]>
Commit | Line | Data |
---|---|---|
1bcfbf03 WX |
1 | /* |
2 | * Tiny Code Generator for QEMU | |
3 | * | |
4 | * Copyright (c) 2021 WANG Xuerui <git@xen0n.name> | |
5 | * | |
6 | * Based on tcg/riscv/tcg-target.c.inc | |
7 | * | |
8 | * Copyright (c) 2018 SiFive, Inc | |
9 | * Copyright (c) 2008-2009 Arnaud Patard <arnaud.patard@rtp-net.org> | |
10 | * Copyright (c) 2009 Aurelien Jarno <aurelien@aurel32.net> | |
11 | * Copyright (c) 2008 Fabrice Bellard | |
12 | * | |
13 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
14 | * of this software and associated documentation files (the "Software"), to deal | |
15 | * in the Software without restriction, including without limitation the rights | |
16 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
17 | * copies of the Software, and to permit persons to whom the Software is | |
18 | * furnished to do so, subject to the following conditions: | |
19 | * | |
20 | * The above copyright notice and this permission notice shall be included in | |
21 | * all copies or substantial portions of the Software. | |
22 | * | |
23 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
24 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
25 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
26 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
27 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
28 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
29 | * THE SOFTWARE. | |
30 | */ | |
31 | ||
32 | #ifdef CONFIG_DEBUG_TCG | |
33 | static const char * const tcg_target_reg_names[TCG_TARGET_NB_REGS] = { | |
34 | "zero", | |
35 | "ra", | |
36 | "tp", | |
37 | "sp", | |
38 | "a0", | |
39 | "a1", | |
40 | "a2", | |
41 | "a3", | |
42 | "a4", | |
43 | "a5", | |
44 | "a6", | |
45 | "a7", | |
46 | "t0", | |
47 | "t1", | |
48 | "t2", | |
49 | "t3", | |
50 | "t4", | |
51 | "t5", | |
52 | "t6", | |
53 | "t7", | |
54 | "t8", | |
55 | "r21", /* reserved in the LP64* ABI, hence no ABI name */ | |
56 | "s9", | |
57 | "s0", | |
58 | "s1", | |
59 | "s2", | |
60 | "s3", | |
61 | "s4", | |
62 | "s5", | |
63 | "s6", | |
64 | "s7", | |
65 | "s8" | |
66 | }; | |
67 | #endif | |
68 | ||
69 | static const int tcg_target_reg_alloc_order[] = { | |
70 | /* Registers preserved across calls */ | |
71 | /* TCG_REG_S0 reserved for TCG_AREG0 */ | |
72 | TCG_REG_S1, | |
73 | TCG_REG_S2, | |
74 | TCG_REG_S3, | |
75 | TCG_REG_S4, | |
76 | TCG_REG_S5, | |
77 | TCG_REG_S6, | |
78 | TCG_REG_S7, | |
79 | TCG_REG_S8, | |
80 | TCG_REG_S9, | |
81 | ||
82 | /* Registers (potentially) clobbered across calls */ | |
83 | TCG_REG_T0, | |
84 | TCG_REG_T1, | |
85 | TCG_REG_T2, | |
86 | TCG_REG_T3, | |
87 | TCG_REG_T4, | |
88 | TCG_REG_T5, | |
89 | TCG_REG_T6, | |
90 | TCG_REG_T7, | |
91 | TCG_REG_T8, | |
92 | ||
93 | /* Argument registers, opposite order of allocation. */ | |
94 | TCG_REG_A7, | |
95 | TCG_REG_A6, | |
96 | TCG_REG_A5, | |
97 | TCG_REG_A4, | |
98 | TCG_REG_A3, | |
99 | TCG_REG_A2, | |
100 | TCG_REG_A1, | |
101 | TCG_REG_A0, | |
102 | }; | |
103 | ||
104 | static const int tcg_target_call_iarg_regs[] = { | |
105 | TCG_REG_A0, | |
106 | TCG_REG_A1, | |
107 | TCG_REG_A2, | |
108 | TCG_REG_A3, | |
109 | TCG_REG_A4, | |
110 | TCG_REG_A5, | |
111 | TCG_REG_A6, | |
112 | TCG_REG_A7, | |
113 | }; | |
114 | ||
115 | static const int tcg_target_call_oarg_regs[] = { | |
116 | TCG_REG_A0, | |
117 | TCG_REG_A1, | |
118 | }; | |
ba0cdd80 WX |
119 | |
120 | #define TCG_CT_CONST_ZERO 0x100 | |
121 | #define TCG_CT_CONST_S12 0x200 | |
122 | #define TCG_CT_CONST_N12 0x400 | |
123 | #define TCG_CT_CONST_U12 0x800 | |
124 | #define TCG_CT_CONST_C12 0x1000 | |
125 | #define TCG_CT_CONST_WSZ 0x2000 | |
126 | ||
127 | #define ALL_GENERAL_REGS MAKE_64BIT_MASK(0, 32) | |
128 | /* | |
129 | * For softmmu, we need to avoid conflicts with the first 5 | |
130 | * argument registers to call the helper. Some of these are | |
131 | * also used for the tlb lookup. | |
132 | */ | |
133 | #ifdef CONFIG_SOFTMMU | |
134 | #define SOFTMMU_RESERVE_REGS MAKE_64BIT_MASK(TCG_REG_A0, 5) | |
135 | #else | |
136 | #define SOFTMMU_RESERVE_REGS 0 | |
137 | #endif | |
138 | ||
139 | ||
