]>
Commit | Line | Data |
---|---|---|
c896fe29 FB |
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 | ||
757e725b | 25 | #include "qemu/osdep.h" |
cca82982 | 26 | |
813da627 RH |
27 | /* Define to jump the ELF file used to communicate with GDB. */ |
28 | #undef DEBUG_JIT | |
29 | ||
72fd2efb | 30 | #include "qemu/error-report.h" |
f348b6d1 | 31 | #include "qemu/cutils.h" |
1de7afc9 | 32 | #include "qemu/host-utils.h" |
d4c51a0a | 33 | #include "qemu/qemu-print.h" |
084cfca1 | 34 | #include "qemu/cacheflush.h" |
ad768e6f | 35 | #include "qemu/cacheinfo.h" |
533206f0 | 36 | #include "qemu/timer.h" |
cac9b0fd | 37 | #include "exec/translation-block.h" |
d0a9bb5e | 38 | #include "exec/tlb-common.h" |
d7ec12f8 | 39 | #include "tcg/startup.h" |
ad3d0e4d | 40 | #include "tcg/tcg-op-common.h" |
813da627 | 41 | |
edee2579 | 42 | #if UINTPTR_MAX == UINT32_MAX |
813da627 | 43 | # define ELF_CLASS ELFCLASS32 |
edee2579 RH |
44 | #else |
45 | # define ELF_CLASS ELFCLASS64 | |
813da627 | 46 | #endif |
e03b5686 | 47 | #if HOST_BIG_ENDIAN |
813da627 RH |
48 | # define ELF_DATA ELFDATA2MSB |
49 | #else | |
50 | # define ELF_DATA ELFDATA2LSB | |
51 | #endif | |
52 | ||
c896fe29 | 53 | #include "elf.h" |
508127e2 | 54 | #include "exec/log.h" |
d2ba8026 | 55 | #include "tcg/tcg-ldst.h" |
47f7313d | 56 | #include "tcg/tcg-temp-internal.h" |
5ff7258c | 57 | #include "tcg-internal.h" |
327b75a4 | 58 | #include "tcg/perf.h" |
7d478306 RH |
59 | #ifdef CONFIG_USER_ONLY |
60 | #include "exec/user/guest-base.h" | |
61 | #endif | |
c896fe29 | 62 | |
139c1837 | 63 | /* Forward declarations for functions declared in tcg-target.c.inc and |
ce151109 | 64 | used here. */ |
e4d58b41 RH |
65 | static void tcg_target_init(TCGContext *s); |
66 | static void tcg_target_qemu_prologue(TCGContext *s); | |
6ac17786 | 67 | static bool patch_reloc(tcg_insn_unit *code_ptr, int type, |
2ba7fae2 | 68 | intptr_t value, intptr_t addend); |
c896fe29 | 69 | |
497a22eb RH |
70 | /* The CIE and FDE header definitions will be common to all hosts. */ |
71 | typedef struct { | |
72 | uint32_t len __attribute__((aligned((sizeof(void *))))); | |
73 | uint32_t id; | |
74 | uint8_t version; | |
75 | char augmentation[1]; | |
76 | uint8_t code_align; | |
77 | uint8_t data_align; | |
78 | uint8_t return_column; | |
79 | } DebugFrameCIE; | |
80 | ||
81 | typedef struct QEMU_PACKED { | |
82 | uint32_t len __attribute__((aligned((sizeof(void *))))); | |
83 | uint32_t cie_offset; | |
edee2579 RH |
84 | uintptr_t func_start; |
85 | uintptr_t func_len; | |
497a22eb RH |
86 | } DebugFrameFDEHeader; |
87 | ||
2c90784a RH |
88 | typedef struct QEMU_PACKED { |
89 | DebugFrameCIE cie; | |
90 | DebugFrameFDEHeader fde; | |
91 | } DebugFrameHeader; | |
92 | ||
2528f771 RH |
93 | typedef struct TCGLabelQemuLdst { |
94 | bool is_ld; /* qemu_ld: true, qemu_st: false */ | |
95 | MemOpIdx oi; | |
96 | TCGType type; /* result type of a load */ | |
97 | TCGReg addrlo_reg; /* reg index for low word of guest virtual addr */ | |
98 | TCGReg addrhi_reg; /* reg index for high word of guest virtual addr */ | |
99 | TCGReg datalo_reg; /* reg index for low word to be loaded or stored */ | |
100 | TCGReg datahi_reg; /* reg index for high word to be loaded or stored */ | |
101 | const tcg_insn_unit *raddr; /* addr of the next IR of qemu_ld/st IR */ | |
102 | tcg_insn_unit *label_ptr[2]; /* label pointers to be updated */ | |
103 | QSIMPLEQ_ENTRY(TCGLabelQemuLdst) next; | |
104 | } TCGLabelQemuLdst; | |
105 | ||
755bf9e5 | 106 | static void tcg_register_jit_int(const void *buf, size_t size, |
2c90784a RH |
107 | const void *debug_frame, |
108 | size_t debug_frame_size) | |
813da627 RH |
109 | __attribute__((unused)); |
110 | ||
139c1837 | 111 | /* Forward declarations for functions declared and used in tcg-target.c.inc. */ |
9358fbbf | 112 | static void tcg_out_tb_start(TCGContext *s); |
2a534aff | 113 | static void tcg_out_ld(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg1, |
a05b5b9b | 114 | intptr_t arg2); |
78113e83 | 115 | static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg); |
c0ad3001 | 116 | static void tcg_out_movi(TCGContext *s, TCGType type, |
2a534aff | 117 | TCGReg ret, tcg_target_long arg); |
678155b2 | 118 | static void tcg_out_ext8s(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg); |
753e42ea | 119 | static void tcg_out_ext16s(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg); |
d0e66c89 | 120 | static void tcg_out_ext8u(TCGContext *s, TCGReg ret, TCGReg arg); |
379afdff | 121 | static void tcg_out_ext16u(TCGContext *s, TCGReg ret, TCGReg arg); |
52bf3398 | 122 | static void tcg_out_ext32s(TCGContext *s, TCGReg ret, TCGReg arg); |
9ecf5f61 | 123 | static void tcg_out_ext32u(TCGContext *s, TCGReg ret, TCGReg arg); |
9c6aa274 | 124 | static void tcg_out_exts_i32_i64(TCGContext *s, TCGReg ret, TCGReg arg); |
b9bfe000 | 125 | static void tcg_out_extu_i32_i64(TCGContext *s, TCGReg ret, TCGReg arg); |
b8b94ac6 | 126 | static void tcg_out_extrl_i64_i32(TCGContext *s, TCGReg ret, TCGReg arg); |
313bdea8 | 127 | static void tcg_out_addi_ptr(TCGContext *s, TCGReg, TCGReg, tcg_target_long); |
129f1f9e | 128 | static bool tcg_out_xchg(TCGContext *s, TCGType type, TCGReg r1, TCGReg r2); |
b55a8d9d | 129 | static void tcg_out_exit_tb(TCGContext *s, uintptr_t arg); |
cf7d6b8e | 130 | static void tcg_out_goto_tb(TCGContext *s, int which); |
5e8892db MR |
131 | static void tcg_out_op(TCGContext *s, TCGOpcode opc, |
132 | const TCGArg args[TCG_MAX_OP_ARGS], | |
133 | const int const_args[TCG_MAX_OP_ARGS]); | |
d2fd745f | 134 | #if TCG_TARGET_MAYBE_vec |
e7632cfa RH |
135 | static bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece, |
136 | TCGReg dst, TCGReg src); | |
d6ecb4a9 RH |
137 | static bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece, |
138 | TCGReg dst, TCGReg base, intptr_t offset); | |
4e186175 RH |
139 | static void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece, |
140 | TCGReg dst, int64_t arg); | |
5e8892db MR |
141 | static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc, |
142 | unsigned vecl, unsigned vece, | |
143 | const TCGArg args[TCG_MAX_OP_ARGS], | |
144 | const int const_args[TCG_MAX_OP_ARGS]); | |
d2fd745f | 145 | #else |
e7632cfa RH |
146 | static inline bool tcg_out_dup_vec(TCGContext *s, TCGType type, unsigned vece, |
147 | TCGReg dst, TCGReg src) | |
148 | { | |
149 | g_assert_not_reached(); | |
150 | } | |
d6ecb4a9 RH |
151 | static inline bool tcg_out_dupm_vec(TCGContext *s, TCGType type, unsigned vece, |
152 | TCGReg dst, TCGReg base, intptr_t offset) | |
153 | { | |
154 | g_assert_not_reached(); | |
155 | } | |
4e186175 RH |
156 | static inline void tcg_out_dupi_vec(TCGContext *s, TCGType type, unsigned vece, |
157 | TCGReg dst, int64_t arg) | |
e7632cfa RH |
158 | { |
159 | g_assert_not_reached(); | |
160 | } | |
5e8892db MR |
161 | static inline void tcg_out_vec_op(TCGContext *s, TCGOpcode opc, |
162 | unsigned vecl, unsigned vece, | |
163 | const TCGArg args[TCG_MAX_OP_ARGS], | |
164 | const int const_args[TCG_MAX_OP_ARGS]) | |
d2fd745f RH |
165 | { |
166 | g_assert_not_reached(); | |
167 | } | |
168 | #endif | |
2a534aff | 169 | static void tcg_out_st(TCGContext *s, TCGType type, TCGReg arg, TCGReg arg1, |
a05b5b9b | 170 | intptr_t arg2); |
59d7c14e RH |
171 | static bool tcg_out_sti(TCGContext *s, TCGType type, TCGArg val, |
172 | TCGReg base, intptr_t ofs); | |
7b7d8b2d | 173 | static void tcg_out_call(TCGContext *s, const tcg_insn_unit *target, |
cee44b03 | 174 | const TCGHelperInfo *info); |
5e3d0c19 | 175 | static TCGReg tcg_target_call_oarg_reg(TCGCallReturnKind kind, int slot); |
ebe92db2 | 176 | static bool tcg_target_const_match(int64_t val, TCGType type, int ct, int vece); |
659ef5cb | 177 | #ifdef TCG_TARGET_NEED_LDST_LABELS |
aeee05f5 | 178 | static int tcg_out_ldst_finalize(TCGContext *s); |
659ef5cb | 179 | #endif |
c896fe29 | 180 | |
23088ca0 RH |
181 | #ifndef CONFIG_USER_ONLY |
182 | #define guest_base ({ qemu_build_not_reached(); (uintptr_t)0; }) | |
183 | #endif | |
184 | ||
8429a1ca RH |
185 | typedef struct TCGLdstHelperParam { |
186 | TCGReg (*ra_gen)(TCGContext *s, const TCGLabelQemuLdst *l, int arg_reg); | |
187 | unsigned ntmp; | |
188 | int tmp[3]; | |
189 | } TCGLdstHelperParam; | |
190 | ||
191 | static void tcg_out_ld_helper_args(TCGContext *s, const TCGLabelQemuLdst *l, | |
192 | const TCGLdstHelperParam *p) | |
193 | __attribute__((unused)); | |
194 | static void tcg_out_ld_helper_ret(TCGContext *s, const TCGLabelQemuLdst *l, | |
195 | bool load_sign, const TCGLdstHelperParam *p) | |
196 | __attribute__((unused)); | |
197 | static void tcg_out_st_helper_args(TCGContext *s, const TCGLabelQemuLdst *l, | |
198 | const TCGLdstHelperParam *p) | |
199 | __attribute__((unused)); | |
200 | ||
de95016d | 201 | static void * const qemu_ld_helpers[MO_SSIZE + 1] __attribute__((unused)) = { |
0cadc1ed RH |
202 | [MO_UB] = helper_ldub_mmu, |
203 | [MO_SB] = helper_ldsb_mmu, | |
204 | [MO_UW] = helper_lduw_mmu, | |
205 | [MO_SW] = helper_ldsw_mmu, | |
206 | [MO_UL] = helper_ldul_mmu, | |
207 | [MO_UQ] = helper_ldq_mmu, | |
208 | #if TCG_TARGET_REG_BITS == 64 | |
209 | [MO_SL] = helper_ldsl_mmu, | |
ebebea53 | 210 | [MO_128] = helper_ld16_mmu, |
0cadc1ed RH |
211 | #endif |
212 | }; | |
213 | ||
de95016d | 214 | static void * const qemu_st_helpers[MO_SIZE + 1] __attribute__((unused)) = { |
0cadc1ed RH |
215 | [MO_8] = helper_stb_mmu, |
216 | [MO_16] = helper_stw_mmu, | |
217 | [MO_32] = helper_stl_mmu, | |
218 | [MO_64] = helper_stq_mmu, | |
ebebea53 RH |
219 | #if TCG_TARGET_REG_BITS == 64 |
220 | [MO_128] = helper_st16_mmu, | |
221 | #endif | |
0cadc1ed | 222 | }; |
0cadc1ed | 223 | |
e63b8a29 RH |
224 | typedef struct { |
225 | MemOp atom; /* lg2 bits of atomicity required */ | |
226 | MemOp align; /* lg2 bits of alignment to use */ | |
227 | } TCGAtomAlign; | |
228 | ||
229 | static TCGAtomAlign atom_and_align_for_opc(TCGContext *s, MemOp opc, | |
230 | MemOp host_atom, bool allow_two_ops) | |
231 | __attribute__((unused)); | |
232 | ||
397cabaa RH |
233 | #ifdef CONFIG_USER_ONLY |
234 | bool tcg_use_softmmu; | |
235 | #endif | |
236 | ||
42eb6dfc RH |
237 | TCGContext tcg_init_ctx; |
238 | __thread TCGContext *tcg_ctx; | |
239 | ||
5ff7258c | 240 | TCGContext **tcg_ctxs; |
0e2d61cf RH |
241 | unsigned int tcg_cur_ctxs; |
242 | unsigned int tcg_max_ctxs; | |
ad75a51e | 243 | TCGv_env tcg_env; |
c8bc1168 | 244 | const void *tcg_code_gen_epilogue; |
db0c51a3 | 245 | uintptr_t tcg_splitwx_diff; |
df2cce29 | 246 | |
b91ccb31 RH |
247 | #ifndef CONFIG_TCG_INTERPRETER |
248 | tcg_prologue_fn *tcg_qemu_tb_exec; | |
249 | #endif | |
250 | ||
d2fd745f | 251 | static TCGRegSet tcg_target_available_regs[TCG_TYPE_COUNT]; |
b1d8e52e | 252 | static TCGRegSet tcg_target_call_clobber_regs; |
c896fe29 | 253 | |
1813e175 | 254 | #if TCG_TARGET_INSN_UNIT_SIZE == 1 |
4196dca6 | 255 | static __attribute__((unused)) inline void tcg_out8(TCGContext *s, uint8_t v) |
c896fe29 FB |
256 | { |
257 | *s->code_ptr++ = v; | |
258 | } | |
259 | ||
4196dca6 PM |
260 | static __attribute__((unused)) inline void tcg_patch8(tcg_insn_unit *p, |
261 | uint8_t v) | |
5c53bb81 | 262 | { |
1813e175 | 263 | *p = v; |
5c53bb81 | 264 | } |
1813e175 | 265 | #endif |
5c53bb81 | 266 | |
1813e175 | 267 | #if TCG_TARGET_INSN_UNIT_SIZE <= 2 |
4196dca6 | 268 | static __attribute__((unused)) inline void tcg_out16(TCGContext *s, uint16_t v) |
c896fe29 | 269 | { |
1813e175 RH |
270 | if (TCG_TARGET_INSN_UNIT_SIZE == 2) { |
271 | *s->code_ptr++ = v; | |
272 | } else { | |
273 | tcg_insn_unit *p = s->code_ptr; | |
274 | memcpy(p, &v, sizeof(v)); | |
275 | s->code_ptr = p + (2 / TCG_TARGET_INSN_UNIT_SIZE); | |
276 | } | |
c896fe29 FB |
277 | } |
278 | ||
4196dca6 PM |
279 | static __attribute__((unused)) inline void tcg_patch16(tcg_insn_unit *p, |
280 | uint16_t v) | |
5c53bb81 | 281 | { |
1813e175 RH |
282 | if (TCG_TARGET_INSN_UNIT_SIZE == 2) { |
283 | *p = v; | |
284 | } else { | |
285 | memcpy(p, &v, sizeof(v)); | |
286 | } | |
5c53bb81 | 287 | } |
1813e175 | 288 | #endif |
5c53bb81 | 289 | |
1813e175 | 290 | #if TCG_TARGET_INSN_UNIT_SIZE <= 4 |
4196dca6 | 291 | static __attribute__((unused)) inline void tcg_out32(TCGContext *s, uint32_t v) |
c896fe29 | 292 | { |
1813e175 RH |
293 | if (TCG_TARGET_INSN_UNIT_SIZE == 4) { |
294 | *s->code_ptr++ = v; | |
295 | } else { | |
296 | tcg_insn_unit *p = s->code_ptr; | |
297 | memcpy(p, &v, sizeof(v)); | |
298 | s->code_ptr = p + (4 / TCG_TARGET_INSN_UNIT_SIZE); | |
299 | } | |
c896fe29 FB |
300 | } |
301 | ||
4196dca6 PM |
302 | static __attribute__((unused)) inline void tcg_patch32(tcg_insn_unit *p, |
303 | uint32_t v) | |
5c53bb81 | 304 | { |
1813e175 RH |
305 | if (TCG_TARGET_INSN_UNIT_SIZE == 4) { |
306 | *p = v; | |
307 | } else { | |
308 | memcpy(p, &v, sizeof(v)); | |
309 | } | |
5c53bb81 | 310 | } |
1813e175 | 311 | #endif |
5c53bb81 | 312 | |
1813e175 | 313 | #if TCG_TARGET_INSN_UNIT_SIZE <= 8 |
4196dca6 | 314 | static __attribute__((unused)) inline void tcg_out64(TCGContext *s, uint64_t v) |
ac26eb69 | 315 | { |
1813e175 RH |
316 | if (TCG_TARGET_INSN_UNIT_SIZE == 8) { |
317 | *s->code_ptr++ = v; | |
318 | } else { | |
319 | tcg_insn_unit *p = s->code_ptr; | |
320 | memcpy(p, &v, sizeof(v)); | |
321 | s->code_ptr = p + (8 / TCG_TARGET_INSN_UNIT_SIZE); | |
322 | } | |
ac26eb69 RH |
323 | } |
324 | ||
4196dca6 PM |
325 | static __attribute__((unused)) inline void tcg_patch64(tcg_insn_unit *p, |
326 | uint64_t v) | |
5c53bb81 | 327 | { |
1813e175 RH |
328 | if (TCG_TARGET_INSN_UNIT_SIZE == 8) { |
329 | *p = v; | |
330 | } else { | |
331 | memcpy(p, &v, sizeof(v)); | |
332 | } | |
5c53bb81 | 333 | } |
1813e175 | 334 | #endif |
5c53bb81 | 335 | |
c896fe29 FB |
336 | /* label relocation processing */ |
337 | ||
1813e175 | 338 | static void tcg_out_reloc(TCGContext *s, tcg_insn_unit *code_ptr, int type, |
bec16311 | 339 | TCGLabel *l, intptr_t addend) |
c896fe29 | 340 | { |
7ecd02a0 | 341 | TCGRelocation *r = tcg_malloc(sizeof(TCGRelocation)); |
c896fe29 | 342 | |
7ecd02a0 RH |
343 | r->type = type; |
344 | r->ptr = code_ptr; | |
345 | r->addend = addend; | |
346 | QSIMPLEQ_INSERT_TAIL(&l->relocs, r, next); | |
c896fe29 FB |
347 | } |
348 | ||
92ab8e7d | 349 | static void tcg_out_label(TCGContext *s, TCGLabel *l) |
c896fe29 | 350 | { |
eabb7b91 | 351 | tcg_debug_assert(!l->has_value); |
c896fe29 | 352 | l->has_value = 1; |
92ab8e7d | 353 | l->u.value_ptr = tcg_splitwx_to_rx(s->code_ptr); |
c896fe29 FB |
354 | } |
355 | ||
42a268c2 | 356 | TCGLabel *gen_new_label(void) |
c896fe29 | 357 | { |
b1311c4a | 358 | TCGContext *s = tcg_ctx; |
51e3972c | 359 | TCGLabel *l = tcg_malloc(sizeof(TCGLabel)); |
c896fe29 | 360 | |
7ecd02a0 RH |
361 | memset(l, 0, sizeof(TCGLabel)); |
362 | l->id = s->nb_labels++; | |
f85b1fc4 | 363 | QSIMPLEQ_INIT(&l->branches); |
7ecd02a0 RH |
364 | QSIMPLEQ_INIT(&l->relocs); |
365 | ||
bef16ab4 | 366 | QSIMPLEQ_INSERT_TAIL(&s->labels, l, next); |
42a268c2 RH |
367 | |
368 | return l; | |
c896fe29 FB |
369 | } |
370 | ||
7ecd02a0 RH |
371 | static bool tcg_resolve_relocs(TCGContext *s) |
372 | { | |
373 | TCGLabel *l; | |
374 | ||
375 | QSIMPLEQ_FOREACH(l, &s->labels, next) { | |
376 | TCGRelocation *r; | |
377 | uintptr_t value = l->u.value; | |
378 | ||
379 | QSIMPLEQ_FOREACH(r, &l->relocs, next) { | |
380 | if (!patch_reloc(r->ptr, r->type, value, r->addend)) { | |
381 | return false; | |
382 | } | |
383 | } | |
384 | } | |
385 | return true; | |
386 | } | |
387 | ||
9f754620 RH |
388 | static void set_jmp_reset_offset(TCGContext *s, int which) |
389 | { | |
f14bed3f RH |
390 | /* |
391 | * We will check for overflow at the end of the opcode loop in | |
392 | * tcg_gen_code, where we bound tcg_current_code_size to UINT16_MAX. | |
393 | */ | |
b7e4afbd | 394 | s->gen_tb->jmp_reset_offset[which] = tcg_current_code_size(s); |
9f754620 RH |
395 | } |
396 | ||
b52a2c03 RH |
397 | static void G_GNUC_UNUSED set_jmp_insn_offset(TCGContext *s, int which) |
398 | { | |
399 | /* | |
400 | * We will check for overflow at the end of the opcode loop in | |
401 | * tcg_gen_code, where we bound tcg_current_code_size to UINT16_MAX. | |
402 | */ | |
9da6079b | 403 | s->gen_tb->jmp_insn_offset[which] = tcg_current_code_size(s); |
b52a2c03 RH |
404 | } |
405 | ||
becc452a RH |
406 | static uintptr_t G_GNUC_UNUSED get_jmp_target_addr(TCGContext *s, int which) |
407 | { | |
408 | /* | |
409 | * Return the read-execute version of the pointer, for the benefit | |
410 | * of any pc-relative addressing mode. | |
411 | */ | |
9da6079b | 412 | return (uintptr_t)tcg_splitwx_to_rx(&s->gen_tb->jmp_target_addr[which]); |
becc452a RH |
413 | } |
414 | ||
397cabaa RH |
415 | static int __attribute__((unused)) |
416 | tlb_mask_table_ofs(TCGContext *s, int which) | |
d0a9bb5e | 417 | { |
7857ee11 RH |
418 | return (offsetof(CPUNegativeOffsetState, tlb.f[which]) - |
419 | sizeof(CPUNegativeOffsetState)); | |
d0a9bb5e | 420 | } |
d0a9bb5e | 421 | |
db6b7d0c | 422 | /* Signal overflow, starting over with fewer guest insns. */ |
8905770b MAL |
423 | static G_NORETURN |
424 | void tcg_raise_tb_overflow(TCGContext *s) | |
db6b7d0c RH |
425 | { |
426 | siglongjmp(s->jmp_trans, -2); | |
427 | } | |
428 | ||
8429a1ca RH |
429 | /* |
430 | * Used by tcg_out_movext{1,2} to hold the arguments for tcg_out_movext. | |
431 | * By the time we arrive at tcg_out_movext1, @dst is always a TCGReg. | |
432 | * | |
433 | * However, tcg_out_helper_load_slots reuses this field to hold an | |
434 | * argument slot number (which may designate a argument register or an | |
435 | * argument stack slot), converting to TCGReg once all arguments that | |
436 | * are destined for the stack are processed. | |
437 | */ | |
129f1f9e | 438 | typedef struct TCGMovExtend { |
8429a1ca | 439 | unsigned dst; |
129f1f9e RH |
440 | TCGReg src; |
441 | TCGType dst_type; | |
442 | TCGType src_type; | |
443 | MemOp src_ext; | |
444 | } TCGMovExtend; | |
445 | ||
b3dfd5fc RH |
446 | /** |
447 | * tcg_out_movext -- move and extend | |
448 | * @s: tcg context | |
449 | * @dst_type: integral type for destination | |
450 | * @dst: destination register | |
451 | * @src_type: integral type for source | |
452 | * @src_ext: extension to apply to source | |
453 | * @src: source register | |
454 | * | |
455 | * Move or extend @src into @dst, depending on @src_ext and the types. | |
456 | */ | |
129f1f9e RH |
457 | static void tcg_out_movext(TCGContext *s, TCGType dst_type, TCGReg dst, |
458 | TCGType src_type, MemOp src_ext, TCGReg src) | |
b3dfd5fc RH |
459 | { |
460 | switch (src_ext) { | |
461 | case MO_UB: | |
462 | tcg_out_ext8u(s, dst, src); | |
463 | break; | |
464 | case MO_SB: | |
465 | tcg_out_ext8s(s, dst_type, dst, src); | |
466 | break; | |
467 | case MO_UW: | |
468 | tcg_out_ext16u(s, dst, src); | |
469 | break; | |
470 | case MO_SW: | |
471 | tcg_out_ext16s(s, dst_type, dst, src); | |
472 | break; | |
473 | case MO_UL: | |
474 | case MO_SL: | |
475 | if (dst_type == TCG_TYPE_I32) { | |
476 | if (src_type == TCG_TYPE_I32) { | |
477 | tcg_out_mov(s, TCG_TYPE_I32, dst, src); | |
478 | } else { | |
479 | tcg_out_extrl_i64_i32(s, dst, src); | |
480 | } | |
481 | } else if (src_type == TCG_TYPE_I32) { | |
482 | if (src_ext & MO_SIGN) { | |
483 | tcg_out_exts_i32_i64(s, dst, src); | |
484 | } else { | |
485 | tcg_out_extu_i32_i64(s, dst, src); | |
486 | } | |
487 | } else { | |
488 | if (src_ext & MO_SIGN) { | |
489 | tcg_out_ext32s(s, dst, src); | |
490 | } else { | |
491 | tcg_out_ext32u(s, dst, src); | |
492 | } | |
493 | } | |
494 | break; | |
495 | case MO_UQ: | |
496 | tcg_debug_assert(TCG_TARGET_REG_BITS == 64); | |
497 | if (dst_type == TCG_TYPE_I32) { | |
498 | tcg_out_extrl_i64_i32(s, dst, src); | |
499 | } else { | |
500 | tcg_out_mov(s, TCG_TYPE_I64, dst, src); | |
501 | } | |
502 | break; | |
503 | default: | |
504 | g_assert_not_reached(); | |
505 | } | |
506 | } | |
507 | ||
129f1f9e RH |
508 | /* Minor variations on a theme, using a structure. */ |
509 | static void tcg_out_movext1_new_src(TCGContext *s, const TCGMovExtend *i, | |
510 | TCGReg src) | |
511 | { | |
512 | tcg_out_movext(s, i->dst_type, i->dst, i->src_type, i->src_ext, src); | |
513 | } | |
514 | ||
515 | static void tcg_out_movext1(TCGContext *s, const TCGMovExtend *i) | |
516 | { | |
517 | tcg_out_movext1_new_src(s, i, i->src); | |
518 | } | |
519 | ||
520 | /** | |
521 | * tcg_out_movext2 -- move and extend two pair | |
522 | * @s: tcg context | |
523 | * @i1: first move description | |
524 | * @i2: second move description | |
525 | * @scratch: temporary register, or -1 for none | |
526 | * | |
527 | * As tcg_out_movext, for both @i1 and @i2, caring for overlap | |
528 | * between the sources and destinations. | |
529 | */ | |
530 | ||
8429a1ca RH |
531 | static void tcg_out_movext2(TCGContext *s, const TCGMovExtend *i1, |
532 | const TCGMovExtend *i2, int scratch) | |
129f1f9e RH |
533 | { |
534 | TCGReg src1 = i1->src; | |
535 | TCGReg src2 = i2->src; | |
536 | ||
537 | if (i1->dst != src2) { | |
538 | tcg_out_movext1(s, i1); | |
539 | tcg_out_movext1(s, i2); | |
540 | return; | |
541 | } | |
542 | if (i2->dst == src1) { | |
543 | TCGType src1_type = i1->src_type; | |
544 | TCGType src2_type = i2->src_type; | |
545 | ||
546 | if (tcg_out_xchg(s, MAX(src1_type, src2_type), src1, src2)) { | |
547 | /* The data is now in the correct registers, now extend. */ | |
548 | src1 = i2->src; | |
549 | src2 = i1->src; | |
550 | } else { | |
551 | tcg_debug_assert(scratch >= 0); | |
552 | tcg_out_mov(s, src1_type, scratch, src1); | |
553 | src1 = scratch; | |
554 | } | |
555 | } | |
556 | tcg_out_movext1_new_src(s, i2, src2); | |
557 | tcg_out_movext1_new_src(s, i1, src1); | |
558 | } | |
559 | ||
2462e30e RH |
560 | /** |
561 | * tcg_out_movext3 -- move and extend three pair | |
562 | * @s: tcg context | |
563 | * @i1: first move description | |
564 | * @i2: second move description | |
565 | * @i3: third move description | |
566 | * @scratch: temporary register, or -1 for none | |
567 | * | |
568 | * As tcg_out_movext, for all of @i1, @i2 and @i3, caring for overlap | |
569 | * between the sources and destinations. | |
570 | */ | |
571 | ||
572 | static void tcg_out_movext3(TCGContext *s, const TCGMovExtend *i1, | |
573 | const TCGMovExtend *i2, const TCGMovExtend *i3, | |
574 | int scratch) | |
575 | { | |
576 | TCGReg src1 = i1->src; | |
577 | TCGReg src2 = i2->src; | |
578 | TCGReg src3 = i3->src; | |
579 | ||
580 | if (i1->dst != src2 && i1->dst != src3) { | |
581 | tcg_out_movext1(s, i1); | |
582 | tcg_out_movext2(s, i2, i3, scratch); | |
583 | return; | |
584 | } | |
585 | if (i2->dst != src1 && i2->dst != src3) { | |
586 | tcg_out_movext1(s, i2); | |
587 | tcg_out_movext2(s, i1, i3, scratch); | |
588 | return; | |
589 | } | |
590 | if (i3->dst != src1 && i3->dst != src2) { | |
591 | tcg_out_movext1(s, i3); | |
592 | tcg_out_movext2(s, i1, i2, scratch); | |
593 | return; | |
594 | } | |
595 | ||
596 | /* | |
597 | * There is a cycle. Since there are only 3 nodes, the cycle is | |
598 | * either "clockwise" or "anti-clockwise", and can be solved with | |
599 | * a single scratch or two xchg. | |
600 | */ | |
601 | if (i1->dst == src2 && i2->dst == src3 && i3->dst == src1) { | |
602 | /* "Clockwise" */ | |
603 | if (tcg_out_xchg(s, MAX(i1->src_type, i2->src_type), src1, src2)) { | |
604 | tcg_out_xchg(s, MAX(i2->src_type, i3->src_type), src2, src3); | |
605 | /* The data is now in the correct registers, now extend. */ | |
606 | tcg_out_movext1_new_src(s, i1, i1->dst); | |
607 | tcg_out_movext1_new_src(s, i2, i2->dst); | |
608 | tcg_out_movext1_new_src(s, i3, i3->dst); | |
609 | } else { | |
610 | tcg_debug_assert(scratch >= 0); | |
611 | tcg_out_mov(s, i1->src_type, scratch, src1); | |
612 | tcg_out_movext1(s, i3); | |
613 | tcg_out_movext1(s, i2); | |
614 | tcg_out_movext1_new_src(s, i1, scratch); | |
615 | } | |
616 | } else if (i1->dst == src3 && i2->dst == src1 && i3->dst == src2) { | |
617 | /* "Anti-clockwise" */ | |
618 | if (tcg_out_xchg(s, MAX(i2->src_type, i3->src_type), src2, src3)) { | |
619 | tcg_out_xchg(s, MAX(i1->src_type, i2->src_type), src1, src2); | |
620 | /* The data is now in the correct registers, now extend. */ | |
621 | tcg_out_movext1_new_src(s, i1, i1->dst); | |
622 | tcg_out_movext1_new_src(s, i2, i2->dst); | |
623 | tcg_out_movext1_new_src(s, i3, i3->dst); | |
624 | } else { | |
625 | tcg_debug_assert(scratch >= 0); | |
626 | tcg_out_mov(s, i1->src_type, scratch, src1); | |
627 | tcg_out_movext1(s, i2); | |
628 | tcg_out_movext1(s, i3); | |
629 | tcg_out_movext1_new_src(s, i1, scratch); | |
630 | } | |
631 | } else { | |
632 | g_assert_not_reached(); | |
633 | } | |
634 | } | |
635 | ||
4c22e840 RH |
636 | #define C_PFX1(P, A) P##A |
637 | #define C_PFX2(P, A, B) P##A##_##B | |
638 | #define C_PFX3(P, A, B, C) P##A##_##B##_##C | |
639 | #define C_PFX4(P, A, B, C, D) P##A##_##B##_##C##_##D | |
640 | #define C_PFX5(P, A, B, C, D, E) P##A##_##B##_##C##_##D##_##E | |
641 | #define C_PFX6(P, A, B, C, D, E, F) P##A##_##B##_##C##_##D##_##E##_##F | |
642 | ||
643 | /* Define an enumeration for the various combinations. */ | |
644 | ||
645 | #define C_O0_I1(I1) C_PFX1(c_o0_i1_, I1), | |
646 | #define C_O0_I2(I1, I2) C_PFX2(c_o0_i2_, I1, I2), | |
647 | #define C_O0_I3(I1, I2, I3) C_PFX3(c_o0_i3_, I1, I2, I3), | |
648 | #define C_O0_I4(I1, I2, I3, I4) C_PFX4(c_o0_i4_, I1, I2, I3, I4), | |
649 | ||
650 | #define C_O1_I1(O1, I1) C_PFX2(c_o1_i1_, O1, I1), | |
651 | #define C_O1_I2(O1, I1, I2) C_PFX3(c_o1_i2_, O1, I1, I2), | |
652 | #define C_O1_I3(O1, I1, I2, I3) C_PFX4(c_o1_i3_, O1, I1, I2, I3), | |
653 | #define C_O1_I4(O1, I1, I2, I3, I4) C_PFX5(c_o1_i4_, O1, I1, I2, I3, I4), | |
654 | ||
655 | #define C_N1_I2(O1, I1, I2) C_PFX3(c_n1_i2_, O1, I1, I2), | |
ca5bed07 | 656 | #define C_N1O1_I1(O1, O2, I1) C_PFX3(c_n1o1_i1_, O1, O2, I1), |
fa645b48 | 657 | #define C_N2_I1(O1, O2, I1) C_PFX3(c_n2_i1_, O1, O2, I1), |
4c22e840 RH |
658 | |
659 | #define C_O2_I1(O1, O2, I1) C_PFX3(c_o2_i1_, O1, O2, I1), | |
660 | #define C_O2_I2(O1, O2, I1, I2) C_PFX4(c_o2_i2_, O1, O2, I1, I2), | |
661 | #define C_O2_I3(O1, O2, I1, I2, I3) C_PFX5(c_o2_i3_, O1, O2, I1, I2, I3), | |
662 | #define C_O2_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_o2_i4_, O1, O2, I1, I2, I3, I4), | |
22d2e535 | 663 | #define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_n1_o1_i4_, O1, O2, I1, I2, I3, I4), |
4c22e840 RH |
664 | |
665 | typedef enum { | |
666 | #include "tcg-target-con-set.h" | |
667 | } TCGConstraintSetIndex; | |
668 | ||
669 | static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode); | |
670 | ||
671 | #undef C_O0_I1 | |
672 | #undef C_O0_I2 | |
673 | #undef C_O0_I3 | |
674 | #undef C_O0_I4 | |
675 | #undef C_O1_I1 | |
676 | #undef C_O1_I2 | |
677 | #undef C_O1_I3 | |
678 | #undef C_O1_I4 | |
679 | #undef C_N1_I2 | |
ca5bed07 | 680 | #undef C_N1O1_I1 |
fa645b48 | 681 | #undef C_N2_I1 |
4c22e840 RH |
682 | #undef C_O2_I1 |
683 | #undef C_O2_I2 | |
684 | #undef C_O2_I3 | |
685 | #undef C_O2_I4 | |
22d2e535 | 686 | #undef C_N1_O1_I4 |
4c22e840 RH |
687 | |
688 | /* Put all of the constraint sets into an array, indexed by the enum. */ | |
689 | ||
690 | #define C_O0_I1(I1) { .args_ct_str = { #I1 } }, | |
691 | #define C_O0_I2(I1, I2) { .args_ct_str = { #I1, #I2 } }, | |
692 | #define C_O0_I3(I1, I2, I3) { .args_ct_str = { #I1, #I2, #I3 } }, | |
693 | #define C_O0_I4(I1, I2, I3, I4) { .args_ct_str = { #I1, #I2, #I3, #I4 } }, | |
694 | ||
695 | #define C_O1_I1(O1, I1) { .args_ct_str = { #O1, #I1 } }, | |
696 | #define C_O1_I2(O1, I1, I2) { .args_ct_str = { #O1, #I1, #I2 } }, | |
697 | #define C_O1_I3(O1, I1, I2, I3) { .args_ct_str = { #O1, #I1, #I2, #I3 } }, | |
698 | #define C_O1_I4(O1, I1, I2, I3, I4) { .args_ct_str = { #O1, #I1, #I2, #I3, #I4 } }, | |
699 | ||
700 | #define C_N1_I2(O1, I1, I2) { .args_ct_str = { "&" #O1, #I1, #I2 } }, | |
ca5bed07 | 701 | #define C_N1O1_I1(O1, O2, I1) { .args_ct_str = { "&" #O1, #O2, #I1 } }, |
fa645b48 | 702 | #define C_N2_I1(O1, O2, I1) { .args_ct_str = { "&" #O1, "&" #O2, #I1 } }, |
4c22e840 RH |
703 | |
704 | #define C_O2_I1(O1, O2, I1) { .args_ct_str = { #O1, #O2, #I1 } }, | |
705 | #define C_O2_I2(O1, O2, I1, I2) { .args_ct_str = { #O1, #O2, #I1, #I2 } }, | |
706 | #define C_O2_I3(O1, O2, I1, I2, I3) { .args_ct_str = { #O1, #O2, #I1, #I2, #I3 } }, | |
707 | #define C_O2_I4(O1, O2, I1, I2, I3, I4) { .args_ct_str = { #O1, #O2, #I1, #I2, #I3, #I4 } }, | |
22d2e535 | 708 | #define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) { .args_ct_str = { "&" #O1, #O2, #I1, #I2, #I3, #I4 } }, |
4c22e840 RH |
709 | |
710 | static const TCGTargetOpDef constraint_sets[] = { | |
711 | #include "tcg-target-con-set.h" | |
712 | }; | |
713 | ||
714 | ||
715 | #undef C_O0_I1 | |
716 | #undef C_O0_I2 | |
717 | #undef C_O0_I3 | |
718 | #undef C_O0_I4 | |
719 | #undef C_O1_I1 | |
720 | #undef C_O1_I2 | |
721 | #undef C_O1_I3 | |
722 | #undef C_O1_I4 | |
723 | #undef C_N1_I2 | |
ca5bed07 | 724 | #undef C_N1O1_I1 |
fa645b48 | 725 | #undef C_N2_I1 |
4c22e840 RH |
726 | #undef C_O2_I1 |
727 | #undef C_O2_I2 | |
728 | #undef C_O2_I3 | |
729 | #undef C_O2_I4 | |
22d2e535 | 730 | #undef C_N1_O1_I4 |
4c22e840 RH |
731 | |
732 | /* Expand the enumerator to be returned from tcg_target_op_def(). */ | |
733 | ||
734 | #define C_O0_I1(I1) C_PFX1(c_o0_i1_, I1) | |
735 | #define C_O0_I2(I1, I2) C_PFX2(c_o0_i2_, I1, I2) | |
736 | #define C_O0_I3(I1, I2, I3) C_PFX3(c_o0_i3_, I1, I2, I3) | |
737 | #define C_O0_I4(I1, I2, I3, I4) C_PFX4(c_o0_i4_, I1, I2, I3, I4) | |
738 | ||
739 | #define C_O1_I1(O1, I1) C_PFX2(c_o1_i1_, O1, I1) | |
740 | #define C_O1_I2(O1, I1, I2) C_PFX3(c_o1_i2_, O1, I1, I2) | |
741 | #define C_O1_I3(O1, I1, I2, I3) C_PFX4(c_o1_i3_, O1, I1, I2, I3) | |
742 | #define C_O1_I4(O1, I1, I2, I3, I4) C_PFX5(c_o1_i4_, O1, I1, I2, I3, I4) | |
743 | ||
744 | #define C_N1_I2(O1, I1, I2) C_PFX3(c_n1_i2_, O1, I1, I2) | |
ca5bed07 | 745 | #define C_N1O1_I1(O1, O2, I1) C_PFX3(c_n1o1_i1_, O1, O2, I1) |
fa645b48 | 746 | #define C_N2_I1(O1, O2, I1) C_PFX3(c_n2_i1_, O1, O2, I1) |
4c22e840 RH |
747 | |
748 | #define C_O2_I1(O1, O2, I1) C_PFX3(c_o2_i1_, O1, O2, I1) | |
749 | #define C_O2_I2(O1, O2, I1, I2) C_PFX4(c_o2_i2_, O1, O2, I1, I2) | |
750 | #define C_O2_I3(O1, O2, I1, I2, I3) C_PFX5(c_o2_i3_, O1, O2, I1, I2, I3) | |
751 | #define C_O2_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_o2_i4_, O1, O2, I1, I2, I3, I4) | |
22d2e535 | 752 | #define C_N1_O1_I4(O1, O2, I1, I2, I3, I4) C_PFX6(c_n1_o1_i4_, O1, O2, I1, I2, I3, I4) |
4c22e840 | 753 | |
139c1837 | 754 | #include "tcg-target.c.inc" |
c896fe29 | 755 | |
7857ee11 RH |
756 | #ifndef CONFIG_TCG_INTERPRETER |
757 | /* Validate CPUTLBDescFast placement. */ | |
758 | QEMU_BUILD_BUG_ON((int)(offsetof(CPUNegativeOffsetState, tlb.f[0]) - | |
759 | sizeof(CPUNegativeOffsetState)) | |
760 | < MIN_TLB_MASK_TABLE_OFS); | |
761 | #endif | |
762 | ||
38b47b19 EC |
763 | static void alloc_tcg_plugin_context(TCGContext *s) |
764 | { | |
765 | #ifdef CONFIG_PLUGIN | |
766 | s->plugin_tb = g_new0(struct qemu_plugin_tb, 1); | |
