]>
Commit | Line | Data |
---|---|---|
8289b279 BS |
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 | ||
8d8fdbae | 25 | #ifdef CONFIG_DEBUG_TCG |
8289b279 BS |
26 | static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { |
27 | "%g0", | |
28 | "%g1", | |
29 | "%g2", | |
30 | "%g3", | |
31 | "%g4", | |
32 | "%g5", | |
33 | "%g6", | |
34 | "%g7", | |
35 | "%o0", | |
36 | "%o1", | |
37 | "%o2", | |
38 | "%o3", | |
39 | "%o4", | |
40 | "%o5", | |
41 | "%o6", | |
42 | "%o7", | |
43 | "%l0", | |
44 | "%l1", | |
45 | "%l2", | |
46 | "%l3", | |
47 | "%l4", | |
48 | "%l5", | |
49 | "%l6", | |
50 | "%l7", | |
51 | "%i0", | |
52 | "%i1", | |
53 | "%i2", | |
54 | "%i3", | |
55 | "%i4", | |
56 | "%i5", | |
57 | "%i6", | |
58 | "%i7", | |
59 | }; | |
d4a9eb1f | 60 | #endif |
8289b279 | 61 | |
9f44adc5 RH |
62 | #ifdef __arch64__ |
63 | # define SPARC64 1 | |
64 | #else | |
65 | # define SPARC64 0 | |
66 | #endif | |
67 | ||
34b1a49c RH |
68 | /* Note that sparcv8plus can only hold 64 bit quantities in %g and %o |
69 | registers. These are saved manually by the kernel in full 64-bit | |
70 | slots. The %i and %l registers are saved by the register window | |
71 | mechanism, which only allocates space for 32 bits. Given that this | |
72 | window spill/fill can happen on any signal, we must consider the | |
73 | high bits of the %i and %l registers garbage at all times. */ | |
74 | #if SPARC64 | |
75 | # define ALL_64 0xffffffffu | |
76 | #else | |
77 | # define ALL_64 0xffffu | |
78 | #endif | |
79 | ||
375816f8 RH |
80 | /* Define some temporary registers. T2 is used for constant generation. */ |
81 | #define TCG_REG_T1 TCG_REG_G1 | |
82 | #define TCG_REG_T2 TCG_REG_O7 | |
83 | ||
4cbea598 | 84 | #ifndef CONFIG_SOFTMMU |
375816f8 | 85 | # define TCG_GUEST_BASE_REG TCG_REG_I5 |
c6f7e4fb | 86 | #endif |
e141ab52 | 87 | |
ab20bdc1 RH |
88 | #define TCG_REG_TB TCG_REG_I1 |
89 | #define USE_REG_TB (sizeof(void *) > 4) | |
90 | ||
0954d0d9 | 91 | static const int tcg_target_reg_alloc_order[] = { |
8289b279 BS |
92 | TCG_REG_L0, |
93 | TCG_REG_L1, | |
94 | TCG_REG_L2, | |
95 | TCG_REG_L3, | |
96 | TCG_REG_L4, | |
97 | TCG_REG_L5, | |
98 | TCG_REG_L6, | |
99 | TCG_REG_L7, | |
26adfb75 | 100 | |
8289b279 BS |
101 | TCG_REG_I0, |
102 | TCG_REG_I1, | |
103 | TCG_REG_I2, | |
104 | TCG_REG_I3, | |
105 | TCG_REG_I4, | |
375816f8 | 106 | TCG_REG_I5, |
26adfb75 RH |
107 | |
108 | TCG_REG_G2, | |
109 | TCG_REG_G3, | |
110 | TCG_REG_G4, | |
111 | TCG_REG_G5, | |
112 | ||
113 | TCG_REG_O0, | |
114 | TCG_REG_O1, | |
115 | TCG_REG_O2, | |
116 | TCG_REG_O3, | |
117 | TCG_REG_O4, | |
118 | TCG_REG_O5, | |
8289b279 BS |
119 | }; |
120 | ||
121 | static const int tcg_target_call_iarg_regs[6] = { | |
122 | TCG_REG_O0, | |
123 | TCG_REG_O1, | |
124 | TCG_REG_O2, | |
125 | TCG_REG_O3, | |
126 | TCG_REG_O4, | |
127 | TCG_REG_O5, | |
128 | }; | |
129 | ||
26a74ae3 | 130 | static const int tcg_target_call_oarg_regs[] = { |
8289b279 | 131 | TCG_REG_O0, |
e141ab52 BS |
132 | TCG_REG_O1, |
133 | TCG_REG_O2, | |
134 | TCG_REG_O3, | |
8289b279 BS |
135 | }; |
136 | ||
8289b279 BS |
137 | #define INSN_OP(x) ((x) << 30) |
138 | #define INSN_OP2(x) ((x) << 22) | |
139 | #define INSN_OP3(x) ((x) << 19) | |
140 | #define INSN_OPF(x) ((x) << 5) | |
141 | #define INSN_RD(x) ((x) << 25) | |
142 | #define INSN_RS1(x) ((x) << 14) | |
143 | #define INSN_RS2(x) (x) | |
8384dd67 | 144 | #define INSN_ASI(x) ((x) << 5) |
8289b279 | 145 | |
203342d8 | 146 | #define INSN_IMM10(x) ((1 << 13) | ((x) & 0x3ff)) |
dbfe80e1 | 147 | #define INSN_IMM11(x) ((1 << 13) | ((x) & 0x7ff)) |
8289b279 | 148 | #define INSN_IMM13(x) ((1 << 13) | ((x) & 0x1fff)) |
ab1339b9 | 149 | #define INSN_OFF16(x) ((((x) >> 2) & 0x3fff) | ((((x) >> 16) & 3) << 20)) |
1da92db2 | 150 | #define INSN_OFF19(x) (((x) >> 2) & 0x07ffff) |
a115f3ea | 151 | #define INSN_COND(x) ((x) << 25) |
8289b279 | 152 | |
cf7c2ca5 BS |
153 | #define COND_N 0x0 |
154 | #define COND_E 0x1 | |
155 | #define COND_LE 0x2 | |
156 | #define COND_L 0x3 | |
157 | #define COND_LEU 0x4 | |
158 | #define COND_CS 0x5 | |
159 | #define COND_NEG 0x6 | |
160 | #define COND_VS 0x7 | |
b3db8758 | 161 | #define COND_A 0x8 |
cf7c2ca5 BS |
162 | #define COND_NE 0x9 |
163 | #define COND_G 0xa | |
164 | #define COND_GE 0xb | |
165 | #define COND_GU 0xc | |
166 | #define COND_CC 0xd | |
167 | #define COND_POS 0xe | |
168 | #define COND_VC 0xf | |
a115f3ea | 169 | #define BA (INSN_OP(0) | INSN_COND(COND_A) | INSN_OP2(0x2)) |
8289b279 | 170 | |
ab1339b9 RH |
171 | #define RCOND_Z 1 |
172 | #define RCOND_LEZ 2 | |
173 | #define RCOND_LZ 3 | |
174 | #define RCOND_NZ 5 | |
175 | #define RCOND_GZ 6 | |
176 | #define RCOND_GEZ 7 | |
177 | ||
dbfe80e1 RH |
178 | #define MOVCC_ICC (1 << 18) |
179 | #define MOVCC_XCC (1 << 18 | 1 << 12) | |
180 | ||
a115f3ea RH |
181 | #define BPCC_ICC 0 |
182 | #define BPCC_XCC (2 << 20) | |
183 | #define BPCC_PT (1 << 19) | |
184 | #define BPCC_PN 0 | |
185 | #define BPCC_A (1 << 29) | |
186 | ||
ab1339b9 RH |
187 | #define BPR_PT BPCC_PT |
188 | ||
8289b279 | 189 | #define ARITH_ADD (INSN_OP(2) | INSN_OP3(0x00)) |
7a3766f3 | 190 | #define ARITH_ADDCC (INSN_OP(2) | INSN_OP3(0x10)) |
8289b279 | 191 | #define ARITH_AND (INSN_OP(2) | INSN_OP3(0x01)) |
dc69960d | 192 | #define ARITH_ANDN (INSN_OP(2) | INSN_OP3(0x05)) |
8289b279 | 193 | #define ARITH_OR (INSN_OP(2) | INSN_OP3(0x02)) |
9a7f3228 | 194 | #define ARITH_ORCC (INSN_OP(2) | INSN_OP3(0x12)) |
be6551b1 | 195 | #define ARITH_ORN (INSN_OP(2) | INSN_OP3(0x06)) |
8289b279 | 196 | #define ARITH_XOR (INSN_OP(2) | INSN_OP3(0x03)) |
f5ef6aac BS |
197 | #define ARITH_SUB (INSN_OP(2) | INSN_OP3(0x04)) |
198 | #define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14)) | |
c470b663 RH |
199 | #define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08)) |
200 | #define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c)) | |
8289b279 | 201 | #define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a)) |
f4c16661 | 202 | #define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b)) |
8289b279 BS |
203 | #define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e)) |
204 | #define ARITH_SDIV (INSN_OP(2) | INSN_OP3(0x0f)) | |
205 | #define ARITH_MULX (INSN_OP(2) | INSN_OP3(0x09)) | |
206 | #define ARITH_UDIVX (INSN_OP(2) | INSN_OP3(0x0d)) | |
207 | #define ARITH_SDIVX (INSN_OP(2) | INSN_OP3(0x2d)) | |
dbfe80e1 | 208 | #define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c)) |
203342d8 | 209 | #define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f)) |
8289b279 | 210 | |
90379ca8 | 211 | #define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11)) |
de8301e5 | 212 | #define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16)) |
90379ca8 | 213 | |
8289b279 BS |
214 | #define SHIFT_SLL (INSN_OP(2) | INSN_OP3(0x25)) |
215 | #define SHIFT_SRL (INSN_OP(2) | INSN_OP3(0x26)) | |
216 | #define SHIFT_SRA (INSN_OP(2) | INSN_OP3(0x27)) | |
217 | ||
218 | #define SHIFT_SLLX (INSN_OP(2) | INSN_OP3(0x25) | (1 << 12)) | |
219 | #define SHIFT_SRLX (INSN_OP(2) | INSN_OP3(0x26) | (1 << 12)) | |
220 | #define SHIFT_SRAX (INSN_OP(2) | INSN_OP3(0x27) | (1 << 12)) | |
221 | ||
7a3766f3 | 222 | #define RDY (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(0)) |
583d1215 | 223 | #define WRY (INSN_OP(2) | INSN_OP3(0x30) | INSN_RD(0)) |
8289b279 | 224 | #define JMPL (INSN_OP(2) | INSN_OP3(0x38)) |
8b66eefe | 225 | #define RETURN (INSN_OP(2) | INSN_OP3(0x39)) |
8289b279 BS |
226 | #define SAVE (INSN_OP(2) | INSN_OP3(0x3c)) |
227 | #define RESTORE (INSN_OP(2) | INSN_OP3(0x3d)) | |
228 | #define SETHI (INSN_OP(0) | INSN_OP2(0x4)) | |
229 | #define CALL INSN_OP(1) | |
230 | #define LDUB (INSN_OP(3) | INSN_OP3(0x01)) | |
231 | #define LDSB (INSN_OP(3) | INSN_OP3(0x09)) | |
232 | #define LDUH (INSN_OP(3) | INSN_OP3(0x02)) | |
233 | #define LDSH (INSN_OP(3) | INSN_OP3(0x0a)) | |
234 | #define LDUW (INSN_OP(3) | INSN_OP3(0x00)) | |
235 | #define LDSW (INSN_OP(3) | INSN_OP3(0x08)) | |
236 | #define LDX (INSN_OP(3) | INSN_OP3(0x0b)) | |
237 | #define STB (INSN_OP(3) | INSN_OP3(0x05)) | |
238 | #define STH (INSN_OP(3) | INSN_OP3(0x06)) | |
239 | #define STW (INSN_OP(3) | INSN_OP3(0x04)) | |
240 | #define STX (INSN_OP(3) | INSN_OP3(0x0e)) | |
8384dd67 BS |
241 | #define LDUBA (INSN_OP(3) | INSN_OP3(0x11)) |
242 | #define LDSBA (INSN_OP(3) | INSN_OP3(0x19)) | |
243 | #define LDUHA (INSN_OP(3) | INSN_OP3(0x12)) | |
244 | #define LDSHA (INSN_OP(3) | INSN_OP3(0x1a)) | |
245 | #define LDUWA (INSN_OP(3) | INSN_OP3(0x10)) | |
246 | #define LDSWA (INSN_OP(3) | INSN_OP3(0x18)) | |
247 | #define LDXA (INSN_OP(3) | INSN_OP3(0x1b)) | |
248 | #define STBA (INSN_OP(3) | INSN_OP3(0x15)) | |
249 | #define STHA (INSN_OP(3) | INSN_OP3(0x16)) | |
250 | #define STWA (INSN_OP(3) | INSN_OP3(0x14)) | |
251 | #define STXA (INSN_OP(3) | INSN_OP3(0x1e)) | |
252 | ||
f8f03b37 PK |
253 | #define MEMBAR (INSN_OP(2) | INSN_OP3(0x28) | INSN_RS1(15) | (1 << 13)) |
254 | ||
ab20bdc1 RH |
255 | #define NOP (SETHI | INSN_RD(TCG_REG_G0) | 0) |
256 | ||
8384dd67 BS |
257 | #ifndef ASI_PRIMARY_LITTLE |
258 | #define ASI_PRIMARY_LITTLE 0x88 | |
259 | #endif | |
8289b279 | 260 | |
a0ce341a RH |
261 | #define LDUH_LE (LDUHA | INSN_ASI(ASI_PRIMARY_LITTLE)) |
262 | #define LDSH_LE (LDSHA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
