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) {
456 tcg_out_sethi(s, ret, arg);
457 if (arg & 0x3ff) {
458 tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR);
459 }
460 return;
461 }
462
463 /* A 32-bit constant sign-extended to 64-bits. */
464 if (arg == lo) {
465 tcg_out_sethi(s, ret, ~arg);
466 tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR);
467 return;
468 }
469
470 /* A 21-bit constant, shifted. */
471 lsb = ctz64(arg);
472 test = (tcg_target_long)arg >> lsb;
473 if (check_fit_tl(test, 13)) {
474 tcg_out_movi_imm13(s, ret, test);
475 tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX);
476 return;
477 } else if (lsb > 10 && test == extract64(test, 0, 21)) {
478 tcg_out_sethi(s, ret, test << 10);
479 tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX);
480 return;
481 }
482
483 /* A 64-bit constant decomposed into 2 32-bit pieces. */
484 if (check_fit_i32(lo, 13)) {
485 hi = (arg - lo) >> 32;
486 tcg_out_movi(s, TCG_TYPE_I32, ret, hi);
487 tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
488 tcg_out_arithi(s, ret, ret, lo, ARITH_ADD);
489 } else {
490 hi = arg >> 32;
491 tcg_out_movi(s, TCG_TYPE_I32, ret, hi);
492 tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_T2, lo);
493 tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX);
494 tcg_out_arith(s, ret, ret, TCG_REG_T2, ARITH_OR);
495 }
496 }
497
498 static inline void tcg_out_movi(TCGContext *s, TCGType type,
499 TCGReg ret, tcg_target_long arg)
500 {
501 tcg_out_movi_int(s, type, ret, arg, false);
502 }
503
504 static inline void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1,
505 TCGReg a2, int op)
506 {
507 tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2));
508 }
509
510 static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr,
511 intptr_t offset, int op)
512 {
513 if (check_fit_ptr(offset, 13)) {
514 tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) |
515 INSN_IMM13(offset));
516 } else {
517 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset);
518 tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op);
519 }
520 }
521
522 static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret,
523 TCGReg arg1, intptr_t arg2)
524 {
525 tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX));
526 }
527
528 static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg,
529 TCGReg arg1, intptr_t arg2)
530 {
531 tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX));
532 }
533
534 static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val,
535 TCGReg base, intptr_t ofs)
536 {
537 if (val == 0) {
538 tcg_out_st(s, type, TCG_REG_G0, base, ofs);
539 return true;
540 }
541 return false;
542 }
543
544 static void tcg_out_ld_ptr(TCGContext *s, TCGReg ret, uintptr_t arg)
545 {
546 intptr_t diff = arg - (uintptr_t)s->code_gen_ptr;
547 if (USE_REG_TB && check_fit_ptr(diff, 13)) {
548 tcg_out_ld(s, TCG_TYPE_PTR, ret, TCG_REG_TB, diff);
549 return;
550 }
551 tcg_out_movi(s, TCG_TYPE_PTR, ret, arg & ~0x3ff);
552 tcg_out_ld(s, TCG_TYPE_PTR, ret, ret, arg & 0x3ff);
553 }
554
555 static inline void tcg_out_sety(TCGContext *s, TCGReg rs)
556 {
557 tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs));
558 }
559
560 static inline void tcg_out_rdy(TCGContext *s, TCGReg rd)
561 {
562 tcg_out32(s, RDY | INSN_RD(rd));
563 }
564
565 static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1,
566 int32_t val2, int val2const, int uns)
567 {
568 /* Load Y with the sign/zero extension of RS1 to 64-bits. */
569 if (uns) {
570 tcg_out_sety(s, TCG_REG_G0);
571 } else {
572 tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA);
573 tcg_out_sety(s, TCG_REG_T1);
574 }
575
576 tcg_out_arithc(s, rd, rs1, val2, val2const,
577 uns ? ARITH_UDIV : ARITH_SDIV);
578 }
579
580 static inline void tcg_out_nop(TCGContext *s)
581 {
582 tcg_out32(s, NOP);
583 }
584
585 static const uint8_t tcg_cond_to_bcond[] = {
586 [TCG_COND_EQ] = COND_E,
587 [TCG_COND_NE] = COND_NE,
588 [TCG_COND_LT] = COND_L,
589 [TCG_COND_GE] = COND_GE,
590 [TCG_COND_LE] = COND_LE,
591 [TCG_COND_GT] = COND_G,
592 [TCG_COND_LTU] = COND_CS,
593 [TCG_COND_GEU] = COND_CC,
594 [TCG_COND_LEU] = COND_LEU,
595 [TCG_COND_GTU] = COND_GU,
596 };
597
598 static const uint8_t tcg_cond_to_rcond[] = {
599 [TCG_COND_EQ] = RCOND_Z,
600 [TCG_COND_NE] = RCOND_NZ,
601 [TCG_COND_LT] = RCOND_LZ,
602 [TCG_COND_GT] = RCOND_GZ,
603 [TCG_COND_LE] = RCOND_LEZ,
604 [TCG_COND_GE] = RCOND_GEZ
605 };
606
607 static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19)
608 {
609 tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19);
610 }
611
612 static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l)
613 {
614 int off19 = 0;
615
616 if (l->has_value) {
617 off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr));
618 } else {
619 tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0);
620 }
621 tcg_out_bpcc0(s, scond, flags, off19);
622 }
623
624 static void tcg_out_cmp(TCGContext *s, TCGReg c1, int32_t c2, int c2const)
625 {
626 tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const, ARITH_SUBCC);
627 }
628
629 static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1,
630 int32_t arg2, int const_arg2, TCGLabel *l)
631 {
632 tcg_out_cmp(s, arg1, arg2, const_arg2);
633 tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l);
634 tcg_out_nop(s);
635 }
636
637 static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret,
638 int32_t v1, int v1const)
639 {
640 tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret)
641 | INSN_RS1(tcg_cond_to_bcond[cond])
642 | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1)));
643 }
644
645 static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
646 TCGReg c1, int32_t c2, int c2const,
647 int32_t v1, int v1const)
648 {
649 tcg_out_cmp(s, c1, c2, c2const);
650 tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const);
651 }
652
653 static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1,
654 int32_t arg2, int const_arg2, TCGLabel *l)
655 {
656 /* For 64-bit signed comparisons vs zero, we can avoid the compare. */
657 if (arg2 == 0 && !is_unsigned_cond(cond)) {
658 int off16 = 0;
659
660 if (l->has_value) {
661 off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr));
662 } else {
663 tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0);
664 }
665 tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1)
666 | INSN_COND(tcg_cond_to_rcond[cond]) | off16);
667 } else {
668 tcg_out_cmp(s, arg1, arg2, const_arg2);
669 tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l);
670 }
671 tcg_out_nop(s);
672 }
673
674 static void tcg_out_movr(TCGContext *s, TCGCond cond, TCGReg ret, TCGReg c1,
675 int32_t v1, int v1const)
676 {
677 tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1)
678 | (tcg_cond_to_rcond[cond] << 10)
679 | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1)));
680 }
681
682 static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
683 TCGReg c1, int32_t c2, int c2const,
684 int32_t v1, int v1const)
685 {
686 /* For 64-bit signed comparisons vs zero, we can avoid the compare.