140 | static inline tcg_target_long sextreg(tcg_target_long val, int pos, int len) | |
141 | { | |
142 | return sextract64(val, pos, len); | |
143 | } | |
144 | ||
145 | /* test if a constant matches the constraint */ | |
146 | static bool tcg_target_const_match(int64_t val, TCGType type, int ct) | |
147 | { | |
148 | if (ct & TCG_CT_CONST) { | |
149 | return true; | |
150 | } | |
151 | if ((ct & TCG_CT_CONST_ZERO) && val == 0) { | |
152 | return true; | |
153 | } | |
154 | if ((ct & TCG_CT_CONST_S12) && val == sextreg(val, 0, 12)) { | |
155 | return true; | |
156 | } | |
157 | if ((ct & TCG_CT_CONST_N12) && -val == sextreg(-val, 0, 12)) { | |
158 | return true; | |
159 | } | |
160 | if ((ct & TCG_CT_CONST_U12) && val >= 0 && val <= 0xfff) { | |
161 | return true; | |
162 | } | |
163 | if ((ct & TCG_CT_CONST_C12) && ~val >= 0 && ~val <= 0xfff) { | |
164 | return true; | |
165 | } | |
166 | if ((ct & TCG_CT_CONST_WSZ) && val == (type == TCG_TYPE_I32 ? 32 : 64)) { | |
167 | return true; | |
168 | } | |
169 | return false; | |
170 | } | |
bf8c1c81 WX |
171 | |
172 | /* | |
173 | * Relocations | |
174 | */ | |
175 | ||
176 | /* | |
177 | * Relocation records defined in LoongArch ELF psABI v1.00 is way too | |
178 | * complicated; a whopping stack machine is needed to stuff the fields, at | |
179 | * the very least one SOP_PUSH and one SOP_POP (of the correct format) are | |
180 | * needed. | |
181 | * | |
182 | * Hence, define our own simpler relocation types. Numbers are chosen as to | |
183 | * not collide with potential future additions to the true ELF relocation | |
184 | * type enum. | |
185 | */ | |
186 | ||
187 | /* Field Sk16, shifted right by 2; suitable for conditional jumps */ | |
188 | #define R_LOONGARCH_BR_SK16 256 | |
189 | /* Field Sd10k16, shifted right by 2; suitable for B and BL */ | |
190 | #define R_LOONGARCH_BR_SD10K16 257 | |
191 | ||
192 | static bool reloc_br_sk16(tcg_insn_unit *src_rw, const tcg_insn_unit *target) | |
193 | { | |
194 | const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); | |
195 | intptr_t offset = (intptr_t)target - (intptr_t)src_rx; | |
196 | ||
197 | tcg_debug_assert((offset & 3) == 0); | |
198 | offset >>= 2; | |
199 | if (offset == sextreg(offset, 0, 16)) { | |
200 | *src_rw = deposit64(*src_rw, 10, 16, offset); | |
201 | return true; | |
202 | } | |
203 | ||
204 | return false; | |
205 | } | |
206 | ||
207 | static bool reloc_br_sd10k16(tcg_insn_unit *src_rw, | |
208 | const tcg_insn_unit *target) | |
209 | { | |
210 | const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw); | |
211 | intptr_t offset = (intptr_t)target - (intptr_t)src_rx; | |
212 | ||
213 | tcg_debug_assert((offset & 3) == 0); | |
214 | offset >>= 2; | |
215 | if (offset == sextreg(offset, 0, 26)) { | |
216 | *src_rw = deposit64(*src_rw, 0, 10, offset >> 16); /* slot d10 */ | |
217 | *src_rw = deposit64(*src_rw, 10, 16, offset); /* slot k16 */ | |
218 | return true; | |
219 | } | |
220 | ||
221 | return false; | |
222 | } | |
223 | ||
224 | static bool patch_reloc(tcg_insn_unit *code_ptr, int type, | |
225 | intptr_t value, intptr_t addend) | |
226 | { | |
227 | tcg_debug_assert(addend == 0); | |
228 | switch (type) { | |
229 | case R_LOONGARCH_BR_SK16: | |
230 | return reloc_br_sk16(code_ptr, (tcg_insn_unit *)value); | |
231 | case R_LOONGARCH_BR_SD10K16: | |
232 | return reloc_br_sd10k16(code_ptr, (tcg_insn_unit *)value); | |
233 | default: | |
234 | g_assert_not_reached(); | |
235 | } | |
236 | } | |
fae2361d WX |
237 | |
238 | #include "tcg-insn-defs.c.inc" | |
239 | ||
240 | /* | |
241 | * TCG intrinsics | |
242 | */ | |
243 | ||
244 | static void tcg_out_mb(TCGContext *s, TCGArg a0) | |
245 | { | |
246 | /* Baseline LoongArch only has the full barrier, unfortunately. */ | |
247 | tcg_out_opc_dbar(s, 0); | |
248 | } | |
249 | ||
dacc5172 WX |
250 | static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) |
251 | { | |
252 | if (ret == arg) { | |
253 | return true; | |
254 | } | |
255 | switch (type) { | |
256 | case TCG_TYPE_I32: | |
257 | case TCG_TYPE_I64: | |
258 | /* | |
259 | * Conventional register-register move used in LoongArch is | |