263 | #define LDUW_LE (LDUWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
264 | #define LDSW_LE (LDSWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
265 | #define LDX_LE (LDXA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
266 | ||
267 | #define STH_LE (STHA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
268 | #define STW_LE (STWA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
269 | #define STX_LE (STXA | INSN_ASI(ASI_PRIMARY_LITTLE)) | |
270 | ||
90379ca8 RH |
271 | #ifndef use_vis3_instructions |
272 | bool use_vis3_instructions; | |
273 | #endif | |
274 | ||
425532d7 | 275 | static inline int check_fit_i64(int64_t val, unsigned int bits) |
a115f3ea | 276 | { |
425532d7 | 277 | return val == sextract64(val, 0, bits); |
a115f3ea RH |
278 | } |
279 | ||
425532d7 | 280 | static inline int check_fit_i32(int32_t val, unsigned int bits) |
a115f3ea | 281 | { |
425532d7 | 282 | return val == sextract32(val, 0, bits); |
a115f3ea RH |
283 | } |
284 | ||
425532d7 RH |
285 | #define check_fit_tl check_fit_i64 |
286 | #if SPARC64 | |
287 | # define check_fit_ptr check_fit_i64 | |
288 | #else | |
289 | # define check_fit_ptr check_fit_i32 | |
290 | #endif | |
291 | ||
abce5964 | 292 | static void patch_reloc(tcg_insn_unit *code_ptr, int type, |
2ba7fae2 | 293 | intptr_t value, intptr_t addend) |
a115f3ea RH |
294 | { |
295 | uint32_t insn; | |
abce5964 | 296 | |
eabb7b91 | 297 | tcg_debug_assert(addend == 0); |
abce5964 RH |
298 | value = tcg_ptr_byte_diff((tcg_insn_unit *)value, code_ptr); |
299 | ||
a115f3ea | 300 | switch (type) { |
ab1339b9 | 301 | case R_SPARC_WDISP16: |
425532d7 | 302 | if (!check_fit_ptr(value >> 2, 16)) { |
ab1339b9 RH |
303 | tcg_abort(); |
304 | } | |
abce5964 | 305 | insn = *code_ptr; |
ab1339b9 RH |
306 | insn &= ~INSN_OFF16(-1); |
307 | insn |= INSN_OFF16(value); | |
abce5964 | 308 | *code_ptr = insn; |
ab1339b9 | 309 | break; |
a115f3ea | 310 | case R_SPARC_WDISP19: |
425532d7 | 311 | if (!check_fit_ptr(value >> 2, 19)) { |
a115f3ea RH |
312 | tcg_abort(); |
313 | } | |
abce5964 | 314 | insn = *code_ptr; |
a115f3ea RH |
315 | insn &= ~INSN_OFF19(-1); |
316 | insn |= INSN_OFF19(value); | |
abce5964 | 317 | *code_ptr = insn; |
a115f3ea RH |
318 | break; |
319 | default: | |
320 | tcg_abort(); | |
321 | } | |
322 | } | |
323 | ||
324 | /* parse target specific constraints */ | |
069ea736 RH |
325 | static const char *target_parse_constraint(TCGArgConstraint *ct, |
326 | const char *ct_str, TCGType type) | |
a115f3ea | 327 | { |
069ea736 | 328 | switch (*ct_str++) { |
a115f3ea RH |
329 | case 'r': |
330 | ct->ct |= TCG_CT_REG; | |
331 | tcg_regset_set32(ct->u.regs, 0, 0xffffffff); | |
332 | break; | |
34b1a49c | 333 | case 'R': |
a115f3ea | 334 | ct->ct |= TCG_CT_REG; |
34b1a49c RH |
335 | tcg_regset_set32(ct->u.regs, 0, ALL_64); |
336 | break; | |
337 | case 'A': /* qemu_ld/st address constraint */ | |
338 | ct->ct |= TCG_CT_REG; | |
339 | tcg_regset_set32(ct->u.regs, 0, | |
340 | TARGET_LONG_BITS == 64 ? ALL_64 : 0xffffffff); | |
341 | reserve_helpers: | |
a115f3ea RH |
342 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O0); |
343 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O1); | |
344 | tcg_regset_reset_reg(ct->u.regs, TCG_REG_O2); | |
345 | break; | |
34b1a49c RH |
346 | case 's': /* qemu_st data 32-bit constraint */ |
347 | ct->ct |= TCG_CT_REG; | |
348 | tcg_regset_set32(ct->u.regs, 0, 0xffffffff); | |
349 | goto reserve_helpers; | |
350 | case 'S': /* qemu_st data 64-bit constraint */ | |
351 | ct->ct |= TCG_CT_REG; | |
352 | tcg_regset_set32(ct->u.regs, 0, ALL_64); | |
353 | goto reserve_helpers; | |
a115f3ea RH |
354 | case 'I': |
355 | ct->ct |= TCG_CT_CONST_S11; | |
356 | break; | |
357 | case 'J': | |
358 | ct->ct |= TCG_CT_CONST_S13; | |
359 | break; | |
360 | case 'Z': | |
361 | ct->ct |= TCG_CT_CONST_ZERO; | |
362 | break; | |
363 | default: | |
069ea736 | 364 | return NULL; |
a115f3ea | 365 | } |
069ea736 | 366 | return ct_str; |
a115f3ea RH |
367 | } |
368 | ||
369 | /* test if a constant matches the constraint */ | |
f6c6afc1 | 370 | static inline int tcg_target_const_match(tcg_target_long val, TCGType type, |
a115f3ea RH |
371 | const TCGArgConstraint *arg_ct) |
372 | { | |
373 | int ct = arg_ct->ct; | |
374 | ||
375 | if (ct & TCG_CT_CONST) { | |
376 | return 1; | |
4b304cfa RH |
377 | } |
378 | ||
379 | if (type == TCG_TYPE_I32) { | |
380 | val = (int32_t)val; | |
381 | } | |
382 | ||
383 | if ((ct & TCG_CT_CONST_ZERO) && val == 0) { | |
a115f3ea RH |
384 | return 1; |
385 | } else if ((ct & TCG_CT_CONST_S11) && check_fit_tl(val, 11)) { | |
386 | return 1; | |
387 | } else if ((ct & TCG_CT_CONST_S13) && check_fit_tl(val, 13)) { | |
388 | return 1; | |
389 | } else { | |
390 | return 0; | |
391 | } | |
392 | } | |
393 | ||
35e2da15 RH |
394 | static inline void tcg_out_arith(TCGContext *s, TCGReg rd, TCGReg rs1, |
395 | TCGReg rs2, int op) | |
26cc915c | 396 | { |
35e2da15 | 397 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_RS2(rs2)); |
26cc915c BS |
398 | } |
399 | ||
35e2da15 RH |
400 | static inline void tcg_out_arithi(TCGContext *s, TCGReg rd, TCGReg rs1, |
401 | int32_t offset, int op) | |
26cc915c | 402 | { |
35e2da15 | 403 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | INSN_IMM13(offset)); |
26cc915c BS |
404 | } |
405 | ||
35e2da15 RH |
406 | static void tcg_out_arithc(TCGContext *s, TCGReg rd, TCGReg rs1, |
407 | int32_t val2, int val2const, int op) | |
ba225198 RH |
408 | { |
409 | tcg_out32(s, op | INSN_RD(rd) | INSN_RS1(rs1) | |
410 | | (val2const ? INSN_IMM13(val2) : INSN_RS2(val2))); | |
411 | } | |
412 | ||
2a534aff RH |
413 | static inline void tcg_out_mov(TCGContext *s, TCGType type, |
414 | TCGReg ret, TCGReg arg) | |
8289b279 | 415 | { |
dda73c78 RH |
416 | if (ret != arg) { |
417 | tcg_out_arith(s, ret, arg, TCG_REG_G0, ARITH_OR); | |
418 | } | |
26cc915c BS |
419 | } |
420 | ||
35e2da15 | 421 | static inline void tcg_out_sethi(TCGContext *s, TCGReg ret, uint32_t arg) |
26cc915c BS |
422 | { |
423 | tcg_out32(s, SETHI | INSN_RD(ret) | ((arg & 0xfffffc00) >> 10)); | |
8289b279 BS |
424 | } |
425 | ||
35e2da15 | 426 | static inline void tcg_out_movi_imm13(TCGContext *s, TCGReg ret, int32_t arg) |
b101234a BS |
427 | { |
428 | tcg_out_arithi(s, ret, TCG_REG_G0, arg, ARITH_OR); | |
429 | } | |
430 | ||
ab20bdc1 RH |
431 | static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, |
432 | tcg_target_long arg, bool in_prologue) | |
8289b279 | 433 | { |
425532d7 | 434 | tcg_target_long hi, lo = (int32_t)arg; |
ab20bdc1 | 435 | tcg_target_long test, lsb; |
a9c7d27b | 436 | |
035b2398 RH |
437 | /* Make sure we test 32-bit constants for imm13 properly. */ |
438 | if (type == TCG_TYPE_I32) { | |
439 | arg = lo; | |
440 | } | |
441 | ||
a9c7d27b RH |
442 | /* A 13-bit constant sign-extended to 64-bits. */ |
443 | if (check_fit_tl(arg, 13)) { | |
b101234a | 444 | tcg_out_movi_imm13(s, ret, arg); |
a9c7d27b | 445 | return; |
8289b279 | 446 | } |
8289b279 | 447 | |
a9c7d27b | 448 | /* A 32-bit constant, or 32-bit zero-extended to 64-bits. */ |
34b1a49c | 449 | if (type == TCG_TYPE_I32 || arg == (uint32_t)arg) { |
a9c7d27b RH |
450 | tcg_out_sethi(s, ret, arg); |
451 | if (arg & 0x3ff) { | |
452 | tcg_out_arithi(s, ret, ret, arg & 0x3ff, ARITH_OR); | |
453 | } | |
454 | return; | |
455 | } | |
456 | ||
457 | /* A 32-bit constant sign-extended to 64-bits. */ | |
425532d7 | 458 | if (arg == lo) { |
43172207 RH |
459 | tcg_out_sethi(s, ret, ~arg); |
460 | tcg_out_arithi(s, ret, ret, (arg & 0x3ff) | -0x400, ARITH_XOR); | |
a9c7d27b RH |
461 | return; |
462 | } | |
463 | ||
ab20bdc1 RH |
464 | /* A 21-bit constant, shifted. */ |
465 | lsb = ctz64(arg); | |
466 | test = (tcg_target_long)arg >> lsb; | |
467 | if (check_fit_tl(test, 13)) { | |
468 | tcg_out_movi_imm13(s, ret, test); | |
469 | tcg_out_arithi(s, ret, ret, lsb, SHIFT_SLLX); | |
470 | return; | |
471 | } else if (lsb > 10 && test == extract64(test, 0, 21)) { | |
472 | tcg_out_sethi(s, ret, test << 10); | |
473 | tcg_out_arithi(s, ret, ret, lsb - 10, SHIFT_SLLX); | |
474 | return; | |
475 | } | |
476 | ||
477 | if (USE_REG_TB && !in_prologue) { | |
478 | intptr_t diff = arg - (uintptr_t)s->code_gen_ptr; | |
479 | if (check_fit_ptr(diff, 13)) { | |
480 | tcg_out_arithi(s, ret, TCG_REG_TB, diff, ARITH_ADD); | |
481 | return; | |
482 | } | |
483 | } | |
484 | ||
a9c7d27b | 485 | /* A 64-bit constant decomposed into 2 32-bit pieces. */ |
425532d7 | 486 | if (check_fit_i32(lo, 13)) { |
34b1a49c | 487 | hi = (arg - lo) >> 32; |
a9c7d27b RH |
488 | tcg_out_movi(s, TCG_TYPE_I32, ret, hi); |
489 | tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); | |
490 | tcg_out_arithi(s, ret, ret, lo, ARITH_ADD); | |
43172207 | 491 | } else { |
34b1a49c | 492 | hi = arg >> 32; |
a9c7d27b RH |
493 | tcg_out_movi(s, TCG_TYPE_I32, ret, hi); |
494 | tcg_out_movi(s, TCG_TYPE_I32, TCG_REG_T2, lo); | |
375816f8 | 495 | tcg_out_arithi(s, ret, ret, 32, SHIFT_SLLX); |
375816f8 | 496 | tcg_out_arith(s, ret, ret, TCG_REG_T2, ARITH_OR); |
6f41b777 | 497 | } |
b101234a BS |
498 | } |
499 | ||
ab20bdc1 RH |
500 | static inline void tcg_out_movi(TCGContext *s, TCGType type, |
501 | TCGReg ret, tcg_target_long arg) | |
502 | { | |
503 | tcg_out_movi_int(s, type, ret, arg, false); | |
504 | } | |
