5 // GCC/SDL port by Niels Wagenaar (Linux/WIN32) and Caz (BeOS)
6 // Cleanups/fixes/rewrites by James L. Hammons
15 //#define OP_DEBUG_BMP
17 #define BLEND_Y(dst, src) op_blend_y[(((uint16)dst<<8)) | ((uint16)(src))]
18 #define BLEND_CR(dst, src) op_blend_cr[(((uint16)dst)<<8) | ((uint16)(src))]
20 #define OBJECT_TYPE_BITMAP 0 // 000
21 #define OBJECT_TYPE_SCALE 1 // 001
22 #define OBJECT_TYPE_GPU 2 // 010
23 #define OBJECT_TYPE_BRANCH 3 // 011
24 #define OBJECT_TYPE_STOP 4 // 100
26 #define CONDITION_EQUAL 0
27 #define CONDITION_LESS_THAN 1
28 #define CONDITION_GREATER_THAN 2
29 #define CONDITION_OP_FLAG_SET 3
30 #define CONDITION_SECOND_HALF_LINE 4
32 #define OPFLAG_RELEASE 8 // Bus release bit
33 #define OPFLAG_TRANS 4 // Transparency bit
34 #define OPFLAG_RMW 2 // Read-Modify-Write bit
35 #define OPFLAG_REFLECT 1 // Horizontal mirror bit
37 // Private function prototypes
39 void OPProcessFixedBitmap(int scanline, uint64 p0, uint64 p1, bool render);
40 void OPProcessScaledBitmap(int scanline, uint64 p0, uint64 p1, uint64 p2, bool render);
41 void DumpScaledObject(uint64 p0, uint64 p1, uint64 p2);
42 void DumpFixedObject(uint64 p0, uint64 p1);
43 uint64 op_load_phrase(uint32 offset);
45 // External global variables
47 extern uint32 jaguar_mainRom_crc32;
49 // Local global variables
51 static uint8 * op_blend_y;
52 static uint8 * op_blend_cr;
53 // There may be a problem with this "RAM" overlapping (and thus being independent of)
54 // some of the regular TOM RAM...
55 static uint8 objectp_ram[0x40]; // This is based at $F00000
56 uint8 objectp_running;
57 bool objectp_stop_reading_list;
59 static uint8 op_bitmap_bit_depth[8] = { 1, 2, 4, 8, 16, 24, 32, 0 };
60 //static uint32 op_bitmap_bit_size[8] =
61 // { (uint32)(0.125*65536), (uint32)(0.25*65536), (uint32)(0.5*65536), (uint32)(1*65536),
62 // (uint32)(2*65536), (uint32)(1*65536), (uint32)(1*65536), (uint32)(1*65536) };
63 static uint32 op_pointer;
65 int32 phraseWidthToPixels[8] = { 64, 32, 16, 8, 4, 2, 0, 0 };
69 // Object Processor initialization
73 // Blend tables (64K each)
74 memory_malloc_secure((void **)&op_blend_y, 0x10000, "Jaguar Object processor Y blend lookup table");
75 memory_malloc_secure((void **)&op_blend_cr, 0x10000, "Jaguar Object processor CR blend lookup table");
77 // Here we calculate the saturating blend of a signed 4-bit value and an
78 // existing Cyan/Red value as well as a signed 8-bit value and an existing intensity...