687 Note that the immediate range is one bit smaller, so we must check
688 for that as well. */
689 if (c2 == 0 && !is_unsigned_cond(cond)
690 && (!v1const || check_fit_i32(v1, 10))) {
691 tcg_out_movr(s, cond, ret, c1, v1, v1const);
692 } else {
693 tcg_out_cmp(s, c1, c2, c2const);
694 tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const);
695 }
696 }
697
698 static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
699 TCGReg c1, int32_t c2, int c2const)
700 {
701 /* For 32-bit comparisons, we can play games with ADDC/SUBC. */
702 switch (cond) {
703 case TCG_COND_LTU:
704 case TCG_COND_GEU:
705 /* The result of the comparison is in the carry bit. */
706 break;
707
708 case TCG_COND_EQ:
709 case TCG_COND_NE:
710 /* For equality, we can transform to inequality vs zero. */
711 if (c2 != 0) {
712 tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
713 c2 = TCG_REG_T1;
714 } else {
715 c2 = c1;
716 }
717 c1 = TCG_REG_G0, c2const = 0;
718 cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
719 break;
720
721 case TCG_COND_GTU:
722 case TCG_COND_LEU:
723 /* If we don't need to load a constant into a register, we can
724 swap the operands on GTU/LEU. There's no benefit to loading
725 the constant into a temporary register. */
726 if (!c2const || c2 == 0) {
727 TCGReg t = c1;
728 c1 = c2;
729 c2 = t;
730 c2const = 0;
731 cond = tcg_swap_cond(cond);
732 break;
733 }
734 /* FALLTHRU */
735
736 default:
737 tcg_out_cmp(s, c1, c2, c2const);
738 tcg_out_movi_imm13(s, ret, 0);
739 tcg_out_movcc(s, cond, MOVCC_ICC, ret, 1, 1);
740 return;
741 }
742
743 tcg_out_cmp(s, c1, c2, c2const);
744 if (cond == TCG_COND_LTU) {
745 tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
746 } else {
747 tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
748 }
749 }
750
751 static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
752 TCGReg c1, int32_t c2, int c2const)
753 {
754 if (use_vis3_instructions) {
755 switch (cond) {
756 case TCG_COND_NE:
757 if (c2 != 0) {
758 break;
759 }
760 c2 = c1, c2const = 0, c1 = TCG_REG_G0;
761 /* FALLTHRU */
762 case TCG_COND_LTU:
763 tcg_out_cmp(s, c1, c2, c2const);
764 tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
765 return;
766 default:
767 break;
768 }
769 }
770
771 /* For 64-bit signed comparisons vs zero, we can avoid the compare
772 if the input does not overlap the output. */
773 if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) {
774 tcg_out_movi_imm13(s, ret, 0);
775 tcg_out_movr(s, cond, ret, c1, 1, 1);
776 } else {
777 tcg_out_cmp(s, c1, c2, c2const);
778 tcg_out_movi_imm13(s, ret, 0);
779 tcg_out_movcc(s, cond, MOVCC_XCC, ret, 1, 1);
780 }
781 }
782
783 static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh,
784 TCGReg al, TCGReg ah, int32_t bl, int blconst,
785 int32_t bh, int bhconst, int opl, int oph)
786 {
787 TCGReg tmp = TCG_REG_T1;
788
789 /* Note that the low parts are fully consumed before tmp is set. */
790 if (rl != ah && (bhconst || rl != bh)) {
791 tmp = rl;
792 }
793
794 tcg_out_arithc(s, tmp, al, bl, blconst, opl);
795 tcg_out_arithc(s, rh, ah, bh, bhconst, oph);
796 tcg_out_mov(s, TCG_TYPE_I32, rl, tmp);
797 }
798
799 static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
800 TCGReg al, TCGReg ah, int32_t bl, int blconst,
801 int32_t bh, int bhconst, bool is_sub)
802 {
803 TCGReg tmp = TCG_REG_T1;
804
805 /* Note that the low parts are fully consumed before tmp is set. */
806 if (rl != ah && (bhconst || rl != bh)) {
807 tmp = rl;
808 }
809
810 tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
811
812 if (use_vis3_instructions && !is_sub) {
813 /* Note that ADDXC doesn't accept immediates. */
814 if (bhconst && bh != 0) {
815 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh);
816 bh = TCG_REG_T2;
817 }
818 tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC);
819 } else if (bh == TCG_REG_G0) {
820 /* If we have a zero, we can perform the operation in two insns,
821 with the arithmetic first, and a conditional move into place. */
822 if (rh == ah) {
823 tcg_out_arithi(s, TCG_REG_T2, ah, 1,
824 is_sub ? ARITH_SUB : ARITH_ADD);
825 tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0);
826 } else {
827 tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD);
828 tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0);
829 }
830 } else {
831 /* Otherwise adjust BH as if there is carry into T2 ... */
832 if (bhconst) {
833 tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh + (is_sub ? -1 : 1));
834 } else {
835 tcg_out_arithi(s, TCG_REG_T2, bh, 1,
836 is_sub ? ARITH_SUB : ARITH_ADD);
837 }
838 /* ... smoosh T2 back to original BH if carry is clear ... */
839 tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst);
840 /* ... and finally perform the arithmetic with the new operand. */
841 tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD);
842 }
843
844 tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
845 }
846
847 static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest,
848 bool in_prologue)
849 {
850 ptrdiff_t disp = tcg_pcrel_diff(s, dest);
851
852 if (disp == (int32_t)disp) {
853 tcg_out32(s, CALL | (uint32_t)disp >> 2);
854 } else {
855 uintptr_t desti = (uintptr_t)dest;
856 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1,
857 desti & ~0xfff, in_prologue);
858 tcg_out_arithi(s, TCG_REG_O7, TCG_REG_T1, desti & 0xfff, JMPL);
859 }
860 }
861
862 static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest)
863 {
864 tcg_out_call_nodelay(s, dest, false);
865 tcg_out_nop(s);
866 }
867
868 static void tcg_out_mb(TCGContext *s, TCGArg a0)
869 {
870 /* Note that the TCG memory order constants mirror the Sparc MEMBAR. */
871 tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL));
872 }
873
874 #ifdef CONFIG_SOFTMMU
875 static tcg_insn_unit *qemu_ld_trampoline[16];
876 static tcg_insn_unit *qemu_st_trampoline[16];
877
878 static void emit_extend(TCGContext *s, TCGReg r, int op)
879 {
880 /* Emit zero extend of 8, 16 or 32 bit data as
881 * required by the MO_* value op; do nothing for 64 bit.
