| 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 <stdlib.h> |
| 13 | #include <string.h> |
| 14 | #include "gpu.h" |
| 15 | #include "gpu_timing.h" |
| 16 | #include "../../libpcsxcore/gpu.h" // meh |
| 17 | #include "../../frontend/plugin_lib.h" |
| 18 | |
| 19 | #ifndef ARRAY_SIZE |
| 20 | #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0])) |
| 21 | #endif |
| 22 | #ifdef __GNUC__ |
| 23 | #define unlikely(x) __builtin_expect((x), 0) |
| 24 | #define preload __builtin_prefetch |
| 25 | #define noinline __attribute__((noinline)) |
| 26 | #else |
| 27 | #define unlikely(x) |
| 28 | #define preload(...) |
| 29 | #define noinline |
| 30 | #endif |
| 31 | |
| 32 | //#define log_io gpu_log |
| 33 | #define log_io(...) |
| 34 | |
| 35 | struct psx_gpu gpu; |
| 36 | |
| 37 | static noinline int do_cmd_buffer(uint32_t *data, int count, |
| 38 | int *cycles_sum, int *cycles_last); |
| 39 | static void finish_vram_transfer(int is_read); |
| 40 | |
| 41 | static noinline void do_cmd_reset(void) |
| 42 | { |
| 43 | int dummy = 0; |
| 44 | if (unlikely(gpu.cmd_len > 0)) |
| 45 | do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy, &dummy); |
| 46 | gpu.cmd_len = 0; |
| 47 | |
| 48 | if (unlikely(gpu.dma.h > 0)) |
| 49 | finish_vram_transfer(gpu.dma_start.is_read); |
| 50 | gpu.dma.h = 0; |
| 51 | } |
| 52 | |
| 53 | static noinline void do_reset(void) |
| 54 | { |
| 55 | unsigned int i; |
| 56 | |
| 57 | do_cmd_reset(); |
| 58 | |
| 59 | memset(gpu.regs, 0, sizeof(gpu.regs)); |
| 60 | for (i = 0; i < sizeof(gpu.ex_regs) / sizeof(gpu.ex_regs[0]); i++) |
| 61 | gpu.ex_regs[i] = (0xe0 + i) << 24; |
| 62 | gpu.status = 0x14802000; |
| 63 | gpu.gp0 = 0; |
| 64 | gpu.regs[3] = 1; |
| 65 | gpu.screen.hres = gpu.screen.w = 256; |
| 66 | gpu.screen.vres = gpu.screen.h = 240; |
| 67 | gpu.screen.x = gpu.screen.y = 0; |
| 68 | renderer_sync_ecmds(gpu.ex_regs); |
| 69 | renderer_notify_res_change(); |
| 70 | } |
| 71 | |
| 72 | static noinline void update_width(void) |
| 73 | { |
| 74 | static const short hres_all[8] = { 256, 368, 320, 368, 512, 368, 640, 368 }; |
| 75 | static const uint8_t hdivs[8] = { 10, 7, 8, 7, 5, 7, 4, 7 }; |
| 76 | uint8_t hdiv = hdivs[(gpu.status >> 16) & 7]; |
| 77 | int hres = hres_all[(gpu.status >> 16) & 7]; |
| 78 | int pal = gpu.status & PSX_GPU_STATUS_PAL; |
| 79 | int sw = gpu.screen.x2 - gpu.screen.x1; |
| 80 | int type = gpu.state.screen_centering_type; |
| 81 | int x = 0, x_auto; |
| 82 | if (type == C_AUTO) |
| 83 | type = gpu.state.screen_centering_type_default; |
| 84 | if (sw <= 0) |
| 85 | /* nothing displayed? */; |
| 86 | else { |
| 87 | int s = pal ? 656 : 608; // or 600? pal is just a guess |
| 88 | x = (gpu.screen.x1 - s) / hdiv; |
| 89 | x = (x + 1) & ~1; // blitter limitation |
| 90 | sw /= hdiv; |
| 91 | sw = (sw + 2) & ~3; // according to nocash |
| 92 | |
| 93 | if (gpu.state.show_overscan == 2) // widescreen hack |
| 94 | sw = (sw + 63) & ~63; |
| 95 | if (gpu.state.show_overscan && sw >= hres) |
| 96 | x = 0, hres = sw; |
| 97 | switch (type) { |
| 98 | case C_INGAME: |
| 99 | break; |
| 100 | case C_MANUAL: |
| 101 | x = gpu.state.screen_centering_x; |
| 102 | break; |
| 103 | default: |
| 104 | // correct if slightly miscentered |
| 105 | x_auto = (hres - sw) / 2 & ~3; |
| 106 | if ((uint32_t)x_auto <= 8u && abs(x) < 24) |
| 107 | x = x_auto; |
| 108 | } |
| 109 | if (x + sw > hres) |
| 110 | sw = hres - x; |
| 111 | // .x range check is done in vout_update() |
| 112 | } |
| 113 | // reduce the unpleasant right border that a few games have |
| 114 | if (gpu.state.screen_centering_type == 0 |
| 115 | && x <= 4 && hres - (x + sw) >= 4) |
| 116 | hres -= 4; |
| 117 | gpu.screen.x = x; |
| 118 | gpu.screen.w = sw; |
| 119 | gpu.screen.hres = hres; |
| 120 | gpu.state.dims_changed = 1; |
| 121 | //printf("xx %d %d (%d) -> %2d, %d / %d\n", gpu.screen.x1, |
| 122 | // gpu.screen.x2, gpu.screen.x2 - gpu.screen.x1, x, sw, hres); |
| 123 | } |
| 124 | |
