Merge pull request #836 from pcercuei/update-lightrec-20240611
[pcsx_rearmed.git] / deps / lightrec / lightrec-private.h
... / ...
CommitLineData
1/* SPDX-License-Identifier: LGPL-2.1-or-later */
2/*
3 * Copyright (C) 2016-2021 Paul Cercueil <paul@crapouillou.net>
4 */
5
6#ifndef __LIGHTREC_PRIVATE_H__
7#define __LIGHTREC_PRIVATE_H__
8
9#include "lightning-wrapper.h"
10#include "lightrec-config.h"
11#include "disassembler.h"
12#include "lightrec.h"
13#include "regcache.h"
14
15#if ENABLE_THREADED_COMPILER
16#include <stdatomic.h>
17#endif
18
19#ifdef _MSC_BUILD
20#include <immintrin.h>
21#endif
22
23#include <inttypes.h>
24#include <stdint.h>
25
26#define X32_FMT "0x%08"PRIx32
27#define PC_FMT "PC "X32_FMT
28
29#define ARRAY_SIZE(x) (sizeof(x) ? sizeof(x) / sizeof((x)[0]) : 0)
30
31#define GENMASK(h, l) \
32 (((uintptr_t)-1 << (l)) & ((uintptr_t)-1 >> (__WORDSIZE - 1 - (h))))
33
34#ifdef __GNUC__
35# define likely(x) __builtin_expect(!!(x),1)
36# define unlikely(x) __builtin_expect(!!(x),0)
37#else
38# define likely(x) (x)
39# define unlikely(x) (x)
40#endif
41
42#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
43# define LE32TOH(x) __builtin_bswap32(x)
44# define HTOLE32(x) __builtin_bswap32(x)
45# define LE16TOH(x) __builtin_bswap16(x)
46# define HTOLE16(x) __builtin_bswap16(x)
47#else
48# define LE32TOH(x) (x)
49# define HTOLE32(x) (x)
50# define LE16TOH(x) (x)
51# define HTOLE16(x) (x)
52#endif
53
54#if HAS_DEFAULT_ELM
55#define SET_DEFAULT_ELM(table, value) [0 ... ARRAY_SIZE(table) - 1] = value
56#else
57#define SET_DEFAULT_ELM(table, value) [0] = NULL
58#endif
59
60#if __has_attribute(__fallthrough__)
61# define fallthrough __attribute__((__fallthrough__))
62#else
63# define fallthrough do {} while (0) /* fallthrough */
64#endif
65
66#define container_of(ptr, type, member) \
67 ((type *)((void *)(ptr) - offsetof(type, member)))
68
69#ifdef _MSC_BUILD
70# define popcount32(x) __popcnt(x)
71# define clz32(x) _lzcnt_u32(x)
72# define ctz32(x) _tzcnt_u32(x)
73#else
74# define popcount32(x) __builtin_popcount(x)
75# define clz32(x) __builtin_clz(x)
76# define ctz32(x) __builtin_ctz(x)
77#endif
78
79/* Flags for (struct block *)->flags */
80#define BLOCK_NEVER_COMPILE BIT(0)
81#define BLOCK_SHOULD_RECOMPILE BIT(1)
82#define BLOCK_FULLY_TAGGED BIT(2)
83#define BLOCK_IS_DEAD BIT(3)
84#define BLOCK_IS_MEMSET BIT(4)
85#define BLOCK_NO_OPCODE_LIST BIT(5)
86#define BLOCK_PRELOAD_PC BIT(6)
87
88#define RAM_SIZE 0x200000
89#define BIOS_SIZE 0x80000
90
91#define CODE_LUT_SIZE ((RAM_SIZE + BIOS_SIZE) >> 2)
92
93#define REG_LO 32
94#define REG_HI 33
95#define REG_TEMP (offsetof(struct lightrec_state, temp_reg) / sizeof(u32))
96
97/* Definition of jit_state_t (avoids inclusion of <lightning.h>) */
98struct jit_node;
99struct jit_state;
100typedef struct jit_state jit_state_t;
101
102struct blockcache;
103struct recompiler;
104struct regcache;
105struct opcode;
106struct reaper;
107
108struct u16x2 {
109#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
110 u16 h, l;
111#else
112 u16 l, h;
113#endif
114};
115
116struct block {
117 jit_state_t *_jit;
118 struct opcode *opcode_list;
119 void (*function)(void);
120 const u32 *code;
121 struct block *next;
122 u32 pc;
123 u32 hash;
124 u32 precompile_date;
125 unsigned int code_size;
126 u16 nb_ops;
127#if ENABLE_THREADED_COMPILER
128 _Atomic u8 flags;
129#else
130 u8 flags;