79 // Note: CRY is 4 bits Cyan, 4 bits Red, 16 bits intensitY
80 for(int i=0; i<256*256; i++)
82 int y = (i >> 8) & 0xFF;
83 int dy = (INT8)i; // Sign extend the Y index
84 int c1 = (i >> 8) & 0x0F;
85 int dc1 = (INT8)(i << 4) >> 4; // Sign extend the R index
86 int c2 = (i >> 12) & 0x0F;
87 int dc2 = (INT8)(i & 0xF0) >> 4; // Sign extend the C index
107 op_blend_cr[i] = (c2 << 4) | c1;
114 // Object Processor reset
118 memset(objectp_ram, 0x00, 0x40);
125 { "(BITMAP)", "(SCALED BITMAP)", "(GPU INT)", "(BRANCH)", "(STOP)", "???", "???", "???" };
127 { "\"==\"", "\"<\"", "\">\"", "(opflag set)", "(second half line)", "?", "?", "?" };
129 uint32 olp = op_get_list_pointer();
130 WriteLog("OP: OLP = %08X\n", olp);
131 WriteLog("OP: Phrase dump\n ----------\n");
132 for(uint32 i=0; i<0x100; i+=8)
134 uint32 hi = JaguarReadLong(olp + i, OP), lo = JaguarReadLong(olp + i + 4, OP);
135 WriteLog("\t%08X: %08X %08X %s", olp + i, hi, lo, opType[lo & 0x07]);
136 if ((lo & 0x07) == 3)
138 uint16 ypos = (lo >> 3) & 0x7FF;
139 uint8 cc = (lo >> 14) & 0x03;
140 uint32 link = ((hi << 11) | (lo >> 21)) & 0x3FFFF8;
141 WriteLog(" YPOS=%u, CC=%s, link=%08X", ypos, ccType[cc], link);
144 if ((lo & 0x07) == 0)
145 DumpFixedObject(op_load_phrase(olp+i), op_load_phrase(olp+i+8));
146 if ((lo & 0x07) == 1)
147 DumpScaledObject(op_load_phrase(olp+i), op_load_phrase(olp+i+8), op_load_phrase(olp+i+16));
153 // Object Processor memory access
154 // Memory range: F00010 - F00027
156 uint8 OPReadByte(uint32 offset, uint32 who/*=UNKNOWN*/)
159 return objectp_ram[offset];
162 uint16 OPReadWord(uint32 offset, uint32 who/*=UNKNOWN*/)
165 return GET16(objectp_ram, offset);
168 // F00010-F00017 R xxxxxxxx xxxxxxxx OB - current object code from the graphics processor
169 // F00020-F00023 W xxxxxxxx xxxxxxxx OLP - start of the object list
170 // F00026 W -------- -------x OBF - object processor flag
172 void OPWriteByte(uint32 offset, uint8 data, uint32 who/*=UNKNOWN*/)
175 objectp_ram[offset] = data;
178 void OPWriteWord(uint32 offset, uint16 data, uint32 who/*=UNKNOWN*/)
181 SET16(objectp_ram, offset, data);
183 /*if (offset == 0x20)
184 WriteLog("OP: Setting lo list pointer: %04X\n", data);
186 WriteLog("OP: Setting hi list pointer: %04X\n", data);//*/
189 uint32 op_get_list_pointer(void)
191 // Note: This register is LO / HI WORD, hence the funky look of this...
192 // return (objectp_ram[0x22] << 24) | (objectp_ram[0x23] << 16) | (objectp_ram[0x20] << 8) | objectp_ram[0x21];
193 return GET16(objectp_ram, 0x20) | (GET16(objectp_ram, 0x22) << 16);
196 // This is WRONG, since the OBF is only 16 bits wide!!! [FIXED]
198 uint32 op_get_status_register(void)
200 // return (objectp_ram[0x26] << 24) | (objectp_ram[0x27] << 16) | (objectp_ram[0x28] << 8) | objectp_ram[0x29];
201 // return GET32(objectp_ram, 0x26);
202 return GET16(objectp_ram, 0x26);
205 // This is WRONG, since the OBF is only 16 bits wide!!! [FIXED]
207 void op_set_status_register(uint32 data)
209 /* objectp_ram[0x26] = (data & 0xFF000000) >> 24;
210 objectp_ram[0x27] = (data & 0x00FF0000) >> 16;
211 objectp_ram[0x28] = (data & 0x0000FF00) >> 8;
212 objectp_ram[0x29] |= (data & 0xFE);*/
213 objectp_ram[0x26] = (data & 0x0000FF00) >> 8;
214 objectp_ram[0x27] |= (data & 0xFE);
217 void op_set_current_object(uint64 object)
219 //Not sure this is right... Wouldn't it just be stored 64 bit BE?
