| 1 | /* |
| 2 | * (C) GraÅžvydas "notaz" Ignotas, 2011-2012 |
| 3 | * |
| 4 | * This work is licensed under the terms of any of these licenses |
| 5 | * (at your option): |
| 6 | * - GNU GPL, version 2 or later. |
| 7 | * - GNU LGPL, version 2.1 or later. |
| 8 | * See the COPYING file in the top-level directory. |
| 9 | */ |
| 10 | |
| 11 | #include <stdio.h> |
| 12 | #include <string.h> |
| 13 | #include "gpu.h" |
| 14 | |
| 15 | #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0])) |
| 16 | #ifdef __GNUC__ |
| 17 | #define unlikely(x) __builtin_expect((x), 0) |
| 18 | #define preload __builtin_prefetch |
| 19 | #define noinline __attribute__((noinline)) |
| 20 | #else |
| 21 | #define unlikely(x) |
| 22 | #define preload(...) |
| 23 | #define noinline |
| 24 | #endif |
| 25 | |
| 26 | #define gpu_log(fmt, ...) \ |
| 27 | printf("%d:%03d: " fmt, *gpu.state.frame_count, *gpu.state.hcnt, ##__VA_ARGS__) |
| 28 | |
| 29 | //#define log_io gpu_log |
| 30 | #define log_io(...) |
| 31 | //#define log_anomaly gpu_log |
| 32 | #define log_anomaly(...) |
| 33 | |
| 34 | struct psx_gpu gpu; |
| 35 | |
| 36 | static noinline int do_cmd_buffer(uint32_t *data, int count); |
| 37 | static void finish_vram_transfer(int is_read); |
| 38 | |
| 39 | static noinline void do_cmd_reset(void) |
| 40 | { |
| 41 | if (unlikely(gpu.cmd_len > 0)) |
| 42 | do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len); |
| 43 | gpu.cmd_len = 0; |
| 44 | |
| 45 | if (unlikely(gpu.dma.h > 0)) |
| 46 | finish_vram_transfer(gpu.dma_start.is_read); |
| 47 | gpu.dma.h = 0; |
| 48 | } |
| 49 | |
| 50 | static noinline void do_reset(void) |
| 51 | { |
| 52 | unsigned int i; |
| 53 | |
| 54 | do_cmd_reset(); |
| 55 | |
| 56 | memset(gpu.regs, 0, sizeof(gpu.regs)); |
| 57 | for (i = 0; i < sizeof(gpu.ex_regs) / sizeof(gpu.ex_regs[0]); i++) |
| 58 | gpu.ex_regs[i] = (0xe0 + i) << 24; |
| 59 | gpu.status = 0x14802000; |
| 60 | gpu.gp0 = 0; |
| 61 | gpu.regs[3] = 1; |
| 62 | gpu.screen.hres = gpu.screen.w = 256; |
| 63 | gpu.screen.vres = gpu.screen.h = 240; |
| 64 | } |
| 65 | |
| 66 | static noinline void update_width(void) |
| 67 | { |
| 68 | int sw = gpu.screen.x2 - gpu.screen.x1; |
| 69 | if (sw <= 0 || sw >= 2560) |
| 70 | // full width |
| 71 | gpu.screen.w = gpu.screen.hres; |
| 72 | else |
| 73 | gpu.screen.w = sw * gpu.screen.hres / 2560; |
| 74 | } |
| 75 | |
| 76 | static noinline void update_height(void) |
| 77 | { |
| 78 | // TODO: emulate this properly.. |
| 79 | int sh = gpu.screen.y2 - gpu.screen.y1; |
| 80 | if (gpu.status & PSX_GPU_STATUS_DHEIGHT) |
| 81 | sh *= 2; |
| 82 | if (sh <= 0 || sh > gpu.screen.vres) |
| 83 | sh = gpu.screen.vres; |
| 84 | |
| 85 | gpu.screen.h = sh; |
| 86 | } |
| 87 | |
| 88 | static noinline void decide_frameskip(void) |
| 89 | { |
| 90 | if (gpu.frameskip.active) |
| 91 | gpu.frameskip.cnt++; |
| 92 | else { |
| 93 | gpu.frameskip.cnt = 0; |
| 94 | gpu.frameskip.frame_ready = 1; |
| 95 | } |
| 96 | |
| 97 | if (!gpu.frameskip.active && *gpu.frameskip.advice) |
| 98 | gpu.frameskip.active = 1; |