767 | s->plugin_tb->insns = | |
768 | g_ptr_array_new_with_free_func(qemu_plugin_insn_cleanup_fn); | |
769 | #endif | |
770 | } | |
771 | ||
3468b59e EC |
772 | /* |
773 | * All TCG threads except the parent (i.e. the one that called tcg_context_init | |
774 | * and registered the target's TCG globals) must register with this function | |
775 | * before initiating translation. | |
776 | * | |
777 | * In user-mode we just point tcg_ctx to tcg_init_ctx. See the documentation | |
778 | * of tcg_region_init() for the reasoning behind this. | |
779 | * | |
7893e42d PMD |
780 | * In system-mode each caller registers its context in tcg_ctxs[]. Note that in |
781 | * system-mode tcg_ctxs[] does not track tcg_ctx_init, since the initial context | |
3468b59e EC |
782 | * is not used anymore for translation once this function is called. |
783 | * | |
7893e42d PMD |
784 | * Not tracking tcg_init_ctx in tcg_ctxs[] in system-mode keeps code that |
785 | * iterates over the array (e.g. tcg_code_size() the same for both system/user | |
786 | * modes. | |
3468b59e EC |
787 | */ |
788 | #ifdef CONFIG_USER_ONLY | |
789 | void tcg_register_thread(void) | |
790 | { | |
791 | tcg_ctx = &tcg_init_ctx; | |
792 | } | |
793 | #else | |
794 | void tcg_register_thread(void) | |
795 | { | |
796 | TCGContext *s = g_malloc(sizeof(*s)); | |
797 | unsigned int i, n; | |
3468b59e EC |
798 | |
799 | *s = tcg_init_ctx; | |
800 | ||
801 | /* Relink mem_base. */ | |
802 | for (i = 0, n = tcg_init_ctx.nb_globals; i < n; ++i) { | |
803 | if (tcg_init_ctx.temps[i].mem_base) { | |
804 | ptrdiff_t b = tcg_init_ctx.temps[i].mem_base - tcg_init_ctx.temps; | |
805 | tcg_debug_assert(b >= 0 && b < n); | |
806 | s->temps[i].mem_base = &s->temps[b]; | |
807 | } | |
808 | } | |
809 | ||
810 | /* Claim an entry in tcg_ctxs */ | |
0e2d61cf RH |
811 | n = qatomic_fetch_inc(&tcg_cur_ctxs); |
812 | g_assert(n < tcg_max_ctxs); | |
d73415a3 | 813 | qatomic_set(&tcg_ctxs[n], s); |
3468b59e | 814 | |
38b47b19 EC |
815 | if (n > 0) { |
816 | alloc_tcg_plugin_context(s); | |
bf042e8e | 817 | tcg_region_initial_alloc(s); |
38b47b19 EC |
818 | } |
819 | ||
3468b59e | 820 | tcg_ctx = s; |
e8feb96f | 821 | } |
3468b59e | 822 | #endif /* !CONFIG_USER_ONLY */ |
e8feb96f | 823 | |
c896fe29 FB |
824 | /* pool based memory allocation */ |
825 | void *tcg_malloc_internal(TCGContext *s, int size) | |
826 | { | |
827 | TCGPool *p; | |
828 | int pool_size; | |
a813e36f | 829 | |
c896fe29 FB |
830 | if (size > TCG_POOL_CHUNK_SIZE) { |
831 | /* big malloc: insert a new pool (XXX: could optimize) */ | |
7267c094 | 832 | p = g_malloc(sizeof(TCGPool) + size); |
c896fe29 | 833 | p->size = size; |
4055299e KB |
834 | p->next = s->pool_first_large; |
835 | s->pool_first_large = p; | |
836 | return p->data; | |
c896fe29 FB |
837 | } else { |
838 | p = s->pool_current; | |
839 | if (!p) { | |
840 | p = s->pool_first; | |
841 | if (!p) | |
842 | goto new_pool; | |
843 | } else { | |
844 | if (!p->next) { | |
845 | new_pool: | |
846 | pool_size = TCG_POOL_CHUNK_SIZE; | |
7267c094 | 847 | p = g_malloc(sizeof(TCGPool) + pool_size); |
c896fe29 FB |
848 | p->size = pool_size; |
849 | p->next = NULL; | |
a813e36f | 850 | if (s->pool_current) { |
c896fe29 | 851 | s->pool_current->next = p; |
a813e36f | 852 | } else { |
c896fe29 | 853 | s->pool_first = p; |
a813e36f | 854 | } |
c896fe29 FB |
855 | } else { |
856 | p = p->next; | |
857 | } | |
858 | } | |
859 | } | |
860 | s->pool_current = p; | |
861 | s->pool_cur = p->data + size; | |
862 | s->pool_end = p->data + p->size; | |
863 | return p->data; | |
864 | } | |
865 | ||
866 | void tcg_pool_reset(TCGContext *s) | |
867 | { | |
4055299e KB |
868 | TCGPool *p, *t; |
869 | for (p = s->pool_first_large; p; p = t) { | |
870 | t = p->next; | |
871 | g_free(p); | |
872 | } | |
873 | s->pool_first_large = NULL; | |
c896fe29 FB |
874 | s->pool_cur = s->pool_end = NULL; |
875 | s->pool_current = NULL; | |
876 | } | |
877 | ||
8429a1ca RH |
878 | /* |
879 | * Create TCGHelperInfo structures for "tcg/tcg-ldst.h" functions, | |
880 | * akin to what "exec/helper-tcg.h" does with DEF_HELPER_FLAGS_N. | |
881 | * We only use these for layout in tcg_out_ld_helper_ret and | |
882 | * tcg_out_st_helper_args, and share them between several of | |
883 | * the helpers, with the end result that it's easier to build manually. | |
884 | */ | |
885 | ||
886 | #if TCG_TARGET_REG_BITS == 32 | |
887 | # define dh_typecode_ttl dh_typecode_i32 | |
888 | #else | |
889 | # define dh_typecode_ttl dh_typecode_i64 | |
890 | #endif | |
891 | ||
892 | static TCGHelperInfo info_helper_ld32_mmu = { | |
893 | .flags = TCG_CALL_NO_WG, | |
894 | .typemask = dh_typemask(ttl, 0) /* return tcg_target_ulong */ | |
895 | | dh_typemask(env, 1) | |
24e46e6c | 896 | | dh_typemask(i64, 2) /* uint64_t addr */ |
8429a1ca RH |
897 | | dh_typemask(i32, 3) /* unsigned oi */ |
898 | | dh_typemask(ptr, 4) /* uintptr_t ra */ | |
899 | }; | |
900 | ||
901 | static TCGHelperInfo info_helper_ld64_mmu = { | |
902 | .flags = TCG_CALL_NO_WG, | |
903 | .typemask = dh_typemask(i64, 0) /* return uint64_t */ | |
904 | | dh_typemask(env, 1) | |
24e46e6c | 905 | | dh_typemask(i64, 2) /* uint64_t addr */ |
8429a1ca RH |
906 | | dh_typemask(i32, 3) /* unsigned oi */ |
907 | | dh_typemask(ptr, 4) /* uintptr_t ra */ | |
908 | }; | |
909 | ||
ebebea53 RH |
910 | static TCGHelperInfo info_helper_ld128_mmu = { |
911 | .flags = TCG_CALL_NO_WG, | |
912 | .typemask = dh_typemask(i128, 0) /* return Int128 */ | |
913 | | dh_typemask(env, 1) | |
24e46e6c | 914 | | dh_typemask(i64, 2) /* uint64_t addr */ |
ebebea53 RH |
915 | | dh_typemask(i32, 3) /* unsigned oi */ |
916 | | dh_typemask(ptr, 4) /* uintptr_t ra */ | |
917 | }; | |
918 | ||
8429a1ca RH |
919 | static TCGHelperInfo info_helper_st32_mmu = { |
920 | .flags = TCG_CALL_NO_WG, | |
921 | .typemask = dh_typemask(void, 0) | |
922 | | dh_typemask(env, 1) | |
24e46e6c | 923 | | dh_typemask(i64, 2) /* uint64_t addr */ |
8429a1ca RH |
924 | | dh_typemask(i32, 3) /* uint32_t data */ |
925 | | dh_typemask(i32, 4) /* unsigned oi */ | |
926 | | dh_typemask(ptr, 5) /* uintptr_t ra */ | |
927 | }; | |
928 | ||
929 | static TCGHelperInfo info_helper_st64_mmu = { | |
930 | .flags = TCG_CALL_NO_WG, | |
931 | .typemask = dh_typemask(void, 0) | |
932 | | dh_typemask(env, 1) | |
24e46e6c | 933 | | dh_typemask(i64, 2) /* uint64_t addr */ |
8429a1ca RH |
934 | | dh_typemask(i64, 3) /* uint64_t data */ |
935 | | dh_typemask(i32, 4) /* unsigned oi */ | |
936 | | dh_typemask(ptr, 5) /* uintptr_t ra */ | |
937 | }; | |
938 | ||
ebebea53 RH |
939 | static TCGHelperInfo info_helper_st128_mmu = { |
940 | .flags = TCG_CALL_NO_WG, | |
941 | .typemask = dh_typemask(void, 0) | |
942 | | dh_typemask(env, 1) | |
24e46e6c | 943 | | dh_typemask(i64, 2) /* uint64_t addr */ |
ebebea53 RH |
944 | | dh_typemask(i128, 3) /* Int128 data */ |
945 | | dh_typemask(i32, 4) /* unsigned oi */ | |
946 | | dh_typemask(ptr, 5) /* uintptr_t ra */ | |
947 | }; | |
948 | ||
22f15579 | 949 | #ifdef CONFIG_TCG_INTERPRETER |
c6ef8c7b PMD |
950 | static ffi_type *typecode_to_ffi(int argmask) |
951 | { | |
e9709e17 RH |
952 | /* |
953 | * libffi does not support __int128_t, so we have forced Int128 | |
954 | * to use the structure definition instead of the builtin type. | |
955 | */ | |
956 | static ffi_type *ffi_type_i128_elements[3] = { | |
957 | &ffi_type_uint64, | |
958 | &ffi_type_uint64, | |
959 | NULL | |
960 | }; | |
961 | static ffi_type ffi_type_i128 = { | |
962 | .size = 16, | |
963 | .alignment = __alignof__(Int128), | |
964 | .type = FFI_TYPE_STRUCT, | |
965 | .elements = ffi_type_i128_elements, | |
966 | }; | |
967 | ||
c6ef8c7b PMD |
968 | switch (argmask) { |
969 | case dh_typecode_void: | |
970 | return &ffi_type_void; | |
971 | case dh_typecode_i32: | |
972 | return &ffi_type_uint32; | |
973 | case dh_typecode_s32: | |
974 | return &ffi_type_sint32; | |
975 | case dh_typecode_i64: | |
976 | return &ffi_type_uint64; | |
977 | case dh_typecode_s64: | |
978 | return &ffi_type_sint64; | |
979 | case dh_typecode_ptr: | |
980 | return &ffi_type_pointer; | |
e9709e17 RH |
981 | case dh_typecode_i128: |
982 | return &ffi_type_i128; | |
c6ef8c7b PMD |
983 | } |
984 | g_assert_not_reached(); | |
985 | } | |
0c22e176 | 986 | |
d53106c9 | 987 | static ffi_cif *init_ffi_layout(TCGHelperInfo *info) |
0c22e176 | 988 | { |
d53106c9 RH |
989 | unsigned typemask = info->typemask; |
990 | struct { | |
991 | ffi_cif cif; | |
992 | ffi_type *args[]; | |
993 | } *ca; | |
994 | ffi_status status; | |
995 | int nargs; | |
996 | ||
997 | /* Ignoring the return type, find the last non-zero field. */ | |
998 | nargs = 32 - clz32(typemask >> 3); | |
999 | nargs = DIV_ROUND_UP(nargs, 3); | |
1000 | assert(nargs <= MAX_CALL_IARGS); | |
1001 | ||
1002 | ca = g_malloc0(sizeof(*ca) + nargs * sizeof(ffi_type *)); | |
1003 | ca->cif.rtype = typecode_to_ffi(typemask & 7); | |
1004 | ca->cif.nargs = nargs; | |
1005 | ||
1006 | if (nargs != 0) { | |
1007 | ca->cif.arg_types = ca->args; | |
1008 | for (int j = 0; j < nargs; ++j) { | |
1009 | int typecode = extract32(typemask, (j + 1) * 3, 3); | |
1010 | ca->args[j] = typecode_to_ffi(typecode); | |
0c22e176 | 1011 | } |
0c22e176 | 1012 | } |
f9c4bb80 | 1013 | |
d53106c9 RH |
1014 | status = ffi_prep_cif(&ca->cif, FFI_DEFAULT_ABI, nargs, |
1015 | ca->cif.rtype, ca->cif.arg_types); | |
1016 | assert(status == FFI_OK); | |
1017 | ||
1018 | return &ca->cif; | |
0c22e176 | 1019 | } |
d53106c9 RH |
1020 | |
1021 | #define HELPER_INFO_INIT(I) (&(I)->cif) | |
1022 | #define HELPER_INFO_INIT_VAL(I) init_ffi_layout(I) | |
1023 | #else | |
1024 | #define HELPER_INFO_INIT(I) (&(I)->init) | |
1025 | #define HELPER_INFO_INIT_VAL(I) 1 | |
0c22e176 | 1026 | #endif /* CONFIG_TCG_INTERPRETER */ |
22f15579 | 1027 | |
338b61e9 RH |
1028 | static inline bool arg_slot_reg_p(unsigned arg_slot) |
1029 | { | |
1030 | /* | |
1031 | * Split the sizeof away from the comparison to avoid Werror from | |
1032 | * "unsigned < 0 is always false", when iarg_regs is empty. | |
1033 | */ | |
1034 | unsigned nreg = ARRAY_SIZE(tcg_target_call_iarg_regs); | |
1035 | return arg_slot < nreg; | |
1036 | } | |
1037 | ||
d78e4a4f RH |
1038 | static inline int arg_slot_stk_ofs(unsigned arg_slot) |
1039 | { | |
1040 | unsigned max = TCG_STATIC_CALL_ARGS_SIZE / sizeof(tcg_target_long); | |
1041 | unsigned stk_slot = arg_slot - ARRAY_SIZE(tcg_target_call_iarg_regs); | |
1042 | ||
1043 | tcg_debug_assert(stk_slot < max); | |
1044 | return TCG_TARGET_CALL_STACK_OFFSET + stk_slot * sizeof(tcg_target_long); | |
1045 | } | |
1046 | ||
39004a71 RH |
1047 | typedef struct TCGCumulativeArgs { |
1048 | int arg_idx; /* tcg_gen_callN args[] */ | |
1049 | int info_in_idx; /* TCGHelperInfo in[] */ | |
1050 | int arg_slot; /* regs+stack slot */ | |
1051 | int ref_slot; /* stack slots for references */ | |
1052 | } TCGCumulativeArgs; | |
1053 | ||
1054 | static void layout_arg_even(TCGCumulativeArgs *cum) | |
1055 | { | |
1056 | cum->arg_slot += cum->arg_slot & 1; | |
1057 | } | |
1058 | ||
1059 | static void layout_arg_1(TCGCumulativeArgs *cum, TCGHelperInfo *info, | |
1060 | TCGCallArgumentKind kind) | |
1061 | { | |
1062 | TCGCallArgumentLoc *loc = &info->in[cum->info_in_idx]; | |
1063 | ||
1064 | *loc = (TCGCallArgumentLoc){ | |
1065 | .kind = kind, | |
1066 | .arg_idx = cum->arg_idx, | |
1067 | .arg_slot = cum->arg_slot, | |
1068 | }; | |
1069 | cum->info_in_idx++; | |
1070 | cum->arg_slot++; | |
1071 | } | |
1072 | ||
1073 | static void layout_arg_normal_n(TCGCumulativeArgs *cum, | |
1074 | TCGHelperInfo *info, int n) | |
1075 | { | |
1076 | TCGCallArgumentLoc *loc = &info->in[cum->info_in_idx]; | |
1077 | ||
1078 | for (int i = 0; i < n; ++i) { | |
1079 | /* Layout all using the same arg_idx, adjusting the subindex. */ | |
1080 | loc[i] = (TCGCallArgumentLoc){ | |
1081 | .kind = TCG_CALL_ARG_NORMAL, | |
1082 | .arg_idx = cum->arg_idx, | |
1083 | .tmp_subindex = i, | |
1084 | .arg_slot = cum->arg_slot + i, | |
1085 | }; | |
1086 | } | |
1087 | cum->info_in_idx += n; | |
1088 | cum->arg_slot += n; | |
1089 | } | |
1090 | ||
313bdea8 RH |
1091 | static void layout_arg_by_ref(TCGCumulativeArgs *cum, TCGHelperInfo *info) |
1092 | { | |
1093 | TCGCallArgumentLoc *loc = &info->in[cum->info_in_idx]; | |
1094 | int n = 128 / TCG_TARGET_REG_BITS; | |
1095 | ||
1096 | /* The first subindex carries the pointer. */ | |
1097 | layout_arg_1(cum, info, TCG_CALL_ARG_BY_REF); | |
1098 | ||
1099 | /* | |
1100 | * The callee is allowed to clobber memory associated with | |
1101 | * structure pass by-reference. Therefore we must make copies. | |
1102 | * Allocate space from "ref_slot", which will be adjusted to | |
1103 | * follow the parameters on the stack. | |
1104 | */ | |
1105 | loc[0].ref_slot = cum->ref_slot; | |
1106 | ||
1107 | /* | |
1108 | * Subsequent words also go into the reference slot, but | |
1109 | * do not accumulate into the regular arguments. | |
1110 | */ | |
1111 | for (int i = 1; i < n; ++i) { | |
1112 | loc[i] = (TCGCallArgumentLoc){ | |
1113 | .kind = TCG_CALL_ARG_BY_REF_N, | |
1114 | .arg_idx = cum->arg_idx, | |
1115 | .tmp_subindex = i, | |
1116 | .ref_slot = cum->ref_slot + i, | |
1117 | }; | |
1118 | } | |
e18ed26c | 1119 | cum->info_in_idx += n - 1; /* i=0 accounted for in layout_arg_1 */ |
313bdea8 RH |
1120 | cum->ref_slot += n; |
1121 | } | |
1122 | ||
39004a71 RH |
1123 | static void init_call_layout(TCGHelperInfo *info) |
1124 | { | |
1125 | int max_reg_slots = ARRAY_SIZE(tcg_target_call_iarg_regs); | |
1126 | int max_stk_slots = TCG_STATIC_CALL_ARGS_SIZE / sizeof(tcg_target_long); | |
1127 | unsigned typemask = info->typemask; | |
1128 | unsigned typecode; | |
1129 | TCGCumulativeArgs cum = { }; | |
1130 | ||
1131 | /* | |
1132 | * Parse and place any function return value. | |
1133 | */ | |
1134 | typecode = typemask & 7; | |
1135 | switch (typecode) { | |
1136 | case dh_typecode_void: | |
1137 | info->nr_out = 0; | |
1138 | break; | |
1139 | case dh_typecode_i32: | |
1140 | case dh_typecode_s32: | |
1141 | case dh_typecode_ptr: | |
1142 | info->nr_out = 1; | |
1143 | info->out_kind = TCG_CALL_RET_NORMAL; | |
1144 | break; | |
1145 | case dh_typecode_i64: | |
1146 | case dh_typecode_s64: | |
1147 | info->nr_out = 64 / TCG_TARGET_REG_BITS; | |
1148 | info->out_kind = TCG_CALL_RET_NORMAL; | |
5e3d0c19 RH |
1149 | /* Query the last register now to trigger any assert early. */ |
1150 | tcg_target_call_oarg_reg(info->out_kind, info->nr_out - 1); | |
466d3759 RH |
1151 | break; |
1152 | case dh_typecode_i128: | |
1153 | info->nr_out = 128 / TCG_TARGET_REG_BITS; | |
5427a9a7 RH |
1154 | info->out_kind = TCG_TARGET_CALL_RET_I128; |
1155 | switch (TCG_TARGET_CALL_RET_I128) { | |
466d3759 | 1156 | case TCG_CALL_RET_NORMAL: |
5e3d0c19 RH |
1157 | /* Query the last register now to trigger any assert early. */ |
1158 | tcg_target_call_oarg_reg(info->out_kind, info->nr_out - 1); | |
466d3759 | 1159 | break; |
c6556aa0 RH |
1160 | case TCG_CALL_RET_BY_VEC: |
1161 | /* Query the single register now to trigger any assert early. */ | |
1162 | tcg_target_call_oarg_reg(TCG_CALL_RET_BY_VEC, 0); | |
1163 | break; | |
313bdea8 RH |
1164 | case TCG_CALL_RET_BY_REF: |
1165 | /* | |
1166 | * Allocate the first argument to the output. | |
1167 | * We don't need to store this anywhere, just make it | |
1168 | * unavailable for use in the input loop below. | |
1169 | */ | |
1170 | cum.arg_slot = 1; | |
1171 | break; | |
466d3759 RH |
1172 | default: |
1173 | qemu_build_not_reached(); | |
1174 | } | |
39004a71 RH |
1175 | break; |
1176 | default: | |
1177 | g_assert_not_reached(); | |
1178 | } | |
39004a71 RH |
1179 | |
1180 | /* | |
1181 | * Parse and place function arguments. | |
1182 | */ | |
1183 | for (typemask >>= 3; typemask; typemask >>= 3, cum.arg_idx++) { | |
1184 | TCGCallArgumentKind kind; | |
1185 | TCGType type; | |
1186 | ||
1187 | typecode = typemask & 7; | |
1188 | switch (typecode) { | |
1189 | case dh_typecode_i32: | |
1190 | case dh_typecode_s32: | |
1191 | type = TCG_TYPE_I32; | |
1192 | break; | |
1193 | case dh_typecode_i64: | |
1194 | case dh_typecode_s64: | |
1195 | type = TCG_TYPE_I64; | |
1196 | break; | |
1197 | case dh_typecode_ptr: | |
1198 | type = TCG_TYPE_PTR; | |
1199 | break; | |
466d3759 RH |
1200 | case dh_typecode_i128: |
1201 | type = TCG_TYPE_I128; | |
1202 | break; | |
39004a71 RH |
1203 | default: |
1204 | g_assert_not_reached(); | |
1205 | } | |
1206 | ||
1207 | switch (type) { | |
1208 | case TCG_TYPE_I32: | |
1209 | switch (TCG_TARGET_CALL_ARG_I32) { | |
1210 | case TCG_CALL_ARG_EVEN: | |
1211 | layout_arg_even(&cum); | |
1212 | /* fall through */ | |
1213 | case TCG_CALL_ARG_NORMAL: | |
1214 | layout_arg_1(&cum, info, TCG_CALL_ARG_NORMAL); | |
1215 | break; | |
1216 | case TCG_CALL_ARG_EXTEND: | |
1217 | kind = TCG_CALL_ARG_EXTEND_U + (typecode & 1); | |
1218 | layout_arg_1(&cum, info, kind); | |
1219 | break; | |
1220 | default: | |
1221 | qemu_build_not_reached(); | |
1222 | } | |
1223 | break; | |
1224 | ||
1225 | case TCG_TYPE_I64: | |
1226 | switch (TCG_TARGET_CALL_ARG_I64) { | |
1227 | case TCG_CALL_ARG_EVEN: | |
1228 | layout_arg_even(&cum); | |
1229 | /* fall through */ | |
1230 | case TCG_CALL_ARG_NORMAL: | |
1231 | if (TCG_TARGET_REG_BITS == 32) { | |
1232 | layout_arg_normal_n(&cum, info, 2); | |
1233 | } else { | |
1234 | layout_arg_1(&cum, info, TCG_CALL_ARG_NORMAL); | |
1235 | } | |
1236 | break; | |
1237 | default: | |
1238 | qemu_build_not_reached(); | |
1239 | } | |
1240 | break; | |
1241 | ||
466d3759 | 1242 | case TCG_TYPE_I128: |
5427a9a7 | 1243 | switch (TCG_TARGET_CALL_ARG_I128) { |
466d3759 RH |
1244 | case TCG_CALL_ARG_EVEN: |
1245 | layout_arg_even(&cum); | |
1246 | /* fall through */ | |
1247 | case TCG_CALL_ARG_NORMAL: | |
1248 | layout_arg_normal_n(&cum, info, 128 / TCG_TARGET_REG_BITS); | |
1249 | break; | |
313bdea8 RH |
1250 | case TCG_CALL_ARG_BY_REF: |
1251 | layout_arg_by_ref(&cum, info); | |
1252 | break; | |
466d3759 RH |
1253 | default: |
1254 | qemu_build_not_reached(); | |
1255 | } | |
1256 | break; | |
1257 | ||
39004a71 RH |
1258 | default: |
1259 | g_assert_not_reached(); | |
1260 | } | |
1261 | } | |
1262 | info->nr_in = cum.info_in_idx; | |
1263 | ||
1264 | /* Validate that we didn't overrun the input array. */ | |
1265 | assert(cum.info_in_idx <= ARRAY_SIZE(info->in)); | |
1266 | /* Validate the backend has enough argument space. */ | |
1267 | assert(cum.arg_slot <= max_reg_slots + max_stk_slots); | |
313bdea8 RH |
1268 | |
1269 | /* | |
1270 | * Relocate the "ref_slot" area to the end of the parameters. | |
1271 | * Minimizing this stack offset helps code size for x86, | |
1272 | * which has a signed 8-bit offset encoding. | |
1273 | */ | |
1274 | if (cum.ref_slot != 0) { | |
1275 | int ref_base = 0; | |
1276 | ||
1277 | if (cum.arg_slot > max_reg_slots) { | |
1278 | int align = __alignof(Int128) / sizeof(tcg_target_long); | |
1279 | ||
1280 | ref_base = cum.arg_slot - max_reg_slots; | |
1281 | if (align > 1) { | |
1282 | ref_base = ROUND_UP(ref_base, align); | |
1283 | } | |
1284 | } | |
1285 | assert(ref_base + cum.ref_slot <= max_stk_slots); | |
d78e4a4f | 1286 | ref_base += max_reg_slots; |
313bdea8 RH |
1287 | |
1288 | if (ref_base != 0) { | |
1289 | for (int i = cum.info_in_idx - 1; i >= 0; --i) { | |
1290 | TCGCallArgumentLoc *loc = &info->in[i]; | |
1291 | switch (loc->kind) { | |
1292 | case TCG_CALL_ARG_BY_REF: | |
1293 | case TCG_CALL_ARG_BY_REF_N: | |
1294 | loc->ref_slot += ref_base; | |
1295 | break; | |
1296 | default: | |
1297 | break; | |
1298 | } | |
1299 | } | |
1300 | } | |
1301 | } | |
39004a71 RH |
1302 | } |
1303 | ||
91478cef | 1304 | static int indirect_reg_alloc_order[ARRAY_SIZE(tcg_target_reg_alloc_order)]; |
f69d277e | 1305 | static void process_op_defs(TCGContext *s); |
1c2adb95 RH |
1306 | static TCGTemp *tcg_global_reg_new_internal(TCGContext *s, TCGType type, |
1307 | TCGReg reg, const char *name); | |
91478cef | 1308 | |
43b972b7 | 1309 | static void tcg_context_init(unsigned max_cpus) |
c896fe29 | 1310 | { |
a76aabd3 | 1311 | TCGContext *s = &tcg_init_ctx; |
100b5e01 | 1312 | int op, total_args, n, i; |
c896fe29 FB |
1313 | TCGOpDef *def; |
1314 | TCGArgConstraint *args_ct; | |
1c2adb95 | 1315 | TCGTemp *ts; |
c896fe29 FB |
1316 | |
1317 | memset(s, 0, sizeof(*s)); | |
c896fe29 | 1318 | s->nb_globals = 0; |
c70fbf0a | 1319 | |
c896fe29 FB |
1320 | /* Count total number of arguments and allocate the corresponding |
1321 | space */ | |
1322 | total_args = 0; | |
1323 | for(op = 0; op < NB_OPS; op++) { | |
1324 | def = &tcg_op_defs[op]; | |
1325 | n = def->nb_iargs + def->nb_oargs; | |
1326 | total_args += n; | |
1327 | } | |
1328 | ||
bc2b17e6 | 1329 | args_ct = g_new0(TCGArgConstraint, total_args); |
c896fe29 FB |
1330 | |
1331 | for(op = 0; op < NB_OPS; op++) { | |
1332 | def = &tcg_op_defs[op]; | |
1333 | def->args_ct = args_ct; | |
c896fe29 | 1334 | n = def->nb_iargs + def->nb_oargs; |
c896fe29 FB |
1335 | args_ct += n; |
1336 | } | |
5cd8f621 | 1337 | |
8429a1ca RH |
1338 | init_call_layout(&info_helper_ld32_mmu); |
1339 | init_call_layout(&info_helper_ld64_mmu); | |
ebebea53 | 1340 | init_call_layout(&info_helper_ld128_mmu); |
8429a1ca RH |
1341 | init_call_layout(&info_helper_st32_mmu); |
1342 | init_call_layout(&info_helper_st64_mmu); | |
ebebea53 | 1343 | init_call_layout(&info_helper_st128_mmu); |
8429a1ca | 1344 | |
c896fe29 | 1345 | tcg_target_init(s); |
f69d277e | 1346 | process_op_defs(s); |
91478cef RH |
1347 | |
1348 | /* Reverse the order of the saved registers, assuming they're all at | |
1349 | the start of tcg_target_reg_alloc_order. */ | |
1350 | for (n = 0; n < ARRAY_SIZE(tcg_target_reg_alloc_order); ++n) { | |
1351 | int r = tcg_target_reg_alloc_order[n]; | |
1352 | if (tcg_regset_test_reg(tcg_target_call_clobber_regs, r)) { | |
1353 | break; | |
1354 | } | |
1355 | } | |
1356 | for (i = 0; i < n; ++i) { | |
1357 | indirect_reg_alloc_order[i] = tcg_target_reg_alloc_order[n - 1 - i]; | |
1358 | } | |
1359 | for (; i < ARRAY_SIZE(tcg_target_reg_alloc_order); ++i) { | |
1360 | indirect_reg_alloc_order[i] = tcg_target_reg_alloc_order[i]; | |
1361 | } | |
b1311c4a | 1362 | |
38b47b19 EC |
1363 | alloc_tcg_plugin_context(s); |
1364 | ||
b1311c4a | 1365 | tcg_ctx = s; |
3468b59e EC |
1366 | /* |
1367 | * In user-mode we simply share the init context among threads, since we | |
1368 | * use a single region. See the documentation tcg_region_init() for the | |
1369 | * reasoning behind this. | |
7893e42d | 1370 | * In system-mode we will have at most max_cpus TCG threads. |
3468b59e EC |
1371 | */ |
1372 | #ifdef CONFIG_USER_ONLY | |
df2cce29 | 1373 | tcg_ctxs = &tcg_ctx; |
0e2d61cf RH |
1374 | tcg_cur_ctxs = 1; |
1375 | tcg_max_ctxs = 1; | |
3468b59e | 1376 | #else |
0e2d61cf RH |
1377 | tcg_max_ctxs = max_cpus; |
1378 | tcg_ctxs = g_new0(TCGContext *, max_cpus); | |
3468b59e | 1379 | #endif |
1c2adb95 RH |
1380 | |
1381 | tcg_debug_assert(!tcg_regset_test_reg(s->reserved_regs, TCG_AREG0)); | |
1382 | ts = tcg_global_reg_new_internal(s, TCG_TYPE_PTR, TCG_AREG0, "env"); | |
ad75a51e | 1383 | tcg_env = temp_tcgv_ptr(ts); |
9002ec79 | 1384 | } |
b03cce8e | 1385 | |
43b972b7 | 1386 | void tcg_init(size_t tb_size, int splitwx, unsigned max_cpus) |
a76aabd3 | 1387 | { |
43b972b7 RH |
1388 | tcg_context_init(max_cpus); |
1389 | tcg_region_init(tb_size, splitwx, max_cpus); | |
a76aabd3 RH |
1390 | } |
1391 | ||
6e3b2bfd EC |
1392 | /* |
1393 | * Allocate TBs right before their corresponding translated code, making | |
1394 | * sure that TBs and code are on different cache lines. | |
1395 | */ | |
1396 | TranslationBlock *tcg_tb_alloc(TCGContext *s) | |
1397 | { | |
1398 | uintptr_t align = qemu_icache_linesize; | |
1399 | TranslationBlock *tb; | |
1400 | void *next; | |
1401 | ||
e8feb96f | 1402 | retry: |
6e3b2bfd EC |
1403 | tb = (void *)ROUND_UP((uintptr_t)s->code_gen_ptr, align); |
1404 | next = (void *)ROUND_UP((uintptr_t)(tb + 1), align); | |
1405 | ||
1406 | if (unlikely(next > s->code_gen_highwater)) { | |
e8feb96f EC |
1407 | if (tcg_region_alloc(s)) { |
1408 | return NULL; | |
1409 | } | |
1410 | goto retry; | |
6e3b2bfd | 1411 | } |
d73415a3 | 1412 | qatomic_set(&s->code_gen_ptr, next); |
57a26946 | 1413 | s->data_gen_ptr = NULL; |
6e3b2bfd EC |
1414 | return tb; |
1415 | } | |
1416 | ||
935f75ae | 1417 | void tcg_prologue_init(void) |
9002ec79 | 1418 | { |
935f75ae | 1419 | TCGContext *s = tcg_ctx; |
b0a0794a | 1420 | size_t prologue_size; |
8163b749 | 1421 | |
b0a0794a RH |
1422 | s->code_ptr = s->code_gen_ptr; |
1423 | s->code_buf = s->code_gen_ptr; | |
5b38ee31 | 1424 | s->data_gen_ptr = NULL; |
b91ccb31 RH |
1425 | |
1426 | #ifndef CONFIG_TCG_INTERPRETER | |
b0a0794a | 1427 | tcg_qemu_tb_exec = (tcg_prologue_fn *)tcg_splitwx_to_rx(s->code_ptr); |
b91ccb31 | 1428 | #endif |
8163b749 | 1429 | |
5b38ee31 RH |
1430 | #ifdef TCG_TARGET_NEED_POOL_LABELS |
1431 | s->pool_labels = NULL; | |
1432 | #endif | |
1433 | ||
653b87eb | 1434 | qemu_thread_jit_write(); |
8163b749 | 1435 | /* Generate the prologue. */ |
b03cce8e | 1436 | tcg_target_qemu_prologue(s); |
5b38ee31 RH |
1437 | |
1438 | #ifdef TCG_TARGET_NEED_POOL_LABELS | |
1439 | /* Allow the prologue to put e.g. guest_base into a pool entry. */ | |
1440 | { | |
1768987b RH |
1441 | int result = tcg_out_pool_finalize(s); |
1442 | tcg_debug_assert(result == 0); | |
5b38ee31 RH |
1443 | } |
1444 | #endif | |
1445 | ||
b0a0794a | 1446 | prologue_size = tcg_current_code_size(s); |
5584e2db | 1447 | perf_report_prologue(s->code_gen_ptr, prologue_size); |
b0a0794a | 1448 | |
df5d2b16 | 1449 | #ifndef CONFIG_TCG_INTERPRETER |
b0a0794a RH |
1450 | flush_idcache_range((uintptr_t)tcg_splitwx_to_rx(s->code_buf), |
1451 | (uintptr_t)s->code_buf, prologue_size); | |
df5d2b16 | 1452 | #endif |
8163b749 | 1453 | |
d6b64b2b | 1454 | if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM)) { |
c60f599b | 1455 | FILE *logfile = qemu_log_trylock(); |
78b54858 RH |
1456 | if (logfile) { |
1457 | fprintf(logfile, "PROLOGUE: [size=%zu]\n", prologue_size); | |
1458 | if (s->data_gen_ptr) { | |
1459 | size_t code_size = s->data_gen_ptr - s->code_gen_ptr; | |
1460 | size_t data_size = prologue_size - code_size; | |
1461 | size_t i; | |
1462 | ||
1463 | disas(logfile, s->code_gen_ptr, code_size); | |
1464 | ||
1465 | for (i = 0; i < data_size; i += sizeof(tcg_target_ulong)) { | |
1466 | if (sizeof(tcg_target_ulong) == 8) { | |
1467 | fprintf(logfile, | |
1468 | "0x%08" PRIxPTR ": .quad 0x%016" PRIx64 "\n", | |
1469 | (uintptr_t)s->data_gen_ptr + i, | |
1470 | *(uint64_t *)(s->data_gen_ptr + i)); | |
1471 | } else { | |
1472 | fprintf(logfile, | |
1473 | "0x%08" PRIxPTR ": .long 0x%08x\n", | |
1474 | (uintptr_t)s->data_gen_ptr + i, | |
1475 | *(uint32_t *)(s->data_gen_ptr + i)); | |
1476 | } | |
5b38ee31 | 1477 | } |
78b54858 RH |
1478 | } else { |
1479 | disas(logfile, s->code_gen_ptr, prologue_size); | |
5b38ee31 | 1480 | } |
78b54858 | 1481 | fprintf(logfile, "\n"); |
78b54858 | 1482 | qemu_log_unlock(logfile); |
5b38ee31 | 1483 | } |
d6b64b2b | 1484 | } |
cedbcb01 | 1485 | |
6eea0434 RH |
1486 | #ifndef CONFIG_TCG_INTERPRETER |
1487 | /* | |
1488 | * Assert that goto_ptr is implemented completely, setting an epilogue. | |
1489 | * For tci, we use NULL as the signal to return from the interpreter, | |
1490 | * so skip this check. | |
1491 | */ | |
f4e01e30 | 1492 | tcg_debug_assert(tcg_code_gen_epilogue != NULL); |
6eea0434 | 1493 | #endif |
d1c74ab3 RH |
1494 | |
1495 | tcg_region_prologue_set(s); | |
c896fe29 FB |
1496 | } |
1497 | ||
c896fe29 FB |
1498 | void tcg_func_start(TCGContext *s) |
1499 | { | |
1500 | tcg_pool_reset(s); | |
1501 | s->nb_temps = s->nb_globals; | |
0ec9eabc RH |
1502 | |
1503 | /* No temps have been previously allocated for size or locality. */ | |
1504 | memset(s->free_temps, 0, sizeof(s->free_temps)); | |
1505 | ||
c0522136 RH |
1506 | /* No constant temps have been previously allocated. */ |
1507 | for (int i = 0; i < TCG_TYPE_COUNT; ++i) { | |
1508 | if (s->const_table[i]) { | |
1509 | g_hash_table_remove_all(s->const_table[i]); | |
1510 | } | |
1511 | } | |
1512 | ||
abebf925 | 1513 | s->nb_ops = 0; |
c896fe29 FB |
1514 | s->nb_labels = 0; |
1515 | s->current_frame_offset = s->frame_start; | |
1516 | ||
0a209d4b RH |
1517 | #ifdef CONFIG_DEBUG_TCG |
1518 | s->goto_tb_issue_mask = 0; | |
1519 | #endif | |
1520 | ||
15fa08f8 RH |
1521 | QTAILQ_INIT(&s->ops); |
1522 | QTAILQ_INIT(&s->free_ops); | |
bef16ab4 | 1523 | QSIMPLEQ_INIT(&s->labels); |
4baf3978 RH |
1524 | |
1525 | tcg_debug_assert(s->addr_type == TCG_TYPE_I32 || | |
1526 | s->addr_type == TCG_TYPE_I64); | |
d0a9bb5e | 1527 | |
747bd69d | 1528 | tcg_debug_assert(s->insn_start_words > 0); |
c896fe29 FB |
1529 | } |
1530 | ||
ae30e866 | 1531 | static TCGTemp *tcg_temp_alloc(TCGContext *s) |
7ca4b752 RH |
1532 | { |
1533 | int n = s->nb_temps++; | |
ae30e866 RH |
1534 | |
1535 | if (n >= TCG_MAX_TEMPS) { | |
db6b7d0c | 1536 | tcg_raise_tb_overflow(s); |
ae30e866 | 1537 | } |
7ca4b752 RH |
1538 | return memset(&s->temps[n], 0, sizeof(TCGTemp)); |
1539 | } | |
1540 | ||
ae30e866 | 1541 | static TCGTemp *tcg_global_alloc(TCGContext *s) |
7ca4b752 | 1542 | { |
fa477d25 RH |
1543 | TCGTemp *ts; |
1544 | ||
7ca4b752 | 1545 | tcg_debug_assert(s->nb_globals == s->nb_temps); |
ae30e866 | 1546 | tcg_debug_assert(s->nb_globals < TCG_MAX_TEMPS); |
7ca4b752 | 1547 | s->nb_globals++; |
fa477d25 | 1548 | ts = tcg_temp_alloc(s); |
ee17db83 | 1549 | ts->kind = TEMP_GLOBAL; |
fa477d25 RH |
1550 | |
1551 | return ts; | |
c896fe29 FB |
1552 | } |
1553 | ||
085272b3 RH |
1554 | static TCGTemp *tcg_global_reg_new_internal(TCGContext *s, TCGType type, |
1555 | TCGReg reg, const char *name) | |
c896fe29 | 1556 | { |
c896fe29 | 1557 | TCGTemp *ts; |
c896fe29 | 1558 | |
1a057554 | 1559 | tcg_debug_assert(TCG_TARGET_REG_BITS == 64 || type == TCG_TYPE_I32); |
7ca4b752 RH |
1560 | |
1561 | ts = tcg_global_alloc(s); | |
c896fe29 FB |
1562 | ts->base_type = type; |
1563 | ts->type = type; | |
ee17db83 | 1564 | ts->kind = TEMP_FIXED; |
c896fe29 | 1565 | ts->reg = reg; |
c896fe29 | 1566 | ts->name = name; |
c896fe29 | 1567 | tcg_regset_set_reg(s->reserved_regs, reg); |
7ca4b752 | 1568 | |
085272b3 | 1569 | return ts; |
a7812ae4 PB |
1570 | } |
1571 | ||
b6638662 | 1572 | void tcg_set_frame(TCGContext *s, TCGReg reg, intptr_t start, intptr_t size) |
b3a62939 | 1573 | { |
b3a62939 RH |
1574 | s->frame_start = start; |
1575 | s->frame_end = start + size; | |
085272b3 RH |
1576 | s->frame_temp |
1577 | = tcg_global_reg_new_internal(s, TCG_TYPE_PTR, reg, "_frame"); | |
b3a62939 RH |
1578 | } |
1579 | ||
4643f3e0 RH |
1580 | static TCGTemp *tcg_global_mem_new_internal(TCGv_ptr base, intptr_t offset, |
1581 | const char *name, TCGType type) | |
c896fe29 | 1582 | { |
b1311c4a | 1583 | TCGContext *s = tcg_ctx; |
dc41aa7d | 1584 | TCGTemp *base_ts = tcgv_ptr_temp(base); |
7ca4b752 | 1585 | TCGTemp *ts = tcg_global_alloc(s); |
aef85402 | 1586 | int indirect_reg = 0; |
c896fe29 | 1587 | |
c0522136 RH |
1588 | switch (base_ts->kind) { |
1589 | case TEMP_FIXED: | |
1590 | break; | |