220 // Stored as least significant 32 bits first, ms32 last in big endian
221 /* objectp_ram[0x13] = object & 0xFF; object >>= 8;
222 objectp_ram[0x12] = object & 0xFF; object >>= 8;
223 objectp_ram[0x11] = object & 0xFF; object >>= 8;
224 objectp_ram[0x10] = object & 0xFF; object >>= 8;
226 objectp_ram[0x17] = object & 0xFF; object >>= 8;
227 objectp_ram[0x16] = object & 0xFF; object >>= 8;
228 objectp_ram[0x15] = object & 0xFF; object >>= 8;
229 objectp_ram[0x14] = object & 0xFF;*/
230 // Let's try regular good old big endian...
231 objectp_ram[0x17] = object & 0xFF; object >>= 8;
232 objectp_ram[0x16] = object & 0xFF; object >>= 8;
233 objectp_ram[0x15] = object & 0xFF; object >>= 8;
234 objectp_ram[0x14] = object & 0xFF; object >>= 8;
236 objectp_ram[0x13] = object & 0xFF; object >>= 8;
237 objectp_ram[0x12] = object & 0xFF; object >>= 8;
238 objectp_ram[0x11] = object & 0xFF; object >>= 8;
239 objectp_ram[0x10] = object & 0xFF;
242 uint64 op_load_phrase(uint32 offset)
244 offset &= ~0x07; // 8 byte alignment
245 return ((uint64)JaguarReadLong(offset, OP) << 32) | (uint64)JaguarReadLong(offset+4, OP);
248 void OPStorePhrase(uint32 offset, uint64 p)
250 offset &= ~0x07; // 8 byte alignment
251 JaguarWriteLong(offset, p >> 32, OP);
252 JaguarWriteLong(offset + 4, p & 0xFFFFFFFF, OP);
256 // Debugging routines
258 void DumpScaledObject(uint64 p0, uint64 p1, uint64 p2)
260 WriteLog(" (SCALED BITMAP)");
261 WriteLog(" %08X --> phrase %08X %08X\n", op_pointer, (uint32)(p1>>32), (uint32)(p1&0xFFFFFFFF));
262 WriteLog(" %08X --> phrase %08X %08X ", op_pointer+8, (uint32)(p2>>32), (uint32)(p2&0xFFFFFFFF));
263 uint8 bitdepth = (p1 >> 12) & 0x07;
264 int16 ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)?
265 int32 xpos = p1 & 0xFFF;
266 xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos);
267 uint32 iwidth = ((p1 >> 28) & 0x3FF);
268 uint32 dwidth = ((p1 >> 18) & 0x3FF); // Unsigned!
269 uint16 height = ((p0 >> 14) & 0x3FF);
270 uint32 link = ((p0 >> 24) & 0x7FFFF) << 3;
271 uint32 ptr = ((p0 >> 43) & 0x1FFFFF) << 3;
272 uint32 firstPix = (p1 >> 49) & 0x3F;
273 uint8 flags = (p1 >> 45) & 0x0F;
274 uint8 idx = (p1 >> 38) & 0x7F;
275 uint32 pitch = (p1 >> 15) & 0x07;
276 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",
277 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);
278 uint32 hscale = p2 & 0xFF;
279 uint32 vscale = (p2 >> 8) & 0xFF;
280 uint32 remainder = (p2 >> 16) & 0xFF;
281 WriteLog(" [hsc: %02X, vsc: %02X, rem: %02X]\n", hscale, vscale, remainder);
284 void DumpFixedObject(uint64 p0, uint64 p1)
286 WriteLog(" (BITMAP)");
287 WriteLog(" %08X --> phrase %08X %08X\n", op_pointer, (uint32)(p1>>32), (uint32)(p1&0xFFFFFFFF));
288 uint8 bitdepth = (p1 >> 12) & 0x07;
289 int16 ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)?