| 99 | else if (gpu.frameskip.set > 0 && gpu.frameskip.cnt < gpu.frameskip.set) |
| 100 | gpu.frameskip.active = 1; |
| 101 | else |
| 102 | gpu.frameskip.active = 0; |
| 103 | |
| 104 | if (!gpu.frameskip.active && gpu.frameskip.pending_fill[0] != 0) { |
| 105 | int dummy; |
| 106 | do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy); |
| 107 | gpu.frameskip.pending_fill[0] = 0; |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | static noinline int decide_frameskip_allow(uint32_t cmd_e3) |
| 112 | { |
| 113 | // no frameskip if it decides to draw to display area, |
| 114 | // but not for interlace since it'll most likely always do that |
| 115 | uint32_t x = cmd_e3 & 0x3ff; |
| 116 | uint32_t y = (cmd_e3 >> 10) & 0x3ff; |
| 117 | gpu.frameskip.allow = (gpu.status & PSX_GPU_STATUS_INTERLACE) || |
| 118 | (uint32_t)(x - gpu.screen.x) >= (uint32_t)gpu.screen.w || |
| 119 | (uint32_t)(y - gpu.screen.y) >= (uint32_t)gpu.screen.h; |
| 120 | return gpu.frameskip.allow; |
| 121 | } |
| 122 | |
| 123 | static noinline void get_gpu_info(uint32_t data) |
| 124 | { |
| 125 | switch (data & 0x0f) { |
| 126 | case 0x02: |
| 127 | case 0x03: |
| 128 | case 0x04: |
| 129 | gpu.gp0 = gpu.ex_regs[data & 7] & 0xfffff; |
| 130 | break; |
| 131 | case 0x05: |
| 132 | gpu.gp0 = gpu.ex_regs[5] & 0x3fffff; |
| 133 | break; |
| 134 | case 0x07: |
| 135 | gpu.gp0 = 2; |
| 136 | break; |
| 137 | default: |
| 138 | // gpu.gp0 unchanged |
| 139 | break; |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | // double, for overdraw guard |
| 144 | #define VRAM_SIZE (1024 * 512 * 2 * 2) |
| 145 | |
| 146 | static int map_vram(void) |
| 147 | { |
| 148 | gpu.vram = gpu.mmap(VRAM_SIZE); |
| 149 | if (gpu.vram != NULL) { |
| 150 | gpu.vram += 4096 / 2; |
| 151 | return 0; |
| 152 | } |
| 153 | else { |
| 154 | fprintf(stderr, "could not map vram, expect crashes\n"); |
| 155 | return -1; |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | long GPUinit(void) |
| 160 | { |
| 161 | int ret; |
| 162 | ret = vout_init(); |
| 163 | ret |= renderer_init(); |
| 164 | |
| 165 | gpu.state.frame_count = &gpu.zero; |
| 166 | gpu.state.hcnt = &gpu.zero; |
| 167 | gpu.frameskip.active = 0; |
| 168 | gpu.cmd_len = 0; |
| 169 | do_reset(); |
| 170 | |
| 171 | if (gpu.mmap != NULL) { |
| 172 | if (map_vram() != 0) |
| 173 | ret = -1; |
| 174 | } |
| 175 | return ret; |
| 176 | } |
| 177 | |
| 178 | long GPUshutdown(void) |
| 179 | { |
| 180 | long ret; |
| 181 | |
| 182 | renderer_finish(); |
| 183 | ret = vout_finish(); |
| 184 | if (gpu.vram != NULL) { |
| 185 | gpu.vram -= 4096 / 2; |
| 186 | gpu.munmap(gpu.vram, VRAM_SIZE); |
| 187 | } |
| 188 | gpu.vram = NULL; |
| 189 | |
| 190 | return ret; |
| 191 | } |
| 192 | |
| 193 | void GPUwriteStatus(uint32_t data) |
| 194 | { |
| 195 | static const short hres[8] = { 256, 368, 320, 384, 512, 512, 640, 640 }; |
| 196 | static const short vres[4] = { 240, 480, 256, 480 }; |
| 197 | uint32_t cmd = data >> 24; |
| 198 | |
| 199 | if (cmd < ARRAY_SIZE(gpu.regs)) { |