1591 | case TEMP_GLOBAL: | |
5a18407f RH |
1592 | /* We do not support double-indirect registers. */ |
1593 | tcg_debug_assert(!base_ts->indirect_reg); | |
b3915dbb | 1594 | base_ts->indirect_base = 1; |
5a18407f RH |
1595 | s->nb_indirects += (TCG_TARGET_REG_BITS == 32 && type == TCG_TYPE_I64 |
1596 | ? 2 : 1); | |
1597 | indirect_reg = 1; | |
c0522136 RH |
1598 | break; |
1599 | default: | |
1600 | g_assert_not_reached(); | |
b3915dbb RH |
1601 | } |
1602 | ||
7ca4b752 RH |
1603 | if (TCG_TARGET_REG_BITS == 32 && type == TCG_TYPE_I64) { |
1604 | TCGTemp *ts2 = tcg_global_alloc(s); | |
c896fe29 | 1605 | char buf[64]; |
7ca4b752 RH |
1606 | |
1607 | ts->base_type = TCG_TYPE_I64; | |
c896fe29 | 1608 | ts->type = TCG_TYPE_I32; |
b3915dbb | 1609 | ts->indirect_reg = indirect_reg; |
c896fe29 | 1610 | ts->mem_allocated = 1; |
b3a62939 | 1611 | ts->mem_base = base_ts; |
aef85402 | 1612 | ts->mem_offset = offset; |
c896fe29 FB |
1613 | pstrcpy(buf, sizeof(buf), name); |
1614 | pstrcat(buf, sizeof(buf), "_0"); | |
1615 | ts->name = strdup(buf); | |
c896fe29 | 1616 | |
7ca4b752 RH |
1617 | tcg_debug_assert(ts2 == ts + 1); |
1618 | ts2->base_type = TCG_TYPE_I64; | |
1619 | ts2->type = TCG_TYPE_I32; | |
b3915dbb | 1620 | ts2->indirect_reg = indirect_reg; |
7ca4b752 RH |
1621 | ts2->mem_allocated = 1; |
1622 | ts2->mem_base = base_ts; | |
aef85402 | 1623 | ts2->mem_offset = offset + 4; |
fac87bd2 | 1624 | ts2->temp_subindex = 1; |
c896fe29 FB |
1625 | pstrcpy(buf, sizeof(buf), name); |
1626 | pstrcat(buf, sizeof(buf), "_1"); | |
120c1084 | 1627 | ts2->name = strdup(buf); |
7ca4b752 | 1628 | } else { |
c896fe29 FB |
1629 | ts->base_type = type; |
1630 | ts->type = type; | |
b3915dbb | 1631 | ts->indirect_reg = indirect_reg; |
c896fe29 | 1632 | ts->mem_allocated = 1; |
b3a62939 | 1633 | ts->mem_base = base_ts; |
c896fe29 | 1634 | ts->mem_offset = offset; |
c896fe29 | 1635 | ts->name = name; |
c896fe29 | 1636 | } |
085272b3 | 1637 | return ts; |
a7812ae4 PB |
1638 | } |
1639 | ||
4643f3e0 RH |
1640 | TCGv_i32 tcg_global_mem_new_i32(TCGv_ptr reg, intptr_t off, const char *name) |
1641 | { | |
1642 | TCGTemp *ts = tcg_global_mem_new_internal(reg, off, name, TCG_TYPE_I32); | |
1643 | return temp_tcgv_i32(ts); | |
1644 | } | |
1645 | ||
1646 | TCGv_i64 tcg_global_mem_new_i64(TCGv_ptr reg, intptr_t off, const char *name) | |
1647 | { | |
1648 | TCGTemp *ts = tcg_global_mem_new_internal(reg, off, name, TCG_TYPE_I64); | |
1649 | return temp_tcgv_i64(ts); | |
1650 | } | |
1651 | ||
1652 | TCGv_ptr tcg_global_mem_new_ptr(TCGv_ptr reg, intptr_t off, const char *name) | |
1653 | { | |
1654 | TCGTemp *ts = tcg_global_mem_new_internal(reg, off, name, TCG_TYPE_PTR); | |
1655 | return temp_tcgv_ptr(ts); | |
1656 | } | |
1657 | ||
1658 | static TCGTemp *tcg_temp_new_internal(TCGType type, TCGTempKind kind) | |
c896fe29 | 1659 | { |
b1311c4a | 1660 | TCGContext *s = tcg_ctx; |
c896fe29 | 1661 | TCGTemp *ts; |
e1c08b00 | 1662 | int n; |
7ca4b752 | 1663 | |
e1c08b00 RH |
1664 | if (kind == TEMP_EBB) { |
1665 | int idx = find_first_bit(s->free_temps[type].l, TCG_MAX_TEMPS); | |
1666 | ||
1667 | if (idx < TCG_MAX_TEMPS) { | |
1668 | /* There is already an available temp with the right type. */ | |
1669 | clear_bit(idx, s->free_temps[type].l); | |
1670 | ||
1671 | ts = &s->temps[idx]; | |
1672 | ts->temp_allocated = 1; | |
1673 | tcg_debug_assert(ts->base_type == type); | |
1674 | tcg_debug_assert(ts->kind == kind); | |
2f2e911d | 1675 | return ts; |
43eef72f | 1676 | } |
e1c08b00 RH |
1677 | } else { |
1678 | tcg_debug_assert(kind == TEMP_TB); | |
1679 | } | |
7ca4b752 | 1680 | |
e1c08b00 RH |
1681 | switch (type) { |
1682 | case TCG_TYPE_I32: | |
1683 | case TCG_TYPE_V64: | |
1684 | case TCG_TYPE_V128: | |
1685 | case TCG_TYPE_V256: | |
1686 | n = 1; | |
1687 | break; | |
1688 | case TCG_TYPE_I64: | |
1689 | n = 64 / TCG_TARGET_REG_BITS; | |
1690 | break; | |
1691 | case TCG_TYPE_I128: | |
1692 | n = 128 / TCG_TARGET_REG_BITS; | |
1693 | break; | |
1694 | default: | |
1695 | g_assert_not_reached(); | |
1696 | } | |
43eef72f | 1697 | |
e1c08b00 RH |
1698 | ts = tcg_temp_alloc(s); |
1699 | ts->base_type = type; | |
1700 | ts->temp_allocated = 1; | |
1701 | ts->kind = kind; | |
1702 | ||
1703 | if (n == 1) { | |
1704 | ts->type = type; | |
1705 | } else { | |
1706 | ts->type = TCG_TYPE_REG; | |
43eef72f | 1707 | |
e1c08b00 RH |
1708 | for (int i = 1; i < n; ++i) { |
1709 | TCGTemp *ts2 = tcg_temp_alloc(s); | |
43eef72f | 1710 | |
e1c08b00 RH |
1711 | tcg_debug_assert(ts2 == ts + i); |
1712 | ts2->base_type = type; | |
1713 | ts2->type = TCG_TYPE_REG; | |
1714 | ts2->temp_allocated = 1; | |
1715 | ts2->temp_subindex = i; | |
1716 | ts2->kind = kind; | |
e8996ee0 | 1717 | } |
c896fe29 | 1718 | } |
085272b3 | 1719 | return ts; |
c896fe29 FB |
1720 | } |
1721 | ||
4643f3e0 RH |
1722 | TCGv_i32 tcg_temp_new_i32(void) |
1723 | { | |
1724 | return temp_tcgv_i32(tcg_temp_new_internal(TCG_TYPE_I32, TEMP_TB)); | |
1725 | } | |
1726 | ||
1727 | TCGv_i32 tcg_temp_ebb_new_i32(void) | |
1728 | { | |
1729 | return temp_tcgv_i32(tcg_temp_new_internal(TCG_TYPE_I32, TEMP_EBB)); | |
1730 | } | |
1731 | ||
1732 | TCGv_i64 tcg_temp_new_i64(void) | |
1733 | { | |
1734 | return temp_tcgv_i64(tcg_temp_new_internal(TCG_TYPE_I64, TEMP_TB)); | |
1735 | } | |
1736 | ||
1737 | TCGv_i64 tcg_temp_ebb_new_i64(void) | |
1738 | { | |
1739 | return temp_tcgv_i64(tcg_temp_new_internal(TCG_TYPE_I64, TEMP_EBB)); | |
1740 | } | |
1741 | ||
1742 | TCGv_ptr tcg_temp_new_ptr(void) | |
1743 | { | |
1744 | return temp_tcgv_ptr(tcg_temp_new_internal(TCG_TYPE_PTR, TEMP_TB)); | |
1745 | } | |
1746 | ||
1747 | TCGv_ptr tcg_temp_ebb_new_ptr(void) | |
1748 | { | |
1749 | return temp_tcgv_ptr(tcg_temp_new_internal(TCG_TYPE_PTR, TEMP_EBB)); | |
1750 | } | |
1751 | ||
1752 | TCGv_i128 tcg_temp_new_i128(void) | |
1753 | { | |
1754 | return temp_tcgv_i128(tcg_temp_new_internal(TCG_TYPE_I128, TEMP_TB)); | |
1755 | } | |
1756 | ||
1757 | TCGv_i128 tcg_temp_ebb_new_i128(void) | |
1758 | { | |
1759 | return temp_tcgv_i128(tcg_temp_new_internal(TCG_TYPE_I128, TEMP_EBB)); | |
1760 | } | |
1761 | ||
d2fd745f RH |
1762 | TCGv_vec tcg_temp_new_vec(TCGType type) |
1763 | { | |
1764 | TCGTemp *t; | |
1765 | ||
1766 | #ifdef CONFIG_DEBUG_TCG | |
1767 | switch (type) { | |
1768 | case TCG_TYPE_V64: | |
1769 | assert(TCG_TARGET_HAS_v64); | |
1770 | break; | |
1771 | case TCG_TYPE_V128: | |
1772 | assert(TCG_TARGET_HAS_v128); | |
1773 | break; | |
1774 | case TCG_TYPE_V256: | |
1775 | assert(TCG_TARGET_HAS_v256); | |
1776 | break; | |
1777 | default: | |
1778 | g_assert_not_reached(); | |
1779 | } | |
1780 | #endif | |
1781 | ||
bbf989bf | 1782 | t = tcg_temp_new_internal(type, TEMP_EBB); |
d2fd745f RH |
1783 | return temp_tcgv_vec(t); |
1784 | } | |
1785 | ||
1786 | /* Create a new temp of the same type as an existing temp. */ | |
1787 | TCGv_vec tcg_temp_new_vec_matching(TCGv_vec match) | |
1788 | { | |
1789 | TCGTemp *t = tcgv_vec_temp(match); | |
1790 | ||
1791 | tcg_debug_assert(t->temp_allocated != 0); | |
1792 | ||
bbf989bf | 1793 | t = tcg_temp_new_internal(t->base_type, TEMP_EBB); |
d2fd745f RH |
1794 | return temp_tcgv_vec(t); |
1795 | } | |
1796 | ||
5bfa8034 | 1797 | void tcg_temp_free_internal(TCGTemp *ts) |
c896fe29 | 1798 | { |
b1311c4a | 1799 | TCGContext *s = tcg_ctx; |
c896fe29 | 1800 | |
c7482438 RH |
1801 | switch (ts->kind) { |
1802 | case TEMP_CONST: | |
f57c6915 | 1803 | case TEMP_TB: |
2f2e911d RH |
1804 | /* Silently ignore free. */ |
1805 | break; | |
1806 | case TEMP_EBB: | |
1807 | tcg_debug_assert(ts->temp_allocated != 0); | |
1808 | ts->temp_allocated = 0; | |
1809 | set_bit(temp_idx(ts), s->free_temps[ts->base_type].l); | |
c7482438 RH |
1810 | break; |
1811 | default: | |
2f2e911d | 1812 | /* It never made sense to free TEMP_FIXED or TEMP_GLOBAL. */ |
c7482438 | 1813 | g_assert_not_reached(); |
c0522136 | 1814 | } |
c896fe29 FB |
1815 | } |
1816 | ||
58b79713 RH |
1817 | void tcg_temp_free_i32(TCGv_i32 arg) |
1818 | { | |
1819 | tcg_temp_free_internal(tcgv_i32_temp(arg)); | |
1820 | } | |
1821 | ||
1822 | void tcg_temp_free_i64(TCGv_i64 arg) | |
1823 | { | |
1824 | tcg_temp_free_internal(tcgv_i64_temp(arg)); | |
1825 | } | |
1826 | ||
1827 | void tcg_temp_free_i128(TCGv_i128 arg) | |
1828 | { | |
1829 | tcg_temp_free_internal(tcgv_i128_temp(arg)); | |
1830 | } | |
1831 | ||
1832 | void tcg_temp_free_ptr(TCGv_ptr arg) | |
1833 | { | |
1834 | tcg_temp_free_internal(tcgv_ptr_temp(arg)); | |
1835 | } | |
1836 | ||
1837 | void tcg_temp_free_vec(TCGv_vec arg) | |
1838 | { | |
1839 | tcg_temp_free_internal(tcgv_vec_temp(arg)); | |
1840 | } | |
1841 | ||
c0522136 RH |
1842 | TCGTemp *tcg_constant_internal(TCGType type, int64_t val) |
1843 | { | |
1844 | TCGContext *s = tcg_ctx; | |
1845 | GHashTable *h = s->const_table[type]; | |
1846 | TCGTemp *ts; | |
1847 | ||
1848 | if (h == NULL) { | |
1849 | h = g_hash_table_new(g_int64_hash, g_int64_equal); | |
1850 | s->const_table[type] = h; | |
1851 | } | |
1852 | ||
1853 | ts = g_hash_table_lookup(h, &val); | |
1854 | if (ts == NULL) { | |
aef85402 RH |
1855 | int64_t *val_ptr; |
1856 | ||
c0522136 RH |
1857 | ts = tcg_temp_alloc(s); |
1858 | ||
1859 | if (TCG_TARGET_REG_BITS == 32 && type == TCG_TYPE_I64) { | |
1860 | TCGTemp *ts2 = tcg_temp_alloc(s); | |
1861 | ||
aef85402 RH |
1862 | tcg_debug_assert(ts2 == ts + 1); |
1863 | ||
c0522136 RH |
1864 | ts->base_type = TCG_TYPE_I64; |
1865 | ts->type = TCG_TYPE_I32; | |
1866 | ts->kind = TEMP_CONST; | |
1867 | ts->temp_allocated = 1; | |
c0522136 | 1868 | |
c0522136 RH |
1869 | ts2->base_type = TCG_TYPE_I64; |
1870 | ts2->type = TCG_TYPE_I32; | |
1871 | ts2->kind = TEMP_CONST; | |
1872 | ts2->temp_allocated = 1; | |
fac87bd2 | 1873 | ts2->temp_subindex = 1; |
aef85402 RH |
1874 | |
1875 | /* | |
1876 | * Retain the full value of the 64-bit constant in the low | |
1877 | * part, so that the hash table works. Actual uses will | |
1878 | * truncate the value to the low part. | |
1879 | */ | |
1880 | ts[HOST_BIG_ENDIAN].val = val; | |
1881 | ts[!HOST_BIG_ENDIAN].val = val >> 32; | |
1882 | val_ptr = &ts[HOST_BIG_ENDIAN].val; | |
c0522136 RH |
1883 | } else { |
1884 | ts->base_type = type; | |
1885 | ts->type = type; | |
1886 | ts->kind = TEMP_CONST; | |
1887 | ts->temp_allocated = 1; | |
1888 | ts->val = val; | |
aef85402 | 1889 | val_ptr = &ts->val; |
c0522136 | 1890 | } |
aef85402 | 1891 | g_hash_table_insert(h, val_ptr, ts); |
c0522136 RH |
1892 | } |
1893 | ||
1894 | return ts; | |
1895 | } | |
1896 | ||
16edaee7 RH |
1897 | TCGv_i32 tcg_constant_i32(int32_t val) |
1898 | { | |
1899 | return temp_tcgv_i32(tcg_constant_internal(TCG_TYPE_I32, val)); | |
1900 | } | |
1901 | ||
1902 | TCGv_i64 tcg_constant_i64(int64_t val) | |
1903 | { | |
1904 | return temp_tcgv_i64(tcg_constant_internal(TCG_TYPE_I64, val)); | |
1905 | } | |
1906 | ||
1907 | TCGv_ptr tcg_constant_ptr_int(intptr_t val) | |
1908 | { | |
1909 | return temp_tcgv_ptr(tcg_constant_internal(TCG_TYPE_PTR, val)); | |
1910 | } | |
1911 | ||
c0522136 RH |
1912 | TCGv_vec tcg_constant_vec(TCGType type, unsigned vece, int64_t val) |
1913 | { | |
1914 | val = dup_const(vece, val); | |
1915 | return temp_tcgv_vec(tcg_constant_internal(type, val)); | |
1916 | } | |
1917 | ||
88d4005b RH |
1918 | TCGv_vec tcg_constant_vec_matching(TCGv_vec match, unsigned vece, int64_t val) |
1919 | { | |
1920 | TCGTemp *t = tcgv_vec_temp(match); | |
1921 | ||
1922 | tcg_debug_assert(t->temp_allocated != 0); | |
1923 | return tcg_constant_vec(t->base_type, vece, val); | |
1924 | } | |
1925 | ||
177f648f RH |
1926 | #ifdef CONFIG_DEBUG_TCG |
1927 | size_t temp_idx(TCGTemp *ts) | |
1928 | { | |
1929 | ptrdiff_t n = ts - tcg_ctx->temps; | |
1930 | assert(n >= 0 && n < tcg_ctx->nb_temps); | |
1931 | return n; | |
1932 | } | |
1933 | ||
1934 | TCGTemp *tcgv_i32_temp(TCGv_i32 v) | |
1935 | { | |
1936 | uintptr_t o = (uintptr_t)v - offsetof(TCGContext, temps); | |
1937 | ||
1938 | assert(o < sizeof(TCGTemp) * tcg_ctx->nb_temps); | |
1939 | assert(o % sizeof(TCGTemp) == 0); | |
1940 | ||
1941 | return (void *)tcg_ctx + (uintptr_t)v; | |
1942 | } | |
1943 | #endif /* CONFIG_DEBUG_TCG */ | |
1944 | ||
be0f34b5 RH |
1945 | /* Return true if OP may appear in the opcode stream. |
1946 | Test the runtime variable that controls each opcode. */ | |
1947 | bool tcg_op_supported(TCGOpcode op) | |
1948 | { | |
d2fd745f RH |
1949 | const bool have_vec |
1950 | = TCG_TARGET_HAS_v64 | TCG_TARGET_HAS_v128 | TCG_TARGET_HAS_v256; | |
1951 | ||
be0f34b5 RH |
1952 | switch (op) { |
1953 | case INDEX_op_discard: | |
1954 | case INDEX_op_set_label: | |
1955 | case INDEX_op_call: | |
1956 | case INDEX_op_br: | |
1957 | case INDEX_op_mb: | |
1958 | case INDEX_op_insn_start: | |
1959 | case INDEX_op_exit_tb: | |
1960 | case INDEX_op_goto_tb: | |
f4e01e30 | 1961 | case INDEX_op_goto_ptr: |
fecccfcc RH |
1962 | case INDEX_op_qemu_ld_a32_i32: |
1963 | case INDEX_op_qemu_ld_a64_i32: | |
1964 | case INDEX_op_qemu_st_a32_i32: | |
1965 | case INDEX_op_qemu_st_a64_i32: | |
1966 | case INDEX_op_qemu_ld_a32_i64: | |
1967 | case INDEX_op_qemu_ld_a64_i64: | |
1968 | case INDEX_op_qemu_st_a32_i64: | |
1969 | case INDEX_op_qemu_st_a64_i64: | |
be0f34b5 RH |
1970 | return true; |
1971 | ||
fecccfcc RH |
1972 | case INDEX_op_qemu_st8_a32_i32: |
1973 | case INDEX_op_qemu_st8_a64_i32: | |
07ce0b05 RH |
1974 | return TCG_TARGET_HAS_qemu_st8_i32; |
1975 | ||
fecccfcc RH |
1976 | case INDEX_op_qemu_ld_a32_i128: |
1977 | case INDEX_op_qemu_ld_a64_i128: | |
1978 | case INDEX_op_qemu_st_a32_i128: | |
1979 | case INDEX_op_qemu_st_a64_i128: | |
12fde9bc RH |
1980 | return TCG_TARGET_HAS_qemu_ldst_i128; |
1981 | ||
be0f34b5 | 1982 | case INDEX_op_mov_i32: |
be0f34b5 RH |
1983 | case INDEX_op_setcond_i32: |
1984 | case INDEX_op_brcond_i32: | |
3871be75 | 1985 | case INDEX_op_movcond_i32: |
be0f34b5 RH |
1986 | case INDEX_op_ld8u_i32: |
1987 | case INDEX_op_ld8s_i32: | |
1988 | case INDEX_op_ld16u_i32: | |
1989 | case INDEX_op_ld16s_i32: | |
1990 | case INDEX_op_ld_i32: | |
1991 | case INDEX_op_st8_i32: | |
1992 | case INDEX_op_st16_i32: | |
1993 | case INDEX_op_st_i32: | |
1994 | case INDEX_op_add_i32: | |
1995 | case INDEX_op_sub_i32: | |
b701f195 | 1996 | case INDEX_op_neg_i32: |
be0f34b5 RH |
1997 | case INDEX_op_mul_i32: |
1998 | case INDEX_op_and_i32: | |
1999 | case INDEX_op_or_i32: | |
2000 | case INDEX_op_xor_i32: | |
2001 | case INDEX_op_shl_i32: | |
2002 | case INDEX_op_shr_i32: | |
2003 | case INDEX_op_sar_i32: | |
2004 | return true; | |
2005 | ||
3635502d RH |
2006 | case INDEX_op_negsetcond_i32: |
2007 | return TCG_TARGET_HAS_negsetcond_i32; | |
be0f34b5 RH |
2008 | case INDEX_op_div_i32: |
2009 | case INDEX_op_divu_i32: | |
2010 | return TCG_TARGET_HAS_div_i32; | |
2011 | case INDEX_op_rem_i32: | |
2012 | case INDEX_op_remu_i32: | |
2013 | return TCG_TARGET_HAS_rem_i32; | |
2014 | case INDEX_op_div2_i32: | |
2015 | case INDEX_op_divu2_i32: | |
2016 | return TCG_TARGET_HAS_div2_i32; | |
2017 | case INDEX_op_rotl_i32: | |
2018 | case INDEX_op_rotr_i32: | |
2019 | return TCG_TARGET_HAS_rot_i32; | |
2020 | case INDEX_op_deposit_i32: | |
2021 | return TCG_TARGET_HAS_deposit_i32; | |
2022 | case INDEX_op_extract_i32: | |
2023 | return TCG_TARGET_HAS_extract_i32; | |
2024 | case INDEX_op_sextract_i32: | |
2025 | return TCG_TARGET_HAS_sextract_i32; | |
fce1296f RH |
2026 | case INDEX_op_extract2_i32: |
2027 | return TCG_TARGET_HAS_extract2_i32; | |
be0f34b5 RH |
2028 | case INDEX_op_add2_i32: |
2029 | return TCG_TARGET_HAS_add2_i32; | |
2030 | case INDEX_op_sub2_i32: | |
2031 | return TCG_TARGET_HAS_sub2_i32; | |
2032 | case INDEX_op_mulu2_i32: | |
2033 | return TCG_TARGET_HAS_mulu2_i32; | |
2034 | case INDEX_op_muls2_i32: | |
2035 | return TCG_TARGET_HAS_muls2_i32; | |
2036 | case INDEX_op_muluh_i32: | |
2037 | return TCG_TARGET_HAS_muluh_i32; | |
2038 | case INDEX_op_mulsh_i32: | |
2039 | return TCG_TARGET_HAS_mulsh_i32; | |
2040 | case INDEX_op_ext8s_i32: | |
2041 | return TCG_TARGET_HAS_ext8s_i32; | |
2042 | case INDEX_op_ext16s_i32: | |
2043 | return TCG_TARGET_HAS_ext16s_i32; | |
2044 | case INDEX_op_ext8u_i32: | |
2045 | return TCG_TARGET_HAS_ext8u_i32; | |
2046 | case INDEX_op_ext16u_i32: | |
2047 | return TCG_TARGET_HAS_ext16u_i32; | |
2048 | case INDEX_op_bswap16_i32: | |
2049 | return TCG_TARGET_HAS_bswap16_i32; | |
2050 | case INDEX_op_bswap32_i32: | |
2051 | return TCG_TARGET_HAS_bswap32_i32; | |
2052 | case INDEX_op_not_i32: | |
2053 | return TCG_TARGET_HAS_not_i32; | |
be0f34b5 RH |
2054 | case INDEX_op_andc_i32: |
2055 | return TCG_TARGET_HAS_andc_i32; | |
2056 | case INDEX_op_orc_i32: | |
2057 | return TCG_TARGET_HAS_orc_i32; | |
2058 | case INDEX_op_eqv_i32: | |
2059 | return TCG_TARGET_HAS_eqv_i32; | |
2060 | case INDEX_op_nand_i32: | |
2061 | return TCG_TARGET_HAS_nand_i32; | |
2062 | case INDEX_op_nor_i32: | |
2063 | return TCG_TARGET_HAS_nor_i32; | |
2064 | case INDEX_op_clz_i32: | |
2065 | return TCG_TARGET_HAS_clz_i32; | |
2066 | case INDEX_op_ctz_i32: | |
2067 | return TCG_TARGET_HAS_ctz_i32; | |
2068 | case INDEX_op_ctpop_i32: | |
2069 | return TCG_TARGET_HAS_ctpop_i32; | |
2070 | ||
2071 | case INDEX_op_brcond2_i32: | |
2072 | case INDEX_op_setcond2_i32: | |
2073 | return TCG_TARGET_REG_BITS == 32; | |
2074 | ||
2075 | case INDEX_op_mov_i64: | |
be0f34b5 RH |
2076 | case INDEX_op_setcond_i64: |
2077 | case INDEX_op_brcond_i64: | |
3871be75 | 2078 | case INDEX_op_movcond_i64: |
be0f34b5 RH |
2079 | case INDEX_op_ld8u_i64: |
2080 | case INDEX_op_ld8s_i64: | |
2081 | case INDEX_op_ld16u_i64: | |
2082 | case INDEX_op_ld16s_i64: | |
2083 | case INDEX_op_ld32u_i64: | |
2084 | case INDEX_op_ld32s_i64: | |
2085 | case INDEX_op_ld_i64: | |
2086 | case INDEX_op_st8_i64: | |
2087 | case INDEX_op_st16_i64: | |
2088 | case INDEX_op_st32_i64: | |
2089 | case INDEX_op_st_i64: | |
2090 | case INDEX_op_add_i64: | |
2091 | case INDEX_op_sub_i64: | |
b701f195 | 2092 | case INDEX_op_neg_i64: |
be0f34b5 RH |
2093 | case INDEX_op_mul_i64: |
2094 | case INDEX_op_and_i64: | |
2095 | case INDEX_op_or_i64: | |
2096 | case INDEX_op_xor_i64: | |
2097 | case INDEX_op_shl_i64: | |
2098 | case INDEX_op_shr_i64: | |
2099 | case INDEX_op_sar_i64: | |
2100 | case INDEX_op_ext_i32_i64: | |
2101 | case INDEX_op_extu_i32_i64: | |
2102 | return TCG_TARGET_REG_BITS == 64; | |
2103 | ||
3635502d RH |
2104 | case INDEX_op_negsetcond_i64: |
2105 | return TCG_TARGET_HAS_negsetcond_i64; | |
be0f34b5 RH |
2106 | case INDEX_op_div_i64: |
2107 | case INDEX_op_divu_i64: | |
2108 | return TCG_TARGET_HAS_div_i64; | |
2109 | case INDEX_op_rem_i64: | |
2110 | case INDEX_op_remu_i64: | |
2111 | return TCG_TARGET_HAS_rem_i64; | |
2112 | case INDEX_op_div2_i64: | |
2113 | case INDEX_op_divu2_i64: | |
2114 | return TCG_TARGET_HAS_div2_i64; | |
2115 | case INDEX_op_rotl_i64: | |
2116 | case INDEX_op_rotr_i64: | |
2117 | return TCG_TARGET_HAS_rot_i64; | |
2118 | case INDEX_op_deposit_i64: | |
2119 | return TCG_TARGET_HAS_deposit_i64; | |
2120 | case INDEX_op_extract_i64: | |
2121 | return TCG_TARGET_HAS_extract_i64; | |
2122 | case INDEX_op_sextract_i64: | |
2123 | return TCG_TARGET_HAS_sextract_i64; | |
fce1296f RH |
2124 | case INDEX_op_extract2_i64: |
2125 | return TCG_TARGET_HAS_extract2_i64; | |
be0f34b5 | 2126 | case INDEX_op_extrl_i64_i32: |
be0f34b5 | 2127 | case INDEX_op_extrh_i64_i32: |
13d885b0 | 2128 | return TCG_TARGET_HAS_extr_i64_i32; |
be0f34b5 RH |
2129 | case INDEX_op_ext8s_i64: |
2130 | return TCG_TARGET_HAS_ext8s_i64; | |
2131 | case INDEX_op_ext16s_i64: | |
2132 | return TCG_TARGET_HAS_ext16s_i64; | |
2133 | case INDEX_op_ext32s_i64: | |
2134 | return TCG_TARGET_HAS_ext32s_i64; | |
2135 | case INDEX_op_ext8u_i64: | |
2136 | return TCG_TARGET_HAS_ext8u_i64; | |
2137 | case INDEX_op_ext16u_i64: | |
2138 | return TCG_TARGET_HAS_ext16u_i64; | |
2139 | case INDEX_op_ext32u_i64: | |
2140 | return TCG_TARGET_HAS_ext32u_i64; | |
2141 | case INDEX_op_bswap16_i64: | |
2142 | return TCG_TARGET_HAS_bswap16_i64; | |
2143 | case INDEX_op_bswap32_i64: | |
2144 | return TCG_TARGET_HAS_bswap32_i64; | |
2145 | case INDEX_op_bswap64_i64: | |
2146 | return TCG_TARGET_HAS_bswap64_i64; | |
2147 | case INDEX_op_not_i64: | |
2148 | return TCG_TARGET_HAS_not_i64; | |
be0f34b5 RH |
2149 | case INDEX_op_andc_i64: |
2150 | return TCG_TARGET_HAS_andc_i64; | |
2151 | case INDEX_op_orc_i64: | |
2152 | return TCG_TARGET_HAS_orc_i64; | |
2153 | case INDEX_op_eqv_i64: | |
2154 | return TCG_TARGET_HAS_eqv_i64; | |
2155 | case INDEX_op_nand_i64: | |
2156 | return TCG_TARGET_HAS_nand_i64; | |
2157 | case INDEX_op_nor_i64: | |
2158 | return TCG_TARGET_HAS_nor_i64; | |
2159 | case INDEX_op_clz_i64: | |
2160 | return TCG_TARGET_HAS_clz_i64; | |
2161 | case INDEX_op_ctz_i64: | |
2162 | return TCG_TARGET_HAS_ctz_i64; | |
2163 | case INDEX_op_ctpop_i64: | |
2164 | return TCG_TARGET_HAS_ctpop_i64; | |
2165 | case INDEX_op_add2_i64: | |
2166 | return TCG_TARGET_HAS_add2_i64; | |
2167 | case INDEX_op_sub2_i64: | |
2168 | return TCG_TARGET_HAS_sub2_i64; | |
2169 | case INDEX_op_mulu2_i64: | |
2170 | return TCG_TARGET_HAS_mulu2_i64; | |
2171 | case INDEX_op_muls2_i64: | |
2172 | return TCG_TARGET_HAS_muls2_i64; | |
2173 | case INDEX_op_muluh_i64: | |
2174 | return TCG_TARGET_HAS_muluh_i64; | |
2175 | case INDEX_op_mulsh_i64: | |
2176 | return TCG_TARGET_HAS_mulsh_i64; | |
2177 | ||
d2fd745f RH |
2178 | case INDEX_op_mov_vec: |
2179 | case INDEX_op_dup_vec: | |
37ee55a0 | 2180 | case INDEX_op_dupm_vec: |
d2fd745f RH |
2181 | case INDEX_op_ld_vec: |
2182 | case INDEX_op_st_vec: | |
2183 | case INDEX_op_add_vec: | |
2184 | case INDEX_op_sub_vec: | |
2185 | case INDEX_op_and_vec: | |
2186 | case INDEX_op_or_vec: | |
2187 | case INDEX_op_xor_vec: | |
212be173 | 2188 | case INDEX_op_cmp_vec: |
d2fd745f RH |
2189 | return have_vec; |
2190 | case INDEX_op_dup2_vec: | |
2191 | return have_vec && TCG_TARGET_REG_BITS == 32; | |
2192 | case INDEX_op_not_vec: | |
2193 | return have_vec && TCG_TARGET_HAS_not_vec; | |
2194 | case INDEX_op_neg_vec: | |
2195 | return have_vec && TCG_TARGET_HAS_neg_vec; | |
bcefc902 RH |
2196 | case INDEX_op_abs_vec: |
2197 | return have_vec && TCG_TARGET_HAS_abs_vec; | |
d2fd745f RH |
2198 | case INDEX_op_andc_vec: |
2199 | return have_vec && TCG_TARGET_HAS_andc_vec; | |
2200 | case INDEX_op_orc_vec: | |
2201 | return have_vec && TCG_TARGET_HAS_orc_vec; | |
ed523473 RH |
2202 | case INDEX_op_nand_vec: |
2203 | return have_vec && TCG_TARGET_HAS_nand_vec; | |
2204 | case INDEX_op_nor_vec: | |
2205 | return have_vec && TCG_TARGET_HAS_nor_vec; | |
2206 | case INDEX_op_eqv_vec: | |
2207 | return have_vec && TCG_TARGET_HAS_eqv_vec; | |
3774030a RH |
2208 | case INDEX_op_mul_vec: |
2209 | return have_vec && TCG_TARGET_HAS_mul_vec; | |
d0ec9796 RH |
2210 | case INDEX_op_shli_vec: |
2211 | case INDEX_op_shri_vec: | |
2212 | case INDEX_op_sari_vec: | |
2213 | return have_vec && TCG_TARGET_HAS_shi_vec; | |
2214 | case INDEX_op_shls_vec: | |
2215 | case INDEX_op_shrs_vec: | |
2216 | case INDEX_op_sars_vec: | |
2217 | return have_vec && TCG_TARGET_HAS_shs_vec; | |
2218 | case INDEX_op_shlv_vec: | |
2219 | case INDEX_op_shrv_vec: | |
2220 | case INDEX_op_sarv_vec: | |
2221 | return have_vec && TCG_TARGET_HAS_shv_vec; | |
b0f7e744 RH |
2222 | case INDEX_op_rotli_vec: |
2223 | return have_vec && TCG_TARGET_HAS_roti_vec; | |
23850a74 RH |
2224 | case INDEX_op_rotls_vec: |
2225 | return have_vec && TCG_TARGET_HAS_rots_vec; | |
5d0ceda9 RH |
2226 | case INDEX_op_rotlv_vec: |
2227 | case INDEX_op_rotrv_vec: | |
2228 | return have_vec && TCG_TARGET_HAS_rotv_vec; | |
8afaf050 RH |
2229 | case INDEX_op_ssadd_vec: |
2230 | case INDEX_op_usadd_vec: | |
2231 | case INDEX_op_sssub_vec: | |
2232 | case INDEX_op_ussub_vec: | |
2233 | return have_vec && TCG_TARGET_HAS_sat_vec; | |
dd0a0fcd RH |
2234 | case INDEX_op_smin_vec: |
2235 | case INDEX_op_umin_vec: | |
2236 | case INDEX_op_smax_vec: | |
2237 | case INDEX_op_umax_vec: | |
2238 | return have_vec && TCG_TARGET_HAS_minmax_vec; | |
38dc1294 RH |
2239 | case INDEX_op_bitsel_vec: |
2240 | return have_vec && TCG_TARGET_HAS_bitsel_vec; | |
f75da298 RH |
2241 | case INDEX_op_cmpsel_vec: |
2242 | return have_vec && TCG_TARGET_HAS_cmpsel_vec; | |
d2fd745f | 2243 | |
db432672 RH |
2244 | default: |
2245 | tcg_debug_assert(op > INDEX_op_last_generic && op < NB_OPS); | |
2246 | return true; | |
be0f34b5 | 2247 | } |
be0f34b5 RH |
2248 | } |
2249 | ||
39004a71 RH |
2250 | static TCGOp *tcg_op_alloc(TCGOpcode opc, unsigned nargs); |
2251 | ||
a3a692b8 | 2252 | static void tcg_gen_callN(TCGHelperInfo *info, TCGTemp *ret, TCGTemp **args) |
c896fe29 | 2253 | { |
39004a71 RH |
2254 | TCGv_i64 extend_free[MAX_CALL_IARGS]; |
2255 | int n_extend = 0; | |
75e8b9b7 | 2256 | TCGOp *op; |
39004a71 | 2257 | int i, n, pi = 0, total_args; |
afb49896 | 2258 | |
d53106c9 RH |
2259 | if (unlikely(g_once_init_enter(HELPER_INFO_INIT(info)))) { |
2260 | init_call_layout(info); | |
2261 | g_once_init_leave(HELPER_INFO_INIT(info), HELPER_INFO_INIT_VAL(info)); | |
2262 | } | |
2263 | ||
39004a71 RH |
2264 | total_args = info->nr_out + info->nr_in + 2; |
2265 | op = tcg_op_alloc(INDEX_op_call, total_args); | |
2bece2c8 | 2266 | |
38b47b19 | 2267 | #ifdef CONFIG_PLUGIN |
17083f6f EC |
2268 | /* Flag helpers that may affect guest state */ |
2269 | if (tcg_ctx->plugin_insn && | |
2270 | !(info->flags & TCG_CALL_PLUGIN) && | |
2271 | !(info->flags & TCG_CALL_NO_SIDE_EFFECTS)) { | |
38b47b19 EC |
2272 | tcg_ctx->plugin_insn->calls_helpers = true; |
2273 | } | |
2274 | #endif | |
2275 | ||
39004a71 RH |
2276 | TCGOP_CALLO(op) = n = info->nr_out; |
2277 | switch (n) { | |
2278 | case 0: | |
2279 | tcg_debug_assert(ret == NULL); | |
2280 | break; | |
2281 | case 1: | |
2282 | tcg_debug_assert(ret != NULL); | |
2283 | op->args[pi++] = temp_arg(ret); | |
2284 | break; | |
2285 | case 2: | |
466d3759 | 2286 | case 4: |
39004a71 | 2287 | tcg_debug_assert(ret != NULL); |
466d3759 | 2288 | tcg_debug_assert(ret->base_type == ret->type + ctz32(n)); |
39004a71 | 2289 | tcg_debug_assert(ret->temp_subindex == 0); |
466d3759 RH |
2290 | for (i = 0; i < n; ++i) { |
2291 | op->args[pi++] = temp_arg(ret + i); | |
2292 | } | |
39004a71 RH |
2293 | break; |
2294 | default: | |
2295 | g_assert_not_reached(); | |
2296 | } | |
2297 | ||
2298 | TCGOP_CALLI(op) = n = info->nr_in; | |
2299 | for (i = 0; i < n; i++) { | |
2300 | const TCGCallArgumentLoc *loc = &info->in[i]; | |
2301 | TCGTemp *ts = args[loc->arg_idx] + loc->tmp_subindex; | |
2302 | ||
2303 | switch (loc->kind) { | |
2304 | case TCG_CALL_ARG_NORMAL: | |
313bdea8 RH |
2305 | case TCG_CALL_ARG_BY_REF: |
2306 | case TCG_CALL_ARG_BY_REF_N: | |
39004a71 RH |
2307 | op->args[pi++] = temp_arg(ts); |
2308 | break; | |
eb8b0224 | 2309 | |
39004a71 RH |
2310 | case TCG_CALL_ARG_EXTEND_U: |
2311 | case TCG_CALL_ARG_EXTEND_S: | |
2312 | { | |
5dd48602 | 2313 | TCGv_i64 temp = tcg_temp_ebb_new_i64(); |
39004a71 RH |
2314 | TCGv_i32 orig = temp_tcgv_i32(ts); |
2315 | ||
2316 | if (loc->kind == TCG_CALL_ARG_EXTEND_S) { | |
eb8b0224 RH |
2317 | tcg_gen_ext_i32_i64(temp, orig); |
2318 | } else { | |
2319 | tcg_gen_extu_i32_i64(temp, orig); | |
2320 | } | |
39004a71 RH |
2321 | op->args[pi++] = tcgv_i64_arg(temp); |
2322 | extend_free[n_extend++] = temp; | |
2bece2c8 | 2323 | } |
e2a9dd6b | 2324 | break; |
7b7d8b2d | 2325 | |
e2a9dd6b RH |
2326 | default: |
2327 | g_assert_not_reached(); | |
c896fe29 FB |
2328 | } |
2329 | } | |
d53106c9 | 2330 | op->args[pi++] = (uintptr_t)info->func; |
3e92aa34 | 2331 | op->args[pi++] = (uintptr_t)info; |
39004a71 | 2332 | tcg_debug_assert(pi == total_args); |
a7812ae4 | 2333 | |
39004a71 | 2334 | QTAILQ_INSERT_TAIL(&tcg_ctx->ops, op, link); |
7319d83a | 2335 | |
39004a71 RH |
2336 | tcg_debug_assert(n_extend < ARRAY_SIZE(extend_free)); |
2337 | for (i = 0; i < n_extend; ++i) { | |
2338 | tcg_temp_free_i64(extend_free[i]); | |
2bece2c8 | 2339 | } |
c896fe29 | 2340 | } |
c896fe29 | 2341 | |
a3a692b8 RH |
2342 | void tcg_gen_call0(TCGHelperInfo *info, TCGTemp *ret) |
2343 | { | |
2344 | tcg_gen_callN(info, ret, NULL); | |
2345 | } | |
2346 | ||
2347 | void tcg_gen_call1(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1) | |
2348 | { | |
2349 | tcg_gen_callN(info, ret, &t1); | |
2350 | } | |
2351 | ||
2352 | void tcg_gen_call2(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, TCGTemp *t2) | |
2353 | { | |
2354 | TCGTemp *args[2] = { t1, t2 }; | |
2355 | tcg_gen_callN(info, ret, args); | |
2356 | } | |
2357 | ||
2358 | void tcg_gen_call3(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, | |
2359 | TCGTemp *t2, TCGTemp *t3) | |
2360 | { | |
2361 | TCGTemp *args[3] = { t1, t2, t3 }; | |
2362 | tcg_gen_callN(info, ret, args); | |
2363 | } | |
2364 | ||
2365 | void tcg_gen_call4(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, | |
2366 | TCGTemp *t2, TCGTemp *t3, TCGTemp *t4) | |
2367 | { | |
2368 | TCGTemp *args[4] = { t1, t2, t3, t4 }; | |
2369 | tcg_gen_callN(info, ret, args); | |
2370 | } | |
2371 | ||
2372 | void tcg_gen_call5(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, | |
2373 | TCGTemp *t2, TCGTemp *t3, TCGTemp *t4, TCGTemp *t5) | |
2374 | { | |
2375 | TCGTemp *args[5] = { t1, t2, t3, t4, t5 }; | |
2376 | tcg_gen_callN(info, ret, args); | |
2377 | } | |
2378 | ||
2379 | void tcg_gen_call6(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, TCGTemp *t2, | |
2380 | TCGTemp *t3, TCGTemp *t4, TCGTemp *t5, TCGTemp *t6) | |
2381 | { | |
2382 | TCGTemp *args[6] = { t1, t2, t3, t4, t5, t6 }; | |
2383 | tcg_gen_callN(info, ret, args); | |
2384 | } | |
2385 | ||
2386 | void tcg_gen_call7(TCGHelperInfo *info, TCGTemp *ret, TCGTemp *t1, | |
2387 | TCGTemp *t2, TCGTemp *t3, TCGTemp *t4, | |
2388 | TCGTemp *t5, TCGTemp *t6, TCGTemp *t7) | |
2389 | { | |
2390 | TCGTemp *args[7] = { t1, t2, t3, t4, t5, t6, t7 }; | |
2391 | tcg_gen_callN(info, ret, args); | |
2392 | } | |
2393 | ||
8fcd3692 | 2394 | static void tcg_reg_alloc_start(TCGContext *s) |
c896fe29 | 2395 | { |
ac3b8891 | 2396 | int i, n; |
ac3b8891 | 2397 | |
ee17db83 RH |
2398 | for (i = 0, n = s->nb_temps; i < n; i++) { |
2399 | TCGTemp *ts = &s->temps[i]; | |
2400 | TCGTempVal val = TEMP_VAL_MEM; | |
2401 | ||
2402 | switch (ts->kind) { | |
c0522136 RH |
2403 | case TEMP_CONST: |
2404 | val = TEMP_VAL_CONST; | |
2405 | break; | |
ee17db83 RH |
2406 | case TEMP_FIXED: |
2407 | val = TEMP_VAL_REG; | |
2408 | break; | |
2409 | case TEMP_GLOBAL: | |
2410 | break; | |
c7482438 | 2411 | case TEMP_EBB: |
ee17db83 RH |
2412 | val = TEMP_VAL_DEAD; |
2413 | /* fall through */ | |
f57c6915 | 2414 | case TEMP_TB: |
ee17db83 RH |
2415 | ts->mem_allocated = 0; |
2416 | break; | |
2417 | default: | |
2418 | g_assert_not_reached(); | |
2419 | } | |
2420 | ts->val_type = val; | |
e8996ee0 | 2421 | } |
f8b2f202 RH |
2422 | |
2423 | memset(s->reg_to_temp, 0, sizeof(s->reg_to_temp)); | |
c896fe29 FB |
2424 | } |
2425 | ||
f8b2f202 RH |
2426 | static char *tcg_get_arg_str_ptr(TCGContext *s, char *buf, int buf_size, |
2427 | TCGTemp *ts) | |
c896fe29 | 2428 | { |
1807f4c4 | 2429 | int idx = temp_idx(ts); |
ac56dd48 | 2430 | |
ee17db83 RH |
2431 | switch (ts->kind) { |