| 125 | static noinline void update_height(void) |
| 126 | { |
| 127 | int pal = gpu.status & PSX_GPU_STATUS_PAL; |
| 128 | int dheight = gpu.status & PSX_GPU_STATUS_DHEIGHT; |
| 129 | int y = gpu.screen.y1 - (pal ? 39 : 16); // 39 for spyro |
| 130 | int sh = gpu.screen.y2 - gpu.screen.y1; |
| 131 | int center_tol = 16; |
| 132 | int vres = 240; |
| 133 | |
| 134 | if (pal && (sh > 240 || gpu.screen.vres == 256)) |
| 135 | vres = 256; |
| 136 | if (dheight) |
| 137 | y *= 2, sh *= 2, vres *= 2, center_tol *= 2; |
| 138 | if (sh <= 0) |
| 139 | /* nothing displayed? */; |
| 140 | else { |
| 141 | switch (gpu.state.screen_centering_type) { |
| 142 | case C_INGAME: |
| 143 | break; |
| 144 | case C_BORDERLESS: |
| 145 | y = 0; |
| 146 | break; |
| 147 | case C_MANUAL: |
| 148 | y = gpu.state.screen_centering_y; |
| 149 | break; |
| 150 | default: |
| 151 | // correct if slightly miscentered |
| 152 | if ((uint32_t)(vres - sh) <= 1 && abs(y) <= center_tol) |
| 153 | y = 0; |
| 154 | } |
| 155 | if (y + sh > vres) |
| 156 | sh = vres - y; |
| 157 | } |
| 158 | gpu.screen.y = y; |
| 159 | gpu.screen.h = sh; |
| 160 | gpu.screen.vres = vres; |
| 161 | gpu.state.dims_changed = 1; |
| 162 | //printf("yy %d %d -> %d, %d / %d\n", |
| 163 | // gpu.screen.y1, gpu.screen.y2, y, sh, vres); |
| 164 | } |
| 165 | |
| 166 | static noinline void decide_frameskip(void) |
| 167 | { |
| 168 | if (gpu.frameskip.active) |
| 169 | gpu.frameskip.cnt++; |
| 170 | else { |
| 171 | gpu.frameskip.cnt = 0; |
| 172 | gpu.frameskip.frame_ready = 1; |
| 173 | } |
| 174 | |
| 175 | if (!gpu.frameskip.active && *gpu.frameskip.advice) |
| 176 | gpu.frameskip.active = 1; |
| 177 | else if (gpu.frameskip.set > 0 && gpu.frameskip.cnt < gpu.frameskip.set) |
| 178 | gpu.frameskip.active = 1; |
| 179 | else |
| 180 | gpu.frameskip.active = 0; |
| 181 | |
| 182 | if (!gpu.frameskip.active && gpu.frameskip.pending_fill[0] != 0) { |
| 183 | int dummy = 0; |
| 184 | do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy, &dummy, &dummy); |
| 185 | gpu.frameskip.pending_fill[0] = 0; |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | static noinline int decide_frameskip_allow(uint32_t cmd_e3) |
| 190 | { |
| 191 | // no frameskip if it decides to draw to display area, |
| 192 | // but not for interlace since it'll most likely always do that |
| 193 | uint32_t x = cmd_e3 & 0x3ff; |
| 194 | uint32_t y = (cmd_e3 >> 10) & 0x3ff; |
| 195 | gpu.frameskip.allow = (gpu.status & PSX_GPU_STATUS_INTERLACE) || |
| 196 | (uint32_t)(x - gpu.screen.src_x) >= (uint32_t)gpu.screen.w || |
| 197 | (uint32_t)(y - gpu.screen.src_y) >= (uint32_t)gpu.screen.h; |
| 198 | return gpu.frameskip.allow; |
| 199 | } |
| 200 | |
| 201 | static void flush_cmd_buffer(void); |
| 202 | |
| 203 | static noinline void get_gpu_info(uint32_t data) |
| 204 | { |
| 205 | if (unlikely(gpu.cmd_len > 0)) |
| 206 | flush_cmd_buffer(); |
| 207 | switch (data & 0x0f) { |
| 208 | case 0x02: |
| 209 | case 0x03: |
| 210 | case 0x04: |
| 211 | gpu.gp0 = gpu.ex_regs[data & 7] & 0xfffff; |
| 212 | break; |
| 213 | case 0x05: |
| 214 | gpu.gp0 = gpu.ex_regs[5] & 0x3fffff; |
| 215 | break; |
| 216 | case 0x07: |
| 217 | gpu.gp0 = 2; |
| 218 | break; |
| 219 | default: |
| 220 | // gpu.gp0 unchanged |
| 221 | break; |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | // double, for overdraw guard |
| 226 | #define VRAM_SIZE (1024 * 512 * 2 * 2) |
| 227 | |
| 228 | static int map_vram(void) |
| 229 | { |
| 230 | gpu.vram = gpu.mmap(VRAM_SIZE); |
| 231 | if (gpu.vram != NULL) { |
| 232 | gpu.vram += 4096 / 2; |
| 233 | return 0; |
| 234 | } |
| 235 | else { |
| 236 | fprintf(stderr, "could not map vram, expect crashes\n"); |
| 237 | return -1; |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | long GPUinit(void) |
| 242 | { |
| 243 | int ret; |
| 244 | ret = vout_init(); |
| 245 | ret |= renderer_init(); |
| 246 | |
| 247 | memset(&gpu.state, 0, sizeof(gpu.state)); |
| 248 | memset(&gpu.frameskip, 0, sizeof(gpu.frameskip)); |