882 */
883 switch (op & MO_SIZE) {
884 case MO_8:
885 tcg_out_arithi(s, r, r, 0xff, ARITH_AND);
886 break;
887 case MO_16:
888 tcg_out_arithi(s, r, r, 16, SHIFT_SLL);
889 tcg_out_arithi(s, r, r, 16, SHIFT_SRL);
890 break;
891 case MO_32:
892 if (SPARC64) {
893 tcg_out_arith(s, r, r, 0, SHIFT_SRL);
894 }
895 break;
896 case MO_64:
897 break;
898 }
899 }
900
901 static void build_trampolines(TCGContext *s)
902 {
903 static void * const qemu_ld_helpers[16] = {
904 [MO_UB] = helper_ret_ldub_mmu,
905 [MO_SB] = helper_ret_ldsb_mmu,
906 [MO_LEUW] = helper_le_lduw_mmu,
907 [MO_LESW] = helper_le_ldsw_mmu,
908 [MO_LEUL] = helper_le_ldul_mmu,
909 [MO_LEQ] = helper_le_ldq_mmu,
910 [MO_BEUW] = helper_be_lduw_mmu,
911 [MO_BESW] = helper_be_ldsw_mmu,
912 [MO_BEUL] = helper_be_ldul_mmu,
913 [MO_BEQ] = helper_be_ldq_mmu,
914 };
915 static void * const qemu_st_helpers[16] = {
916 [MO_UB] = helper_ret_stb_mmu,
917 [MO_LEUW] = helper_le_stw_mmu,
918 [MO_LEUL] = helper_le_stl_mmu,
919 [MO_LEQ] = helper_le_stq_mmu,
920 [MO_BEUW] = helper_be_stw_mmu,
921 [MO_BEUL] = helper_be_stl_mmu,
922 [MO_BEQ] = helper_be_stq_mmu,
923 };
924
925 int i;
926 TCGReg ra;
927
928 for (i = 0; i < 16; ++i) {
929 if (qemu_ld_helpers[i] == NULL) {
930 continue;
931 }
932
933 /* May as well align the trampoline. */
934 while ((uintptr_t)s->code_ptr & 15) {
935 tcg_out_nop(s);
936 }
937 qemu_ld_trampoline[i] = s->code_ptr;
938
939 if (SPARC64 || TARGET_LONG_BITS == 32) {
940 ra = TCG_REG_O3;
941 } else {
942 /* Install the high part of the address. */
943 tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O2, 32, SHIFT_SRLX);
944 ra = TCG_REG_O4;
945 }
946
947 /* Set the retaddr operand. */
948 tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7);
949 /* Set the env operand. */
950 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0);
951 /* Tail call. */
952 tcg_out_call_nodelay(s, qemu_ld_helpers[i], true);
953 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra);
954 }
955
956 for (i = 0; i < 16; ++i) {
957 if (qemu_st_helpers[i] == NULL) {
958 continue;
959 }
960
961 /* May as well align the trampoline. */
962 while ((uintptr_t)s->code_ptr & 15) {
963 tcg_out_nop(s);
964 }
965 qemu_st_trampoline[i] = s->code_ptr;
966
967 if (SPARC64) {
968 emit_extend(s, TCG_REG_O2, i);
969 ra = TCG_REG_O4;
970 } else {
971 ra = TCG_REG_O1;
972 if (TARGET_LONG_BITS == 64) {
973 /* Install the high part of the address. */
974 tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX);
975 ra += 2;
976 } else {
977 ra += 1;
978 }
979 if ((i & MO_SIZE) == MO_64) {
980 /* Install the high part of the data. */
981 tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX);
982 ra += 2;
983 } else {
984 emit_extend(s, ra, i);
985 ra += 1;
986 }
987 /* Skip the oi argument. */
988 ra += 1;
989 }
990
991 /* Set the retaddr operand. */
992 if (ra >= TCG_REG_O6) {
993 tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_O7, TCG_REG_CALL_STACK,
994 TCG_TARGET_CALL_STACK_OFFSET);
995 ra = TCG_REG_G1;
996 }
997 tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7);
998 /* Set the env operand. */
999 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0);
1000 /* Tail call. */
1001 tcg_out_call_nodelay(s, qemu_st_helpers[i], true);
1002 tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra);
1003 }
1004 }
1005 #endif
1006
1007 /* Generate global QEMU prologue and epilogue code */
1008 static void tcg_target_qemu_prologue(TCGContext *s)
1009 {
1010 int tmp_buf_size, frame_size;
1011
1012 /* The TCG temp buffer is at the top of the frame, immediately
1013 below the frame pointer. */
1014 tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
1015 tcg_set_frame(s, TCG_REG_I6, TCG_TARGET_STACK_BIAS - tmp_buf_size,
1016 tmp_buf_size);
1017
1018 /* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
1019 otherwise the minimal frame usable by callees. */
1020 frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
1021 frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
1022 frame_size += TCG_TARGET_STACK_ALIGN - 1;
1023 frame_size &= -TCG_TARGET_STACK_ALIGN;
1024 tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) |
1025 INSN_IMM13(-frame_size));
1026
1027 #ifndef CONFIG_SOFTMMU
1028 if (guest_base != 0) {
1029 tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true);
1030 tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG);
1031 }
1032 #endif
1033
1034 /* We choose TCG_REG_TB such that no move is required. */
1035 if (USE_REG_TB) {
1036 QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1);
1037 tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB);
1038 }
1039
1040 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL);
1041 /* delay slot */
1042 tcg_out_nop(s);
1043
1044 /* Epilogue for goto_ptr. */
1045 s->code_gen_epilogue = s->code_ptr;
1046 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1047 /* delay slot */
1048 tcg_out_movi_imm13(s, TCG_REG_O0, 0);
1049
1050 #ifdef CONFIG_SOFTMMU
1051 build_trampolines(s);
1052 #endif
1053 }
1054
1055 static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
1056 {
1057 int i;
1058 for (i = 0; i < count; ++i) {
1059 p[i] = NOP;
1060 }
1061 }
1062
1063 #if defined(CONFIG_SOFTMMU)
1064 /* Perform the TLB load and compare.