260 | * `or dst, src, zero`. | |
261 | */ | |
262 | tcg_out_opc_or(s, ret, arg, TCG_REG_ZERO); | |
263 | break; | |
264 | default: | |
265 | g_assert_not_reached(); | |
266 | } | |
267 | return true; | |
268 | } | |
269 | ||
270 | static bool imm_part_needs_loading(bool high_bits_are_ones, | |
271 | tcg_target_long part) | |
272 | { | |
273 | if (high_bits_are_ones) { | |
274 | return part != -1; | |
275 | } else { | |
276 | return part != 0; | |
277 | } | |
278 | } | |
279 | ||
280 | /* Loads a 32-bit immediate into rd, sign-extended. */ | |
281 | static void tcg_out_movi_i32(TCGContext *s, TCGReg rd, int32_t val) | |
282 | { | |
283 | tcg_target_long lo = sextreg(val, 0, 12); | |
284 | tcg_target_long hi12 = sextreg(val, 12, 20); | |
285 | ||
286 | /* Single-instruction cases. */ | |
287 | if (lo == val) { | |
288 | /* val fits in simm12: addi.w rd, zero, val */ | |
289 | tcg_out_opc_addi_w(s, rd, TCG_REG_ZERO, val); | |
290 | return; | |
291 | } | |
292 | if (0x800 <= val && val <= 0xfff) { | |
293 | /* val fits in uimm12: ori rd, zero, val */ | |
294 | tcg_out_opc_ori(s, rd, TCG_REG_ZERO, val); | |
295 | return; | |
296 | } | |
297 | ||
298 | /* High bits must be set; load with lu12i.w + optional ori. */ | |
299 | tcg_out_opc_lu12i_w(s, rd, hi12); | |
300 | if (lo != 0) { | |
301 | tcg_out_opc_ori(s, rd, rd, lo & 0xfff); | |
302 | } | |
303 | } | |
304 | ||
305 | static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, | |
306 | tcg_target_long val) | |
307 | { | |
308 | /* | |
309 | * LoongArch conventionally loads 64-bit immediates in at most 4 steps, | |
310 | * with dedicated instructions for filling the respective bitfields | |
311 | * below: | |
312 | * | |
313 | * 6 5 4 3 | |
314 | * 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 | |
315 | * +-----------------------+---------------------------------------+... | |
316 | * | hi52 | hi32 | | |
317 | * +-----------------------+---------------------------------------+... | |
318 | * 3 2 1 | |
319 | * 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 | |
320 | * ...+-------------------------------------+-------------------------+ | |
321 | * | hi12 | lo | | |
322 | * ...+-------------------------------------+-------------------------+ | |
323 | * | |
324 | * Check if val belong to one of the several fast cases, before falling | |
325 | * back to the slow path. | |
326 | */ | |
327 | ||
328 | intptr_t pc_offset; | |
329 | tcg_target_long val_lo, val_hi, pc_hi, offset_hi; | |
330 | tcg_target_long hi32, hi52; | |
331 | bool rd_high_bits_are_ones; | |
332 | ||
333 | /* Value fits in signed i32. */ | |
334 | if (type == TCG_TYPE_I32 || val == (int32_t)val) { | |
335 | tcg_out_movi_i32(s, rd, val); | |
336 | return; | |
337 | } | |
338 | ||
339 | /* PC-relative cases. */ | |
340 | pc_offset = tcg_pcrel_diff(s, (void *)val); | |
341 | if (pc_offset == sextreg(pc_offset, 0, 22) && (pc_offset & 3) == 0) { | |
342 | /* Single pcaddu2i. */ | |
343 | tcg_out_opc_pcaddu2i(s, rd, pc_offset >> 2); | |
344 | return; | |
345 | } | |
346 | ||
347 | if (pc_offset == (int32_t)pc_offset) { | |
348 | /* Offset within 32 bits; load with pcalau12i + ori. */ | |
349 | val_lo = sextreg(val, 0, 12); | |
350 | val_hi = val >> 12; | |
351 | pc_hi = (val - pc_offset) >> 12; | |
352 | offset_hi = val_hi - pc_hi; | |
353 | ||
354 | tcg_debug_assert(offset_hi == sextreg(offset_hi, 0, 20)); | |
355 | tcg_out_opc_pcalau12i(s, rd, offset_hi); | |
356 | if (val_lo != 0) { | |
357 | tcg_out_opc_ori(s, rd, rd, val_lo & 0xfff); | |
358 | } | |
359 | return; | |
360 | } | |
361 | ||
362 | hi32 = sextreg(val, 32, 20); | |
363 | hi52 = sextreg(val, 52, 12); | |
364 | ||
365 | /* Single cu52i.d case. */ | |
366 | if (ctz64(val) >= 52) { | |
367 | tcg_out_opc_cu52i_d(s, rd, TCG_REG_ZERO, hi52); | |
368 | return; | |
369 | } | |
370 | ||
371 | /* Slow path. Initialize the low 32 bits, then concat high bits. */ | |
372 | tcg_out_movi_i32(s, rd, val); | |
373 | rd_high_bits_are_ones = (int32_t)val < 0; | |
374 | ||
375 | if (imm_part_needs_loading(rd_high_bits_are_ones, hi32)) { | |
376 | tcg_out_opc_cu32i_d(s, rd, hi32); | |