| 200 | if (cmd > 1 && cmd != 5 && gpu.regs[cmd] == data) |
| 201 | return; |
| 202 | gpu.regs[cmd] = data; |
| 203 | } |
| 204 | |
| 205 | gpu.state.fb_dirty = 1; |
| 206 | |
| 207 | switch (cmd) { |
| 208 | case 0x00: |
| 209 | do_reset(); |
| 210 | break; |
| 211 | case 0x01: |
| 212 | do_cmd_reset(); |
| 213 | break; |
| 214 | case 0x03: |
| 215 | if (data & 1) |
| 216 | gpu.status |= PSX_GPU_STATUS_BLANKING; |
| 217 | else |
| 218 | gpu.status &= ~PSX_GPU_STATUS_BLANKING; |
| 219 | break; |
| 220 | case 0x04: |
| 221 | gpu.status &= ~PSX_GPU_STATUS_DMA_MASK; |
| 222 | gpu.status |= PSX_GPU_STATUS_DMA(data & 3); |
| 223 | break; |
| 224 | case 0x05: |
| 225 | gpu.screen.x = data & 0x3ff; |
| 226 | gpu.screen.y = (data >> 10) & 0x1ff; |
| 227 | if (gpu.frameskip.set) { |
| 228 | decide_frameskip_allow(gpu.ex_regs[3]); |
| 229 | if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) { |
| 230 | decide_frameskip(); |
| 231 | gpu.frameskip.last_flip_frame = *gpu.state.frame_count; |
| 232 | } |
| 233 | } |
| 234 | break; |
| 235 | case 0x06: |
| 236 | gpu.screen.x1 = data & 0xfff; |
| 237 | gpu.screen.x2 = (data >> 12) & 0xfff; |
| 238 | update_width(); |
| 239 | break; |
| 240 | case 0x07: |
| 241 | gpu.screen.y1 = data & 0x3ff; |
| 242 | gpu.screen.y2 = (data >> 10) & 0x3ff; |
| 243 | update_height(); |
| 244 | break; |
| 245 | case 0x08: |
| 246 | gpu.status = (gpu.status & ~0x7f0000) | ((data & 0x3F) << 17) | ((data & 0x40) << 10); |
| 247 | gpu.screen.hres = hres[(gpu.status >> 16) & 7]; |
| 248 | gpu.screen.vres = vres[(gpu.status >> 19) & 3]; |
| 249 | update_width(); |
| 250 | update_height(); |
| 251 | renderer_notify_res_change(); |
| 252 | break; |
| 253 | default: |
| 254 | if ((cmd & 0xf0) == 0x10) |
| 255 | get_gpu_info(data); |
| 256 | break; |
| 257 | } |
| 258 | |
| 259 | #ifdef GPUwriteStatus_ext |
| 260 | GPUwriteStatus_ext(data); |
| 261 | #endif |
| 262 | } |
| 263 | |
| 264 | const unsigned char cmd_lengths[256] = |
| 265 | { |
| 266 | 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 267 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 268 | 3, 3, 3, 3, 6, 6, 6, 6, 4, 4, 4, 4, 8, 8, 8, 8, // 20 |
| 269 | 5, 5, 5, 5, 8, 8, 8, 8, 7, 7, 7, 7, 11, 11, 11, 11, |
| 270 | 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, // 40 |
| 271 | 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, |
| 272 | 2, 2, 2, 2, 3, 3, 3, 3, 1, 1, 1, 1, 0, 0, 0, 0, // 60 |
| 273 | 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 2, 2, 2, 2, |
| 274 | 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // 80 |
| 275 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 276 | 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // a0 |
| 277 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 278 | 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // c0 |
| 279 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 280 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // e0 |
| 281 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 |
| 282 | }; |
| 283 | |