| 249 | gpu.zero = 0; |
| 250 | gpu.state.frame_count = &gpu.zero; |
| 251 | gpu.state.hcnt = &gpu.zero; |
| 252 | gpu.cmd_len = 0; |
| 253 | do_reset(); |
| 254 | |
| 255 | if (gpu.mmap != NULL) { |
| 256 | if (map_vram() != 0) |
| 257 | ret = -1; |
| 258 | } |
| 259 | return ret; |
| 260 | } |
| 261 | |
| 262 | long GPUshutdown(void) |
| 263 | { |
| 264 | long ret; |
| 265 | |
| 266 | renderer_finish(); |
| 267 | ret = vout_finish(); |
| 268 | if (gpu.vram != NULL) { |
| 269 | gpu.vram -= 4096 / 2; |
| 270 | gpu.munmap(gpu.vram, VRAM_SIZE); |
| 271 | } |
| 272 | gpu.vram = NULL; |
| 273 | |
| 274 | return ret; |
| 275 | } |
| 276 | |
| 277 | void GPUwriteStatus(uint32_t data) |
| 278 | { |
| 279 | uint32_t cmd = data >> 24; |
| 280 | int src_x, src_y; |
| 281 | |
| 282 | if (cmd < ARRAY_SIZE(gpu.regs)) { |
| 283 | if (cmd > 1 && cmd != 5 && gpu.regs[cmd] == data) |
| 284 | return; |
| 285 | gpu.regs[cmd] = data; |
| 286 | } |
| 287 | |
| 288 | gpu.state.fb_dirty = 1; |
| 289 | |
| 290 | switch (cmd) { |
| 291 | case 0x00: |
| 292 | do_reset(); |
| 293 | break; |
| 294 | case 0x01: |
| 295 | do_cmd_reset(); |
| 296 | break; |
| 297 | case 0x03: |
| 298 | if (data & 1) { |
| 299 | gpu.status |= PSX_GPU_STATUS_BLANKING; |
| 300 | gpu.state.dims_changed = 1; // for hud clearing |
| 301 | } |
| 302 | else |
| 303 | gpu.status &= ~PSX_GPU_STATUS_BLANKING; |
| 304 | break; |
| 305 | case 0x04: |
| 306 | gpu.status &= ~PSX_GPU_STATUS_DMA_MASK; |
| 307 | gpu.status |= PSX_GPU_STATUS_DMA(data & 3); |
| 308 | break; |
| 309 | case 0x05: |
| 310 | src_x = data & 0x3ff; src_y = (data >> 10) & 0x1ff; |
| 311 | if (src_x != gpu.screen.src_x || src_y != gpu.screen.src_y) { |
| 312 | gpu.screen.src_x = src_x; |
| 313 | gpu.screen.src_y = src_y; |
| 314 | renderer_notify_scanout_change(src_x, src_y); |
| 315 | if (gpu.frameskip.set) { |
| 316 | decide_frameskip_allow(gpu.ex_regs[3]); |
| 317 | if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) { |
| 318 | decide_frameskip(); |
| 319 | gpu.frameskip.last_flip_frame = *gpu.state.frame_count; |
| 320 | } |
| 321 | } |
| 322 | } |
| 323 | break; |
| 324 | case 0x06: |
| 325 | gpu.screen.x1 = data & 0xfff; |
| 326 | gpu.screen.x2 = (data >> 12) & 0xfff; |
| 327 | update_width(); |
| 328 | break; |
| 329 | case 0x07: |
| 330 | gpu.screen.y1 = data & 0x3ff; |
| 331 | gpu.screen.y2 = (data >> 10) & 0x3ff; |
| 332 | update_height(); |
| 333 | break; |
| 334 | case 0x08: |
| 335 | gpu.status = (gpu.status & ~0x7f0000) | ((data & 0x3F) << 17) | ((data & 0x40) << 10); |
| 336 | update_width(); |
| 337 | update_height(); |
| 338 | renderer_notify_res_change(); |
| 339 | break; |
| 340 | default: |
| 341 | if ((cmd & 0xf0) == 0x10) |
| 342 | get_gpu_info(data); |
| 343 | break; |
| 344 | } |
| 345 | |
| 346 | #ifdef GPUwriteStatus_ext |
| 347 | GPUwriteStatus_ext(data); |
| 348 | #endif |
| 349 | } |
| 350 | |
| 351 | const unsigned char cmd_lengths[256] = |
| 352 | { |
| 353 | 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 354 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 355 | 3, 3, 3, 3, 6, 6, 6, 6, 4, 4, 4, 4, 8, 8, 8, 8, // 20 |
| 356 | 5, 5, 5, 5, 8, 8, 8, 8, 7, 7, 7, 7, 11, 11, 11, 11, |
| 357 | 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, // 40 |
| 358 | 3, 3, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, |
| 359 | 2, 2, 2, 2, 3, 3, 3, 3, 1, 1, 1, 1, 0, 0, 0, 0, // 60 |
| 360 | 1, 1, 1, 1, 2, 2, 2, 2, 1, 1, 1, 1, 2, 2, 2, 2, |
| 361 | 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, // 80 |
| 362 | 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, |
| 363 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, // a0 |
| 364 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 365 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, // c0 |
| 366 | 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, |
| 367 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, // e0 |
| 368 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 |