290 int32 xpos = p1 & 0xFFF;
291 xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos);
292 uint32 iwidth = ((p1 >> 28) & 0x3FF);
293 uint32 dwidth = ((p1 >> 18) & 0x3FF); // Unsigned!
294 uint16 height = ((p0 >> 14) & 0x3FF);
295 uint32 link = ((p0 >> 24) & 0x7FFFF) << 3;
296 uint32 ptr = ((p0 >> 43) & 0x1FFFFF) << 3;
297 uint32 firstPix = (p1 >> 49) & 0x3F;
298 uint8 flags = (p1 >> 45) & 0x0F;
299 uint8 idx = (p1 >> 38) & 0x7F;
300 uint32 pitch = (p1 >> 15) & 0x07;
301 WriteLog(" [%u (%u) x %u @ (%i, %u) (%u bpp), l: %08X, p: %08X fp: %02X, fl:%s%s%s%s, idx:%02X, pt:%02X]\n",
302 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);
306 // Object Processor main routine
308 void OPProcessList(int scanline, bool render)
310 extern int op_start_log;
311 // char * condition_to_str[8] =
312 // { "==", "<", ">", "(opflag set)", "(second half line)", "?", "?", "?" };
314 // If jaguar_exec() is working right, we should *never* have to check for this
316 /* if (scanline < tom_get_vdb())
319 if (scanline >= 525)//tom_getVideoModeHeight()+tom_get_vdb())
322 op_pointer = op_get_list_pointer();
324 objectp_stop_reading_list = false;
326 // *** BEGIN OP PROCESSOR TESTING ONLY ***
327 extern bool interactiveMode;
329 extern int objectPtr;
331 int bitmapCounter = 0;
332 // *** END OP PROCESSOR TESTING ONLY ***
334 // if (op_pointer) WriteLog(" new op list at 0x%.8x scanline %i\n",op_pointer,scanline);
337 // *** BEGIN OP PROCESSOR TESTING ONLY ***
338 if (interactiveMode && bitmapCounter == objectPtr)
342 // *** END OP PROCESSOR TESTING ONLY ***
343 if (objectp_stop_reading_list)
346 uint64 p0 = op_load_phrase(op_pointer);
348 if (scanline == tom_get_vdb() && op_start_log)
349 //if (scanline == 215 && op_start_log)
351 WriteLog("%08X --> phrase %08X %08X", op_pointer - 8, (int)(p0>>32), (int)(p0&0xFFFFFFFF));
352 if ((p0 & 0x07) == OBJECT_TYPE_BITMAP)
354 WriteLog(" (BITMAP) ");
355 uint64 p1 = op_load_phrase(op_pointer);
356 WriteLog("\n%08X --> phrase %08X %08X ", op_pointer, (int)(p1>>32), (int)(p1&0xFFFFFFFF));
357 uint8 bitdepth = (p1 >> 12) & 0x07;
358 int16 ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)?
359 int32 xpos = p1 & 0xFFF;
360 xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos);
361 uint32 iwidth = ((p1 >> 28) & 0x3FF);
362 uint32 dwidth = ((p1 >> 18) & 0x3FF); // Unsigned!
363 uint16 height = ((p0 >> 14) & 0x3FF);
364 uint32 link = ((p0 >> 24) & 0x7FFFF) << 3;
365 uint32 ptr = ((p0 >> 43) & 0x1FFFFF) << 3;
366 uint32 firstPix = (p1 >> 49) & 0x3F;
367 uint8 flags = (p1 >> 45) & 0x0F;
368 uint8 idx = (p1 >> 38) & 0x7F;
369 uint32 pitch = (p1 >> 15) & 0x07;
370 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",
371 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);
373 if ((p0 & 0x07) == OBJECT_TYPE_SCALE)
375 WriteLog(" (SCALED BITMAP)");
376 uint64 p1 = op_load_phrase(op_pointer), p2 = op_load_phrase(op_pointer+8);
377 WriteLog("\n%08X --> phrase %08X %08X ", op_pointer, (int)(p1>>32), (int)(p1&0xFFFFFFFF));
378 WriteLog("\n%08X --> phrase %08X %08X ", op_pointer+8, (int)(p2>>32), (int)(p2&0xFFFFFFFF));
379 uint8 bitdepth = (p1 >> 12) & 0x07;
380 int16 ypos = ((p0 >> 3) & 0x3FF); // ??? What if not interlaced (/2)?