505 | ||
35e2da15 RH |
506 | static inline void tcg_out_ldst_rr(TCGContext *s, TCGReg data, TCGReg a1, |
507 | TCGReg a2, int op) | |
8289b279 | 508 | { |
a0ce341a | 509 | tcg_out32(s, op | INSN_RD(data) | INSN_RS1(a1) | INSN_RS2(a2)); |
8289b279 BS |
510 | } |
511 | ||
35e2da15 RH |
512 | static void tcg_out_ldst(TCGContext *s, TCGReg ret, TCGReg addr, |
513 | intptr_t offset, int op) | |
8289b279 | 514 | { |
425532d7 | 515 | if (check_fit_ptr(offset, 13)) { |
8289b279 BS |
516 | tcg_out32(s, op | INSN_RD(ret) | INSN_RS1(addr) | |
517 | INSN_IMM13(offset)); | |
a0ce341a | 518 | } else { |
375816f8 RH |
519 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, offset); |
520 | tcg_out_ldst_rr(s, ret, addr, TCG_REG_T1, op); | |
cf7c2ca5 | 521 | } |
8289b279 BS |
522 | } |
523 | ||
2a534aff | 524 | static inline void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, |
a05b5b9b | 525 | TCGReg arg1, intptr_t arg2) |
8289b279 | 526 | { |
a0ce341a | 527 | tcg_out_ldst(s, ret, arg1, arg2, (type == TCG_TYPE_I32 ? LDUW : LDX)); |
8289b279 BS |
528 | } |
529 | ||
2a534aff | 530 | static inline void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, |
a05b5b9b | 531 | TCGReg arg1, intptr_t arg2) |
8289b279 | 532 | { |
a0ce341a RH |
533 | tcg_out_ldst(s, arg, arg1, arg2, (type == TCG_TYPE_I32 ? STW : STX)); |
534 | } | |
535 | ||
59d7c14e RH |
536 | static inline bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, |
537 | TCGReg base, intptr_t ofs) | |
538 | { | |
539 | if (val == 0) { | |
540 | tcg_out_st(s, type, TCG_REG_G0, base, ofs); | |
541 | return true; | |
542 | } | |
543 | return false; | |
544 | } | |
545 | ||
35e2da15 | 546 | static void tcg_out_ld_ptr(TCGContext *s, TCGReg ret, uintptr_t arg) |
a0ce341a | 547 | { |
ab20bdc1 RH |
548 | intptr_t diff = arg - (uintptr_t)s->code_gen_ptr; |
549 | if (USE_REG_TB && check_fit_ptr(diff, 13)) { | |
550 | tcg_out_ld(s, TCG_TYPE_PTR, ret, TCG_REG_TB, diff); | |
551 | return; | |
552 | } | |
35e2da15 RH |
553 | tcg_out_movi(s, TCG_TYPE_PTR, ret, arg & ~0x3ff); |
554 | tcg_out_ld(s, TCG_TYPE_PTR, ret, ret, arg & 0x3ff); | |
8289b279 BS |
555 | } |
556 | ||
35e2da15 | 557 | static inline void tcg_out_sety(TCGContext *s, TCGReg rs) |
8289b279 | 558 | { |
583d1215 | 559 | tcg_out32(s, WRY | INSN_RS1(TCG_REG_G0) | INSN_RS2(rs)); |
8289b279 BS |
560 | } |
561 | ||
35e2da15 | 562 | static inline void tcg_out_rdy(TCGContext *s, TCGReg rd) |
7a3766f3 RH |
563 | { |
564 | tcg_out32(s, RDY | INSN_RD(rd)); | |
565 | } | |
566 | ||
35e2da15 RH |
567 | static void tcg_out_div32(TCGContext *s, TCGReg rd, TCGReg rs1, |
568 | int32_t val2, int val2const, int uns) | |
583d1215 RH |
569 | { |
570 | /* Load Y with the sign/zero extension of RS1 to 64-bits. */ | |
571 | if (uns) { | |
572 | tcg_out_sety(s, TCG_REG_G0); | |
573 | } else { | |
375816f8 RH |
574 | tcg_out_arithi(s, TCG_REG_T1, rs1, 31, SHIFT_SRA); |
575 | tcg_out_sety(s, TCG_REG_T1); | |
583d1215 RH |
576 | } |
577 | ||
578 | tcg_out_arithc(s, rd, rs1, val2, val2const, | |
579 | uns ? ARITH_UDIV : ARITH_SDIV); | |
580 | } | |
581 | ||
8289b279 BS |
582 | static inline void tcg_out_nop(TCGContext *s) |
583 | { | |
ab20bdc1 | 584 | tcg_out32(s, NOP); |
8289b279 BS |
585 | } |
586 | ||
0aed257f | 587 | static const uint8_t tcg_cond_to_bcond[] = { |
cf7c2ca5 BS |
588 | [TCG_COND_EQ] = COND_E, |
589 | [TCG_COND_NE] = COND_NE, | |
590 | [TCG_COND_LT] = COND_L, | |
591 | [TCG_COND_GE] = COND_GE, | |
592 | [TCG_COND_LE] = COND_LE, | |
593 | [TCG_COND_GT] = COND_G, | |
594 | [TCG_COND_LTU] = COND_CS, | |
595 | [TCG_COND_GEU] = COND_CC, | |
596 | [TCG_COND_LEU] = COND_LEU, | |
597 | [TCG_COND_GTU] = COND_GU, | |
598 | }; | |
599 | ||
ab1339b9 RH |
600 | static const uint8_t tcg_cond_to_rcond[] = { |
601 | [TCG_COND_EQ] = RCOND_Z, | |
602 | [TCG_COND_NE] = RCOND_NZ, | |
603 | [TCG_COND_LT] = RCOND_LZ, | |
604 | [TCG_COND_GT] = RCOND_GZ, | |
605 | [TCG_COND_LE] = RCOND_LEZ, | |
606 | [TCG_COND_GE] = RCOND_GEZ | |
607 | }; | |
608 | ||
a115f3ea RH |
609 | static void tcg_out_bpcc0(TCGContext *s, int scond, int flags, int off19) |
610 | { | |
611 | tcg_out32(s, INSN_OP(0) | INSN_OP2(1) | INSN_COND(scond) | flags | off19); | |
612 | } | |
613 | ||
bec16311 | 614 | static void tcg_out_bpcc(TCGContext *s, int scond, int flags, TCGLabel *l) |
a115f3ea | 615 | { |
a115f3ea RH |
616 | int off19; |
617 | ||
618 | if (l->has_value) { | |
abce5964 | 619 | off19 = INSN_OFF19(tcg_pcrel_diff(s, l->u.value_ptr)); |
a115f3ea RH |
620 | } else { |
621 | /* Make sure to preserve destinations during retranslation. */ | |
abce5964 | 622 | off19 = *s->code_ptr & INSN_OFF19(-1); |
bec16311 | 623 | tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP19, l, 0); |
a115f3ea RH |
624 | } |
625 | tcg_out_bpcc0(s, scond, flags, off19); | |
626 | } | |
627 | ||
35e2da15 | 628 | static void tcg_out_cmp(TCGContext *s, TCGReg c1, int32_t c2, int c2const) |
56f4927e | 629 | { |
ba225198 | 630 | tcg_out_arithc(s, TCG_REG_G0, c1, c2, c2const, ARITH_SUBCC); |
56f4927e RH |
631 | } |
632 | ||
35e2da15 | 633 | static void tcg_out_brcond_i32(TCGContext *s, TCGCond cond, TCGReg arg1, |
bec16311 | 634 | int32_t arg2, int const_arg2, TCGLabel *l) |
cf7c2ca5 | 635 | { |
56f4927e | 636 | tcg_out_cmp(s, arg1, arg2, const_arg2); |
bec16311 | 637 | tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_ICC | BPCC_PT, l); |
cf7c2ca5 BS |
638 | tcg_out_nop(s); |
639 | } | |
640 | ||
35e2da15 RH |
641 | static void tcg_out_movcc(TCGContext *s, TCGCond cond, int cc, TCGReg ret, |
642 | int32_t v1, int v1const) | |
ded37f0d RH |
643 | { |
644 | tcg_out32(s, ARITH_MOVCC | cc | INSN_RD(ret) | |
645 | | INSN_RS1(tcg_cond_to_bcond[cond]) | |
646 | | (v1const ? INSN_IMM11(v1) : INSN_RS2(v1))); | |
647 | } | |
648 | ||
35e2da15 RH |
649 | static void tcg_out_movcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, |
650 | TCGReg c1, int32_t c2, int c2const, | |
651 | int32_t v1, int v1const) | |
ded37f0d RH |
652 | { |
653 | tcg_out_cmp(s, c1, c2, c2const); | |
654 | tcg_out_movcc(s, cond, MOVCC_ICC, ret, v1, v1const); | |
655 | } | |
656 | ||
35e2da15 | 657 | static void tcg_out_brcond_i64(TCGContext *s, TCGCond cond, TCGReg arg1, |
bec16311 | 658 | int32_t arg2, int const_arg2, TCGLabel *l) |
1da92db2 | 659 | { |
ab1339b9 RH |
660 | /* For 64-bit signed comparisons vs zero, we can avoid the compare. */ |
661 | if (arg2 == 0 && !is_unsigned_cond(cond)) { | |
ab1339b9 RH |
662 | int off16; |
663 | ||
664 | if (l->has_value) { | |
abce5964 | 665 | off16 = INSN_OFF16(tcg_pcrel_diff(s, l->u.value_ptr)); |
ab1339b9 RH |
666 | } else { |
667 | /* Make sure to preserve destinations during retranslation. */ | |
abce5964 | 668 | off16 = *s->code_ptr & INSN_OFF16(-1); |
bec16311 | 669 | tcg_out_reloc(s, s->code_ptr, R_SPARC_WDISP16, l, 0); |
ab1339b9 RH |
670 | } |
671 | tcg_out32(s, INSN_OP(0) | INSN_OP2(3) | BPR_PT | INSN_RS1(arg1) | |
672 | | INSN_COND(tcg_cond_to_rcond[cond]) | off16); | |
673 | } else { | |
674 | tcg_out_cmp(s, arg1, arg2, const_arg2); | |
bec16311 | 675 | tcg_out_bpcc(s, tcg_cond_to_bcond[cond], BPCC_XCC | BPCC_PT, l); |
ab1339b9 | 676 | } |
1da92db2 BS |
677 | tcg_out_nop(s); |
678 | } | |
ded37f0d | 679 | |
35e2da15 RH |
680 | static void tcg_out_movr(TCGContext *s, TCGCond cond, TCGReg ret, TCGReg c1, |
681 | int32_t v1, int v1const) | |
203342d8 RH |
682 | { |
683 | tcg_out32(s, ARITH_MOVR | INSN_RD(ret) | INSN_RS1(c1) | |
684 | | (tcg_cond_to_rcond[cond] << 10) | |
685 | | (v1const ? INSN_IMM10(v1) : INSN_RS2(v1))); | |
686 | } | |
687 | ||
35e2da15 RH |
688 | static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, |
689 | TCGReg c1, int32_t c2, int c2const, | |
690 | int32_t v1, int v1const) | |
ded37f0d | 691 | { |
203342d8 RH |
692 | /* For 64-bit signed comparisons vs zero, we can avoid the compare. |
693 | Note that the immediate range is one bit smaller, so we must check | |
694 | for that as well. */ | |
695 | if (c2 == 0 && !is_unsigned_cond(cond) | |
35e2da15 | 696 | && (!v1const || check_fit_i32(v1, 10))) { |
203342d8 RH |
697 | tcg_out_movr(s, cond, ret, c1, v1, v1const); |
698 | } else { | |
699 | tcg_out_cmp(s, c1, c2, c2const); | |
700 | tcg_out_movcc(s, cond, MOVCC_XCC, ret, v1, v1const); | |
701 | } | |
ded37f0d | 702 | } |
1da92db2 | 703 | |
35e2da15 RH |
704 | static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret, |
705 | TCGReg c1, int32_t c2, int c2const) | |
dbfe80e1 | 706 | { |
c470b663 | 707 | /* For 32-bit comparisons, we can play games with ADDC/SUBC. */ |
dbfe80e1 | 708 | switch (cond) { |
7d458a75 RH |
709 | case TCG_COND_LTU: |
710 | case TCG_COND_GEU: | |
711 | /* The result of the comparison is in the carry bit. */ | |
712 | break; | |
713 | ||
dbfe80e1 RH |
714 | case TCG_COND_EQ: |
715 | case TCG_COND_NE: | |
7d458a75 | 716 | /* For equality, we can transform to inequality vs zero. */ |
dbfe80e1 | 717 | if (c2 != 0) { |
321b6c05 RH |
718 | tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR); |
719 | c2 = TCG_REG_T1; | |
720 | } else { | |
721 | c2 = c1; | |
dbfe80e1 | 722 | } |
321b6c05 | 723 | c1 = TCG_REG_G0, c2const = 0; |
7d458a75 | 724 | cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU); |
dbfe80e1 RH |
725 | break; |
726 | ||
727 | case TCG_COND_GTU: | |
dbfe80e1 | 728 | case TCG_COND_LEU: |
7d458a75 RH |