1065
1066 Inputs:
1067 ADDRLO and ADDRHI contain the possible two parts of the address.
1068
1069 MEM_INDEX and S_BITS are the memory context and log2 size of the load.
1070
1071 WHICH is the offset into the CPUTLBEntry structure of the slot to read.
1072 This should be offsetof addr_read or addr_write.
1073
1074 The result of the TLB comparison is in %[ix]cc. The sanitized address
1075 is in the returned register, maybe %o0. The TLB addend is in %o1. */
1076
1077 /* We expect tlb_mask to be before tlb_table. */
1078 QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
1079 offsetof(CPUArchState, tlb_mask));
1080
1081 /* We expect tlb_mask to be "near" tlb_table. */
1082 QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) -
1083 offsetof(CPUArchState, tlb_mask) >= (1 << 13));
1084
1085 static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addr, int mem_index,
1086 TCGMemOp opc, int which)
1087 {
1088 int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
1089 int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
1090 TCGReg base = TCG_AREG0;
1091 const TCGReg r0 = TCG_REG_O0;
1092 const TCGReg r1 = TCG_REG_O1;
1093 const TCGReg r2 = TCG_REG_O2;
1094 unsigned s_bits = opc & MO_SIZE;
1095 unsigned a_bits = get_alignment_bits(opc);
1096 tcg_target_long compare_mask;
1097
1098 if (!check_fit_i32(table_off, 13)) {
1099 int table_hi;
1100
1101 base = r1;
1102 if (table_off <= 2 * 0xfff) {
1103 table_hi = 0xfff;
1104 tcg_out_arithi(s, base, TCG_AREG0, table_hi, ARITH_ADD);
1105 } else {
1106 table_hi = table_off & ~0x3ff;
1107 tcg_out_sethi(s, base, table_hi);
1108 tcg_out_arith(s, base, TCG_AREG0, base, ARITH_ADD);
1109 }
1110 mask_off -= table_hi;
1111 table_off -= table_hi;
1112 tcg_debug_assert(check_fit_i32(mask_off, 13));
1113 }
1114
1115 /* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
1116 tcg_out_ld(s, TCG_TYPE_PTR, r0, base, mask_off);
1117 tcg_out_ld(s, TCG_TYPE_PTR, r1, base, table_off);
1118
1119 /* Extract the page index, shifted into place for tlb index. */
1120 tcg_out_arithi(s, r2, addr, TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS,
1121 SHIFT_SRL);
1122 tcg_out_arith(s, r2, r2, r0, ARITH_AND);
1123
1124 /* Add the tlb_table pointer, creating the CPUTLBEntry address into R2. */
1125 tcg_out_arith(s, r2, r2, r1, ARITH_ADD);
1126
1127 /* Load the tlb comparator and the addend. */
1128 tcg_out_ld(s, TCG_TYPE_TL, r0, r2, which);
1129 tcg_out_ld(s, TCG_TYPE_PTR, r1, r2, offsetof(CPUTLBEntry, addend));
1130
1131 /* Mask out the page offset, except for the required alignment.
1132 We don't support unaligned accesses. */
1133 if (a_bits < s_bits) {
1134 a_bits = s_bits;
1135 }
1136 compare_mask = (tcg_target_ulong)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
1137 if (check_fit_tl(compare_mask, 13)) {
1138 tcg_out_arithi(s, r2, addr, compare_mask, ARITH_AND);
1139 } else {
1140 tcg_out_movi(s, TCG_TYPE_TL, r2, compare_mask);
1141 tcg_out_arith(s, r2, addr, r2, ARITH_AND);
1142 }
1143 tcg_out_cmp(s, r0, r2, 0);
1144
1145 /* If the guest address must be zero-extended, do so now. */
1146 if (SPARC64 && TARGET_LONG_BITS == 32) {
1147 tcg_out_arithi(s, r0, addr, 0, SHIFT_SRL);
1148 return r0;
1149 }
1150 return addr;
1151 }
1152 #endif /* CONFIG_SOFTMMU */
1153
1154 static const int qemu_ld_opc[16] = {
1155 [MO_UB] = LDUB,
1156 [MO_SB] = LDSB,
1157
1158 [MO_BEUW] = LDUH,
1159 [MO_BESW] = LDSH,
1160 [MO_BEUL] = LDUW,
1161 [MO_BESL] = LDSW,
1162 [MO_BEQ] = LDX,
1163
1164 [MO_LEUW] = LDUH_LE,
1165 [MO_LESW] = LDSH_LE,
1166 [MO_LEUL] = LDUW_LE,
1167 [MO_LESL] = LDSW_LE,
1168 [MO_LEQ] = LDX_LE,
1169 };
1170
1171 static const int qemu_st_opc[16] = {
1172 [MO_UB] = STB,
1173
1174 [MO_BEUW] = STH,
1175 [MO_BEUL] = STW,
1176 [MO_BEQ] = STX,
1177
1178 [MO_LEUW] = STH_LE,
1179 [MO_LEUL] = STW_LE,
1180 [MO_LEQ] = STX_LE,
1181 };
1182
1183 static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr,
1184 TCGMemOpIdx oi, bool is_64)
1185 {
1186 TCGMemOp memop = get_memop(oi);
1187 #ifdef CONFIG_SOFTMMU
1188 unsigned memi = get_mmuidx(oi);
1189 TCGReg addrz, param;
1190 tcg_insn_unit *func;
1191 tcg_insn_unit *label_ptr;
1192
1193 addrz = tcg_out_tlb_load(s, addr, memi, memop,
1194 offsetof(CPUTLBEntry, addr_read));
1195
1196 /* The fast path is exactly one insn. Thus we can perform the
1197 entire TLB Hit in the (annulled) delay slot of the branch
1198 over the TLB Miss case. */
1199
1200 /* beq,a,pt %[xi]cc, label0 */
1201 label_ptr = s->code_ptr;
1202 tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT
1203 | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0);
1204 /* delay slot */
1205 tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1,
1206 qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]);
1207
1208 /* TLB Miss. */
1209
1210 param = TCG_REG_O1;
1211 if (!SPARC64 && TARGET_LONG_BITS == 64) {
1212 /* Skip the high-part; we'll perform the extract in the trampoline. */
1213 param++;
1214 }
1215 tcg_out_mov(s, TCG_TYPE_REG, param++, addrz);
1216
1217 /* We use the helpers to extend SB and SW data, leaving the case
1218 of SL needing explicit extending below. */
1219 if ((memop & MO_SSIZE) == MO_SL) {
1220 func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SIZE)];
1221 } else {
1222 func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SSIZE)];
1223 }
1224 tcg_debug_assert(func != NULL);
1225 tcg_out_call_nodelay(s, func, false);
1226 /* delay slot */
1227 tcg_out_movi(s, TCG_TYPE_I32, param, oi);
1228
1229 /* Recall that all of the helpers return 64-bit results.