381 int32 xpos = p1 & 0xFFF;
382 xpos = (xpos & 0x800 ? xpos | 0xFFFFF000 : xpos);
383 uint32 iwidth = ((p1 >> 28) & 0x3FF);
384 uint32 dwidth = ((p1 >> 18) & 0x3FF); // Unsigned!
385 uint16 height = ((p0 >> 14) & 0x3FF);
386 uint32 link = ((p0 >> 24) & 0x7FFFF) << 3;
387 uint32 ptr = ((p0 >> 43) & 0x1FFFFF) << 3;
388 uint32 firstPix = (p1 >> 49) & 0x3F;
389 uint8 flags = (p1 >> 45) & 0x0F;
390 uint8 idx = (p1 >> 38) & 0x7F;
391 uint32 pitch = (p1 >> 15) & 0x07;
392 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",
393 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);
394 uint32 hscale = p2 & 0xFF;
395 uint32 vscale = (p2 >> 8) & 0xFF;
396 uint32 remainder = (p2 >> 16) & 0xFF;
397 WriteLog(" [hsc: %02X, vsc: %02X, rem: %02X]\n", hscale, vscale, remainder);
399 if ((p0 & 0x07) == OBJECT_TYPE_GPU)
400 WriteLog(" (GPU)\n");
401 if ((p0 & 0x07) == OBJECT_TYPE_BRANCH)
403 WriteLog(" (BRANCH)\n");
404 uint8 * jaguar_mainRam = GetRamPtr();
405 WriteLog("[RAM] --> ");
406 for(int k=0; k<8; k++)
407 WriteLog("%02X ", jaguar_mainRam[op_pointer-8 + k]);
410 if ((p0 & 0x07) == OBJECT_TYPE_STOP)
411 WriteLog(" --> List end\n");
414 // WriteLog("%08X type %i\n", op_pointer, (uint8)p0 & 0x07);
415 switch ((uint8)p0 & 0x07)
417 case OBJECT_TYPE_BITMAP:
419 // Would *not* be /2 if interlaced...!
420 uint16 ypos = ((p0 >> 3) & 0x3FF) / 2;
421 // This is only theory implied by Rayman...!
422 // It seems that if the YPOS is zero, then bump the YPOS value so that it coincides with
423 // the VDB value. With interlacing, this would be slightly more tricky.
424 // There's probably another bit somewhere that enables this mode--but so far, doesn't seem
425 // to affect any other game in a negative way (that I've seen).
426 // Either that, or it's an undocumented bug...
428 //No, the reason this was needed is that the OP code before was wrong. Any value
429 //less than VDB will get written to the top line of the display!
431 // ypos = TOMReadWord(0xF00046, OP) / 2; // Get the VDB value
432 uint32 height = (p0 & 0xFFC000) >> 14;
433 uint32 oldOPP = op_pointer - 8;
434 // *** BEGIN OP PROCESSOR TESTING ONLY ***
435 if (inhibit && op_start_log)
436 WriteLog("!!! ^^^ This object is INHIBITED! ^^^ !!!\n");
438 if (!inhibit) // For OP testing only!