| 284 | #define VRAM_MEM_XY(x, y) &gpu.vram[(y) * 1024 + (x)] |
| 285 | |
| 286 | static inline void do_vram_line(int x, int y, uint16_t *mem, int l, int is_read) |
| 287 | { |
| 288 | uint16_t *vram = VRAM_MEM_XY(x, y); |
| 289 | if (is_read) |
| 290 | memcpy(mem, vram, l * 2); |
| 291 | else |
| 292 | memcpy(vram, mem, l * 2); |
| 293 | } |
| 294 | |
| 295 | static int do_vram_io(uint32_t *data, int count, int is_read) |
| 296 | { |
| 297 | int count_initial = count; |
| 298 | uint16_t *sdata = (uint16_t *)data; |
| 299 | int x = gpu.dma.x, y = gpu.dma.y; |
| 300 | int w = gpu.dma.w, h = gpu.dma.h; |
| 301 | int o = gpu.dma.offset; |
| 302 | int l; |
| 303 | count *= 2; // operate in 16bpp pixels |
| 304 | |
| 305 | if (gpu.dma.offset) { |
| 306 | l = w - gpu.dma.offset; |
| 307 | if (count < l) |
| 308 | l = count; |
| 309 | |
| 310 | do_vram_line(x + o, y, sdata, l, is_read); |
| 311 | |
| 312 | if (o + l < w) |
| 313 | o += l; |
| 314 | else { |
| 315 | o = 0; |
| 316 | y++; |
| 317 | h--; |
| 318 | } |
| 319 | sdata += l; |
| 320 | count -= l; |
| 321 | } |
| 322 | |
| 323 | for (; h > 0 && count >= w; sdata += w, count -= w, y++, h--) { |
| 324 | y &= 511; |
| 325 | do_vram_line(x, y, sdata, w, is_read); |
| 326 | } |
| 327 | |
| 328 | if (h > 0) { |
| 329 | if (count > 0) { |
| 330 | y &= 511; |
| 331 | do_vram_line(x, y, sdata, count, is_read); |
| 332 | o = count; |
| 333 | count = 0; |
| 334 | } |
| 335 | } |
| 336 | else |
| 337 | finish_vram_transfer(is_read); |
| 338 | gpu.dma.y = y; |
| 339 | gpu.dma.h = h; |
| 340 | gpu.dma.offset = o; |
| 341 | |
| 342 | return count_initial - count / 2; |
| 343 | } |
| 344 | |
| 345 | static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_read) |
| 346 | { |
| 347 | if (gpu.dma.h) |
| 348 | log_anomaly("start_vram_transfer while old unfinished\n"); |
| 349 | |
| 350 | gpu.dma.x = pos_word & 0x3ff; |
| 351 | gpu.dma.y = (pos_word >> 16) & 0x1ff; |
| 352 | gpu.dma.w = ((size_word - 1) & 0x3ff) + 1; |
| 353 | gpu.dma.h = (((size_word >> 16) - 1) & 0x1ff) + 1; |
| 354 | gpu.dma.offset = 0; |
| 355 | gpu.dma.is_read = is_read; |
| 356 | gpu.dma_start = gpu.dma; |
| 357 | |
| 358 | renderer_flush_queues(); |
| 359 | if (is_read) { |
| 360 | gpu.status |= PSX_GPU_STATUS_IMG; |
| 361 | // XXX: wrong for width 1 |
| 362 | gpu.gp0 = LE32TOH(*(uint32_t *) VRAM_MEM_XY(gpu.dma.x, gpu.dma.y)); |
| 363 | gpu.state.last_vram_read_frame = *gpu.state.frame_count; |
| 364 | } |
| 365 | |
| 366 | log_io("start_vram_transfer %c (%d, %d) %dx%d\n", is_read ? 'r' : 'w', |
| 367 | gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h); |
| 368 | } |
| 369 | |
| 370 | static void finish_vram_transfer(int is_read) |
| 371 | { |
| 372 | if (is_read) |
| 373 | gpu.status &= ~PSX_GPU_STATUS_IMG; |
| 374 | else |
| 375 | renderer_update_caches(gpu.dma_start.x, gpu.dma_start.y, |
| 376 | gpu.dma_start.w, gpu.dma_start.h); |
| 377 | } |
| 378 | |
| 379 | static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd) |
| 380 | { |
| 381 | int cmd = 0, pos = 0, len, dummy, v; |
| 382 | int skip = 1; |