1230 Which complicates things for sparcv8plus. */
1231 if (SPARC64) {
1232 /* We let the helper sign-extend SB and SW, but leave SL for here. */
1233 if (is_64 && (memop & MO_SSIZE) == MO_SL) {
1234 tcg_out_arithi(s, data, TCG_REG_O0, 0, SHIFT_SRA);
1235 } else {
1236 tcg_out_mov(s, TCG_TYPE_REG, data, TCG_REG_O0);
1237 }
1238 } else {
1239 if ((memop & MO_SIZE) == MO_64) {
1240 tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, 32, SHIFT_SLLX);
1241 tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O1, 0, SHIFT_SRL);
1242 tcg_out_arith(s, data, TCG_REG_O0, TCG_REG_O1, ARITH_OR);
1243 } else if (is_64) {
1244 /* Re-extend from 32-bit rather than reassembling when we
1245 know the high register must be an extension. */
1246 tcg_out_arithi(s, data, TCG_REG_O1, 0,
1247 memop & MO_SIGN ? SHIFT_SRA : SHIFT_SRL);
1248 } else {
1249 tcg_out_mov(s, TCG_TYPE_I32, data, TCG_REG_O1);
1250 }
1251 }
1252
1253 *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1254 #else
1255 if (SPARC64 && TARGET_LONG_BITS == 32) {
1256 tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL);
1257 addr = TCG_REG_T1;
1258 }
1259 tcg_out_ldst_rr(s, data, addr,
1260 (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0),
1261 qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]);
1262 #endif /* CONFIG_SOFTMMU */
1263 }
1264
1265 static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr,
1266 TCGMemOpIdx oi)
1267 {
1268 TCGMemOp memop = get_memop(oi);
1269 #ifdef CONFIG_SOFTMMU
1270 unsigned memi = get_mmuidx(oi);
1271 TCGReg addrz, param;
1272 tcg_insn_unit *func;
1273 tcg_insn_unit *label_ptr;
1274
1275 addrz = tcg_out_tlb_load(s, addr, memi, memop,
1276 offsetof(CPUTLBEntry, addr_write));
1277
1278 /* The fast path is exactly one insn. Thus we can perform the entire
1279 TLB Hit in the (annulled) delay slot of the branch over TLB Miss. */
1280 /* beq,a,pt %[xi]cc, label0 */
1281 label_ptr = s->code_ptr;
1282 tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT
1283 | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0);
1284 /* delay slot */
1285 tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1,
1286 qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]);
1287
1288 /* TLB Miss. */
1289
1290 param = TCG_REG_O1;
1291 if (!SPARC64 && TARGET_LONG_BITS == 64) {
1292 /* Skip the high-part; we'll perform the extract in the trampoline. */
1293 param++;
1294 }
1295 tcg_out_mov(s, TCG_TYPE_REG, param++, addrz);
1296 if (!SPARC64 && (memop & MO_SIZE) == MO_64) {
1297 /* Skip the high-part; we'll perform the extract in the trampoline. */
1298 param++;
1299 }
1300 tcg_out_mov(s, TCG_TYPE_REG, param++, data);
1301
1302 func = qemu_st_trampoline[memop & (MO_BSWAP | MO_SIZE)];
1303 tcg_debug_assert(func != NULL);
1304 tcg_out_call_nodelay(s, func, false);
1305 /* delay slot */
1306 tcg_out_movi(s, TCG_TYPE_I32, param, oi);
1307
1308 *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr));
1309 #else
1310 if (SPARC64 && TARGET_LONG_BITS == 32) {
1311 tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL);
1312 addr = TCG_REG_T1;
1313 }
1314 tcg_out_ldst_rr(s, data, addr,
1315 (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0),
1316 qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]);
1317 #endif /* CONFIG_SOFTMMU */
1318 }
1319
1320 static void tcg_out_op(TCGContext *s, TCGOpcode opc,
1321 const TCGArg args[TCG_MAX_OP_ARGS],
1322 const int const_args[TCG_MAX_OP_ARGS])
1323 {
1324 TCGArg a0, a1, a2;
1325 int c, c2;
1326
1327 /* Hoist the loads of the most common arguments. */
1328 a0 = args[0];
1329 a1 = args[1];
1330 a2 = args[2];
1331 c2 = const_args[2];
1332
1333 switch (opc) {
1334 case INDEX_op_exit_tb:
1335 if (check_fit_ptr(a0, 13)) {
1336 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1337 tcg_out_movi_imm13(s, TCG_REG_O0, a0);
1338 break;
1339 } else if (USE_REG_TB) {
1340 intptr_t tb_diff = a0 - (uintptr_t)s->code_gen_ptr;
1341 if (check_fit_ptr(tb_diff, 13)) {
1342 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1343 /* Note that TCG_REG_TB has been unwound to O1. */
1344 tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD);
1345 break;
1346 }
1347 }
1348 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff);
1349 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN);
1350 tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR);
1351 break;
1352 case INDEX_op_goto_tb:
1353 if (s->tb_jmp_insn_offset) {
1354 /* direct jump method */
1355 if (USE_REG_TB) {
1356 /* make sure the patch is 8-byte aligned. */
1357 if ((intptr_t)s->code_ptr & 4) {
1358 tcg_out_nop(s);
1359 }
1360 s->tb_jmp_insn_offset[a0] = tcg_current_code_size(s);
1361 tcg_out_sethi(s, TCG_REG_T1, 0);
1362 tcg_out_arithi(s, TCG_REG_T1, TCG_REG_T1, 0, ARITH_OR);
1363 tcg_out_arith(s, TCG_REG_G0, TCG_REG_TB, TCG_REG_T1, JMPL);
1364 tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD);
1365 } else {
1366 s->tb_jmp_insn_offset[a0] = tcg_current_code_size(s);
1367 tcg_out32(s, CALL);
1368 tcg_out_nop(s);
1369 }
1370 } else {
1371 /* indirect jump method */
1372 tcg_out_ld_ptr(s, TCG_REG_TB,
1373 (uintptr_t)(s->tb_jmp_target_addr + a0));
1374 tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL);
1375 tcg_out_nop(s);
1376 }
1377 set_jmp_reset_offset(s, a0);
1378
1379 /* For the unlinked path of goto_tb, we need to reset