377 | rd_high_bits_are_ones = hi32 < 0; | |
378 | } | |
379 | ||
380 | if (imm_part_needs_loading(rd_high_bits_are_ones, hi52)) { | |
381 | tcg_out_opc_cu52i_d(s, rd, rd, hi52); | |
382 | } | |
383 | } | |
384 | ||
6be08fcf WX |
385 | static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg) |
386 | { | |
387 | tcg_out_opc_andi(s, ret, arg, 0xff); | |
388 | } | |
389 | ||
390 | static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg) | |
391 | { | |
392 | tcg_out_opc_bstrpick_w(s, ret, arg, 0, 15); | |
393 | } | |
394 | ||
395 | static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg) | |
396 | { | |
397 | tcg_out_opc_bstrpick_d(s, ret, arg, 0, 31); | |
398 | } | |
399 | ||
400 | static void tcg_out_ext8s(TCGContext *s, TCGReg ret, TCGReg arg) | |
401 | { | |
402 | tcg_out_opc_sext_b(s, ret, arg); | |
403 | } | |
404 | ||
405 | static void tcg_out_ext16s(TCGContext *s, TCGReg ret, TCGReg arg) | |
406 | { | |
407 | tcg_out_opc_sext_h(s, ret, arg); | |
408 | } | |
409 | ||
410 | static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg) | |
411 | { | |
412 | tcg_out_opc_addi_w(s, ret, arg, 0); | |
413 | } | |
414 | ||
fde69301 WX |
415 | static void tcg_out_clzctz(TCGContext *s, LoongArchInsn opc, |
416 | TCGReg a0, TCGReg a1, TCGReg a2, | |
417 | bool c2, bool is_32bit) | |
418 | { | |
419 | if (c2) { | |
420 | /* | |
421 | * Fast path: semantics already satisfied due to constraint and | |
422 | * insn behavior, single instruction is enough. | |
423 | */ | |
424 | tcg_debug_assert(a2 == (is_32bit ? 32 : 64)); | |
425 | /* all clz/ctz insns belong to DJ-format */ | |
426 | tcg_out32(s, encode_dj_insn(opc, a0, a1)); | |
427 | return; | |
428 | } | |
429 | ||
430 | tcg_out32(s, encode_dj_insn(opc, TCG_REG_TMP0, a1)); | |
431 | /* a0 = a1 ? REG_TMP0 : a2 */ | |
432 | tcg_out_opc_maskeqz(s, TCG_REG_TMP0, TCG_REG_TMP0, a1); | |
433 | tcg_out_opc_masknez(s, a0, a2, a1); | |
434 | tcg_out_opc_or(s, a0, TCG_REG_TMP0, a0); | |
435 | } | |
436 | ||
9ee775cf WX |
437 | static void tcg_out_setcond(TCGContext *s, TCGCond cond, TCGReg ret, |
438 | TCGReg arg1, TCGReg arg2, bool c2) | |
439 | { | |
440 | TCGReg tmp; | |
441 | ||
442 | if (c2) { | |
443 | tcg_debug_assert(arg2 == 0); | |
444 | } | |
445 | ||
446 | switch (cond) { | |
447 | case TCG_COND_EQ: | |
448 | if (c2) { | |
449 | tmp = arg1; | |
450 | } else { | |
451 | tcg_out_opc_sub_d(s, ret, arg1, arg2); | |
452 | tmp = ret; | |
453 | } | |
454 | tcg_out_opc_sltui(s, ret, tmp, 1); | |
455 | break; | |
456 | case TCG_COND_NE: | |
457 | if (c2) { | |
458 | tmp = arg1; | |
459 | } else { | |
460 | tcg_out_opc_sub_d(s, ret, arg1, arg2); | |
461 | tmp = ret; | |
462 | } | |
463 | tcg_out_opc_sltu(s, ret, TCG_REG_ZERO, tmp); | |
464 | break; | |
465 | case TCG_COND_LT: | |
466 | tcg_out_opc_slt(s, ret, arg1, arg2); | |
467 | break; | |
468 | case TCG_COND_GE: | |
469 | tcg_out_opc_slt(s, ret, arg1, arg2); | |
470 | tcg_out_opc_xori(s, ret, ret, 1); | |
471 | break; | |
472 | case TCG_COND_LE: | |
473 | tcg_out_setcond(s, TCG_COND_GE, ret, arg2, arg1, false); | |
474 | break; | |
475 | case TCG_COND_GT: | |
476 | tcg_out_setcond(s, TCG_COND_LT, ret, arg2, arg1, false); | |
477 | break; | |
478 | case TCG_COND_LTU: | |
479 | tcg_out_opc_sltu(s, ret, arg1, arg2); | |
480 | break; | |
481 | case TCG_COND_GEU: | |
482 | tcg_out_opc_sltu(s, ret, arg1, arg2); | |
483 | tcg_out_opc_xori(s, ret, ret, 1); | |
484 | break; | |
485 | case TCG_COND_LEU: | |
486 | tcg_out_setcond(s, TCG_COND_GEU, ret, arg2, arg1, false); | |
487 | break; | |
488 | case TCG_COND_GTU: | |
489 | tcg_out_setcond(s, TCG_COND_LTU, ret, arg2, arg1, false); | |
490 | break; | |
491 | default: | |
492 | g_assert_not_reached(); | |
493 | break; | |
494 | } | |
495 | } | |
496 | ||
94505c02 WX |
497 | /* |
498 | * Branch helpers | |
499 | */ | |
500 | ||
501 | static const struct { | |
502 | LoongArchInsn op; | |
503 | bool swap; | |
504 | } tcg_brcond_to_loongarch[] = { | |
505 | [TCG_COND_EQ] = { OPC_BEQ, false }, | |
506 | [TCG_COND_NE] = { OPC_BNE, false }, | |
507 | [TCG_COND_LT] = { OPC_BGT, true }, | |
508 | [TCG_COND_GE] = { OPC_BLE, true }, | |
509 | [TCG_COND_LE] = { OPC_BLE, false }, | |