2432 | case TEMP_FIXED: | |
2433 | case TEMP_GLOBAL: | |
ac56dd48 | 2434 | pstrcpy(buf, buf_size, ts->name); |
ee17db83 | 2435 | break; |
f57c6915 | 2436 | case TEMP_TB: |
f8b2f202 | 2437 | snprintf(buf, buf_size, "loc%d", idx - s->nb_globals); |
ee17db83 | 2438 | break; |
c7482438 | 2439 | case TEMP_EBB: |
f8b2f202 | 2440 | snprintf(buf, buf_size, "tmp%d", idx - s->nb_globals); |
ee17db83 | 2441 | break; |
c0522136 RH |
2442 | case TEMP_CONST: |
2443 | switch (ts->type) { | |
2444 | case TCG_TYPE_I32: | |
2445 | snprintf(buf, buf_size, "$0x%x", (int32_t)ts->val); | |
2446 | break; | |
2447 | #if TCG_TARGET_REG_BITS > 32 | |
2448 | case TCG_TYPE_I64: | |
2449 | snprintf(buf, buf_size, "$0x%" PRIx64, ts->val); | |
2450 | break; | |
2451 | #endif | |
2452 | case TCG_TYPE_V64: | |
2453 | case TCG_TYPE_V128: | |
2454 | case TCG_TYPE_V256: | |
2455 | snprintf(buf, buf_size, "v%d$0x%" PRIx64, | |
2456 | 64 << (ts->type - TCG_TYPE_V64), ts->val); | |
2457 | break; | |
2458 | default: | |
2459 | g_assert_not_reached(); | |
2460 | } | |
2461 | break; | |
c896fe29 FB |
2462 | } |
2463 | return buf; | |
2464 | } | |
2465 | ||
43439139 RH |
2466 | static char *tcg_get_arg_str(TCGContext *s, char *buf, |
2467 | int buf_size, TCGArg arg) | |
f8b2f202 | 2468 | { |
43439139 | 2469 | return tcg_get_arg_str_ptr(s, buf, buf_size, arg_temp(arg)); |
f8b2f202 RH |
2470 | } |
2471 | ||
f48f3ede BS |
2472 | static const char * const cond_name[] = |
2473 | { | |
0aed257f RH |
2474 | [TCG_COND_NEVER] = "never", |
2475 | [TCG_COND_ALWAYS] = "always", | |
f48f3ede BS |
2476 | [TCG_COND_EQ] = "eq", |
2477 | [TCG_COND_NE] = "ne", | |
2478 | [TCG_COND_LT] = "lt", | |
2479 | [TCG_COND_GE] = "ge", | |
2480 | [TCG_COND_LE] = "le", | |
2481 | [TCG_COND_GT] = "gt", | |
2482 | [TCG_COND_LTU] = "ltu", | |
2483 | [TCG_COND_GEU] = "geu", | |
2484 | [TCG_COND_LEU] = "leu", | |
2485 | [TCG_COND_GTU] = "gtu" | |
2486 | }; | |
2487 | ||
12fde9bc | 2488 | static const char * const ldst_name[(MO_BSWAP | MO_SSIZE) + 1] = |
f713d6ad RH |
2489 | { |
2490 | [MO_UB] = "ub", | |
2491 | [MO_SB] = "sb", | |
2492 | [MO_LEUW] = "leuw", | |
2493 | [MO_LESW] = "lesw", | |
2494 | [MO_LEUL] = "leul", | |
2495 | [MO_LESL] = "lesl", | |
fc313c64 | 2496 | [MO_LEUQ] = "leq", |
f713d6ad RH |
2497 | [MO_BEUW] = "beuw", |
2498 | [MO_BESW] = "besw", | |
2499 | [MO_BEUL] = "beul", | |
2500 | [MO_BESL] = "besl", | |
fc313c64 | 2501 | [MO_BEUQ] = "beq", |
12fde9bc RH |
2502 | [MO_128 + MO_BE] = "beo", |
2503 | [MO_128 + MO_LE] = "leo", | |
f713d6ad RH |
2504 | }; |
2505 | ||
1f00b27f | 2506 | static const char * const alignment_name[(MO_AMASK >> MO_ASHIFT) + 1] = { |
1f00b27f | 2507 | [MO_UNALN >> MO_ASHIFT] = "un+", |
1f00b27f | 2508 | [MO_ALIGN >> MO_ASHIFT] = "al+", |
1f00b27f SS |
2509 | [MO_ALIGN_2 >> MO_ASHIFT] = "al2+", |
2510 | [MO_ALIGN_4 >> MO_ASHIFT] = "al4+", | |
2511 | [MO_ALIGN_8 >> MO_ASHIFT] = "al8+", | |
2512 | [MO_ALIGN_16 >> MO_ASHIFT] = "al16+", | |
2513 | [MO_ALIGN_32 >> MO_ASHIFT] = "al32+", | |
2514 | [MO_ALIGN_64 >> MO_ASHIFT] = "al64+", | |
2515 | }; | |
2516 | ||
37031fef RH |
2517 | static const char * const atom_name[(MO_ATOM_MASK >> MO_ATOM_SHIFT) + 1] = { |
2518 | [MO_ATOM_IFALIGN >> MO_ATOM_SHIFT] = "", | |
2519 | [MO_ATOM_IFALIGN_PAIR >> MO_ATOM_SHIFT] = "pair+", | |
2520 | [MO_ATOM_WITHIN16 >> MO_ATOM_SHIFT] = "w16+", | |
2521 | [MO_ATOM_WITHIN16_PAIR >> MO_ATOM_SHIFT] = "w16p+", | |
2522 | [MO_ATOM_SUBALIGN >> MO_ATOM_SHIFT] = "sub+", | |
2523 | [MO_ATOM_NONE >> MO_ATOM_SHIFT] = "noat+", | |
2524 | }; | |
2525 | ||
587195bd RH |
2526 | static const char bswap_flag_name[][6] = { |
2527 | [TCG_BSWAP_IZ] = "iz", | |
2528 | [TCG_BSWAP_OZ] = "oz", | |
2529 | [TCG_BSWAP_OS] = "os", | |
2530 | [TCG_BSWAP_IZ | TCG_BSWAP_OZ] = "iz,oz", | |
2531 | [TCG_BSWAP_IZ | TCG_BSWAP_OS] = "iz,os", | |
2532 | }; | |
2533 | ||
b016486e RH |
2534 | static inline bool tcg_regset_single(TCGRegSet d) |
2535 | { | |
2536 | return (d & (d - 1)) == 0; | |
2537 | } | |
2538 | ||
2539 | static inline TCGReg tcg_regset_first(TCGRegSet d) | |
2540 | { | |
2541 | if (TCG_TARGET_NB_REGS <= 32) { | |
2542 | return ctz32(d); | |
2543 | } else { | |
2544 | return ctz64(d); | |
2545 | } | |
2546 | } | |
2547 | ||
b7a83ff8 RH |
2548 | /* Return only the number of characters output -- no error return. */ |
2549 | #define ne_fprintf(...) \ | |
2550 | ({ int ret_ = fprintf(__VA_ARGS__); ret_ >= 0 ? ret_ : 0; }) | |
2551 | ||
2552 | static void tcg_dump_ops(TCGContext *s, FILE *f, bool have_prefs) | |
c896fe29 | 2553 | { |
c896fe29 | 2554 | char buf[128]; |
c45cb8bb | 2555 | TCGOp *op; |
c45cb8bb | 2556 | |
15fa08f8 | 2557 | QTAILQ_FOREACH(op, &s->ops, link) { |
c45cb8bb RH |
2558 | int i, k, nb_oargs, nb_iargs, nb_cargs; |
2559 | const TCGOpDef *def; | |
c45cb8bb | 2560 | TCGOpcode c; |
bdfb460e | 2561 | int col = 0; |
c896fe29 | 2562 | |
c45cb8bb | 2563 | c = op->opc; |
c896fe29 | 2564 | def = &tcg_op_defs[c]; |
c45cb8bb | 2565 | |
765b842a | 2566 | if (c == INDEX_op_insn_start) { |
b016486e | 2567 | nb_oargs = 0; |
b7a83ff8 | 2568 | col += ne_fprintf(f, "\n ----"); |
9aef40ed | 2569 | |
747bd69d | 2570 | for (i = 0, k = s->insn_start_words; i < k; ++i) { |
c9ad8d27 RH |
2571 | col += ne_fprintf(f, " %016" PRIx64, |
2572 | tcg_get_insn_start_param(op, i)); | |
eeacee4d | 2573 | } |
7e4597d7 | 2574 | } else if (c == INDEX_op_call) { |
3e92aa34 | 2575 | const TCGHelperInfo *info = tcg_call_info(op); |
fa52e660 | 2576 | void *func = tcg_call_func(op); |
3e92aa34 | 2577 | |
c896fe29 | 2578 | /* variable number of arguments */ |
cd9090aa RH |
2579 | nb_oargs = TCGOP_CALLO(op); |
2580 | nb_iargs = TCGOP_CALLI(op); | |
c896fe29 | 2581 | nb_cargs = def->nb_cargs; |
c896fe29 | 2582 | |
b7a83ff8 | 2583 | col += ne_fprintf(f, " %s ", def->name); |
3e92aa34 RH |
2584 | |
2585 | /* | |
2586 | * Print the function name from TCGHelperInfo, if available. | |
2587 | * Note that plugins have a template function for the info, | |
2588 | * but the actual function pointer comes from the plugin. | |
2589 | */ | |
3e92aa34 | 2590 | if (func == info->func) { |
b7a83ff8 | 2591 | col += ne_fprintf(f, "%s", info->name); |
3e92aa34 | 2592 | } else { |
b7a83ff8 | 2593 | col += ne_fprintf(f, "plugin(%p)", func); |
3e92aa34 RH |
2594 | } |
2595 | ||
b7a83ff8 | 2596 | col += ne_fprintf(f, ",$0x%x,$%d", info->flags, nb_oargs); |
cf066674 | 2597 | for (i = 0; i < nb_oargs; i++) { |
b7a83ff8 RH |
2598 | col += ne_fprintf(f, ",%s", tcg_get_arg_str(s, buf, sizeof(buf), |
2599 | op->args[i])); | |
b03cce8e | 2600 | } |
cf066674 | 2601 | for (i = 0; i < nb_iargs; i++) { |
efee3746 | 2602 | TCGArg arg = op->args[nb_oargs + i]; |
39004a71 | 2603 | const char *t = tcg_get_arg_str(s, buf, sizeof(buf), arg); |
b7a83ff8 | 2604 | col += ne_fprintf(f, ",%s", t); |
e8996ee0 | 2605 | } |
b03cce8e | 2606 | } else { |
b7a83ff8 | 2607 | col += ne_fprintf(f, " %s ", def->name); |
c45cb8bb RH |
2608 | |
2609 | nb_oargs = def->nb_oargs; | |
2610 | nb_iargs = def->nb_iargs; | |
2611 | nb_cargs = def->nb_cargs; | |
2612 | ||
d2fd745f | 2613 | if (def->flags & TCG_OPF_VECTOR) { |
b7a83ff8 RH |
2614 | col += ne_fprintf(f, "v%d,e%d,", 64 << TCGOP_VECL(op), |
2615 | 8 << TCGOP_VECE(op)); | |
d2fd745f RH |
2616 | } |
2617 | ||
b03cce8e | 2618 | k = 0; |
c45cb8bb | 2619 | for (i = 0; i < nb_oargs; i++) { |
b7a83ff8 RH |
2620 | const char *sep = k ? "," : ""; |
2621 | col += ne_fprintf(f, "%s%s", sep, | |
2622 | tcg_get_arg_str(s, buf, sizeof(buf), | |
2623 | op->args[k++])); | |
b03cce8e | 2624 | } |
c45cb8bb | 2625 | for (i = 0; i < nb_iargs; i++) { |
b7a83ff8 RH |
2626 | const char *sep = k ? "," : ""; |
2627 | col += ne_fprintf(f, "%s%s", sep, | |
2628 | tcg_get_arg_str(s, buf, sizeof(buf), | |
2629 | op->args[k++])); | |
b03cce8e | 2630 | } |
be210acb RH |
2631 | switch (c) { |
2632 | case INDEX_op_brcond_i32: | |
be210acb | 2633 | case INDEX_op_setcond_i32: |
3635502d | 2634 | case INDEX_op_negsetcond_i32: |
ffc5ea09 | 2635 | case INDEX_op_movcond_i32: |
ffc5ea09 | 2636 | case INDEX_op_brcond2_i32: |
be210acb | 2637 | case INDEX_op_setcond2_i32: |
ffc5ea09 | 2638 | case INDEX_op_brcond_i64: |
be210acb | 2639 | case INDEX_op_setcond_i64: |
3635502d | 2640 | case INDEX_op_negsetcond_i64: |
ffc5ea09 | 2641 | case INDEX_op_movcond_i64: |
212be173 | 2642 | case INDEX_op_cmp_vec: |
f75da298 | 2643 | case INDEX_op_cmpsel_vec: |
efee3746 RH |
2644 | if (op->args[k] < ARRAY_SIZE(cond_name) |
2645 | && cond_name[op->args[k]]) { | |
b7a83ff8 | 2646 | col += ne_fprintf(f, ",%s", cond_name[op->args[k++]]); |
eeacee4d | 2647 | } else { |
b7a83ff8 | 2648 | col += ne_fprintf(f, ",$0x%" TCG_PRIlx, op->args[k++]); |
eeacee4d | 2649 | } |
f48f3ede | 2650 | i = 1; |
be210acb | 2651 | break; |
fecccfcc RH |
2652 | case INDEX_op_qemu_ld_a32_i32: |
2653 | case INDEX_op_qemu_ld_a64_i32: | |
2654 | case INDEX_op_qemu_st_a32_i32: | |
2655 | case INDEX_op_qemu_st_a64_i32: | |
2656 | case INDEX_op_qemu_st8_a32_i32: | |
2657 | case INDEX_op_qemu_st8_a64_i32: | |
2658 | case INDEX_op_qemu_ld_a32_i64: | |
2659 | case INDEX_op_qemu_ld_a64_i64: | |
2660 | case INDEX_op_qemu_st_a32_i64: | |
2661 | case INDEX_op_qemu_st_a64_i64: | |
2662 | case INDEX_op_qemu_ld_a32_i128: | |
2663 | case INDEX_op_qemu_ld_a64_i128: | |
2664 | case INDEX_op_qemu_st_a32_i128: | |
2665 | case INDEX_op_qemu_st_a64_i128: | |
59227d5d | 2666 | { |
37031fef | 2667 | const char *s_al, *s_op, *s_at; |
9002ffcb | 2668 | MemOpIdx oi = op->args[k++]; |
9a239c6e | 2669 | MemOp mop = get_memop(oi); |
59227d5d RH |
2670 | unsigned ix = get_mmuidx(oi); |
2671 | ||
9a239c6e PMD |
2672 | s_al = alignment_name[(mop & MO_AMASK) >> MO_ASHIFT]; |
2673 | s_op = ldst_name[mop & (MO_BSWAP | MO_SSIZE)]; | |
2674 | s_at = atom_name[(mop & MO_ATOM_MASK) >> MO_ATOM_SHIFT]; | |
2675 | mop &= ~(MO_AMASK | MO_BSWAP | MO_SSIZE | MO_ATOM_MASK); | |
37031fef RH |
2676 | |
2677 | /* If all fields are accounted for, print symbolically. */ | |
9a239c6e | 2678 | if (!mop && s_al && s_op && s_at) { |
37031fef RH |
2679 | col += ne_fprintf(f, ",%s%s%s,%u", |
2680 | s_at, s_al, s_op, ix); | |
59c4b7e8 | 2681 | } else { |
9a239c6e PMD |
2682 | mop = get_memop(oi); |
2683 | col += ne_fprintf(f, ",$0x%x,%u", mop, ix); | |
59227d5d RH |
2684 | } |
2685 | i = 1; | |
f713d6ad | 2686 | } |
f713d6ad | 2687 | break; |
587195bd RH |
2688 | case INDEX_op_bswap16_i32: |
2689 | case INDEX_op_bswap16_i64: | |
2690 | case INDEX_op_bswap32_i32: | |
2691 | case INDEX_op_bswap32_i64: | |
2692 | case INDEX_op_bswap64_i64: | |
2693 | { | |
2694 | TCGArg flags = op->args[k]; | |
2695 | const char *name = NULL; | |
2696 | ||
2697 | if (flags < ARRAY_SIZE(bswap_flag_name)) { | |
2698 | name = bswap_flag_name[flags]; | |
2699 | } | |
2700 | if (name) { | |
b7a83ff8 | 2701 | col += ne_fprintf(f, ",%s", name); |
587195bd | 2702 | } else { |
b7a83ff8 | 2703 | col += ne_fprintf(f, ",$0x%" TCG_PRIlx, flags); |
587195bd RH |
2704 | } |
2705 | i = k = 1; | |
2706 | } | |
2707 | break; | |
be210acb | 2708 | default: |
f48f3ede | 2709 | i = 0; |
be210acb RH |
2710 | break; |
2711 | } | |
51e3972c RH |
2712 | switch (c) { |
2713 | case INDEX_op_set_label: | |
2714 | case INDEX_op_br: | |
2715 | case INDEX_op_brcond_i32: | |
2716 | case INDEX_op_brcond_i64: | |
2717 | case INDEX_op_brcond2_i32: | |
b7a83ff8 RH |
2718 | col += ne_fprintf(f, "%s$L%d", k ? "," : "", |
2719 | arg_label(op->args[k])->id); | |
51e3972c RH |
2720 | i++, k++; |
2721 | break; | |
3470867b RH |
2722 | case INDEX_op_mb: |
2723 | { | |
2724 | TCGBar membar = op->args[k]; | |
2725 | const char *b_op, *m_op; | |
2726 | ||
2727 | switch (membar & TCG_BAR_SC) { | |
2728 | case 0: | |
2729 | b_op = "none"; | |
2730 | break; | |
2731 | case TCG_BAR_LDAQ: | |
2732 | b_op = "acq"; | |
2733 | break; | |
2734 | case TCG_BAR_STRL: | |
2735 | b_op = "rel"; | |
2736 | break; | |
2737 | case TCG_BAR_SC: | |
2738 | b_op = "seq"; | |
2739 | break; | |
2740 | default: | |
2741 | g_assert_not_reached(); | |
2742 | } | |
2743 | ||
2744 | switch (membar & TCG_MO_ALL) { | |
2745 | case 0: | |
2746 | m_op = "none"; | |
2747 | break; | |
2748 | case TCG_MO_LD_LD: | |
2749 | m_op = "rr"; | |
2750 | break; | |
2751 | case TCG_MO_LD_ST: | |
2752 | m_op = "rw"; | |
2753 | break; | |
2754 | case TCG_MO_ST_LD: | |
2755 | m_op = "wr"; | |
2756 | break; | |
2757 | case TCG_MO_ST_ST: | |
2758 | m_op = "ww"; | |
2759 | break; | |
2760 | case TCG_MO_LD_LD | TCG_MO_LD_ST: | |
2761 | m_op = "rr+rw"; | |
2762 | break; | |
2763 | case TCG_MO_LD_LD | TCG_MO_ST_LD: | |
2764 | m_op = "rr+wr"; | |
2765 | break; | |
2766 | case TCG_MO_LD_LD | TCG_MO_ST_ST: | |
2767 | m_op = "rr+ww"; | |
2768 | break; | |
2769 | case TCG_MO_LD_ST | TCG_MO_ST_LD: | |
2770 | m_op = "rw+wr"; | |
2771 | break; | |
2772 | case TCG_MO_LD_ST | TCG_MO_ST_ST: | |
2773 | m_op = "rw+ww"; | |
2774 | break; | |
2775 | case TCG_MO_ST_LD | TCG_MO_ST_ST: | |
2776 | m_op = "wr+ww"; | |
2777 | break; | |
2778 | case TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_LD: | |
2779 | m_op = "rr+rw+wr"; | |
2780 | break; | |
2781 | case TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_ST: | |
2782 | m_op = "rr+rw+ww"; | |
2783 | break; | |
2784 | case TCG_MO_LD_LD | TCG_MO_ST_LD | TCG_MO_ST_ST: | |
2785 | m_op = "rr+wr+ww"; | |
2786 | break; | |
2787 | case TCG_MO_LD_ST | TCG_MO_ST_LD | TCG_MO_ST_ST: | |
2788 | m_op = "rw+wr+ww"; | |
2789 | break; | |
2790 | case TCG_MO_ALL: | |
2791 | m_op = "all"; | |
2792 | break; | |
2793 | default: | |
2794 | g_assert_not_reached(); | |
2795 | } | |
2796 | ||
2797 | col += ne_fprintf(f, "%s%s:%s", (k ? "," : ""), b_op, m_op); | |
2798 | i++, k++; | |
2799 | } | |
2800 | break; | |
51e3972c RH |
2801 | default: |
2802 | break; | |
2803 | } | |
2804 | for (; i < nb_cargs; i++, k++) { | |
b7a83ff8 RH |
2805 | col += ne_fprintf(f, "%s$0x%" TCG_PRIlx, k ? "," : "", |
2806 | op->args[k]); | |
bdfb460e RH |
2807 | } |
2808 | } | |
bdfb460e | 2809 | |
1894f69a | 2810 | if (have_prefs || op->life) { |
b7a83ff8 RH |
2811 | for (; col < 40; ++col) { |
2812 | putc(' ', f); | |
bdfb460e | 2813 | } |
1894f69a RH |
2814 | } |
2815 | ||
2816 | if (op->life) { | |
2817 | unsigned life = op->life; | |
bdfb460e RH |
2818 | |
2819 | if (life & (SYNC_ARG * 3)) { | |
b7a83ff8 | 2820 | ne_fprintf(f, " sync:"); |
bdfb460e RH |
2821 | for (i = 0; i < 2; ++i) { |
2822 | if (life & (SYNC_ARG << i)) { | |
b7a83ff8 | 2823 | ne_fprintf(f, " %d", i); |
bdfb460e RH |
2824 | } |
2825 | } | |
2826 | } | |
2827 | life /= DEAD_ARG; | |
2828 | if (life) { | |
b7a83ff8 | 2829 | ne_fprintf(f, " dead:"); |
bdfb460e RH |
2830 | for (i = 0; life; ++i, life >>= 1) { |
2831 | if (life & 1) { | |
b7a83ff8 | 2832 | ne_fprintf(f, " %d", i); |
bdfb460e RH |
2833 | } |
2834 | } | |
b03cce8e | 2835 | } |
c896fe29 | 2836 | } |
1894f69a RH |
2837 | |
2838 | if (have_prefs) { | |
2839 | for (i = 0; i < nb_oargs; ++i) { | |
31fd884b | 2840 | TCGRegSet set = output_pref(op, i); |
1894f69a RH |
2841 | |
2842 | if (i == 0) { | |
b7a83ff8 | 2843 | ne_fprintf(f, " pref="); |
1894f69a | 2844 | } else { |
b7a83ff8 | 2845 | ne_fprintf(f, ","); |
1894f69a RH |
2846 | } |
2847 | if (set == 0) { | |
b7a83ff8 | 2848 | ne_fprintf(f, "none"); |
1894f69a | 2849 | } else if (set == MAKE_64BIT_MASK(0, TCG_TARGET_NB_REGS)) { |
b7a83ff8 | 2850 | ne_fprintf(f, "all"); |
1894f69a RH |
2851 | #ifdef CONFIG_DEBUG_TCG |
2852 | } else if (tcg_regset_single(set)) { | |
2853 | TCGReg reg = tcg_regset_first(set); | |
b7a83ff8 | 2854 | ne_fprintf(f, "%s", tcg_target_reg_names[reg]); |
1894f69a RH |
2855 | #endif |
2856 | } else if (TCG_TARGET_NB_REGS <= 32) { | |
b7a83ff8 | 2857 | ne_fprintf(f, "0x%x", (uint32_t)set); |
1894f69a | 2858 | } else { |
b7a83ff8 | 2859 | ne_fprintf(f, "0x%" PRIx64, (uint64_t)set); |
1894f69a RH |
2860 | } |
2861 | } | |
2862 | } | |
2863 | ||
b7a83ff8 | 2864 | putc('\n', f); |
c896fe29 FB |
2865 | } |
2866 | } | |
2867 | ||
2868 | /* we give more priority to constraints with less registers */ | |
2869 | static int get_constraint_priority(const TCGOpDef *def, int k) | |
2870 | { | |
74a11790 | 2871 | const TCGArgConstraint *arg_ct = &def->args_ct[k]; |
29f5e925 | 2872 | int n = ctpop64(arg_ct->regs); |
c896fe29 | 2873 | |
29f5e925 RH |
2874 | /* |
2875 | * Sort constraints of a single register first, which includes output | |
2876 | * aliases (which must exactly match the input already allocated). | |
2877 | */ | |
2878 | if (n == 1 || arg_ct->oalias) { | |
2879 | return INT_MAX; | |
2880 | } | |
2881 | ||
2882 | /* | |
2883 | * Sort register pairs next, first then second immediately after. | |
2884 | * Arbitrarily sort multiple pairs by the index of the first reg; | |
2885 | * there shouldn't be many pairs. | |
2886 | */ | |
2887 | switch (arg_ct->pair) { | |
2888 | case 1: | |
2889 | case 3: | |
2890 | return (k + 1) * 2; | |
2891 | case 2: | |
2892 | return (arg_ct->pair_index + 1) * 2 - 1; | |
c896fe29 | 2893 | } |
29f5e925 RH |
2894 | |
2895 | /* Finally, sort by decreasing register count. */ | |
2896 | assert(n > 1); | |
2897 | return -n; | |
c896fe29 FB |
2898 | } |
2899 | ||
2900 | /* sort from highest priority to lowest */ | |
2901 | static void sort_constraints(TCGOpDef *def, int start, int n) | |
2902 | { | |
66792f90 RH |
2903 | int i, j; |
2904 | TCGArgConstraint *a = def->args_ct; | |
c896fe29 | 2905 | |
66792f90 RH |
2906 | for (i = 0; i < n; i++) { |
2907 | a[start + i].sort_index = start + i; | |
2908 | } | |
2909 | if (n <= 1) { | |
c896fe29 | 2910 | return; |
66792f90 RH |
2911 | } |
2912 | for (i = 0; i < n - 1; i++) { | |
2913 | for (j = i + 1; j < n; j++) { | |
2914 | int p1 = get_constraint_priority(def, a[start + i].sort_index); | |
2915 | int p2 = get_constraint_priority(def, a[start + j].sort_index); | |
c896fe29 | 2916 | if (p1 < p2) { |
66792f90 RH |
2917 | int tmp = a[start + i].sort_index; |
2918 | a[start + i].sort_index = a[start + j].sort_index; | |
2919 | a[start + j].sort_index = tmp; | |
c896fe29 FB |
2920 | } |
2921 | } | |
2922 | } | |
2923 | } | |
2924 | ||
f69d277e | 2925 | static void process_op_defs(TCGContext *s) |
c896fe29 | 2926 | { |
a9751609 | 2927 | TCGOpcode op; |
c896fe29 | 2928 | |
f69d277e RH |
2929 | for (op = 0; op < NB_OPS; op++) { |
2930 | TCGOpDef *def = &tcg_op_defs[op]; | |
2931 | const TCGTargetOpDef *tdefs; | |
29f5e925 RH |
2932 | bool saw_alias_pair = false; |
2933 | int i, o, i2, o2, nb_args; | |
f69d277e RH |
2934 | |
2935 | if (def->flags & TCG_OPF_NOT_PRESENT) { | |
2936 | continue; | |
2937 | } | |
2938 | ||
c896fe29 | 2939 | nb_args = def->nb_iargs + def->nb_oargs; |
f69d277e RH |
2940 | if (nb_args == 0) { |
2941 | continue; | |
2942 | } | |
2943 | ||
4c22e840 RH |
2944 | /* |
2945 | * Macro magic should make it impossible, but double-check that | |
2946 | * the array index is in range. Since the signness of an enum | |
2947 | * is implementation defined, force the result to unsigned. | |
2948 | */ | |
2949 | unsigned con_set = tcg_target_op_def(op); | |
2950 | tcg_debug_assert(con_set < ARRAY_SIZE(constraint_sets)); | |
2951 | tdefs = &constraint_sets[con_set]; | |
f69d277e RH |
2952 | |
2953 | for (i = 0; i < nb_args; i++) { | |
2954 | const char *ct_str = tdefs->args_ct_str[i]; | |
8940ea0d PMD |
2955 | bool input_p = i >= def->nb_oargs; |
2956 | ||
f69d277e | 2957 | /* Incomplete TCGTargetOpDef entry. */ |
eabb7b91 | 2958 | tcg_debug_assert(ct_str != NULL); |
f69d277e | 2959 | |
8940ea0d PMD |
2960 | switch (*ct_str) { |
2961 | case '0' ... '9': | |
2962 | o = *ct_str - '0'; | |
2963 | tcg_debug_assert(input_p); | |
2964 | tcg_debug_assert(o < def->nb_oargs); | |
2965 | tcg_debug_assert(def->args_ct[o].regs != 0); | |
2966 | tcg_debug_assert(!def->args_ct[o].oalias); | |
2967 | def->args_ct[i] = def->args_ct[o]; | |
2968 | /* The output sets oalias. */ | |
2969 | def->args_ct[o].oalias = 1; | |
2970 | def->args_ct[o].alias_index = i; | |
2971 | /* The input sets ialias. */ | |
2972 | def->args_ct[i].ialias = 1; | |
2973 | def->args_ct[i].alias_index = o; | |
29f5e925 RH |
2974 | if (def->args_ct[i].pair) { |
2975 | saw_alias_pair = true; | |
2976 | } | |
8940ea0d PMD |
2977 | tcg_debug_assert(ct_str[1] == '\0'); |
2978 | continue; | |
2979 | ||
2980 | case '&': | |
2981 | tcg_debug_assert(!input_p); | |
2982 | def->args_ct[i].newreg = true; | |
2983 | ct_str++; | |
2984 | break; | |
29f5e925 RH |
2985 | |
2986 | case 'p': /* plus */ | |
2987 | /* Allocate to the register after the previous. */ | |
2988 | tcg_debug_assert(i > (input_p ? def->nb_oargs : 0)); | |
2989 | o = i - 1; | |
2990 | tcg_debug_assert(!def->args_ct[o].pair); | |
2991 | tcg_debug_assert(!def->args_ct[o].ct); | |
2992 | def->args_ct[i] = (TCGArgConstraint){ | |
2993 | .pair = 2, | |
2994 | .pair_index = o, | |
2995 | .regs = def->args_ct[o].regs << 1, | |
ca5bed07 | 2996 | .newreg = def->args_ct[o].newreg, |
29f5e925 RH |
2997 | }; |
2998 | def->args_ct[o].pair = 1; | |
2999 | def->args_ct[o].pair_index = i; | |
3000 | tcg_debug_assert(ct_str[1] == '\0'); | |
3001 | continue; | |
3002 | ||
3003 | case 'm': /* minus */ | |
3004 | /* Allocate to the register before the previous. */ | |
3005 | tcg_debug_assert(i > (input_p ? def->nb_oargs : 0)); | |
3006 | o = i - 1; | |
3007 | tcg_debug_assert(!def->args_ct[o].pair); | |
3008 | tcg_debug_assert(!def->args_ct[o].ct); | |
3009 | def->args_ct[i] = (TCGArgConstraint){ | |
3010 | .pair = 1, | |
3011 | .pair_index = o, | |
3012 | .regs = def->args_ct[o].regs >> 1, | |
ca5bed07 | 3013 | .newreg = def->args_ct[o].newreg, |
29f5e925 RH |
3014 | }; |
3015 | def->args_ct[o].pair = 2; | |
3016 | def->args_ct[o].pair_index = i; | |
3017 | tcg_debug_assert(ct_str[1] == '\0'); | |
3018 | continue; | |
8940ea0d PMD |
3019 | } |
3020 | ||
3021 | do { | |
3022 | switch (*ct_str) { | |
17280ff4 RH |
3023 | case 'i': |
3024 | def->args_ct[i].ct |= TCG_CT_CONST; | |
17280ff4 | 3025 | break; |
358b4923 | 3026 | |
358b4923 RH |
3027 | /* Include all of the target-specific constraints. */ |
3028 | ||
3029 | #undef CONST | |
3030 | #define CONST(CASE, MASK) \ | |
8940ea0d | 3031 | case CASE: def->args_ct[i].ct |= MASK; break; |
358b4923 | 3032 | #define REGS(CASE, MASK) \ |
8940ea0d | 3033 | case CASE: def->args_ct[i].regs |= MASK; break; |
358b4923 RH |
3034 | |
3035 | #include "tcg-target-con-str.h" | |
3036 | ||
3037 | #undef REGS | |
3038 | #undef CONST | |
17280ff4 | 3039 | default: |
8940ea0d PMD |
3040 | case '0' ... '9': |
3041 | case '&': | |
29f5e925 RH |
3042 | case 'p': |
3043 | case 'm': | |
17280ff4 | 3044 | /* Typo in TCGTargetOpDef constraint. */ |
358b4923 | 3045 | g_assert_not_reached(); |
c896fe29 | 3046 | } |
8940ea0d | 3047 | } while (*++ct_str != '\0'); |
c896fe29 FB |
3048 | } |
3049 | ||
c68aaa18 | 3050 | /* TCGTargetOpDef entry with too much information? */ |
eabb7b91 | 3051 | tcg_debug_assert(i == TCG_MAX_OP_ARGS || tdefs->args_ct_str[i] == NULL); |
c68aaa18 | 3052 | |
29f5e925 RH |
3053 | /* |
3054 | * Fix up output pairs that are aliased with inputs. | |
3055 | * When we created the alias, we copied pair from the output. | |
3056 | * There are three cases: | |
3057 | * (1a) Pairs of inputs alias pairs of outputs. | |
3058 | * (1b) One input aliases the first of a pair of outputs. | |
3059 | * (2) One input aliases the second of a pair of outputs. | |
3060 | * | |
3061 | * Case 1a is handled by making sure that the pair_index'es are | |
3062 | * properly updated so that they appear the same as a pair of inputs. | |
3063 | * | |
3064 | * Case 1b is handled by setting the pair_index of the input to | |
3065 | * itself, simply so it doesn't point to an unrelated argument. | |
3066 | * Since we don't encounter the "second" during the input allocation | |
3067 | * phase, nothing happens with the second half of the input pair. | |
3068 | * | |
3069 | * Case 2 is handled by setting the second input to pair=3, the | |
3070 | * first output to pair=3, and the pair_index'es to match. | |
3071 | */ | |
3072 | if (saw_alias_pair) { | |
3073 | for (i = def->nb_oargs; i < nb_args; i++) { | |
3074 | /* | |
3075 | * Since [0-9pm] must be alone in the constraint string, | |
3076 | * the only way they can both be set is if the pair comes | |
3077 | * from the output alias. | |
3078 | */ | |
3079 | if (!def->args_ct[i].ialias) { | |
3080 | continue; | |
3081 | } | |
3082 | switch (def->args_ct[i].pair) { | |
3083 | case 0: | |
3084 | break; | |
3085 | case 1: | |
3086 | o = def->args_ct[i].alias_index; | |
3087 | o2 = def->args_ct[o].pair_index; | |
3088 | tcg_debug_assert(def->args_ct[o].pair == 1); | |
3089 | tcg_debug_assert(def->args_ct[o2].pair == 2); | |
3090 | if (def->args_ct[o2].oalias) { | |
3091 | /* Case 1a */ | |
3092 | i2 = def->args_ct[o2].alias_index; | |
3093 | tcg_debug_assert(def->args_ct[i2].pair == 2); | |
3094 | def->args_ct[i2].pair_index = i; | |
3095 | def->args_ct[i].pair_index = i2; | |
3096 | } else { | |
3097 | /* Case 1b */ | |
3098 | def->args_ct[i].pair_index = i; | |
3099 | } | |
3100 | break; | |
3101 | case 2: | |
3102 | o = def->args_ct[i].alias_index; | |
3103 | o2 = def->args_ct[o].pair_index; | |
3104 | tcg_debug_assert(def->args_ct[o].pair == 2); | |
3105 | tcg_debug_assert(def->args_ct[o2].pair == 1); | |
3106 | if (def->args_ct[o2].oalias) { | |
3107 | /* Case 1a */ | |
3108 | i2 = def->args_ct[o2].alias_index; | |
3109 | tcg_debug_assert(def->args_ct[i2].pair == 1); | |
3110 | def->args_ct[i2].pair_index = i; | |
3111 | def->args_ct[i].pair_index = i2; | |
3112 | } else { | |
3113 | /* Case 2 */ | |
3114 | def->args_ct[i].pair = 3; | |
3115 | def->args_ct[o2].pair = 3; | |
3116 | def->args_ct[i].pair_index = o2; | |
3117 | def->args_ct[o2].pair_index = i; | |
3118 | } | |
3119 | break; | |
3120 | default: | |
3121 | g_assert_not_reached(); | |
3122 | } | |
3123 | } | |
3124 | } | |
3125 | ||
c896fe29 FB |
3126 | /* sort the constraints (XXX: this is just an heuristic) */ |
3127 | sort_constraints(def, 0, def->nb_oargs); | |
3128 | sort_constraints(def, def->nb_oargs, def->nb_iargs); | |
a9751609 | 3129 | } |
c896fe29 FB |
3130 | } |
3131 | ||
f85b1fc4 | 3132 | static void remove_label_use(TCGOp *op, int idx) |
0c627cdc | 3133 | { |
f85b1fc4 RH |
3134 | TCGLabel *label = arg_label(op->args[idx]); |
3135 | TCGLabelUse *use; | |
d88a117e | 3136 | |
f85b1fc4 RH |
3137 | QSIMPLEQ_FOREACH(use, &label->branches, next) { |
3138 | if (use->op == op) { | |
3139 | QSIMPLEQ_REMOVE(&label->branches, use, TCGLabelUse, next); | |
3140 | return; | |
3141 | } | |
3142 | } | |
3143 | g_assert_not_reached(); | |
3144 | } | |
3145 | ||
3146 | void tcg_op_remove(TCGContext *s, TCGOp *op) | |
3147 | { | |
d88a117e RH |
3148 | switch (op->opc) { |
3149 | case INDEX_op_br: | |
f85b1fc4 | 3150 | remove_label_use(op, 0); |
d88a117e RH |
3151 | break; |
3152 | case INDEX_op_brcond_i32: | |
3153 | case INDEX_op_brcond_i64: | |
f85b1fc4 | 3154 | remove_label_use(op, 3); |
d88a117e RH |
3155 | break; |
3156 | case INDEX_op_brcond2_i32: | |
f85b1fc4 | 3157 | remove_label_use(op, 5); |
d88a117e RH |
3158 | break; |
3159 | default: | |
3160 | break; | |
3161 | } | |
3162 | ||
15fa08f8 RH |
3163 | QTAILQ_REMOVE(&s->ops, op, link); |
3164 | QTAILQ_INSERT_TAIL(&s->free_ops, op, link); | |
abebf925 | 3165 | s->nb_ops--; |
0c627cdc RH |
3166 | } |
3167 | ||
a80cdd31 RH |
3168 | void tcg_remove_ops_after(TCGOp *op) |
3169 | { | |
3170 | TCGContext *s = tcg_ctx; | |
3171 | ||
3172 | while (true) { | |
3173 | TCGOp *last = tcg_last_op(); | |
3174 | if (last == op) { | |
3175 | return; | |
3176 | } | |
3177 | tcg_op_remove(s, last); | |
3178 | } | |
3179 | } | |
3180 | ||
d4478943 | 3181 | static TCGOp *tcg_op_alloc(TCGOpcode opc, unsigned nargs) |
5a18407f | 3182 | { |
15fa08f8 | 3183 | TCGContext *s = tcg_ctx; |
cb10bc63 RH |
3184 | TCGOp *op = NULL; |
3185 | ||
3186 | if (unlikely(!QTAILQ_EMPTY(&s->free_ops))) { | |
3187 | QTAILQ_FOREACH(op, &s->free_ops, link) { | |
3188 | if (nargs <= op->nargs) { | |
3189 | QTAILQ_REMOVE(&s->free_ops, op, link); | |
3190 | nargs = op->nargs; | |
3191 | goto found; | |
3192 | } | |
3193 | } | |
15fa08f8 | 3194 | } |
cb10bc63 RH |
3195 | |
3196 | /* Most opcodes have 3 or 4 operands: reduce fragmentation. */ | |
3197 | nargs = MAX(4, nargs); | |
3198 | op = tcg_malloc(sizeof(TCGOp) + sizeof(TCGArg) * nargs); | |
3199 | ||
3200 | found: | |
15fa08f8 RH |
3201 | memset(op, 0, offsetof(TCGOp, link)); |
3202 | op->opc = opc; | |
cb10bc63 RH |
3203 | op->nargs = nargs; |
3204 | ||
3205 | /* Check for bitfield overflow. */ | |
3206 | tcg_debug_assert(op->nargs == nargs); | |
5a18407f | 3207 | |
cb10bc63 | 3208 | s->nb_ops++; |
15fa08f8 RH |
3209 | return op; |
3210 | } | |
3211 | ||
d4478943 | 3212 | TCGOp *tcg_emit_op(TCGOpcode opc, unsigned nargs) |
15fa08f8 | 3213 | { |
d4478943 | 3214 | TCGOp *op = tcg_op_alloc(opc, nargs); |
15fa08f8 RH |
3215 | QTAILQ_INSERT_TAIL(&tcg_ctx->ops, op, link); |
3216 | return op; | |
3217 | } | |
5a18407f | 3218 | |
d4478943 PMD |
3219 | TCGOp *tcg_op_insert_before(TCGContext *s, TCGOp *old_op, |
3220 | TCGOpcode opc, unsigned nargs) | |
15fa08f8 | 3221 | { |
d4478943 | 3222 | TCGOp *new_op = tcg_op_alloc(opc, nargs); |
15fa08f8 | 3223 | QTAILQ_INSERT_BEFORE(old_op, new_op, link); |
5a18407f RH |
3224 | return new_op; |
3225 | } | |
3226 | ||
d4478943 PMD |
3227 | TCGOp *tcg_op_insert_after(TCGContext *s, TCGOp *old_op, |
3228 | TCGOpcode opc, unsigned nargs) | |
5a18407f | 3229 | { |
d4478943 | 3230 | TCGOp *new_op = tcg_op_alloc(opc, nargs); |
15fa08f8 | 3231 | QTAILQ_INSERT_AFTER(&s->ops, old_op, new_op, link); |
5a18407f RH |
3232 | return new_op; |
3233 | } | |
3234 | ||
968f305e RH |
3235 | static void move_label_uses(TCGLabel *to, TCGLabel *from) |
3236 | { | |
3237 | TCGLabelUse *u; | |
3238 | ||
3239 | QSIMPLEQ_FOREACH(u, &from->branches, next) { | |
3240 | TCGOp *op = u->op; | |
3241 | switch (op->opc) { | |
3242 | case INDEX_op_br: | |
3243 | op->args[0] = label_arg(to); | |
3244 | break; | |
3245 | case INDEX_op_brcond_i32: | |
3246 | case INDEX_op_brcond_i64: | |
3247 | op->args[3] = label_arg(to); | |
3248 | break; | |
3249 | case INDEX_op_brcond2_i32: | |
3250 | op->args[5] = label_arg(to); | |
3251 | break; | |
3252 | default: | |
3253 | g_assert_not_reached(); | |
3254 | } | |
3255 | } | |
3256 | ||
3257 | QSIMPLEQ_CONCAT(&to->branches, &from->branches); | |
3258 | } | |
3259 | ||
b4fc67c7 | 3260 | /* Reachable analysis : remove unreachable code. */ |
9bbee4c0 RH |
3261 | static void __attribute__((noinline)) |
3262 | reachable_code_pass(TCGContext *s) | |
b4fc67c7 | 3263 | { |
4d89d0bb | 3264 | TCGOp *op, *op_next, *op_prev; |
b4fc67c7 RH |
3265 | bool dead = false; |
3266 | ||
3267 | QTAILQ_FOREACH_SAFE(op, &s->ops, link, op_next) { | |
3268 | bool remove = dead; | |
3269 | TCGLabel *label; | |
b4fc67c7 RH |
3270 | |
3271 | switch (op->opc) { | |
3272 | case INDEX_op_set_label: | |
3273 | label = arg_label(op->args[0]); | |
4d89d0bb | 3274 | |
968f305e RH |
3275 | /* |
3276 | * Note that the first op in the TB is always a load, | |
3277 | * so there is always something before a label. | |
3278 | */ | |
3279 | op_prev = QTAILQ_PREV(op, link); | |
3280 | ||
3281 | /* | |
3282 | * If we find two sequential labels, move all branches to | |