729 | /* If we don't need to load a constant into a register, we can |
730 | swap the operands on GTU/LEU. There's no benefit to loading | |
731 | the constant into a temporary register. */ | |
732 | if (!c2const || c2 == 0) { | |
35e2da15 | 733 | TCGReg t = c1; |
7d458a75 RH |
734 | c1 = c2; |
735 | c2 = t; | |
736 | c2const = 0; | |
737 | cond = tcg_swap_cond(cond); | |
738 | break; | |
739 | } | |
740 | /* FALLTHRU */ | |
dbfe80e1 RH |
741 | |
742 | default: | |
743 | tcg_out_cmp(s, c1, c2, c2const); | |
dbfe80e1 | 744 | tcg_out_movi_imm13(s, ret, 0); |
ded37f0d | 745 | tcg_out_movcc(s, cond, MOVCC_ICC, ret, 1, 1); |
dbfe80e1 RH |
746 | return; |
747 | } | |
748 | ||
749 | tcg_out_cmp(s, c1, c2, c2const); | |
750 | if (cond == TCG_COND_LTU) { | |
c470b663 | 751 | tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC); |
dbfe80e1 | 752 | } else { |
c470b663 | 753 | tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC); |
dbfe80e1 RH |
754 | } |
755 | } | |
756 | ||
35e2da15 RH |
757 | static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret, |
758 | TCGReg c1, int32_t c2, int c2const) | |
dbfe80e1 | 759 | { |
9d6a7a85 RH |
760 | if (use_vis3_instructions) { |
761 | switch (cond) { | |
762 | case TCG_COND_NE: | |
763 | if (c2 != 0) { | |
764 | break; | |
765 | } | |
766 | c2 = c1, c2const = 0, c1 = TCG_REG_G0; | |
767 | /* FALLTHRU */ | |
768 | case TCG_COND_LTU: | |
769 | tcg_out_cmp(s, c1, c2, c2const); | |
770 | tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC); | |
771 | return; | |
772 | default: | |
773 | break; | |
774 | } | |
775 | } | |
776 | ||
203342d8 RH |
777 | /* For 64-bit signed comparisons vs zero, we can avoid the compare |
778 | if the input does not overlap the output. */ | |
779 | if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) { | |
780 | tcg_out_movi_imm13(s, ret, 0); | |
781 | tcg_out_movr(s, cond, ret, c1, 1, 1); | |
782 | } else { | |
783 | tcg_out_cmp(s, c1, c2, c2const); | |
784 | tcg_out_movi_imm13(s, ret, 0); | |
785 | tcg_out_movcc(s, cond, MOVCC_XCC, ret, 1, 1); | |
786 | } | |
dbfe80e1 | 787 | } |
4ec28e25 | 788 | |
609ac1e1 RH |
789 | static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh, |
790 | TCGReg al, TCGReg ah, int32_t bl, int blconst, | |
791 | int32_t bh, int bhconst, int opl, int oph) | |
4ec28e25 | 792 | { |
35e2da15 | 793 | TCGReg tmp = TCG_REG_T1; |
4ec28e25 RH |
794 | |
795 | /* Note that the low parts are fully consumed before tmp is set. */ | |
796 | if (rl != ah && (bhconst || rl != bh)) { | |
797 | tmp = rl; | |
798 | } | |
799 | ||
800 | tcg_out_arithc(s, tmp, al, bl, blconst, opl); | |
801 | tcg_out_arithc(s, rh, ah, bh, bhconst, oph); | |
802 | tcg_out_mov(s, TCG_TYPE_I32, rl, tmp); | |
803 | } | |
dbfe80e1 | 804 | |
609ac1e1 RH |
805 | static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh, |
806 | TCGReg al, TCGReg ah, int32_t bl, int blconst, | |
807 | int32_t bh, int bhconst, bool is_sub) | |
808 | { | |
809 | TCGReg tmp = TCG_REG_T1; | |
810 | ||
811 | /* Note that the low parts are fully consumed before tmp is set. */ | |
812 | if (rl != ah && (bhconst || rl != bh)) { | |
813 | tmp = rl; | |
814 | } | |
815 | ||
816 | tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC); | |
817 | ||
90379ca8 RH |
818 | if (use_vis3_instructions && !is_sub) { |
819 | /* Note that ADDXC doesn't accept immediates. */ | |
820 | if (bhconst && bh != 0) { | |
821 | tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh); | |
822 | bh = TCG_REG_T2; | |
823 | } | |
824 | tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC); | |
825 | } else if (bh == TCG_REG_G0) { | |
609ac1e1 RH |
826 | /* If we have a zero, we can perform the operation in two insns, |
827 | with the arithmetic first, and a conditional move into place. */ | |
828 | if (rh == ah) { | |
829 | tcg_out_arithi(s, TCG_REG_T2, ah, 1, | |
830 | is_sub ? ARITH_SUB : ARITH_ADD); | |
831 | tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0); | |
832 | } else { | |
833 | tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD); | |
834 | tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0); | |
835 | } | |
836 | } else { | |
837 | /* Otherwise adjust BH as if there is carry into T2 ... */ | |
838 | if (bhconst) { | |
839 | tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh + (is_sub ? -1 : 1)); | |
840 | } else { | |
841 | tcg_out_arithi(s, TCG_REG_T2, bh, 1, | |
842 | is_sub ? ARITH_SUB : ARITH_ADD); | |
843 | } | |
844 | /* ... smoosh T2 back to original BH if carry is clear ... */ | |
845 | tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst); | |
846 | /* ... and finally perform the arithmetic with the new operand. */ | |
847 | tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD); | |
848 | } | |
849 | ||
850 | tcg_out_mov(s, TCG_TYPE_I64, rl, tmp); | |
851 | } | |
852 | ||
ab20bdc1 RH |
853 | static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest, |
854 | bool in_prologue) | |
aad2f06a | 855 | { |
abce5964 | 856 | ptrdiff_t disp = tcg_pcrel_diff(s, dest); |
aad2f06a RH |
857 | |
858 | if (disp == (int32_t)disp) { | |
859 | tcg_out32(s, CALL | (uint32_t)disp >> 2); | |
860 | } else { | |
abce5964 | 861 | uintptr_t desti = (uintptr_t)dest; |
ab20bdc1 RH |
862 | tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_REG_T1, |
863 | desti & ~0xfff, in_prologue); | |
abce5964 | 864 | tcg_out_arithi(s, TCG_REG_O7, TCG_REG_T1, desti & 0xfff, JMPL); |
aad2f06a RH |
865 | } |
866 | } | |
867 | ||
4e9cf840 RH |
868 | static void tcg_out_call(TCGContext *s, tcg_insn_unit *dest) |
869 | { | |
ab20bdc1 | 870 | tcg_out_call_nodelay(s, dest, false); |
4e9cf840 RH |
871 | tcg_out_nop(s); |
872 | } | |
873 | ||
f8f03b37 PK |
874 | static void tcg_out_mb(TCGContext *s, TCGArg a0) |
875 | { | |
876 | /* Note that the TCG memory order constants mirror the Sparc MEMBAR. */ | |
877 | tcg_out32(s, MEMBAR | (a0 & TCG_MO_ALL)); | |
878 | } | |
879 | ||
7ea5d725 | 880 | #ifdef CONFIG_SOFTMMU |
abce5964 RH |
881 | static tcg_insn_unit *qemu_ld_trampoline[16]; |
882 | static tcg_insn_unit *qemu_st_trampoline[16]; | |
7ea5d725 | 883 | |
709a340d PM |
884 | static void emit_extend(TCGContext *s, TCGReg r, int op) |
885 | { | |
886 | /* Emit zero extend of 8, 16 or 32 bit data as | |
887 | * required by the MO_* value op; do nothing for 64 bit. | |
888 | */ | |
889 | switch (op & MO_SIZE) { | |
890 | case MO_8: | |
891 | tcg_out_arithi(s, r, r, 0xff, ARITH_AND); | |
892 | break; | |
893 | case MO_16: | |
894 | tcg_out_arithi(s, r, r, 16, SHIFT_SLL); | |
895 | tcg_out_arithi(s, r, r, 16, SHIFT_SRL); | |
896 | break; | |
897 | case MO_32: | |
898 | if (SPARC64) { | |
899 | tcg_out_arith(s, r, r, 0, SHIFT_SRL); | |
900 | } | |
901 | break; | |
902 | case MO_64: | |
903 | break; | |
904 | } | |
905 | } | |
906 | ||
7ea5d725 RH |
907 | static void build_trampolines(TCGContext *s) |
908 | { | |
abce5964 RH |
909 | static void * const qemu_ld_helpers[16] = { |
910 | [MO_UB] = helper_ret_ldub_mmu, | |
911 | [MO_SB] = helper_ret_ldsb_mmu, | |
912 | [MO_LEUW] = helper_le_lduw_mmu, | |
913 | [MO_LESW] = helper_le_ldsw_mmu, | |
914 | [MO_LEUL] = helper_le_ldul_mmu, | |
915 | [MO_LEQ] = helper_le_ldq_mmu, | |
916 | [MO_BEUW] = helper_be_lduw_mmu, | |
917 | [MO_BESW] = helper_be_ldsw_mmu, | |
918 | [MO_BEUL] = helper_be_ldul_mmu, | |
919 | [MO_BEQ] = helper_be_ldq_mmu, | |
7ea5d725 | 920 | }; |
abce5964 RH |
921 | static void * const qemu_st_helpers[16] = { |
922 | [MO_UB] = helper_ret_stb_mmu, | |
923 | [MO_LEUW] = helper_le_stw_mmu, | |
924 | [MO_LEUL] = helper_le_stl_mmu, | |
925 | [MO_LEQ] = helper_le_stq_mmu, | |
926 | [MO_BEUW] = helper_be_stw_mmu, | |
927 | [MO_BEUL] = helper_be_stl_mmu, | |
928 | [MO_BEQ] = helper_be_stq_mmu, | |
7ea5d725 RH |
929 | }; |
930 | ||
931 | int i; | |
932 | TCGReg ra; | |
7ea5d725 RH |
933 | |
934 | for (i = 0; i < 16; ++i) { | |
abce5964 | 935 | if (qemu_ld_helpers[i] == NULL) { |
7ea5d725 RH |
936 | continue; |
937 | } | |
938 | ||
939 | /* May as well align the trampoline. */ | |
abce5964 | 940 | while ((uintptr_t)s->code_ptr & 15) { |
7ea5d725 | 941 | tcg_out_nop(s); |
7ea5d725 | 942 | } |
abce5964 | 943 | qemu_ld_trampoline[i] = s->code_ptr; |
7ea5d725 | 944 | |
34b1a49c RH |
945 | if (SPARC64 || TARGET_LONG_BITS == 32) { |
946 | ra = TCG_REG_O3; | |
947 | } else { | |
948 | /* Install the high part of the address. */ | |
949 | tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O2, 32, SHIFT_SRLX); | |
950 | ra = TCG_REG_O4; | |
951 | } | |
7ea5d725 RH |
952 | |
953 | /* Set the retaddr operand. */ | |
954 | tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7); | |
955 | /* Set the env operand. */ | |
956 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0); | |
957 | /* Tail call. */ | |
ab20bdc1 | 958 | tcg_out_call_nodelay(s, qemu_ld_helpers[i], true); |
7ea5d725 RH |
959 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra); |
960 | } | |
961 | ||
962 | for (i = 0; i < 16; ++i) { | |
abce5964 | 963 | if (qemu_st_helpers[i] == NULL) { |
7ea5d725 RH |
964 | continue; |
965 | } | |
966 | ||
967 | /* May as well align the trampoline. */ | |
abce5964 | 968 | while ((uintptr_t)s->code_ptr & 15) { |
7ea5d725 | 969 | tcg_out_nop(s); |
7ea5d725 | 970 | } |
abce5964 | 971 | qemu_st_trampoline[i] = s->code_ptr; |
7ea5d725 | 972 | |
34b1a49c | 973 | if (SPARC64) { |