| 369 | }; |
| 370 | |
| 371 | #define VRAM_MEM_XY(x, y) &gpu.vram[(y) * 1024 + (x)] |
| 372 | |
| 373 | // this isn't very useful so should be rare |
| 374 | static void cpy_mask(uint16_t *dst, const uint16_t *src, int l, uint32_t r6) |
| 375 | { |
| 376 | int i; |
| 377 | if (r6 == 1) { |
| 378 | for (i = 0; i < l; i++) |
| 379 | dst[i] = src[i] | 0x8000; |
| 380 | } |
| 381 | else { |
| 382 | uint16_t msb = r6 << 15; |
| 383 | for (i = 0; i < l; i++) { |
| 384 | uint16_t mask = (int16_t)dst[i] >> 15; |
| 385 | dst[i] = (dst[i] & mask) | ((src[i] | msb) & ~mask); |
| 386 | } |
| 387 | } |
| 388 | } |
| 389 | |
| 390 | static inline void do_vram_line(int x, int y, uint16_t *mem, int l, |
| 391 | int is_read, uint32_t r6) |
| 392 | { |
| 393 | uint16_t *vram = VRAM_MEM_XY(x, y); |
| 394 | if (unlikely(is_read)) |
| 395 | memcpy(mem, vram, l * 2); |
| 396 | else if (unlikely(r6)) |
| 397 | cpy_mask(vram, mem, l, r6); |
| 398 | else |
| 399 | memcpy(vram, mem, l * 2); |
| 400 | } |
| 401 | |
| 402 | static int do_vram_io(uint32_t *data, int count, int is_read) |
| 403 | { |
| 404 | int count_initial = count; |
| 405 | uint32_t r6 = gpu.ex_regs[6] & 3; |
| 406 | uint16_t *sdata = (uint16_t *)data; |
| 407 | int x = gpu.dma.x, y = gpu.dma.y; |
| 408 | int w = gpu.dma.w, h = gpu.dma.h; |
| 409 | int o = gpu.dma.offset; |
| 410 | int l; |
| 411 | count *= 2; // operate in 16bpp pixels |
| 412 | |
| 413 | if (gpu.dma.offset) { |
| 414 | l = w - gpu.dma.offset; |
| 415 | if (count < l) |
| 416 | l = count; |
| 417 | |
| 418 | do_vram_line(x + o, y, sdata, l, is_read, r6); |
| 419 | |
| 420 | if (o + l < w) |
| 421 | o += l; |
| 422 | else { |
| 423 | o = 0; |
| 424 | y++; |
| 425 | h--; |
| 426 | } |
| 427 | sdata += l; |
| 428 | count -= l; |
| 429 | } |
| 430 | |
| 431 | for (; h > 0 && count >= w; sdata += w, count -= w, y++, h--) { |
| 432 | y &= 511; |
| 433 | do_vram_line(x, y, sdata, w, is_read, r6); |
| 434 | } |
| 435 | |
| 436 | if (h > 0) { |
| 437 | if (count > 0) { |
| 438 | y &= 511; |
| 439 | do_vram_line(x, y, sdata, count, is_read, r6); |
| 440 | o = count; |
| 441 | count = 0; |
| 442 | } |
| 443 | } |
| 444 | else |
| 445 | finish_vram_transfer(is_read); |
| 446 | gpu.dma.y = y; |
| 447 | gpu.dma.h = h; |
| 448 | gpu.dma.offset = o; |
| 449 | |
| 450 | return count_initial - count / 2; |
| 451 | } |
| 452 | |
| 453 | static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_read) |
| 454 | { |
| 455 | if (gpu.dma.h) |
| 456 | log_anomaly("start_vram_transfer while old unfinished\n"); |
| 457 | |
| 458 | gpu.dma.x = pos_word & 0x3ff; |
| 459 | gpu.dma.y = (pos_word >> 16) & 0x1ff; |
| 460 | gpu.dma.w = ((size_word - 1) & 0x3ff) + 1; |
| 461 | gpu.dma.h = (((size_word >> 16) - 1) & 0x1ff) + 1; |
| 462 | gpu.dma.offset = 0; |
| 463 | gpu.dma.is_read = is_read; |
| 464 | gpu.dma_start = gpu.dma; |
| 465 | |
| 466 | renderer_flush_queues(); |
| 467 | if (is_read) { |
| 468 | gpu.status |= PSX_GPU_STATUS_IMG; |
| 469 | // XXX: wrong for width 1 |
| 470 | gpu.gp0 = LE32TOH(*(uint32_t *) VRAM_MEM_XY(gpu.dma.x, gpu.dma.y)); |
| 471 | gpu.state.last_vram_read_frame = *gpu.state.frame_count; |
| 472 | } |
| 473 | |
| 474 | log_io("start_vram_transfer %c (%d, %d) %dx%d\n", is_read ? 'r' : 'w', |
| 475 | gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h); |
| 476 | if (gpu.gpu_state_change) |
| 477 | gpu.gpu_state_change(PGS_VRAM_TRANSFER_START); |
| 478 | } |
| 479 | |
| 480 | static void finish_vram_transfer(int is_read) |
| 481 | { |
| 482 | if (is_read) |
| 483 | gpu.status &= ~PSX_GPU_STATUS_IMG; |
| 484 | else { |
| 485 | gpu.state.fb_dirty = 1; |
| 486 | renderer_update_caches(gpu.dma_start.x, gpu.dma_start.y, |
| 487 | gpu.dma_start.w, gpu.dma_start.h, 0); |
| 488 | } |
| 489 | if (gpu.gpu_state_change) |
| 490 | gpu.gpu_state_change(PGS_VRAM_TRANSFER_END); |
| 491 | } |
| 492 | |
| 493 | static void do_vram_copy(const uint32_t *params, int *cpu_cycles) |
| 494 | { |
| 495 | const uint32_t sx = LE32TOH(params[0]) & 0x3FF; |