3283 | * reference the second label and remove the first label. | |
3284 | * Do this before branch to next optimization, so that the | |
3285 | * middle label is out of the way. | |
3286 | */ | |
3287 | if (op_prev->opc == INDEX_op_set_label) { | |
3288 | move_label_uses(label, arg_label(op_prev->args[0])); | |
3289 | tcg_op_remove(s, op_prev); | |
3290 | op_prev = QTAILQ_PREV(op, link); | |
3291 | } | |
3292 | ||
4d89d0bb RH |
3293 | /* |
3294 | * Optimization can fold conditional branches to unconditional. | |
3295 | * If we find a label which is preceded by an unconditional | |
3296 | * branch to next, remove the branch. We couldn't do this when | |
3297 | * processing the branch because any dead code between the branch | |
3298 | * and label had not yet been removed. | |
3299 | */ | |
4d89d0bb RH |
3300 | if (op_prev->opc == INDEX_op_br && |
3301 | label == arg_label(op_prev->args[0])) { | |
3302 | tcg_op_remove(s, op_prev); | |
3303 | /* Fall through means insns become live again. */ | |
3304 | dead = false; | |
3305 | } | |
3306 | ||
f85b1fc4 | 3307 | if (QSIMPLEQ_EMPTY(&label->branches)) { |
b4fc67c7 RH |
3308 | /* |
3309 | * While there is an occasional backward branch, virtually | |
3310 | * all branches generated by the translators are forward. | |
3311 | * Which means that generally we will have already removed | |
3312 | * all references to the label that will be, and there is | |
3313 | * little to be gained by iterating. | |
3314 | */ | |
3315 | remove = true; | |
3316 | } else { | |
3317 | /* Once we see a label, insns become live again. */ | |
3318 | dead = false; | |
3319 | remove = false; | |
b4fc67c7 RH |
3320 | } |
3321 | break; | |
3322 | ||
3323 | case INDEX_op_br: | |
3324 | case INDEX_op_exit_tb: | |
3325 | case INDEX_op_goto_ptr: | |
3326 | /* Unconditional branches; everything following is dead. */ | |
3327 | dead = true; | |
3328 | break; | |
3329 | ||
3330 | case INDEX_op_call: | |
3331 | /* Notice noreturn helper calls, raising exceptions. */ | |
90163900 | 3332 | if (tcg_call_flags(op) & TCG_CALL_NO_RETURN) { |
b4fc67c7 RH |
3333 | dead = true; |
3334 | } | |
3335 | break; | |
3336 | ||
3337 | case INDEX_op_insn_start: | |
3338 | /* Never remove -- we need to keep these for unwind. */ | |
3339 | remove = false; | |
3340 | break; | |
3341 | ||
3342 | default: | |
3343 | break; | |
3344 | } | |
3345 | ||
3346 | if (remove) { | |
3347 | tcg_op_remove(s, op); | |
3348 | } | |
3349 | } | |
3350 | } | |
3351 | ||
c70fbf0a RH |
3352 | #define TS_DEAD 1 |
3353 | #define TS_MEM 2 | |
3354 | ||
5a18407f RH |
3355 | #define IS_DEAD_ARG(n) (arg_life & (DEAD_ARG << (n))) |
3356 | #define NEED_SYNC_ARG(n) (arg_life & (SYNC_ARG << (n))) | |
3357 | ||
25f49c5f RH |
3358 | /* For liveness_pass_1, the register preferences for a given temp. */ |
3359 | static inline TCGRegSet *la_temp_pref(TCGTemp *ts) | |
3360 | { | |
3361 | return ts->state_ptr; | |
3362 | } | |
3363 | ||
3364 | /* For liveness_pass_1, reset the preferences for a given temp to the | |
3365 | * maximal regset for its type. | |
3366 | */ | |
3367 | static inline void la_reset_pref(TCGTemp *ts) | |
3368 | { | |
3369 | *la_temp_pref(ts) | |
3370 | = (ts->state == TS_DEAD ? 0 : tcg_target_available_regs[ts->type]); | |
3371 | } | |
3372 | ||
9c43b68d AJ |
3373 | /* liveness analysis: end of function: all temps are dead, and globals |
3374 | should be in memory. */ | |
2616c808 | 3375 | static void la_func_end(TCGContext *s, int ng, int nt) |
c896fe29 | 3376 | { |
b83eabea RH |
3377 | int i; |
3378 | ||
3379 | for (i = 0; i < ng; ++i) { | |
3380 | s->temps[i].state = TS_DEAD | TS_MEM; | |
25f49c5f | 3381 | la_reset_pref(&s->temps[i]); |
b83eabea RH |
3382 | } |
3383 | for (i = ng; i < nt; ++i) { | |
3384 | s->temps[i].state = TS_DEAD; | |
25f49c5f | 3385 | la_reset_pref(&s->temps[i]); |
b83eabea | 3386 | } |
c896fe29 FB |
3387 | } |
3388 | ||
9c43b68d AJ |
3389 | /* liveness analysis: end of basic block: all temps are dead, globals |
3390 | and local temps should be in memory. */ | |
2616c808 | 3391 | static void la_bb_end(TCGContext *s, int ng, int nt) |
641d5fbe | 3392 | { |
b83eabea | 3393 | int i; |
641d5fbe | 3394 | |
ee17db83 RH |
3395 | for (i = 0; i < nt; ++i) { |
3396 | TCGTemp *ts = &s->temps[i]; | |
3397 | int state; | |
3398 | ||
3399 | switch (ts->kind) { | |
3400 | case TEMP_FIXED: | |
3401 | case TEMP_GLOBAL: | |
f57c6915 | 3402 | case TEMP_TB: |
ee17db83 RH |
3403 | state = TS_DEAD | TS_MEM; |
3404 | break; | |
c7482438 | 3405 | case TEMP_EBB: |
c0522136 | 3406 | case TEMP_CONST: |
ee17db83 RH |
3407 | state = TS_DEAD; |
3408 | break; | |
3409 | default: | |
3410 | g_assert_not_reached(); | |
3411 | } | |
3412 | ts->state = state; | |
3413 | la_reset_pref(ts); | |
641d5fbe FB |
3414 | } |
3415 | } | |
3416 | ||
f65a061c RH |
3417 | /* liveness analysis: sync globals back to memory. */ |
3418 | static void la_global_sync(TCGContext *s, int ng) | |
3419 | { | |
3420 | int i; | |
3421 | ||
3422 | for (i = 0; i < ng; ++i) { | |
25f49c5f RH |
3423 | int state = s->temps[i].state; |
3424 | s->temps[i].state = state | TS_MEM; | |
3425 | if (state == TS_DEAD) { | |
3426 | /* If the global was previously dead, reset prefs. */ | |
3427 | la_reset_pref(&s->temps[i]); | |
3428 | } | |
f65a061c RH |
3429 | } |
3430 | } | |
3431 | ||
b4cb76e6 | 3432 | /* |
c7482438 RH |
3433 | * liveness analysis: conditional branch: all temps are dead unless |
3434 | * explicitly live-across-conditional-branch, globals and local temps | |
3435 | * should be synced. | |
b4cb76e6 RH |
3436 | */ |
3437 | static void la_bb_sync(TCGContext *s, int ng, int nt) | |
3438 | { | |
3439 | la_global_sync(s, ng); | |
3440 | ||
3441 | for (int i = ng; i < nt; ++i) { | |
c0522136 RH |
3442 | TCGTemp *ts = &s->temps[i]; |
3443 | int state; | |
3444 | ||
3445 | switch (ts->kind) { | |
f57c6915 | 3446 | case TEMP_TB: |
c0522136 RH |
3447 | state = ts->state; |
3448 | ts->state = state | TS_MEM; | |
b4cb76e6 RH |
3449 | if (state != TS_DEAD) { |
3450 | continue; | |
3451 | } | |
c0522136 | 3452 | break; |
c7482438 | 3453 | case TEMP_EBB: |
c0522136 RH |
3454 | case TEMP_CONST: |
3455 | continue; | |
3456 | default: | |
3457 | g_assert_not_reached(); | |
b4cb76e6 RH |
3458 | } |
3459 | la_reset_pref(&s->temps[i]); | |
3460 | } | |
3461 | } | |
3462 | ||
f65a061c RH |
3463 | /* liveness analysis: sync globals back to memory and kill. */ |
3464 | static void la_global_kill(TCGContext *s, int ng) | |
3465 | { | |
3466 | int i; | |
3467 | ||
3468 | for (i = 0; i < ng; i++) { | |
3469 | s->temps[i].state = TS_DEAD | TS_MEM; | |
25f49c5f RH |
3470 | la_reset_pref(&s->temps[i]); |
3471 | } | |
3472 | } | |
3473 | ||
3474 | /* liveness analysis: note live globals crossing calls. */ | |
3475 | static void la_cross_call(TCGContext *s, int nt) | |
3476 | { | |
3477 | TCGRegSet mask = ~tcg_target_call_clobber_regs; | |
3478 | int i; | |
3479 | ||
3480 | for (i = 0; i < nt; i++) { | |
3481 | TCGTemp *ts = &s->temps[i]; | |
3482 | if (!(ts->state & TS_DEAD)) { | |
3483 | TCGRegSet *pset = la_temp_pref(ts); | |
3484 | TCGRegSet set = *pset; | |
3485 | ||
3486 | set &= mask; | |
3487 | /* If the combination is not possible, restart. */ | |
3488 | if (set == 0) { | |
3489 | set = tcg_target_available_regs[ts->type] & mask; | |
3490 | } | |
3491 | *pset = set; | |
3492 | } | |
f65a061c RH |
3493 | } |
3494 | } | |
3495 | ||
874b8574 RH |
3496 | /* |
3497 | * Liveness analysis: Verify the lifetime of TEMP_TB, and reduce | |
3498 | * to TEMP_EBB, if possible. | |
3499 | */ | |
3500 | static void __attribute__((noinline)) | |
3501 | liveness_pass_0(TCGContext *s) | |
3502 | { | |
3503 | void * const multiple_ebb = (void *)(uintptr_t)-1; | |
3504 | int nb_temps = s->nb_temps; | |
3505 | TCGOp *op, *ebb; | |
3506 | ||
3507 | for (int i = s->nb_globals; i < nb_temps; ++i) { | |
3508 | s->temps[i].state_ptr = NULL; | |
3509 | } | |
3510 | ||
3511 | /* | |
3512 | * Represent each EBB by the op at which it begins. In the case of | |
3513 | * the first EBB, this is the first op, otherwise it is a label. | |
3514 | * Collect the uses of each TEMP_TB: NULL for unused, EBB for use | |
3515 | * within a single EBB, else MULTIPLE_EBB. | |
3516 | */ | |
3517 | ebb = QTAILQ_FIRST(&s->ops); | |
3518 | QTAILQ_FOREACH(op, &s->ops, link) { | |
3519 | const TCGOpDef *def; | |
3520 | int nb_oargs, nb_iargs; | |
3521 | ||
3522 | switch (op->opc) { | |
3523 | case INDEX_op_set_label: | |
3524 | ebb = op; | |
3525 | continue; | |
3526 | case INDEX_op_discard: | |
3527 | continue; | |
3528 | case INDEX_op_call: | |
3529 | nb_oargs = TCGOP_CALLO(op); | |
3530 | nb_iargs = TCGOP_CALLI(op); | |
3531 | break; | |
3532 | default: | |
3533 | def = &tcg_op_defs[op->opc]; | |
3534 | nb_oargs = def->nb_oargs; | |
3535 | nb_iargs = def->nb_iargs; | |
3536 | break; | |
3537 | } | |
3538 | ||
3539 | for (int i = 0; i < nb_oargs + nb_iargs; ++i) { | |
3540 | TCGTemp *ts = arg_temp(op->args[i]); | |
3541 | ||
3542 | if (ts->kind != TEMP_TB) { | |
3543 | continue; | |
3544 | } | |
3545 | if (ts->state_ptr == NULL) { | |
3546 | ts->state_ptr = ebb; | |
3547 | } else if (ts->state_ptr != ebb) { | |
3548 | ts->state_ptr = multiple_ebb; | |
3549 | } | |
3550 | } | |
3551 | } | |
3552 | ||
3553 | /* | |
3554 | * For TEMP_TB that turned out not to be used beyond one EBB, | |
3555 | * reduce the liveness to TEMP_EBB. | |
3556 | */ | |
3557 | for (int i = s->nb_globals; i < nb_temps; ++i) { | |
3558 | TCGTemp *ts = &s->temps[i]; | |
3559 | if (ts->kind == TEMP_TB && ts->state_ptr != multiple_ebb) { | |
3560 | ts->kind = TEMP_EBB; | |
3561 | } | |
3562 | } | |
3563 | } | |
3564 | ||
a1b3c48d | 3565 | /* Liveness analysis : update the opc_arg_life array to tell if a |
c896fe29 FB |
3566 | given input arguments is dead. Instructions updating dead |
3567 | temporaries are removed. */ | |
9bbee4c0 RH |
3568 | static void __attribute__((noinline)) |
3569 | liveness_pass_1(TCGContext *s) | |
c896fe29 | 3570 | { |
c70fbf0a | 3571 | int nb_globals = s->nb_globals; |
2616c808 | 3572 | int nb_temps = s->nb_temps; |
15fa08f8 | 3573 | TCGOp *op, *op_prev; |
25f49c5f RH |
3574 | TCGRegSet *prefs; |
3575 | int i; | |
3576 | ||
3577 | prefs = tcg_malloc(sizeof(TCGRegSet) * nb_temps); | |
3578 | for (i = 0; i < nb_temps; ++i) { | |
3579 | s->temps[i].state_ptr = prefs + i; | |
3580 | } | |
a1b3c48d | 3581 | |
ae36a246 | 3582 | /* ??? Should be redundant with the exit_tb that ends the TB. */ |
2616c808 | 3583 | la_func_end(s, nb_globals, nb_temps); |
c896fe29 | 3584 | |
eae3eb3e | 3585 | QTAILQ_FOREACH_REVERSE_SAFE(op, &s->ops, link, op_prev) { |
25f49c5f | 3586 | int nb_iargs, nb_oargs; |
c45cb8bb RH |
3587 | TCGOpcode opc_new, opc_new2; |
3588 | bool have_opc_new2; | |
a1b3c48d | 3589 | TCGLifeData arg_life = 0; |
25f49c5f | 3590 | TCGTemp *ts; |
c45cb8bb RH |
3591 | TCGOpcode opc = op->opc; |
3592 | const TCGOpDef *def = &tcg_op_defs[opc]; | |
3593 | ||
c45cb8bb | 3594 | switch (opc) { |
c896fe29 | 3595 | case INDEX_op_call: |
c6e113f5 | 3596 | { |
39004a71 RH |
3597 | const TCGHelperInfo *info = tcg_call_info(op); |
3598 | int call_flags = tcg_call_flags(op); | |
c896fe29 | 3599 | |
cd9090aa RH |
3600 | nb_oargs = TCGOP_CALLO(op); |
3601 | nb_iargs = TCGOP_CALLI(op); | |
c6e113f5 | 3602 | |
c45cb8bb | 3603 | /* pure functions can be removed if their result is unused */ |
78505279 | 3604 | if (call_flags & TCG_CALL_NO_SIDE_EFFECTS) { |
cf066674 | 3605 | for (i = 0; i < nb_oargs; i++) { |
25f49c5f RH |
3606 | ts = arg_temp(op->args[i]); |
3607 | if (ts->state != TS_DEAD) { | |
c6e113f5 | 3608 | goto do_not_remove_call; |
9c43b68d | 3609 | } |
c6e113f5 | 3610 | } |
c45cb8bb | 3611 | goto do_remove; |
152c35aa RH |
3612 | } |
3613 | do_not_remove_call: | |
c896fe29 | 3614 | |
25f49c5f | 3615 | /* Output args are dead. */ |
152c35aa | 3616 | for (i = 0; i < nb_oargs; i++) { |
25f49c5f RH |
3617 | ts = arg_temp(op->args[i]); |
3618 | if (ts->state & TS_DEAD) { | |
152c35aa RH |
3619 | arg_life |= DEAD_ARG << i; |
3620 | } | |
25f49c5f | 3621 | if (ts->state & TS_MEM) { |
152c35aa | 3622 | arg_life |= SYNC_ARG << i; |
c6e113f5 | 3623 | } |
25f49c5f RH |
3624 | ts->state = TS_DEAD; |
3625 | la_reset_pref(ts); | |
152c35aa | 3626 | } |
78505279 | 3627 | |
31fd884b RH |
3628 | /* Not used -- it will be tcg_target_call_oarg_reg(). */ |
3629 | memset(op->output_pref, 0, sizeof(op->output_pref)); | |
3630 | ||
152c35aa RH |
3631 | if (!(call_flags & (TCG_CALL_NO_WRITE_GLOBALS | |
3632 | TCG_CALL_NO_READ_GLOBALS))) { | |
f65a061c | 3633 | la_global_kill(s, nb_globals); |
152c35aa | 3634 | } else if (!(call_flags & TCG_CALL_NO_READ_GLOBALS)) { |
f65a061c | 3635 | la_global_sync(s, nb_globals); |
152c35aa | 3636 | } |
b9c18f56 | 3637 | |
25f49c5f | 3638 | /* Record arguments that die in this helper. */ |
152c35aa | 3639 | for (i = nb_oargs; i < nb_iargs + nb_oargs; i++) { |
25f49c5f | 3640 | ts = arg_temp(op->args[i]); |
39004a71 | 3641 | if (ts->state & TS_DEAD) { |
152c35aa | 3642 | arg_life |= DEAD_ARG << i; |
c6e113f5 | 3643 | } |
152c35aa | 3644 | } |
25f49c5f RH |
3645 | |
3646 | /* For all live registers, remove call-clobbered prefs. */ | |
3647 | la_cross_call(s, nb_temps); | |
3648 | ||
39004a71 RH |
3649 | /* |
3650 | * Input arguments are live for preceding opcodes. | |
3651 | * | |
3652 | * For those arguments that die, and will be allocated in | |
3653 | * registers, clear the register set for that arg, to be | |
3654 | * filled in below. For args that will be on the stack, | |
3655 | * reset to any available reg. Process arguments in reverse | |
3656 | * order so that if a temp is used more than once, the stack | |
3657 | * reset to max happens before the register reset to 0. | |
3658 | */ | |
3659 | for (i = nb_iargs - 1; i >= 0; i--) { | |
3660 | const TCGCallArgumentLoc *loc = &info->in[i]; | |
3661 | ts = arg_temp(op->args[nb_oargs + i]); | |
25f49c5f | 3662 | |
39004a71 RH |
3663 | if (ts->state & TS_DEAD) { |
3664 | switch (loc->kind) { | |
3665 | case TCG_CALL_ARG_NORMAL: | |
3666 | case TCG_CALL_ARG_EXTEND_U: | |
3667 | case TCG_CALL_ARG_EXTEND_S: | |
338b61e9 | 3668 | if (arg_slot_reg_p(loc->arg_slot)) { |
39004a71 RH |
3669 | *la_temp_pref(ts) = 0; |
3670 | break; | |
3671 | } | |
3672 | /* fall through */ | |
3673 | default: | |
3674 | *la_temp_pref(ts) = | |
3675 | tcg_target_available_regs[ts->type]; | |
3676 | break; | |
3677 | } | |
25f49c5f RH |
3678 | ts->state &= ~TS_DEAD; |
3679 | } | |
3680 | } | |
3681 | ||
39004a71 RH |
3682 | /* |
3683 | * For each input argument, add its input register to prefs. | |
3684 | * If a temp is used once, this produces a single set bit; | |
3685 | * if a temp is used multiple times, this produces a set. | |
3686 | */ | |
3687 | for (i = 0; i < nb_iargs; i++) { | |
3688 | const TCGCallArgumentLoc *loc = &info->in[i]; | |
3689 | ts = arg_temp(op->args[nb_oargs + i]); | |
3690 | ||
3691 | switch (loc->kind) { | |
3692 | case TCG_CALL_ARG_NORMAL: | |
3693 | case TCG_CALL_ARG_EXTEND_U: | |
3694 | case TCG_CALL_ARG_EXTEND_S: | |
338b61e9 | 3695 | if (arg_slot_reg_p(loc->arg_slot)) { |
39004a71 RH |
3696 | tcg_regset_set_reg(*la_temp_pref(ts), |
3697 | tcg_target_call_iarg_regs[loc->arg_slot]); | |
3698 | } | |
3699 | break; | |
3700 | default: | |
3701 | break; | |
c19f47bf | 3702 | } |
c896fe29 | 3703 | } |
c896fe29 | 3704 | } |
c896fe29 | 3705 | break; |
765b842a | 3706 | case INDEX_op_insn_start: |
c896fe29 | 3707 | break; |
5ff9d6a4 | 3708 | case INDEX_op_discard: |
5ff9d6a4 | 3709 | /* mark the temporary as dead */ |
25f49c5f RH |
3710 | ts = arg_temp(op->args[0]); |
3711 | ts->state = TS_DEAD; | |
3712 | la_reset_pref(ts); | |
5ff9d6a4 | 3713 | break; |
1305c451 RH |
3714 | |
3715 | case INDEX_op_add2_i32: | |
c45cb8bb | 3716 | opc_new = INDEX_op_add_i32; |
f1fae40c | 3717 | goto do_addsub2; |
1305c451 | 3718 | case INDEX_op_sub2_i32: |
c45cb8bb | 3719 | opc_new = INDEX_op_sub_i32; |
f1fae40c RH |
3720 | goto do_addsub2; |
3721 | case INDEX_op_add2_i64: | |
c45cb8bb | 3722 | opc_new = INDEX_op_add_i64; |
f1fae40c RH |
3723 | goto do_addsub2; |
3724 | case INDEX_op_sub2_i64: | |
c45cb8bb | 3725 | opc_new = INDEX_op_sub_i64; |
f1fae40c | 3726 | do_addsub2: |
1305c451 RH |
3727 | nb_iargs = 4; |
3728 | nb_oargs = 2; | |
3729 | /* Test if the high part of the operation is dead, but not | |
3730 | the low part. The result can be optimized to a simple | |
3731 | add or sub. This happens often for x86_64 guest when the | |
3732 | cpu mode is set to 32 bit. */ | |
b83eabea RH |
3733 | if (arg_temp(op->args[1])->state == TS_DEAD) { |
3734 | if (arg_temp(op->args[0])->state == TS_DEAD) { | |
1305c451 RH |
3735 | goto do_remove; |
3736 | } | |
c45cb8bb RH |
3737 | /* Replace the opcode and adjust the args in place, |
3738 | leaving 3 unused args at the end. */ | |
3739 | op->opc = opc = opc_new; | |
efee3746 RH |
3740 | op->args[1] = op->args[2]; |
3741 | op->args[2] = op->args[4]; | |
1305c451 RH |
3742 | /* Fall through and mark the single-word operation live. */ |
3743 | nb_iargs = 2; | |
3744 | nb_oargs = 1; | |
3745 | } | |
3746 | goto do_not_remove; | |
3747 | ||
1414968a | 3748 | case INDEX_op_mulu2_i32: |
c45cb8bb RH |
3749 | opc_new = INDEX_op_mul_i32; |
3750 | opc_new2 = INDEX_op_muluh_i32; | |
3751 | have_opc_new2 = TCG_TARGET_HAS_muluh_i32; | |
03271524 | 3752 | goto do_mul2; |
f1fae40c | 3753 | case INDEX_op_muls2_i32: |
c45cb8bb RH |
3754 | opc_new = INDEX_op_mul_i32; |
3755 | opc_new2 = INDEX_op_mulsh_i32; | |
3756 | have_opc_new2 = TCG_TARGET_HAS_mulsh_i32; | |
f1fae40c RH |
3757 | goto do_mul2; |
3758 | case INDEX_op_mulu2_i64: | |
c45cb8bb RH |
3759 | opc_new = INDEX_op_mul_i64; |
3760 | opc_new2 = INDEX_op_muluh_i64; | |
3761 | have_opc_new2 = TCG_TARGET_HAS_muluh_i64; | |
03271524 | 3762 | goto do_mul2; |
f1fae40c | 3763 | case INDEX_op_muls2_i64: |
c45cb8bb RH |
3764 | opc_new = INDEX_op_mul_i64; |
3765 | opc_new2 = INDEX_op_mulsh_i64; | |
3766 | have_opc_new2 = TCG_TARGET_HAS_mulsh_i64; | |
03271524 | 3767 | goto do_mul2; |
f1fae40c | 3768 | do_mul2: |
1414968a RH |
3769 | nb_iargs = 2; |
3770 | nb_oargs = 2; | |
b83eabea RH |
3771 | if (arg_temp(op->args[1])->state == TS_DEAD) { |
3772 | if (arg_temp(op->args[0])->state == TS_DEAD) { | |
03271524 | 3773 | /* Both parts of the operation are dead. */ |
1414968a RH |
3774 | goto do_remove; |
3775 | } | |
03271524 | 3776 | /* The high part of the operation is dead; generate the low. */ |
c45cb8bb | 3777 | op->opc = opc = opc_new; |
efee3746 RH |
3778 | op->args[1] = op->args[2]; |
3779 | op->args[2] = op->args[3]; | |
b83eabea | 3780 | } else if (arg_temp(op->args[0])->state == TS_DEAD && have_opc_new2) { |
c45cb8bb RH |
3781 | /* The low part of the operation is dead; generate the high. */ |
3782 | op->opc = opc = opc_new2; | |
efee3746 RH |
3783 | op->args[0] = op->args[1]; |
3784 | op->args[1] = op->args[2]; | |
3785 | op->args[2] = op->args[3]; | |
03271524 RH |
3786 | } else { |
3787 | goto do_not_remove; | |
1414968a | 3788 | } |
03271524 RH |
3789 | /* Mark the single-word operation live. */ |
3790 | nb_oargs = 1; | |
1414968a RH |
3791 | goto do_not_remove; |
3792 | ||
c896fe29 | 3793 | default: |
1305c451 | 3794 | /* XXX: optimize by hardcoding common cases (e.g. triadic ops) */ |
49516bc0 AJ |
3795 | nb_iargs = def->nb_iargs; |
3796 | nb_oargs = def->nb_oargs; | |
c896fe29 | 3797 | |
49516bc0 AJ |
3798 | /* Test if the operation can be removed because all |
3799 | its outputs are dead. We assume that nb_oargs == 0 | |
3800 | implies side effects */ | |
3801 | if (!(def->flags & TCG_OPF_SIDE_EFFECTS) && nb_oargs != 0) { | |
c45cb8bb | 3802 | for (i = 0; i < nb_oargs; i++) { |
b83eabea | 3803 | if (arg_temp(op->args[i])->state != TS_DEAD) { |
49516bc0 | 3804 | goto do_not_remove; |
9c43b68d | 3805 | } |
49516bc0 | 3806 | } |
152c35aa RH |
3807 | goto do_remove; |
3808 | } | |
3809 | goto do_not_remove; | |
49516bc0 | 3810 | |
152c35aa RH |
3811 | do_remove: |
3812 | tcg_op_remove(s, op); | |
3813 | break; | |
3814 | ||
3815 | do_not_remove: | |
152c35aa | 3816 | for (i = 0; i < nb_oargs; i++) { |
25f49c5f RH |
3817 | ts = arg_temp(op->args[i]); |
3818 | ||
3819 | /* Remember the preference of the uses that followed. */ | |
31fd884b RH |
3820 | if (i < ARRAY_SIZE(op->output_pref)) { |
3821 | op->output_pref[i] = *la_temp_pref(ts); | |
3822 | } | |
25f49c5f RH |
3823 | |
3824 | /* Output args are dead. */ | |
3825 | if (ts->state & TS_DEAD) { | |
152c35aa | 3826 | arg_life |= DEAD_ARG << i; |
49516bc0 | 3827 | } |
25f49c5f | 3828 | if (ts->state & TS_MEM) { |
152c35aa RH |
3829 | arg_life |= SYNC_ARG << i; |
3830 | } | |
25f49c5f RH |
3831 | ts->state = TS_DEAD; |
3832 | la_reset_pref(ts); | |
152c35aa | 3833 | } |
49516bc0 | 3834 | |
25f49c5f | 3835 | /* If end of basic block, update. */ |
ae36a246 RH |
3836 | if (def->flags & TCG_OPF_BB_EXIT) { |
3837 | la_func_end(s, nb_globals, nb_temps); | |
b4cb76e6 RH |
3838 | } else if (def->flags & TCG_OPF_COND_BRANCH) { |
3839 | la_bb_sync(s, nb_globals, nb_temps); | |
ae36a246 | 3840 | } else if (def->flags & TCG_OPF_BB_END) { |
2616c808 | 3841 | la_bb_end(s, nb_globals, nb_temps); |
152c35aa | 3842 | } else if (def->flags & TCG_OPF_SIDE_EFFECTS) { |
f65a061c | 3843 | la_global_sync(s, nb_globals); |
25f49c5f RH |
3844 | if (def->flags & TCG_OPF_CALL_CLOBBER) { |
3845 | la_cross_call(s, nb_temps); | |
3846 | } | |
152c35aa RH |
3847 | } |
3848 | ||
25f49c5f | 3849 | /* Record arguments that die in this opcode. */ |
152c35aa | 3850 | for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) { |
25f49c5f RH |
3851 | ts = arg_temp(op->args[i]); |
3852 | if (ts->state & TS_DEAD) { | |
152c35aa | 3853 | arg_life |= DEAD_ARG << i; |
c896fe29 | 3854 | } |
c896fe29 | 3855 | } |
25f49c5f RH |
3856 | |
3857 | /* Input arguments are live for preceding opcodes. */ | |
152c35aa | 3858 | for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) { |
25f49c5f RH |
3859 | ts = arg_temp(op->args[i]); |
3860 | if (ts->state & TS_DEAD) { | |
3861 | /* For operands that were dead, initially allow | |
3862 | all regs for the type. */ | |
3863 | *la_temp_pref(ts) = tcg_target_available_regs[ts->type]; | |
3864 | ts->state &= ~TS_DEAD; | |
3865 | } | |
3866 | } | |
3867 | ||
3868 | /* Incorporate constraints for this operand. */ | |
3869 | switch (opc) { | |
3870 | case INDEX_op_mov_i32: | |
3871 | case INDEX_op_mov_i64: | |
3872 | /* Note that these are TCG_OPF_NOT_PRESENT and do not | |
3873 | have proper constraints. That said, special case | |
3874 | moves to propagate preferences backward. */ | |
3875 | if (IS_DEAD_ARG(1)) { | |
3876 | *la_temp_pref(arg_temp(op->args[0])) | |
3877 | = *la_temp_pref(arg_temp(op->args[1])); | |
3878 | } | |
3879 | break; | |
3880 | ||
3881 | default: | |
3882 | for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) { | |
3883 | const TCGArgConstraint *ct = &def->args_ct[i]; | |
3884 | TCGRegSet set, *pset; | |
3885 | ||
3886 | ts = arg_temp(op->args[i]); | |
3887 | pset = la_temp_pref(ts); | |
3888 | set = *pset; | |
3889 | ||
9be0d080 | 3890 | set &= ct->regs; |
bc2b17e6 | 3891 | if (ct->ialias) { |
31fd884b | 3892 | set &= output_pref(op, ct->alias_index); |
25f49c5f RH |
3893 | } |
3894 | /* If the combination is not possible, restart. */ | |
3895 | if (set == 0) { | |
9be0d080 | 3896 | set = ct->regs; |
25f49c5f RH |
3897 | } |
3898 | *pset = set; | |
3899 | } | |
3900 | break; | |
152c35aa | 3901 | } |
c896fe29 FB |
3902 | break; |
3903 | } | |
bee158cb | 3904 | op->life = arg_life; |
1ff0a2c5 | 3905 | } |
c896fe29 | 3906 | } |
c896fe29 | 3907 | |
5a18407f | 3908 | /* Liveness analysis: Convert indirect regs to direct temporaries. */ |
9bbee4c0 RH |
3909 | static bool __attribute__((noinline)) |
3910 | liveness_pass_2(TCGContext *s) | |
5a18407f RH |
3911 | { |
3912 | int nb_globals = s->nb_globals; | |
15fa08f8 | 3913 | int nb_temps, i; |
5a18407f | 3914 | bool changes = false; |
15fa08f8 | 3915 | TCGOp *op, *op_next; |
5a18407f | 3916 | |
5a18407f RH |
3917 | /* Create a temporary for each indirect global. */ |
3918 | for (i = 0; i < nb_globals; ++i) { | |
3919 | TCGTemp *its = &s->temps[i]; | |
3920 | if (its->indirect_reg) { | |
3921 | TCGTemp *dts = tcg_temp_alloc(s); | |
3922 | dts->type = its->type; | |
3923 | dts->base_type = its->base_type; | |
e1e64652 | 3924 | dts->temp_subindex = its->temp_subindex; |
c7482438 | 3925 | dts->kind = TEMP_EBB; |
b83eabea RH |
3926 | its->state_ptr = dts; |
3927 | } else { | |
3928 | its->state_ptr = NULL; | |
5a18407f | 3929 | } |
b83eabea RH |
3930 | /* All globals begin dead. */ |
3931 | its->state = TS_DEAD; | |
3932 | } | |
3933 | for (nb_temps = s->nb_temps; i < nb_temps; ++i) { | |
3934 | TCGTemp *its = &s->temps[i]; | |
3935 | its->state_ptr = NULL; | |
3936 | its->state = TS_DEAD; | |
5a18407f | 3937 | } |
5a18407f | 3938 | |
15fa08f8 | 3939 | QTAILQ_FOREACH_SAFE(op, &s->ops, link, op_next) { |
5a18407f RH |
3940 | TCGOpcode opc = op->opc; |
3941 | const TCGOpDef *def = &tcg_op_defs[opc]; | |
3942 | TCGLifeData arg_life = op->life; | |
3943 | int nb_iargs, nb_oargs, call_flags; | |
b83eabea | 3944 | TCGTemp *arg_ts, *dir_ts; |
5a18407f | 3945 | |
5a18407f | 3946 | if (opc == INDEX_op_call) { |
cd9090aa RH |
3947 | nb_oargs = TCGOP_CALLO(op); |
3948 | nb_iargs = TCGOP_CALLI(op); | |
90163900 | 3949 | call_flags = tcg_call_flags(op); |
5a18407f RH |
3950 | } else { |
3951 | nb_iargs = def->nb_iargs; | |
3952 | nb_oargs = def->nb_oargs; | |
3953 | ||
3954 | /* Set flags similar to how calls require. */ | |
b4cb76e6 RH |
3955 | if (def->flags & TCG_OPF_COND_BRANCH) { |
3956 | /* Like reading globals: sync_globals */ | |
3957 | call_flags = TCG_CALL_NO_WRITE_GLOBALS; | |
3958 | } else if (def->flags & TCG_OPF_BB_END) { | |
5a18407f RH |
3959 | /* Like writing globals: save_globals */ |
3960 | call_flags = 0; | |
3961 | } else if (def->flags & TCG_OPF_SIDE_EFFECTS) { | |
3962 | /* Like reading globals: sync_globals */ | |
3963 | call_flags = TCG_CALL_NO_WRITE_GLOBALS; | |
3964 | } else { | |
3965 | /* No effect on globals. */ | |
3966 | call_flags = (TCG_CALL_NO_READ_GLOBALS | | |
3967 | TCG_CALL_NO_WRITE_GLOBALS); | |
3968 | } | |
3969 | } | |
3970 | ||
3971 | /* Make sure that input arguments are available. */ | |
3972 | for (i = nb_oargs; i < nb_iargs + nb_oargs; i++) { | |
b83eabea | 3973 | arg_ts = arg_temp(op->args[i]); |
39004a71 RH |
3974 | dir_ts = arg_ts->state_ptr; |
3975 | if (dir_ts && arg_ts->state == TS_DEAD) { | |
3976 | TCGOpcode lopc = (arg_ts->type == TCG_TYPE_I32 | |
3977 | ? INDEX_op_ld_i32 | |
3978 | : INDEX_op_ld_i64); | |
3979 | TCGOp *lop = tcg_op_insert_before(s, op, lopc, 3); | |
3980 | ||
3981 | lop->args[0] = temp_arg(dir_ts); | |
3982 | lop->args[1] = temp_arg(arg_ts->mem_base); | |
3983 | lop->args[2] = arg_ts->mem_offset; | |
3984 | ||
3985 | /* Loaded, but synced with memory. */ | |
3986 | arg_ts->state = TS_MEM; | |
5a18407f RH |
3987 | } |
3988 | } | |
3989 | ||
3990 | /* Perform input replacement, and mark inputs that became dead. | |
3991 | No action is required except keeping temp_state up to date | |
3992 | so that we reload when needed. */ | |
3993 | for (i = nb_oargs; i < nb_iargs + nb_oargs; i++) { | |
b83eabea | 3994 | arg_ts = arg_temp(op->args[i]); |
39004a71 RH |
3995 | dir_ts = arg_ts->state_ptr; |
3996 | if (dir_ts) { | |
3997 | op->args[i] = temp_arg(dir_ts); | |
3998 | changes = true; | |
3999 | if (IS_DEAD_ARG(i)) { | |
4000 | arg_ts->state = TS_DEAD; | |
5a18407f RH |
4001 | } |
4002 | } | |
4003 | } | |
4004 | ||
4005 | /* Liveness analysis should ensure that the following are | |
4006 | all correct, for call sites and basic block end points. */ | |
4007 | if (call_flags & TCG_CALL_NO_READ_GLOBALS) { | |
4008 | /* Nothing to do */ | |
4009 | } else if (call_flags & TCG_CALL_NO_WRITE_GLOBALS) { | |
4010 | for (i = 0; i < nb_globals; ++i) { | |
4011 | /* Liveness should see that globals are synced back, | |
4012 | that is, either TS_DEAD or TS_MEM. */ | |
b83eabea RH |
4013 | arg_ts = &s->temps[i]; |
4014 | tcg_debug_assert(arg_ts->state_ptr == 0 | |
4015 | || arg_ts->state != 0); | |
5a18407f RH |
4016 | } |
4017 | } else { | |
4018 | for (i = 0; i < nb_globals; ++i) { | |
4019 | /* Liveness should see that globals are saved back, | |
4020 | that is, TS_DEAD, waiting to be reloaded. */ | |
b83eabea RH |
4021 | arg_ts = &s->temps[i]; |
4022 | tcg_debug_assert(arg_ts->state_ptr == 0 | |
4023 | || arg_ts->state == TS_DEAD); | |
5a18407f RH |
4024 | } |
4025 | } | |
4026 | ||
4027 | /* Outputs become available. */ | |
61f15c48 RH |
4028 | if (opc == INDEX_op_mov_i32 || opc == INDEX_op_mov_i64) { |
4029 | arg_ts = arg_temp(op->args[0]); | |
b83eabea | 4030 | dir_ts = arg_ts->state_ptr; |
61f15c48 RH |
4031 | if (dir_ts) { |
4032 | op->args[0] = temp_arg(dir_ts); | |
4033 | changes = true; | |
4034 | ||
4035 | /* The output is now live and modified. */ | |
4036 | arg_ts->state = 0; | |
4037 | ||
4038 | if (NEED_SYNC_ARG(0)) { | |
4039 | TCGOpcode sopc = (arg_ts->type == TCG_TYPE_I32 | |
4040 | ? INDEX_op_st_i32 | |
4041 | : INDEX_op_st_i64); | |
d4478943 | 4042 | TCGOp *sop = tcg_op_insert_after(s, op, sopc, 3); |
61f15c48 RH |
4043 | TCGTemp *out_ts = dir_ts; |
4044 | ||
4045 | if (IS_DEAD_ARG(0)) { | |
4046 | out_ts = arg_temp(op->args[1]); | |
4047 | arg_ts->state = TS_DEAD; | |
4048 | tcg_op_remove(s, op); | |
4049 | } else { | |
4050 | arg_ts->state = TS_MEM; | |
4051 | } | |
4052 | ||
4053 | sop->args[0] = temp_arg(out_ts); | |
4054 | sop->args[1] = temp_arg(arg_ts->mem_base); | |
4055 | sop->args[2] = arg_ts->mem_offset; | |
4056 | } else { | |
4057 | tcg_debug_assert(!IS_DEAD_ARG(0)); | |
4058 | } | |
5a18407f | 4059 | } |
61f15c48 RH |
4060 | } else { |
4061 | for (i = 0; i < nb_oargs; i++) { | |
4062 | arg_ts = arg_temp(op->args[i]); | |
4063 | dir_ts = arg_ts->state_ptr; | |
4064 | if (!dir_ts) { | |
4065 | continue; | |
4066 | } | |
4067 | op->args[i] = temp_arg(dir_ts); | |
4068 | changes = true; | |
5a18407f | 4069 | |
61f15c48 RH |
4070 | /* The output is now live and modified. */ |
4071 | arg_ts->state = 0; | |
5a18407f | 4072 | |
61f15c48 RH |
4073 | /* Sync outputs upon their last write. */ |
4074 | if (NEED_SYNC_ARG(i)) { | |
4075 | TCGOpcode sopc = (arg_ts->type == TCG_TYPE_I32 | |
4076 | ? INDEX_op_st_i32 | |
4077 | : INDEX_op_st_i64); | |
d4478943 | 4078 | TCGOp *sop = tcg_op_insert_after(s, op, sopc, 3); |
5a18407f | 4079 | |
61f15c48 RH |
4080 | sop->args[0] = temp_arg(dir_ts); |
4081 | sop->args[1] = temp_arg(arg_ts->mem_base); | |
4082 | sop->args[2] = arg_ts->mem_offset; | |
5a18407f | 4083 | |
61f15c48 RH |
4084 | arg_ts->state = TS_MEM; |
4085 | } | |
4086 | /* Drop outputs that are dead. */ | |
4087 | if (IS_DEAD_ARG(i)) { | |
4088 | arg_ts->state = TS_DEAD; | |
4089 | } | |
5a18407f RH |
4090 | } |
4091 | } | |
4092 | } | |
4093 | ||
4094 | return changes; | |
4095 | } | |
4096 | ||
2272e4a7 | 4097 | static void temp_allocate_frame(TCGContext *s, TCGTemp *ts) |
c896fe29 | 4098 | { |
31c96417 | 4099 | intptr_t off; |
273eb50c | 4100 | int size, align; |
c1c09194 | 4101 | |
273eb50c RH |
4102 | /* When allocating an object, look at the full type. */ |
4103 | size = tcg_type_size(ts->base_type); | |
