| 1 | // |
| 2 | // Object Processor |
| 3 | // |
| 4 | // Original source by David Raingeard (Cal2) |
| 5 | // GCC/SDL port by Niels Wagenaar (Linux/WIN32) and Caz (BeOS) |
| 6 | // Extensive cleanups/fixes/rewrites by James Hammons |
| 7 | // (C) 2010 Underground Software |
| 8 | // |
| 9 | // JLH = James Hammons <jlhamm@acm.org> |
| 10 | // JPM = Jean-Paul Mari <djipi.mari@gmail.com> |
| 11 | // |
| 12 | // Who When What |
| 13 | // --- ---------- ----------------------------------------------------------- |
| 14 | // JLH 01/16/2010 Created this log ;-) |
| 15 | // JPM 06/06/2016 Visual Studio support |
| 16 | // |
| 17 | |
| 18 | #include "op.h" |
| 19 | |
| 20 | #include <stdlib.h> |
| 21 | #include <string.h> |
| 22 | #include "gpu.h" |
| 23 | #include "jaguar.h" |
| 24 | #include "log.h" |
| 25 | #include "m68000/m68kinterface.h" |
| 26 | #include "memory.h" |
| 27 | #include "tom.h" |
| 28 | |
| 29 | //#define OP_DEBUG |
| 30 | //#define OP_DEBUG_BMP |
| 31 | |
| 32 | #define BLEND_Y(dst, src) op_blend_y[(((uint16_t)dst<<8)) | ((uint16_t)(src))] |
| 33 | #define BLEND_CR(dst, src) op_blend_cr[(((uint16_t)dst)<<8) | ((uint16_t)(src))] |
| 34 | |
| 35 | #define OBJECT_TYPE_BITMAP 0 // 000 |
| 36 | #define OBJECT_TYPE_SCALE 1 // 001 |
| 37 | #define OBJECT_TYPE_GPU 2 // 010 |
| 38 | #define OBJECT_TYPE_BRANCH 3 // 011 |
| 39 | #define OBJECT_TYPE_STOP 4 // 100 |
| 40 | |
| 41 | #define CONDITION_EQUAL 0 // VC == YPOS |
| 42 | #define CONDITION_LESS_THAN 1 // VC < YPOS |
| 43 | #define CONDITION_GREATER_THAN 2 // VC > YPOS |
| 44 | #define CONDITION_OP_FLAG_SET 3 |
| 45 | #define CONDITION_SECOND_HALF_LINE 4 |
| 46 | |
| 47 | #if 0 |
| 48 | #define OPFLAG_RELEASE 8 // Bus release bit |
| 49 | #define OPFLAG_TRANS 4 // Transparency bit |
| 50 | #define OPFLAG_RMW 2 // Read-Modify-Write bit |
| 51 | #define OPFLAG_REFLECT 1 // Horizontal mirror bit |
| 52 | #endif |
| 53 | |
| 54 | // Private function prototypes |
| 55 | |
| 56 | void OPProcessFixedBitmap(uint64_t p0, uint64_t p1, bool render); |
| 57 | void OPProcessScaledBitmap(uint64_t p0, uint64_t p1, uint64_t p2, bool render); |
| 58 | void OPDiscoverObjects(uint32_t address); |
| 59 | void OPDumpObjectList(void); |
| 60 | void DumpScaledObject(uint64_t p0, uint64_t p1, uint64_t p2); |
| 61 | void DumpFixedObject(uint64_t p0, uint64_t p1); |
| 62 | void DumpBitmapCore(uint64_t p0, uint64_t p1); |
| 63 | uint64_t OPLoadPhrase(uint32_t offset); |
| 64 | |
| 65 | // Local global variables |
| 66 | |
| 67 | // Blend tables (64K each) |
| 68 | static uint8_t op_blend_y[0x10000]; |
| 69 | static uint8_t op_blend_cr[0x10000]; |
| 70 | // There may be a problem with this "RAM" overlapping (and thus being independent of) |
| 71 | // some of the regular TOM RAM... |
| 72 | //#warning objectp_ram is separated from TOM RAM--need to fix that! |
| 73 | //static uint8_t objectp_ram[0x40]; // This is based at $F00000 |
| 74 | uint8_t objectp_running = 0; |
| 75 | //bool objectp_stop_reading_list; |
| 76 | |
| 77 | static uint8_t op_bitmap_bit_depth[8] = { 1, 2, 4, 8, 16, 24, 32, 0 }; |
| 78 | //static uint32_t op_bitmap_bit_size[8] = |
| 79 | // { (uint32_t)(0.125*65536), (uint32_t)(0.25*65536), (uint32_t)(0.5*65536), (uint32_t)(1*65536), |
| 80 | // (uint32_t)(2*65536), (uint32_t)(1*65536), (uint32_t)(1*65536), (uint32_t)(1*65536) }; |
| 81 | static uint32_t op_pointer; |
| 82 | |
| 83 | int32_t phraseWidthToPixels[8] = { 64, 32, 16, 8, 4, 2, 0, 0 }; |
| 84 | |
| 85 | |
| 86 | // |
| 87 | // Object Processor initialization |
| 88 | // |
| 89 | void OPInit(void) |
| 90 | { |
| 91 | // Here we calculate the saturating blend of a signed 4-bit value and an |
| 92 | // existing Cyan/Red value as well as a signed 8-bit value and an existing intensity... |
| 93 | // Note: CRY is 4 bits Cyan, 4 bits Red, 16 bits intensitY |
| 94 | for(int i=0; i<256*256; i++) |
| 95 | { |
| 96 | int y = (i >> 8) & 0xFF; |
| 97 | int dy = (int8_t)i; // Sign extend the Y index |
| 98 | int c1 = (i >> 8) & 0x0F; |
| 99 | int dc1 = (int8_t)(i << 4) >> 4; // Sign extend the R index |
| 100 | int c2 = (i >> 12) & 0x0F; |
| 101 | int dc2 = (int8_t)(i & 0xF0) >> 4; // Sign extend the C index |
| 102 | |
| 103 | y += dy; |
| 104 | |
| 105 | if (y < 0) |
| 106 | y = 0; |
| 107 | else if (y > 0xFF) |
| 108 | y = 0xFF; |
| 109 | |
| 110 | op_blend_y[i] = y; |
| 111 | |
| 112 | c1 += dc1; |
| 113 | |
| 114 | if (c1 < 0) |
| 115 | c1 = 0; |
| 116 | else if (c1 > 0x0F) |
| 117 | c1 = 0x0F; |
| 118 | |
| 119 | c2 += dc2; |
| 120 | |
| 121 | if (c2 < 0) |
| 122 | c2 = 0; |
| 123 | else if (c2 > 0x0F) |
| 124 | c2 = 0x0F; |
| 125 | |
| 126 | op_blend_cr[i] = (c2 << 4) | c1; |
| 127 | } |
| 128 | |
| 129 | OPReset(); |
| 130 | } |
| 131 | |
| 132 | |
| 133 | // |
| 134 | // Object Processor reset |
| 135 | // |
| 136 | void OPReset(void) |
| 137 | { |
| 138 | // memset(objectp_ram, 0x00, 0x40); |
| 139 | objectp_running = 0; |
| 140 | } |
| 141 | |
| 142 | |
| 143 | static const char * opType[8] = |
| 144 | { "(BITMAP)", "(SCALED BITMAP)", "(GPU INT)", "(BRANCH)", "(STOP)", "???", "???", "???" }; |
| 145 | static const char * ccType[8] = |
| 146 | { "==", "<", ">", "(opflag set)", "(second half line)", "?", "?", "?" }; |
| 147 | static uint32_t object[8192]; |
| 148 | static uint32_t numberOfObjects; |
| 149 | //static uint32_t objectLink[8192]; |
| 150 | //static uint32_t numberOfLinks; |
| 151 | |
| 152 | |
| 153 | void OPDone(void) |
| 154 | { |
| 155 | //#warning "!!! Fix OL dump so that it follows links !!!" |
| 156 | // const char * opType[8] = |
| 157 | // { "(BITMAP)", "(SCALED BITMAP)", "(GPU INT)", "(BRANCH)", "(STOP)", "???", "???", "???" }; |
| 158 | // const char * ccType[8] = |
| 159 | // { "\"==\"", "\"<\"", "\">\"", "(opflag set)", "(second half line)", "?", "?", "?" }; |
| 160 | |
| 161 | uint32_t olp = OPGetListPointer(); |
| 162 | WriteLog("\nOP: OLP = $%08X\n", olp); |
| 163 | WriteLog("OP: Phrase dump\n ----------\n"); |
| 164 | |
| 165 | #if 0 |
| 166 | for(uint32_t i=0; i<0x100; i+=8) |
| 167 | { |
| 168 | uint32_t hi = JaguarReadLong(olp + i, OP), lo = JaguarReadLong(olp + i + 4, OP); |
| 169 | WriteLog("\t%08X: %08X %08X %s", olp + i, hi, lo, opType[lo & 0x07]); |
| 170 | |
| 171 | if ((lo & 0x07) == 3) |
| 172 | { |
| 173 | uint16_t ypos = (lo >> 3) & 0x7FF; |
| 174 | uint8_t cc = (lo >> 14) & 0x03; |
| 175 | uint32_t link = ((hi << 11) | (lo >> 21)) & 0x3FFFF8; |
| 176 | WriteLog(" YPOS=%u, CC=%s, link=%08X", ypos, ccType[cc], link); |
| 177 | } |
| 178 | |
| 179 | WriteLog("\n"); |
| 180 | |
| 181 | if ((lo & 0x07) == 0) |
| 182 | DumpFixedObject(OPLoadPhrase(olp+i), OPLoadPhrase(olp+i+8)); |
| 183 | |
| 184 | if ((lo & 0x07) == 1) |
| 185 | DumpScaledObject(OPLoadPhrase(olp+i), OPLoadPhrase(olp+i+8), OPLoadPhrase(olp+i+16)); |
| 186 | } |
| 187 | |
| 188 | WriteLog("\n"); |
| 189 | #else |
| 190 | //#warning "!!! Fix lockup in OPDiscoverObjects() !!!" |
| 191 | //temp, to keep the following function from locking up on bad/weird OLs |
| 192 | //return; |
| 193 | |
| 194 | numberOfObjects = 0; |
| 195 | OPDiscoverObjects(olp); |
| 196 | OPDumpObjectList(); |
| 197 | #endif |
| 198 | } |
| 199 | |
| 200 | |
| 201 | bool OPObjectExists(uint32_t address) |
| 202 | { |
| 203 | // Yes, we really do a linear search, every time. :-/ |
| 204 | for(uint32_t i=0; i<numberOfObjects; i++) |
| 205 | { |
| 206 | if (address == object[i]) |
| 207 | return true; |
| 208 | } |
| 209 | |
| 210 | return false; |
| 211 | } |
| 212 | |
| 213 | |
| 214 | void OPDiscoverObjects(uint32_t address) |
| 215 | { |
| 216 | uint8_t objectType = 0; |
| 217 | |
| 218 | do |
| 219 | { |
| 220 | // If we've seen this object already, bail out! |
| 221 | // Otherwise, add it to the list |
| 222 | if (OPObjectExists(address)) |
| 223 | return; |
| 224 | |
| 225 | object[numberOfObjects++] = address; |
| 226 | |
| 227 | // Get the object & decode its type, link address |
| 228 | uint32_t hi = JaguarReadLong(address + 0, OP); |
| 229 | uint32_t lo = JaguarReadLong(address + 4, OP); |
| 230 | objectType = lo & 0x07; |
| 231 | uint32_t link = ((hi << 11) | (lo >> 21)) & 0x3FFFF8; |
| 232 | |
| 233 | if (objectType == 3) |
| 234 | { |
| 235 | // Branch if YPOS < 2047 (or YPOS > 0) can be treated as a GOTO, so |
| 236 | // don't do any discovery in that case. Otherwise, have at it: |
| 237 | if (((lo & 0xFFFF) != 0x7FFB) && ((lo & 0xFFFF) != 0x8003)) |
| 238 | // Recursion needed to follow all links! This does depth-first |
| 239 | // recursion on the not-taken objects |
| 240 | OPDiscoverObjects(address + 8); |
| 241 | } |
| 242 | |
| 243 | // Get the next object... |
| 244 | address = link; |
| 245 | } |
| 246 | while (objectType != 4); |
| 247 | } |
| 248 | |
| 249 | |
| 250 | void OPDumpObjectList(void) |
| 251 | { |
| 252 | for(uint32_t i=0; i<numberOfObjects; i++) |
| 253 | { |
| 254 | uint32_t address = object[i]; |
| 255 | |
| 256 | uint32_t hi = JaguarReadLong(address + 0, OP); |
| 257 | uint32_t lo = JaguarReadLong(address + 4, OP); |
| 258 | uint8_t objectType = lo & 0x07; |
| 259 | uint32_t link = ((hi << 11) | (lo >> 21)) & 0x3FFFF8; |
| 260 | WriteLog("%08X: %08X %08X %s -> $%08X", address, hi, lo, opType[objectType], link); |
| 261 | |
| 262 | if (objectType == 3) |
| 263 | { |
| 264 | uint16_t ypos = (lo >> 3) & 0x7FF; |
| 265 | uint8_t cc = (lo >> 14) & 0x07; // Proper # of bits == 3 |
| 266 | WriteLog(" YPOS %s %u", ccType[cc], ypos); |
| 267 | } |
| 268 | |
| 269 | WriteLog("\n"); |
| 270 | |
| 271 | // Yes, this is how the OP finds follow-on phrases for bitmap/scaled |
| 272 | // bitmap objects...! |
| 273 | if (objectType == 0) |
| 274 | DumpFixedObject(OPLoadPhrase(address + 0), |
| 275 | OPLoadPhrase(address | 0x08)); |
| 276 | |
| 277 | if (objectType == 1) |
| 278 | DumpScaledObject(OPLoadPhrase(address + 0), |
| 279 | OPLoadPhrase(address | 0x08), OPLoadPhrase(address | 0x10)); |
| 280 | |
| 281 | if (address == link) // Ruh roh... |
| 282 | { |
| 283 | // Runaway recursive link is bad! |
| 284 | WriteLog("***** SELF REFERENTIAL LINK *****\n\n"); |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | WriteLog("\n"); |
| 289 | } |
| 290 | |
| 291 | |
| 292 | // |
| 293 | // Object Processor memory access |
| 294 | // Memory range: F00010 - F00027 |
| 295 | // |
| 296 | // F00010-F00017 R xxxxxxxx xxxxxxxx OB - current object code from the graphics processor |
| 297 | // F00020-F00023 W xxxxxxxx xxxxxxxx OLP - start of the object list |
| 298 | // F00026 W -------- -------x OBF - object processor flag |
| 299 | // |
| 300 | |
| 301 | #if 0 |
| 302 | uint8_t OPReadByte(uint32_t offset, uint32_t who/*=UNKNOWN*/) |
| 303 | { |
| 304 | offset &= 0x3F; |
| 305 | return objectp_ram[offset]; |
| 306 | } |
| 307 | |
| 308 | uint16_t OPReadWord(uint32_t offset, uint32_t who/*=UNKNOWN*/) |