| 496 | const uint32_t sy = (LE32TOH(params[0]) >> 16) & 0x1FF; |
| 497 | const uint32_t dx = LE32TOH(params[1]) & 0x3FF; |
| 498 | const uint32_t dy = (LE32TOH(params[1]) >> 16) & 0x1FF; |
| 499 | uint32_t w = ((LE32TOH(params[2]) - 1) & 0x3FF) + 1; |
| 500 | uint32_t h = (((LE32TOH(params[2]) >> 16) - 1) & 0x1FF) + 1; |
| 501 | uint16_t msb = gpu.ex_regs[6] << 15; |
| 502 | uint16_t lbuf[128]; |
| 503 | uint32_t x, y; |
| 504 | |
| 505 | *cpu_cycles += gput_copy(w, h); |
| 506 | if (sx == dx && sy == dy && msb == 0) |
| 507 | return; |
| 508 | |
| 509 | renderer_flush_queues(); |
| 510 | |
| 511 | if (unlikely((sx < dx && dx < sx + w) || sx + w > 1024 || dx + w > 1024 || msb)) |
| 512 | { |
| 513 | for (y = 0; y < h; y++) |
| 514 | { |
| 515 | const uint16_t *src = VRAM_MEM_XY(0, (sy + y) & 0x1ff); |
| 516 | uint16_t *dst = VRAM_MEM_XY(0, (dy + y) & 0x1ff); |
| 517 | for (x = 0; x < w; x += ARRAY_SIZE(lbuf)) |
| 518 | { |
| 519 | uint32_t x1, w1 = w - x; |
| 520 | if (w1 > ARRAY_SIZE(lbuf)) |
| 521 | w1 = ARRAY_SIZE(lbuf); |
| 522 | for (x1 = 0; x1 < w1; x1++) |
| 523 | lbuf[x1] = src[(sx + x + x1) & 0x3ff]; |
| 524 | for (x1 = 0; x1 < w1; x1++) |
| 525 | dst[(dx + x + x1) & 0x3ff] = lbuf[x1] | msb; |
| 526 | } |
| 527 | } |
| 528 | } |
| 529 | else |
| 530 | { |
| 531 | uint32_t sy1 = sy, dy1 = dy; |
| 532 | for (y = 0; y < h; y++, sy1++, dy1++) |
| 533 | memcpy(VRAM_MEM_XY(dx, dy1 & 0x1ff), VRAM_MEM_XY(sx, sy1 & 0x1ff), w * 2); |
| 534 | } |
| 535 | |
| 536 | renderer_update_caches(dx, dy, w, h, 0); |
| 537 | } |
| 538 | |
| 539 | static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd) |
| 540 | { |
| 541 | int cmd = 0, pos = 0, len, dummy = 0, v; |
| 542 | int skip = 1; |
| 543 | |
| 544 | gpu.frameskip.pending_fill[0] = 0; |
| 545 | |
| 546 | while (pos < count && skip) { |
| 547 | uint32_t *list = data + pos; |
| 548 | cmd = LE32TOH(list[0]) >> 24; |
| 549 | len = 1 + cmd_lengths[cmd]; |
| 550 | if (pos + len > count) { |
| 551 | cmd = -1; |
| 552 | break; // incomplete cmd |
| 553 | } |
| 554 | |
| 555 | switch (cmd) { |
| 556 | case 0x02: |
| 557 | if ((LE32TOH(list[2]) & 0x3ff) > gpu.screen.w || ((LE32TOH(list[2]) >> 16) & 0x1ff) > gpu.screen.h) |
| 558 | // clearing something large, don't skip |
| 559 | do_cmd_list(list, 3, &dummy, &dummy, &dummy); |
| 560 | else |
| 561 | memcpy(gpu.frameskip.pending_fill, list, 3 * 4); |
| 562 | break; |
| 563 | case 0x24 ... 0x27: |
| 564 | case 0x2c ... 0x2f: |
| 565 | case 0x34 ... 0x37: |
| 566 | case 0x3c ... 0x3f: |
| 567 | gpu.ex_regs[1] &= ~0x1ff; |
| 568 | gpu.ex_regs[1] |= LE32TOH(list[4 + ((cmd >> 4) & 1)]) & 0x1ff; |
| 569 | break; |
| 570 | case 0x48 ... 0x4F: |
| 571 | for (v = 3; pos + v < count; v++) |
| 572 | { |
| 573 | if ((list[v] & HTOLE32(0xf000f000)) == HTOLE32(0x50005000)) |
| 574 | break; |
| 575 | } |
| 576 | len += v - 3; |
| 577 | break; |
| 578 | case 0x58 ... 0x5F: |
| 579 | for (v = 4; pos + v < count; v += 2) |
| 580 | { |
| 581 | if ((list[v] & HTOLE32(0xf000f000)) == HTOLE32(0x50005000)) |
| 582 | break; |
| 583 | } |
| 584 | len += v - 4; |
| 585 | break; |
| 586 | default: |
| 587 | if (cmd == 0xe3) |
| 588 | skip = decide_frameskip_allow(LE32TOH(list[0])); |
| 589 | if ((cmd & 0xf8) == 0xe0) |
| 590 | gpu.ex_regs[cmd & 7] = LE32TOH(list[0]); |
| 591 | break; |
| 592 | } |
| 593 | if (0x80 <= cmd && cmd <= 0xdf) |
| 594 | break; // image i/o |
| 595 | |
| 596 | pos += len; |
| 597 | } |
| 598 | |
| 599 | renderer_sync_ecmds(gpu.ex_regs); |
| 600 | *last_cmd = cmd; |
| 601 | return pos; |
| 602 | } |
| 603 | |
| 604 | static noinline int do_cmd_buffer(uint32_t *data, int count, |
| 605 | int *cycles_sum, int *cycles_last) |
| 606 | { |
| 607 | int cmd, pos; |
| 608 | uint32_t old_e3 = gpu.ex_regs[3]; |
| 609 | int vram_dirty = 0; |
| 610 | |
| 611 | // process buffer |
| 612 | for (pos = 0; pos < count; ) |
| 613 | { |
| 614 | if (gpu.dma.h && !gpu.dma_start.is_read) { // XXX: need to verify |
| 615 | vram_dirty = 1; |