439 // *** END OP PROCESSOR TESTING ONLY ***
440 if (scanline >= ypos && height > 0)
442 uint64 p1 = op_load_phrase(op_pointer);
444 //WriteLog("OP: Writing scanline %d with ypos == %d...\n", scanline, ypos);
445 //WriteLog("--> Writing %u BPP bitmap...\n", op_bitmap_bit_depth[(p1 >> 12) & 0x07]);
446 OPProcessFixedBitmap(scanline, p0, p1, render);
450 //???Does this really happen??? Doesn't seem to work if you do this...!
451 // uint32 link = (p0 & 0x7FFFF000000) >> 21;
452 // SET16(objectp_ram, 0x20, link & 0xFFFF); // OLP
453 // SET16(objectp_ram, 0x22, link >> 16);
454 /* uint32 height = (p0 & 0xFFC000) >> 14;
457 // NOTE: Would subtract 2 if in interlaced mode...!
458 // uint64 height = ((p0 & 0xFFC000) - 0x4000) & 0xFFC000;
462 uint64 data = (p0 & 0xFFFFF80000000000) >> 40;
463 uint64 dwidth = (p1 & 0xFFC0000) >> 15;
466 p0 &= ~0xFFFFF80000FFC000; // Mask out old data...
467 p0 |= (uint64)height << 14;
469 OPStorePhrase(oldOPP, p0);
471 op_pointer = (p0 & 0x000007FFFF000000) >> 21;
474 case OBJECT_TYPE_SCALE:
476 // Would *not* be /2 if interlaced...!
477 uint16 ypos = ((p0 >> 3) & 0x3FF) / 2;
478 // This is only theory implied by Rayman...!
479 // It seems that if the YPOS is zero, then bump the YPOS value so that it coincides with
480 // the VDB value. With interlacing, this would be slightly more tricky.
481 // There's probably another bit somewhere that enables this mode--but so far, doesn't seem
482 // to affect any other game in a negative way (that I've seen).
483 // Either that, or it's an undocumented bug...
485 //No, the reason this was needed is that the OP code before was wrong. Any value
486 //less than VDB will get written to the top line of the display!
488 // ypos = TOMReadWord(0xF00046, OP) / 2; // Get the VDB value
489 uint32 height = (p0 & 0xFFC000) >> 14;
490 uint32 oldOPP = op_pointer - 8;
491 // *** BEGIN OP PROCESSOR TESTING ONLY ***
492 if (inhibit && op_start_log)
494 WriteLog("!!! ^^^ This object is INHIBITED! ^^^ !!! (scanline=%u, ypos=%u, height=%u)\n", scanline, ypos, height);
495 DumpScaledObject(p0, op_load_phrase(op_pointer), op_load_phrase(op_pointer+8));
498 if (!inhibit) // For OP testing only!
499 // *** END OP PROCESSOR TESTING ONLY ***
500 if (scanline >= ypos && height > 0)
502 uint64 p1 = op_load_phrase(op_pointer);
504 uint64 p2 = op_load_phrase(op_pointer);
506 //WriteLog("OP: %08X (%d) %08X%08X %08X%08X %08X%08X\n", oldOPP, scanline, (uint32)(p0>>32), (uint32)(p0&0xFFFFFFFF), (uint32)(p1>>32), (uint32)(p1&0xFFFFFFFF), (uint32)(p2>>32), (uint32)(p2&0xFFFFFFFF));
507 OPProcessScaledBitmap(scanline, p0, p1, p2, render);
511 //???Does this really happen??? Doesn't seem to work if you do this...!
512 // uint32 link = (p0 & 0x7FFFF000000) >> 21;
513 // SET16(objectp_ram, 0x20, link & 0xFFFF); // OLP
514 // SET16(objectp_ram, 0x22, link >> 16);
515 /* uint32 height = (p0 & 0xFFC000) >> 14;
518 // NOTE: Would subtract 2 if in interlaced mode...!
519 // uint64 height = ((p0 & 0xFFC000) - 0x4000) & 0xFFC000;
521 uint8 remainder = p2 >> 16, vscale = p2 >> 8;
522 //Actually, we should skip this object if it has a vscale of zero.