| 309 | { |
| 310 | offset &= 0x3F; |
| 311 | return GET16(objectp_ram, offset); |
| 312 | } |
| 313 | |
| 314 | void OPWriteByte(uint32_t offset, uint8_t data, uint32_t who/*=UNKNOWN*/) |
| 315 | { |
| 316 | offset &= 0x3F; |
| 317 | objectp_ram[offset] = data; |
| 318 | } |
| 319 | |
| 320 | void OPWriteWord(uint32_t offset, uint16_t data, uint32_t who/*=UNKNOWN*/) |
| 321 | { |
| 322 | offset &= 0x3F; |
| 323 | SET16(objectp_ram, offset, data); |
| 324 | |
| 325 | /*if (offset == 0x20) |
| 326 | WriteLog("OP: Setting lo list pointer: %04X\n", data); |
| 327 | if (offset == 0x22) |
| 328 | WriteLog("OP: Setting hi list pointer: %04X\n", data);//*/ |
| 329 | } |
| 330 | #endif |
| 331 | |
| 332 | |
| 333 | uint32_t OPGetListPointer(void) |
| 334 | { |
| 335 | // Note: This register is LO / HI WORD, hence the funky look of this... |
| 336 | return GET16(tomRam8, 0x20) | (GET16(tomRam8, 0x22) << 16); |
| 337 | } |
| 338 | |
| 339 | |
| 340 | // This is WRONG, since the OBF is only 16 bits wide!!! [FIXED] |
| 341 | |
| 342 | uint32_t OPGetStatusRegister(void) |
| 343 | { |
| 344 | return GET16(tomRam8, 0x26); |
| 345 | } |
| 346 | |
| 347 | |
| 348 | // This is WRONG, since the OBF is only 16 bits wide!!! [FIXED] |
| 349 | |
| 350 | void OPSetStatusRegister(uint32_t data) |
| 351 | { |
| 352 | tomRam8[0x26] = (data & 0x0000FF00) >> 8; |
| 353 | tomRam8[0x27] |= (data & 0xFE); |
| 354 | } |
| 355 | |
| 356 | |
| 357 | void OPSetCurrentObject(uint64_t object) |
| 358 | { |
| 359 | //Not sure this is right... Wouldn't it just be stored 64 bit BE? |
| 360 | // Stored as least significant 32 bits first, ms32 last in big endian |
| 361 | /* objectp_ram[0x13] = object & 0xFF; object >>= 8; |
| 362 | objectp_ram[0x12] = object & 0xFF; object >>= 8; |
| 363 | objectp_ram[0x11] = object & 0xFF; object >>= 8; |
| 364 | objectp_ram[0x10] = object & 0xFF; object >>= 8; |
| 365 | |
| 366 | objectp_ram[0x17] = object & 0xFF; object >>= 8; |
| 367 | objectp_ram[0x16] = object & 0xFF; object >>= 8; |
| 368 | objectp_ram[0x15] = object & 0xFF; object >>= 8; |
| 369 | objectp_ram[0x14] = object & 0xFF;*/ |
| 370 | // Let's try regular good old big endian... |
| 371 | tomRam8[0x17] = object & 0xFF; object >>= 8; |
| 372 | tomRam8[0x16] = object & 0xFF; object >>= 8; |
| 373 | tomRam8[0x15] = object & 0xFF; object >>= 8; |
| 374 | tomRam8[0x14] = object & 0xFF; object >>= 8; |
| 375 | |
| 376 | tomRam8[0x13] = object & 0xFF; object >>= 8; |
| 377 | tomRam8[0x12] = object & 0xFF; object >>= 8; |
| 378 | tomRam8[0x11] = object & 0xFF; object >>= 8; |
| 379 | tomRam8[0x10] = object & 0xFF; |
| 380 | } |
| 381 | |
| 382 | |
| 383 | uint64_t OPLoadPhrase(uint32_t offset) |
| 384 | { |
| 385 | offset &= ~0x07; // 8 byte alignment |
| 386 | return ((uint64_t)JaguarReadLong(offset, OP) << 32) | (uint64_t)JaguarReadLong(offset+4, OP); |
| 387 | } |
| 388 | |
| 389 | |
| 390 | void OPStorePhrase(uint32_t offset, uint64_t p) |
| 391 | { |
| 392 | offset &= ~0x07; // 8 byte alignment |
| 393 | JaguarWriteLong(offset, p >> 32, OP); |
| 394 | JaguarWriteLong(offset + 4, p & 0xFFFFFFFF, OP); |
| 395 | } |
| 396 | |
| 397 | |
| 398 | // |
| 399 | // Debugging routines |
| 400 | // |
| 401 | void DumpScaledObject(uint64_t p0, uint64_t p1, uint64_t p2) |
| 402 | { |
| 403 | WriteLog(" %08X %08X\n", (uint32_t)(p1>>32), (uint32_t)(p1&0xFFFFFFFF)); |
| 404 | WriteLog(" %08X %08X\n", (uint32_t)(p2>>32), (uint32_t)(p2&0xFFFFFFFF)); |
| 405 | DumpBitmapCore(p0, p1); |
| 406 | uint32_t hscale = p2 & 0xFF; |
| 407 | uint32_t vscale = (p2 >> 8) & 0xFF; |
| 408 | uint32_t remainder = (p2 >> 16) & 0xFF; |
| 409 | WriteLog(" [hsc: %02X, vsc: %02X, rem: %02X]\n", hscale, vscale, remainder); |
| 410 | } |
| 411 | |
| 412 | |
| 413 | void DumpFixedObject(uint64_t p0, uint64_t p1) |
| 414 | { |
| 415 | WriteLog(" %08X %08X\n", (uint32_t)(p1>>32), (uint32_t)(p1&0xFFFFFFFF)); |
| 416 | DumpBitmapCore(p0, p1); |
| 417 | } |
| 418 | |
| 419 | |
| 420 | void DumpBitmapCore(uint64_t p0, uint64_t p1) |
| 421 | { |
| 422 | uint32_t bdMultiplier[8] = { 64, 32, 16, 8, 4, 2, 1, 1 }; |
| 423 | uint8_t bitdepth = (p1 >> 12) & 0x07; |
| 424 | //WAS: int16_t ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)? |
| 425 | int16_t ypos = ((p0 >> 3) & 0x7FF); // ??? What if not interlaced (/2)? |
| 426 | int32_t xpos = p1 & 0xFFF; |
| 427 | xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos); // Sign extend that mutha! |
| 428 | uint32_t iwidth = ((p1 >> 28) & 0x3FF); |
| 429 | uint32_t dwidth = ((p1 >> 18) & 0x3FF); // Unsigned! |
| 430 | uint16_t height = ((p0 >> 14) & 0x3FF); |
| 431 | uint32_t link = ((p0 >> 24) & 0x7FFFF) << 3; |
| 432 | uint32_t ptr = ((p0 >> 43) & 0x1FFFFF) << 3; |
| 433 | uint32_t firstPix = (p1 >> 49) & 0x3F; |
| 434 | uint8_t flags = (p1 >> 45) & 0x0F; |
| 435 | uint8_t idx = (p1 >> 38) & 0x7F; |
| 436 | uint32_t pitch = (p1 >> 15) & 0x07; |
| 437 | WriteLog(" [%u x %u @ (%i, %u) (iw:%u, dw:%u) (%u bpp), p:%08X fp:%02X, fl:%s%s%s%s, idx:%02X, pt:%02X]\n", |
| 438 | iwidth * bdMultiplier[bitdepth], |
| 439 | height, xpos, ypos, iwidth, dwidth, op_bitmap_bit_depth[bitdepth], |
| 440 | ptr, firstPix, (flags&OPFLAG_REFLECT ? "REFLECT " : ""), |
| 441 | (flags&OPFLAG_RMW ? "RMW " : ""), (flags&OPFLAG_TRANS ? "TRANS " : ""), |
| 442 | (flags&OPFLAG_RELEASE ? "RELEASE" : ""), idx, pitch); |
| 443 | } |
| 444 | |
| 445 | |
| 446 | // |
| 447 | // Object Processor main routine |
| 448 | // |
| 449 | #ifdef _MSC_VER |
| 450 | #pragma message("Warning: Need to fix this so that when an GPU object IRQ happens, we can pick up OP processing where we left off. !!! FIX !!!") |
| 451 | #else |
| 452 | #warning "Need to fix this so that when an GPU object IRQ happens, we can pick up OP processing where we left off. !!! FIX !!!" |
| 453 | #endif // _MSC_VER |
| 454 | void OPProcessList(int halfline, bool render) |
| 455 | { |
| 456 | #ifdef _MSC_VER |
| 457 | #pragma message("Warning: !!! NEED TO HANDLE MULTIPLE FIELDS PROPERLY !!!") |
| 458 | #else |
| 459 | #warning "!!! NEED TO HANDLE MULTIPLE FIELDS PROPERLY !!!" |
| 460 | #endif // _MSC_VER |
| 461 | // We ignore them, for now; not good D-: |
| 462 | // N.B.: Half-lines are exactly that, half-lines. When in interlaced mode, it |
| 463 | // draws the screen exactly the same way as it does in non, one line at a |
| 464 | // time. The only way you know you're in field #2 is that the topmost bit |
| 465 | // of VC is set. Half-line mode is so you can draw higher horizontal |
| 466 | // resolutions than you normally could, as the line buffer is only 720 |
| 467 | // pixels wide... |
| 468 | halfline &= 0x7FF; |
| 469 | |
| 470 | extern int op_start_log; |
| 471 | |
| 472 | op_pointer = OPGetListPointer(); |
| 473 | |
| 474 | // objectp_stop_reading_list = false; |
| 475 | |
| 476 | //WriteLog("OP: Processing line #%u (OLP=%08X)...\n", halfline, op_pointer); |
| 477 | //op_done(); |
| 478 | |
| 479 | // *** BEGIN OP PROCESSOR TESTING ONLY *** |
| 480 | extern bool interactiveMode; |
| 481 | extern bool iToggle; |
| 482 | extern int objectPtr; |
| 483 | bool inhibit; |
| 484 | int bitmapCounter = 0; |
| 485 | // *** END OP PROCESSOR TESTING ONLY *** |
| 486 | |
| 487 | uint32_t opCyclesToRun = 30000; // This is a pulled-out-of-the-air value (will need to be fixed, obviously!) |
| 488 | |
| 489 | // if (op_pointer) WriteLog(" new op list at 0x%.8x halfline %i\n",op_pointer,halfline); |
| 490 | while (op_pointer) |
| 491 | { |
| 492 | // *** BEGIN OP PROCESSOR TESTING ONLY *** |
| 493 | if (interactiveMode && bitmapCounter == objectPtr) |
| 494 | inhibit = iToggle; |
| 495 | else |
| 496 | inhibit = false; |
| 497 | // *** END OP PROCESSOR TESTING ONLY *** |
| 498 | // if (objectp_stop_reading_list) |
| 499 | // return; |
| 500 | |
| 501 | uint64_t p0 = OPLoadPhrase(op_pointer); |
| 502 | op_pointer += 8; |
| 503 | //WriteLog("\t%08X type %i\n", op_pointer, (uint8_t)p0 & 0x07); |
| 504 | |
| 505 | #if 1 |
| 506 | if (halfline == TOMGetVDB() && op_start_log) |
| 507 | //if (halfline == 215 && op_start_log) |
| 508 | //if (halfline == 28 && op_start_log) |
| 509 | //if (halfline == 0) |
| 510 | { |
| 511 | WriteLog("%08X --> phrase %08X %08X", op_pointer - 8, (int)(p0>>32), (int)(p0&0xFFFFFFFF)); |
| 512 | if ((p0 & 0x07) == OBJECT_TYPE_BITMAP) |
| 513 | { |
| 514 | WriteLog(" (BITMAP) "); |
| 515 | uint64_t p1 = OPLoadPhrase(op_pointer); |
| 516 | WriteLog("\n%08X --> phrase %08X %08X ", op_pointer, (int)(p1>>32), (int)(p1&0xFFFFFFFF)); |
| 517 | uint8_t bitdepth = (p1 >> 12) & 0x07; |
| 518 | //WAS: int16_t ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)? |
| 519 | int16_t ypos = ((p0 >> 3) & 0x7FF); // ??? What if not interlaced (/2)? |
| 520 | int32_t xpos = p1 & 0xFFF; |
| 521 | xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos); |
| 522 | uint32_t iwidth = ((p1 >> 28) & 0x3FF); |
| 523 | uint32_t dwidth = ((p1 >> 18) & 0x3FF); // Unsigned! |
| 524 | uint16_t height = ((p0 >> 14) & 0x3FF); |
| 525 | uint32_t link = ((p0 >> 24) & 0x7FFFF) << 3; |
| 526 | uint32_t ptr = ((p0 >> 43) & 0x1FFFFF) << 3; |
| 527 | uint32_t firstPix = (p1 >> 49) & 0x3F; |
| 528 | uint8_t flags = (p1 >> 45) & 0x0F; |
| 529 | uint8_t idx = (p1 >> 38) & 0x7F; |
| 530 | uint32_t pitch = (p1 >> 15) & 0x07; |
| 531 | WriteLog("\n [%u (%u) x %u @ (%i, %u) (%u bpp), l: %08X, p: %08X fp: %02X, fl:%s%s%s%s, idx:%02X, pt:%02X]\n", |
| 532 | iwidth, dwidth, height, xpos, ypos, op_bitmap_bit_depth[bitdepth], link, ptr, firstPix, (flags&OPFLAG_REFLECT ? "REFLECT " : ""), (flags&OPFLAG_RMW ? "RMW " : ""), (flags&OPFLAG_TRANS ? "TRANS " : ""), (flags&OPFLAG_RELEASE ? "RELEASE" : ""), idx, pitch); |
| 533 | } |
| 534 | if ((p0 & 0x07) == OBJECT_TYPE_SCALE) |
| 535 | { |
| 536 | WriteLog(" (SCALED BITMAP)"); |
| 537 | uint64_t p1 = OPLoadPhrase(op_pointer), p2 = OPLoadPhrase(op_pointer+8); |
| 538 | WriteLog("\n%08X --> phrase %08X %08X ", op_pointer, (int)(p1>>32), (int)(p1&0xFFFFFFFF)); |
| 539 | WriteLog("\n%08X --> phrase %08X %08X ", op_pointer+8, (int)(p2>>32), (int)(p2&0xFFFFFFFF)); |
| 540 | uint8_t bitdepth = (p1 >> 12) & 0x07; |
| 541 | //WAS: int16_t ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)? |
| 542 | int16_t ypos = ((p0 >> 3) & 0x7FF); // ??? What if not interlaced (/2)? |
| 543 | int32_t xpos = p1 & 0xFFF; |
| 544 | xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos); |
| 545 | uint32_t iwidth = ((p1 >> 28) & 0x3FF); |
| 546 | uint32_t dwidth = ((p1 >> 18) & 0x3FF); // Unsigned! |
| 547 | uint16_t height = ((p0 >> 14) & 0x3FF); |
| 548 | uint32_t link = ((p0 >> 24) & 0x7FFFF) << 3; |
| 549 | uint32_t ptr = ((p0 >> 43) & 0x1FFFFF) << 3; |
| 550 | uint32_t firstPix = (p1 >> 49) & 0x3F; |
| 551 | uint8_t flags = (p1 >> 45) & 0x0F; |