| 616 | pos += do_vram_io(data + pos, count - pos, 0); |
| 617 | if (pos == count) |
| 618 | break; |
| 619 | } |
| 620 | |
| 621 | cmd = LE32TOH(data[pos]) >> 24; |
| 622 | if (0xa0 <= cmd && cmd <= 0xdf) { |
| 623 | if (unlikely((pos+2) >= count)) { |
| 624 | // incomplete vram write/read cmd, can't consume yet |
| 625 | cmd = -1; |
| 626 | break; |
| 627 | } |
| 628 | |
| 629 | // consume vram write/read cmd |
| 630 | start_vram_transfer(LE32TOH(data[pos + 1]), LE32TOH(data[pos + 2]), (cmd & 0xe0) == 0xc0); |
| 631 | pos += 3; |
| 632 | continue; |
| 633 | } |
| 634 | else if ((cmd & 0xe0) == 0x80) { |
| 635 | if (unlikely((pos+3) >= count)) { |
| 636 | cmd = -1; // incomplete cmd, can't consume yet |
| 637 | break; |
| 638 | } |
| 639 | *cycles_sum += *cycles_last; |
| 640 | *cycles_last = 0; |
| 641 | do_vram_copy(data + pos + 1, cycles_last); |
| 642 | vram_dirty = 1; |
| 643 | pos += 4; |
| 644 | continue; |
| 645 | } |
| 646 | else if (cmd == 0x1f) { |
| 647 | log_anomaly("irq1?\n"); |
| 648 | pos++; |
| 649 | continue; |
| 650 | } |
| 651 | |
| 652 | // 0xex cmds might affect frameskip.allow, so pass to do_cmd_list_skip |
| 653 | if (gpu.frameskip.active && (gpu.frameskip.allow || ((LE32TOH(data[pos]) >> 24) & 0xf0) == 0xe0)) |
| 654 | pos += do_cmd_list_skip(data + pos, count - pos, &cmd); |
| 655 | else { |
| 656 | pos += do_cmd_list(data + pos, count - pos, cycles_sum, cycles_last, &cmd); |
| 657 | vram_dirty = 1; |
| 658 | } |
| 659 | |
| 660 | if (cmd == -1) |
| 661 | // incomplete cmd |
| 662 | break; |
| 663 | } |
| 664 | |
| 665 | gpu.status &= ~0x1fff; |
| 666 | gpu.status |= gpu.ex_regs[1] & 0x7ff; |
| 667 | gpu.status |= (gpu.ex_regs[6] & 3) << 11; |
| 668 | |
| 669 | gpu.state.fb_dirty |= vram_dirty; |
| 670 | |
| 671 | if (old_e3 != gpu.ex_regs[3]) |
| 672 | decide_frameskip_allow(gpu.ex_regs[3]); |
| 673 | |
| 674 | return count - pos; |
| 675 | } |
| 676 | |
| 677 | static noinline void flush_cmd_buffer(void) |
| 678 | { |
| 679 | int dummy = 0, left; |
| 680 | left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy, &dummy); |
| 681 | if (left > 0) |
| 682 | memmove(gpu.cmd_buffer, gpu.cmd_buffer + gpu.cmd_len - left, left * 4); |
| 683 | if (left != gpu.cmd_len) { |
| 684 | if (!gpu.dma.h && gpu.gpu_state_change) |
| 685 | gpu.gpu_state_change(PGS_PRIMITIVE_START); |
| 686 | gpu.cmd_len = left; |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | void GPUwriteDataMem(uint32_t *mem, int count) |
| 691 | { |
| 692 | int dummy = 0, left; |
| 693 | |
| 694 | log_io("gpu_dma_write %p %d\n", mem, count); |
| 695 | |
| 696 | if (unlikely(gpu.cmd_len > 0)) |
| 697 | flush_cmd_buffer(); |
| 698 | |
| 699 | left = do_cmd_buffer(mem, count, &dummy, &dummy); |
| 700 | if (left) |
| 701 | log_anomaly("GPUwriteDataMem: discarded %d/%d words\n", left, count); |
| 702 | } |
| 703 | |
| 704 | void GPUwriteData(uint32_t data) |
| 705 | { |
| 706 | log_io("gpu_write %08x\n", data); |
| 707 | gpu.cmd_buffer[gpu.cmd_len++] = HTOLE32(data); |
| 708 | if (gpu.cmd_len >= CMD_BUFFER_LEN) |
| 709 | flush_cmd_buffer(); |
| 710 | } |
| 711 | |
| 712 | long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, |
| 713 | uint32_t *progress_addr, int32_t *cycles_last_cmd) |
| 714 | { |
| 715 | uint32_t addr, *list, ld_addr; |
| 716 | int len, left, count, ld_count = 32; |
| 717 | int cpu_cycles_sum = 0; |
| 718 | int cpu_cycles_last = 0; |
| 719 | |
| 720 | preload(rambase + (start_addr & 0x1fffff) / 4); |
| 721 | |
| 722 | if (unlikely(gpu.cmd_len > 0)) |
| 723 | flush_cmd_buffer(); |
| 724 | |
| 725 | log_io("gpu_dma_chain\n"); |
| 726 | addr = ld_addr = start_addr & 0xffffff; |
| 727 | for (count = 0; (addr & 0x800000) == 0; count++) |
| 728 | { |
| 729 | list = rambase + (addr & 0x1fffff) / 4; |
| 730 | len = LE32TOH(list[0]) >> 24; |
| 731 | addr = LE32TOH(list[0]) & 0xffffff; |
| 732 | preload(rambase + (addr & 0x1fffff) / 4); |
| 733 | |
| 734 | cpu_cycles_sum += 10; |
| 735 | if (len > 0) |