1380 TCG_REG_TB to the beginning of this TB. */
1381 if (USE_REG_TB) {
1382 c = -tcg_current_code_size(s);
1383 if (check_fit_i32(c, 13)) {
1384 tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, c, ARITH_ADD);
1385 } else {
1386 tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, c);
1387 tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB,
1388 TCG_REG_T1, ARITH_ADD);
1389 }
1390 }
1391 break;
1392 case INDEX_op_goto_ptr:
1393 tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL);
1394 if (USE_REG_TB) {
1395 tcg_out_arith(s, TCG_REG_TB, a0, TCG_REG_G0, ARITH_OR);
1396 } else {
1397 tcg_out_nop(s);
1398 }
1399 break;
1400 case INDEX_op_br:
1401 tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0));
1402 tcg_out_nop(s);
1403 break;
1404
1405 #define OP_32_64(x) \
1406 glue(glue(case INDEX_op_, x), _i32): \
1407 glue(glue(case INDEX_op_, x), _i64)
1408
1409 OP_32_64(ld8u):
1410 tcg_out_ldst(s, a0, a1, a2, LDUB);
1411 break;
1412 OP_32_64(ld8s):
1413 tcg_out_ldst(s, a0, a1, a2, LDSB);
1414 break;
1415 OP_32_64(ld16u):
1416 tcg_out_ldst(s, a0, a1, a2, LDUH);
1417 break;
1418 OP_32_64(ld16s):
1419 tcg_out_ldst(s, a0, a1, a2, LDSH);
1420 break;
1421 case INDEX_op_ld_i32:
1422 case INDEX_op_ld32u_i64:
1423 tcg_out_ldst(s, a0, a1, a2, LDUW);
1424 break;
1425 OP_32_64(st8):
1426 tcg_out_ldst(s, a0, a1, a2, STB);
1427 break;
1428 OP_32_64(st16):
1429 tcg_out_ldst(s, a0, a1, a2, STH);
1430 break;
1431 case INDEX_op_st_i32:
1432 case INDEX_op_st32_i64:
1433 tcg_out_ldst(s, a0, a1, a2, STW);
1434 break;
1435 OP_32_64(add):
1436 c = ARITH_ADD;
1437 goto gen_arith;
1438 OP_32_64(sub):
1439 c = ARITH_SUB;
1440 goto gen_arith;
1441 OP_32_64(and):
1442 c = ARITH_AND;
1443 goto gen_arith;
1444 OP_32_64(andc):
1445 c = ARITH_ANDN;
1446 goto gen_arith;
1447 OP_32_64(or):
1448 c = ARITH_OR;
1449 goto gen_arith;
1450 OP_32_64(orc):
1451 c = ARITH_ORN;
1452 goto gen_arith;
1453 OP_32_64(xor):
1454 c = ARITH_XOR;
1455 goto gen_arith;
1456 case INDEX_op_shl_i32:
1457 c = SHIFT_SLL;
1458 do_shift32:
1459 /* Limit immediate shift count lest we create an illegal insn. */
1460 tcg_out_arithc(s, a0, a1, a2 & 31, c2, c);
1461 break;
1462 case INDEX_op_shr_i32:
1463 c = SHIFT_SRL;
1464 goto do_shift32;
1465 case INDEX_op_sar_i32:
1466 c = SHIFT_SRA;
1467 goto do_shift32;
1468 case INDEX_op_mul_i32:
1469 c = ARITH_UMUL;
1470 goto gen_arith;
1471
1472 OP_32_64(neg):
1473 c = ARITH_SUB;
1474 goto gen_arith1;
1475 OP_32_64(not):
1476 c = ARITH_ORN;
1477 goto gen_arith1;
1478
1479 case INDEX_op_div_i32:
1480 tcg_out_div32(s, a0, a1, a2, c2, 0);
1481 break;
1482 case INDEX_op_divu_i32:
1483 tcg_out_div32(s, a0, a1, a2, c2, 1);
1484 break;
1485
1486 case INDEX_op_brcond_i32:
1487 tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1488 break;
1489 case INDEX_op_setcond_i32:
1490 tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2);
1491 break;
1492 case INDEX_op_movcond_i32:
1493 tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1494 break;
1495
1496 case INDEX_op_add2_i32:
1497 tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1498 args[4], const_args[4], args[5], const_args[5],
1499 ARITH_ADDCC, ARITH_ADDC);
1500 break;
1501 case INDEX_op_sub2_i32:
1502 tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
1503 args[4], const_args[4], args[5], const_args[5],
1504 ARITH_SUBCC, ARITH_SUBC);
1505 break;
1506 case INDEX_op_mulu2_i32:
1507 c = ARITH_UMUL;
1508 goto do_mul2;
1509 case INDEX_op_muls2_i32:
1510 c = ARITH_SMUL;
1511 do_mul2:
1512 /* The 32-bit multiply insns produce a full 64-bit result. If the
1513 destination register can hold it, we can avoid the slower RDY. */
1514 tcg_out_arithc(s, a0, a2, args[3], const_args[3], c);
1515 if (SPARC64 || a0 <= TCG_REG_O7) {
1516 tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX);
1517 } else {
1518 tcg_out_rdy(s, a1);
1519 }
1520 break;
1521
1522 case INDEX_op_qemu_ld_i32:
1523 tcg_out_qemu_ld(s, a0, a1, a2, false);
1524 break;
1525 case INDEX_op_qemu_ld_i64:
1526 tcg_out_qemu_ld(s, a0, a1, a2, true);
1527 break;
1528 case INDEX_op_qemu_st_i32:
1529 case INDEX_op_qemu_st_i64:
1530 tcg_out_qemu_st(s, a0, a1, a2);
1531 break;
1532
1533 case INDEX_op_ld32s_i64:
1534 tcg_out_ldst(s, a0, a1, a2, LDSW);
1535 break;
1536 case INDEX_op_ld_i64:
1537 tcg_out_ldst(s, a0, a1, a2, LDX);
1538 break;
1539 case INDEX_op_st_i64:
1540 tcg_out_ldst(s, a0, a1, a2, STX);
1541 break;
1542 case INDEX_op_shl_i64:
1543 c = SHIFT_SLLX;
1544 do_shift64:
1545 /* Limit immediate shift count lest we create an illegal insn. */
1546 tcg_out_arithc(s, a0, a1, a2 & 63, c2, c);
1547 break;
1548 case INDEX_op_shr_i64:
1549 c = SHIFT_SRLX;
1550 goto do_shift64;
1551 case INDEX_op_sar_i64:
1552 c = SHIFT_SRAX;
1553 goto do_shift64;
1554 case INDEX_op_mul_i64:
1555 c = ARITH_MULX;
1556 goto gen_arith;
1557 case INDEX_op_div_i64:
1558 c = ARITH_SDIVX;
1559 goto gen_arith;
1560 case INDEX_op_divu_i64:
1561 c = ARITH_UDIVX;
1562 goto gen_arith;
1563 case INDEX_op_ext_i32_i64:
1564 case INDEX_op_ext32s_i64:
1565 tcg_out_arithi(s, a0, a1, 0, SHIFT_SRA);
1566 break;
1567 case INDEX_op_extu_i32_i64:
1568 case INDEX_op_ext32u_i64:
1569 tcg_out_arithi(s, a0, a1, 0, SHIFT_SRL);
1570 break;
1571 case INDEX_op_extrl_i64_i32:
1572 tcg_out_mov(s, TCG_TYPE_I32, a0, a1);
1573 break;
1574 case INDEX_op_extrh_i64_i32:
1575 tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX);
1576 break;
1577
1578 case INDEX_op_brcond_i64:
1579 tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3]));
1580 break;
1581 case INDEX_op_setcond_i64:
1582 tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2);
1583 break;
1584 case INDEX_op_movcond_i64:
1585 tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
1586 break;
1587 case INDEX_op_add2_i64:
1588 tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1589 const_args[4], args[5], const_args[5], false);
1590 break;
1591 case INDEX_op_sub2_i64:
1592 tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
1593 const_args[4], args[5], const_args[5], true);
1594 break;
1595 case INDEX_op_muluh_i64:
1596 tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI);
1597 break;
1598
1599 gen_arith:
1600 tcg_out_arithc(s, a0, a1, a2, c2, c);
1601 break;
1602
1603 gen_arith1:
1604 tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c);
1605 break;
1606
1607 case INDEX_op_mb:
1608 tcg_out_mb(s, a0);
1609 break;
1610
1611 case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */
1612 case INDEX_op_mov_i64:
1613 case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */
1614 case INDEX_op_movi_i64:
1615 case INDEX_op_call: /* Always emitted via tcg_out_call. */
1616 default:
1617 tcg_abort();
1618 }
1619 }
1620
1621 static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
1622 {
1623 static const TCGTargetOpDef r = { .args_ct_str = { "r" } };
1624 static const TCGTargetOpDef r_r = { .args_ct_str = { "r", "r" } };
1625 static const TCGTargetOpDef R_r = { .args_ct_str = { "R", "r" } };
1626 static const TCGTargetOpDef r_R = { .args_ct_str = { "r", "R" } };
1627 static const TCGTargetOpDef R_R = { .args_ct_str = { "R", "R" } };
1628 static const TCGTargetOpDef r_A = { .args_ct_str = { "r", "A" } };
1629 static const TCGTargetOpDef R_A = { .args_ct_str = { "R", "A" } };
1630 static const TCGTargetOpDef rZ_r = { .args_ct_str = { "rZ", "r" } };
1631 static const TCGTargetOpDef RZ_r = { .args_ct_str = { "RZ", "r" } };
1632 static const TCGTargetOpDef sZ_A = { .args_ct_str = { "sZ", "A" } };
1633 static const TCGTargetOpDef SZ_A = { .args_ct_str = { "SZ", "A" } };
1634 static const TCGTargetOpDef rZ_rJ = { .args_ct_str = { "rZ", "rJ" } };
1635 static const TCGTargetOpDef RZ_RJ = { .args_ct_str = { "RZ", "RJ" } };
1636 static const TCGTargetOpDef R_R_R = { .args_ct_str = { "R", "R", "R" } };
1637 static const TCGTargetOpDef r_rZ_rJ
1638 = { .args_ct_str = { "r", "rZ", "rJ" } };
1639 static const TCGTargetOpDef R_RZ_RJ
1640 = { .args_ct_str = { "R", "RZ", "RJ" } };
1641 static const TCGTargetOpDef r_r_rZ_rJ
1642 = { .args_ct_str = { "r", "r", "rZ", "rJ" } };
1643 static const TCGTargetOpDef movc_32
1644 = { .args_ct_str = { "r", "rZ", "rJ", "rI", "0" } };
1645 static const TCGTargetOpDef movc_64
1646 = { .args_ct_str = { "R", "RZ", "RJ", "RI", "0" } };
1647 static const TCGTargetOpDef add2_32
1648 = { .args_ct_str = { "r", "r", "rZ", "rZ", "rJ", "rJ" } };
1649 static const TCGTargetOpDef add2_64
1650 = { .args_ct_str = { "R", "R", "RZ", "RZ", "RJ", "RI" } };
1651
1652 switch (op) {
1653 case INDEX_op_goto_ptr:
1654 return &r;
1655
1656 case INDEX_op_ld8u_i32:
1657 case INDEX_op_ld8s_i32:
1658 case INDEX_op_ld16u_i32:
1659 case INDEX_op_ld16s_i32:
1660 case INDEX_op_ld_i32:
1661 case INDEX_op_neg_i32:
1662 case INDEX_op_not_i32:
1663 return &r_r;
1664
1665 case INDEX_op_st8_i32:
1666 case INDEX_op_st16_i32:
1667 case INDEX_op_st_i32:
1668 return &rZ_r;
1669
1670 case INDEX_op_add_i32:
1671 case INDEX_op_mul_i32:
1672 case INDEX_op_div_i32:
1673 case INDEX_op_divu_i32:
1674 case INDEX_op_sub_i32:
1675 case INDEX_op_and_i32:
1676 case INDEX_op_andc_i32:
1677 case INDEX_op_or_i32:
1678 case INDEX_op_orc_i32:
1679 case INDEX_op_xor_i32:
1680 case INDEX_op_shl_i32:
1681 case INDEX_op_shr_i32:
1682 case INDEX_op_sar_i32:
1683 case INDEX_op_setcond_i32:
1684 return &r_rZ_rJ;
1685
1686 case INDEX_op_brcond_i32:
1687 return &rZ_rJ;
1688 case INDEX_op_movcond_i32:
1689 return &movc_32;
1690 case INDEX_op_add2_i32:
1691 case INDEX_op_sub2_i32:
1692 return &add2_32;
1693 case INDEX_op_mulu2_i32:
1694 case INDEX_op_muls2_i32:
1695 return &r_r_rZ_rJ;
1696
1697 case INDEX_op_ld8u_i64:
1698 case INDEX_op_ld8s_i64:
1699 case INDEX_op_ld16u_i64:
1700 case INDEX_op_ld16s_i64:
1701 case INDEX_op_ld32u_i64:
1702 case INDEX_op_ld32s_i64:
1703 case INDEX_op_ld_i64:
1704 case INDEX_op_ext_i32_i64:
1705 case INDEX_op_extu_i32_i64:
1706 return &R_r;
1707
1708 case INDEX_op_st8_i64:
1709 case INDEX_op_st16_i64:
1710 case INDEX_op_st32_i64:
1711 case INDEX_op_st_i64:
1712 return &RZ_r;
1713
1714 case INDEX_op_add_i64:
1715 case INDEX_op_mul_i64:
1716 case INDEX_op_div_i64:
1717 case INDEX_op_divu_i64:
1718 case INDEX_op_sub_i64:
1719 case INDEX_op_and_i64:
1720 case INDEX_op_andc_i64:
1721 case INDEX_op_or_i64:
1722 case INDEX_op_orc_i64:
1723 case INDEX_op_xor_i64:
1724 case INDEX_op_shl_i64:
1725 case INDEX_op_shr_i64:
1726 case INDEX_op_sar_i64:
1727 case INDEX_op_setcond_i64:
1728 return &R_RZ_RJ;
1729
1730 case INDEX_op_neg_i64:
1731 case INDEX_op_not_i64:
1732 case INDEX_op_ext32s_i64:
1733 case INDEX_op_ext32u_i64:
1734 return &R_R;
1735
1736 case INDEX_op_extrl_i64_i32:
1737 case INDEX_op_extrh_i64_i32:
1738 return &r_R;
1739
1740 case INDEX_op_brcond_i64:
1741 return &RZ_RJ;
1742 case INDEX_op_movcond_i64:
1743 return &movc_64;
1744 case INDEX_op_add2_i64:
1745 case INDEX_op_sub2_i64:
1746 return &add2_64;
1747 case INDEX_op_muluh_i64:
1748 return &R_R_R;
1749
1750 case INDEX_op_qemu_ld_i32:
1751 return &r_A;
1752 case INDEX_op_qemu_ld_i64:
1753 return &R_A;
1754 case INDEX_op_qemu_st_i32:
1755 return &sZ_A;
1756 case INDEX_op_qemu_st_i64:
1757 return &SZ_A;
1758
1759 default:
1760 return NULL;
1761 }
1762 }
1763
1764 static void tcg_target_init(TCGContext *s)
1765 {
1766 /* Only probe for the platform and capabilities if we havn't already
1767 determined maximum values at compile time. */
1768 #ifndef use_vis3_instructions
1769 {
1770 unsigned long hwcap = qemu_getauxval(AT_HWCAP);
1771 use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
1772 }
1773 #endif
1774
1775 tcg_target_available_regs[TCG_TYPE_I32] = 0xffffffff;
1776 tcg_target_available_regs[TCG_TYPE_I64] = ALL_64;
1777
1778 tcg_target_call_clobber_regs = 0;
1779 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G1);
1780 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G2);
1781 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G3);
1782 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G4);
1783 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G5);
1784 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G6);
1785 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_G7);
1786 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O0);
1787 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O1);
1788 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O2);
1789 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O3);
1790 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O4);
1791 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O5);
1792 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O6);
1793 tcg_regset_set_reg(tcg_target_call_clobber_regs, TCG_REG_O7);
1794
1795 s->reserved_regs = 0;
1796 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */
1797 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */
1798 tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */
1799 tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */
1800 tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */
1801 tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */
1802 tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */
1803 tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */
1804 }
1805
1806 #if SPARC64
1807 # define ELF_HOST_MACHINE EM_SPARCV9
1808 #else
1809 # define ELF_HOST_MACHINE EM_SPARC32PLUS
1810 # define ELF_HOST_FLAGS EF_SPARC_32PLUS
1811 #endif
1812
1813 typedef struct {
1814 DebugFrameHeader h;
1815 uint8_t fde_def_cfa[SPARC64 ? 4 : 2];
1816 uint8_t fde_win_save;
1817 uint8_t fde_ret_save[3];
1818 } DebugFrame;
1819
1820 static const DebugFrame debug_frame = {
1821 .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */
1822 .h.cie.id = -1,
1823 .h.cie.version = 1,
1824 .h.cie.code_align = 1,
1825 .h.cie.data_align = -sizeof(void *) & 0x7f,
1826 .h.cie.return_column = 15, /* o7 */
1827
1828 /* Total FDE size does not include the "len" member. */
1829 .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset),
1830
1831 .fde_def_cfa = {
1832 #if SPARC64
1833 12, 30, /* DW_CFA_def_cfa i6, 2047 */
1834 (2047 & 0x7f) | 0x80, (2047 >> 7)
1835 #else
1836 13, 30 /* DW_CFA_def_cfa_register i6 */
1837 #endif
1838 },
1839 .fde_win_save = 0x2d, /* DW_CFA_GNU_window_save */
1840 .fde_ret_save = { 9, 15, 31 }, /* DW_CFA_register o7, i7 */
1841 };
1842
1843 void tcg_register_jit(void *buf, size_t buf_size)
1844 {
1845 tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame));
1846 }
1847
1848 void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr,
1849 uintptr_t addr)
1850 {
1851 intptr_t tb_disp = addr - tc_ptr;
1852 intptr_t br_disp = addr - jmp_addr;
1853 tcg_insn_unit i1, i2;
1854
1855 /* We can reach the entire address space for ILP32.
1856 For LP64, the code_gen_buffer can't be larger than 2GB. */
1857 tcg_debug_assert(tb_disp == (int32_t)tb_disp);
1858 tcg_debug_assert(br_disp == (int32_t)br_disp);
1859
1860 if (!USE_REG_TB) {
1861 atomic_set((uint32_t *)jmp_addr, deposit32(CALL, 0, 30, br_disp >> 2));
1862 flush_icache_range(jmp_addr, jmp_addr + 4);
1863 return;
1864 }
1865
1866 /* This does not exercise the range of the branch, but we do
1867 still need to be able to load the new value of TCG_REG_TB.
1868 But this does still happen quite often. */
1869 if (check_fit_ptr(tb_disp, 13)) {
1870 /* ba,pt %icc, addr */
1871 i1 = (INSN_OP(0) | INSN_OP2(1) | INSN_COND(COND_A)
1872 | BPCC_ICC | BPCC_PT | INSN_OFF19(br_disp));
1873 i2 = (ARITH_ADD | INSN_RD(TCG_REG_TB) | INSN_RS1(TCG_REG_TB)
1874 | INSN_IMM13(tb_disp));
1875 } else if (tb_disp >= 0) {
1876 i1 = SETHI | INSN_RD(TCG_REG_T1) | ((tb_disp & 0xfffffc00) >> 10);
1877 i2 = (ARITH_OR | INSN_RD(TCG_REG_T1) | INSN_RS1(TCG_REG_T1)
1878 | INSN_IMM13(tb_disp & 0x3ff));
1879 } else {
1880 i1 = SETHI | INSN_RD(TCG_REG_T1) | ((~tb_disp & 0xfffffc00) >> 10);
1881 i2 = (ARITH_XOR | INSN_RD(TCG_REG_T1) | INSN_RS1(TCG_REG_T1)
1882 | INSN_IMM13((tb_disp & 0x3ff) | -0x400));
1883 }
1884
1885 atomic_set((uint64_t *)jmp_addr, deposit64(i2, 32, 32, i1));
1886 flush_icache_range(jmp_addr, jmp_addr + 8);
1887 }
Random patches