| 552 | uint8_t idx = (p1 >> 38) & 0x7F; |
| 553 | uint32_t pitch = (p1 >> 15) & 0x07; |
| 554 | WriteLog("\n [%u (%u) x %u @ (%i, %u) (%u bpp), l: %08X, p: %08X fp: %02X, fl:%s%s%s%s, idx:%02X, pt:%02X]\n", |
| 555 | iwidth, dwidth, height, xpos, ypos, op_bitmap_bit_depth[bitdepth], link, ptr, firstPix, (flags&OPFLAG_REFLECT ? "REFLECT " : ""), (flags&OPFLAG_RMW ? "RMW " : ""), (flags&OPFLAG_TRANS ? "TRANS " : ""), (flags&OPFLAG_RELEASE ? "RELEASE" : ""), idx, pitch); |
| 556 | uint32_t hscale = p2 & 0xFF; |
| 557 | uint32_t vscale = (p2 >> 8) & 0xFF; |
| 558 | uint32_t remainder = (p2 >> 16) & 0xFF; |
| 559 | WriteLog(" [hsc: %02X, vsc: %02X, rem: %02X]\n", hscale, vscale, remainder); |
| 560 | } |
| 561 | if ((p0 & 0x07) == OBJECT_TYPE_GPU) |
| 562 | WriteLog(" (GPU)\n"); |
| 563 | if ((p0 & 0x07) == OBJECT_TYPE_BRANCH) |
| 564 | { |
| 565 | WriteLog(" (BRANCH)\n"); |
| 566 | uint8_t * jaguarMainRam = GetRamPtr(); |
| 567 | WriteLog("[RAM] --> "); |
| 568 | for(int k=0; k<8; k++) |
| 569 | WriteLog("%02X ", jaguarMainRam[op_pointer-8 + k]); |
| 570 | WriteLog("\n"); |
| 571 | } |
| 572 | if ((p0 & 0x07) == OBJECT_TYPE_STOP) |
| 573 | WriteLog(" --> List end\n\n"); |
| 574 | } |
| 575 | #endif |
| 576 | |
| 577 | switch ((uint8_t)p0 & 0x07) |
| 578 | { |
| 579 | case OBJECT_TYPE_BITMAP: |
| 580 | { |
| 581 | uint16_t ypos = (p0 >> 3) & 0x7FF; |
| 582 | // This is only theory implied by Rayman...! |
| 583 | // It seems that if the YPOS is zero, then bump the YPOS value so that it |
| 584 | // coincides with the VDB value. With interlacing, this would be slightly more |
| 585 | // tricky. There's probably another bit somewhere that enables this mode--but |
| 586 | // so far, doesn't seem to affect any other game in a negative way (that I've |
| 587 | // seen). Either that, or it's an undocumented bug... |
| 588 | |
| 589 | //No, the reason this was needed is that the OP code before was wrong. Any value |
| 590 | //less than VDB will get written to the top line of the display! |
| 591 | #if 0 |
| 592 | // Not so sure... Let's see what happens here... |
| 593 | // No change... |
| 594 | if (ypos == 0) |
| 595 | ypos = TOMReadWord(0xF00046, OP) / 2; // Get the VDB value |
| 596 | #endif |
| 597 | // Actually, no. Any item less than VDB will get only the lines that hang over |
| 598 | // VDB displayed. Actually, this is incorrect. It seems that VDB value is wrong |
| 599 | // somewhere and that's what's causing things to fuck up. Still no idea why. |
| 600 | |
| 601 | uint32_t height = (p0 & 0xFFC000) >> 14; |
| 602 | uint32_t oldOPP = op_pointer - 8; |
| 603 | // *** BEGIN OP PROCESSOR TESTING ONLY *** |
| 604 | if (inhibit && op_start_log) |
| 605 | WriteLog("!!! ^^^ This object is INHIBITED! ^^^ !!!\n"); |
| 606 | bitmapCounter++; |
| 607 | if (!inhibit) // For OP testing only! |
| 608 | // *** END OP PROCESSOR TESTING ONLY *** |
| 609 | if (halfline >= ypos && height > 0) |
| 610 | { |
| 611 | // Believe it or not, this is what the OP actually does... |
| 612 | // which is why they're required to be on a dphrase boundary! |
| 613 | uint64_t p1 = OPLoadPhrase(oldOPP | 0x08); |
| 614 | //unneeded op_pointer += 8; |
| 615 | //WriteLog("OP: Writing halfline %d with ypos == %d...\n", halfline, ypos); |
| 616 | //WriteLog("--> Writing %u BPP bitmap...\n", op_bitmap_bit_depth[(p1 >> 12) & 0x07]); |
| 617 | // OPProcessFixedBitmap(halfline, p0, p1, render); |
| 618 | OPProcessFixedBitmap(p0, p1, render); |
| 619 | |
| 620 | // OP write-backs |
| 621 | |
| 622 | height--; |
| 623 | |
| 624 | uint64_t data = (p0 & 0xFFFFF80000000000LL) >> 40; |
| 625 | uint64_t dwidth = (p1 & 0xFFC0000) >> 15; |
| 626 | data += dwidth; |
| 627 | |
| 628 | p0 &= ~0xFFFFF80000FFC000LL; // Mask out old data... |
| 629 | p0 |= (uint64_t)height << 14; |
| 630 | p0 |= data << 40; |
| 631 | OPStorePhrase(oldOPP, p0); |
| 632 | } |
| 633 | |
| 634 | // OP bottom 3 bits are hardwired to zero. The link address |
| 635 | // reflects this, so we only need the top 19 bits of the address |
| 636 | // (which is why we only shift 21, and not 24). |
| 637 | op_pointer = (p0 & 0x000007FFFF000000LL) >> 21; |
| 638 | |
| 639 | // KLUDGE: Seems that memory access is mirrored in the first 8MB of |
| 640 | // memory... |
| 641 | if (op_pointer > 0x1FFFFF && op_pointer < 0x800000) |
| 642 | op_pointer &= 0xFF1FFFFF; // Knock out bits 21-23 |
| 643 | |
| 644 | break; |
| 645 | } |
| 646 | case OBJECT_TYPE_SCALE: |
| 647 | { |
| 648 | //WAS: uint16_t ypos = (p0 >> 3) & 0x3FF; |
| 649 | uint16_t ypos = (p0 >> 3) & 0x7FF; |
| 650 | uint32_t height = (p0 & 0xFFC000) >> 14; |
| 651 | uint32_t oldOPP = op_pointer - 8; |
| 652 | //WriteLog("OP: Scaled Object (ypos=%04X, height=%04X", ypos, height); |
| 653 | // *** BEGIN OP PROCESSOR TESTING ONLY *** |
| 654 | if (inhibit && op_start_log) |
| 655 | { |
| 656 | WriteLog("!!! ^^^ This object is INHIBITED! ^^^ !!! (halfline=%u, ypos=%u, height=%u)\n", halfline, ypos, height); |
| 657 | DumpScaledObject(p0, OPLoadPhrase(op_pointer), OPLoadPhrase(op_pointer+8)); |
| 658 | } |
| 659 | bitmapCounter++; |
| 660 | if (!inhibit) // For OP testing only! |
| 661 | // *** END OP PROCESSOR TESTING ONLY *** |
| 662 | if (halfline >= ypos && height > 0) |
| 663 | { |
| 664 | // Believe it or not, this is what the OP actually does... |
| 665 | // which is why they're required to be on a qphrase boundary! |
| 666 | uint64_t p1 = OPLoadPhrase(oldOPP | 0x08); |
| 667 | uint64_t p2 = OPLoadPhrase(oldOPP | 0x10); |
| 668 | //unneeded op_pointer += 16; |
| 669 | OPProcessScaledBitmap(p0, p1, p2, render); |
| 670 | |
| 671 | // OP write-backs |
| 672 | |
| 673 | uint16_t remainder = (p2 >> 16) & 0xFF;//, vscale = p2 >> 8; |
| 674 | uint8_t /*remainder = p2 >> 16,*/ vscale = p2 >> 8; |
| 675 | //Actually, we should skip this object if it has a vscale of zero. |
| 676 | //Or do we? Not sure... Atari Karts has a few lines that look like: |
| 677 | // (SCALED BITMAP) |
| 678 | //000E8268 --> phrase 00010000 7000B00D |
| 679 | // [7 (0) x 1 @ (13, 0) (8 bpp), l: 000E82A0, p: 000E0FC0 fp: 00, fl:RELEASE, idx:00, pt:01] |
| 680 | // [hsc: 9A, vsc: 00, rem: 00] |
| 681 | // Could it be the vscale is overridden if the DWIDTH is zero? Hmm... |
| 682 | //WriteLog("OP: Scaled bitmap processing (rem=%02X, vscale=%02X)...\n", remainder, vscale);//*/ |
| 683 | |
| 684 | if (vscale == 0) |
| 685 | vscale = 0x20; // OP bug??? Nope, it isn't...! Or is it? |
| 686 | |
| 687 | //extern int start_logging; |
| 688 | //if (start_logging) |
| 689 | // WriteLog("--> Returned from scaled bitmap processing (rem=%02X, vscale=%02X)...\n", remainder, vscale);//*/ |
| 690 | //Locks up here: |
| 691 | //--> Returned from scaled bitmap processing (rem=20, vscale=80)... |
| 692 | //There are other problems here, it looks like... |
| 693 | //Another lock up: |
| 694 | //About to execute OP (508)... |
| 695 | /* |
| 696 | OP: Scaled bitmap 4x? 4bpp at 38,? hscale=7C fpix=0 data=00075E28 pitch 1 hflipped=no dwidth=? (linked to 00071118) Transluency=no |
| 697 | --> Returned from scaled bitmap processing (rem=50, vscale=7C)... |
| 698 | OP: Scaled bitmap 4x? 4bpp at 38,? hscale=7C fpix=0 data=00075E28 pitch 1 hflipped=no dwidth=? (linked to 00071118) Transluency=no |
| 699 | --> Returned from scaled bitmap processing (rem=30, vscale=7C)... |
| 700 | OP: Scaled bitmap 4x? 4bpp at 38,? hscale=7C fpix=0 data=00075E28 pitch 1 hflipped=no dwidth=? (linked to 00071118) Transluency=no |
| 701 | --> Returned from scaled bitmap processing (rem=10, vscale=7C)... |
| 702 | OP: Scaled bitmap 4x? 4bpp at 36,? hscale=7E fpix=0 data=000756A8 pitch 1 hflipped=no dwidth=? (linked to 00073058) Transluency=no |
| 703 | --> Returned from scaled bitmap processing (rem=00, vscale=7E)... |
| 704 | OP: Scaled bitmap 4x? 4bpp at 34,? hscale=80 fpix=0 data=000756C8 pitch 1 hflipped=no dwidth=? (linked to 00073078) Transluency=no |
| 705 | --> Returned from scaled bitmap processing (rem=00, vscale=80)... |
| 706 | OP: Scaled bitmap 4x? 4bpp at 36,? hscale=7E fpix=0 data=000756C8 pitch 1 hflipped=no dwidth=? (linked to 00073058) Transluency=no |
| 707 | --> Returned from scaled bitmap processing (rem=5E, vscale=7E)... |
| 708 | OP: Scaled bitmap 4x? 4bpp at 34,? hscale=80 fpix=0 data=000756E8 pitch 1 hflipped=no dwidth=? (linked to 00073078) Transluency=no |
| 709 | --> Returned from scaled bitmap processing (rem=60, vscale=80)... |
| 710 | OP: Scaled bitmap 4x? 4bpp at 36,? hscale=7E fpix=0 data=000756C8 pitch 1 hflipped=no dwidth=? (linked to 00073058) Transluency=no |
| 711 | --> Returned from scaled bitmap processing (rem=3E, vscale=7E)... |
| 712 | OP: Scaled bitmap 4x? 4bpp at 34,? hscale=80 fpix=0 data=000756E8 pitch 1 hflipped=no dwidth=? (linked to 00073078) Transluency=no |
| 713 | --> Returned from scaled bitmap processing (rem=40, vscale=80)... |
| 714 | OP: Scaled bitmap 4x? 4bpp at 36,? hscale=7E fpix=0 data=000756C8 pitch 1 hflipped=no dwidth=? (linked to 00073058) Transluency=no |
| 715 | --> Returned from scaled bitmap processing (rem=1E, vscale=7E)... |
| 716 | OP: Scaled bitmap 4x? 4bpp at 34,? hscale=80 fpix=0 data=000756E8 pitch 1 hflipped=no dwidth=? (linked to 00073078) Transluency=no |
| 717 | --> Returned from scaled bitmap processing (rem=20, vscale=80)... |
| 718 | */ |
| 719 | //Here's another problem: |
| 720 | // [hsc: 20, vsc: 20, rem: 00] |
| 721 | // Since we're not checking for $E0 (but that's what we get from the above), we |
| 722 | // end up repeating this halfline unnecessarily... !!! FIX !!! [DONE, but... |
| 723 | // still not quite right. Either that, or the Accolade team that wrote Bubsy |
| 724 | // screwed up royal.] |
| 725 | //Also note: $E0 = 7.0 which IS a legal vscale value... |
| 726 | |
| 727 | // if (remainder & 0x80) // I.e., it's negative |
| 728 | // if ((remainder & 0x80) || remainder == 0) // I.e., it's <= 0 |
| 729 | // if ((remainder - 1) >= 0xE0) // I.e., it's <= 0 |
| 730 | // if ((remainder >= 0xE1) || remainder == 0)// I.e., it's <= 0 |
| 731 | // if ((remainder >= 0xE1 && remainder <= 0xFF) || remainder == 0)// I.e., it's <= 0 |
| 732 | // if (remainder <= 0x20) // I.e., it's <= 1.0 |
| 733 | // I.e., it's < 1.0f -> means it'll go negative when we subtract 1.0f. |
| 734 | if (remainder < 0x20) |
| 735 | { |
| 736 | uint64_t data = (p0 & 0xFFFFF80000000000LL) >> 40; |
| 737 | uint64_t dwidth = (p1 & 0xFFC0000) >> 15; |
| 738 | |
| 739 | // while (remainder & 0x80) |
| 740 | // while ((remainder & 0x80) || remainder == 0) |
| 741 | // while ((remainder - 1) >= 0xE0) |
| 742 | // while ((remainder >= 0xE1) || remainder == 0) |
| 743 | // while ((remainder >= 0xE1 && remainder <= 0xFF) || remainder == 0) |
| 744 | // while (remainder <= 0x20) |
| 745 | while (remainder < 0x20) |
| 746 | { |
| 747 | remainder += vscale; |