4104 | switch (ts->base_type) { | |
c1c09194 | 4105 | case TCG_TYPE_I32: |
31c96417 | 4106 | align = 4; |
c1c09194 RH |
4107 | break; |
4108 | case TCG_TYPE_I64: | |
4109 | case TCG_TYPE_V64: | |
31c96417 | 4110 | align = 8; |
c1c09194 | 4111 | break; |
43eef72f | 4112 | case TCG_TYPE_I128: |
c1c09194 | 4113 | case TCG_TYPE_V128: |
c1c09194 | 4114 | case TCG_TYPE_V256: |
43eef72f RH |
4115 | /* |
4116 | * Note that we do not require aligned storage for V256, | |
4117 | * and that we provide alignment for I128 to match V128, | |
4118 | * even if that's above what the host ABI requires. | |
4119 | */ | |
31c96417 | 4120 | align = 16; |
c1c09194 RH |
4121 | break; |
4122 | default: | |
4123 | g_assert_not_reached(); | |
b591dc59 | 4124 | } |
c1c09194 | 4125 | |
b9537d59 RH |
4126 | /* |
4127 | * Assume the stack is sufficiently aligned. | |
4128 | * This affects e.g. ARM NEON, where we have 8 byte stack alignment | |
4129 | * and do not require 16 byte vector alignment. This seems slightly | |
4130 | * easier than fully parameterizing the above switch statement. | |
4131 | */ | |
4132 | align = MIN(TCG_TARGET_STACK_ALIGN, align); | |
c1c09194 | 4133 | off = ROUND_UP(s->current_frame_offset, align); |
732d5897 RH |
4134 | |
4135 | /* If we've exhausted the stack frame, restart with a smaller TB. */ | |
4136 | if (off + size > s->frame_end) { | |
4137 | tcg_raise_tb_overflow(s); | |
4138 | } | |
c1c09194 | 4139 | s->current_frame_offset = off + size; |
9defd1bd | 4140 | #if defined(__sparc__) |
273eb50c | 4141 | off += TCG_TARGET_STACK_BIAS; |
9defd1bd | 4142 | #endif |
273eb50c RH |
4143 | |
4144 | /* If the object was subdivided, assign memory to all the parts. */ | |
4145 | if (ts->base_type != ts->type) { | |
4146 | int part_size = tcg_type_size(ts->type); | |
4147 | int part_count = size / part_size; | |
4148 | ||
4149 | /* | |
4150 | * Each part is allocated sequentially in tcg_temp_new_internal. | |
4151 | * Jump back to the first part by subtracting the current index. | |
4152 | */ | |
4153 | ts -= ts->temp_subindex; | |
4154 | for (int i = 0; i < part_count; ++i) { | |
4155 | ts[i].mem_offset = off + i * part_size; | |
4156 | ts[i].mem_base = s->frame_temp; | |
4157 | ts[i].mem_allocated = 1; | |
4158 | } | |
4159 | } else { | |
4160 | ts->mem_offset = off; | |
4161 | ts->mem_base = s->frame_temp; | |
4162 | ts->mem_allocated = 1; | |
4163 | } | |
c896fe29 FB |
4164 | } |
4165 | ||
098859f1 RH |
4166 | /* Assign @reg to @ts, and update reg_to_temp[]. */ |
4167 | static void set_temp_val_reg(TCGContext *s, TCGTemp *ts, TCGReg reg) | |
4168 | { | |
4169 | if (ts->val_type == TEMP_VAL_REG) { | |
4170 | TCGReg old = ts->reg; | |
4171 | tcg_debug_assert(s->reg_to_temp[old] == ts); | |
4172 | if (old == reg) { | |
4173 | return; | |
4174 | } | |
4175 | s->reg_to_temp[old] = NULL; | |
4176 | } | |
4177 | tcg_debug_assert(s->reg_to_temp[reg] == NULL); | |
4178 | s->reg_to_temp[reg] = ts; | |
4179 | ts->val_type = TEMP_VAL_REG; | |
4180 | ts->reg = reg; | |
4181 | } | |
4182 | ||
4183 | /* Assign a non-register value type to @ts, and update reg_to_temp[]. */ | |
4184 | static void set_temp_val_nonreg(TCGContext *s, TCGTemp *ts, TCGTempVal type) | |
4185 | { | |
4186 | tcg_debug_assert(type != TEMP_VAL_REG); | |
4187 | if (ts->val_type == TEMP_VAL_REG) { | |
4188 | TCGReg reg = ts->reg; | |
4189 | tcg_debug_assert(s->reg_to_temp[reg] == ts); | |
4190 | s->reg_to_temp[reg] = NULL; | |
4191 | } | |
4192 | ts->val_type = type; | |
4193 | } | |
4194 | ||
b722452a | 4195 | static void temp_load(TCGContext *, TCGTemp *, TCGRegSet, TCGRegSet, TCGRegSet); |
b3915dbb | 4196 | |
59d7c14e RH |
4197 | /* Mark a temporary as free or dead. If 'free_or_dead' is negative, |
4198 | mark it free; otherwise mark it dead. */ | |
4199 | static void temp_free_or_dead(TCGContext *s, TCGTemp *ts, int free_or_dead) | |
7f6ceedf | 4200 | { |
c0522136 RH |
4201 | TCGTempVal new_type; |
4202 | ||
4203 | switch (ts->kind) { | |
4204 | case TEMP_FIXED: | |
59d7c14e | 4205 | return; |
c0522136 | 4206 | case TEMP_GLOBAL: |
f57c6915 | 4207 | case TEMP_TB: |
c0522136 RH |
4208 | new_type = TEMP_VAL_MEM; |
4209 | break; | |
c7482438 | 4210 | case TEMP_EBB: |
c0522136 RH |
4211 | new_type = free_or_dead < 0 ? TEMP_VAL_MEM : TEMP_VAL_DEAD; |
4212 | break; | |
4213 | case TEMP_CONST: | |
4214 | new_type = TEMP_VAL_CONST; | |
4215 | break; | |
4216 | default: | |
4217 | g_assert_not_reached(); | |
59d7c14e | 4218 | } |
098859f1 | 4219 | set_temp_val_nonreg(s, ts, new_type); |
59d7c14e | 4220 | } |
7f6ceedf | 4221 | |
59d7c14e RH |
4222 | /* Mark a temporary as dead. */ |
4223 | static inline void temp_dead(TCGContext *s, TCGTemp *ts) | |
4224 | { | |
4225 | temp_free_or_dead(s, ts, 1); | |
4226 | } | |
4227 | ||
4228 | /* Sync a temporary to memory. 'allocated_regs' is used in case a temporary | |
4229 | registers needs to be allocated to store a constant. If 'free_or_dead' | |
4230 | is non-zero, subsequently release the temporary; if it is positive, the | |
4231 | temp is dead; if it is negative, the temp is free. */ | |
98b4e186 RH |
4232 | static void temp_sync(TCGContext *s, TCGTemp *ts, TCGRegSet allocated_regs, |
4233 | TCGRegSet preferred_regs, int free_or_dead) | |
59d7c14e | 4234 | { |
c0522136 | 4235 | if (!temp_readonly(ts) && !ts->mem_coherent) { |
7f6ceedf | 4236 | if (!ts->mem_allocated) { |
2272e4a7 | 4237 | temp_allocate_frame(s, ts); |
59d7c14e | 4238 | } |
59d7c14e RH |
4239 | switch (ts->val_type) { |
4240 | case TEMP_VAL_CONST: | |
4241 | /* If we're going to free the temp immediately, then we won't | |
4242 | require it later in a register, so attempt to store the | |
4243 | constant to memory directly. */ | |
4244 | if (free_or_dead | |
4245 | && tcg_out_sti(s, ts->type, ts->val, | |
4246 | ts->mem_base->reg, ts->mem_offset)) { | |
4247 | break; | |
4248 | } | |
4249 | temp_load(s, ts, tcg_target_available_regs[ts->type], | |
98b4e186 | 4250 | allocated_regs, preferred_regs); |
59d7c14e RH |
4251 | /* fallthrough */ |
4252 | ||
4253 | case TEMP_VAL_REG: | |
4254 | tcg_out_st(s, ts->type, ts->reg, | |
4255 | ts->mem_base->reg, ts->mem_offset); | |
4256 | break; | |
4257 | ||
4258 | case TEMP_VAL_MEM: | |
4259 | break; | |
4260 | ||
4261 | case TEMP_VAL_DEAD: | |
4262 | default: | |
732e89f4 | 4263 | g_assert_not_reached(); |
59d7c14e RH |
4264 | } |
4265 | ts->mem_coherent = 1; | |
4266 | } | |
4267 | if (free_or_dead) { | |
4268 | temp_free_or_dead(s, ts, free_or_dead); | |
7f6ceedf | 4269 | } |
7f6ceedf AJ |
4270 | } |
4271 | ||
c896fe29 | 4272 | /* free register 'reg' by spilling the corresponding temporary if necessary */ |
b3915dbb | 4273 | static void tcg_reg_free(TCGContext *s, TCGReg reg, TCGRegSet allocated_regs) |
c896fe29 | 4274 | { |
f8b2f202 | 4275 | TCGTemp *ts = s->reg_to_temp[reg]; |
f8b2f202 | 4276 | if (ts != NULL) { |
98b4e186 | 4277 | temp_sync(s, ts, allocated_regs, 0, -1); |
c896fe29 FB |
4278 | } |
4279 | } | |
4280 | ||
b016486e RH |
4281 | /** |
4282 | * tcg_reg_alloc: | |
4283 | * @required_regs: Set of registers in which we must allocate. | |
4284 | * @allocated_regs: Set of registers which must be avoided. | |
4285 | * @preferred_regs: Set of registers we should prefer. | |
4286 | * @rev: True if we search the registers in "indirect" order. | |
4287 | * | |
4288 | * The allocated register must be in @required_regs & ~@allocated_regs, | |
4289 | * but if we can put it in @preferred_regs we may save a move later. | |
4290 | */ | |
4291 | static TCGReg tcg_reg_alloc(TCGContext *s, TCGRegSet required_regs, | |
4292 | TCGRegSet allocated_regs, | |
4293 | TCGRegSet preferred_regs, bool rev) | |
c896fe29 | 4294 | { |
b016486e RH |
4295 | int i, j, f, n = ARRAY_SIZE(tcg_target_reg_alloc_order); |
4296 | TCGRegSet reg_ct[2]; | |
91478cef | 4297 | const int *order; |
c896fe29 | 4298 | |
b016486e RH |
4299 | reg_ct[1] = required_regs & ~allocated_regs; |
4300 | tcg_debug_assert(reg_ct[1] != 0); | |
4301 | reg_ct[0] = reg_ct[1] & preferred_regs; | |
4302 | ||
4303 | /* Skip the preferred_regs option if it cannot be satisfied, | |
4304 | or if the preference made no difference. */ | |
4305 | f = reg_ct[0] == 0 || reg_ct[0] == reg_ct[1]; | |
4306 | ||
91478cef | 4307 | order = rev ? indirect_reg_alloc_order : tcg_target_reg_alloc_order; |
c896fe29 | 4308 | |
b016486e RH |
4309 | /* Try free registers, preferences first. */ |
4310 | for (j = f; j < 2; j++) { | |
4311 | TCGRegSet set = reg_ct[j]; | |
4312 | ||
4313 | if (tcg_regset_single(set)) { | |
4314 | /* One register in the set. */ | |
4315 | TCGReg reg = tcg_regset_first(set); | |
4316 | if (s->reg_to_temp[reg] == NULL) { | |
4317 | return reg; | |
4318 | } | |
4319 | } else { | |
4320 | for (i = 0; i < n; i++) { | |
4321 | TCGReg reg = order[i]; | |
4322 | if (s->reg_to_temp[reg] == NULL && | |
4323 | tcg_regset_test_reg(set, reg)) { | |
4324 | return reg; | |
4325 | } | |
4326 | } | |
4327 | } | |
c896fe29 FB |
4328 | } |
4329 | ||
b016486e RH |
4330 | /* We must spill something. */ |
4331 | for (j = f; j < 2; j++) { | |
4332 | TCGRegSet set = reg_ct[j]; | |
4333 | ||
4334 | if (tcg_regset_single(set)) { | |
4335 | /* One register in the set. */ | |
4336 | TCGReg reg = tcg_regset_first(set); | |
b3915dbb | 4337 | tcg_reg_free(s, reg, allocated_regs); |
c896fe29 | 4338 | return reg; |
b016486e RH |
4339 | } else { |
4340 | for (i = 0; i < n; i++) { | |
4341 | TCGReg reg = order[i]; | |
4342 | if (tcg_regset_test_reg(set, reg)) { | |
4343 | tcg_reg_free(s, reg, allocated_regs); | |
4344 | return reg; | |
4345 | } | |
4346 | } | |
c896fe29 FB |
4347 | } |
4348 | } | |
4349 | ||
732e89f4 | 4350 | g_assert_not_reached(); |
c896fe29 FB |
4351 | } |
4352 | ||
29f5e925 RH |
4353 | static TCGReg tcg_reg_alloc_pair(TCGContext *s, TCGRegSet required_regs, |
4354 | TCGRegSet allocated_regs, | |
4355 | TCGRegSet preferred_regs, bool rev) | |
4356 | { | |
4357 | int i, j, k, fmin, n = ARRAY_SIZE(tcg_target_reg_alloc_order); | |
4358 | TCGRegSet reg_ct[2]; | |
4359 | const int *order; | |
4360 | ||
4361 | /* Ensure that if I is not in allocated_regs, I+1 is not either. */ | |
4362 | reg_ct[1] = required_regs & ~(allocated_regs | (allocated_regs >> 1)); | |
4363 | tcg_debug_assert(reg_ct[1] != 0); | |
4364 | reg_ct[0] = reg_ct[1] & preferred_regs; | |
4365 | ||
4366 | order = rev ? indirect_reg_alloc_order : tcg_target_reg_alloc_order; | |
4367 | ||
4368 | /* | |
4369 | * Skip the preferred_regs option if it cannot be satisfied, | |
4370 | * or if the preference made no difference. | |
4371 | */ | |
4372 | k = reg_ct[0] == 0 || reg_ct[0] == reg_ct[1]; | |
4373 | ||
4374 | /* | |
4375 | * Minimize the number of flushes by looking for 2 free registers first, | |
4376 | * then a single flush, then two flushes. | |
4377 | */ | |
4378 | for (fmin = 2; fmin >= 0; fmin--) { | |
4379 | for (j = k; j < 2; j++) { | |
4380 | TCGRegSet set = reg_ct[j]; | |
4381 | ||
4382 | for (i = 0; i < n; i++) { | |
4383 | TCGReg reg = order[i]; | |
4384 | ||
4385 | if (tcg_regset_test_reg(set, reg)) { | |
4386 | int f = !s->reg_to_temp[reg] + !s->reg_to_temp[reg + 1]; | |
4387 | if (f >= fmin) { | |
4388 | tcg_reg_free(s, reg, allocated_regs); | |
4389 | tcg_reg_free(s, reg + 1, allocated_regs); | |
4390 | return reg; | |
4391 | } | |
4392 | } | |
4393 | } | |
4394 | } | |
4395 | } | |
732e89f4 | 4396 | g_assert_not_reached(); |
29f5e925 RH |
4397 | } |
4398 | ||
40ae5c62 RH |
4399 | /* Make sure the temporary is in a register. If needed, allocate the register |
4400 | from DESIRED while avoiding ALLOCATED. */ | |
4401 | static void temp_load(TCGContext *s, TCGTemp *ts, TCGRegSet desired_regs, | |
b722452a | 4402 | TCGRegSet allocated_regs, TCGRegSet preferred_regs) |
40ae5c62 RH |
4403 | { |
4404 | TCGReg reg; | |
4405 | ||
4406 | switch (ts->val_type) { | |
4407 | case TEMP_VAL_REG: | |
4408 | return; | |
4409 | case TEMP_VAL_CONST: | |
b016486e | 4410 | reg = tcg_reg_alloc(s, desired_regs, allocated_regs, |
b722452a | 4411 | preferred_regs, ts->indirect_base); |
0a6a8bc8 RH |
4412 | if (ts->type <= TCG_TYPE_I64) { |
4413 | tcg_out_movi(s, ts->type, reg, ts->val); | |
4414 | } else { | |
4e186175 RH |
4415 | uint64_t val = ts->val; |
4416 | MemOp vece = MO_64; | |
4417 | ||
4418 | /* | |
4419 | * Find the minimal vector element that matches the constant. | |
4420 | * The targets will, in general, have to do this search anyway, | |
4421 | * do this generically. | |
4422 | */ | |
4e186175 RH |
4423 | if (val == dup_const(MO_8, val)) { |
4424 | vece = MO_8; | |
4425 | } else if (val == dup_const(MO_16, val)) { | |
4426 | vece = MO_16; | |
0b4286dd | 4427 | } else if (val == dup_const(MO_32, val)) { |
4e186175 RH |
4428 | vece = MO_32; |
4429 | } | |
4430 | ||
4431 | tcg_out_dupi_vec(s, ts->type, vece, reg, ts->val); | |
0a6a8bc8 | 4432 | } |
40ae5c62 RH |
4433 | ts->mem_coherent = 0; |
4434 | break; | |
4435 | case TEMP_VAL_MEM: | |
b016486e | 4436 | reg = tcg_reg_alloc(s, desired_regs, allocated_regs, |
b722452a | 4437 | preferred_regs, ts->indirect_base); |
40ae5c62 RH |
4438 | tcg_out_ld(s, ts->type, reg, ts->mem_base->reg, ts->mem_offset); |
4439 | ts->mem_coherent = 1; | |
4440 | break; | |
4441 | case TEMP_VAL_DEAD: | |
4442 | default: | |
732e89f4 | 4443 | g_assert_not_reached(); |
40ae5c62 | 4444 | } |
098859f1 | 4445 | set_temp_val_reg(s, ts, reg); |
40ae5c62 RH |
4446 | } |
4447 | ||
59d7c14e RH |
4448 | /* Save a temporary to memory. 'allocated_regs' is used in case a |
4449 | temporary registers needs to be allocated to store a constant. */ | |
4450 | static void temp_save(TCGContext *s, TCGTemp *ts, TCGRegSet allocated_regs) | |
1ad80729 | 4451 | { |
5a18407f RH |
4452 | /* The liveness analysis already ensures that globals are back |
4453 | in memory. Keep an tcg_debug_assert for safety. */ | |
e01fa97d | 4454 | tcg_debug_assert(ts->val_type == TEMP_VAL_MEM || temp_readonly(ts)); |
1ad80729 AJ |
4455 | } |
4456 | ||
9814dd27 | 4457 | /* save globals to their canonical location and assume they can be |
e8996ee0 FB |
4458 | modified be the following code. 'allocated_regs' is used in case a |
4459 | temporary registers needs to be allocated to store a constant. */ | |
4460 | static void save_globals(TCGContext *s, TCGRegSet allocated_regs) | |
c896fe29 | 4461 | { |
ac3b8891 | 4462 | int i, n; |
c896fe29 | 4463 | |
ac3b8891 | 4464 | for (i = 0, n = s->nb_globals; i < n; i++) { |
b13eb728 | 4465 | temp_save(s, &s->temps[i], allocated_regs); |
c896fe29 | 4466 | } |
e5097dc8 FB |
4467 | } |
4468 | ||
3d5c5f87 AJ |
4469 | /* sync globals to their canonical location and assume they can be |
4470 | read by the following code. 'allocated_regs' is used in case a | |
4471 | temporary registers needs to be allocated to store a constant. */ | |
4472 | static void sync_globals(TCGContext *s, TCGRegSet allocated_regs) | |
4473 | { | |
ac3b8891 | 4474 | int i, n; |
3d5c5f87 | 4475 | |
ac3b8891 | 4476 | for (i = 0, n = s->nb_globals; i < n; i++) { |
12b9b11a | 4477 | TCGTemp *ts = &s->temps[i]; |
5a18407f | 4478 | tcg_debug_assert(ts->val_type != TEMP_VAL_REG |
ee17db83 | 4479 | || ts->kind == TEMP_FIXED |
5a18407f | 4480 | || ts->mem_coherent); |
3d5c5f87 AJ |
4481 | } |
4482 | } | |
4483 | ||
e5097dc8 | 4484 | /* at the end of a basic block, we assume all temporaries are dead and |
e8996ee0 FB |
4485 | all globals are stored at their canonical location. */ |
4486 | static void tcg_reg_alloc_bb_end(TCGContext *s, TCGRegSet allocated_regs) | |
e5097dc8 | 4487 | { |
e5097dc8 FB |
4488 | int i; |
4489 | ||
b13eb728 RH |
4490 | for (i = s->nb_globals; i < s->nb_temps; i++) { |
4491 | TCGTemp *ts = &s->temps[i]; | |
c0522136 RH |
4492 | |
4493 | switch (ts->kind) { | |
f57c6915 | 4494 | case TEMP_TB: |
b13eb728 | 4495 | temp_save(s, ts, allocated_regs); |
c0522136 | 4496 | break; |
c7482438 | 4497 | case TEMP_EBB: |
5a18407f RH |
4498 | /* The liveness analysis already ensures that temps are dead. |
4499 | Keep an tcg_debug_assert for safety. */ | |
4500 | tcg_debug_assert(ts->val_type == TEMP_VAL_DEAD); | |
c0522136 RH |
4501 | break; |
4502 | case TEMP_CONST: | |
4503 | /* Similarly, we should have freed any allocated register. */ | |
4504 | tcg_debug_assert(ts->val_type == TEMP_VAL_CONST); | |
4505 | break; | |
4506 | default: | |
4507 | g_assert_not_reached(); | |
c896fe29 FB |
4508 | } |
4509 | } | |
e8996ee0 FB |
4510 | |
4511 | save_globals(s, allocated_regs); | |
c896fe29 FB |
4512 | } |
4513 | ||
b4cb76e6 | 4514 | /* |
c7482438 RH |
4515 | * At a conditional branch, we assume all temporaries are dead unless |
4516 | * explicitly live-across-conditional-branch; all globals and local | |
4517 | * temps are synced to their location. | |
b4cb76e6 RH |
4518 | */ |
4519 | static void tcg_reg_alloc_cbranch(TCGContext *s, TCGRegSet allocated_regs) | |
4520 | { | |
4521 | sync_globals(s, allocated_regs); | |
4522 | ||
4523 | for (int i = s->nb_globals; i < s->nb_temps; i++) { | |
4524 | TCGTemp *ts = &s->temps[i]; | |
4525 | /* | |
4526 | * The liveness analysis already ensures that temps are dead. | |
4527 | * Keep tcg_debug_asserts for safety. | |
4528 | */ | |
c0522136 | 4529 | switch (ts->kind) { |
f57c6915 | 4530 | case TEMP_TB: |
b4cb76e6 | 4531 | tcg_debug_assert(ts->val_type != TEMP_VAL_REG || ts->mem_coherent); |
c0522136 | 4532 | break; |
c7482438 | 4533 | case TEMP_EBB: |
c0522136 RH |
4534 | case TEMP_CONST: |
4535 | break; | |
4536 | default: | |
4537 | g_assert_not_reached(); | |
b4cb76e6 RH |
4538 | } |
4539 | } | |
4540 | } | |
4541 | ||
bab1671f | 4542 | /* |
c58f4c97 | 4543 | * Specialized code generation for INDEX_op_mov_* with a constant. |
bab1671f | 4544 | */ |
0fe4fca4 | 4545 | static void tcg_reg_alloc_do_movi(TCGContext *s, TCGTemp *ots, |
ba87719c RH |
4546 | tcg_target_ulong val, TCGLifeData arg_life, |
4547 | TCGRegSet preferred_regs) | |
e8996ee0 | 4548 | { |
d63e3b6e | 4549 | /* ENV should not be modified. */ |
e01fa97d | 4550 | tcg_debug_assert(!temp_readonly(ots)); |
59d7c14e RH |
4551 | |
4552 | /* The movi is not explicitly generated here. */ | |
098859f1 | 4553 | set_temp_val_nonreg(s, ots, TEMP_VAL_CONST); |
59d7c14e RH |
4554 | ots->val = val; |
4555 | ots->mem_coherent = 0; | |
4556 | if (NEED_SYNC_ARG(0)) { | |
ba87719c | 4557 | temp_sync(s, ots, s->reserved_regs, preferred_regs, IS_DEAD_ARG(0)); |
59d7c14e | 4558 | } else if (IS_DEAD_ARG(0)) { |
f8bf00f1 | 4559 | temp_dead(s, ots); |
4c4e1ab2 | 4560 | } |
e8996ee0 FB |
4561 | } |
4562 | ||
bab1671f RH |
4563 | /* |
4564 | * Specialized code generation for INDEX_op_mov_*. | |
4565 | */ | |
dd186292 | 4566 | static void tcg_reg_alloc_mov(TCGContext *s, const TCGOp *op) |
c896fe29 | 4567 | { |
dd186292 | 4568 | const TCGLifeData arg_life = op->life; |
69e3706d | 4569 | TCGRegSet allocated_regs, preferred_regs; |
c896fe29 | 4570 | TCGTemp *ts, *ots; |
450445d5 | 4571 | TCGType otype, itype; |
098859f1 | 4572 | TCGReg oreg, ireg; |
c896fe29 | 4573 | |
d21369f5 | 4574 | allocated_regs = s->reserved_regs; |
31fd884b | 4575 | preferred_regs = output_pref(op, 0); |
43439139 RH |
4576 | ots = arg_temp(op->args[0]); |
4577 | ts = arg_temp(op->args[1]); | |
450445d5 | 4578 | |
d63e3b6e | 4579 | /* ENV should not be modified. */ |
e01fa97d | 4580 | tcg_debug_assert(!temp_readonly(ots)); |
d63e3b6e | 4581 | |
450445d5 RH |
4582 | /* Note that otype != itype for no-op truncation. */ |
4583 | otype = ots->type; | |
4584 | itype = ts->type; | |
c29c1d7e | 4585 | |
0fe4fca4 PB |
4586 | if (ts->val_type == TEMP_VAL_CONST) { |
4587 | /* propagate constant or generate sti */ | |
4588 | tcg_target_ulong val = ts->val; | |
4589 | if (IS_DEAD_ARG(1)) { | |
4590 | temp_dead(s, ts); | |
4591 | } | |
69e3706d | 4592 | tcg_reg_alloc_do_movi(s, ots, val, arg_life, preferred_regs); |
0fe4fca4 PB |
4593 | return; |
4594 | } | |
4595 | ||
4596 | /* If the source value is in memory we're going to be forced | |
4597 | to have it in a register in order to perform the copy. Copy | |
4598 | the SOURCE value into its own register first, that way we | |
4599 | don't have to reload SOURCE the next time it is used. */ | |
4600 | if (ts->val_type == TEMP_VAL_MEM) { | |
69e3706d RH |
4601 | temp_load(s, ts, tcg_target_available_regs[itype], |
4602 | allocated_regs, preferred_regs); | |
c29c1d7e | 4603 | } |
0fe4fca4 | 4604 | tcg_debug_assert(ts->val_type == TEMP_VAL_REG); |
098859f1 RH |
4605 | ireg = ts->reg; |
4606 | ||
d63e3b6e | 4607 | if (IS_DEAD_ARG(0)) { |
c29c1d7e AJ |
4608 | /* mov to a non-saved dead register makes no sense (even with |
4609 | liveness analysis disabled). */ | |
eabb7b91 | 4610 | tcg_debug_assert(NEED_SYNC_ARG(0)); |
c29c1d7e | 4611 | if (!ots->mem_allocated) { |
2272e4a7 | 4612 | temp_allocate_frame(s, ots); |
c29c1d7e | 4613 | } |
098859f1 | 4614 | tcg_out_st(s, otype, ireg, ots->mem_base->reg, ots->mem_offset); |
c29c1d7e | 4615 | if (IS_DEAD_ARG(1)) { |
f8bf00f1 | 4616 | temp_dead(s, ts); |
c29c1d7e | 4617 | } |
f8bf00f1 | 4618 | temp_dead(s, ots); |
098859f1 RH |
4619 | return; |
4620 | } | |
4621 | ||
4622 | if (IS_DEAD_ARG(1) && ts->kind != TEMP_FIXED) { | |
4623 | /* | |
4624 | * The mov can be suppressed. Kill input first, so that it | |
4625 | * is unlinked from reg_to_temp, then set the output to the | |
4626 | * reg that we saved from the input. | |
4627 | */ | |
4628 | temp_dead(s, ts); | |
4629 | oreg = ireg; | |
c29c1d7e | 4630 | } else { |
098859f1 RH |
4631 | if (ots->val_type == TEMP_VAL_REG) { |
4632 | oreg = ots->reg; | |
c896fe29 | 4633 | } else { |
098859f1 RH |
4634 | /* Make sure to not spill the input register during allocation. */ |
4635 | oreg = tcg_reg_alloc(s, tcg_target_available_regs[otype], | |
4636 | allocated_regs | ((TCGRegSet)1 << ireg), | |
4637 | preferred_regs, ots->indirect_base); | |
c896fe29 | 4638 | } |
098859f1 RH |
4639 | if (!tcg_out_mov(s, otype, oreg, ireg)) { |
4640 | /* | |
4641 | * Cross register class move not supported. | |
4642 | * Store the source register into the destination slot | |
4643 | * and leave the destination temp as TEMP_VAL_MEM. | |
4644 | */ | |
4645 | assert(!temp_readonly(ots)); | |
4646 | if (!ts->mem_allocated) { | |
4647 | temp_allocate_frame(s, ots); | |
4648 | } | |
4649 | tcg_out_st(s, ts->type, ireg, ots->mem_base->reg, ots->mem_offset); | |
4650 | set_temp_val_nonreg(s, ts, TEMP_VAL_MEM); | |
4651 | ots->mem_coherent = 1; | |
4652 | return; | |
c896fe29 | 4653 | } |
ec7a869d | 4654 | } |
098859f1 RH |
4655 | set_temp_val_reg(s, ots, oreg); |
4656 | ots->mem_coherent = 0; | |
4657 | ||
4658 | if (NEED_SYNC_ARG(0)) { | |
4659 | temp_sync(s, ots, allocated_regs, 0, 0); | |
4660 | } | |
c896fe29 FB |
4661 | } |
4662 | ||
bab1671f RH |
4663 | /* |
4664 | * Specialized code generation for INDEX_op_dup_vec. | |
4665 | */ | |
4666 | static void tcg_reg_alloc_dup(TCGContext *s, const TCGOp *op) | |
4667 | { | |
4668 | const TCGLifeData arg_life = op->life; | |
4669 | TCGRegSet dup_out_regs, dup_in_regs; | |
4670 | TCGTemp *its, *ots; | |
4671 | TCGType itype, vtype; | |
4672 | unsigned vece; | |
31c96417 | 4673 | int lowpart_ofs; |
bab1671f RH |
4674 | bool ok; |
4675 | ||
4676 | ots = arg_temp(op->args[0]); | |
4677 | its = arg_temp(op->args[1]); | |
4678 | ||
4679 | /* ENV should not be modified. */ | |
e01fa97d | 4680 | tcg_debug_assert(!temp_readonly(ots)); |
bab1671f RH |
4681 | |
4682 | itype = its->type; | |
4683 | vece = TCGOP_VECE(op); | |
4684 | vtype = TCGOP_VECL(op) + TCG_TYPE_V64; | |
4685 | ||
4686 | if (its->val_type == TEMP_VAL_CONST) { | |
4687 | /* Propagate constant via movi -> dupi. */ | |
4688 | tcg_target_ulong val = its->val; | |
4689 | if (IS_DEAD_ARG(1)) { | |
4690 | temp_dead(s, its); | |
4691 | } | |
31fd884b | 4692 | tcg_reg_alloc_do_movi(s, ots, val, arg_life, output_pref(op, 0)); |
bab1671f RH |
4693 | return; |
4694 | } | |
4695 | ||
9be0d080 RH |
4696 | dup_out_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[0].regs; |
4697 | dup_in_regs = tcg_op_defs[INDEX_op_dup_vec].args_ct[1].regs; | |
bab1671f RH |
4698 | |
4699 | /* Allocate the output register now. */ | |
4700 | if (ots->val_type != TEMP_VAL_REG) { | |
4701 | TCGRegSet allocated_regs = s->reserved_regs; | |
098859f1 | 4702 | TCGReg oreg; |
bab1671f RH |
4703 | |
4704 | if (!IS_DEAD_ARG(1) && its->val_type == TEMP_VAL_REG) { | |
4705 | /* Make sure to not spill the input register. */ | |
4706 | tcg_regset_set_reg(allocated_regs, its->reg); | |
4707 | } | |
098859f1 | 4708 | oreg = tcg_reg_alloc(s, dup_out_regs, allocated_regs, |
31fd884b | 4709 | output_pref(op, 0), ots->indirect_base); |
098859f1 | 4710 | set_temp_val_reg(s, ots, oreg); |
bab1671f RH |
4711 | } |
4712 | ||
4713 | switch (its->val_type) { | |
4714 | case TEMP_VAL_REG: | |
4715 | /* | |
4716 | * The dup constriaints must be broad, covering all possible VECE. | |
4717 | * However, tcg_op_dup_vec() gets to see the VECE and we allow it | |
4718 | * to fail, indicating that extra moves are required for that case. | |
4719 | */ | |
4720 | if (tcg_regset_test_reg(dup_in_regs, its->reg)) { | |
4721 | if (tcg_out_dup_vec(s, vtype, vece, ots->reg, its->reg)) { | |
4722 | goto done; | |
4723 | } | |
4724 | /* Try again from memory or a vector input register. */ | |
4725 | } | |
4726 | if (!its->mem_coherent) { | |
4727 | /* | |
4728 | * The input register is not synced, and so an extra store | |
4729 | * would be required to use memory. Attempt an integer-vector | |
4730 | * register move first. We do not have a TCGRegSet for this. | |
4731 | */ | |
4732 | if (tcg_out_mov(s, itype, ots->reg, its->reg)) { | |
4733 | break; | |
4734 | } | |
4735 | /* Sync the temp back to its slot and load from there. */ | |
4736 | temp_sync(s, its, s->reserved_regs, 0, 0); | |
4737 | } | |
4738 | /* fall through */ | |
4739 | ||
4740 | case TEMP_VAL_MEM: | |
31c96417 RH |
4741 | lowpart_ofs = 0; |
4742 | if (HOST_BIG_ENDIAN) { | |
4743 | lowpart_ofs = tcg_type_size(itype) - (1 << vece); | |
4744 | } | |
d6ecb4a9 | 4745 | if (tcg_out_dupm_vec(s, vtype, vece, ots->reg, its->mem_base->reg, |
31c96417 | 4746 | its->mem_offset + lowpart_ofs)) { |
d6ecb4a9 RH |
4747 | goto done; |
4748 | } | |
098859f1 | 4749 | /* Load the input into the destination vector register. */ |
bab1671f RH |
4750 | tcg_out_ld(s, itype, ots->reg, its->mem_base->reg, its->mem_offset); |
4751 | break; | |
4752 | ||
4753 | default: | |
4754 | g_assert_not_reached(); | |
4755 | } | |
4756 | ||
4757 | /* We now have a vector input register, so dup must succeed. */ | |
4758 | ok = tcg_out_dup_vec(s, vtype, vece, ots->reg, ots->reg); | |
4759 | tcg_debug_assert(ok); | |
4760 | ||
4761 | done: | |
36f5539c | 4762 | ots->mem_coherent = 0; |
bab1671f RH |
4763 | if (IS_DEAD_ARG(1)) { |
4764 | temp_dead(s, its); | |
4765 | } | |
4766 | if (NEED_SYNC_ARG(0)) { | |
4767 | temp_sync(s, ots, s->reserved_regs, 0, 0); | |
4768 | } | |
4769 | if (IS_DEAD_ARG(0)) { | |
4770 | temp_dead(s, ots); | |
4771 | } | |
4772 | } | |
4773 | ||
dd186292 | 4774 | static void tcg_reg_alloc_op(TCGContext *s, const TCGOp *op) |
c896fe29 | 4775 | { |
dd186292 RH |
4776 | const TCGLifeData arg_life = op->life; |
4777 | const TCGOpDef * const def = &tcg_op_defs[op->opc]; | |
82790a87 RH |
4778 | TCGRegSet i_allocated_regs; |
4779 | TCGRegSet o_allocated_regs; | |
b6638662 RH |
4780 | int i, k, nb_iargs, nb_oargs; |
4781 | TCGReg reg; | |
c896fe29 FB |
4782 | TCGArg arg; |
4783 | const TCGArgConstraint *arg_ct; | |
4784 | TCGTemp *ts; | |
4785 | TCGArg new_args[TCG_MAX_OP_ARGS]; | |
4786 | int const_args[TCG_MAX_OP_ARGS]; | |
4787 | ||
4788 | nb_oargs = def->nb_oargs; | |
4789 | nb_iargs = def->nb_iargs; | |
4790 | ||
4791 | /* copy constants */ | |
a813e36f | 4792 | memcpy(new_args + nb_oargs + nb_iargs, |
dd186292 | 4793 | op->args + nb_oargs + nb_iargs, |
c896fe29 FB |
4794 | sizeof(TCGArg) * def->nb_cargs); |
4795 | ||
d21369f5 RH |
4796 | i_allocated_regs = s->reserved_regs; |
4797 | o_allocated_regs = s->reserved_regs; | |
82790a87 | 4798 | |
a813e36f | 4799 | /* satisfy input constraints */ |
dd186292 | 4800 | for (k = 0; k < nb_iargs; k++) { |
29f5e925 RH |
4801 | TCGRegSet i_preferred_regs, i_required_regs; |
4802 | bool allocate_new_reg, copyto_new_reg; | |
4803 | TCGTemp *ts2; | |
4804 | int i1, i2; | |
d62816f2 | 4805 | |
66792f90 | 4806 | i = def->args_ct[nb_oargs + k].sort_index; |
dd186292 | 4807 | arg = op->args[i]; |
c896fe29 | 4808 | arg_ct = &def->args_ct[i]; |
43439139 | 4809 | ts = arg_temp(arg); |
40ae5c62 RH |
4810 | |
4811 | if (ts->val_type == TEMP_VAL_CONST | |
ebe92db2 | 4812 | && tcg_target_const_match(ts->val, ts->type, arg_ct->ct, TCGOP_VECE(op))) { |
40ae5c62 RH |
4813 | /* constant is OK for instruction */ |
4814 | const_args[i] = 1; | |
4815 | new_args[i] = ts->val; | |
d62816f2 | 4816 | continue; |
c896fe29 | 4817 | } |
40ae5c62 | 4818 | |
1c1824dc RH |
4819 | reg = ts->reg; |
4820 | i_preferred_regs = 0; | |
29f5e925 | 4821 | i_required_regs = arg_ct->regs; |
1c1824dc | 4822 | allocate_new_reg = false; |
29f5e925 RH |
4823 | copyto_new_reg = false; |
4824 | ||
4825 | switch (arg_ct->pair) { | |
4826 | case 0: /* not paired */ | |
4827 | if (arg_ct->ialias) { | |
31fd884b | 4828 | i_preferred_regs = output_pref(op, arg_ct->alias_index); |
29f5e925 RH |
4829 | |
4830 | /* | |
4831 | * If the input is readonly, then it cannot also be an | |
4832 | * output and aliased to itself. If the input is not | |
4833 | * dead after the instruction, we must allocate a new | |
4834 | * register and move it. | |
4835 | */ | |
22d2e535 IL |
4836 | if (temp_readonly(ts) || !IS_DEAD_ARG(i) |
4837 | || def->args_ct[arg_ct->alias_index].newreg) { | |
29f5e925 RH |
4838 | allocate_new_reg = true; |
4839 | } else if (ts->val_type == TEMP_VAL_REG) { | |
4840 | /* | |
4841 | * Check if the current register has already been | |
4842 | * allocated for another input. | |
4843 | */ | |
4844 | allocate_new_reg = | |
4845 | tcg_regset_test_reg(i_allocated_regs, reg); | |
4846 | } | |
4847 | } | |
4848 | if (!allocate_new_reg) { | |
4849 | temp_load(s, ts, i_required_regs, i_allocated_regs, | |
4850 | i_preferred_regs); | |
4851 | reg = ts->reg; | |
4852 | allocate_new_reg = !tcg_regset_test_reg(i_required_regs, reg); | |
4853 | } | |
4854 | if (allocate_new_reg) { | |
4855 | /* | |
4856 | * Allocate a new register matching the constraint | |
4857 | * and move the temporary register into it. | |
4858 | */ | |
4859 | temp_load(s, ts, tcg_target_available_regs[ts->type], | |
4860 | i_allocated_regs, 0); | |
4861 | reg = tcg_reg_alloc(s, i_required_regs, i_allocated_regs, | |
4862 | i_preferred_regs, ts->indirect_base); | |
4863 | copyto_new_reg = true; | |
4864 | } | |
4865 | break; | |
4866 | ||
4867 | case 1: | |
4868 | /* First of an input pair; if i1 == i2, the second is an output. */ | |
4869 | i1 = i; | |
4870 | i2 = arg_ct->pair_index; | |
4871 | ts2 = i1 != i2 ? arg_temp(op->args[i2]) : NULL; | |
4872 | ||
4873 | /* | |
4874 | * It is easier to default to allocating a new pair | |
4875 | * and to identify a few cases where it's not required. | |
4876 | */ | |
4877 | if (arg_ct->ialias) { | |
31fd884b | 4878 | i_preferred_regs = output_pref(op, arg_ct->alias_index); |
29f5e925 RH |