709a340d | 974 | emit_extend(s, TCG_REG_O2, i); |
34b1a49c RH |
975 | ra = TCG_REG_O4; |
976 | } else { | |
977 | ra = TCG_REG_O1; | |
978 | if (TARGET_LONG_BITS == 64) { | |
979 | /* Install the high part of the address. */ | |
980 | tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX); | |
981 | ra += 2; | |
982 | } else { | |
983 | ra += 1; | |
984 | } | |
985 | if ((i & MO_SIZE) == MO_64) { | |
986 | /* Install the high part of the data. */ | |
987 | tcg_out_arithi(s, ra, ra + 1, 32, SHIFT_SRLX); | |
988 | ra += 2; | |
989 | } else { | |
709a340d | 990 | emit_extend(s, ra, i); |
34b1a49c RH |
991 | ra += 1; |
992 | } | |
3972ef6f | 993 | /* Skip the oi argument. */ |
34b1a49c RH |
994 | ra += 1; |
995 | } | |
996 | ||
7ea5d725 RH |
997 | /* Set the retaddr operand. */ |
998 | if (ra >= TCG_REG_O6) { | |
999 | tcg_out_st(s, TCG_TYPE_PTR, TCG_REG_O7, TCG_REG_CALL_STACK, | |
1000 | TCG_TARGET_CALL_STACK_OFFSET); | |
1001 | ra = TCG_REG_G1; | |
1002 | } | |
1003 | tcg_out_mov(s, TCG_TYPE_PTR, ra, TCG_REG_O7); | |
1004 | /* Set the env operand. */ | |
1005 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O0, TCG_AREG0); | |
1006 | /* Tail call. */ | |
ab20bdc1 | 1007 | tcg_out_call_nodelay(s, qemu_st_helpers[i], true); |
7ea5d725 RH |
1008 | tcg_out_mov(s, TCG_TYPE_PTR, TCG_REG_O7, ra); |
1009 | } | |
1010 | } | |
1011 | #endif | |
1012 | ||
7d551702 | 1013 | /* Generate global QEMU prologue and epilogue code */ |
e4d58b41 | 1014 | static void tcg_target_qemu_prologue(TCGContext *s) |
b3db8758 | 1015 | { |
4c3204cb RH |
1016 | int tmp_buf_size, frame_size; |
1017 | ||
1018 | /* The TCG temp buffer is at the top of the frame, immediately | |
1019 | below the frame pointer. */ | |
1020 | tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long); | |
1021 | tcg_set_frame(s, TCG_REG_I6, TCG_TARGET_STACK_BIAS - tmp_buf_size, | |
1022 | tmp_buf_size); | |
1023 | ||
1024 | /* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is | |
1025 | otherwise the minimal frame usable by callees. */ | |
1026 | frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS; | |
1027 | frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size; | |
1028 | frame_size += TCG_TARGET_STACK_ALIGN - 1; | |
1029 | frame_size &= -TCG_TARGET_STACK_ALIGN; | |
b3db8758 | 1030 | tcg_out32(s, SAVE | INSN_RD(TCG_REG_O6) | INSN_RS1(TCG_REG_O6) | |
4c3204cb | 1031 | INSN_IMM13(-frame_size)); |
c6f7e4fb | 1032 | |
4cbea598 | 1033 | #ifndef CONFIG_SOFTMMU |
b76f21a7 | 1034 | if (guest_base != 0) { |
ab20bdc1 | 1035 | tcg_out_movi_int(s, TCG_TYPE_PTR, TCG_GUEST_BASE_REG, guest_base, true); |
c6f7e4fb RH |
1036 | tcg_regset_set_reg(s->reserved_regs, TCG_GUEST_BASE_REG); |
1037 | } | |
1038 | #endif | |
1039 | ||
ab20bdc1 RH |
1040 | /* We choose TCG_REG_TB such that no move is required. */ |
1041 | if (USE_REG_TB) { | |
1042 | QEMU_BUILD_BUG_ON(TCG_REG_TB != TCG_REG_I1); | |
1043 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_TB); | |
1044 | } | |
1045 | ||
aad2f06a | 1046 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I1, 0, JMPL); |
0c554161 RH |
1047 | /* delay slot */ |
1048 | tcg_out_nop(s); | |
4c3204cb | 1049 | |
38f81dc5 RH |
1050 | /* Epilogue for goto_ptr. */ |
1051 | s->code_gen_epilogue = s->code_ptr; | |
1052 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); | |
1053 | /* delay slot */ | |
1054 | tcg_out_movi_imm13(s, TCG_REG_O0, 0); | |
7ea5d725 RH |
1055 | |
1056 | #ifdef CONFIG_SOFTMMU | |
1057 | build_trampolines(s); | |
1058 | #endif | |
b3db8758 BS |
1059 | } |
1060 | ||
f5ef6aac | 1061 | #if defined(CONFIG_SOFTMMU) |
a0ce341a | 1062 | /* Perform the TLB load and compare. |
bffe1431 | 1063 | |
a0ce341a | 1064 | Inputs: |
a8b12c10 | 1065 | ADDRLO and ADDRHI contain the possible two parts of the address. |
a0ce341a RH |
1066 | |
1067 | MEM_INDEX and S_BITS are the memory context and log2 size of the load. | |
1068 | ||
1069 | WHICH is the offset into the CPUTLBEntry structure of the slot to read. | |
1070 | This should be offsetof addr_read or addr_write. | |
1071 | ||
1072 | The result of the TLB comparison is in %[ix]cc. The sanitized address | |
1073 | is in the returned register, maybe %o0. The TLB addend is in %o1. */ | |
1074 | ||
34b1a49c | 1075 | static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addr, int mem_index, |
85aa8081 | 1076 | TCGMemOp opc, int which) |
a0ce341a | 1077 | { |
a8b12c10 RH |
1078 | const TCGReg r0 = TCG_REG_O0; |
1079 | const TCGReg r1 = TCG_REG_O1; | |
1080 | const TCGReg r2 = TCG_REG_O2; | |
85aa8081 RH |
1081 | unsigned s_bits = opc & MO_SIZE; |
1082 | unsigned a_bits = get_alignment_bits(opc); | |
a0ce341a RH |
1083 | int tlb_ofs; |
1084 | ||
d801a8f2 | 1085 | /* Shift the page number down. */ |
34b1a49c | 1086 | tcg_out_arithi(s, r1, addr, TARGET_PAGE_BITS, SHIFT_SRL); |
a0ce341a | 1087 | |
85aa8081 RH |
1088 | /* Mask out the page offset, except for the required alignment. |
1089 | We don't support unaligned accesses. */ | |
1090 | if (a_bits < s_bits) { | |
1091 | a_bits = s_bits; | |
1092 | } | |
d801a8f2 | 1093 | tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_T1, |
85aa8081 | 1094 | TARGET_PAGE_MASK | ((1 << a_bits) - 1)); |
d801a8f2 RH |
1095 | |
1096 | /* Mask the tlb index. */ | |
1097 | tcg_out_arithi(s, r1, r1, CPU_TLB_SIZE - 1, ARITH_AND); | |
1098 | ||
1099 | /* Mask page, part 2. */ | |
1100 | tcg_out_arith(s, r0, addr, TCG_REG_T1, ARITH_AND); | |
a0ce341a | 1101 | |
d801a8f2 RH |
1102 | /* Shift the tlb index into place. */ |
1103 | tcg_out_arithi(s, r1, r1, CPU_TLB_ENTRY_BITS, SHIFT_SLL); | |
a0ce341a RH |
1104 | |
1105 | /* Relative to the current ENV. */ | |
1106 | tcg_out_arith(s, r1, TCG_AREG0, r1, ARITH_ADD); | |
1107 | ||
1108 | /* Find a base address that can load both tlb comparator and addend. */ | |
1109 | tlb_ofs = offsetof(CPUArchState, tlb_table[mem_index][0]); | |
425532d7 | 1110 | if (!check_fit_ptr(tlb_ofs + sizeof(CPUTLBEntry), 13)) { |
35e2da15 RH |
1111 | if (tlb_ofs & ~0x3ff) { |
1112 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, tlb_ofs & ~0x3ff); | |
1113 | tcg_out_arith(s, r1, r1, TCG_REG_T1, ARITH_ADD); | |
1114 | } | |
d801a8f2 | 1115 | tlb_ofs &= 0x3ff; |
a0ce341a RH |
1116 | } |
1117 | ||
1118 | /* Load the tlb comparator and the addend. */ | |
1119 | tcg_out_ld(s, TCG_TYPE_TL, r2, r1, tlb_ofs + which); | |
1120 | tcg_out_ld(s, TCG_TYPE_PTR, r1, r1, tlb_ofs+offsetof(CPUTLBEntry, addend)); | |
1121 | ||
1122 | /* subcc arg0, arg2, %g0 */ | |
1123 | tcg_out_cmp(s, r0, r2, 0); | |
1124 | ||
1125 | /* If the guest address must be zero-extended, do so now. */ | |
9f44adc5 | 1126 | if (SPARC64 && TARGET_LONG_BITS == 32) { |
34b1a49c | 1127 | tcg_out_arithi(s, r0, addr, 0, SHIFT_SRL); |
a0ce341a RH |
1128 | return r0; |
1129 | } | |
34b1a49c | 1130 | return addr; |
a0ce341a RH |
1131 | } |
1132 | #endif /* CONFIG_SOFTMMU */ | |
1133 | ||
eef0d9e7 RH |
1134 | static const int qemu_ld_opc[16] = { |
1135 | [MO_UB] = LDUB, | |
1136 | [MO_SB] = LDSB, | |
1137 | ||
1138 | [MO_BEUW] = LDUH, | |
1139 | [MO_BESW] = LDSH, | |
1140 | [MO_BEUL] = LDUW, | |
1141 | [MO_BESL] = LDSW, | |
1142 | [MO_BEQ] = LDX, | |
1143 | ||
1144 | [MO_LEUW] = LDUH_LE, | |
1145 | [MO_LESW] = LDSH_LE, | |
1146 | [MO_LEUL] = LDUW_LE, | |
1147 | [MO_LESL] = LDSW_LE, | |
1148 | [MO_LEQ] = LDX_LE, | |
a0ce341a | 1149 | }; |
9d0efc88 | 1150 | |
eef0d9e7 RH |
1151 | static const int qemu_st_opc[16] = { |
1152 | [MO_UB] = STB, | |
1153 | ||
1154 | [MO_BEUW] = STH, | |
1155 | [MO_BEUL] = STW, | |
1156 | [MO_BEQ] = STX, | |
1157 | ||
1158 | [MO_LEUW] = STH_LE, | |
1159 | [MO_LEUL] = STW_LE, | |
1160 | [MO_LEQ] = STX_LE, | |
a0ce341a | 1161 | }; |
bffe1431 | 1162 | |
34b1a49c | 1163 | static void tcg_out_qemu_ld(TCGContext *s, TCGReg data, TCGReg addr, |
59227d5d | 1164 | TCGMemOpIdx oi, bool is_64) |
f5ef6aac | 1165 | { |
59227d5d | 1166 | TCGMemOp memop = get_memop(oi); |
34b1a49c | 1167 | #ifdef CONFIG_SOFTMMU |
59227d5d | 1168 | unsigned memi = get_mmuidx(oi); |
cab0a7ea | 1169 | TCGReg addrz, param; |
abce5964 RH |
1170 | tcg_insn_unit *func; |
1171 | tcg_insn_unit *label_ptr; | |
f5ef6aac | 1172 | |
85aa8081 | 1173 | addrz = tcg_out_tlb_load(s, addr, memi, memop, |
cab0a7ea | 1174 | offsetof(CPUTLBEntry, addr_read)); |
a0ce341a | 1175 | |
34b1a49c RH |
1176 | /* The fast path is exactly one insn. Thus we can perform the |
1177 | entire TLB Hit in the (annulled) delay slot of the branch | |
1178 | over the TLB Miss case. */ | |
a0ce341a | 1179 | |
34b1a49c | 1180 | /* beq,a,pt %[xi]cc, label0 */ |
abce5964 | 1181 | label_ptr = s->code_ptr; |
34b1a49c RH |
1182 | tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT |
1183 | | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0); | |
1184 | /* delay slot */ | |
2b7ec66f RH |
1185 | tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1, |
1186 | qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]); | |
53c37487 | 1187 | |
a0ce341a | 1188 | /* TLB Miss. */ |
f5ef6aac | 1189 | |
7ea5d725 | 1190 | param = TCG_REG_O1; |
34b1a49c RH |
1191 | if (!SPARC64 && TARGET_LONG_BITS == 64) { |
1192 | /* Skip the high-part; we'll perform the extract in the trampoline. */ | |
1193 | param++; | |
a0ce341a | 1194 | } |
5c32be5b | 1195 | tcg_out_mov(s, TCG_TYPE_REG, param++, addrz); |
f5ef6aac | 1196 | |
7ea5d725 RH |