| 748 | |
| 749 | if (height) |
| 750 | height--; |
| 751 | |
| 752 | data += dwidth; |
| 753 | } |
| 754 | |
| 755 | p0 &= ~0xFFFFF80000FFC000LL; // Mask out old data... |
| 756 | p0 |= (uint64_t)height << 14; |
| 757 | p0 |= data << 40; |
| 758 | OPStorePhrase(oldOPP, p0); |
| 759 | } |
| 760 | |
| 761 | remainder -= 0x20; // 1.0f in [3.5] fixed point format |
| 762 | |
| 763 | //if (start_logging) |
| 764 | // WriteLog("--> Finished writebacks...\n");//*/ |
| 765 | |
| 766 | //WriteLog(" [%08X%08X -> ", (uint32_t)(p2>>32), (uint32_t)(p2&0xFFFFFFFF)); |
| 767 | p2 &= ~0x0000000000FF0000LL; |
| 768 | p2 |= (uint64_t)remainder << 16; |
| 769 | //WriteLog("%08X%08X]\n", (uint32_t)(p2>>32), (uint32_t)(p2&0xFFFFFFFF)); |
| 770 | OPStorePhrase(oldOPP + 16, p2); |
| 771 | //remainder = (uint8_t)(p2 >> 16), vscale = (uint8_t)(p2 >> 8); |
| 772 | //WriteLog(" [after]: rem=%02X, vscale=%02X\n", remainder, vscale); |
| 773 | } |
| 774 | |
| 775 | // OP bottom 3 bits are hardwired to zero. The link address |
| 776 | // reflects this, so we only need the top 19 bits of the address |
| 777 | // (which is why we only shift 21, and not 24). |
| 778 | op_pointer = (p0 & 0x000007FFFF000000LL) >> 21; |
| 779 | |
| 780 | // KLUDGE: Seems that memory access is mirrored in the first 8MB of |
| 781 | // memory... |
| 782 | if (op_pointer > 0x1FFFFF && op_pointer < 0x800000) |
| 783 | op_pointer &= 0xFF1FFFFF; // Knock out bits 21-23 |
| 784 | |
| 785 | break; |
| 786 | } |
| 787 | case OBJECT_TYPE_GPU: |
| 788 | { |
| 789 | //WriteLog("OP: Asserting GPU IRQ #3...\n"); |
| 790 | #ifdef _MSC_VER |
| 791 | #pragma message("Warning: Need to fix OP GPU IRQ handling! !!! FIX !!!") |
| 792 | #else |
| 793 | #warning "Need to fix OP GPU IRQ handling! !!! FIX !!!" |
| 794 | #endif // _MSC_VER |
| 795 | OPSetCurrentObject(p0); |
| 796 | GPUSetIRQLine(3, ASSERT_LINE); |
| 797 | //Also, OP processing is suspended from this point until OBF (F00026) is written to... |
| 798 | // !!! FIX !!! |
| 799 | //Do something like: |
| 800 | //OPSuspendedByGPU = true; |
| 801 | //Dunno if the OP keeps processing from where it was interrupted, or if it just continues |
| 802 | //on the next halfline... |
| 803 | // --> It continues from where it was interrupted! !!! FIX !!! |
| 804 | break; |
| 805 | } |
| 806 | case OBJECT_TYPE_BRANCH: |
| 807 | { |
| 808 | uint16_t ypos = (p0 >> 3) & 0x7FF; |
| 809 | // JTRM is wrong: CC is bits 14-16 (3 bits, *not* 2) |
| 810 | uint8_t cc = (p0 >> 14) & 0x07; |
| 811 | uint32_t link = (p0 >> 21) & 0x3FFFF8; |
| 812 | |
| 813 | switch (cc) |
| 814 | { |
| 815 | case CONDITION_EQUAL: |
| 816 | if (halfline == ypos || ypos == 0x7FF) |
| 817 | op_pointer = link; |
| 818 | break; |
| 819 | case CONDITION_LESS_THAN: |
| 820 | if (halfline < ypos) |
| 821 | op_pointer = link; |
| 822 | break; |
| 823 | case CONDITION_GREATER_THAN: |
| 824 | if (halfline > ypos) |
| 825 | op_pointer = link; |
| 826 | break; |
| 827 | case CONDITION_OP_FLAG_SET: |
| 828 | if (OPGetStatusRegister() & 0x01) |
| 829 | op_pointer = link; |
| 830 | break; |
| 831 | case CONDITION_SECOND_HALF_LINE: |
| 832 | // Branch if bit 10 of HC is set... |
| 833 | if (TOMGetHC() & 0x0400) |
| 834 | op_pointer = link; |
| 835 | break; |
| 836 | default: |
| 837 | // Basically, if you do this, the OP does nothing. :-) |
| 838 | WriteLog("OP: Unimplemented branch condition %i\n", cc); |
| 839 | } |
| 840 | break; |
| 841 | } |
| 842 | case OBJECT_TYPE_STOP: |
| 843 | { |
| 844 | OPSetCurrentObject(p0); |
| 845 | |
| 846 | if ((p0 & 0x08) && TOMIRQEnabled(IRQ_OPFLAG)) |
| 847 | { |
| 848 | TOMSetPendingObjectInt(); |
| 849 | m68k_set_irq(2); // Cause a 68K IPL 2 to occur... |
| 850 | } |
| 851 | |
| 852 | // Bail out, we're done... |
| 853 | return; |
| 854 | } |
| 855 | default: |
| 856 | WriteLog("OP: Unknown object type %i\n", (uint8_t)p0 & 0x07); |
| 857 | } |
| 858 | |
| 859 | // Here is a little sanity check to keep the OP from locking up the |
| 860 | // machine when fed bad data. Better would be to count how many actual |
| 861 | // cycles it used and bail out/reenter to properly simulate an |
| 862 | // overloaded OP... !!! FIX !!! |
| 863 | #ifdef _MSC_VER |
| 864 | #pragma message("Warning: Better would be to count how many actual cycles it used and bail out/reenter to properly simulate an overloaded OP... !!! FIX !!!") |
| 865 | #else |
| 866 | #warning "Better would be to count how many actual cycles it used and bail out/reenter to properly simulate an overloaded OP... !!! FIX !!!" |
| 867 | #endif // _MSC_VER |
| 868 | opCyclesToRun--; |
| 869 | |
| 870 | if (!opCyclesToRun) |
| 871 | return; |
| 872 | } |
| 873 | } |
| 874 | |
| 875 | |
| 876 | // |
| 877 | // Store fixed size bitmap in line buffer |
| 878 | // |
| 879 | void OPProcessFixedBitmap(uint64_t p0, uint64_t p1, bool render) |
| 880 | { |
| 881 | // Need to make sure that when writing that it stays within the line buffer... |
| 882 | // LBUF ($F01800 - $F01D9E) 360 x 32-bit RAM |
| 883 | uint8_t depth = (p1 >> 12) & 0x07; // Color depth of image |
| 884 | int32_t xpos = ((int16_t)((p1 << 4) & 0xFFFF)) >> 4;// Image xpos in LBUF |
| 885 | uint32_t iwidth = (p1 >> 28) & 0x3FF; // Image width in *phrases* |
| 886 | uint32_t data = (p0 >> 40) & 0xFFFFF8; // Pixel data address |
| 887 | uint32_t firstPix = (p1 >> 49) & 0x3F; |
| 888 | // "The LSB is significant only for scaled objects..." -JTRM |
| 889 | // "In 1 BPP mode, all five bits are significant. In 2 BPP mode, the top |
| 890 | // four are significant..." |
| 891 | firstPix &= 0x3E; |
| 892 | |
| 893 | // We can ignore the RELEASE (high order) bit for now--probably forever...! |
| 894 | // uint8_t flags = (p1 >> 45) & 0x0F; // REFLECT, RMW, TRANS, RELEASE |
| 895 | //Optimize: break these out to their own BOOL values |
| 896 | uint8_t flags = (p1 >> 45) & 0x07; // REFLECT (0), RMW (1), TRANS (2) |
| 897 | bool flagREFLECT = (flags & OPFLAG_REFLECT ? true : false), |
| 898 | flagRMW = (flags & OPFLAG_RMW ? true : false), |
| 899 | flagTRANS = (flags & OPFLAG_TRANS ? true : false); |
| 900 | // "For images with 1 to 4 bits/pixel the top 7 to 4 bits of the index |
| 901 | // provide the most significant bits of the palette address." |
| 902 | uint8_t index = (p1 >> 37) & 0xFE; // CLUT index offset (upper pix, 1-4 bpp) |
| 903 | uint32_t pitch = (p1 >> 15) & 0x07; // Phrase pitch |
| 904 | pitch <<= 3; // Optimization: Multiply pitch by 8 |
| 905 | |
| 906 | // int16_t scanlineWidth = tom_getVideoModeWidth(); |
| 907 | uint8_t * tomRam8 = TOMGetRamPointer(); |
| 908 | uint8_t * paletteRAM = &tomRam8[0x400]; |
| 909 | // This is OK as long as it's used correctly: For 16-bit RAM to RAM direct |
| 910 | // copies--NOT for use when using endian-corrected data (i.e., any of the |
| 911 | // *_word_read functions!) |
| 912 | uint16_t * paletteRAM16 = (uint16_t *)paletteRAM; |
| 913 | |
| 914 | // WriteLog("bitmap %ix? %ibpp at %i,? firstpix=? data=0x%.8x pitch %i hflipped=%s dwidth=? (linked to ?) RMW=%s Tranparent=%s\n", |
| 915 | // iwidth, op_bitmap_bit_depth[bitdepth], xpos, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), (flags&OPFLAG_RMW ? "yes" : "no"), (flags&OPFLAG_TRANS ? "yes" : "no")); |
| 916 | |
| 917 | // Is it OK to have a 0 for the data width??? (i.e., undocumented?) |
| 918 | // Seems to be... Seems that dwidth *can* be zero (i.e., reuse same line) as |
| 919 | // well. |
| 920 | // Pitch == 0 is OK too... |
| 921 | |
| 922 | //kludge: Seems that the OP treats iwidth == 0 as iwidth == 1... Need to |
| 923 | // investigate on real hardware... |
| 924 | #ifdef _MSC_VER |
| 925 | #pragma message("Warning: !!! Need to investigate iwidth == 0 behavior on real hardware !!!") |
| 926 | #else |
| 927 | #warning "!!! Need to investigate iwidth == 0 behavior on real hardware !!!" |
| 928 | #endif // _MSC_VER |
| 929 | if (iwidth == 0) |
| 930 | iwidth = 1; |
| 931 | |
| 932 | // if (!render || op_pointer == 0 || ptr == 0 || pitch == 0) |
| 933 | //I'm not convinced that we need to concern ourselves with data & op_pointer |
| 934 | //here either! |
| 935 | if (!render || iwidth == 0) |
| 936 | return; |
| 937 | |
| 938 | //OK, so we know the position in the line buffer is correct. It's the clipping |
| 939 | //in 24bpp mode that's wrong! |
| 940 | #if 0 |
| 941 | //This is a total kludge, based upon the fact that 24BPP mode puts *4* bytes |
| 942 | //into the line buffer for each pixel. |
| 943 | if (depth == 5) // i.e., 24bpp mode... |
| 944 | xpos >>= 1; // Cut it in half... |
| 945 | #endif |
| 946 | |
| 947 | //#define OP_DEBUG_BMP |
| 948 | //#ifdef OP_DEBUG_BMP |
| 949 | // WriteLog("bitmap %ix%i %ibpp at %i,%i firstpix=%i data=0x%.8x pitch %i hflipped=%s dwidth=%i (linked to 0x%.8x) Transluency=%s\n", |
| 950 | // iwidth, height, op_bitmap_bit_depth[bitdepth], xpos, ypos, firstPix, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), dwidth, op_pointer, (flags&OPFLAG_RMW ? "yes" : "no")); |
| 951 | //#endif |
| 952 | |
| 953 | // int32_t leftMargin = xpos, rightMargin = (xpos + (phraseWidthToPixels[depth] * iwidth)) - 1; |
| 954 | int32_t startPos = xpos, endPos = xpos + |
| 955 | (!flagREFLECT ? (phraseWidthToPixels[depth] * iwidth) - 1 |
| 956 | : -((phraseWidthToPixels[depth] * iwidth) + 1)); |
| 957 | uint32_t clippedWidth = 0, phraseClippedWidth = 0, dataClippedWidth = 0;//, phrasePixel = 0; |
| 958 | bool in24BPPMode = (((GET16(tomRam8, 0x0028) >> 1) & 0x03) == 1 ? true : false); // VMODE |
| 959 | // This is correct, the OP line buffer is a constant size... |
| 960 | int32_t limit = 720; |
| 961 | int32_t lbufWidth = 719; |
| 962 | |
| 963 | // If the image is completely to the left or right of the line buffer, then |
| 964 | // bail. |
| 965 | //If in REFLECT mode, then these values are swapped! !!! FIX !!! [DONE] |
| 966 | //There are four possibilities: |
| 967 | // 1. image sits on left edge and no REFLECT; starts out of bounds but ends in bounds. |
| 968 | // 2. image sits on left edge and REFLECT; starts in bounds but ends out of bounds. |
| 969 | // 3. image sits on right edge and REFLECT; starts out of bounds but ends in bounds. |
| 970 | // 4. image sits on right edge and no REFLECT; starts in bounds but ends out of bounds. |
| 971 | //Numbers 2 & 4 can be caught by checking the LBUF clip while in the inner loop, |
| 972 | // numbers 1 & 3 are of concern. |
| 973 | // This *indirectly* handles only cases 2 & 4! And is WRONG is REFLECT is set...! |
| 974 | // if (rightMargin < 0 || leftMargin > lbufWidth) |
| 975 | |
| 976 | // It might be easier to swap these (if REFLECTed) and just use XPOS down below... |
| 977 | // That way, you could simply set XPOS to leftMargin if !REFLECT and to rightMargin otherwise. |
| 978 | // Still have to be careful with the DATA and IWIDTH values though... |