| 736 | cpu_cycles_sum += 5 + len; |
| 737 | |
| 738 | log_io(".chain %08lx #%d+%d %u+%u\n", |
| 739 | (long)(list - rambase) * 4, len, gpu.cmd_len, cpu_cycles_sum, cpu_cycles_last); |
| 740 | if (unlikely(gpu.cmd_len > 0)) { |
| 741 | if (gpu.cmd_len + len > ARRAY_SIZE(gpu.cmd_buffer)) { |
| 742 | log_anomaly("cmd_buffer overflow, likely garbage commands\n"); |
| 743 | gpu.cmd_len = 0; |
| 744 | } |
| 745 | memcpy(gpu.cmd_buffer + gpu.cmd_len, list + 1, len * 4); |
| 746 | gpu.cmd_len += len; |
| 747 | flush_cmd_buffer(); |
| 748 | continue; |
| 749 | } |
| 750 | |
| 751 | if (len) { |
| 752 | left = do_cmd_buffer(list + 1, len, &cpu_cycles_sum, &cpu_cycles_last); |
| 753 | if (left) { |
| 754 | memcpy(gpu.cmd_buffer, list + 1 + len - left, left * 4); |
| 755 | gpu.cmd_len = left; |
| 756 | log_anomaly("GPUdmaChain: %d/%d words left\n", left, len); |
| 757 | } |
| 758 | } |
| 759 | |
| 760 | if (progress_addr) { |
| 761 | *progress_addr = addr; |
| 762 | break; |
| 763 | } |
| 764 | if (addr == ld_addr) { |
| 765 | log_anomaly("GPUdmaChain: loop @ %08x, cnt=%u\n", addr, count); |
| 766 | break; |
| 767 | } |
| 768 | if (count == ld_count) { |
| 769 | ld_addr = addr; |
| 770 | ld_count *= 2; |
| 771 | } |
| 772 | } |
| 773 | |
| 774 | //printf(" -> %d %d\n", cpu_cycles_sum, cpu_cycles_last); |
| 775 | gpu.state.last_list.frame = *gpu.state.frame_count; |
| 776 | gpu.state.last_list.hcnt = *gpu.state.hcnt; |
| 777 | gpu.state.last_list.cycles = cpu_cycles_sum + cpu_cycles_last; |
| 778 | gpu.state.last_list.addr = start_addr; |
| 779 | |
| 780 | *cycles_last_cmd = cpu_cycles_last; |
| 781 | return cpu_cycles_sum; |
| 782 | } |
| 783 | |
| 784 | void GPUreadDataMem(uint32_t *mem, int count) |
| 785 | { |
| 786 | log_io("gpu_dma_read %p %d\n", mem, count); |
| 787 | |
| 788 | if (unlikely(gpu.cmd_len > 0)) |
| 789 | flush_cmd_buffer(); |
| 790 | |
| 791 | if (gpu.dma.h) |
| 792 | do_vram_io(mem, count, 1); |
| 793 | } |
| 794 | |
| 795 | uint32_t GPUreadData(void) |
| 796 | { |
| 797 | uint32_t ret; |
| 798 | |
| 799 | if (unlikely(gpu.cmd_len > 0)) |
| 800 | flush_cmd_buffer(); |
| 801 | |
| 802 | ret = gpu.gp0; |
| 803 | if (gpu.dma.h) { |
| 804 | ret = HTOLE32(ret); |
| 805 | do_vram_io(&ret, 1, 1); |
| 806 | ret = LE32TOH(ret); |
| 807 | } |
| 808 | |
| 809 | log_io("gpu_read %08x\n", ret); |
| 810 | return ret; |
| 811 | } |
| 812 | |
| 813 | uint32_t GPUreadStatus(void) |
| 814 | { |
| 815 | uint32_t ret; |
| 816 | |
| 817 | if (unlikely(gpu.cmd_len > 0)) |
| 818 | flush_cmd_buffer(); |
| 819 | |
| 820 | ret = gpu.status; |
| 821 | log_io("gpu_read_status %08x\n", ret); |
| 822 | return ret; |
| 823 | } |
| 824 | |
| 825 | struct GPUFreeze |
| 826 | { |
| 827 | uint32_t ulFreezeVersion; // should be always 1 for now (set by main emu) |
| 828 | uint32_t ulStatus; // current gpu status |
| 829 | uint32_t ulControl[256]; // latest control register values |
| 830 | unsigned char psxVRam[1024*1024*2]; // current VRam image (full 2 MB for ZN) |
| 831 | }; |
| 832 | |
| 833 | long GPUfreeze(uint32_t type, struct GPUFreeze *freeze) |
| 834 | { |
| 835 | int i; |
| 836 | |
| 837 | switch (type) { |
| 838 | case 1: // save |
| 839 | if (gpu.cmd_len > 0) |
| 840 | flush_cmd_buffer(); |
| 841 | memcpy(freeze->psxVRam, gpu.vram, 1024 * 512 * 2); |
| 842 | memcpy(freeze->ulControl, gpu.regs, sizeof(gpu.regs)); |
| 843 | memcpy(freeze->ulControl + 0xe0, gpu.ex_regs, sizeof(gpu.ex_regs)); |
| 844 | freeze->ulStatus = gpu.status; |
| 845 | break; |
| 846 | case 0: // load |
| 847 | memcpy(gpu.vram, freeze->psxVRam, 1024 * 512 * 2); |
| 848 | memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs)); |
| 849 | memcpy(gpu.ex_regs, freeze->ulControl + 0xe0, sizeof(gpu.ex_regs)); |
| 850 | gpu.status = freeze->ulStatus; |
| 851 | gpu.cmd_len = 0; |
| 852 | for (i = 8; i > 0; i--) { |
| 853 | gpu.regs[i] ^= 1; // avoid reg change detection |
| 854 | GPUwriteStatus((i << 24) | (gpu.regs[i] ^ 1)); |
| 855 | } |
| 856 | renderer_sync_ecmds(gpu.ex_regs); |