510 | [TCG_COND_GT] = { OPC_BGT, false }, | |
511 | [TCG_COND_LTU] = { OPC_BGTU, true }, | |
512 | [TCG_COND_GEU] = { OPC_BLEU, true }, | |
513 | [TCG_COND_LEU] = { OPC_BLEU, false }, | |
514 | [TCG_COND_GTU] = { OPC_BGTU, false } | |
515 | }; | |
516 | ||
517 | static void tcg_out_brcond(TCGContext *s, TCGCond cond, TCGReg arg1, | |
518 | TCGReg arg2, TCGLabel *l) | |
519 | { | |
520 | LoongArchInsn op = tcg_brcond_to_loongarch[cond].op; | |
521 | ||
522 | tcg_debug_assert(op != 0); | |
523 | ||
524 | if (tcg_brcond_to_loongarch[cond].swap) { | |
525 | TCGReg t = arg1; | |
526 | arg1 = arg2; | |
527 | arg2 = t; | |
528 | } | |
529 | ||
530 | /* all conditional branch insns belong to DJSk16-format */ | |
531 | tcg_out_reloc(s, s->code_ptr, R_LOONGARCH_BR_SK16, l, 0); | |
532 | tcg_out32(s, encode_djsk16_insn(op, arg1, arg2, 0)); | |
533 | } | |
534 | ||
a26d99d7 WX |
535 | static void tcg_out_call_int(TCGContext *s, const tcg_insn_unit *arg, bool tail) |
536 | { | |
537 | TCGReg link = tail ? TCG_REG_ZERO : TCG_REG_RA; | |
538 | ptrdiff_t offset = tcg_pcrel_diff(s, arg); | |
539 | ||
540 | tcg_debug_assert((offset & 3) == 0); | |
541 | if (offset == sextreg(offset, 0, 28)) { | |
542 | /* short jump: +/- 256MiB */ | |
543 | if (tail) { | |
544 | tcg_out_opc_b(s, offset >> 2); | |
545 | } else { | |
546 | tcg_out_opc_bl(s, offset >> 2); | |
547 | } | |
548 | } else if (offset == sextreg(offset, 0, 38)) { | |
549 | /* long jump: +/- 256GiB */ | |
550 | tcg_target_long lo = sextreg(offset, 0, 18); | |
551 | tcg_target_long hi = offset - lo; | |
552 | tcg_out_opc_pcaddu18i(s, TCG_REG_TMP0, hi >> 18); | |
553 | tcg_out_opc_jirl(s, link, TCG_REG_TMP0, lo >> 2); | |
554 | } else { | |
555 | /* far jump: 64-bit */ | |
556 | tcg_target_long lo = sextreg((tcg_target_long)arg, 0, 18); | |
557 | tcg_target_long hi = (tcg_target_long)arg - lo; | |
558 | tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_TMP0, hi); | |
559 | tcg_out_opc_jirl(s, link, TCG_REG_TMP0, lo >> 2); | |
560 | } | |
561 | } | |
562 | ||
563 | static void tcg_out_call(TCGContext *s, const tcg_insn_unit *arg) | |
564 | { | |
565 | tcg_out_call_int(s, arg, false); | |
566 | } | |
567 | ||
fae2361d WX |
568 | /* |
569 | * Entry-points | |
570 | */ | |
571 | ||
572 | static void tcg_out_op(TCGContext *s, TCGOpcode opc, | |
573 | const TCGArg args[TCG_MAX_OP_ARGS], | |
574 | const int const_args[TCG_MAX_OP_ARGS]) | |
575 | { | |
576 | TCGArg a0 = args[0]; | |
6be08fcf | 577 | TCGArg a1 = args[1]; |
97b2fafb WX |
578 | TCGArg a2 = args[2]; |
579 | int c2 = const_args[2]; | |
fae2361d WX |
580 | |
581 | switch (opc) { | |
582 | case INDEX_op_mb: | |
583 | tcg_out_mb(s, a0); | |
584 | break; | |
585 | ||
e3b15766 WX |
586 | case INDEX_op_goto_ptr: |
587 | tcg_out_opc_jirl(s, TCG_REG_ZERO, a0, 0); | |
588 | break; | |
589 | ||
94505c02 WX |
590 | case INDEX_op_br: |
591 | tcg_out_reloc(s, s->code_ptr, R_LOONGARCH_BR_SD10K16, arg_label(a0), | |
592 | 0); | |
593 | tcg_out_opc_b(s, 0); | |
594 | break; | |
595 | ||
596 | case INDEX_op_brcond_i32: | |
597 | case INDEX_op_brcond_i64: | |
598 | tcg_out_brcond(s, a2, a0, a1, arg_label(args[3])); | |
599 | break; | |
600 | ||
6be08fcf WX |
601 | case INDEX_op_ext8s_i32: |
602 | case INDEX_op_ext8s_i64: | |
603 | tcg_out_ext8s(s, a0, a1); | |
604 | break; | |
605 | ||
606 | case INDEX_op_ext8u_i32: | |
607 | case INDEX_op_ext8u_i64: | |
608 | tcg_out_ext8u(s, a0, a1); | |
609 | break; | |
610 | ||
611 | case INDEX_op_ext16s_i32: | |
612 | case INDEX_op_ext16s_i64: | |
613 | tcg_out_ext16s(s, a0, a1); | |
614 | break; | |
615 | ||
616 | case INDEX_op_ext16u_i32: | |
617 | case INDEX_op_ext16u_i64: | |
618 | tcg_out_ext16u(s, a0, a1); | |
619 | break; | |
620 | ||
621 | case INDEX_op_ext32u_i64: | |
622 | case INDEX_op_extu_i32_i64: | |
623 | tcg_out_ext32u(s, a0, a1); | |
624 | break; | |
625 | ||
626 | case INDEX_op_ext32s_i64: | |
627 | case INDEX_op_extrl_i64_i32: | |
628 | case INDEX_op_ext_i32_i64: | |
629 | tcg_out_ext32s(s, a0, a1); | |
630 | break; | |
631 | ||
632 | case INDEX_op_extrh_i64_i32: | |
633 | tcg_out_opc_srai_d(s, a0, a1, 32); | |