| 979 | |
| 980 | // if ((!flagREFLECT && (rightMargin < 0 || leftMargin > lbufWidth)) |
| 981 | // || (flagREFLECT && (leftMargin < 0 || rightMargin > lbufWidth))) |
| 982 | // return; |
| 983 | if ((!flagREFLECT && (endPos < 0 || startPos > lbufWidth)) |
| 984 | || (flagREFLECT && (startPos < 0 || endPos > lbufWidth))) |
| 985 | return; |
| 986 | |
| 987 | // Otherwise, find the clip limits and clip the phrase as well... |
| 988 | // NOTE: I'm fudging here by letting the actual blit overstep the bounds of the |
| 989 | // line buffer, but it shouldn't matter since there are two unused line |
| 990 | // buffers below and nothing above and I'll at most write 8 bytes outside |
| 991 | // the line buffer... I could use a fractional clip begin/end value, but |
| 992 | // this makes the blit a *lot* more hairy. I might fix this in the future |
| 993 | // if it becomes necessary. (JLH) |
| 994 | // Probably wouldn't be *that* hairy. Just use a delta that tells the inner loop |
| 995 | // which pixel in the phrase is being written, and quit when either end of phrases |
| 996 | // is reached or line buffer extents are surpassed. |
| 997 | |
| 998 | //This stuff is probably wrong as well... !!! FIX !!! |
| 999 | //The strange thing is that it seems to work, but that's no guarantee that it's bulletproof! |
| 1000 | //Yup. Seems that JagMania doesn't work correctly with this... |
| 1001 | //Dunno if this is the problem, but Atari Karts is showing *some* of the road now... |
| 1002 | // if (!flagREFLECT) |
| 1003 | |
| 1004 | /* |
| 1005 | if (leftMargin < 0) |
| 1006 | clippedWidth = 0 - leftMargin, |
| 1007 | phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth], |
| 1008 | leftMargin = 0 - (clippedWidth % phraseWidthToPixels[depth]); |
| 1009 | // leftMargin = 0; |
| 1010 | |
| 1011 | if (rightMargin > lbufWidth) |
| 1012 | clippedWidth = rightMargin - lbufWidth, |
| 1013 | phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];//, |
| 1014 | // rightMargin = lbufWidth + (clippedWidth % phraseWidthToPixels[depth]); |
| 1015 | // rightMargin = lbufWidth; |
| 1016 | */ |
| 1017 | if (depth > 5) |
| 1018 | WriteLog("OP: We're about to encounter a divide by zero error!\n"); |
| 1019 | // NOTE: We're just using endPos to figure out how much, if any, to clip by. |
| 1020 | // ALSO: There may be another case where we start out of bounds and end out |
| 1021 | // of bounds...! |
| 1022 | // !!! FIX !!! |
| 1023 | if (startPos < 0) // Case #1: Begin out, end in, L to R |
| 1024 | clippedWidth = 0 - startPos, |
| 1025 | dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth], |
| 1026 | startPos = 0 - (clippedWidth % phraseWidthToPixels[depth]); |
| 1027 | |
| 1028 | if (endPos < 0) // Case #2: Begin in, end out, R to L |
| 1029 | clippedWidth = 0 - endPos, |
| 1030 | phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth]; |
| 1031 | |
| 1032 | if (endPos > lbufWidth) // Case #3: Begin in, end out, L to R |
| 1033 | clippedWidth = endPos - lbufWidth, |
| 1034 | phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth]; |
| 1035 | |
| 1036 | if (startPos > lbufWidth) // Case #4: Begin out, end in, R to L |
| 1037 | clippedWidth = startPos - lbufWidth, |
| 1038 | dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth], |
| 1039 | startPos = lbufWidth + (clippedWidth % phraseWidthToPixels[depth]); |
| 1040 | //printf("<OP:spos=%i,epos=%i]", startPos, endPos); |
| 1041 | |
| 1042 | // If the image is sitting on the line buffer left or right edge, we need to compensate |
| 1043 | // by decreasing the image phrase width accordingly. |
| 1044 | iwidth -= phraseClippedWidth; |
| 1045 | |
| 1046 | // Also, if we're clipping the phrase we need to make sure we're in the correct part of |
| 1047 | // the pixel data. |
| 1048 | // data += phraseClippedWidth * (pitch << 3); |
| 1049 | data += dataClippedWidth * pitch; |
| 1050 | |
| 1051 | // NOTE: When the bitmap is in REFLECT mode, the XPOS marks the *right* side of the |
| 1052 | // bitmap! This makes clipping & etc. MUCH, much easier...! |
| 1053 | // uint32_t lbufAddress = 0x1800 + (!in24BPPMode ? leftMargin * 2 : leftMargin * 4); |
| 1054 | //Why does this work right when multiplying startPos by 2 (instead of 4) for 24 BPP mode? |
| 1055 | //Is this a bug in the OP? |
| 1056 | //It's because in 24bpp mode, each pixel takes *4* bytes, instead of the usual 2. |
| 1057 | //Though it looks like we're doing it here no matter what... |
| 1058 | // uint32_t lbufAddress = 0x1800 + (!in24BPPMode ? startPos * 2 : startPos * 2); |
| 1059 | //Let's try this: |
| 1060 | uint32_t lbufAddress = 0x1800 + (startPos * 2); |
| 1061 | uint8_t * currentLineBuffer = &tomRam8[lbufAddress]; |
| 1062 | |
| 1063 | // Render. |
| 1064 | |
| 1065 | // Hmm. We check above for 24 BPP mode, but don't do anything about it below... |
| 1066 | // If we *were* in 24 BPP mode, how would you convert CRY to RGB24? Seems to me |
| 1067 | // that if you're in CRY mode then you wouldn't be able to use 24 BPP bitmaps |
| 1068 | // anyway. |
| 1069 | // This seems to be the case (at least according to the Midsummer docs)...! |
| 1070 | |
| 1071 | // This is to test using palette zeroes instead of bit zeroes... |
| 1072 | // And it seems that this is wrong, index == 0 is transparent apparently... :-/ |
| 1073 | //#define OP_USES_PALETTE_ZERO |
| 1074 | |
| 1075 | if (depth == 0) // 1 BPP |
| 1076 | { |
| 1077 | // The LSB of flags is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1078 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1079 | |
| 1080 | // Fetch 1st phrase... |
| 1081 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1082 | //Note that firstPix should only be honored *if* we start with the 1st phrase of the bitmap |
| 1083 | //i.e., we didn't clip on the margin... !!! FIX !!! |
| 1084 | pixels <<= firstPix; // Skip first N pixels (N=firstPix)... |
| 1085 | int i = firstPix; // Start counter at right spot... |
| 1086 | |
| 1087 | while (iwidth--) |
| 1088 | { |
| 1089 | while (i++ < 64) |
| 1090 | { |
| 1091 | uint8_t bit = pixels >> 63; |
| 1092 | #ifndef OP_USES_PALETTE_ZERO |
| 1093 | if (flagTRANS && bit == 0) |
| 1094 | #else |
| 1095 | if (flagTRANS && (paletteRAM16[index | bit] == 0)) |
| 1096 | #endif |
| 1097 | ; // Do nothing... |
| 1098 | else |
| 1099 | { |
| 1100 | if (!flagRMW) |
| 1101 | //Optimize: Set palleteRAM16 to beginning of palette RAM + index*2 and use only [bit] as index... |
| 1102 | //Won't optimize RMW case though... |
| 1103 | // This is the *only* correct use of endian-dependent code |
| 1104 | // (i.e., mem-to-mem direct copying)! |
| 1105 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bit]; |
| 1106 | else |
| 1107 | *currentLineBuffer = |
| 1108 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bit) << 1]), |
| 1109 | *(currentLineBuffer + 1) = |
| 1110 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bit) << 1) + 1]); |
| 1111 | } |
| 1112 | |
| 1113 | currentLineBuffer += lbufDelta; |
| 1114 | pixels <<= 1; |
| 1115 | } |
| 1116 | i = 0; |
| 1117 | // Fetch next phrase... |
| 1118 | data += pitch; |
| 1119 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1120 | } |
| 1121 | } |
| 1122 | else if (depth == 1) // 2 BPP |
| 1123 | { |
| 1124 | if (firstPix) |
| 1125 | WriteLog("OP: Fixed bitmap @ 2 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix); |
| 1126 | index &= 0xFC; // Top six bits form CLUT index |
| 1127 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1128 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1129 | |
| 1130 | while (iwidth--) |
| 1131 | { |
| 1132 | // Fetch phrase... |
| 1133 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1134 | data += pitch; |
| 1135 | |
| 1136 | for(int i=0; i<32; i++) |
| 1137 | { |
| 1138 | uint8_t bits = pixels >> 62; |
| 1139 | // Seems to me that both of these are in the same endian, so we could cast it as |
| 1140 | // uint16_t * and do straight across copies (what about 24 bpp? Treat it differently...) |
| 1141 | // This only works for the palettized modes (1 - 8 BPP), since we actually have to |
| 1142 | // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?) |
| 1143 | // No, it isn't because we read the memory in an endian safe way--this *won't* work... |
| 1144 | #ifndef OP_USES_PALETTE_ZERO |
| 1145 | if (flagTRANS && bits == 0) |
| 1146 | #else |
| 1147 | if (flagTRANS && (paletteRAM16[index | bits] == 0)) |
| 1148 | #endif |
| 1149 | ; // Do nothing... |
| 1150 | else |
| 1151 | { |
| 1152 | if (!flagRMW) |
| 1153 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bits]; |
| 1154 | else |
| 1155 | *currentLineBuffer = |
| 1156 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]), |
| 1157 | *(currentLineBuffer + 1) = |
| 1158 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]); |
| 1159 | } |
| 1160 | |
| 1161 | currentLineBuffer += lbufDelta; |
| 1162 | pixels <<= 2; |
| 1163 | } |
| 1164 | } |
| 1165 | } |
| 1166 | else if (depth == 2) // 4 BPP |
| 1167 | { |
| 1168 | if (firstPix) |
| 1169 | WriteLog("OP: Fixed bitmap @ 4 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix); |
| 1170 | index &= 0xF0; // Top four bits form CLUT index |
| 1171 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1172 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1173 | |
| 1174 | while (iwidth--) |
| 1175 | { |
| 1176 | // Fetch phrase... |
| 1177 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1178 | data += pitch; |
| 1179 | |
| 1180 | for(int i=0; i<16; i++) |
| 1181 | { |
| 1182 | uint8_t bits = pixels >> 60; |
| 1183 | // Seems to me that both of these are in the same endian, so we could cast it as |
| 1184 | // uint16_t * and do straight across copies (what about 24 bpp? Treat it differently...) |
| 1185 | // This only works for the palettized modes (1 - 8 BPP), since we actually have to |
| 1186 | // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?) |
| 1187 | // No, it isn't because we read the memory in an endian safe way--this *won't* work... |
| 1188 | #ifndef OP_USES_PALETTE_ZERO |
| 1189 | if (flagTRANS && bits == 0) |
| 1190 | #else |
| 1191 | if (flagTRANS && (paletteRAM16[index | bits] == 0)) |
| 1192 | #endif |
| 1193 | ; // Do nothing... |
| 1194 | else |
| 1195 | { |
| 1196 | if (!flagRMW) |
| 1197 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bits]; |
| 1198 | else |
| 1199 | *currentLineBuffer = |
| 1200 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]), |
| 1201 | *(currentLineBuffer + 1) = |
| 1202 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]); |
| 1203 | } |
| 1204 | |
| 1205 | currentLineBuffer += lbufDelta; |
| 1206 | pixels <<= 4; |
| 1207 | } |
| 1208 | } |
| 1209 | } |
| 1210 | else if (depth == 3) // 8 BPP |
| 1211 | { |