| 383 | |
| 384 | gpu.frameskip.pending_fill[0] = 0; |
| 385 | |
| 386 | while (pos < count && skip) { |
| 387 | uint32_t *list = data + pos; |
| 388 | cmd = LE32TOH(list[0]) >> 24; |
| 389 | len = 1 + cmd_lengths[cmd]; |
| 390 | |
| 391 | switch (cmd) { |
| 392 | case 0x02: |
| 393 | if ((LE32TOH(list[2]) & 0x3ff) > gpu.screen.w || ((LE32TOH(list[2]) >> 16) & 0x1ff) > gpu.screen.h) |
| 394 | // clearing something large, don't skip |
| 395 | do_cmd_list(list, 3, &dummy); |
| 396 | else |
| 397 | memcpy(gpu.frameskip.pending_fill, list, 3 * 4); |
| 398 | break; |
| 399 | case 0x24 ... 0x27: |
| 400 | case 0x2c ... 0x2f: |
| 401 | case 0x34 ... 0x37: |
| 402 | case 0x3c ... 0x3f: |
| 403 | gpu.ex_regs[1] &= ~0x1ff; |
| 404 | gpu.ex_regs[1] |= LE32TOH(list[4 + ((cmd >> 4) & 1)]) & 0x1ff; |
| 405 | break; |
| 406 | case 0x48 ... 0x4F: |
| 407 | for (v = 3; pos + v < count; v++) |
| 408 | { |
| 409 | if ((list[v] & HTOLE32(0xf000f000)) == HTOLE32(0x50005000)) |
| 410 | break; |
| 411 | } |
| 412 | len += v - 3; |
| 413 | break; |
| 414 | case 0x58 ... 0x5F: |
| 415 | for (v = 4; pos + v < count; v += 2) |
| 416 | { |
| 417 | if ((list[v] & HTOLE32(0xf000f000)) == HTOLE32(0x50005000)) |
| 418 | break; |
| 419 | } |
| 420 | len += v - 4; |
| 421 | break; |
| 422 | default: |
| 423 | if (cmd == 0xe3) |
| 424 | skip = decide_frameskip_allow(LE32TOH(list[0])); |
| 425 | if ((cmd & 0xf8) == 0xe0) |
| 426 | gpu.ex_regs[cmd & 7] = LE32TOH(list[0]); |
| 427 | break; |
| 428 | } |
| 429 | |
| 430 | if (pos + len > count) { |
| 431 | cmd = -1; |
| 432 | break; // incomplete cmd |
| 433 | } |
| 434 | if (0xa0 <= cmd && cmd <= 0xdf) |
| 435 | break; // image i/o |
| 436 | |
| 437 | pos += len; |
| 438 | } |
| 439 | |
| 440 | renderer_sync_ecmds(gpu.ex_regs); |
| 441 | *last_cmd = cmd; |
| 442 | return pos; |
| 443 | } |
| 444 | |
| 445 | static noinline int do_cmd_buffer(uint32_t *data, int count) |
| 446 | { |
| 447 | int cmd, pos; |
| 448 | uint32_t old_e3 = gpu.ex_regs[3]; |
| 449 | int vram_dirty = 0; |
| 450 | |
| 451 | // process buffer |
| 452 | for (pos = 0; pos < count; ) |
| 453 | { |
| 454 | if (gpu.dma.h && !gpu.dma_start.is_read) { // XXX: need to verify |
| 455 | vram_dirty = 1; |
| 456 | pos += do_vram_io(data + pos, count - pos, 0); |
| 457 | if (pos == count) |
| 458 | break; |
| 459 | } |
| 460 | |
| 461 | cmd = LE32TOH(data[pos]) >> 24; |
| 462 | if (0xa0 <= cmd && cmd <= 0xdf) { |
| 463 | if (unlikely((pos+2) >= count)) { |
| 464 | // incomplete vram write/read cmd, can't consume yet |
| 465 | cmd = -1; |
| 466 | break; |
| 467 | } |
| 468 | |
| 469 | // consume vram write/read cmd |
| 470 | start_vram_transfer(LE32TOH(data[pos + 1]), LE32TOH(data[pos + 2]), (cmd & 0xe0) == 0xc0); |
| 471 | pos += 3; |
| 472 | continue; |
| 473 | } |
| 474 | |
| 475 | // 0xex cmds might affect frameskip.allow, so pass to do_cmd_list_skip |
| 476 | if (gpu.frameskip.active && (gpu.frameskip.allow || ((LE32TOH(data[pos]) >> 24) & 0xf0) == 0xe0)) |
| 477 | pos += do_cmd_list_skip(data + pos, count - pos, &cmd); |