1197 | /* We use the helpers to extend SB and SW data, leaving the case |
1198 | of SL needing explicit extending below. */ | |
2b7ec66f RH |
1199 | if ((memop & MO_SSIZE) == MO_SL) { |
1200 | func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SIZE)]; | |
7ea5d725 | 1201 | } else { |
2b7ec66f | 1202 | func = qemu_ld_trampoline[memop & (MO_BSWAP | MO_SSIZE)]; |
7ea5d725 | 1203 | } |
eabb7b91 | 1204 | tcg_debug_assert(func != NULL); |
ab20bdc1 | 1205 | tcg_out_call_nodelay(s, func, false); |
a0ce341a | 1206 | /* delay slot */ |
3972ef6f | 1207 | tcg_out_movi(s, TCG_TYPE_I32, param, oi); |
7ea5d725 | 1208 | |
34b1a49c RH |
1209 | /* Recall that all of the helpers return 64-bit results. |
1210 | Which complicates things for sparcv8plus. */ | |
1211 | if (SPARC64) { | |
1212 | /* We let the helper sign-extend SB and SW, but leave SL for here. */ | |
2b7ec66f | 1213 | if (is_64 && (memop & MO_SSIZE) == MO_SL) { |
34b1a49c RH |
1214 | tcg_out_arithi(s, data, TCG_REG_O0, 0, SHIFT_SRA); |
1215 | } else { | |
1216 | tcg_out_mov(s, TCG_TYPE_REG, data, TCG_REG_O0); | |
1217 | } | |
1218 | } else { | |
2b7ec66f | 1219 | if ((memop & MO_SIZE) == MO_64) { |
34b1a49c RH |
1220 | tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, 32, SHIFT_SLLX); |
1221 | tcg_out_arithi(s, TCG_REG_O1, TCG_REG_O1, 0, SHIFT_SRL); | |
1222 | tcg_out_arith(s, data, TCG_REG_O0, TCG_REG_O1, ARITH_OR); | |
1223 | } else if (is_64) { | |
1224 | /* Re-extend from 32-bit rather than reassembling when we | |
1225 | know the high register must be an extension. */ | |
1226 | tcg_out_arithi(s, data, TCG_REG_O1, 0, | |
1227 | memop & MO_SIGN ? SHIFT_SRA : SHIFT_SRL); | |
1228 | } else { | |
1229 | tcg_out_mov(s, TCG_TYPE_I32, data, TCG_REG_O1); | |
a0ce341a | 1230 | } |
f5ef6aac BS |
1231 | } |
1232 | ||
abce5964 | 1233 | *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr)); |
90cbed46 | 1234 | #else |
9f44adc5 | 1235 | if (SPARC64 && TARGET_LONG_BITS == 32) { |
34b1a49c RH |
1236 | tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL); |
1237 | addr = TCG_REG_T1; | |
f5ef6aac | 1238 | } |
34b1a49c | 1239 | tcg_out_ldst_rr(s, data, addr, |
b76f21a7 | 1240 | (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0), |
2b7ec66f | 1241 | qemu_ld_opc[memop & (MO_BSWAP | MO_SSIZE)]); |
a0ce341a | 1242 | #endif /* CONFIG_SOFTMMU */ |
f5ef6aac BS |
1243 | } |
1244 | ||
34b1a49c | 1245 | static void tcg_out_qemu_st(TCGContext *s, TCGReg data, TCGReg addr, |
59227d5d | 1246 | TCGMemOpIdx oi) |
f5ef6aac | 1247 | { |
59227d5d | 1248 | TCGMemOp memop = get_memop(oi); |
34b1a49c | 1249 | #ifdef CONFIG_SOFTMMU |
59227d5d | 1250 | unsigned memi = get_mmuidx(oi); |
34b1a49c | 1251 | TCGReg addrz, param; |
abce5964 RH |
1252 | tcg_insn_unit *func; |
1253 | tcg_insn_unit *label_ptr; | |
f5ef6aac | 1254 | |
85aa8081 | 1255 | addrz = tcg_out_tlb_load(s, addr, memi, memop, |
cab0a7ea | 1256 | offsetof(CPUTLBEntry, addr_write)); |
a0ce341a | 1257 | |
a0ce341a RH |
1258 | /* The fast path is exactly one insn. Thus we can perform the entire |
1259 | TLB Hit in the (annulled) delay slot of the branch over TLB Miss. */ | |
1260 | /* beq,a,pt %[xi]cc, label0 */ | |
abce5964 | 1261 | label_ptr = s->code_ptr; |
a115f3ea RH |
1262 | tcg_out_bpcc0(s, COND_E, BPCC_A | BPCC_PT |
1263 | | (TARGET_LONG_BITS == 64 ? BPCC_XCC : BPCC_ICC), 0); | |
a0ce341a | 1264 | /* delay slot */ |
2b7ec66f RH |
1265 | tcg_out_ldst_rr(s, data, addrz, TCG_REG_O1, |
1266 | qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]); | |
a0ce341a RH |
1267 | |
1268 | /* TLB Miss. */ | |
1269 | ||
7ea5d725 | 1270 | param = TCG_REG_O1; |
34b1a49c RH |
1271 | if (!SPARC64 && TARGET_LONG_BITS == 64) { |
1272 | /* Skip the high-part; we'll perform the extract in the trampoline. */ | |
1273 | param++; | |
a0ce341a | 1274 | } |
5c32be5b | 1275 | tcg_out_mov(s, TCG_TYPE_REG, param++, addrz); |
2b7ec66f | 1276 | if (!SPARC64 && (memop & MO_SIZE) == MO_64) { |
34b1a49c RH |
1277 | /* Skip the high-part; we'll perform the extract in the trampoline. */ |
1278 | param++; | |
a0ce341a | 1279 | } |
34b1a49c | 1280 | tcg_out_mov(s, TCG_TYPE_REG, param++, data); |
53c37487 | 1281 | |
2b7ec66f | 1282 | func = qemu_st_trampoline[memop & (MO_BSWAP | MO_SIZE)]; |
eabb7b91 | 1283 | tcg_debug_assert(func != NULL); |
ab20bdc1 | 1284 | tcg_out_call_nodelay(s, func, false); |
a0ce341a | 1285 | /* delay slot */ |
3972ef6f | 1286 | tcg_out_movi(s, TCG_TYPE_I32, param, oi); |
f5ef6aac | 1287 | |
abce5964 | 1288 | *label_ptr |= INSN_OFF19(tcg_ptr_byte_diff(s->code_ptr, label_ptr)); |
8384dd67 | 1289 | #else |
9f44adc5 | 1290 | if (SPARC64 && TARGET_LONG_BITS == 32) { |
34b1a49c RH |
1291 | tcg_out_arithi(s, TCG_REG_T1, addr, 0, SHIFT_SRL); |
1292 | addr = TCG_REG_T1; | |
a0ce341a | 1293 | } |
34b1a49c | 1294 | tcg_out_ldst_rr(s, data, addr, |
b76f21a7 | 1295 | (guest_base ? TCG_GUEST_BASE_REG : TCG_REG_G0), |
2b7ec66f | 1296 | qemu_st_opc[memop & (MO_BSWAP | MO_SIZE)]); |
a0ce341a | 1297 | #endif /* CONFIG_SOFTMMU */ |
f5ef6aac BS |
1298 | } |
1299 | ||
b357f902 RH |
1300 | static void tcg_out_op(TCGContext *s, TCGOpcode opc, |
1301 | const TCGArg args[TCG_MAX_OP_ARGS], | |
1302 | const int const_args[TCG_MAX_OP_ARGS]) | |
8289b279 | 1303 | { |
b357f902 RH |
1304 | TCGArg a0, a1, a2; |
1305 | int c, c2; | |
1306 | ||
1307 | /* Hoist the loads of the most common arguments. */ | |
1308 | a0 = args[0]; | |
1309 | a1 = args[1]; | |
1310 | a2 = args[2]; | |
1311 | c2 = const_args[2]; | |
8289b279 BS |
1312 | |
1313 | switch (opc) { | |
1314 | case INDEX_op_exit_tb: | |
b357f902 | 1315 | if (check_fit_ptr(a0, 13)) { |
8b66eefe | 1316 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); |
b357f902 | 1317 | tcg_out_movi_imm13(s, TCG_REG_O0, a0); |
ab20bdc1 RH |
1318 | break; |
1319 | } else if (USE_REG_TB) { | |
1320 | intptr_t tb_diff = a0 - (uintptr_t)s->code_gen_ptr; | |
1321 | if (check_fit_ptr(tb_diff, 13)) { | |
1322 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); | |
1323 | /* Note that TCG_REG_TB has been unwound to O1. */ | |
1324 | tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O1, tb_diff, ARITH_ADD); | |
1325 | break; | |
1326 | } | |
8b66eefe | 1327 | } |
ab20bdc1 RH |
1328 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_I0, a0 & ~0x3ff); |
1329 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_I7, 8, RETURN); | |
1330 | tcg_out_arithi(s, TCG_REG_O0, TCG_REG_O0, a0 & 0x3ff, ARITH_OR); | |
8289b279 BS |
1331 | break; |
1332 | case INDEX_op_goto_tb: | |
f309101c | 1333 | if (s->tb_jmp_insn_offset) { |
8289b279 | 1334 | /* direct jump method */ |
ab20bdc1 RH |
1335 | if (USE_REG_TB) { |
1336 | /* make sure the patch is 8-byte aligned. */ | |
1337 | if ((intptr_t)s->code_ptr & 4) { | |
1338 | tcg_out_nop(s); | |
1339 | } | |
1340 | s->tb_jmp_insn_offset[a0] = tcg_current_code_size(s); | |
1341 | tcg_out_sethi(s, TCG_REG_T1, 0); | |
1342 | tcg_out_arithi(s, TCG_REG_T1, TCG_REG_T1, 0, ARITH_OR); | |
1343 | tcg_out_arith(s, TCG_REG_G0, TCG_REG_TB, TCG_REG_T1, JMPL); | |
1344 | tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, TCG_REG_T1, ARITH_ADD); | |
1345 | } else { | |
1346 | s->tb_jmp_insn_offset[a0] = tcg_current_code_size(s); | |
1347 | tcg_out32(s, CALL); | |
1348 | tcg_out_nop(s); | |
1349 | } | |
8289b279 BS |
1350 | } else { |
1351 | /* indirect jump method */ | |
ab20bdc1 | 1352 | tcg_out_ld_ptr(s, TCG_REG_TB, |
f309101c | 1353 | (uintptr_t)(s->tb_jmp_target_addr + a0)); |
ab20bdc1 RH |
1354 | tcg_out_arithi(s, TCG_REG_G0, TCG_REG_TB, 0, JMPL); |
1355 | tcg_out_nop(s); | |
1356 | } | |
1357 | s->tb_jmp_reset_offset[a0] = c = tcg_current_code_size(s); | |
1358 | ||
1359 | /* For the unlinked path of goto_tb, we need to reset | |
1360 | TCG_REG_TB to the beginning of this TB. */ | |
1361 | if (USE_REG_TB) { | |
1362 | c = -c; | |
1363 | if (check_fit_i32(c, 13)) { | |
1364 | tcg_out_arithi(s, TCG_REG_TB, TCG_REG_TB, c, ARITH_ADD); | |
1365 | } else { | |
1366 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, c); | |
1367 | tcg_out_arith(s, TCG_REG_TB, TCG_REG_TB, | |
1368 | TCG_REG_T1, ARITH_ADD); | |
1369 | } | |
8289b279 | 1370 | } |
8289b279 | 1371 | break; |
38f81dc5 RH |
1372 | case INDEX_op_goto_ptr: |
1373 | tcg_out_arithi(s, TCG_REG_G0, a0, 0, JMPL); | |
ab20bdc1 RH |
1374 | if (USE_REG_TB) { |
1375 | tcg_out_arith(s, TCG_REG_TB, a0, TCG_REG_G0, ARITH_OR); | |
1376 | } else { | |
1377 | tcg_out_nop(s); | |
1378 | } | |
38f81dc5 | 1379 | break; |
8289b279 | 1380 | case INDEX_op_br: |
bec16311 | 1381 | tcg_out_bpcc(s, COND_A, BPCC_PT, arg_label(a0)); |
f5ef6aac | 1382 | tcg_out_nop(s); |
8289b279 | 1383 | break; |
8289b279 | 1384 | |
8289b279 | 1385 | #define OP_32_64(x) \ |
ba225198 RH |
1386 | glue(glue(case INDEX_op_, x), _i32): \ |
1387 | glue(glue(case INDEX_op_, x), _i64) | |
34b1a49c | 1388 | |
ba225198 | 1389 | OP_32_64(ld8u): |
b357f902 | 1390 | tcg_out_ldst(s, a0, a1, a2, LDUB); |
8289b279 | 1391 | break; |
ba225198 | 1392 | OP_32_64(ld8s): |
b357f902 | 1393 | tcg_out_ldst(s, a0, a1, a2, LDSB); |
8289b279 | 1394 | break; |
ba225198 | 1395 | OP_32_64(ld16u): |
b357f902 | 1396 | tcg_out_ldst(s, a0, a1, a2, LDUH); |
8289b279 | 1397 | break; |
ba225198 | 1398 | OP_32_64(ld16s): |
b357f902 | 1399 | tcg_out_ldst(s, a0, a1, a2, LDSH); |
8289b279 BS |
1400 | break; |
1401 | case INDEX_op_ld_i32: | |