4879 | if (IS_DEAD_ARG(i1) && |
4880 | IS_DEAD_ARG(i2) && | |
4881 | !temp_readonly(ts) && | |
4882 | ts->val_type == TEMP_VAL_REG && | |
4883 | ts->reg < TCG_TARGET_NB_REGS - 1 && | |
4884 | tcg_regset_test_reg(i_required_regs, reg) && | |
4885 | !tcg_regset_test_reg(i_allocated_regs, reg) && | |
4886 | !tcg_regset_test_reg(i_allocated_regs, reg + 1) && | |
4887 | (ts2 | |
4888 | ? ts2->val_type == TEMP_VAL_REG && | |
4889 | ts2->reg == reg + 1 && | |
4890 | !temp_readonly(ts2) | |
4891 | : s->reg_to_temp[reg + 1] == NULL)) { | |
4892 | break; | |
4893 | } | |
4894 | } else { | |
4895 | /* Without aliasing, the pair must also be an input. */ | |
4896 | tcg_debug_assert(ts2); | |
4897 | if (ts->val_type == TEMP_VAL_REG && | |
4898 | ts2->val_type == TEMP_VAL_REG && | |
4899 | ts2->reg == reg + 1 && | |
4900 | tcg_regset_test_reg(i_required_regs, reg)) { | |
4901 | break; | |
4902 | } | |
4903 | } | |
4904 | reg = tcg_reg_alloc_pair(s, i_required_regs, i_allocated_regs, | |
4905 | 0, ts->indirect_base); | |
4906 | goto do_pair; | |
4907 | ||
4908 | case 2: /* pair second */ | |
4909 | reg = new_args[arg_ct->pair_index] + 1; | |
4910 | goto do_pair; | |
1c1824dc | 4911 | |
29f5e925 RH |
4912 | case 3: /* ialias with second output, no first input */ |
4913 | tcg_debug_assert(arg_ct->ialias); | |
31fd884b | 4914 | i_preferred_regs = output_pref(op, arg_ct->alias_index); |
d62816f2 | 4915 | |
29f5e925 RH |
4916 | if (IS_DEAD_ARG(i) && |
4917 | !temp_readonly(ts) && | |
4918 | ts->val_type == TEMP_VAL_REG && | |
4919 | reg > 0 && | |
4920 | s->reg_to_temp[reg - 1] == NULL && | |
4921 | tcg_regset_test_reg(i_required_regs, reg) && | |
4922 | !tcg_regset_test_reg(i_allocated_regs, reg) && | |
4923 | !tcg_regset_test_reg(i_allocated_regs, reg - 1)) { | |
4924 | tcg_regset_set_reg(i_allocated_regs, reg - 1); | |
4925 | break; | |
4926 | } | |
4927 | reg = tcg_reg_alloc_pair(s, i_required_regs >> 1, | |
4928 | i_allocated_regs, 0, | |
4929 | ts->indirect_base); | |
4930 | tcg_regset_set_reg(i_allocated_regs, reg); | |
4931 | reg += 1; | |
4932 | goto do_pair; | |
4933 | ||
4934 | do_pair: | |
c0522136 | 4935 | /* |
29f5e925 RH |
4936 | * If an aliased input is not dead after the instruction, |
4937 | * we must allocate a new register and move it. | |
c0522136 | 4938 | */ |
29f5e925 RH |
4939 | if (arg_ct->ialias && (!IS_DEAD_ARG(i) || temp_readonly(ts))) { |
4940 | TCGRegSet t_allocated_regs = i_allocated_regs; | |
4941 | ||
1c1824dc | 4942 | /* |
29f5e925 RH |
4943 | * Because of the alias, and the continued life, make sure |
4944 | * that the temp is somewhere *other* than the reg pair, | |
4945 | * and we get a copy in reg. | |
1c1824dc | 4946 | */ |
29f5e925 RH |
4947 | tcg_regset_set_reg(t_allocated_regs, reg); |
4948 | tcg_regset_set_reg(t_allocated_regs, reg + 1); | |
4949 | if (ts->val_type == TEMP_VAL_REG && ts->reg == reg) { | |
4950 | /* If ts was already in reg, copy it somewhere else. */ | |
4951 | TCGReg nr; | |
4952 | bool ok; | |
4953 | ||
4954 | tcg_debug_assert(ts->kind != TEMP_FIXED); | |
4955 | nr = tcg_reg_alloc(s, tcg_target_available_regs[ts->type], | |
4956 | t_allocated_regs, 0, ts->indirect_base); | |
4957 | ok = tcg_out_mov(s, ts->type, nr, reg); | |
4958 | tcg_debug_assert(ok); | |
4959 | ||
4960 | set_temp_val_reg(s, ts, nr); | |
4961 | } else { | |
4962 | temp_load(s, ts, tcg_target_available_regs[ts->type], | |
4963 | t_allocated_regs, 0); | |
4964 | copyto_new_reg = true; | |
4965 | } | |
4966 | } else { | |
4967 | /* Preferably allocate to reg, otherwise copy. */ | |
4968 | i_required_regs = (TCGRegSet)1 << reg; | |
4969 | temp_load(s, ts, i_required_regs, i_allocated_regs, | |
4970 | i_preferred_regs); | |
4971 | copyto_new_reg = ts->reg != reg; | |
5ff9d6a4 | 4972 | } |
29f5e925 | 4973 | break; |
d62816f2 | 4974 | |
29f5e925 RH |
4975 | default: |
4976 | g_assert_not_reached(); | |
1c1824dc | 4977 | } |
d62816f2 | 4978 | |
29f5e925 | 4979 | if (copyto_new_reg) { |
78113e83 | 4980 | if (!tcg_out_mov(s, ts->type, reg, ts->reg)) { |
240c08d0 RH |
4981 | /* |
4982 | * Cross register class move not supported. Sync the | |
4983 | * temp back to its slot and load from there. | |
4984 | */ | |
4985 | temp_sync(s, ts, i_allocated_regs, 0, 0); | |
4986 | tcg_out_ld(s, ts->type, reg, | |
4987 | ts->mem_base->reg, ts->mem_offset); | |
78113e83 | 4988 | } |
c896fe29 | 4989 | } |
c896fe29 FB |
4990 | new_args[i] = reg; |
4991 | const_args[i] = 0; | |
82790a87 | 4992 | tcg_regset_set_reg(i_allocated_regs, reg); |
c896fe29 | 4993 | } |
a813e36f | 4994 | |
a52ad07e AJ |
4995 | /* mark dead temporaries and free the associated registers */ |
4996 | for (i = nb_oargs; i < nb_oargs + nb_iargs; i++) { | |
4997 | if (IS_DEAD_ARG(i)) { | |
43439139 | 4998 | temp_dead(s, arg_temp(op->args[i])); |
a52ad07e AJ |
4999 | } |
5000 | } | |
5001 | ||
b4cb76e6 RH |
5002 | if (def->flags & TCG_OPF_COND_BRANCH) { |
5003 | tcg_reg_alloc_cbranch(s, i_allocated_regs); | |
5004 | } else if (def->flags & TCG_OPF_BB_END) { | |
82790a87 | 5005 | tcg_reg_alloc_bb_end(s, i_allocated_regs); |
e8996ee0 | 5006 | } else { |
e8996ee0 | 5007 | if (def->flags & TCG_OPF_CALL_CLOBBER) { |
a813e36f | 5008 | /* XXX: permit generic clobber register list ? */ |
c8074023 RH |
5009 | for (i = 0; i < TCG_TARGET_NB_REGS; i++) { |
5010 | if (tcg_regset_test_reg(tcg_target_call_clobber_regs, i)) { | |
82790a87 | 5011 | tcg_reg_free(s, i, i_allocated_regs); |
e8996ee0 | 5012 | } |
c896fe29 | 5013 | } |
3d5c5f87 AJ |
5014 | } |
5015 | if (def->flags & TCG_OPF_SIDE_EFFECTS) { | |
5016 | /* sync globals if the op has side effects and might trigger | |
5017 | an exception. */ | |
82790a87 | 5018 | sync_globals(s, i_allocated_regs); |
c896fe29 | 5019 | } |
a813e36f | 5020 | |
e8996ee0 | 5021 | /* satisfy the output constraints */ |
e8996ee0 | 5022 | for(k = 0; k < nb_oargs; k++) { |
66792f90 | 5023 | i = def->args_ct[k].sort_index; |
dd186292 | 5024 | arg = op->args[i]; |
e8996ee0 | 5025 | arg_ct = &def->args_ct[i]; |
43439139 | 5026 | ts = arg_temp(arg); |
d63e3b6e RH |
5027 | |
5028 | /* ENV should not be modified. */ | |
e01fa97d | 5029 | tcg_debug_assert(!temp_readonly(ts)); |
d63e3b6e | 5030 | |
29f5e925 RH |
5031 | switch (arg_ct->pair) { |
5032 | case 0: /* not paired */ | |
5033 | if (arg_ct->oalias && !const_args[arg_ct->alias_index]) { | |
5034 | reg = new_args[arg_ct->alias_index]; | |
5035 | } else if (arg_ct->newreg) { | |
5036 | reg = tcg_reg_alloc(s, arg_ct->regs, | |
5037 | i_allocated_regs | o_allocated_regs, | |
31fd884b | 5038 | output_pref(op, k), ts->indirect_base); |
29f5e925 RH |
5039 | } else { |
5040 | reg = tcg_reg_alloc(s, arg_ct->regs, o_allocated_regs, | |
31fd884b | 5041 | output_pref(op, k), ts->indirect_base); |
29f5e925 RH |
5042 | } |
5043 | break; | |
5044 | ||
5045 | case 1: /* first of pair */ | |
29f5e925 RH |
5046 | if (arg_ct->oalias) { |
5047 | reg = new_args[arg_ct->alias_index]; | |
ca5bed07 RH |
5048 | } else if (arg_ct->newreg) { |
5049 | reg = tcg_reg_alloc_pair(s, arg_ct->regs, | |
5050 | i_allocated_regs | o_allocated_regs, | |
5051 | output_pref(op, k), | |
5052 | ts->indirect_base); | |
5053 | } else { | |
5054 | reg = tcg_reg_alloc_pair(s, arg_ct->regs, o_allocated_regs, | |
5055 | output_pref(op, k), | |
5056 | ts->indirect_base); | |
29f5e925 | 5057 | } |
29f5e925 RH |
5058 | break; |
5059 | ||
5060 | case 2: /* second of pair */ | |
29f5e925 RH |
5061 | if (arg_ct->oalias) { |
5062 | reg = new_args[arg_ct->alias_index]; | |
5063 | } else { | |
5064 | reg = new_args[arg_ct->pair_index] + 1; | |
5065 | } | |
5066 | break; | |
5067 | ||
5068 | case 3: /* first of pair, aliasing with a second input */ | |
5069 | tcg_debug_assert(!arg_ct->newreg); | |
5070 | reg = new_args[arg_ct->pair_index] - 1; | |
5071 | break; | |
5072 | ||
5073 | default: | |
5074 | g_assert_not_reached(); | |
c896fe29 | 5075 | } |
82790a87 | 5076 | tcg_regset_set_reg(o_allocated_regs, reg); |
098859f1 | 5077 | set_temp_val_reg(s, ts, reg); |
d63e3b6e | 5078 | ts->mem_coherent = 0; |
e8996ee0 | 5079 | new_args[i] = reg; |
c896fe29 | 5080 | } |
c896fe29 FB |
5081 | } |
5082 | ||
c896fe29 | 5083 | /* emit instruction */ |
678155b2 RH |
5084 | switch (op->opc) { |
5085 | case INDEX_op_ext8s_i32: | |
5086 | tcg_out_ext8s(s, TCG_TYPE_I32, new_args[0], new_args[1]); | |
5087 | break; | |
5088 | case INDEX_op_ext8s_i64: | |
5089 | tcg_out_ext8s(s, TCG_TYPE_I64, new_args[0], new_args[1]); | |
5090 | break; | |
d0e66c89 RH |
5091 | case INDEX_op_ext8u_i32: |
5092 | case INDEX_op_ext8u_i64: | |
5093 | tcg_out_ext8u(s, new_args[0], new_args[1]); | |
5094 | break; | |
753e42ea RH |
5095 | case INDEX_op_ext16s_i32: |
5096 | tcg_out_ext16s(s, TCG_TYPE_I32, new_args[0], new_args[1]); | |
5097 | break; | |
5098 | case INDEX_op_ext16s_i64: | |
5099 | tcg_out_ext16s(s, TCG_TYPE_I64, new_args[0], new_args[1]); | |
5100 | break; | |
379afdff RH |
5101 | case INDEX_op_ext16u_i32: |
5102 | case INDEX_op_ext16u_i64: | |
5103 | tcg_out_ext16u(s, new_args[0], new_args[1]); | |
5104 | break; | |
52bf3398 RH |
5105 | case INDEX_op_ext32s_i64: |
5106 | tcg_out_ext32s(s, new_args[0], new_args[1]); | |
5107 | break; | |
9ecf5f61 RH |
5108 | case INDEX_op_ext32u_i64: |
5109 | tcg_out_ext32u(s, new_args[0], new_args[1]); | |
5110 | break; | |
9c6aa274 RH |
5111 | case INDEX_op_ext_i32_i64: |
5112 | tcg_out_exts_i32_i64(s, new_args[0], new_args[1]); | |
5113 | break; | |
b9bfe000 RH |
5114 | case INDEX_op_extu_i32_i64: |
5115 | tcg_out_extu_i32_i64(s, new_args[0], new_args[1]); | |
5116 | break; | |
b8b94ac6 RH |
5117 | case INDEX_op_extrl_i64_i32: |
5118 | tcg_out_extrl_i64_i32(s, new_args[0], new_args[1]); | |
5119 | break; | |
678155b2 RH |
5120 | default: |
5121 | if (def->flags & TCG_OPF_VECTOR) { | |
5122 | tcg_out_vec_op(s, op->opc, TCGOP_VECL(op), TCGOP_VECE(op), | |
5123 | new_args, const_args); | |
5124 | } else { | |
5125 | tcg_out_op(s, op->opc, new_args, const_args); | |
5126 | } | |
5127 | break; | |
d2fd745f RH |
5128 | } |
5129 | ||
c896fe29 FB |
5130 | /* move the outputs in the correct register if needed */ |
5131 | for(i = 0; i < nb_oargs; i++) { | |
43439139 | 5132 | ts = arg_temp(op->args[i]); |
d63e3b6e RH |
5133 | |
5134 | /* ENV should not be modified. */ | |
e01fa97d | 5135 | tcg_debug_assert(!temp_readonly(ts)); |
d63e3b6e | 5136 | |
ec7a869d | 5137 | if (NEED_SYNC_ARG(i)) { |
98b4e186 | 5138 | temp_sync(s, ts, o_allocated_regs, 0, IS_DEAD_ARG(i)); |
59d7c14e | 5139 | } else if (IS_DEAD_ARG(i)) { |
f8bf00f1 | 5140 | temp_dead(s, ts); |
ec7a869d | 5141 | } |
c896fe29 FB |
5142 | } |
5143 | } | |
5144 | ||
efe86b21 RH |
5145 | static bool tcg_reg_alloc_dup2(TCGContext *s, const TCGOp *op) |
5146 | { | |
5147 | const TCGLifeData arg_life = op->life; | |
5148 | TCGTemp *ots, *itsl, *itsh; | |
5149 | TCGType vtype = TCGOP_VECL(op) + TCG_TYPE_V64; | |
5150 | ||
5151 | /* This opcode is only valid for 32-bit hosts, for 64-bit elements. */ | |
5152 | tcg_debug_assert(TCG_TARGET_REG_BITS == 32); | |
5153 | tcg_debug_assert(TCGOP_VECE(op) == MO_64); | |
5154 | ||
5155 | ots = arg_temp(op->args[0]); | |
5156 | itsl = arg_temp(op->args[1]); | |
5157 | itsh = arg_temp(op->args[2]); | |
5158 | ||
5159 | /* ENV should not be modified. */ | |
5160 | tcg_debug_assert(!temp_readonly(ots)); | |
5161 | ||
5162 | /* Allocate the output register now. */ | |
5163 | if (ots->val_type != TEMP_VAL_REG) { | |
5164 | TCGRegSet allocated_regs = s->reserved_regs; | |
5165 | TCGRegSet dup_out_regs = | |
5166 | tcg_op_defs[INDEX_op_dup_vec].args_ct[0].regs; | |
098859f1 | 5167 | TCGReg oreg; |
efe86b21 RH |
5168 | |
5169 | /* Make sure to not spill the input registers. */ | |
5170 | if (!IS_DEAD_ARG(1) && itsl->val_type == TEMP_VAL_REG) { | |
5171 | tcg_regset_set_reg(allocated_regs, itsl->reg); | |
5172 | } | |
5173 | if (!IS_DEAD_ARG(2) && itsh->val_type == TEMP_VAL_REG) { | |
5174 | tcg_regset_set_reg(allocated_regs, itsh->reg); | |
5175 | } | |
5176 | ||
098859f1 | 5177 | oreg = tcg_reg_alloc(s, dup_out_regs, allocated_regs, |
31fd884b | 5178 | output_pref(op, 0), ots->indirect_base); |
098859f1 | 5179 | set_temp_val_reg(s, ots, oreg); |
efe86b21 RH |
5180 | } |
5181 | ||
5182 | /* Promote dup2 of immediates to dupi_vec. */ | |
5183 | if (itsl->val_type == TEMP_VAL_CONST && itsh->val_type == TEMP_VAL_CONST) { | |
5184 | uint64_t val = deposit64(itsl->val, 32, 32, itsh->val); | |
5185 | MemOp vece = MO_64; | |
5186 | ||
5187 | if (val == dup_const(MO_8, val)) { | |
5188 | vece = MO_8; | |
5189 | } else if (val == dup_const(MO_16, val)) { | |
5190 | vece = MO_16; | |
5191 | } else if (val == dup_const(MO_32, val)) { | |
5192 | vece = MO_32; | |
5193 | } | |
5194 | ||
5195 | tcg_out_dupi_vec(s, vtype, vece, ots->reg, val); | |
5196 | goto done; | |
5197 | } | |
5198 | ||
5199 | /* If the two inputs form one 64-bit value, try dupm_vec. */ | |
aef85402 RH |
5200 | if (itsl->temp_subindex == HOST_BIG_ENDIAN && |
5201 | itsh->temp_subindex == !HOST_BIG_ENDIAN && | |
5202 | itsl == itsh + (HOST_BIG_ENDIAN ? 1 : -1)) { | |
5203 | TCGTemp *its = itsl - HOST_BIG_ENDIAN; | |
5204 | ||
5205 | temp_sync(s, its + 0, s->reserved_regs, 0, 0); | |
5206 | temp_sync(s, its + 1, s->reserved_regs, 0, 0); | |
5207 | ||
efe86b21 RH |
5208 | if (tcg_out_dupm_vec(s, vtype, MO_64, ots->reg, |
5209 | its->mem_base->reg, its->mem_offset)) { | |
5210 | goto done; | |
5211 | } | |
5212 | } | |
5213 | ||
5214 | /* Fall back to generic expansion. */ | |
5215 | return false; | |
5216 | ||
5217 | done: | |
36f5539c | 5218 | ots->mem_coherent = 0; |
efe86b21 RH |
5219 | if (IS_DEAD_ARG(1)) { |
5220 | temp_dead(s, itsl); | |
5221 | } | |
5222 | if (IS_DEAD_ARG(2)) { | |
5223 | temp_dead(s, itsh); | |
5224 | } | |
5225 | if (NEED_SYNC_ARG(0)) { | |
5226 | temp_sync(s, ots, s->reserved_regs, 0, IS_DEAD_ARG(0)); | |
5227 | } else if (IS_DEAD_ARG(0)) { | |
5228 | temp_dead(s, ots); | |
5229 | } | |
5230 | return true; | |
5231 | } | |
5232 | ||
39004a71 RH |
5233 | static void load_arg_reg(TCGContext *s, TCGReg reg, TCGTemp *ts, |
5234 | TCGRegSet allocated_regs) | |
c896fe29 | 5235 | { |
39004a71 RH |
5236 | if (ts->val_type == TEMP_VAL_REG) { |
5237 | if (ts->reg != reg) { | |
5238 | tcg_reg_free(s, reg, allocated_regs); | |
5239 | if (!tcg_out_mov(s, ts->type, reg, ts->reg)) { | |
5240 | /* | |
5241 | * Cross register class move not supported. Sync the | |
5242 | * temp back to its slot and load from there. | |
5243 | */ | |
5244 | temp_sync(s, ts, allocated_regs, 0, 0); | |
5245 | tcg_out_ld(s, ts->type, reg, | |
5246 | ts->mem_base->reg, ts->mem_offset); | |
5247 | } | |
5248 | } | |
5249 | } else { | |
5250 | TCGRegSet arg_set = 0; | |
c896fe29 | 5251 | |
39004a71 RH |
5252 | tcg_reg_free(s, reg, allocated_regs); |
5253 | tcg_regset_set_reg(arg_set, reg); | |
5254 | temp_load(s, ts, arg_set, allocated_regs, 0); | |
b03cce8e | 5255 | } |
39004a71 | 5256 | } |
39cf05d3 | 5257 | |
d78e4a4f | 5258 | static void load_arg_stk(TCGContext *s, unsigned arg_slot, TCGTemp *ts, |
39004a71 RH |
5259 | TCGRegSet allocated_regs) |
5260 | { | |
5261 | /* | |
5262 | * When the destination is on the stack, load up the temp and store. | |
5263 | * If there are many call-saved registers, the temp might live to | |
5264 | * see another use; otherwise it'll be discarded. | |
5265 | */ | |
5266 | temp_load(s, ts, tcg_target_available_regs[ts->type], allocated_regs, 0); | |
5267 | tcg_out_st(s, ts->type, ts->reg, TCG_REG_CALL_STACK, | |
d78e4a4f | 5268 | arg_slot_stk_ofs(arg_slot)); |
39004a71 | 5269 | } |
a813e36f | 5270 | |
39004a71 RH |
5271 | static void load_arg_normal(TCGContext *s, const TCGCallArgumentLoc *l, |
5272 | TCGTemp *ts, TCGRegSet *allocated_regs) | |
5273 | { | |
338b61e9 | 5274 | if (arg_slot_reg_p(l->arg_slot)) { |
39004a71 RH |
5275 | TCGReg reg = tcg_target_call_iarg_regs[l->arg_slot]; |
5276 | load_arg_reg(s, reg, ts, *allocated_regs); | |
5277 | tcg_regset_set_reg(*allocated_regs, reg); | |
5278 | } else { | |
d78e4a4f | 5279 | load_arg_stk(s, l->arg_slot, ts, *allocated_regs); |
39004a71 RH |
5280 | } |
5281 | } | |
40ae5c62 | 5282 | |
d78e4a4f | 5283 | static void load_arg_ref(TCGContext *s, unsigned arg_slot, TCGReg ref_base, |
313bdea8 RH |
5284 | intptr_t ref_off, TCGRegSet *allocated_regs) |
5285 | { | |
5286 | TCGReg reg; | |
313bdea8 | 5287 | |
d78e4a4f | 5288 | if (arg_slot_reg_p(arg_slot)) { |
313bdea8 RH |
5289 | reg = tcg_target_call_iarg_regs[arg_slot]; |
5290 | tcg_reg_free(s, reg, *allocated_regs); | |
5291 | tcg_out_addi_ptr(s, reg, ref_base, ref_off); | |
5292 | tcg_regset_set_reg(*allocated_regs, reg); | |
5293 | } else { | |
5294 | reg = tcg_reg_alloc(s, tcg_target_available_regs[TCG_TYPE_PTR], | |
5295 | *allocated_regs, 0, false); | |
5296 | tcg_out_addi_ptr(s, reg, ref_base, ref_off); | |
5297 | tcg_out_st(s, TCG_TYPE_PTR, reg, TCG_REG_CALL_STACK, | |
d78e4a4f | 5298 | arg_slot_stk_ofs(arg_slot)); |
313bdea8 RH |
5299 | } |
5300 | } | |
5301 | ||
39004a71 RH |
5302 | static void tcg_reg_alloc_call(TCGContext *s, TCGOp *op) |
5303 | { | |
5304 | const int nb_oargs = TCGOP_CALLO(op); | |
5305 | const int nb_iargs = TCGOP_CALLI(op); | |
5306 | const TCGLifeData arg_life = op->life; | |
5307 | const TCGHelperInfo *info = tcg_call_info(op); | |
5308 | TCGRegSet allocated_regs = s->reserved_regs; | |
5309 | int i; | |
40ae5c62 | 5310 | |
39004a71 RH |
5311 | /* |
5312 | * Move inputs into place in reverse order, | |
5313 | * so that we place stacked arguments first. | |
5314 | */ | |
5315 | for (i = nb_iargs - 1; i >= 0; --i) { | |
5316 | const TCGCallArgumentLoc *loc = &info->in[i]; | |
5317 | TCGTemp *ts = arg_temp(op->args[nb_oargs + i]); | |
40ae5c62 | 5318 | |
39004a71 RH |
5319 | switch (loc->kind) { |
5320 | case TCG_CALL_ARG_NORMAL: | |
5321 | case TCG_CALL_ARG_EXTEND_U: | |
5322 | case TCG_CALL_ARG_EXTEND_S: | |
5323 | load_arg_normal(s, loc, ts, &allocated_regs); | |
5324 | break; | |
313bdea8 RH |
5325 | case TCG_CALL_ARG_BY_REF: |
5326 | load_arg_stk(s, loc->ref_slot, ts, allocated_regs); | |
5327 | load_arg_ref(s, loc->arg_slot, TCG_REG_CALL_STACK, | |
d78e4a4f | 5328 | arg_slot_stk_ofs(loc->ref_slot), |
313bdea8 RH |
5329 | &allocated_regs); |
5330 | break; | |
5331 | case TCG_CALL_ARG_BY_REF_N: | |
5332 | load_arg_stk(s, loc->ref_slot, ts, allocated_regs); | |
5333 | break; | |
39004a71 RH |
5334 | default: |
5335 | g_assert_not_reached(); | |
c896fe29 | 5336 | } |
c896fe29 | 5337 | } |
a813e36f | 5338 | |
39004a71 | 5339 | /* Mark dead temporaries and free the associated registers. */ |
dd186292 | 5340 | for (i = nb_oargs; i < nb_iargs + nb_oargs; i++) { |
866cb6cb | 5341 | if (IS_DEAD_ARG(i)) { |
43439139 | 5342 | temp_dead(s, arg_temp(op->args[i])); |
c896fe29 FB |
5343 | } |
5344 | } | |
a813e36f | 5345 | |
39004a71 | 5346 | /* Clobber call registers. */ |
c8074023 RH |
5347 | for (i = 0; i < TCG_TARGET_NB_REGS; i++) { |
5348 | if (tcg_regset_test_reg(tcg_target_call_clobber_regs, i)) { | |
b3915dbb | 5349 | tcg_reg_free(s, i, allocated_regs); |
c896fe29 FB |
5350 | } |
5351 | } | |
78505279 | 5352 | |
39004a71 RH |
5353 | /* |
5354 | * Save globals if they might be written by the helper, | |
5355 | * sync them if they might be read. | |
5356 | */ | |
5357 | if (info->flags & TCG_CALL_NO_READ_GLOBALS) { | |
78505279 | 5358 | /* Nothing to do */ |
39004a71 | 5359 | } else if (info->flags & TCG_CALL_NO_WRITE_GLOBALS) { |
78505279 AJ |
5360 | sync_globals(s, allocated_regs); |
5361 | } else { | |
b9c18f56 AJ |
5362 | save_globals(s, allocated_regs); |
5363 | } | |
c896fe29 | 5364 | |
313bdea8 RH |
5365 | /* |
5366 | * If the ABI passes a pointer to the returned struct as the first | |
5367 | * argument, load that now. Pass a pointer to the output home slot. | |
5368 | */ | |
5369 | if (info->out_kind == TCG_CALL_RET_BY_REF) { | |
5370 | TCGTemp *ts = arg_temp(op->args[0]); | |
5371 | ||
5372 | if (!ts->mem_allocated) { | |
5373 | temp_allocate_frame(s, ts); | |
5374 | } | |
5375 | load_arg_ref(s, 0, ts->mem_base->reg, ts->mem_offset, &allocated_regs); | |
5376 | } | |
5377 | ||
cee44b03 | 5378 | tcg_out_call(s, tcg_call_func(op), info); |
c896fe29 | 5379 | |
39004a71 RH |
5380 | /* Assign output registers and emit moves if needed. */ |
5381 | switch (info->out_kind) { | |
5382 | case TCG_CALL_RET_NORMAL: | |
5383 | for (i = 0; i < nb_oargs; i++) { | |
5384 | TCGTemp *ts = arg_temp(op->args[i]); | |
5e3d0c19 | 5385 | TCGReg reg = tcg_target_call_oarg_reg(TCG_CALL_RET_NORMAL, i); |
d63e3b6e | 5386 | |
39004a71 RH |
5387 | /* ENV should not be modified. */ |
5388 | tcg_debug_assert(!temp_readonly(ts)); | |
d63e3b6e | 5389 | |
39004a71 RH |
5390 | set_temp_val_reg(s, ts, reg); |
5391 | ts->mem_coherent = 0; | |
5392 | } | |
5393 | break; | |
313bdea8 | 5394 | |
c6556aa0 RH |
5395 | case TCG_CALL_RET_BY_VEC: |
5396 | { | |
5397 | TCGTemp *ts = arg_temp(op->args[0]); | |
5398 | ||
5399 | tcg_debug_assert(ts->base_type == TCG_TYPE_I128); | |
5400 | tcg_debug_assert(ts->temp_subindex == 0); | |
5401 | if (!ts->mem_allocated) { | |
5402 | temp_allocate_frame(s, ts); | |
5403 | } | |
5404 | tcg_out_st(s, TCG_TYPE_V128, | |
5405 | tcg_target_call_oarg_reg(TCG_CALL_RET_BY_VEC, 0), | |
5406 | ts->mem_base->reg, ts->mem_offset); | |
5407 | } | |
5408 | /* fall through to mark all parts in memory */ | |
5409 | ||
313bdea8 RH |
5410 | case TCG_CALL_RET_BY_REF: |
5411 | /* The callee has performed a write through the reference. */ | |
5412 | for (i = 0; i < nb_oargs; i++) { | |
5413 | TCGTemp *ts = arg_temp(op->args[i]); | |
5414 | ts->val_type = TEMP_VAL_MEM; | |
5415 | } | |
5416 | break; | |
5417 | ||
39004a71 RH |
5418 | default: |
5419 | g_assert_not_reached(); | |
5420 | } | |
5421 | ||
5422 | /* Flush or discard output registers as needed. */ | |
5423 | for (i = 0; i < nb_oargs; i++) { | |
5424 | TCGTemp *ts = arg_temp(op->args[i]); | |
d63e3b6e | 5425 | if (NEED_SYNC_ARG(i)) { |
39004a71 | 5426 | temp_sync(s, ts, s->reserved_regs, 0, IS_DEAD_ARG(i)); |
d63e3b6e RH |
5427 | } else if (IS_DEAD_ARG(i)) { |
5428 | temp_dead(s, ts); | |
c896fe29 FB |
5429 | } |
5430 | } | |
c896fe29 FB |
5431 | } |
5432 | ||
e63b8a29 RH |
5433 | /** |
5434 | * atom_and_align_for_opc: | |
5435 | * @s: tcg context | |
5436 | * @opc: memory operation code | |
5437 | * @host_atom: MO_ATOM_{IFALIGN,WITHIN16,SUBALIGN} for host operations | |
5438 | * @allow_two_ops: true if we are prepared to issue two operations | |
5439 | * | |
5440 | * Return the alignment and atomicity to use for the inline fast path | |
5441 | * for the given memory operation. The alignment may be larger than | |
5442 | * that specified in @opc, and the correct alignment will be diagnosed | |
5443 | * by the slow path helper. | |
5444 | * | |
5445 | * If @allow_two_ops, the host is prepared to test for 2x alignment, | |
5446 | * and issue two loads or stores for subalignment. | |
5447 | */ | |
5448 | static TCGAtomAlign atom_and_align_for_opc(TCGContext *s, MemOp opc, | |
5449 | MemOp host_atom, bool allow_two_ops) | |
5450 | { | |
5451 | MemOp align = get_alignment_bits(opc); | |
5452 | MemOp size = opc & MO_SIZE; | |
5453 | MemOp half = size ? size - 1 : 0; | |
cbb14556 | 5454 | MemOp atom = opc & MO_ATOM_MASK; |
e63b8a29 | 5455 | MemOp atmax; |
e63b8a29 RH |
5456 | |
5457 | switch (atom) { | |
5458 | case MO_ATOM_NONE: | |
5459 | /* The operation requires no specific atomicity. */ | |
5460 | atmax = MO_8; | |
5461 | break; | |
5462 | ||
5463 | case MO_ATOM_IFALIGN: | |
5464 | atmax = size; | |
5465 | break; | |
5466 | ||
5467 | case MO_ATOM_IFALIGN_PAIR: | |
5468 | atmax = half; | |
5469 | break; | |
5470 | ||
5471 | case MO_ATOM_WITHIN16: | |
5472 | atmax = size; | |
5473 | if (size == MO_128) { | |
5474 | /* Misalignment implies !within16, and therefore no atomicity. */ | |
5475 | } else if (host_atom != MO_ATOM_WITHIN16) { | |
5476 | /* The host does not implement within16, so require alignment. */ | |
5477 | align = MAX(align, size); | |
5478 | } | |
5479 | break; | |
5480 | ||
5481 | case MO_ATOM_WITHIN16_PAIR: | |
5482 | atmax = size; | |
5483 | /* | |
5484 | * Misalignment implies !within16, and therefore half atomicity. | |
5485 | * Any host prepared for two operations can implement this with | |
5486 | * half alignment. | |
5487 | */ | |
5488 | if (host_atom != MO_ATOM_WITHIN16 && allow_two_ops) { | |
5489 | align = MAX(align, half); | |
5490 | } | |
5491 | break; | |
5492 | ||
5493 | case MO_ATOM_SUBALIGN: | |
5494 | atmax = size; | |
5495 | if (host_atom != MO_ATOM_SUBALIGN) { | |
5496 | /* If unaligned but not odd, there are subobjects up to half. */ | |
5497 | if (allow_two_ops) { | |
5498 | align = MAX(align, half); | |
5499 | } else { | |
5500 | align = MAX(align, size); | |
5501 | } | |
5502 | } | |
5503 | break; | |
5504 | ||
5505 | default: | |
5506 | g_assert_not_reached(); | |
5507 | } | |
5508 | ||
5509 | return (TCGAtomAlign){ .atom = atmax, .align = align }; | |
5510 | } | |
5511 | ||
8429a1ca RH |
5512 | /* |
5513 | * Similarly for qemu_ld/st slow path helpers. | |
5514 | * We must re-implement tcg_gen_callN and tcg_reg_alloc_call simultaneously, | |
5515 | * using only the provided backend tcg_out_* functions. | |
5516 | */ | |
5517 | ||
5518 | static int tcg_out_helper_stk_ofs(TCGType type, unsigned slot) | |
5519 | { | |
5520 | int ofs = arg_slot_stk_ofs(slot); | |
5521 | ||
5522 | /* | |
5523 | * Each stack slot is TCG_TARGET_LONG_BITS. If the host does not | |
5524 | * require extension to uint64_t, adjust the address for uint32_t. | |
5525 | */ | |
5526 | if (HOST_BIG_ENDIAN && | |
5527 | TCG_TARGET_REG_BITS == 64 && | |
5528 | type == TCG_TYPE_I32) { | |
5529 | ofs += 4; | |
5530 | } | |
5531 | return ofs; | |
5532 | } | |
5533 | ||
8d314041 RH |
5534 | static void tcg_out_helper_load_slots(TCGContext *s, |
5535 | unsigned nmov, TCGMovExtend *mov, | |
5536 | const TCGLdstHelperParam *parm) | |
8429a1ca | 5537 | { |
8d314041 | 5538 | unsigned i; |
2462e30e RH |
5539 | TCGReg dst3; |
5540 | ||
8d314041 RH |
5541 | /* |
5542 | * Start from the end, storing to the stack first. | |
5543 | * This frees those registers, so we need not consider overlap. | |
5544 | */ | |
5545 | for (i = nmov; i-- > 0; ) { | |
5546 | unsigned slot = mov[i].dst; | |
5547 | ||
5548 | if (arg_slot_reg_p(slot)) { | |
5549 | goto found_reg; | |
5550 | } | |
5551 | ||
5552 | TCGReg src = mov[i].src; | |
5553 | TCGType dst_type = mov[i].dst_type; | |
5554 | MemOp dst_mo = dst_type == TCG_TYPE_I32 ? MO_32 : MO_64; | |
5555 | ||
5556 | /* The argument is going onto the stack; extend into scratch. */ | |
5557 | if ((mov[i].src_ext & MO_SIZE) != dst_mo) { | |
5558 | tcg_debug_assert(parm->ntmp != 0); | |
5559 | mov[i].dst = src = parm->tmp[0]; | |
5560 | tcg_out_movext1(s, &mov[i]); | |
5561 | } | |
5562 | ||
5563 | tcg_out_st(s, dst_type, src, TCG_REG_CALL_STACK, | |
5564 | tcg_out_helper_stk_ofs(dst_type, slot)); | |
5565 | } | |
5566 | return; | |
5567 | ||
5568 | found_reg: | |
5569 | /* | |
5570 | * The remaining arguments are in registers. | |
5571 | * Convert slot numbers to argument registers. | |
5572 | */ | |
5573 | nmov = i + 1; | |
5574 | for (i = 0; i < nmov; ++i) { | |
5575 | mov[i].dst = tcg_target_call_iarg_regs[mov[i].dst]; | |
5576 | } | |
5577 | ||
8429a1ca | 5578 | switch (nmov) { |
2462e30e | 5579 | case 4: |
8429a1ca | 5580 | /* The backend must have provided enough temps for the worst case. */ |
2462e30e | 5581 | tcg_debug_assert(parm->ntmp >= 2); |
8429a1ca | 5582 | |
2462e30e RH |
5583 | dst3 = mov[3].dst; |
5584 | for (unsigned j = 0; j < 3; ++j) { | |
5585 | if (dst3 == mov[j].src) { | |
5586 | /* | |
5587 | * Conflict. Copy the source to a temporary, perform the | |
5588 | * remaining moves, then the extension from our scratch | |
5589 | * on the way out. | |
5590 | */ | |
5591 | TCGReg scratch = parm->tmp[1]; | |
8429a1ca | 5592 | |
2462e30e RH |
5593 | tcg_out_mov(s, mov[3].src_type, scratch, mov[3].src); |
5594 | tcg_out_movext3(s, mov, mov + 1, mov + 2, parm->tmp[0]); | |
5595 | tcg_out_movext1_new_src(s, &mov[3], scratch); | |
5596 | break; | |
8429a1ca | 5597 | } |
8429a1ca | 5598 | } |
8429a1ca | 5599 | |
2462e30e RH |
5600 | /* No conflicts: perform this move and continue. */ |
5601 | tcg_out_movext1(s, &mov[3]); | |
5602 | /* fall through */ | |
5603 | ||
5604 | case 3: | |
5605 | tcg_out_movext3(s, mov, mov + 1, mov + 2, | |
5606 | parm->ntmp ? parm->tmp[0] : -1); | |
5607 | break; | |
8429a1ca | 5608 | case 2: |
2462e30e RH |
5609 | tcg_out_movext2(s, mov, mov + 1, |
5610 | parm->ntmp ? parm->tmp[0] : -1); | |
5611 | break; | |
8429a1ca RH |
5612 | case 1: |
5613 | tcg_out_movext1(s, mov); | |
2462e30e RH |
5614 | break; |
5615 | default: | |
8429a1ca RH |
5616 | g_assert_not_reached(); |
5617 | } | |
5618 | } | |
5619 | ||
8429a1ca RH |
5620 | static void tcg_out_helper_load_imm(TCGContext *s, unsigned slot, |
5621 | TCGType type, tcg_target_long imm, | |
5622 | const TCGLdstHelperParam *parm) | |
5623 | { | |
5624 | if (arg_slot_reg_p(slot)) { | |
5625 | tcg_out_movi(s, type, tcg_target_call_iarg_regs[slot], imm); | |
5626 | } else { | |
5627 | int ofs = tcg_out_helper_stk_ofs(type, slot); | |
5628 | if (!tcg_out_sti(s, type, imm, TCG_REG_CALL_STACK, ofs)) { | |
5629 | tcg_debug_assert(parm->ntmp != 0); | |
5630 | tcg_out_movi(s, type, parm->tmp[0], imm); | |
5631 | tcg_out_st(s, type, parm->tmp[0], TCG_REG_CALL_STACK, ofs); | |
5632 | } | |
5633 | } | |
5634 | } | |
5635 | ||
5636 | static void tcg_out_helper_load_common_args(TCGContext *s, | |
5637 | const TCGLabelQemuLdst *ldst, | |
5638 | const TCGLdstHelperParam *parm, | |
5639 | const TCGHelperInfo *info, | |
5640 | unsigned next_arg) | |
5641 | { | |
5642 | TCGMovExtend ptr_mov = { | |
5643 | .dst_type = TCG_TYPE_PTR, | |
5644 | .src_type = TCG_TYPE_PTR, | |
5645 | .src_ext = sizeof(void *) == 4 ? MO_32 : MO_64 | |
5646 | }; | |
5647 | const TCGCallArgumentLoc *loc = &info->in[0]; | |
5648 | TCGType type; | |
5649 | unsigned slot; | |
5650 | tcg_target_ulong imm; | |
5651 | ||
5652 | /* | |
5653 | * Handle env, which is always first. | |
5654 | */ | |
5655 | ptr_mov.dst = loc->arg_slot; | |
5656 | ptr_mov.src = TCG_AREG0; | |
5657 | tcg_out_helper_load_slots(s, 1, &ptr_mov, parm); | |
5658 | ||
5659 | /* | |
5660 | * Handle oi. | |
5661 | */ | |
5662 | imm = ldst->oi; | |
5663 | loc = &info->in[next_arg]; | |
5664 | type = TCG_TYPE_I32; | |
5665 | switch (loc->kind) { | |
5666 | case TCG_CALL_ARG_NORMAL: | |
5667 | break; | |
5668 | case TCG_CALL_ARG_EXTEND_U: | |
5669 | case TCG_CALL_ARG_EXTEND_S: | |
5670 | /* No extension required for MemOpIdx. */ | |
5671 | tcg_debug_assert(imm <= INT32_MAX); | |
5672 | type = TCG_TYPE_REG; | |
5673 | break; | |
5674 | default: | |
5675 | g_assert_not_reached(); | |
5676 | } | |
5677 | tcg_out_helper_load_imm(s, loc->arg_slot, type, imm, parm); | |
5678 | next_arg++; | |
5679 | ||
5680 | /* | |
5681 | * Handle ra. | |
5682 | */ | |
5683 | loc = &info->in[next_arg]; | |
5684 | slot = loc->arg_slot; | |
5685 | if (parm->ra_gen) { | |
5686 | int arg_reg = -1; | |
5687 | TCGReg ra_reg; | |
5688 | ||
5689 | if (arg_slot_reg_p(slot)) { | |
5690 | arg_reg = tcg_target_call_iarg_regs[slot]; | |
5691 | } | |