| 478 | else { |
| 479 | pos += do_cmd_list(data + pos, count - pos, &cmd); |
| 480 | vram_dirty = 1; |
| 481 | } |
| 482 | |
| 483 | if (cmd == -1) |
| 484 | // incomplete cmd |
| 485 | break; |
| 486 | } |
| 487 | |
| 488 | gpu.status &= ~0x1fff; |
| 489 | gpu.status |= gpu.ex_regs[1] & 0x7ff; |
| 490 | gpu.status |= (gpu.ex_regs[6] & 3) << 11; |
| 491 | |
| 492 | gpu.state.fb_dirty |= vram_dirty; |
| 493 | |
| 494 | if (old_e3 != gpu.ex_regs[3]) |
| 495 | decide_frameskip_allow(gpu.ex_regs[3]); |
| 496 | |
| 497 | return count - pos; |
| 498 | } |
| 499 | |
| 500 | static void flush_cmd_buffer(void) |
| 501 | { |
| 502 | int left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len); |
| 503 | if (left > 0) |
| 504 | memmove(gpu.cmd_buffer, gpu.cmd_buffer + gpu.cmd_len - left, left * 4); |
| 505 | gpu.cmd_len = left; |
| 506 | } |
| 507 | |
| 508 | void GPUwriteDataMem(uint32_t *mem, int count) |
| 509 | { |
| 510 | int left; |
| 511 | |
| 512 | log_io("gpu_dma_write %p %d\n", mem, count); |
| 513 | |
| 514 | if (unlikely(gpu.cmd_len > 0)) |
| 515 | flush_cmd_buffer(); |
| 516 | |
| 517 | left = do_cmd_buffer(mem, count); |
| 518 | if (left) |
| 519 | log_anomaly("GPUwriteDataMem: discarded %d/%d words\n", left, count); |
| 520 | } |
| 521 | |
| 522 | void GPUwriteData(uint32_t data) |
| 523 | { |
| 524 | log_io("gpu_write %08x\n", data); |
| 525 | gpu.cmd_buffer[gpu.cmd_len++] = HTOLE32(data); |
| 526 | if (gpu.cmd_len >= CMD_BUFFER_LEN) |
| 527 | flush_cmd_buffer(); |
| 528 | } |
| 529 | |
| 530 | long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr) |
| 531 | { |
| 532 | uint32_t addr, *list, ld_addr = 0; |
| 533 | int len, left, count; |
| 534 | long cpu_cycles = 0; |
| 535 | |
| 536 | preload(rambase + (start_addr & 0x1fffff) / 4); |
| 537 | |
| 538 | if (unlikely(gpu.cmd_len > 0)) |
| 539 | flush_cmd_buffer(); |
| 540 | |
| 541 | log_io("gpu_dma_chain\n"); |
| 542 | addr = start_addr & 0xffffff; |
| 543 | for (count = 0; (addr & 0x800000) == 0; count++) |
| 544 | { |
| 545 | list = rambase + (addr & 0x1fffff) / 4; |
| 546 | len = LE32TOH(list[0]) >> 24; |
| 547 | addr = LE32TOH(list[0]) & 0xffffff; |
| 548 | preload(rambase + (addr & 0x1fffff) / 4); |
| 549 | |
| 550 | cpu_cycles += 10; |
| 551 | if (len > 0) |
| 552 | cpu_cycles += 5 + len; |
| 553 | |
| 554 | log_io(".chain %08lx #%d+%d\n", |
| 555 | (long)(list - rambase) * 4, len, gpu.cmd_len); |
| 556 | if (unlikely(gpu.cmd_len > 0)) { |
| 557 | memcpy(gpu.cmd_buffer + gpu.cmd_len, list + 1, len * 4); |
| 558 | gpu.cmd_len += len; |
| 559 | flush_cmd_buffer(); |
| 560 | continue; |
| 561 | } |
| 562 | |
| 563 | if (len) { |
| 564 | left = do_cmd_buffer(list + 1, len); |
| 565 | if (left) { |
| 566 | memcpy(gpu.cmd_buffer, list + 1 + len - left, left * 4); |
| 567 | gpu.cmd_len = left; |
| 568 | log_anomaly("GPUdmaChain: %d/%d words left\n", left, len); |
| 569 | } |
| 570 | } |
| 571 | |
| 572 | if (progress_addr) { |
| 573 | *progress_addr = addr; |
| 574 | break; |
| 575 | } |
| 576 | #define LD_THRESHOLD (8*1024) |