523 //Or do we? Not sure... Atari Karts has a few lines that look like:
525 //000E8268 --> phrase 00010000 7000B00D
526 // [7 (0) x 1 @ (13, 0) (8 bpp), l: 000E82A0, p: 000E0FC0 fp: 00, fl:RELEASE, idx:00, pt:01]
527 // [hsc: 9A, vsc: 00, rem: 00]
528 // Could it be the vscale is overridden if the DWIDTH is zero? Hmm...
531 vscale = 0x20; // OP bug??? Nope, it isn't...! Or is it?
533 remainder -= 0x20; // 1.0f in [3.5] fixed point format
534 if (remainder & 0x80) // I.e., it's negative
536 uint64 data = (p0 & 0xFFFFF80000000000) >> 40;
537 uint64 dwidth = (p1 & 0xFFC0000) >> 15;
539 while (remainder & 0x80)
547 p0 &= ~0xFFFFF80000FFC000; // Mask out old data...
548 p0 |= (uint64)height << 14;
550 OPStorePhrase(oldOPP, p0);
553 //WriteLog(" [%08X%08X -> ", (uint32)(p2>>32), (uint32)(p2&0xFFFFFFFF));
554 p2 &= ~0x0000000000FF0000;
555 p2 |= (uint64)remainder << 16;
556 //WriteLog("%08X%08X]\n", (uint32)(p2>>32), (uint32)(p2&0xFFFFFFFF));
557 OPStorePhrase(oldOPP+16, p2);
558 //remainder = (uint8)(p2 >> 16), vscale = (uint8)(p2 >> 8);
559 //WriteLog(" [after]: rem=%02X, vscale=%02X\n", remainder, vscale);
561 op_pointer = (p0 & 0x000007FFFF000000) >> 21;
564 case OBJECT_TYPE_GPU:
566 //WriteLog("OP: Asserting GPU IRQ #3...\n");
567 op_set_current_object(p0);
568 GPUSetIRQLine(3, ASSERT_LINE);
569 //Also, OP processing is suspended from this point until OBF (F00026) is written to...
572 //OPSuspendedByGPU = true;
573 //Dunno if the OP keeps processing from where it was interrupted, or if it just continues
574 //on the next scanline...
577 case OBJECT_TYPE_BRANCH:
579 uint16 ypos = (p0 >> 3) & 0x7FF;
580 uint8 cc = (p0 >> 14) & 0x03;
581 uint32 link = (p0 >> 21) & 0x3FFFF8;
583 // if ((ypos!=507)&&(ypos!=25))
584 // WriteLog("\t%i%s%i link=0x%.8x\n",scanline,condition_to_str[cc],ypos>>1,link);
587 case CONDITION_EQUAL:
588 //Why do this for the equal case? If they wrote an odd YPOS, then it wouldn't be detected!
589 // if (ypos != 0x7FF && (ypos & 0x01))
591 // if ((2 * tom_get_scanline()) == ypos || ypos == 0x7FF)
592 //Here we're using VC instead of the bogus tom_get_scanline() value...
593 if (TOMReadWord(0xF00006, OP) == ypos || ypos == 0x7FF)
596 case CONDITION_LESS_THAN:
597 // if ((2 * tom_get_scanline()) < ypos)
598 if (TOMReadWord(0xF00006, OP) < ypos)
601 case CONDITION_GREATER_THAN:
602 // if ((2 * tom_get_scanline()) > ypos)
603 if (TOMReadWord(0xF00006, OP) > ypos)
606 case CONDITION_OP_FLAG_SET:
607 if (op_get_status_register() & 0x01)
610 case CONDITION_SECOND_HALF_LINE:
611 // This basically means branch if bit 10 of HC is set
612 WriteLog("OP: Unexpected CONDITION_SECOND_HALF_LINE in BRANCH object\nOP: shuting down\n");
617 WriteLog("OP: Unimplemented branch condition %i\n", cc);
621 case OBJECT_TYPE_STOP:
625 //WriteLog("OP: --> STOP\n");
626 // op_set_status_register(((p0>>3) & 0xFFFFFFFF));
627 //This seems more likely...