131#endif
132};
133
134struct lightrec_branch {
135 struct jit_node *branch;
136 u32 target;
137};
138
139struct lightrec_branch_target {
140 struct jit_node *label;
141 u32 offset;
142};
143
144enum c_wrappers {
145 C_WRAPPER_RW,
146 C_WRAPPER_RW_GENERIC,
147 C_WRAPPER_MFC,
148 C_WRAPPER_MTC,
149 C_WRAPPER_CP,
150 C_WRAPPERS_COUNT,
151};
152
153struct lightrec_cstate {
154 struct lightrec_state *state;
155
156 struct lightrec_branch local_branches[512];
157 struct lightrec_branch_target targets[512];
158 u16 movi_temp[32];
159 unsigned int nb_local_branches;
160 unsigned int nb_targets;
161 unsigned int cycles;
162
163 struct regcache *reg_cache;
164
165 _Bool no_load_delay;
166};
167
168struct lightrec_state {
169 struct lightrec_registers regs;
170 u32 temp_reg;
171 u32 curr_pc;
172 u32 next_pc;
173 uintptr_t wrapper_regs[NUM_TEMPS];
174 u8 in_delay_slot_n;
175 u32 current_cycle;
176 u32 target_cycle;
177 u32 exit_flags;
178 u32 old_cycle_counter;
179 u32 cycles_per_op;
180 void *c_wrapper;
181 struct block *dispatcher, *c_wrapper_block;
182 void *c_wrappers[C_WRAPPERS_COUNT];
183 struct blockcache *block_cache;
184 struct recompiler *rec;
185 struct lightrec_cstate *cstate;
186 struct reaper *reaper;
187 void *tlsf;
188 void (*eob_wrapper_func)(void);
189 void (*interpreter_func)(void);
190 void (*ds_check_func)(void);
191 void (*memset_func)(void);
192 void (*get_next_block)(void);
193 struct lightrec_ops ops;
194 unsigned int nb_precompile;
195 unsigned int nb_compile;
196 unsigned int nb_maps;
197 const struct lightrec_mem_map *maps;
198 uintptr_t offset_ram, offset_bios, offset_scratch, offset_io;
199 u32 opt_flags;
200 _Bool with_32bit_lut;
201 _Bool mirrors_mapped;
202 void *code_lut[];
203};
204
205#define lightrec_offset(ptr) \
206 offsetof(struct lightrec_state, ptr)
207
208u32 lightrec_rw(struct lightrec_state *state, union code op, u32 addr,
209 u32 data, u32 *flags, struct block *block, u16 offset);
210
211void lightrec_free_block(struct lightrec_state *state, struct block *block);
212
213void remove_from_code_lut(struct blockcache *cache, struct block *block);
214
215const struct lightrec_mem_map *
216lightrec_get_map(struct lightrec_state *state, void **host, u32 kaddr);
217
218static inline u32 kunseg(u32 addr)
219{
220 if (unlikely(addr >= 0xa0000000))
221 return addr - 0xa0000000;
222 else
223 return addr &~ 0x80000000;
224}
225
226static inline u32 lut_offset(u32 pc)
227{
228 if (pc & BIT(28))
229 return ((pc & (BIOS_SIZE - 1)) + RAM_SIZE) >> 2; // BIOS
230 else
231 return (pc & (RAM_SIZE - 1)) >> 2; // RAM
232}
233
234static inline _Bool is_big_endian(void)
235{
236 return __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__;
237}
238
239static inline _Bool lut_is_32bit(const struct lightrec_state *state)
240{
241 return __WORDSIZE == 32 ||
242 (ENABLE_CODE_BUFFER && state->with_32bit_lut);
243}
244
245static inline size_t lut_elm_size(const struct lightrec_state *state)
246{
247 return lut_is_32bit(state) ? 4 : sizeof(void *);
248}
249
250static inline void ** lut_address(struct lightrec_state *state, u32 offset)
251{
252 if (lut_is_32bit(state))
253 return (void **) ((uintptr_t) state->code_lut + offset * 4);
254 else
255 return &state->code_lut[offset];
256}
257
258static inline void * lut_read(struct lightrec_state *state, u32 offset)
259{
260 void **lut_entry = lut_address(state, offset);