| 577 | if (count >= LD_THRESHOLD) { |
| 578 | if (count == LD_THRESHOLD) { |
| 579 | ld_addr = addr; |
| 580 | continue; |
| 581 | } |
| 582 | |
| 583 | // loop detection marker |
| 584 | // (bit23 set causes DMA error on real machine, so |
| 585 | // unlikely to be ever set by the game) |
| 586 | list[0] |= HTOLE32(0x800000); |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | if (ld_addr != 0) { |
| 591 | // remove loop detection markers |
| 592 | count -= LD_THRESHOLD + 2; |
| 593 | addr = ld_addr & 0x1fffff; |
| 594 | while (count-- > 0) { |
| 595 | list = rambase + addr / 4; |
| 596 | addr = LE32TOH(list[0]) & 0x1fffff; |
| 597 | list[0] &= HTOLE32(~0x800000); |
| 598 | } |
| 599 | } |
| 600 | |
| 601 | gpu.state.last_list.frame = *gpu.state.frame_count; |
| 602 | gpu.state.last_list.hcnt = *gpu.state.hcnt; |
| 603 | gpu.state.last_list.cycles = cpu_cycles; |
| 604 | gpu.state.last_list.addr = start_addr; |
| 605 | |
| 606 | return cpu_cycles; |
| 607 | } |
| 608 | |
| 609 | void GPUreadDataMem(uint32_t *mem, int count) |
| 610 | { |
| 611 | log_io("gpu_dma_read %p %d\n", mem, count); |
| 612 | |
| 613 | if (unlikely(gpu.cmd_len > 0)) |
| 614 | flush_cmd_buffer(); |
| 615 | |
| 616 | if (gpu.dma.h) |
| 617 | do_vram_io(mem, count, 1); |
| 618 | } |
| 619 | |
| 620 | uint32_t GPUreadData(void) |
| 621 | { |
| 622 | uint32_t ret; |
| 623 | |
| 624 | if (unlikely(gpu.cmd_len > 0)) |
| 625 | flush_cmd_buffer(); |
| 626 | |
| 627 | ret = gpu.gp0; |
| 628 | if (gpu.dma.h) { |
| 629 | ret = HTOLE32(ret); |
| 630 | do_vram_io(&ret, 1, 1); |
| 631 | ret = LE32TOH(ret); |
| 632 | } |
| 633 | |
| 634 | log_io("gpu_read %08x\n", ret); |
| 635 | return ret; |
| 636 | } |
| 637 | |
| 638 | uint32_t GPUreadStatus(void) |
| 639 | { |
| 640 | uint32_t ret; |
| 641 | |
| 642 | if (unlikely(gpu.cmd_len > 0)) |
| 643 | flush_cmd_buffer(); |
| 644 | |
| 645 | ret = gpu.status; |
| 646 | log_io("gpu_read_status %08x\n", ret); |
| 647 | return ret; |
| 648 | } |
| 649 | |
| 650 | struct GPUFreeze |
| 651 | { |
| 652 | uint32_t ulFreezeVersion; // should be always 1 for now (set by main emu) |
| 653 | uint32_t ulStatus; // current gpu status |
| 654 | uint32_t ulControl[256]; // latest control register values |
| 655 | unsigned char psxVRam[1024*1024*2]; // current VRam image (full 2 MB for ZN) |
| 656 | }; |
| 657 | |
| 658 | long GPUfreeze(uint32_t type, struct GPUFreeze *freeze) |
| 659 | { |
| 660 | int i; |
| 661 | |
| 662 | switch (type) { |
| 663 | case 1: // save |
| 664 | if (gpu.cmd_len > 0) |
| 665 | flush_cmd_buffer(); |
| 666 | memcpy(freeze->psxVRam, gpu.vram, 1024 * 512 * 2); |
| 667 | memcpy(freeze->ulControl, gpu.regs, sizeof(gpu.regs)); |
| 668 | memcpy(freeze->ulControl + 0xe0, gpu.ex_regs, sizeof(gpu.ex_regs)); |
| 669 | freeze->ulStatus = gpu.status; |
| 670 | break; |
| 671 | case 0: // load |
| 672 | memcpy(gpu.vram, freeze->psxVRam, 1024 * 512 * 2); |
| 673 | memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs)); |
| 674 | memcpy(gpu.ex_regs, freeze->ulControl + 0xe0, sizeof(gpu.ex_regs)); |