634 | break; | |
635 | ||
97b2fafb WX |
636 | case INDEX_op_not_i32: |
637 | case INDEX_op_not_i64: | |
638 | tcg_out_opc_nor(s, a0, a1, TCG_REG_ZERO); | |
639 | break; | |
640 | ||
641 | case INDEX_op_nor_i32: | |
642 | case INDEX_op_nor_i64: | |
643 | if (c2) { | |
644 | tcg_out_opc_ori(s, a0, a1, a2); | |
645 | tcg_out_opc_nor(s, a0, a0, TCG_REG_ZERO); | |
646 | } else { | |
647 | tcg_out_opc_nor(s, a0, a1, a2); | |
648 | } | |
649 | break; | |
650 | ||
651 | case INDEX_op_andc_i32: | |
652 | case INDEX_op_andc_i64: | |
653 | if (c2) { | |
654 | /* guaranteed to fit due to constraint */ | |
655 | tcg_out_opc_andi(s, a0, a1, ~a2); | |
656 | } else { | |
657 | tcg_out_opc_andn(s, a0, a1, a2); | |
658 | } | |
659 | break; | |
660 | ||
661 | case INDEX_op_orc_i32: | |
662 | case INDEX_op_orc_i64: | |
663 | if (c2) { | |
664 | /* guaranteed to fit due to constraint */ | |
665 | tcg_out_opc_ori(s, a0, a1, ~a2); | |
666 | } else { | |
667 | tcg_out_opc_orn(s, a0, a1, a2); | |
668 | } | |
669 | break; | |
670 | ||
671 | case INDEX_op_and_i32: | |
672 | case INDEX_op_and_i64: | |
673 | if (c2) { | |
674 | tcg_out_opc_andi(s, a0, a1, a2); | |
675 | } else { | |
676 | tcg_out_opc_and(s, a0, a1, a2); | |
677 | } | |
678 | break; | |
679 | ||
680 | case INDEX_op_or_i32: | |
681 | case INDEX_op_or_i64: | |
682 | if (c2) { | |
683 | tcg_out_opc_ori(s, a0, a1, a2); | |
684 | } else { | |
685 | tcg_out_opc_or(s, a0, a1, a2); | |
686 | } | |
687 | break; | |
688 | ||
689 | case INDEX_op_xor_i32: | |
690 | case INDEX_op_xor_i64: | |
691 | if (c2) { | |
692 | tcg_out_opc_xori(s, a0, a1, a2); | |
693 | } else { | |
694 | tcg_out_opc_xor(s, a0, a1, a2); | |
695 | } | |
696 | break; | |
697 | ||
7257809f WX |
698 | case INDEX_op_extract_i32: |
699 | tcg_out_opc_bstrpick_w(s, a0, a1, a2, a2 + args[3] - 1); | |
700 | break; | |
701 | case INDEX_op_extract_i64: | |
702 | tcg_out_opc_bstrpick_d(s, a0, a1, a2, a2 + args[3] - 1); | |
703 | break; | |
704 | ||
705 | case INDEX_op_deposit_i32: | |
706 | tcg_out_opc_bstrins_w(s, a0, a2, args[3], args[3] + args[4] - 1); | |
707 | break; | |
708 | case INDEX_op_deposit_i64: | |
709 | tcg_out_opc_bstrins_d(s, a0, a2, args[3], args[3] + args[4] - 1); | |
710 | break; | |
711 | ||
4ab2aff0 WX |
712 | case INDEX_op_bswap16_i32: |
713 | case INDEX_op_bswap16_i64: | |
714 | tcg_out_opc_revb_2h(s, a0, a1); | |
715 | if (a2 & TCG_BSWAP_OS) { | |
716 | tcg_out_ext16s(s, a0, a0); | |
717 | } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { | |
718 | tcg_out_ext16u(s, a0, a0); | |
719 | } | |
720 | break; | |
721 | ||
722 | case INDEX_op_bswap32_i32: | |
723 | /* All 32-bit values are computed sign-extended in the register. */ | |
724 | a2 = TCG_BSWAP_OS; | |
725 | /* fallthrough */ | |
726 | case INDEX_op_bswap32_i64: | |
727 | tcg_out_opc_revb_2w(s, a0, a1); | |
728 | if (a2 & TCG_BSWAP_OS) { | |
729 | tcg_out_ext32s(s, a0, a0); | |
730 | } else if ((a2 & (TCG_BSWAP_IZ | TCG_BSWAP_OZ)) == TCG_BSWAP_OZ) { | |
731 | tcg_out_ext32u(s, a0, a0); | |
732 | } | |
733 | break; | |
734 | ||
735 | case INDEX_op_bswap64_i64: | |
736 | tcg_out_opc_revb_d(s, a0, a1); | |
737 | break; | |
738 | ||
fde69301 WX |
739 | case INDEX_op_clz_i32: |
740 | tcg_out_clzctz(s, OPC_CLZ_W, a0, a1, a2, c2, true); | |
741 | break; | |
742 | case INDEX_op_clz_i64: | |
743 | tcg_out_clzctz(s, OPC_CLZ_D, a0, a1, a2, c2, false); | |
744 | break; | |
745 | ||
746 | case INDEX_op_ctz_i32: | |
747 | tcg_out_clzctz(s, OPC_CTZ_W, a0, a1, a2, c2, true); | |
748 | break; | |
749 | case INDEX_op_ctz_i64: | |
750 | tcg_out_clzctz(s, OPC_CTZ_D, a0, a1, a2, c2, false); | |
751 | break; | |
752 | ||
a164010b WX |
753 | case INDEX_op_shl_i32: |
754 | if (c2) { | |
755 | tcg_out_opc_slli_w(s, a0, a1, a2 & 0x1f); | |
756 | } else { | |
757 | tcg_out_opc_sll_w(s, a0, a1, a2); | |
758 | } | |
759 | break; | |
760 | case INDEX_op_shl_i64: | |
761 | if (c2) { | |
762 | tcg_out_opc_slli_d(s, a0, a1, a2 & 0x3f); | |
763 | } else { | |
764 | tcg_out_opc_sll_d(s, a0, a1, a2); | |
765 | } | |
766 | break; | |
767 | ||
768 | case INDEX_op_shr_i32: | |
769 | if (c2) { | |
770 | tcg_out_opc_srli_w(s, a0, a1, a2 & 0x1f); | |