261
262 if (lut_is_32bit(state))
263 return (void *)(uintptr_t) *(u32 *) lut_entry;
264 else
265 return *lut_entry;
266}
267
268static inline void lut_write(struct lightrec_state *state, u32 offset, void *ptr)
269{
270 void **lut_entry = lut_address(state, offset);
271
272 if (lut_is_32bit(state))
273 *(u32 *) lut_entry = (u32)(uintptr_t) ptr;
274 else
275 *lut_entry = ptr;
276}
277
278static inline u32 get_ds_pc(const struct block *block, u16 offset, s16 imm)
279{
280 u16 flags = block->opcode_list[offset].flags;
281
282 offset += op_flag_no_ds(flags);
283
284 return block->pc + ((offset + imm) << 2);
285}
286
287static inline u32 get_branch_pc(const struct block *block, u16 offset, s16 imm)
288{
289 u16 flags = block->opcode_list[offset].flags;
290
291 offset -= op_flag_no_ds(flags);
292
293 return block->pc + ((offset + imm) << 2);
294}
295
296void lightrec_mtc(struct lightrec_state *state, union code op, u8 reg, u32 data);
297u32 lightrec_mfc(struct lightrec_state *state, union code op);
298void lightrec_rfe(struct lightrec_state *state);
299void lightrec_cp(struct lightrec_state *state, union code op);
300
301struct lightrec_cstate * lightrec_create_cstate(struct lightrec_state *state);
302void lightrec_free_cstate(struct lightrec_cstate *cstate);
303
304union code lightrec_read_opcode(struct lightrec_state *state, u32 pc);
305
306int lightrec_compile_block(struct lightrec_cstate *cstate, struct block *block);
307void lightrec_free_opcode_list(struct lightrec_state *state,
308 struct opcode *list);
309
310unsigned int lightrec_cycles_of_opcode(const struct lightrec_state *state,
311 union code code);
312
313static inline u8 get_mult_div_lo(union code c)
314{
315 return (OPT_FLAG_MULT_DIV && c.r.rd) ? c.r.rd : REG_LO;
316}
317
318static inline u8 get_mult_div_hi(union code c)
319{
320 return (OPT_FLAG_MULT_DIV && c.r.imm) ? c.r.imm : REG_HI;
321}
322
323static inline s16 s16_max(s16 a, s16 b)
324{
325 return a > b ? a : b;
326}
327
328static inline _Bool block_has_flag(struct block *block, u8 flag)
329{
330#if ENABLE_THREADED_COMPILER
331 return atomic_load_explicit(&block->flags, memory_order_relaxed) & flag;
332#else
333 return block->flags & flag;
334#endif
335}
336
337static inline u8 block_set_flags(struct block *block, u8 mask)
338{
339#if ENABLE_THREADED_COMPILER
340 return atomic_fetch_or_explicit(&block->flags, mask,
341 memory_order_relaxed);
342#else
343 u8 flags = block->flags;
344
345 block->flags |= mask;
346
347 return flags;
348#endif
349}
350
351static inline u8 block_clear_flags(struct block *block, u8 mask)
352{
353#if ENABLE_THREADED_COMPILER
354 return atomic_fetch_and_explicit(&block->flags, ~mask,
355 memory_order_relaxed);
356#else
357 u8 flags = block->flags;
358
359 block->flags &= ~mask;
360
361 return flags;
362#endif
363}
364
365static inline _Bool can_sign_extend(s32 value, u8 order)
366{
367 return ((u32)(value >> (order - 1)) + 1) < 2;
368}
369
370static inline _Bool can_zero_extend(u32 value, u8 order)
371{
372 return (value >> order) == 0;
373}
374
375static inline const struct opcode *
376get_delay_slot(const struct opcode *list, u16 i)
377{
378 return op_flag_no_ds(list[i].flags) ? &list[i - 1] : &list[i + 1];
379}
380
381static inline _Bool lightrec_store_next_pc(void)
382{
383 return NUM_REGS + NUM_TEMPS <= 4;
384}
385
386#endif /* __LIGHTREC_PRIVATE_H__ */