| 1212 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1213 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1214 | |
| 1215 | // Fetch 1st phrase... |
| 1216 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1217 | //Note that firstPix should only be honored *if* we start with the 1st phrase of the bitmap |
| 1218 | //i.e., we didn't clip on the margin... !!! FIX !!! |
| 1219 | firstPix &= 0x30; // Only top two bits are valid for 8 BPP |
| 1220 | pixels <<= firstPix; // Skip first N pixels (N=firstPix)... |
| 1221 | int i = firstPix >> 3; // Start counter at right spot... |
| 1222 | |
| 1223 | while (iwidth--) |
| 1224 | { |
| 1225 | while (i++ < 8) |
| 1226 | { |
| 1227 | uint8_t bits = pixels >> 56; |
| 1228 | // Seems to me that both of these are in the same endian, so we could cast it as |
| 1229 | // uint16_t * and do straight across copies (what about 24 bpp? Treat it differently...) |
| 1230 | // This only works for the palettized modes (1 - 8 BPP), since we actually have to |
| 1231 | // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?) |
| 1232 | // No, it isn't because we read the memory in an endian safe way--this *won't* work... |
| 1233 | //This would seem to be problematic... |
| 1234 | //Because it's the palette entry being zero that makes the pixel transparent... |
| 1235 | //Let's try it and see. |
| 1236 | #ifndef OP_USES_PALETTE_ZERO |
| 1237 | if (flagTRANS && bits == 0) |
| 1238 | #else |
| 1239 | if (flagTRANS && (paletteRAM16[bits] == 0)) |
| 1240 | #endif |
| 1241 | ; // Do nothing... |
| 1242 | else |
| 1243 | { |
| 1244 | if (!flagRMW) |
| 1245 | *(uint16_t *)currentLineBuffer = paletteRAM16[bits]; |
| 1246 | else |
| 1247 | *currentLineBuffer = |
| 1248 | BLEND_CR(*currentLineBuffer, paletteRAM[bits << 1]), |
| 1249 | *(currentLineBuffer + 1) = |
| 1250 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[(bits << 1) + 1]); |
| 1251 | } |
| 1252 | |
| 1253 | currentLineBuffer += lbufDelta; |
| 1254 | pixels <<= 8; |
| 1255 | } |
| 1256 | i = 0; |
| 1257 | // Fetch next phrase... |
| 1258 | data += pitch; |
| 1259 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1260 | } |
| 1261 | } |
| 1262 | else if (depth == 4) // 16 BPP |
| 1263 | { |
| 1264 | if (firstPix) |
| 1265 | WriteLog("OP: Fixed bitmap @ 16 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix); |
| 1266 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1267 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1268 | |
| 1269 | while (iwidth--) |
| 1270 | { |
| 1271 | // Fetch phrase... |
| 1272 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1273 | data += pitch; |
| 1274 | |
| 1275 | for(int i=0; i<4; i++) |
| 1276 | { |
| 1277 | uint8_t bitsHi = pixels >> 56, bitsLo = pixels >> 48; |
| 1278 | // Seems to me that both of these are in the same endian, so we could cast it |
| 1279 | // as uint16_t * and do straight across copies (what about 24 bpp? Treat it |
| 1280 | // differently...) This only works for the palettized modes (1 - 8 BPP), since |
| 1281 | // we actually have to copy data from memory in 16 BPP mode (or does it? Isn't |
| 1282 | // this the same as the CLUT case?) No, it isn't because we read the memory in |
| 1283 | // an endian safe way--it *won't* work... |
| 1284 | //This doesn't seem right... Let's try the encoded black value ($8800): |
| 1285 | //Apparently, CRY 0 maps to $8800... |
| 1286 | if (flagTRANS && ((bitsLo | bitsHi) == 0)) |
| 1287 | // if (flagTRANS && (bitsHi == 0x88) && (bitsLo == 0x00)) |
| 1288 | ; // Do nothing... |
| 1289 | else |
| 1290 | { |
| 1291 | if (!flagRMW) |
| 1292 | *currentLineBuffer = bitsHi, |
| 1293 | *(currentLineBuffer + 1) = bitsLo; |
| 1294 | else |
| 1295 | *currentLineBuffer = |
| 1296 | BLEND_CR(*currentLineBuffer, bitsHi), |
| 1297 | *(currentLineBuffer + 1) = |
| 1298 | BLEND_Y(*(currentLineBuffer + 1), bitsLo); |
| 1299 | } |
| 1300 | |
| 1301 | currentLineBuffer += lbufDelta; |
| 1302 | pixels <<= 16; |
| 1303 | } |
| 1304 | } |
| 1305 | } |
| 1306 | else if (depth == 5) // 24 BPP |
| 1307 | { |
| 1308 | //Looks like Iron Soldier is the only game that uses 24BPP mode... |
| 1309 | //There *might* be others... |
| 1310 | //WriteLog("OP: Writing 24 BPP bitmap!\n"); |
| 1311 | if (firstPix) |
| 1312 | WriteLog("OP: Fixed bitmap @ 24 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix); |
| 1313 | // Not sure, but I think RMW only works with 16 BPP and below, and only in CRY mode... |
| 1314 | // The LSB of flags is OPFLAG_REFLECT, so sign extend it and OR 4 into it. |
| 1315 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 4) | 0x04; |
| 1316 | |
| 1317 | while (iwidth--) |
| 1318 | { |
| 1319 | // Fetch phrase... |
| 1320 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1321 | data += pitch; |
| 1322 | |
| 1323 | for(int i=0; i<2; i++) |
| 1324 | { |
| 1325 | // We don't use a 32-bit var here because of endian issues...! |
| 1326 | uint8_t bits3 = pixels >> 56, bits2 = pixels >> 48, |
| 1327 | bits1 = pixels >> 40, bits0 = pixels >> 32; |
| 1328 | |
| 1329 | if (flagTRANS && (bits3 | bits2 | bits1 | bits0) == 0) |
| 1330 | ; // Do nothing... |
| 1331 | else |
| 1332 | *currentLineBuffer = bits3, |
| 1333 | *(currentLineBuffer + 1) = bits2, |
| 1334 | *(currentLineBuffer + 2) = bits1, |
| 1335 | *(currentLineBuffer + 3) = bits0; |
| 1336 | |
| 1337 | currentLineBuffer += lbufDelta; |
| 1338 | pixels <<= 32; |
| 1339 | } |
| 1340 | } |
| 1341 | } |
| 1342 | } |
| 1343 | |
| 1344 | |
| 1345 | // |
| 1346 | // Store scaled bitmap in line buffer |
| 1347 | // |
| 1348 | void OPProcessScaledBitmap(uint64_t p0, uint64_t p1, uint64_t p2, bool render) |
| 1349 | { |
| 1350 | // Need to make sure that when writing that it stays within the line buffer... |
| 1351 | // LBUF ($F01800 - $F01D9E) 360 x 32-bit RAM |
| 1352 | uint8_t depth = (p1 >> 12) & 0x07; // Color depth of image |
| 1353 | int32_t xpos = ((int16_t)((p1 << 4) & 0xFFFF)) >> 4;// Image xpos in LBUF |
| 1354 | uint32_t iwidth = (p1 >> 28) & 0x3FF; // Image width in *phrases* |
| 1355 | uint32_t data = (p0 >> 40) & 0xFFFFF8; // Pixel data address |
| 1356 | //#ifdef OP_DEBUG_BMP |
| 1357 | // Prolly should use this... Though not sure exactly how. |
| 1358 | //Use the upper bits as an offset into the phrase depending on the BPP. That's how! |
| 1359 | uint32_t firstPix = (p1 >> 49) & 0x3F; |
| 1360 | //This is WEIRD! I'm sure I saw Atari Karts request 8 BPP FIRSTPIX! What happened??? |
| 1361 | if (firstPix) |
| 1362 | WriteLog("OP: FIRSTPIX != 0! (Scaled BM)\n"); |
| 1363 | //#endif |
| 1364 | // We can ignore the RELEASE (high order) bit for now--probably forever...! |
| 1365 | // uint8_t flags = (p1 >> 45) & 0x0F; // REFLECT, RMW, TRANS, RELEASE |
| 1366 | //Optimize: break these out to their own BOOL values [DONE] |
| 1367 | uint8_t flags = (p1 >> 45) & 0x07; // REFLECT (0), RMW (1), TRANS (2) |
| 1368 | bool flagREFLECT = (flags & OPFLAG_REFLECT ? true : false), |
| 1369 | flagRMW = (flags & OPFLAG_RMW ? true : false), |
| 1370 | flagTRANS = (flags & OPFLAG_TRANS ? true : false); |
| 1371 | uint8_t index = (p1 >> 37) & 0xFE; // CLUT index offset (upper pix, 1-4 bpp) |
| 1372 | uint32_t pitch = (p1 >> 15) & 0x07; // Phrase pitch |
| 1373 | |
| 1374 | uint8_t * tomRam8 = TOMGetRamPointer(); |
| 1375 | uint8_t * paletteRAM = &tomRam8[0x400]; |
| 1376 | // This is OK as long as it's used correctly: For 16-bit RAM to RAM direct |
| 1377 | // copies--NOT for use when using endian-corrected data (i.e., any of the |
| 1378 | // *ReadWord functions!) |
| 1379 | uint16_t * paletteRAM16 = (uint16_t *)paletteRAM; |
| 1380 | |
| 1381 | uint16_t hscale = p2 & 0xFF; |
| 1382 | // Hmm. It seems that fixing the horizontal scale necessitated re-fixing this. |
| 1383 | // Not sure why, but seems to be consistent with the vertical scaling now (and |
| 1384 | // it may turn out to be wrong!)... |
| 1385 | uint16_t horizontalRemainder = hscale; // Not sure if it starts full, but seems reasonable [It's not!] |
| 1386 | // uint8_t horizontalRemainder = 0; // Let's try zero! Seems to work! Yay! [No, it doesn't!] |
| 1387 | int32_t scaledWidthInPixels = (iwidth * phraseWidthToPixels[depth] * hscale) >> 5; |
| 1388 | uint32_t scaledPhrasePixels = (phraseWidthToPixels[depth] * hscale) >> 5; |
| 1389 | |
| 1390 | // WriteLog("bitmap %ix? %ibpp at %i,? firstpix=? data=0x%.8x pitch %i hflipped=%s dwidth=? (linked to ?) RMW=%s Tranparent=%s\n", |
| 1391 | // iwidth, op_bitmap_bit_depth[bitdepth], xpos, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), (flags&OPFLAG_RMW ? "yes" : "no"), (flags&OPFLAG_TRANS ? "yes" : "no")); |
| 1392 | |
| 1393 | // Looks like an hscale of zero means don't draw! |
| 1394 | if (!render || iwidth == 0 || hscale == 0) |
| 1395 | return; |
| 1396 | |
| 1397 | /*extern int start_logging; |
| 1398 | if (start_logging) |
| 1399 | WriteLog("OP: Scaled bitmap %ix? %ibpp at %i,? hscale=%02X fpix=%i data=%08X pitch %i hflipped=%s dwidth=? (linked to %08X) Transluency=%s\n", |
| 1400 | iwidth, op_bitmap_bit_depth[depth], xpos, hscale, firstPix, data, pitch, (flagREFLECT ? "yes" : "no"), op_pointer, (flagRMW ? "yes" : "no"));*/ |
| 1401 | //#define OP_DEBUG_BMP |
| 1402 | //#ifdef OP_DEBUG_BMP |
| 1403 | // WriteLog("OP: Scaled bitmap %ix%i %ibpp at %i,%i firstpix=%i data=0x%.8x pitch %i hflipped=%s dwidth=%i (linked to 0x%.8x) Transluency=%s\n", |
| 1404 | // iwidth, height, op_bitmap_bit_depth[bitdepth], xpos, ypos, firstPix, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), dwidth, op_pointer, (flags&OPFLAG_RMW ? "yes" : "no")); |
| 1405 | //#endif |
| 1406 | |
| 1407 | int32_t startPos = xpos, endPos = xpos + |
| 1408 | (!flagREFLECT ? scaledWidthInPixels - 1 : -(scaledWidthInPixels + 1)); |
| 1409 | uint32_t clippedWidth = 0, phraseClippedWidth = 0, dataClippedWidth = 0; |
| 1410 | bool in24BPPMode = (((GET16(tomRam8, 0x0028) >> 1) & 0x03) == 1 ? true : false); // VMODE |
| 1411 | // Not sure if this is Jaguar Two only location or what... |
| 1412 | // From the docs, it is... If we want to limit here we should think of something else. |
| 1413 | // int32_t limit = GET16(tom_ram_8, 0x0008); // LIMIT |
| 1414 | int32_t limit = 720; |
| 1415 | // int32_t lbufWidth = (!in24BPPMode ? limit - 1 : (limit / 2) - 1); // Zero based limit... |
| 1416 | int32_t lbufWidth = 719; // Zero based limit... |
| 1417 | |
| 1418 | // If the image is completely to the left or right of the line buffer, then bail. |
| 1419 | //If in REFLECT mode, then these values are swapped! !!! FIX !!! [DONE] |
| 1420 | //There are four possibilities: |
| 1421 | // 1. image sits on left edge and no REFLECT; starts out of bounds but ends in bounds. |
| 1422 | // 2. image sits on left edge and REFLECT; starts in bounds but ends out of bounds. |