53cd9273 | 1402 | case INDEX_op_ld32u_i64: |
b357f902 | 1403 | tcg_out_ldst(s, a0, a1, a2, LDUW); |
8289b279 | 1404 | break; |
ba225198 | 1405 | OP_32_64(st8): |
b357f902 | 1406 | tcg_out_ldst(s, a0, a1, a2, STB); |
8289b279 | 1407 | break; |
ba225198 | 1408 | OP_32_64(st16): |
b357f902 | 1409 | tcg_out_ldst(s, a0, a1, a2, STH); |
8289b279 BS |
1410 | break; |
1411 | case INDEX_op_st_i32: | |
53cd9273 | 1412 | case INDEX_op_st32_i64: |
b357f902 | 1413 | tcg_out_ldst(s, a0, a1, a2, STW); |
8289b279 | 1414 | break; |
ba225198 | 1415 | OP_32_64(add): |
53cd9273 | 1416 | c = ARITH_ADD; |
ba225198 RH |
1417 | goto gen_arith; |
1418 | OP_32_64(sub): | |
8289b279 | 1419 | c = ARITH_SUB; |
ba225198 RH |
1420 | goto gen_arith; |
1421 | OP_32_64(and): | |
8289b279 | 1422 | c = ARITH_AND; |
ba225198 | 1423 | goto gen_arith; |
dc69960d RH |
1424 | OP_32_64(andc): |
1425 | c = ARITH_ANDN; | |
1426 | goto gen_arith; | |
ba225198 | 1427 | OP_32_64(or): |
8289b279 | 1428 | c = ARITH_OR; |
ba225198 | 1429 | goto gen_arith; |
18c8f7a3 RH |
1430 | OP_32_64(orc): |
1431 | c = ARITH_ORN; | |
1432 | goto gen_arith; | |
ba225198 | 1433 | OP_32_64(xor): |
8289b279 | 1434 | c = ARITH_XOR; |
ba225198 | 1435 | goto gen_arith; |
8289b279 BS |
1436 | case INDEX_op_shl_i32: |
1437 | c = SHIFT_SLL; | |
1fd95946 RH |
1438 | do_shift32: |
1439 | /* Limit immediate shift count lest we create an illegal insn. */ | |
b357f902 | 1440 | tcg_out_arithc(s, a0, a1, a2 & 31, c2, c); |
1fd95946 | 1441 | break; |
8289b279 BS |
1442 | case INDEX_op_shr_i32: |
1443 | c = SHIFT_SRL; | |
1fd95946 | 1444 | goto do_shift32; |
8289b279 BS |
1445 | case INDEX_op_sar_i32: |
1446 | c = SHIFT_SRA; | |
1fd95946 | 1447 | goto do_shift32; |
8289b279 BS |
1448 | case INDEX_op_mul_i32: |
1449 | c = ARITH_UMUL; | |
ba225198 | 1450 | goto gen_arith; |
583d1215 | 1451 | |
4b5a85c1 RH |
1452 | OP_32_64(neg): |
1453 | c = ARITH_SUB; | |
1454 | goto gen_arith1; | |
be6551b1 RH |
1455 | OP_32_64(not): |
1456 | c = ARITH_ORN; | |
1457 | goto gen_arith1; | |
4b5a85c1 | 1458 | |
583d1215 | 1459 | case INDEX_op_div_i32: |
b357f902 | 1460 | tcg_out_div32(s, a0, a1, a2, c2, 0); |
583d1215 RH |
1461 | break; |
1462 | case INDEX_op_divu_i32: | |
b357f902 | 1463 | tcg_out_div32(s, a0, a1, a2, c2, 1); |
583d1215 RH |
1464 | break; |
1465 | ||
8289b279 | 1466 | case INDEX_op_brcond_i32: |
bec16311 | 1467 | tcg_out_brcond_i32(s, a2, a0, a1, const_args[1], arg_label(args[3])); |
8289b279 | 1468 | break; |
dbfe80e1 | 1469 | case INDEX_op_setcond_i32: |
b357f902 | 1470 | tcg_out_setcond_i32(s, args[3], a0, a1, a2, c2); |
dbfe80e1 | 1471 | break; |
ded37f0d | 1472 | case INDEX_op_movcond_i32: |
b357f902 | 1473 | tcg_out_movcond_i32(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); |
ded37f0d | 1474 | break; |
dbfe80e1 | 1475 | |
7a3766f3 | 1476 | case INDEX_op_add2_i32: |
609ac1e1 RH |
1477 | tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], |
1478 | args[4], const_args[4], args[5], const_args[5], | |
c470b663 | 1479 | ARITH_ADDCC, ARITH_ADDC); |
7a3766f3 RH |
1480 | break; |
1481 | case INDEX_op_sub2_i32: | |
609ac1e1 RH |
1482 | tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3], |
1483 | args[4], const_args[4], args[5], const_args[5], | |
c470b663 | 1484 | ARITH_SUBCC, ARITH_SUBC); |
7a3766f3 RH |
1485 | break; |
1486 | case INDEX_op_mulu2_i32: | |
f4c16661 RH |
1487 | c = ARITH_UMUL; |
1488 | goto do_mul2; | |
1489 | case INDEX_op_muls2_i32: | |
1490 | c = ARITH_SMUL; | |
1491 | do_mul2: | |
1492 | /* The 32-bit multiply insns produce a full 64-bit result. If the | |
1493 | destination register can hold it, we can avoid the slower RDY. */ | |
b357f902 RH |
1494 | tcg_out_arithc(s, a0, a2, args[3], const_args[3], c); |
1495 | if (SPARC64 || a0 <= TCG_REG_O7) { | |
1496 | tcg_out_arithi(s, a1, a0, 32, SHIFT_SRLX); | |
f4c16661 | 1497 | } else { |
b357f902 | 1498 | tcg_out_rdy(s, a1); |
f4c16661 | 1499 | } |
7a3766f3 | 1500 | break; |
8289b279 | 1501 | |
cab0a7ea | 1502 | case INDEX_op_qemu_ld_i32: |
59227d5d | 1503 | tcg_out_qemu_ld(s, a0, a1, a2, false); |
8289b279 | 1504 | break; |
cab0a7ea | 1505 | case INDEX_op_qemu_ld_i64: |
59227d5d | 1506 | tcg_out_qemu_ld(s, a0, a1, a2, true); |
8289b279 | 1507 | break; |
cab0a7ea | 1508 | case INDEX_op_qemu_st_i32: |
cab0a7ea | 1509 | case INDEX_op_qemu_st_i64: |
59227d5d | 1510 | tcg_out_qemu_st(s, a0, a1, a2); |
a0ce341a | 1511 | break; |
8289b279 | 1512 | |
53cd9273 | 1513 | case INDEX_op_ld32s_i64: |
b357f902 | 1514 | tcg_out_ldst(s, a0, a1, a2, LDSW); |
53cd9273 | 1515 | break; |
8289b279 | 1516 | case INDEX_op_ld_i64: |
b357f902 | 1517 | tcg_out_ldst(s, a0, a1, a2, LDX); |
8289b279 BS |
1518 | break; |
1519 | case INDEX_op_st_i64: | |
b357f902 | 1520 | tcg_out_ldst(s, a0, a1, a2, STX); |
8289b279 BS |
1521 | break; |
1522 | case INDEX_op_shl_i64: | |
1523 | c = SHIFT_SLLX; | |
1fd95946 RH |
1524 | do_shift64: |
1525 | /* Limit immediate shift count lest we create an illegal insn. */ | |
b357f902 | 1526 | tcg_out_arithc(s, a0, a1, a2 & 63, c2, c); |
1fd95946 | 1527 | break; |
8289b279 BS |
1528 | case INDEX_op_shr_i64: |
1529 | c = SHIFT_SRLX; | |
1fd95946 | 1530 | goto do_shift64; |
8289b279 BS |
1531 | case INDEX_op_sar_i64: |
1532 | c = SHIFT_SRAX; | |
1fd95946 | 1533 | goto do_shift64; |
8289b279 BS |
1534 | case INDEX_op_mul_i64: |
1535 | c = ARITH_MULX; | |
ba225198 | 1536 | goto gen_arith; |
583d1215 | 1537 | case INDEX_op_div_i64: |
53cd9273 | 1538 | c = ARITH_SDIVX; |
ba225198 | 1539 | goto gen_arith; |
583d1215 | 1540 | case INDEX_op_divu_i64: |
8289b279 | 1541 | c = ARITH_UDIVX; |
ba225198 | 1542 | goto gen_arith; |
4f2331e5 | 1543 | case INDEX_op_ext_i32_i64: |
cc6dfecf | 1544 | case INDEX_op_ext32s_i64: |
b357f902 | 1545 | tcg_out_arithi(s, a0, a1, 0, SHIFT_SRA); |
cc6dfecf | 1546 | break; |
4f2331e5 | 1547 | case INDEX_op_extu_i32_i64: |
cc6dfecf | 1548 | case INDEX_op_ext32u_i64: |
b357f902 | 1549 | tcg_out_arithi(s, a0, a1, 0, SHIFT_SRL); |
cc6dfecf | 1550 | break; |
609ad705 RH |
1551 | case INDEX_op_extrl_i64_i32: |
1552 | tcg_out_mov(s, TCG_TYPE_I32, a0, a1); | |
1553 | break; | |
1554 | case INDEX_op_extrh_i64_i32: | |
1555 | tcg_out_arithi(s, a0, a1, 32, SHIFT_SRLX); | |
a24fba93 | 1556 | break; |
8289b279 BS |
1557 | |
1558 | case INDEX_op_brcond_i64: | |
bec16311 | 1559 | tcg_out_brcond_i64(s, a2, a0, a1, const_args[1], arg_label(args[3])); |
8289b279 | 1560 | break; |
dbfe80e1 | 1561 | case INDEX_op_setcond_i64: |
b357f902 | 1562 | tcg_out_setcond_i64(s, args[3], a0, a1, a2, c2); |
dbfe80e1 | 1563 | break; |
ded37f0d | 1564 | case INDEX_op_movcond_i64: |
b357f902 | 1565 | tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]); |
ded37f0d | 1566 | break; |
609ac1e1 RH |
1567 | case INDEX_op_add2_i64: |
1568 | tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], | |
1569 | const_args[4], args[5], const_args[5], false); | |
1570 | break; | |
1571 | case INDEX_op_sub2_i64: | |
1572 | tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4], | |
1573 | const_args[4], args[5], const_args[5], true); | |
1574 | break; | |
de8301e5 RH |
1575 | case INDEX_op_muluh_i64: |
1576 | tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI); | |
1577 | break; | |
34b1a49c | 1578 | |
ba225198 | 1579 | gen_arith: |
b357f902 | 1580 | tcg_out_arithc(s, a0, a1, a2, c2, c); |
53cd9273 BS |
1581 | break; |
1582 | ||
4b5a85c1 | 1583 | gen_arith1: |
b357f902 | 1584 | tcg_out_arithc(s, a0, TCG_REG_G0, a1, const_args[1], c); |
4b5a85c1 RH |
1585 | break; |
1586 | ||
f8f03b37 PK |
1587 | case INDEX_op_mb: |
1588 | tcg_out_mb(s, a0); | |
1589 | break; | |
1590 | ||
96d0ee7f | 1591 | case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ |
98b90bab | 1592 | case INDEX_op_mov_i64: |
96d0ee7f | 1593 | case INDEX_op_movi_i32: /* Always emitted via tcg_out_movi. */ |
98b90bab | 1594 | case INDEX_op_movi_i64: |
96d0ee7f | 1595 | case INDEX_op_call: /* Always emitted via tcg_out_call. */ |
8289b279 | 1596 | default: |
8289b279 BS |
1597 | tcg_abort(); |
1598 | } | |
1599 | } | |
1600 | ||
1601 | static const TCGTargetOpDef sparc_op_defs[] = { | |
1602 | { INDEX_op_exit_tb, { } }, | |
b3db8758 | 1603 | { INDEX_op_goto_tb, { } }, |
8289b279 | 1604 | { INDEX_op_br, { } }, |
38f81dc5 | 1605 | { INDEX_op_goto_ptr, { "r" } }, |
8289b279 | 1606 | |
8289b279 BS |
1607 | { INDEX_op_ld8u_i32, { "r", "r" } }, |
1608 | { INDEX_op_ld8s_i32, { "r", "r" } }, | |
1609 | { INDEX_op_ld16u_i32, { "r", "r" } }, | |
1610 | { INDEX_op_ld16s_i32, { "r", "r" } }, | |
1611 | { INDEX_op_ld_i32, { "r", "r" } }, | |
89269f6c RH |
1612 | { INDEX_op_st8_i32, { "rZ", "r" } }, |
1613 | { INDEX_op_st16_i32, { "rZ", "r" } }, | |
1614 | { INDEX_op_st_i32, { "rZ", "r" } }, | |
1615 | ||
1616 | { INDEX_op_add_i32, { "r", "rZ", "rJ" } }, | |
1617 | { INDEX_op_mul_i32, { "r", "rZ", "rJ" } }, | |
1618 | { INDEX_op_div_i32, { "r", "rZ", "rJ" } }, | |
1619 | { INDEX_op_divu_i32, { "r", "rZ", "rJ" } }, | |
89269f6c RH |
1620 | { INDEX_op_sub_i32, { "r", "rZ", "rJ" } }, |
1621 | { INDEX_op_and_i32, { "r", "rZ", "rJ" } }, | |
1622 | { INDEX_op_andc_i32, { "r", "rZ", "rJ" } }, | |
1623 | { INDEX_op_or_i32, { "r", "rZ", "rJ" } }, | |
1624 | { INDEX_op_orc_i32, { "r", "rZ", "rJ" } }, | |
1625 | { INDEX_op_xor_i32, { "r", "rZ", "rJ" } }, | |
1626 | ||
1627 | { INDEX_op_shl_i32, { "r", "rZ", "rJ" } }, | |