| 675 | gpu.status = freeze->ulStatus; |
| 676 | gpu.cmd_len = 0; |
| 677 | for (i = 8; i > 0; i--) { |
| 678 | gpu.regs[i] ^= 1; // avoid reg change detection |
| 679 | GPUwriteStatus((i << 24) | (gpu.regs[i] ^ 1)); |
| 680 | } |
| 681 | renderer_sync_ecmds(gpu.ex_regs); |
| 682 | renderer_update_caches(0, 0, 1024, 512); |
| 683 | break; |
| 684 | } |
| 685 | |
| 686 | return 1; |
| 687 | } |
| 688 | |
| 689 | void GPUupdateLace(void) |
| 690 | { |
| 691 | if (gpu.cmd_len > 0) |
| 692 | flush_cmd_buffer(); |
| 693 | renderer_flush_queues(); |
| 694 | |
| 695 | if (gpu.status & PSX_GPU_STATUS_BLANKING) { |
| 696 | if (!gpu.state.blanked) { |
| 697 | vout_blank(); |
| 698 | gpu.state.blanked = 1; |
| 699 | gpu.state.fb_dirty = 1; |
| 700 | } |
| 701 | return; |
| 702 | } |
| 703 | |
| 704 | if (!gpu.state.fb_dirty) |
| 705 | return; |
| 706 | |
| 707 | if (gpu.frameskip.set) { |
| 708 | if (!gpu.frameskip.frame_ready) { |
| 709 | if (*gpu.state.frame_count - gpu.frameskip.last_flip_frame < 9) |
| 710 | return; |
| 711 | gpu.frameskip.active = 0; |
| 712 | } |
| 713 | gpu.frameskip.frame_ready = 0; |
| 714 | } |
| 715 | |
| 716 | vout_update(); |
| 717 | gpu.state.fb_dirty = 0; |
| 718 | gpu.state.blanked = 0; |
| 719 | } |
| 720 | |
| 721 | void GPUvBlank(int is_vblank, int lcf) |
| 722 | { |
| 723 | int interlace = gpu.state.allow_interlace |
| 724 | && (gpu.status & PSX_GPU_STATUS_INTERLACE) |
| 725 | && (gpu.status & PSX_GPU_STATUS_DHEIGHT); |
| 726 | // interlace doesn't look nice on progressive displays, |
| 727 | // so we have this "auto" mode here for games that don't read vram |
| 728 | if (gpu.state.allow_interlace == 2 |
| 729 | && *gpu.state.frame_count - gpu.state.last_vram_read_frame > 1) |
| 730 | { |
| 731 | interlace = 0; |
| 732 | } |
| 733 | if (interlace || interlace != gpu.state.old_interlace) { |
| 734 | gpu.state.old_interlace = interlace; |
| 735 | |
| 736 | if (gpu.cmd_len > 0) |
| 737 | flush_cmd_buffer(); |
| 738 | renderer_flush_queues(); |
| 739 | renderer_set_interlace(interlace, !lcf); |
| 740 | } |
| 741 | } |
| 742 | |
| 743 | #include "../../frontend/plugin_lib.h" |
| 744 | |
| 745 | void GPUrearmedCallbacks(const struct rearmed_cbs *cbs) |
| 746 | { |
| 747 | gpu.frameskip.set = cbs->frameskip; |
| 748 | gpu.frameskip.advice = &cbs->fskip_advice; |
| 749 | gpu.frameskip.active = 0; |
| 750 | gpu.frameskip.frame_ready = 1; |
| 751 | gpu.state.hcnt = cbs->gpu_hcnt; |
| 752 | gpu.state.frame_count = cbs->gpu_frame_count; |
| 753 | gpu.state.allow_interlace = cbs->gpu_neon.allow_interlace; |
| 754 | gpu.state.enhancement_enable = cbs->gpu_neon.enhancement_enable; |
| 755 | |
| 756 | gpu.mmap = cbs->mmap; |
| 757 | gpu.munmap = cbs->munmap; |
| 758 | |
| 759 | // delayed vram mmap |
| 760 | if (gpu.vram == NULL) |
| 761 | map_vram(); |
| 762 | |
| 763 | if (cbs->pl_vout_set_raw_vram) |
| 764 | cbs->pl_vout_set_raw_vram(gpu.vram); |
| 765 | renderer_set_config(cbs); |
| 766 | vout_set_config(cbs); |
| 767 | } |
| 768 | |
| 769 | // vim:shiftwidth=2:expandtab |