771 | } else { | |
772 | tcg_out_opc_srl_w(s, a0, a1, a2); | |
773 | } | |
774 | break; | |
775 | case INDEX_op_shr_i64: | |
776 | if (c2) { | |
777 | tcg_out_opc_srli_d(s, a0, a1, a2 & 0x3f); | |
778 | } else { | |
779 | tcg_out_opc_srl_d(s, a0, a1, a2); | |
780 | } | |
781 | break; | |
782 | ||
783 | case INDEX_op_sar_i32: | |
784 | if (c2) { | |
785 | tcg_out_opc_srai_w(s, a0, a1, a2 & 0x1f); | |
786 | } else { | |
787 | tcg_out_opc_sra_w(s, a0, a1, a2); | |
788 | } | |
789 | break; | |
790 | case INDEX_op_sar_i64: | |
791 | if (c2) { | |
792 | tcg_out_opc_srai_d(s, a0, a1, a2 & 0x3f); | |
793 | } else { | |
794 | tcg_out_opc_sra_d(s, a0, a1, a2); | |
795 | } | |
796 | break; | |
797 | ||
798 | case INDEX_op_rotl_i32: | |
799 | /* transform into equivalent rotr/rotri */ | |
800 | if (c2) { | |
801 | tcg_out_opc_rotri_w(s, a0, a1, (32 - a2) & 0x1f); | |
802 | } else { | |
803 | tcg_out_opc_sub_w(s, TCG_REG_TMP0, TCG_REG_ZERO, a2); | |
804 | tcg_out_opc_rotr_w(s, a0, a1, TCG_REG_TMP0); | |
805 | } | |
806 | break; | |
807 | case INDEX_op_rotl_i64: | |
808 | /* transform into equivalent rotr/rotri */ | |
809 | if (c2) { | |
810 | tcg_out_opc_rotri_d(s, a0, a1, (64 - a2) & 0x3f); | |
811 | } else { | |
812 | tcg_out_opc_sub_w(s, TCG_REG_TMP0, TCG_REG_ZERO, a2); | |
813 | tcg_out_opc_rotr_d(s, a0, a1, TCG_REG_TMP0); | |
814 | } | |
815 | break; | |
816 | ||
817 | case INDEX_op_rotr_i32: | |
818 | if (c2) { | |
819 | tcg_out_opc_rotri_w(s, a0, a1, a2 & 0x1f); | |
820 | } else { | |
821 | tcg_out_opc_rotr_w(s, a0, a1, a2); | |
822 | } | |
823 | break; | |
824 | case INDEX_op_rotr_i64: | |
825 | if (c2) { | |
826 | tcg_out_opc_rotri_d(s, a0, a1, a2 & 0x3f); | |
827 | } else { | |
828 | tcg_out_opc_rotr_d(s, a0, a1, a2); | |
829 | } | |
830 | break; | |
831 | ||
39f54ce5 WX |
832 | case INDEX_op_add_i32: |
833 | if (c2) { | |
834 | tcg_out_opc_addi_w(s, a0, a1, a2); | |
835 | } else { | |
836 | tcg_out_opc_add_w(s, a0, a1, a2); | |
837 | } | |
838 | break; | |
839 | case INDEX_op_add_i64: | |
840 | if (c2) { | |
841 | tcg_out_opc_addi_d(s, a0, a1, a2); | |
842 | } else { | |
843 | tcg_out_opc_add_d(s, a0, a1, a2); | |
844 | } | |
845 | break; | |
846 | ||
847 | case INDEX_op_sub_i32: | |
848 | if (c2) { | |
849 | tcg_out_opc_addi_w(s, a0, a1, -a2); | |
850 | } else { | |
851 | tcg_out_opc_sub_w(s, a0, a1, a2); | |
852 | } | |
853 | break; | |
854 | case INDEX_op_sub_i64: | |
855 | if (c2) { | |
856 | tcg_out_opc_addi_d(s, a0, a1, -a2); | |
857 | } else { | |
858 | tcg_out_opc_sub_d(s, a0, a1, a2); | |
859 | } | |
860 | break; | |
861 | ||
ff13c196 WX |
862 | case INDEX_op_mul_i32: |
863 | tcg_out_opc_mul_w(s, a0, a1, a2); | |
864 | break; | |
865 | case INDEX_op_mul_i64: | |
866 | tcg_out_opc_mul_d(s, a0, a1, a2); | |
867 | break; | |
868 | ||
869 | case INDEX_op_mulsh_i32: | |
870 | tcg_out_opc_mulh_w(s, a0, a1, a2); | |
871 | break; | |
872 | case INDEX_op_mulsh_i64: | |
873 | tcg_out_opc_mulh_d(s, a0, a1, a2); | |
874 | break; | |
875 | ||
876 | case INDEX_op_muluh_i32: | |
877 | tcg_out_opc_mulh_wu(s, a0, a1, a2); | |
878 | break; | |
879 | case INDEX_op_muluh_i64: | |
880 | tcg_out_opc_mulh_du(s, a0, a1, a2); | |
881 | break; | |
882 | ||
883 | case INDEX_op_div_i32: | |
884 | tcg_out_opc_div_w(s, a0, a1, a2); | |
885 | break; | |
886 | case INDEX_op_div_i64: | |
887 | tcg_out_opc_div_d(s, a0, a1, a2); | |
888 | break; | |
889 | ||
890 | case INDEX_op_divu_i32: | |
891 | tcg_out_opc_div_wu(s, a0, a1, a2); | |
892 | break; | |
893 | case INDEX_op_divu_i64: | |
894 | tcg_out_opc_div_du(s, a0, a1, a2); | |
895 | break; | |
896 | ||
897 | case INDEX_op_rem_i32: | |
898 | tcg_out_opc_mod_w(s, a0, a1, a2); | |
899 | break; | |
900 | case INDEX_op_rem_i64: | |
901 | tcg_out_opc_mod_d(s, a0, a1, a2); | |
902 | break; | |
903 | ||
904 | case INDEX_op_remu_i32: | |
905 | tcg_out_opc_mod_wu(s, a0, a1, a2); | |
906 | break; | |
907 | case INDEX_op_remu_i64: | |
908 | tcg_out_opc_mod_du(s, a0, a1, a2); | |
909 | break; | |
910 | ||
9ee775cf WX |
911 | case INDEX_op_setcond_i32: |
912 | case INDEX_op_setcond_i64: | |
913 | tcg_out_setcond(s, args[3], a0, a1, a2, c2); | |