1628 | { INDEX_op_shr_i32, { "r", "rZ", "rJ" } }, | |
1629 | { INDEX_op_sar_i32, { "r", "rZ", "rJ" } }, | |
8289b279 | 1630 | |
4b5a85c1 | 1631 | { INDEX_op_neg_i32, { "r", "rJ" } }, |
be6551b1 | 1632 | { INDEX_op_not_i32, { "r", "rJ" } }, |
4b5a85c1 | 1633 | |
89269f6c RH |
1634 | { INDEX_op_brcond_i32, { "rZ", "rJ" } }, |
1635 | { INDEX_op_setcond_i32, { "r", "rZ", "rJ" } }, | |
1636 | { INDEX_op_movcond_i32, { "r", "rZ", "rJ", "rI", "0" } }, | |
dbfe80e1 | 1637 | |
89269f6c RH |
1638 | { INDEX_op_add2_i32, { "r", "r", "rZ", "rZ", "rJ", "rJ" } }, |
1639 | { INDEX_op_sub2_i32, { "r", "r", "rZ", "rZ", "rJ", "rJ" } }, | |
1640 | { INDEX_op_mulu2_i32, { "r", "r", "rZ", "rJ" } }, | |
f4c16661 | 1641 | { INDEX_op_muls2_i32, { "r", "r", "rZ", "rJ" } }, |
8289b279 | 1642 | |
34b1a49c RH |
1643 | { INDEX_op_ld8u_i64, { "R", "r" } }, |
1644 | { INDEX_op_ld8s_i64, { "R", "r" } }, | |
1645 | { INDEX_op_ld16u_i64, { "R", "r" } }, | |
1646 | { INDEX_op_ld16s_i64, { "R", "r" } }, | |
1647 | { INDEX_op_ld32u_i64, { "R", "r" } }, | |
1648 | { INDEX_op_ld32s_i64, { "R", "r" } }, | |
1649 | { INDEX_op_ld_i64, { "R", "r" } }, | |
1650 | { INDEX_op_st8_i64, { "RZ", "r" } }, | |
1651 | { INDEX_op_st16_i64, { "RZ", "r" } }, | |
1652 | { INDEX_op_st32_i64, { "RZ", "r" } }, | |
1653 | { INDEX_op_st_i64, { "RZ", "r" } }, | |
1654 | ||
1655 | { INDEX_op_add_i64, { "R", "RZ", "RJ" } }, | |
1656 | { INDEX_op_mul_i64, { "R", "RZ", "RJ" } }, | |
1657 | { INDEX_op_div_i64, { "R", "RZ", "RJ" } }, | |
1658 | { INDEX_op_divu_i64, { "R", "RZ", "RJ" } }, | |
1659 | { INDEX_op_sub_i64, { "R", "RZ", "RJ" } }, | |
1660 | { INDEX_op_and_i64, { "R", "RZ", "RJ" } }, | |
1661 | { INDEX_op_andc_i64, { "R", "RZ", "RJ" } }, | |
1662 | { INDEX_op_or_i64, { "R", "RZ", "RJ" } }, | |
1663 | { INDEX_op_orc_i64, { "R", "RZ", "RJ" } }, | |
1664 | { INDEX_op_xor_i64, { "R", "RZ", "RJ" } }, | |
1665 | ||
1666 | { INDEX_op_shl_i64, { "R", "RZ", "RJ" } }, | |
1667 | { INDEX_op_shr_i64, { "R", "RZ", "RJ" } }, | |
1668 | { INDEX_op_sar_i64, { "R", "RZ", "RJ" } }, | |
1669 | ||
1670 | { INDEX_op_neg_i64, { "R", "RJ" } }, | |
1671 | { INDEX_op_not_i64, { "R", "RJ" } }, | |
1672 | ||
4f2331e5 AJ |
1673 | { INDEX_op_ext32s_i64, { "R", "R" } }, |
1674 | { INDEX_op_ext32u_i64, { "R", "R" } }, | |
1675 | { INDEX_op_ext_i32_i64, { "R", "r" } }, | |
1676 | { INDEX_op_extu_i32_i64, { "R", "r" } }, | |
609ad705 RH |
1677 | { INDEX_op_extrl_i64_i32, { "r", "R" } }, |
1678 | { INDEX_op_extrh_i64_i32, { "r", "R" } }, | |
34b1a49c RH |
1679 | |
1680 | { INDEX_op_brcond_i64, { "RZ", "RJ" } }, | |
1681 | { INDEX_op_setcond_i64, { "R", "RZ", "RJ" } }, | |
1682 | { INDEX_op_movcond_i64, { "R", "RZ", "RJ", "RI", "0" } }, | |
1683 | ||
609ac1e1 RH |
1684 | { INDEX_op_add2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } }, |
1685 | { INDEX_op_sub2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } }, | |
de8301e5 | 1686 | { INDEX_op_muluh_i64, { "R", "RZ", "RZ" } }, |
609ac1e1 | 1687 | |
34b1a49c RH |
1688 | { INDEX_op_qemu_ld_i32, { "r", "A" } }, |
1689 | { INDEX_op_qemu_ld_i64, { "R", "A" } }, | |
ebd0c614 RH |
1690 | { INDEX_op_qemu_st_i32, { "sZ", "A" } }, |
1691 | { INDEX_op_qemu_st_i64, { "SZ", "A" } }, | |
a0ce341a | 1692 | |
f8f03b37 | 1693 | { INDEX_op_mb, { } }, |
8289b279 BS |
1694 | { -1 }, |
1695 | }; | |
1696 | ||
f69d277e RH |
1697 | static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op) |
1698 | { | |
1699 | int i, n = ARRAY_SIZE(sparc_op_defs); | |
1700 | ||
1701 | for (i = 0; i < n; ++i) { | |
1702 | if (sparc_op_defs[i].op == op) { | |
1703 | return &sparc_op_defs[i]; | |
1704 | } | |
1705 | } | |
1706 | return NULL; | |
1707 | } | |
1708 | ||
e4d58b41 | 1709 | static void tcg_target_init(TCGContext *s) |
8289b279 | 1710 | { |
90379ca8 RH |
1711 | /* Only probe for the platform and capabilities if we havn't already |
1712 | determined maximum values at compile time. */ | |
1713 | #ifndef use_vis3_instructions | |
1714 | { | |
1715 | unsigned long hwcap = qemu_getauxval(AT_HWCAP); | |
1716 | use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0; | |
1717 | } | |
1718 | #endif | |
1719 | ||
8289b279 | 1720 | tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffffffff); |
34b1a49c RH |
1721 | tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, ALL_64); |
1722 | ||
8289b279 | 1723 | tcg_regset_set32(tcg_target_call_clobber_regs, 0, |
b3db8758 BS |
1724 | (1 << TCG_REG_G1) | |
1725 | (1 << TCG_REG_G2) | | |
1726 | (1 << TCG_REG_G3) | | |
1727 | (1 << TCG_REG_G4) | | |
1728 | (1 << TCG_REG_G5) | | |
1729 | (1 << TCG_REG_G6) | | |
1730 | (1 << TCG_REG_G7) | | |
8289b279 BS |
1731 | (1 << TCG_REG_O0) | |
1732 | (1 << TCG_REG_O1) | | |
1733 | (1 << TCG_REG_O2) | | |
1734 | (1 << TCG_REG_O3) | | |
1735 | (1 << TCG_REG_O4) | | |
1736 | (1 << TCG_REG_O5) | | |
8289b279 BS |
1737 | (1 << TCG_REG_O7)); |
1738 | ||
1739 | tcg_regset_clear(s->reserved_regs); | |
375816f8 RH |
1740 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G0); /* zero */ |
1741 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G6); /* reserved for os */ | |
1742 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_G7); /* thread pointer */ | |
1743 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_I6); /* frame pointer */ | |
1744 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_I7); /* return address */ | |
1745 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_O6); /* stack pointer */ | |
1746 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_T1); /* for internal use */ | |
1747 | tcg_regset_set_reg(s->reserved_regs, TCG_REG_T2); /* for internal use */ | |
8289b279 | 1748 | } |
cb1977d3 | 1749 | |
9f44adc5 | 1750 | #if SPARC64 |
cb1977d3 | 1751 | # define ELF_HOST_MACHINE EM_SPARCV9 |
9b9c37c3 | 1752 | #else |
cb1977d3 RH |
1753 | # define ELF_HOST_MACHINE EM_SPARC32PLUS |
1754 | # define ELF_HOST_FLAGS EF_SPARC_32PLUS | |
cb1977d3 RH |
1755 | #endif |
1756 | ||
cb1977d3 | 1757 | typedef struct { |
ae18b28d | 1758 | DebugFrameHeader h; |
9f44adc5 | 1759 | uint8_t fde_def_cfa[SPARC64 ? 4 : 2]; |
497a22eb RH |
1760 | uint8_t fde_win_save; |
1761 | uint8_t fde_ret_save[3]; | |
cb1977d3 RH |
1762 | } DebugFrame; |
1763 | ||
ae18b28d RH |
1764 | static const DebugFrame debug_frame = { |
1765 | .h.cie.len = sizeof(DebugFrameCIE)-4, /* length after .len member */ | |
1766 | .h.cie.id = -1, | |
1767 | .h.cie.version = 1, | |
1768 | .h.cie.code_align = 1, | |
1769 | .h.cie.data_align = -sizeof(void *) & 0x7f, | |
1770 | .h.cie.return_column = 15, /* o7 */ | |
cb1977d3 | 1771 | |
497a22eb | 1772 | /* Total FDE size does not include the "len" member. */ |
ae18b28d | 1773 | .h.fde.len = sizeof(DebugFrame) - offsetof(DebugFrame, h.fde.cie_offset), |
497a22eb RH |
1774 | |
1775 | .fde_def_cfa = { | |
9f44adc5 | 1776 | #if SPARC64 |
cb1977d3 RH |
1777 | 12, 30, /* DW_CFA_def_cfa i6, 2047 */ |
1778 | (2047 & 0x7f) | 0x80, (2047 >> 7) | |
1779 | #else | |
1780 | 13, 30 /* DW_CFA_def_cfa_register i6 */ | |
1781 | #endif | |
1782 | }, | |
497a22eb RH |
1783 | .fde_win_save = 0x2d, /* DW_CFA_GNU_window_save */ |
1784 | .fde_ret_save = { 9, 15, 31 }, /* DW_CFA_register o7, i7 */ | |
cb1977d3 RH |
1785 | }; |
1786 | ||
1787 | void tcg_register_jit(void *buf, size_t buf_size) | |
1788 | { | |
cb1977d3 RH |
1789 | tcg_register_jit_int(buf, buf_size, &debug_frame, sizeof(debug_frame)); |
1790 | } | |
5bbd2cae | 1791 | |
a8583393 RH |
1792 | void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_addr, |
1793 | uintptr_t addr) | |
5bbd2cae | 1794 | { |
ab20bdc1 RH |
1795 | intptr_t tb_disp = addr - tc_ptr; |
1796 | intptr_t br_disp = addr - jmp_addr; | |
1797 | tcg_insn_unit i1, i2; | |
1798 | ||
1799 | /* We can reach the entire address space for ILP32. | |
1800 | For LP64, the code_gen_buffer can't be larger than 2GB. */ | |
1801 | tcg_debug_assert(tb_disp == (int32_t)tb_disp); | |
1802 | tcg_debug_assert(br_disp == (int32_t)br_disp); | |
1803 | ||
1804 | if (!USE_REG_TB) { | |
1805 | atomic_set((uint32_t *)jmp_addr, deposit32(CALL, 0, 30, br_disp >> 2)); | |
1806 | flush_icache_range(jmp_addr, jmp_addr + 4); | |
1807 | return; | |
1808 | } | |
5bbd2cae | 1809 | |
ab20bdc1 RH |
1810 | /* This does not exercise the range of the branch, but we do |
1811 | still need to be able to load the new value of TCG_REG_TB. | |
1812 | But this does still happen quite often. */ | |
1813 | if (check_fit_ptr(tb_disp, 13)) { | |
1814 | /* ba,pt %icc, addr */ | |
1815 | i1 = (INSN_OP(0) | INSN_OP2(1) | INSN_COND(COND_A) | |
1816 | | BPCC_ICC | BPCC_PT | INSN_OFF19(br_disp)); | |
1817 | i2 = (ARITH_ADD | INSN_RD(TCG_REG_TB) | INSN_RS1(TCG_REG_TB) | |
1818 | | INSN_IMM13(tb_disp)); | |
1819 | } else if (tb_disp >= 0) { | |
1820 | i1 = SETHI | INSN_RD(TCG_REG_T1) | ((tb_disp & 0xfffffc00) >> 10); | |
1821 | i2 = (ARITH_OR | INSN_RD(TCG_REG_T1) | INSN_RS1(TCG_REG_T1) | |
1822 | | INSN_IMM13(tb_disp & 0x3ff)); | |
1823 | } else { | |
1824 | i1 = SETHI | INSN_RD(TCG_REG_T1) | ((~tb_disp & 0xfffffc00) >> 10); | |
1825 | i2 = (ARITH_XOR | INSN_RD(TCG_REG_T1) | INSN_RS1(TCG_REG_T1) | |
1826 | | INSN_IMM13((tb_disp & 0x3ff) | -0x400)); | |
1827 | } | |
5bbd2cae | 1828 | |
ab20bdc1 RH |
1829 | atomic_set((uint64_t *)jmp_addr, deposit64(i2, 32, 32, i1)); |
1830 | flush_icache_range(jmp_addr, jmp_addr + 8); | |
5bbd2cae | 1831 | } |