| 1423 | // 3. image sits on right edge and REFLECT; starts out of bounds but ends in bounds. |
| 1424 | // 4. image sits on right edge and no REFLECT; starts in bounds but ends out of bounds. |
| 1425 | //Numbers 2 & 4 can be caught by checking the LBUF clip while in the inner loop, |
| 1426 | // numbers 1 & 3 are of concern. |
| 1427 | // This *indirectly* handles only cases 2 & 4! And is WRONG if REFLECT is set...! |
| 1428 | // if (rightMargin < 0 || leftMargin > lbufWidth) |
| 1429 | |
| 1430 | // It might be easier to swap these (if REFLECTed) and just use XPOS down below... |
| 1431 | // That way, you could simply set XPOS to leftMargin if !REFLECT and to rightMargin otherwise. |
| 1432 | // Still have to be careful with the DATA and IWIDTH values though... |
| 1433 | |
| 1434 | if ((!flagREFLECT && (endPos < 0 || startPos > lbufWidth)) |
| 1435 | || (flagREFLECT && (startPos < 0 || endPos > lbufWidth))) |
| 1436 | return; |
| 1437 | |
| 1438 | // Otherwise, find the clip limits and clip the phrase as well... |
| 1439 | // NOTE: I'm fudging here by letting the actual blit overstep the bounds of |
| 1440 | // the line buffer, but it shouldn't matter since there are two |
| 1441 | // unused line buffers below and nothing above and I'll at most write |
| 1442 | // 40 bytes outside the line buffer... I could use a fractional clip |
| 1443 | // begin/end value, but this makes the blit a *lot* more hairy. I |
| 1444 | // might fix this in the future if it becomes necessary. (JLH) |
| 1445 | // Probably wouldn't be *that* hairy. Just use a delta that tells the |
| 1446 | // inner loop which pixel in the phrase is being written, and quit |
| 1447 | // when either end of phrases is reached or line buffer extents are |
| 1448 | // surpassed. |
| 1449 | |
| 1450 | //This stuff is probably wrong as well... !!! FIX !!! |
| 1451 | //The strange thing is that it seems to work, but that's no guarantee that it's |
| 1452 | //bulletproof! |
| 1453 | //Yup. Seems that JagMania doesn't work correctly with this... |
| 1454 | //Dunno if this is the problem, but Atari Karts is showing *some* of the road |
| 1455 | //now... |
| 1456 | //Actually, it is! Or, it was. It doesn't seem to be clipping here, so the |
| 1457 | //problem lies elsewhere! Hmm. Putting the scaling code into the 1/2/8 BPP cases |
| 1458 | //seems to draw the ground a bit more accurately... Strange! |
| 1459 | //It's probably a case of the REFLECT flag being set and the background being |
| 1460 | //written from the right side of the screen... |
| 1461 | //But no, it isn't... At least if the diagnostics are telling the truth! |
| 1462 | |
| 1463 | // NOTE: We're just using endPos to figure out how much, if any, to clip by. |
| 1464 | // ALSO: There may be another case where we start out of bounds and end out |
| 1465 | // of bounds...! |
| 1466 | // !!! FIX !!! |
| 1467 | |
| 1468 | //There's a problem here with scaledPhrasePixels in that it can be forced to |
| 1469 | //zero when the scaling factor is small. So fix it already! !!! FIX !!! |
| 1470 | /*if (scaledPhrasePixels == 0) |
| 1471 | { |
| 1472 | WriteLog("OP: [Scaled] We're about to encounter a divide by zero error!\n"); |
| 1473 | DumpScaledObject(p0, p1, p2); |
| 1474 | }//*/ |
| 1475 | //NOTE: I'm almost 100% sure that this is wrong... And it is! :-p |
| 1476 | |
| 1477 | //Try a simple example... |
| 1478 | // Let's say we have a 8 BPP scanline with an hscale of $80 (4). Our xpos is -10, |
| 1479 | // non-flipped. Pixels in the bitmap are XYZXYZXYZXYZXYZ. |
| 1480 | // Scaled up, they would be XXXXYYYYZZZZXXXXYYYYZZZZXXXXYYYYZZZZ... |
| 1481 | // |
| 1482 | // Normally, we would expect this in the line buffer: |
| 1483 | // ZZXXXXYYYYZZZZXXXXYYYYZZZZ... |
| 1484 | // |
| 1485 | // But instead we're getting: |
| 1486 | // XXXXYYYYZZZZXXXXYYYYZZZZ... |
| 1487 | // |
| 1488 | // or are we??? It would seem so, simply by virtue of the fact that we're NOT starting |
| 1489 | // on negative boundary--or are we? Hmm... |
| 1490 | // cw = 10, dcw = pcw = 10 / ([8 * 4 = 32] 32) = 0, sp = -10 |
| 1491 | // |
| 1492 | // Let's try a real world example: |
| 1493 | // |
| 1494 | //OP: Scaled bitmap (70, 8 BPP, spp=28) sp (-400) < 0... [new sp=-8, cw=400, dcw=pcw=14] |
| 1495 | //OP: Scaled bitmap (6F, 8 BPP, spp=27) sp (-395) < 0... [new sp=-17, cw=395, dcw=pcw=14] |
| 1496 | // |
| 1497 | // Really, spp is 27.75 in the second case... |
| 1498 | // So... If we do 395 / 27.75, we get 14. Ok so far... If we scale that against the |
| 1499 | // start position (14 * 27.75), we get -6.5... NOT -17! |
| 1500 | |
| 1501 | //Now it seems we're working OK, at least for the first case... |
| 1502 | uint32_t scaledPhrasePixelsUS = phraseWidthToPixels[depth] * hscale; |
| 1503 | |
| 1504 | if (startPos < 0) // Case #1: Begin out, end in, L to R |
| 1505 | { |
| 1506 | extern int start_logging; |
| 1507 | if (start_logging) |
| 1508 | WriteLog("OP: Scaled bitmap (%02X, %u BPP, spp=%u) start pos (%i) < 0...", hscale, op_bitmap_bit_depth[depth], scaledPhrasePixels, startPos); |
| 1509 | // clippedWidth = 0 - startPos, |
| 1510 | clippedWidth = (0 - startPos) << 5, |
| 1511 | // dataClippedWidth = phraseClippedWidth = clippedWidth / scaledPhrasePixels, |
| 1512 | dataClippedWidth = phraseClippedWidth = (clippedWidth / scaledPhrasePixelsUS) >> 5, |
| 1513 | // startPos = 0 - (clippedWidth % scaledPhrasePixels); |
| 1514 | startPos += (dataClippedWidth * scaledPhrasePixelsUS) >> 5; |
| 1515 | if (start_logging) |
| 1516 | WriteLog(" [new sp=%i, cw=%i, dcw=pcw=%i]\n", startPos, clippedWidth, dataClippedWidth); |
| 1517 | } |
| 1518 | |
| 1519 | if (endPos < 0) // Case #2: Begin in, end out, R to L |
| 1520 | clippedWidth = 0 - endPos, |
| 1521 | phraseClippedWidth = clippedWidth / scaledPhrasePixels; |
| 1522 | |
| 1523 | if (endPos > lbufWidth) // Case #3: Begin in, end out, L to R |
| 1524 | clippedWidth = endPos - lbufWidth, |
| 1525 | phraseClippedWidth = clippedWidth / scaledPhrasePixels; |
| 1526 | |
| 1527 | if (startPos > lbufWidth) // Case #4: Begin out, end in, R to L |
| 1528 | clippedWidth = startPos - lbufWidth, |
| 1529 | dataClippedWidth = phraseClippedWidth = clippedWidth / scaledPhrasePixels, |
| 1530 | startPos = lbufWidth + (clippedWidth % scaledPhrasePixels); |
| 1531 | |
| 1532 | extern int op_start_log; |
| 1533 | if (op_start_log && clippedWidth != 0) |
| 1534 | WriteLog("OP: Clipped line. SP=%i, EP=%i, clip=%u, iwidth=%u, hscale=%02X\n", startPos, endPos, clippedWidth, iwidth, hscale); |
| 1535 | if (op_start_log && startPos == 13) |
| 1536 | { |
| 1537 | WriteLog("OP: Scaled line. SP=%i, EP=%i, clip=%u, iwidth=%u, hscale=%02X, depth=%u, firstPix=%u\n", startPos, endPos, clippedWidth, iwidth, hscale, depth, firstPix); |
| 1538 | DumpScaledObject(p0, p1, p2); |
| 1539 | if (iwidth == 7) |
| 1540 | { |
| 1541 | WriteLog(" %08X: ", data); |
| 1542 | for(int i=0; i<7*8; i++) |
| 1543 | WriteLog("%02X ", JaguarReadByte(data+i)); |
| 1544 | WriteLog("\n"); |
| 1545 | } |
| 1546 | } |
| 1547 | // If the image is sitting on the line buffer left or right edge, we need to compensate |
| 1548 | // by decreasing the image phrase width accordingly. |
| 1549 | iwidth -= phraseClippedWidth; |
| 1550 | |
| 1551 | // Also, if we're clipping the phrase we need to make sure we're in the correct part of |
| 1552 | // the pixel data. |
| 1553 | // data += phraseClippedWidth * (pitch << 3); |
| 1554 | data += dataClippedWidth * (pitch << 3); |
| 1555 | |
| 1556 | // NOTE: When the bitmap is in REFLECT mode, the XPOS marks the *right* side of the |
| 1557 | // bitmap! This makes clipping & etc. MUCH, much easier...! |
| 1558 | // uint32_t lbufAddress = 0x1800 + (!in24BPPMode ? leftMargin * 2 : leftMargin * 4); |
| 1559 | // uint32_t lbufAddress = 0x1800 + (!in24BPPMode ? startPos * 2 : startPos * 4); |
| 1560 | uint32_t lbufAddress = 0x1800 + startPos * 2; |
| 1561 | uint8_t * currentLineBuffer = &tomRam8[lbufAddress]; |
| 1562 | //uint8_t * lineBufferLowerLimit = &tom_ram_8[0x1800], |
| 1563 | // * lineBufferUpperLimit = &tom_ram_8[0x1800 + 719]; |
| 1564 | |
| 1565 | // Render. |
| 1566 | |
| 1567 | // Hmm. We check above for 24 BPP mode, but don't do anything about it below... |
| 1568 | // If we *were* in 24 BPP mode, how would you convert CRY to RGB24? Seems to me |
| 1569 | // that if you're in CRY mode then you wouldn't be able to use 24 BPP bitmaps |
| 1570 | // anyway. |
| 1571 | // This seems to be the case (at least according to the Midsummer docs)...! |
| 1572 | |
| 1573 | if (depth == 0) // 1 BPP |
| 1574 | { |
| 1575 | if (firstPix != 0) |
| 1576 | WriteLog("OP: Scaled bitmap @ 1 BPP requesting FIRSTPIX!\n"); |
| 1577 | // The LSB of flags is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1578 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1579 | |
| 1580 | int pixCount = 0; |
| 1581 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1582 | |
| 1583 | while ((int32_t)iwidth > 0) |
| 1584 | { |
| 1585 | uint8_t bits = pixels >> 63; |
| 1586 | |
| 1587 | #ifndef OP_USES_PALETTE_ZERO |
| 1588 | if (flagTRANS && bits == 0) |
| 1589 | #else |
| 1590 | if (flagTRANS && (paletteRAM16[index | bits] == 0)) |
| 1591 | #endif |
| 1592 | ; // Do nothing... |
| 1593 | else |
| 1594 | { |
| 1595 | if (!flagRMW) |
| 1596 | // This is the *only* correct use of endian-dependent code |
| 1597 | // (i.e., mem-to-mem direct copying)! |
| 1598 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bits]; |
| 1599 | else |
| 1600 | *currentLineBuffer = |
| 1601 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]), |
| 1602 | *(currentLineBuffer + 1) = |
| 1603 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]); |
| 1604 | } |
| 1605 | |
| 1606 | currentLineBuffer += lbufDelta; |
| 1607 | |
| 1608 | /* |
| 1609 | The reason we subtract the horizontalRemainder *after* the test is because we had too few |
| 1610 | bytes for horizontalRemainder to properly recognize a negative number. But now it's 16 bits |
| 1611 | wide, so we could probably go back to that (as long as we make it an int16_t and not a uint16!) |
| 1612 | */ |
| 1613 | /* horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1614 | while (horizontalRemainder & 0x80) |
| 1615 | { |
| 1616 | horizontalRemainder += hscale; |
| 1617 | pixCount++; |
| 1618 | pixels <<= 1; |
| 1619 | }//*/ |
| 1620 | // while (horizontalRemainder <= 0x20) // I.e., it's <= 1.0 (*before* subtraction) |
| 1621 | while (horizontalRemainder < 0x20) // I.e., it's <= 1.0 (*before* subtraction) |
| 1622 | { |
| 1623 | horizontalRemainder += hscale; |
| 1624 | pixCount++; |
| 1625 | pixels <<= 1; |
| 1626 | } |
| 1627 | horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1628 | |
| 1629 | if (pixCount > 63) |
| 1630 | { |