914 | break; | |
915 | ||
dacc5172 WX |
916 | case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ |
917 | case INDEX_op_mov_i64: | |
a26d99d7 | 918 | case INDEX_op_call: /* Always emitted via tcg_out_call. */ |
fae2361d WX |
919 | default: |
920 | g_assert_not_reached(); | |
921 | } | |
922 | } | |
e3b15766 WX |
923 | |
924 | static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op) | |
925 | { | |
926 | switch (op) { | |
927 | case INDEX_op_goto_ptr: | |
928 | return C_O0_I1(r); | |
929 | ||
94505c02 WX |
930 | case INDEX_op_brcond_i32: |
931 | case INDEX_op_brcond_i64: | |
932 | return C_O0_I2(rZ, rZ); | |
933 | ||
6be08fcf WX |
934 | case INDEX_op_ext8s_i32: |
935 | case INDEX_op_ext8s_i64: | |
936 | case INDEX_op_ext8u_i32: | |
937 | case INDEX_op_ext8u_i64: | |
938 | case INDEX_op_ext16s_i32: | |
939 | case INDEX_op_ext16s_i64: | |
940 | case INDEX_op_ext16u_i32: | |
941 | case INDEX_op_ext16u_i64: | |
942 | case INDEX_op_ext32s_i64: | |
943 | case INDEX_op_ext32u_i64: | |
944 | case INDEX_op_extu_i32_i64: | |
945 | case INDEX_op_extrl_i64_i32: | |
946 | case INDEX_op_extrh_i64_i32: | |
947 | case INDEX_op_ext_i32_i64: | |
97b2fafb WX |
948 | case INDEX_op_not_i32: |
949 | case INDEX_op_not_i64: | |
7257809f WX |
950 | case INDEX_op_extract_i32: |
951 | case INDEX_op_extract_i64: | |
4ab2aff0 WX |
952 | case INDEX_op_bswap16_i32: |
953 | case INDEX_op_bswap16_i64: | |
954 | case INDEX_op_bswap32_i32: | |
955 | case INDEX_op_bswap32_i64: | |
956 | case INDEX_op_bswap64_i64: | |
6be08fcf WX |
957 | return C_O1_I1(r, r); |
958 | ||
97b2fafb WX |
959 | case INDEX_op_andc_i32: |
960 | case INDEX_op_andc_i64: | |
961 | case INDEX_op_orc_i32: | |
962 | case INDEX_op_orc_i64: | |
963 | /* | |
964 | * LoongArch insns for these ops don't have reg-imm forms, but we | |
965 | * can express using andi/ori if ~constant satisfies | |
966 | * TCG_CT_CONST_U12. | |
967 | */ | |
968 | return C_O1_I2(r, r, rC); | |
969 | ||
a164010b WX |
970 | case INDEX_op_shl_i32: |
971 | case INDEX_op_shl_i64: | |
972 | case INDEX_op_shr_i32: | |
973 | case INDEX_op_shr_i64: | |
974 | case INDEX_op_sar_i32: | |
975 | case INDEX_op_sar_i64: | |
976 | case INDEX_op_rotl_i32: | |
977 | case INDEX_op_rotl_i64: | |
978 | case INDEX_op_rotr_i32: | |
979 | case INDEX_op_rotr_i64: | |
980 | return C_O1_I2(r, r, ri); | |
981 | ||
39f54ce5 WX |
982 | case INDEX_op_add_i32: |
983 | case INDEX_op_add_i64: | |
984 | return C_O1_I2(r, r, rI); | |
985 | ||
97b2fafb WX |
986 | case INDEX_op_and_i32: |
987 | case INDEX_op_and_i64: | |
988 | case INDEX_op_nor_i32: | |
989 | case INDEX_op_nor_i64: | |
990 | case INDEX_op_or_i32: | |
991 | case INDEX_op_or_i64: | |
992 | case INDEX_op_xor_i32: | |
993 | case INDEX_op_xor_i64: | |
994 | /* LoongArch reg-imm bitops have their imms ZERO-extended */ | |
995 | return C_O1_I2(r, r, rU); | |
996 | ||
fde69301 WX |
997 | case INDEX_op_clz_i32: |
998 | case INDEX_op_clz_i64: | |
999 | case INDEX_op_ctz_i32: | |
1000 | case INDEX_op_ctz_i64: | |
1001 | return C_O1_I2(r, r, rW); | |
1002 | ||
9ee775cf WX |
1003 | case INDEX_op_setcond_i32: |
1004 | case INDEX_op_setcond_i64: | |
1005 | return C_O1_I2(r, r, rZ); | |
1006 | ||
7257809f WX |
1007 | case INDEX_op_deposit_i32: |
1008 | case INDEX_op_deposit_i64: | |
1009 | /* Must deposit into the same register as input */ | |
1010 | return C_O1_I2(r, 0, rZ); | |
1011 | ||
39f54ce5 WX |
1012 | case INDEX_op_sub_i32: |
1013 | case INDEX_op_sub_i64: | |
1014 | return C_O1_I2(r, rZ, rN); | |
1015 | ||
ff13c196 WX |
1016 | case INDEX_op_mul_i32: |
1017 | case INDEX_op_mul_i64: | |
1018 | case INDEX_op_mulsh_i32: | |
1019 | case INDEX_op_mulsh_i64: | |
1020 | case INDEX_op_muluh_i32: | |
1021 | case INDEX_op_muluh_i64: | |
1022 | case INDEX_op_div_i32: | |
1023 | case INDEX_op_div_i64: | |
1024 | case INDEX_op_divu_i32: | |
1025 | case INDEX_op_divu_i64: | |
1026 | case INDEX_op_rem_i32: | |
1027 | case INDEX_op_rem_i64: | |
1028 | case INDEX_op_remu_i32: | |
1029 | case INDEX_op_remu_i64: | |
1030 | return C_O1_I2(r, rZ, rZ); | |
1031 | ||
e3b15766 WX |
1032 | default: |
1033 | g_assert_not_reached(); | |
1034 | } | |
1035 | } |