| 857 | renderer_update_caches(0, 0, 1024, 512, 0); |
| 858 | break; |
| 859 | } |
| 860 | |
| 861 | return 1; |
| 862 | } |
| 863 | |
| 864 | void GPUupdateLace(void) |
| 865 | { |
| 866 | if (gpu.cmd_len > 0) |
| 867 | flush_cmd_buffer(); |
| 868 | renderer_flush_queues(); |
| 869 | |
| 870 | #ifndef RAW_FB_DISPLAY |
| 871 | if (gpu.status & PSX_GPU_STATUS_BLANKING) { |
| 872 | if (!gpu.state.blanked) { |
| 873 | vout_blank(); |
| 874 | gpu.state.blanked = 1; |
| 875 | gpu.state.fb_dirty = 1; |
| 876 | } |
| 877 | return; |
| 878 | } |
| 879 | |
| 880 | if (!gpu.state.fb_dirty) |
| 881 | return; |
| 882 | #endif |
| 883 | |
| 884 | if (gpu.frameskip.set) { |
| 885 | if (!gpu.frameskip.frame_ready) { |
| 886 | if (*gpu.state.frame_count - gpu.frameskip.last_flip_frame < 9) |
| 887 | return; |
| 888 | gpu.frameskip.active = 0; |
| 889 | } |
| 890 | gpu.frameskip.frame_ready = 0; |
| 891 | } |
| 892 | |
| 893 | vout_update(); |
| 894 | if (gpu.state.enhancement_active && !gpu.state.enhancement_was_active) |
| 895 | renderer_update_caches(0, 0, 1024, 512, 1); |
| 896 | gpu.state.enhancement_was_active = gpu.state.enhancement_active; |
| 897 | gpu.state.fb_dirty = 0; |
| 898 | gpu.state.blanked = 0; |
| 899 | } |
| 900 | |
| 901 | void GPUvBlank(int is_vblank, int lcf) |
| 902 | { |
| 903 | int interlace = gpu.state.allow_interlace |
| 904 | && (gpu.status & PSX_GPU_STATUS_INTERLACE) |
| 905 | && (gpu.status & PSX_GPU_STATUS_DHEIGHT); |
| 906 | // interlace doesn't look nice on progressive displays, |
| 907 | // so we have this "auto" mode here for games that don't read vram |
| 908 | if (gpu.state.allow_interlace == 2 |
| 909 | && *gpu.state.frame_count - gpu.state.last_vram_read_frame > 1) |
| 910 | { |
| 911 | interlace = 0; |
| 912 | } |
| 913 | if (interlace || interlace != gpu.state.old_interlace) { |
| 914 | gpu.state.old_interlace = interlace; |
| 915 | |
| 916 | if (gpu.cmd_len > 0) |
| 917 | flush_cmd_buffer(); |
| 918 | renderer_flush_queues(); |
| 919 | renderer_set_interlace(interlace, !lcf); |
| 920 | } |
| 921 | } |
| 922 | |
| 923 | void GPUgetScreenInfo(int *y, int *base_hres) |
| 924 | { |
| 925 | *y = gpu.screen.y; |
| 926 | *base_hres = gpu.screen.vres; |
| 927 | if (gpu.status & PSX_GPU_STATUS_DHEIGHT) |
| 928 | *base_hres >>= 1; |
| 929 | } |
| 930 | |
| 931 | void GPUrearmedCallbacks(const struct rearmed_cbs *cbs) |
| 932 | { |
| 933 | gpu.frameskip.set = cbs->frameskip; |
| 934 | gpu.frameskip.advice = &cbs->fskip_advice; |
| 935 | gpu.frameskip.active = 0; |
| 936 | gpu.frameskip.frame_ready = 1; |
| 937 | gpu.state.hcnt = (uint32_t *)cbs->gpu_hcnt; |
| 938 | gpu.state.frame_count = (uint32_t *)cbs->gpu_frame_count; |
| 939 | gpu.state.allow_interlace = cbs->gpu_neon.allow_interlace; |
| 940 | gpu.state.enhancement_enable = cbs->gpu_neon.enhancement_enable; |
| 941 | gpu.state.screen_centering_type_default = cbs->screen_centering_type_default; |
| 942 | if (gpu.state.screen_centering_type != cbs->screen_centering_type |
| 943 | || gpu.state.screen_centering_x != cbs->screen_centering_x |
| 944 | || gpu.state.screen_centering_y != cbs->screen_centering_y |
| 945 | || gpu.state.show_overscan != cbs->show_overscan) { |
| 946 | gpu.state.screen_centering_type = cbs->screen_centering_type; |
| 947 | gpu.state.screen_centering_x = cbs->screen_centering_x; |
| 948 | gpu.state.screen_centering_y = cbs->screen_centering_y; |
| 949 | gpu.state.show_overscan = cbs->show_overscan; |
| 950 | update_width(); |
| 951 | update_height(); |
| 952 | } |
| 953 | |
| 954 | gpu.mmap = cbs->mmap; |
| 955 | gpu.munmap = cbs->munmap; |
| 956 | gpu.gpu_state_change = cbs->gpu_state_change; |
| 957 | |
| 958 | // delayed vram mmap |
| 959 | if (gpu.vram == NULL) |
| 960 | map_vram(); |
| 961 | |
| 962 | if (cbs->pl_vout_set_raw_vram) |
| 963 | cbs->pl_vout_set_raw_vram(gpu.vram); |
| 964 | renderer_set_config(cbs); |
| 965 | vout_set_config(cbs); |
| 966 | } |
| 967 | |
| 968 | // vim:shiftwidth=2:expandtab |