| 1631 | int phrasesToSkip = pixCount / 64, pixelShift = pixCount % 64; |
| 1632 | |
| 1633 | data += (pitch << 3) * phrasesToSkip; |
| 1634 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1635 | pixels <<= 1 * pixelShift; |
| 1636 | iwidth -= phrasesToSkip; |
| 1637 | pixCount = pixelShift; |
| 1638 | } |
| 1639 | } |
| 1640 | } |
| 1641 | else if (depth == 1) // 2 BPP |
| 1642 | { |
| 1643 | if (firstPix != 0) |
| 1644 | WriteLog("OP: Scaled bitmap @ 2 BPP requesting FIRSTPIX!\n"); |
| 1645 | index &= 0xFC; // Top six bits form CLUT index |
| 1646 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1647 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1648 | |
| 1649 | int pixCount = 0; |
| 1650 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1651 | |
| 1652 | while ((int32_t)iwidth > 0) |
| 1653 | { |
| 1654 | uint8_t bits = pixels >> 62; |
| 1655 | |
| 1656 | #ifndef OP_USES_PALETTE_ZERO |
| 1657 | if (flagTRANS && bits == 0) |
| 1658 | #else |
| 1659 | if (flagTRANS && (paletteRAM16[index | bits] == 0)) |
| 1660 | #endif |
| 1661 | ; // Do nothing... |
| 1662 | else |
| 1663 | { |
| 1664 | if (!flagRMW) |
| 1665 | // This is the *only* correct use of endian-dependent code |
| 1666 | // (i.e., mem-to-mem direct copying)! |
| 1667 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bits]; |
| 1668 | else |
| 1669 | *currentLineBuffer = |
| 1670 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]), |
| 1671 | *(currentLineBuffer + 1) = |
| 1672 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]); |
| 1673 | } |
| 1674 | |
| 1675 | currentLineBuffer += lbufDelta; |
| 1676 | |
| 1677 | /* horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1678 | while (horizontalRemainder & 0x80) |
| 1679 | { |
| 1680 | horizontalRemainder += hscale; |
| 1681 | pixCount++; |
| 1682 | pixels <<= 2; |
| 1683 | }//*/ |
| 1684 | // while (horizontalRemainder <= 0x20) // I.e., it's <= 0 (*before* subtraction) |
| 1685 | while (horizontalRemainder < 0x20) // I.e., it's <= 1.0 (*before* subtraction) |
| 1686 | { |
| 1687 | horizontalRemainder += hscale; |
| 1688 | pixCount++; |
| 1689 | pixels <<= 2; |
| 1690 | } |
| 1691 | horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1692 | |
| 1693 | if (pixCount > 31) |
| 1694 | { |
| 1695 | int phrasesToSkip = pixCount / 32, pixelShift = pixCount % 32; |
| 1696 | |
| 1697 | data += (pitch << 3) * phrasesToSkip; |
| 1698 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1699 | pixels <<= 2 * pixelShift; |
| 1700 | iwidth -= phrasesToSkip; |
| 1701 | pixCount = pixelShift; |
| 1702 | } |
| 1703 | } |
| 1704 | } |
| 1705 | else if (depth == 2) // 4 BPP |
| 1706 | { |
| 1707 | if (firstPix != 0) |
| 1708 | WriteLog("OP: Scaled bitmap @ 4 BPP requesting FIRSTPIX!\n"); |
| 1709 | index &= 0xF0; // Top four bits form CLUT index |
| 1710 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1711 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1712 | |
| 1713 | int pixCount = 0; |
| 1714 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1715 | |
| 1716 | while ((int32_t)iwidth > 0) |
| 1717 | { |
| 1718 | uint8_t bits = pixels >> 60; |
| 1719 | |
| 1720 | #ifndef OP_USES_PALETTE_ZERO |
| 1721 | if (flagTRANS && bits == 0) |
| 1722 | #else |
| 1723 | if (flagTRANS && (paletteRAM16[index | bits] == 0)) |
| 1724 | #endif |
| 1725 | ; // Do nothing... |
| 1726 | else |
| 1727 | { |
| 1728 | if (!flagRMW) |
| 1729 | // This is the *only* correct use of endian-dependent code |
| 1730 | // (i.e., mem-to-mem direct copying)! |
| 1731 | *(uint16_t *)currentLineBuffer = paletteRAM16[index | bits]; |
| 1732 | else |
| 1733 | *currentLineBuffer = |
| 1734 | BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]), |
| 1735 | *(currentLineBuffer + 1) = |
| 1736 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]); |
| 1737 | } |
| 1738 | |
| 1739 | currentLineBuffer += lbufDelta; |
| 1740 | |
| 1741 | /* horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1742 | while (horizontalRemainder & 0x80) |
| 1743 | { |
| 1744 | horizontalRemainder += hscale; |
| 1745 | pixCount++; |
| 1746 | pixels <<= 4; |
| 1747 | }//*/ |
| 1748 | // while (horizontalRemainder <= 0x20) // I.e., it's <= 0 (*before* subtraction) |
| 1749 | while (horizontalRemainder < 0x20) // I.e., it's <= 0 (*before* subtraction) |
| 1750 | { |
| 1751 | horizontalRemainder += hscale; |
| 1752 | pixCount++; |
| 1753 | pixels <<= 4; |
| 1754 | } |
| 1755 | horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1756 | |
| 1757 | if (pixCount > 15) |
| 1758 | { |
| 1759 | int phrasesToSkip = pixCount / 16, pixelShift = pixCount % 16; |
| 1760 | |
| 1761 | data += (pitch << 3) * phrasesToSkip; |
| 1762 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1763 | pixels <<= 4 * pixelShift; |
| 1764 | iwidth -= phrasesToSkip; |
| 1765 | pixCount = pixelShift; |
| 1766 | } |
| 1767 | } |
| 1768 | } |
| 1769 | else if (depth == 3) // 8 BPP |
| 1770 | { |
| 1771 | if (firstPix) |
| 1772 | WriteLog("OP: Scaled bitmap @ 8 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix); |
| 1773 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it. |
| 1774 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1775 | |
| 1776 | int pixCount = 0; |
| 1777 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1778 | |
| 1779 | while ((int32_t)iwidth > 0) |
| 1780 | { |
| 1781 | uint8_t bits = pixels >> 56; |
| 1782 | |
| 1783 | #ifndef OP_USES_PALETTE_ZERO |
| 1784 | if (flagTRANS && bits == 0) |
| 1785 | #else |
| 1786 | if (flagTRANS && (paletteRAM16[bits] == 0)) |
| 1787 | #endif |
| 1788 | ; // Do nothing... |
| 1789 | else |
| 1790 | { |
| 1791 | if (!flagRMW) |
| 1792 | // This is the *only* correct use of endian-dependent code |
| 1793 | // (i.e., mem-to-mem direct copying)! |
| 1794 | *(uint16_t *)currentLineBuffer = paletteRAM16[bits]; |
| 1795 | /* { |
| 1796 | if (currentLineBuffer >= lineBufferLowerLimit && currentLineBuffer <= lineBufferUpperLimit) |
| 1797 | *(uint16_t *)currentLineBuffer = paletteRAM16[bits]; |
| 1798 | }*/ |
| 1799 | else |
| 1800 | *currentLineBuffer = |
| 1801 | BLEND_CR(*currentLineBuffer, paletteRAM[bits << 1]), |
| 1802 | *(currentLineBuffer + 1) = |
| 1803 | BLEND_Y(*(currentLineBuffer + 1), paletteRAM[(bits << 1) + 1]); |
| 1804 | } |
| 1805 | |
| 1806 | currentLineBuffer += lbufDelta; |
| 1807 | |
| 1808 | // while (horizontalRemainder <= 0x20) // I.e., it's <= 0 (*before* subtraction) |
| 1809 | while (horizontalRemainder < 0x20) // I.e., it's <= 1.0 (*before* subtraction) |
| 1810 | { |
| 1811 | horizontalRemainder += hscale; |
| 1812 | pixCount++; |
| 1813 | pixels <<= 8; |
| 1814 | } |
| 1815 | horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1816 | |
| 1817 | if (pixCount > 7) |
| 1818 | { |
| 1819 | int phrasesToSkip = pixCount / 8, pixelShift = pixCount % 8; |
| 1820 | |
| 1821 | data += (pitch << 3) * phrasesToSkip; |
| 1822 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1823 | pixels <<= 8 * pixelShift; |
| 1824 | iwidth -= phrasesToSkip; |
| 1825 | pixCount = pixelShift; |
| 1826 | } |
| 1827 | } |
| 1828 | } |
| 1829 | else if (depth == 4) // 16 BPP |
| 1830 | { |
| 1831 | if (firstPix != 0) |
| 1832 | WriteLog("OP: Scaled bitmap @ 16 BPP requesting FIRSTPIX!\n"); |
| 1833 | // The LSB is OPFLAG_REFLECT, so sign extend it and OR 2 into it. |
| 1834 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 5) | 0x02; |
| 1835 | |
| 1836 | int pixCount = 0; |
| 1837 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1838 | |
| 1839 | while ((int32_t)iwidth > 0) |
| 1840 | { |
| 1841 | uint8_t bitsHi = pixels >> 56, bitsLo = pixels >> 48; |
| 1842 | |
| 1843 | //This doesn't seem right... Let's try the encoded black value ($8800): |
| 1844 | //Apparently, CRY 0 maps to $8800... |
| 1845 | if (flagTRANS && ((bitsLo | bitsHi) == 0)) |
| 1846 | // if (flagTRANS && (bitsHi == 0x88) && (bitsLo == 0x00)) |
| 1847 | ; // Do nothing... |
| 1848 | else |
| 1849 | { |
| 1850 | if (!flagRMW) |
| 1851 | *currentLineBuffer = bitsHi, |
| 1852 | *(currentLineBuffer + 1) = bitsLo; |
| 1853 | else |
| 1854 | *currentLineBuffer = |
| 1855 | BLEND_CR(*currentLineBuffer, bitsHi), |
| 1856 | *(currentLineBuffer + 1) = |
| 1857 | BLEND_Y(*(currentLineBuffer + 1), bitsLo); |
| 1858 | } |
| 1859 | |
| 1860 | currentLineBuffer += lbufDelta; |
| 1861 | |
| 1862 | /* horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1863 | while (horizontalRemainder & 0x80) |
| 1864 | { |
| 1865 | horizontalRemainder += hscale; |
| 1866 | pixCount++; |
| 1867 | pixels <<= 16; |
| 1868 | }//*/ |
| 1869 | // while (horizontalRemainder <= 0x20) // I.e., it's <= 0 (*before* subtraction) |
| 1870 | while (horizontalRemainder < 0x20) // I.e., it's <= 1.0 (*before* subtraction) |
| 1871 | { |
| 1872 | horizontalRemainder += hscale; |
| 1873 | pixCount++; |
| 1874 | pixels <<= 16; |
| 1875 | } |
| 1876 | horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format |
| 1877 | //*/ |
| 1878 | if (pixCount > 3) |
| 1879 | { |
| 1880 | int phrasesToSkip = pixCount / 4, pixelShift = pixCount % 4; |
| 1881 | |
| 1882 | data += (pitch << 3) * phrasesToSkip; |
| 1883 | pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1884 | pixels <<= 16 * pixelShift; |
| 1885 | |
| 1886 | iwidth -= phrasesToSkip; |
| 1887 | |
| 1888 | pixCount = pixelShift; |
| 1889 | } |
| 1890 | } |
| 1891 | } |
| 1892 | else if (depth == 5) // 24 BPP |
| 1893 | { |
| 1894 | //I'm not sure that you can scale a 24 BPP bitmap properly--the JTRM seem to indicate as much. |
| 1895 | WriteLog("OP: Writing 24 BPP scaled bitmap!\n"); |
| 1896 | if (firstPix != 0) |
| 1897 | WriteLog("OP: Scaled bitmap @ 24 BPP requesting FIRSTPIX!\n"); |
| 1898 | // Not sure, but I think RMW only works with 16 BPP and below, and only in CRY mode... |
| 1899 | // The LSB is OPFLAG_REFLECT, so sign extend it and or 4 into it. |
| 1900 | int32_t lbufDelta = ((int8_t)((flags << 7) & 0xFF) >> 4) | 0x04; |
| 1901 | |
| 1902 | while (iwidth--) |
| 1903 | { |
| 1904 | // Fetch phrase... |
| 1905 | uint64_t pixels = ((uint64_t)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP); |
| 1906 | data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value) |
| 1907 | |
| 1908 | for(int i=0; i<2; i++) |
| 1909 | { |
| 1910 | uint8_t bits3 = pixels >> 56, bits2 = pixels >> 48, |
| 1911 | bits1 = pixels >> 40, bits0 = pixels >> 32; |
| 1912 | |
| 1913 | if (flagTRANS && (bits3 | bits2 | bits1 | bits0) == 0) |
| 1914 | ; // Do nothing... |
| 1915 | else |
| 1916 | *currentLineBuffer = bits3, |
| 1917 | *(currentLineBuffer + 1) = bits2, |
| 1918 | *(currentLineBuffer + 2) = bits1, |
| 1919 | *(currentLineBuffer + 3) = bits0; |
| 1920 | |
| 1921 | currentLineBuffer += lbufDelta; |
| 1922 | pixels <<= 32; |
| 1923 | } |
| 1924 | } |
| 1925 | } |
| 1926 | } |