5692 | ra_reg = parm->ra_gen(s, ldst, arg_reg); | |
5693 | ||
5694 | ptr_mov.dst = slot; | |
5695 | ptr_mov.src = ra_reg; | |
5696 | tcg_out_helper_load_slots(s, 1, &ptr_mov, parm); | |
5697 | } else { | |
5698 | imm = (uintptr_t)ldst->raddr; | |
5699 | tcg_out_helper_load_imm(s, slot, TCG_TYPE_PTR, imm, parm); | |
5700 | } | |
5701 | } | |
5702 | ||
5703 | static unsigned tcg_out_helper_add_mov(TCGMovExtend *mov, | |
5704 | const TCGCallArgumentLoc *loc, | |
5705 | TCGType dst_type, TCGType src_type, | |
5706 | TCGReg lo, TCGReg hi) | |
5707 | { | |
ebebea53 RH |
5708 | MemOp reg_mo; |
5709 | ||
8429a1ca RH |
5710 | if (dst_type <= TCG_TYPE_REG) { |
5711 | MemOp src_ext; | |
5712 | ||
5713 | switch (loc->kind) { | |
5714 | case TCG_CALL_ARG_NORMAL: | |
5715 | src_ext = src_type == TCG_TYPE_I32 ? MO_32 : MO_64; | |
5716 | break; | |
5717 | case TCG_CALL_ARG_EXTEND_U: | |
5718 | dst_type = TCG_TYPE_REG; | |
5719 | src_ext = MO_UL; | |
5720 | break; | |
5721 | case TCG_CALL_ARG_EXTEND_S: | |
5722 | dst_type = TCG_TYPE_REG; | |
5723 | src_ext = MO_SL; | |
5724 | break; | |
5725 | default: | |
5726 | g_assert_not_reached(); | |
5727 | } | |
5728 | ||
5729 | mov[0].dst = loc->arg_slot; | |
5730 | mov[0].dst_type = dst_type; | |
5731 | mov[0].src = lo; | |
5732 | mov[0].src_type = src_type; | |
5733 | mov[0].src_ext = src_ext; | |
5734 | return 1; | |
5735 | } | |
5736 | ||
ebebea53 RH |
5737 | if (TCG_TARGET_REG_BITS == 32) { |
5738 | assert(dst_type == TCG_TYPE_I64); | |
5739 | reg_mo = MO_32; | |
5740 | } else { | |
5741 | assert(dst_type == TCG_TYPE_I128); | |
5742 | reg_mo = MO_64; | |
5743 | } | |
8429a1ca RH |
5744 | |
5745 | mov[0].dst = loc[HOST_BIG_ENDIAN].arg_slot; | |
5746 | mov[0].src = lo; | |
ebebea53 RH |
5747 | mov[0].dst_type = TCG_TYPE_REG; |
5748 | mov[0].src_type = TCG_TYPE_REG; | |
5749 | mov[0].src_ext = reg_mo; | |
8429a1ca RH |
5750 | |
5751 | mov[1].dst = loc[!HOST_BIG_ENDIAN].arg_slot; | |
5752 | mov[1].src = hi; | |
ebebea53 RH |
5753 | mov[1].dst_type = TCG_TYPE_REG; |
5754 | mov[1].src_type = TCG_TYPE_REG; | |
5755 | mov[1].src_ext = reg_mo; | |
8429a1ca RH |
5756 | |
5757 | return 2; | |
5758 | } | |
5759 | ||
5760 | static void tcg_out_ld_helper_args(TCGContext *s, const TCGLabelQemuLdst *ldst, | |
5761 | const TCGLdstHelperParam *parm) | |
5762 | { | |
5763 | const TCGHelperInfo *info; | |
5764 | const TCGCallArgumentLoc *loc; | |
5765 | TCGMovExtend mov[2]; | |
5766 | unsigned next_arg, nmov; | |
5767 | MemOp mop = get_memop(ldst->oi); | |
5768 | ||
5769 | switch (mop & MO_SIZE) { | |
5770 | case MO_8: | |
5771 | case MO_16: | |
5772 | case MO_32: | |
5773 | info = &info_helper_ld32_mmu; | |
5774 | break; | |
5775 | case MO_64: | |
5776 | info = &info_helper_ld64_mmu; | |
5777 | break; | |
ebebea53 RH |
5778 | case MO_128: |
5779 | info = &info_helper_ld128_mmu; | |
5780 | break; | |
8429a1ca RH |
5781 | default: |
5782 | g_assert_not_reached(); | |
5783 | } | |
5784 | ||
5785 | /* Defer env argument. */ | |
5786 | next_arg = 1; | |
5787 | ||
5788 | loc = &info->in[next_arg]; | |
c31e5fa4 | 5789 | if (TCG_TARGET_REG_BITS == 32 && s->addr_type == TCG_TYPE_I32) { |
24e46e6c RH |
5790 | /* |
5791 | * 32-bit host with 32-bit guest: zero-extend the guest address | |
5792 | * to 64-bits for the helper by storing the low part, then | |
5793 | * load a zero for the high part. | |
5794 | */ | |
5795 | tcg_out_helper_add_mov(mov, loc + HOST_BIG_ENDIAN, | |
5796 | TCG_TYPE_I32, TCG_TYPE_I32, | |
5797 | ldst->addrlo_reg, -1); | |
5798 | tcg_out_helper_load_slots(s, 1, mov, parm); | |
8429a1ca | 5799 | |
24e46e6c RH |
5800 | tcg_out_helper_load_imm(s, loc[!HOST_BIG_ENDIAN].arg_slot, |
5801 | TCG_TYPE_I32, 0, parm); | |
5802 | next_arg += 2; | |
c31e5fa4 RH |
5803 | } else { |
5804 | nmov = tcg_out_helper_add_mov(mov, loc, TCG_TYPE_I64, s->addr_type, | |
5805 | ldst->addrlo_reg, ldst->addrhi_reg); | |
5806 | tcg_out_helper_load_slots(s, nmov, mov, parm); | |
5807 | next_arg += nmov; | |
24e46e6c | 5808 | } |
8429a1ca | 5809 | |
ebebea53 RH |
5810 | switch (info->out_kind) { |
5811 | case TCG_CALL_RET_NORMAL: | |
5812 | case TCG_CALL_RET_BY_VEC: | |
5813 | break; | |
5814 | case TCG_CALL_RET_BY_REF: | |
5815 | /* | |
5816 | * The return reference is in the first argument slot. | |
5817 | * We need memory in which to return: re-use the top of stack. | |
5818 | */ | |
5819 | { | |
5820 | int ofs_slot0 = TCG_TARGET_CALL_STACK_OFFSET; | |
5821 | ||
5822 | if (arg_slot_reg_p(0)) { | |
5823 | tcg_out_addi_ptr(s, tcg_target_call_iarg_regs[0], | |
5824 | TCG_REG_CALL_STACK, ofs_slot0); | |
5825 | } else { | |
5826 | tcg_debug_assert(parm->ntmp != 0); | |
5827 | tcg_out_addi_ptr(s, parm->tmp[0], | |
5828 | TCG_REG_CALL_STACK, ofs_slot0); | |
5829 | tcg_out_st(s, TCG_TYPE_PTR, parm->tmp[0], | |
5830 | TCG_REG_CALL_STACK, ofs_slot0); | |
5831 | } | |
5832 | } | |
5833 | break; | |
5834 | default: | |
5835 | g_assert_not_reached(); | |
5836 | } | |
8429a1ca RH |
5837 | |
5838 | tcg_out_helper_load_common_args(s, ldst, parm, info, next_arg); | |
5839 | } | |
5840 | ||
5841 | static void tcg_out_ld_helper_ret(TCGContext *s, const TCGLabelQemuLdst *ldst, | |
5842 | bool load_sign, | |
5843 | const TCGLdstHelperParam *parm) | |
5844 | { | |
ebebea53 | 5845 | MemOp mop = get_memop(ldst->oi); |
8429a1ca | 5846 | TCGMovExtend mov[2]; |
ebebea53 | 5847 | int ofs_slot0; |
8429a1ca | 5848 | |
ebebea53 RH |
5849 | switch (ldst->type) { |
5850 | case TCG_TYPE_I64: | |
5851 | if (TCG_TARGET_REG_BITS == 32) { | |
5852 | break; | |
5853 | } | |
5854 | /* fall through */ | |
8429a1ca | 5855 | |
ebebea53 | 5856 | case TCG_TYPE_I32: |
8429a1ca RH |
5857 | mov[0].dst = ldst->datalo_reg; |
5858 | mov[0].src = tcg_target_call_oarg_reg(TCG_CALL_RET_NORMAL, 0); | |
5859 | mov[0].dst_type = ldst->type; | |
5860 | mov[0].src_type = TCG_TYPE_REG; | |
5861 | ||
5862 | /* | |
5863 | * If load_sign, then we allowed the helper to perform the | |
5864 | * appropriate sign extension to tcg_target_ulong, and all | |
5865 | * we need now is a plain move. | |
5866 | * | |
5867 | * If they do not, then we expect the relevant extension | |
5868 | * instruction to be no more expensive than a move, and | |
5869 | * we thus save the icache etc by only using one of two | |
5870 | * helper functions. | |
5871 | */ | |
5872 | if (load_sign || !(mop & MO_SIGN)) { | |
5873 | if (TCG_TARGET_REG_BITS == 32 || ldst->type == TCG_TYPE_I32) { | |
5874 | mov[0].src_ext = MO_32; | |
5875 | } else { | |
5876 | mov[0].src_ext = MO_64; | |
5877 | } | |
5878 | } else { | |
5879 | mov[0].src_ext = mop & MO_SSIZE; | |
5880 | } | |
5881 | tcg_out_movext1(s, mov); | |
ebebea53 | 5882 | return; |
8429a1ca | 5883 | |
ebebea53 RH |
5884 | case TCG_TYPE_I128: |
5885 | tcg_debug_assert(TCG_TARGET_REG_BITS == 64); | |
5886 | ofs_slot0 = TCG_TARGET_CALL_STACK_OFFSET; | |
5887 | switch (TCG_TARGET_CALL_RET_I128) { | |
5888 | case TCG_CALL_RET_NORMAL: | |
5889 | break; | |
5890 | case TCG_CALL_RET_BY_VEC: | |
5891 | tcg_out_st(s, TCG_TYPE_V128, | |
5892 | tcg_target_call_oarg_reg(TCG_CALL_RET_BY_VEC, 0), | |
5893 | TCG_REG_CALL_STACK, ofs_slot0); | |
5894 | /* fall through */ | |
5895 | case TCG_CALL_RET_BY_REF: | |
5896 | tcg_out_ld(s, TCG_TYPE_I64, ldst->datalo_reg, | |
5897 | TCG_REG_CALL_STACK, ofs_slot0 + 8 * HOST_BIG_ENDIAN); | |
5898 | tcg_out_ld(s, TCG_TYPE_I64, ldst->datahi_reg, | |
5899 | TCG_REG_CALL_STACK, ofs_slot0 + 8 * !HOST_BIG_ENDIAN); | |
5900 | return; | |
5901 | default: | |
5902 | g_assert_not_reached(); | |
5903 | } | |
5904 | break; | |
8429a1ca | 5905 | |
ebebea53 RH |
5906 | default: |
5907 | g_assert_not_reached(); | |
8429a1ca | 5908 | } |
ebebea53 RH |
5909 | |
5910 | mov[0].dst = ldst->datalo_reg; | |
5911 | mov[0].src = | |
5912 | tcg_target_call_oarg_reg(TCG_CALL_RET_NORMAL, HOST_BIG_ENDIAN); | |
723d3a27 RH |
5913 | mov[0].dst_type = TCG_TYPE_REG; |
5914 | mov[0].src_type = TCG_TYPE_REG; | |
ebebea53 RH |
5915 | mov[0].src_ext = TCG_TARGET_REG_BITS == 32 ? MO_32 : MO_64; |
5916 | ||
5917 | mov[1].dst = ldst->datahi_reg; | |
5918 | mov[1].src = | |
5919 | tcg_target_call_oarg_reg(TCG_CALL_RET_NORMAL, !HOST_BIG_ENDIAN); | |
5920 | mov[1].dst_type = TCG_TYPE_REG; | |
5921 | mov[1].src_type = TCG_TYPE_REG; | |
5922 | mov[1].src_ext = TCG_TARGET_REG_BITS == 32 ? MO_32 : MO_64; | |
5923 | ||
5924 | tcg_out_movext2(s, mov, mov + 1, parm->ntmp ? parm->tmp[0] : -1); | |
8429a1ca RH |
5925 | } |
5926 | ||
5927 | static void tcg_out_st_helper_args(TCGContext *s, const TCGLabelQemuLdst *ldst, | |
5928 | const TCGLdstHelperParam *parm) | |
5929 | { | |
5930 | const TCGHelperInfo *info; | |
5931 | const TCGCallArgumentLoc *loc; | |
5932 | TCGMovExtend mov[4]; | |
5933 | TCGType data_type; | |
5934 | unsigned next_arg, nmov, n; | |
5935 | MemOp mop = get_memop(ldst->oi); | |
5936 | ||
5937 | switch (mop & MO_SIZE) { | |
5938 | case MO_8: | |
5939 | case MO_16: | |
5940 | case MO_32: | |
5941 | info = &info_helper_st32_mmu; | |
5942 | data_type = TCG_TYPE_I32; | |
5943 | break; | |
5944 | case MO_64: | |
5945 | info = &info_helper_st64_mmu; | |
5946 | data_type = TCG_TYPE_I64; | |
5947 | break; | |
ebebea53 RH |
5948 | case MO_128: |
5949 | info = &info_helper_st128_mmu; | |
5950 | data_type = TCG_TYPE_I128; | |
5951 | break; | |
8429a1ca RH |
5952 | default: |
5953 | g_assert_not_reached(); | |
5954 | } | |
5955 | ||
5956 | /* Defer env argument. */ | |
5957 | next_arg = 1; | |
5958 | nmov = 0; | |
5959 | ||
5960 | /* Handle addr argument. */ | |
5961 | loc = &info->in[next_arg]; | |
c31e5fa4 | 5962 | if (TCG_TARGET_REG_BITS == 32 && s->addr_type == TCG_TYPE_I32) { |
24e46e6c RH |
5963 | /* |
5964 | * 32-bit host with 32-bit guest: zero-extend the guest address | |
5965 | * to 64-bits for the helper by storing the low part. Later, | |
5966 | * after we have processed the register inputs, we will load a | |
5967 | * zero for the high part. | |
5968 | */ | |
5969 | tcg_out_helper_add_mov(mov, loc + HOST_BIG_ENDIAN, | |
5970 | TCG_TYPE_I32, TCG_TYPE_I32, | |
5971 | ldst->addrlo_reg, -1); | |
5972 | next_arg += 2; | |
5973 | nmov += 1; | |
c31e5fa4 RH |
5974 | } else { |
5975 | n = tcg_out_helper_add_mov(mov, loc, TCG_TYPE_I64, s->addr_type, | |
5976 | ldst->addrlo_reg, ldst->addrhi_reg); | |
5977 | next_arg += n; | |
5978 | nmov += n; | |
24e46e6c | 5979 | } |
8429a1ca RH |
5980 | |
5981 | /* Handle data argument. */ | |
5982 | loc = &info->in[next_arg]; | |
ebebea53 RH |
5983 | switch (loc->kind) { |
5984 | case TCG_CALL_ARG_NORMAL: | |
5985 | case TCG_CALL_ARG_EXTEND_U: | |
5986 | case TCG_CALL_ARG_EXTEND_S: | |
5987 | n = tcg_out_helper_add_mov(mov + nmov, loc, data_type, ldst->type, | |
5988 | ldst->datalo_reg, ldst->datahi_reg); | |
5989 | next_arg += n; | |
5990 | nmov += n; | |
5991 | tcg_out_helper_load_slots(s, nmov, mov, parm); | |
5992 | break; | |
5993 | ||
5994 | case TCG_CALL_ARG_BY_REF: | |
5995 | tcg_debug_assert(TCG_TARGET_REG_BITS == 64); | |
5996 | tcg_debug_assert(data_type == TCG_TYPE_I128); | |
5997 | tcg_out_st(s, TCG_TYPE_I64, | |
5998 | HOST_BIG_ENDIAN ? ldst->datahi_reg : ldst->datalo_reg, | |
5999 | TCG_REG_CALL_STACK, arg_slot_stk_ofs(loc[0].ref_slot)); | |
6000 | tcg_out_st(s, TCG_TYPE_I64, | |
6001 | HOST_BIG_ENDIAN ? ldst->datalo_reg : ldst->datahi_reg, | |
6002 | TCG_REG_CALL_STACK, arg_slot_stk_ofs(loc[1].ref_slot)); | |
6003 | ||
6004 | tcg_out_helper_load_slots(s, nmov, mov, parm); | |
6005 | ||
6006 | if (arg_slot_reg_p(loc->arg_slot)) { | |
6007 | tcg_out_addi_ptr(s, tcg_target_call_iarg_regs[loc->arg_slot], | |
6008 | TCG_REG_CALL_STACK, | |
6009 | arg_slot_stk_ofs(loc->ref_slot)); | |
6010 | } else { | |
6011 | tcg_debug_assert(parm->ntmp != 0); | |
6012 | tcg_out_addi_ptr(s, parm->tmp[0], TCG_REG_CALL_STACK, | |
6013 | arg_slot_stk_ofs(loc->ref_slot)); | |
6014 | tcg_out_st(s, TCG_TYPE_PTR, parm->tmp[0], | |
6015 | TCG_REG_CALL_STACK, arg_slot_stk_ofs(loc->arg_slot)); | |
6016 | } | |
6017 | next_arg += 2; | |
6018 | break; | |
6019 | ||
6020 | default: | |
6021 | g_assert_not_reached(); | |
6022 | } | |
8429a1ca | 6023 | |
c31e5fa4 RH |
6024 | if (TCG_TARGET_REG_BITS == 32 && s->addr_type == TCG_TYPE_I32) { |
6025 | /* Zero extend the address by loading a zero for the high part. */ | |
24e46e6c RH |
6026 | loc = &info->in[1 + !HOST_BIG_ENDIAN]; |
6027 | tcg_out_helper_load_imm(s, loc->arg_slot, TCG_TYPE_I32, 0, parm); | |
6028 | } | |
6029 | ||
8429a1ca RH |
6030 | tcg_out_helper_load_common_args(s, ldst, parm, info, next_arg); |
6031 | } | |
6032 | ||
76cef4b2 | 6033 | int tcg_gen_code(TCGContext *s, TranslationBlock *tb, uint64_t pc_start) |
c896fe29 | 6034 | { |
747bd69d | 6035 | int i, start_words, num_insns; |
15fa08f8 | 6036 | TCGOp *op; |
c896fe29 | 6037 | |
d977e1c2 | 6038 | if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP) |
fbf59aad | 6039 | && qemu_log_in_addr_range(pc_start))) { |
c60f599b | 6040 | FILE *logfile = qemu_log_trylock(); |
78b54858 RH |
6041 | if (logfile) { |
6042 | fprintf(logfile, "OP:\n"); | |
b7a83ff8 | 6043 | tcg_dump_ops(s, logfile, false); |
78b54858 RH |
6044 | fprintf(logfile, "\n"); |
6045 | qemu_log_unlock(logfile); | |
6046 | } | |
c896fe29 | 6047 | } |
c896fe29 | 6048 | |
bef16ab4 RH |
6049 | #ifdef CONFIG_DEBUG_TCG |
6050 | /* Ensure all labels referenced have been emitted. */ | |
6051 | { | |
6052 | TCGLabel *l; | |
6053 | bool error = false; | |
6054 | ||
6055 | QSIMPLEQ_FOREACH(l, &s->labels, next) { | |
f85b1fc4 | 6056 | if (unlikely(!l->present) && !QSIMPLEQ_EMPTY(&l->branches)) { |
bef16ab4 RH |
6057 | qemu_log_mask(CPU_LOG_TB_OP, |
6058 | "$L%d referenced but not present.\n", l->id); | |
6059 | error = true; | |
6060 | } | |
6061 | } | |
6062 | assert(!error); | |
6063 | } | |
6064 | #endif | |
6065 | ||
c45cb8bb | 6066 | tcg_optimize(s); |
8f2e8c07 | 6067 | |
b4fc67c7 | 6068 | reachable_code_pass(s); |
874b8574 | 6069 | liveness_pass_0(s); |
b83eabea | 6070 | liveness_pass_1(s); |
5a18407f | 6071 | |
b83eabea | 6072 | if (s->nb_indirects > 0) { |
b83eabea | 6073 | if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP_IND) |
fbf59aad | 6074 | && qemu_log_in_addr_range(pc_start))) { |
c60f599b | 6075 | FILE *logfile = qemu_log_trylock(); |
78b54858 RH |
6076 | if (logfile) { |
6077 | fprintf(logfile, "OP before indirect lowering:\n"); | |
b7a83ff8 | 6078 | tcg_dump_ops(s, logfile, false); |
78b54858 RH |
6079 | fprintf(logfile, "\n"); |
6080 | qemu_log_unlock(logfile); | |
6081 | } | |
b83eabea | 6082 | } |
645e3a81 | 6083 | |
b83eabea RH |
6084 | /* Replace indirect temps with direct temps. */ |
6085 | if (liveness_pass_2(s)) { | |
6086 | /* If changes were made, re-run liveness. */ | |
6087 | liveness_pass_1(s); | |
5a18407f RH |
6088 | } |
6089 | } | |
c5cc28ff | 6090 | |
d977e1c2 | 6091 | if (unlikely(qemu_loglevel_mask(CPU_LOG_TB_OP_OPT) |
fbf59aad | 6092 | && qemu_log_in_addr_range(pc_start))) { |
c60f599b | 6093 | FILE *logfile = qemu_log_trylock(); |
78b54858 RH |
6094 | if (logfile) { |
6095 | fprintf(logfile, "OP after optimization and liveness analysis:\n"); | |
b7a83ff8 | 6096 | tcg_dump_ops(s, logfile, true); |
78b54858 RH |
6097 | fprintf(logfile, "\n"); |
6098 | qemu_log_unlock(logfile); | |
6099 | } | |
c896fe29 | 6100 | } |
c896fe29 | 6101 | |
35abb009 | 6102 | /* Initialize goto_tb jump offsets. */ |
3a50f424 RH |
6103 | tb->jmp_reset_offset[0] = TB_JMP_OFFSET_INVALID; |
6104 | tb->jmp_reset_offset[1] = TB_JMP_OFFSET_INVALID; | |
9da6079b RH |
6105 | tb->jmp_insn_offset[0] = TB_JMP_OFFSET_INVALID; |
6106 | tb->jmp_insn_offset[1] = TB_JMP_OFFSET_INVALID; | |
35abb009 | 6107 | |
c896fe29 FB |
6108 | tcg_reg_alloc_start(s); |
6109 | ||
db0c51a3 RH |
6110 | /* |
6111 | * Reset the buffer pointers when restarting after overflow. | |
6112 | * TODO: Move this into translate-all.c with the rest of the | |
6113 | * buffer management. Having only this done here is confusing. | |
6114 | */ | |
6115 | s->code_buf = tcg_splitwx_to_rw(tb->tc.ptr); | |
6116 | s->code_ptr = s->code_buf; | |
c896fe29 | 6117 | |
659ef5cb | 6118 | #ifdef TCG_TARGET_NEED_LDST_LABELS |
6001f772 | 6119 | QSIMPLEQ_INIT(&s->ldst_labels); |
659ef5cb | 6120 | #endif |
57a26946 RH |
6121 | #ifdef TCG_TARGET_NEED_POOL_LABELS |
6122 | s->pool_labels = NULL; | |
6123 | #endif | |
9ecefc84 | 6124 | |
747bd69d RH |
6125 | start_words = s->insn_start_words; |
6126 | s->gen_insn_data = | |
6127 | tcg_malloc(sizeof(uint64_t) * s->gen_tb->icount * start_words); | |
6128 | ||
9358fbbf RH |
6129 | tcg_out_tb_start(s); |
6130 | ||
fca8a500 | 6131 | num_insns = -1; |
15fa08f8 | 6132 | QTAILQ_FOREACH(op, &s->ops, link) { |
c45cb8bb | 6133 | TCGOpcode opc = op->opc; |
b3db8758 | 6134 | |
c45cb8bb | 6135 | switch (opc) { |
c896fe29 | 6136 | case INDEX_op_mov_i32: |
c896fe29 | 6137 | case INDEX_op_mov_i64: |
d2fd745f | 6138 | case INDEX_op_mov_vec: |
dd186292 | 6139 | tcg_reg_alloc_mov(s, op); |
c896fe29 | 6140 | break; |
bab1671f RH |
6141 | case INDEX_op_dup_vec: |
6142 | tcg_reg_alloc_dup(s, op); | |
6143 | break; | |
765b842a | 6144 | case INDEX_op_insn_start: |
fca8a500 | 6145 | if (num_insns >= 0) { |
9f754620 RH |
6146 | size_t off = tcg_current_code_size(s); |
6147 | s->gen_insn_end_off[num_insns] = off; | |
6148 | /* Assert that we do not overflow our stored offset. */ | |
6149 | assert(s->gen_insn_end_off[num_insns] == off); | |
fca8a500 RH |
6150 | } |
6151 | num_insns++; | |
747bd69d RH |
6152 | for (i = 0; i < start_words; ++i) { |
6153 | s->gen_insn_data[num_insns * start_words + i] = | |
c9ad8d27 | 6154 | tcg_get_insn_start_param(op, i); |
bad729e2 | 6155 | } |
c896fe29 | 6156 | break; |
5ff9d6a4 | 6157 | case INDEX_op_discard: |
43439139 | 6158 | temp_dead(s, arg_temp(op->args[0])); |
5ff9d6a4 | 6159 | break; |
c896fe29 | 6160 | case INDEX_op_set_label: |
e8996ee0 | 6161 | tcg_reg_alloc_bb_end(s, s->reserved_regs); |
92ab8e7d | 6162 | tcg_out_label(s, arg_label(op->args[0])); |
c896fe29 FB |
6163 | break; |
6164 | case INDEX_op_call: | |
dd186292 | 6165 | tcg_reg_alloc_call(s, op); |
c45cb8bb | 6166 | break; |
b55a8d9d RH |
6167 | case INDEX_op_exit_tb: |
6168 | tcg_out_exit_tb(s, op->args[0]); | |
6169 | break; | |
cf7d6b8e RH |
6170 | case INDEX_op_goto_tb: |
6171 | tcg_out_goto_tb(s, op->args[0]); | |
6172 | break; | |
efe86b21 RH |
6173 | case INDEX_op_dup2_vec: |
6174 | if (tcg_reg_alloc_dup2(s, op)) { | |
6175 | break; | |
6176 | } | |
6177 | /* fall through */ | |
c896fe29 | 6178 | default: |
25c4d9cc | 6179 | /* Sanity check that we've not introduced any unhandled opcodes. */ |
be0f34b5 | 6180 | tcg_debug_assert(tcg_op_supported(opc)); |
c896fe29 FB |
6181 | /* Note: in order to speed up the code, it would be much |
6182 | faster to have specialized register allocator functions for | |
6183 | some common argument patterns */ | |
dd186292 | 6184 | tcg_reg_alloc_op(s, op); |
c896fe29 FB |
6185 | break; |
6186 | } | |
b125f9dc RH |
6187 | /* Test for (pending) buffer overflow. The assumption is that any |
6188 | one operation beginning below the high water mark cannot overrun | |
6189 | the buffer completely. Thus we can test for overflow after | |
6190 | generating code without having to check during generation. */ | |
644da9b3 | 6191 | if (unlikely((void *)s->code_ptr > s->code_gen_highwater)) { |
b125f9dc RH |
6192 | return -1; |
6193 | } | |
6e6c4efe RH |
6194 | /* Test for TB overflow, as seen by gen_insn_end_off. */ |
6195 | if (unlikely(tcg_current_code_size(s) > UINT16_MAX)) { | |
6196 | return -2; | |
6197 | } | |
c896fe29 | 6198 | } |
747bd69d | 6199 | tcg_debug_assert(num_insns + 1 == s->gen_tb->icount); |
fca8a500 | 6200 | s->gen_insn_end_off[num_insns] = tcg_current_code_size(s); |
c45cb8bb | 6201 | |
b76f0d8c | 6202 | /* Generate TB finalization at the end of block */ |
659ef5cb | 6203 | #ifdef TCG_TARGET_NEED_LDST_LABELS |
aeee05f5 RH |
6204 | i = tcg_out_ldst_finalize(s); |
6205 | if (i < 0) { | |
6206 | return i; | |
23dceda6 | 6207 | } |
659ef5cb | 6208 | #endif |
57a26946 | 6209 | #ifdef TCG_TARGET_NEED_POOL_LABELS |
1768987b RH |
6210 | i = tcg_out_pool_finalize(s); |
6211 | if (i < 0) { | |
6212 | return i; | |
57a26946 RH |
6213 | } |
6214 | #endif | |
7ecd02a0 RH |
6215 | if (!tcg_resolve_relocs(s)) { |
6216 | return -2; | |
6217 | } | |
c896fe29 | 6218 | |
df5d2b16 | 6219 | #ifndef CONFIG_TCG_INTERPRETER |
c896fe29 | 6220 | /* flush instruction cache */ |
db0c51a3 RH |
6221 | flush_idcache_range((uintptr_t)tcg_splitwx_to_rx(s->code_buf), |
6222 | (uintptr_t)s->code_buf, | |
1da8de39 | 6223 | tcg_ptr_byte_diff(s->code_ptr, s->code_buf)); |
df5d2b16 | 6224 | #endif |
2aeabc08 | 6225 | |
1813e175 | 6226 | return tcg_current_code_size(s); |
c896fe29 FB |
6227 | } |
6228 | ||
813da627 | 6229 | #ifdef ELF_HOST_MACHINE |
5872bbf2 RH |
6230 | /* In order to use this feature, the backend needs to do three things: |
6231 | ||
6232 | (1) Define ELF_HOST_MACHINE to indicate both what value to | |
6233 | put into the ELF image and to indicate support for the feature. | |
6234 | ||
6235 | (2) Define tcg_register_jit. This should create a buffer containing | |
6236 | the contents of a .debug_frame section that describes the post- | |
6237 | prologue unwind info for the tcg machine. | |
6238 | ||
6239 | (3) Call tcg_register_jit_int, with the constructed .debug_frame. | |
6240 | */ | |
813da627 RH |
6241 | |
6242 | /* Begin GDB interface. THE FOLLOWING MUST MATCH GDB DOCS. */ | |
6243 | typedef enum { | |
6244 | JIT_NOACTION = 0, | |
6245 | JIT_REGISTER_FN, | |
6246 | JIT_UNREGISTER_FN | |
6247 | } jit_actions_t; | |
6248 | ||
6249 | struct jit_code_entry { | |
6250 | struct jit_code_entry *next_entry; | |
6251 | struct jit_code_entry *prev_entry; | |
6252 | const void *symfile_addr; | |
6253 | uint64_t symfile_size; | |
6254 | }; | |
6255 | ||
6256 | struct jit_descriptor { | |
6257 | uint32_t version; | |
6258 | uint32_t action_flag; | |
6259 | struct jit_code_entry *relevant_entry; | |
6260 | struct jit_code_entry *first_entry; | |
6261 | }; | |
6262 | ||
6263 | void __jit_debug_register_code(void) __attribute__((noinline)); | |
6264 | void __jit_debug_register_code(void) | |
6265 | { | |
6266 | asm(""); | |
6267 | } | |
6268 | ||
6269 | /* Must statically initialize the version, because GDB may check | |
6270 | the version before we can set it. */ | |
6271 | struct jit_descriptor __jit_debug_descriptor = { 1, 0, 0, 0 }; | |
6272 | ||
6273 | /* End GDB interface. */ | |
6274 | ||
6275 | static int find_string(const char *strtab, const char *str) | |
6276 | { | |
6277 | const char *p = strtab + 1; | |
6278 | ||
6279 | while (1) { | |
6280 | if (strcmp(p, str) == 0) { | |
6281 | return p - strtab; | |
6282 | } | |
6283 | p += strlen(p) + 1; | |
6284 | } | |
6285 | } | |
6286 | ||
755bf9e5 | 6287 | static void tcg_register_jit_int(const void *buf_ptr, size_t buf_size, |
2c90784a RH |
6288 | const void *debug_frame, |
6289 | size_t debug_frame_size) | |
813da627 | 6290 | { |
5872bbf2 RH |
6291 | struct __attribute__((packed)) DebugInfo { |
6292 | uint32_t len; | |
6293 | uint16_t version; | |
6294 | uint32_t abbrev; | |
6295 | uint8_t ptr_size; | |
6296 | uint8_t cu_die; | |
6297 | uint16_t cu_lang; | |
6298 | uintptr_t cu_low_pc; | |
6299 | uintptr_t cu_high_pc; | |
6300 | uint8_t fn_die; | |
6301 | char fn_name[16]; | |
6302 | uintptr_t fn_low_pc; | |
6303 | uintptr_t fn_high_pc; | |
6304 | uint8_t cu_eoc; | |
6305 | }; | |
813da627 RH |
6306 | |
6307 | struct ElfImage { | |
6308 | ElfW(Ehdr) ehdr; | |
6309 | ElfW(Phdr) phdr; | |
5872bbf2 RH |
6310 | ElfW(Shdr) shdr[7]; |
6311 | ElfW(Sym) sym[2]; | |
6312 | struct DebugInfo di; | |
6313 | uint8_t da[24]; | |
6314 | char str[80]; | |
6315 | }; | |
6316 | ||
6317 | struct ElfImage *img; | |
6318 | ||
6319 | static const struct ElfImage img_template = { | |
6320 | .ehdr = { | |
6321 | .e_ident[EI_MAG0] = ELFMAG0, | |
6322 | .e_ident[EI_MAG1] = ELFMAG1, | |
6323 | .e_ident[EI_MAG2] = ELFMAG2, | |
6324 | .e_ident[EI_MAG3] = ELFMAG3, | |
6325 | .e_ident[EI_CLASS] = ELF_CLASS, | |
6326 | .e_ident[EI_DATA] = ELF_DATA, | |
6327 | .e_ident[EI_VERSION] = EV_CURRENT, | |
6328 | .e_type = ET_EXEC, | |
6329 | .e_machine = ELF_HOST_MACHINE, | |
6330 | .e_version = EV_CURRENT, | |
6331 | .e_phoff = offsetof(struct ElfImage, phdr), | |
6332 | .e_shoff = offsetof(struct ElfImage, shdr), | |
6333 | .e_ehsize = sizeof(ElfW(Shdr)), | |
6334 | .e_phentsize = sizeof(ElfW(Phdr)), | |
6335 | .e_phnum = 1, | |
6336 | .e_shentsize = sizeof(ElfW(Shdr)), | |
6337 | .e_shnum = ARRAY_SIZE(img->shdr), | |
6338 | .e_shstrndx = ARRAY_SIZE(img->shdr) - 1, | |
abbb3eae RH |
6339 | #ifdef ELF_HOST_FLAGS |
6340 | .e_flags = ELF_HOST_FLAGS, | |
6341 | #endif | |
6342 | #ifdef ELF_OSABI | |
6343 | .e_ident[EI_OSABI] = ELF_OSABI, | |
6344 | #endif | |
5872bbf2 RH |
6345 | }, |
6346 | .phdr = { | |
6347 | .p_type = PT_LOAD, | |
6348 | .p_flags = PF_X, | |
6349 | }, | |
6350 | .shdr = { | |
6351 | [0] = { .sh_type = SHT_NULL }, | |
6352 | /* Trick: The contents of code_gen_buffer are not present in | |
6353 | this fake ELF file; that got allocated elsewhere. Therefore | |
6354 | we mark .text as SHT_NOBITS (similar to .bss) so that readers | |
6355 | will not look for contents. We can record any address. */ | |
6356 | [1] = { /* .text */ | |
6357 | .sh_type = SHT_NOBITS, | |
6358 | .sh_flags = SHF_EXECINSTR | SHF_ALLOC, | |
6359 | }, | |
6360 | [2] = { /* .debug_info */ | |
6361 | .sh_type = SHT_PROGBITS, | |
6362 | .sh_offset = offsetof(struct ElfImage, di), | |
6363 | .sh_size = sizeof(struct DebugInfo), | |
6364 | }, | |
6365 | [3] = { /* .debug_abbrev */ | |
6366 | .sh_type = SHT_PROGBITS, | |
6367 | .sh_offset = offsetof(struct ElfImage, da), | |
6368 | .sh_size = sizeof(img->da), | |
6369 | }, | |
6370 | [4] = { /* .debug_frame */ | |
6371 | .sh_type = SHT_PROGBITS, | |
6372 | .sh_offset = sizeof(struct ElfImage), | |
6373 | }, | |
6374 | [5] = { /* .symtab */ | |
6375 | .sh_type = SHT_SYMTAB, | |
6376 | .sh_offset = offsetof(struct ElfImage, sym), | |
6377 | .sh_size = sizeof(img->sym), | |
6378 | .sh_info = 1, | |
6379 | .sh_link = ARRAY_SIZE(img->shdr) - 1, | |
6380 | .sh_entsize = sizeof(ElfW(Sym)), | |
6381 | }, | |
6382 | [6] = { /* .strtab */ | |
6383 | .sh_type = SHT_STRTAB, | |
6384 | .sh_offset = offsetof(struct ElfImage, str), | |
6385 | .sh_size = sizeof(img->str), | |
6386 | } | |
6387 | }, | |
6388 | .sym = { | |
6389 | [1] = { /* code_gen_buffer */ | |
6390 | .st_info = ELF_ST_INFO(STB_GLOBAL, STT_FUNC), | |
6391 | .st_shndx = 1, | |
6392 | } | |
6393 | }, | |
6394 | .di = { | |
6395 | .len = sizeof(struct DebugInfo) - 4, | |
6396 | .version = 2, | |
6397 | .ptr_size = sizeof(void *), | |
6398 | .cu_die = 1, | |
6399 | .cu_lang = 0x8001, /* DW_LANG_Mips_Assembler */ | |
6400 | .fn_die = 2, | |
6401 | .fn_name = "code_gen_buffer" | |
6402 | }, | |
6403 | .da = { | |
6404 | 1, /* abbrev number (the cu) */ | |
6405 | 0x11, 1, /* DW_TAG_compile_unit, has children */ | |
6406 | 0x13, 0x5, /* DW_AT_language, DW_FORM_data2 */ | |
6407 | 0x11, 0x1, /* DW_AT_low_pc, DW_FORM_addr */ | |
6408 | 0x12, 0x1, /* DW_AT_high_pc, DW_FORM_addr */ | |
6409 | 0, 0, /* end of abbrev */ | |
6410 | 2, /* abbrev number (the fn) */ | |
6411 | 0x2e, 0, /* DW_TAG_subprogram, no children */ | |
6412 | 0x3, 0x8, /* DW_AT_name, DW_FORM_string */ | |
6413 | 0x11, 0x1, /* DW_AT_low_pc, DW_FORM_addr */ | |
6414 | 0x12, 0x1, /* DW_AT_high_pc, DW_FORM_addr */ | |
6415 | 0, 0, /* end of abbrev */ | |
6416 | 0 /* no more abbrev */ | |
6417 | }, | |
6418 | .str = "\0" ".text\0" ".debug_info\0" ".debug_abbrev\0" | |
6419 | ".debug_frame\0" ".symtab\0" ".strtab\0" "code_gen_buffer", | |
813da627 RH |
6420 | }; |
6421 | ||
6422 | /* We only need a single jit entry; statically allocate it. */ | |
6423 | static struct jit_code_entry one_entry; | |
6424 | ||
5872bbf2 | 6425 | uintptr_t buf = (uintptr_t)buf_ptr; |
813da627 | 6426 | size_t img_size = sizeof(struct ElfImage) + debug_frame_size; |
2c90784a | 6427 | DebugFrameHeader *dfh; |
813da627 | 6428 | |
5872bbf2 RH |
6429 | img = g_malloc(img_size); |
6430 | *img = img_template; | |
813da627 | 6431 | |
5872bbf2 RH |
6432 | img->phdr.p_vaddr = buf; |
6433 | img->phdr.p_paddr = buf; | |
6434 | img->phdr.p_memsz = buf_size; | |
813da627 | 6435 | |
813da627 | 6436 | img->shdr[1].sh_name = find_string(img->str, ".text"); |
5872bbf2 | 6437 | img->shdr[1].sh_addr = buf; |
813da627 RH |
6438 | img->shdr[1].sh_size = buf_size; |
6439 | ||
5872bbf2 RH |
6440 | img->shdr[2].sh_name = find_string(img->str, ".debug_info"); |
6441 | img->shdr[3].sh_name = find_string(img->str, ".debug_abbrev"); | |
6442 | ||
6443 | img->shdr[4].sh_name = find_string(img->str, ".debug_frame"); | |
6444 | img->shdr[4].sh_size = debug_frame_size; | |
6445 | ||
6446 | img->shdr[5].sh_name = find_string(img->str, ".symtab"); | |
6447 | img->shdr[6].sh_name = find_string(img->str, ".strtab"); | |
6448 | ||
6449 | img->sym[1].st_name = find_string(img->str, "code_gen_buffer"); | |
6450 | img->sym[1].st_value = buf; | |
6451 | img->sym[1].st_size = buf_size; | |
813da627 | 6452 | |
5872bbf2 | 6453 | img->di.cu_low_pc = buf; |
45aba097 | 6454 | img->di.cu_high_pc = buf + buf_size; |
5872bbf2 | 6455 | img->di.fn_low_pc = buf; |
45aba097 | 6456 | img->di.fn_high_pc = buf + buf_size; |
813da627 | 6457 | |
2c90784a RH |
6458 | dfh = (DebugFrameHeader *)(img + 1); |
6459 | memcpy(dfh, debug_frame, debug_frame_size); | |
6460 | dfh->fde.func_start = buf; | |
6461 | dfh->fde.func_len = buf_size; | |
6462 | ||
813da627 RH |
6463 | #ifdef DEBUG_JIT |
6464 | /* Enable this block to be able to debug the ELF image file creation. | |
6465 | One can use readelf, objdump, or other inspection utilities. */ | |
6466 | { | |
eb6b2edf BM |
6467 | g_autofree char *jit = g_strdup_printf("%s/qemu.jit", g_get_tmp_dir()); |
6468 | FILE *f = fopen(jit, "w+b"); | |
813da627 | 6469 | if (f) { |
5872bbf2 | 6470 | if (fwrite(img, img_size, 1, f) != img_size) { |
813da627 RH |
6471 | /* Avoid stupid unused return value warning for fwrite. */ |
6472 | } | |
6473 | fclose(f); | |
6474 | } | |
6475 | } | |
6476 | #endif | |
6477 | ||
6478 | one_entry.symfile_addr = img; | |
6479 | one_entry.symfile_size = img_size; | |
6480 | ||
6481 | __jit_debug_descriptor.action_flag = JIT_REGISTER_FN; | |
6482 | __jit_debug_descriptor.relevant_entry = &one_entry; | |
6483 | __jit_debug_descriptor.first_entry = &one_entry; | |
6484 | __jit_debug_register_code(); | |
6485 | } | |
6486 | #else | |
5872bbf2 RH |
6487 | /* No support for the feature. Provide the entry point expected by exec.c, |
6488 | and implement the internal function we declared earlier. */ | |
813da627 | 6489 | |
755bf9e5 | 6490 | static void tcg_register_jit_int(const void *buf, size_t size, |
2c90784a RH |
6491 | const void *debug_frame, |
6492 | size_t debug_frame_size) | |
813da627 RH |
6493 | { |
6494 | } | |
6495 | ||
755bf9e5 | 6496 | void tcg_register_jit(const void *buf, size_t buf_size) |
813da627 RH |
6497 | { |
6498 | } | |
6499 | #endif /* ELF_HOST_MACHINE */ | |
db432672 RH |
6500 | |
6501 | #if !TCG_TARGET_MAYBE_vec | |
6502 | void tcg_expand_vec_op(TCGOpcode o, TCGType t, unsigned e, TCGArg a0, ...) | |
6503 | { | |
6504 | g_assert_not_reached(); | |
6505 | } | |
6506 | #endif |