628 op_set_current_object(p0);
632 tom_set_pending_object_int();
633 if (tom_irq_enabled(IRQ_OPFLAG) && jaguar_interrupt_handler_is_valid(64))
634 m68k_set_irq(7); // Cause an NMI to occur...
641 WriteLog("op: unknown object type %i\n", ((uint8)p0 & 0x07));
648 // Store fixed size bitmap in line buffer
651 // Interesting thing about Rayman: There seems to be a transparent bitmap (1/8/16 bpp--which?)
652 // being rendered under his feet--doesn't align when walking... Check it out!
654 void OPProcessFixedBitmap(int scanline, uint64 p0, uint64 p1, bool render)
656 // Need to make sure that when writing that it stays within the line buffer...
657 // LBUF ($F01800 - $F01D9E) 360 x 32-bit RAM
658 uint8 depth = (p1 >> 12) & 0x07; // Color depth of image
659 int32 xpos = ((int16)((p1 << 4) & 0xFFFF)) >> 4;// Image xpos in LBUF
660 uint32 iwidth = (p1 >> 28) & 0x3FF; // Image width in *phrases*
661 uint32 data = (p0 >> 40) & 0xFFFFF8; // Pixel data address
662 //#ifdef OP_DEBUG_BMP
663 // Prolly should use this... Though not sure exactly how.
664 uint32 firstPix = (p1 >> 49) & 0x3F;
665 // "The LSB is significant only for scaled objects..." -JTRM
666 // "In 1 BPP mode, all five bits are significant. In 2 BPP mode, the top four are significant..."
669 // We can ignore the RELEASE (high order) bit for now--probably forever...!
670 // uint8 flags = (p1 >> 45) & 0x0F; // REFLECT, RMW, TRANS, RELEASE
671 //Optimize: break these out to their own BOOL values
672 uint8 flags = (p1 >> 45) & 0x07; // REFLECT (0), RMW (1), TRANS (2)
673 bool flagREFLECT = (flags & OPFLAG_REFLECT ? true : false),
674 flagRMW = (flags & OPFLAG_RMW ? true : false),
675 flagTRANS = (flags & OPFLAG_TRANS ? true : false);
676 // "For images with 1 to 4 bits/pixel the top 7 to 4 bits of the index
677 // provide the most significant bits of the palette address."
678 uint8 index = (p1 >> 37) & 0xFE; // CLUT index offset (upper pix, 1-4 bpp)
679 uint32 pitch = (p1 >> 15) & 0x07; // Phrase pitch
681 // int16 scanlineWidth = tom_getVideoModeWidth();
682 uint8 * tom_ram_8 = tom_get_ram_pointer();
683 uint8 * paletteRAM = &tom_ram_8[0x400];
684 // This is OK as long as it's used correctly: For 16-bit RAM to RAM direct copies--NOT
685 // for use when using endian-corrected data (i.e., any of the *_word_read functions!)
686 uint16 * paletteRAM16 = (uint16 *)paletteRAM;
688 // WriteLog("bitmap %ix? %ibpp at %i,? firstpix=? data=0x%.8x pitch %i hflipped=%s dwidth=? (linked to ?) RMW=%s Tranparent=%s\n",
689 // iwidth, op_bitmap_bit_depth[bitdepth], xpos, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), (flags&OPFLAG_RMW ? "yes" : "no"), (flags&OPFLAG_TRANS ? "yes" : "no"));
691 // Is it OK to have a 0 for the data width??? (i.e., undocumented?)
692 // Seems to be... Seems that dwidth *can* be zero (i.e., reuse same line) as well.
693 // Pitch == 0 is OK too...
694 // if (!render || op_pointer == 0 || dwidth == 0 || ptr == 0 || pitch == 0)
695 //I'm not convinced that we need to concern ourselves with data & op_pointer here either!
696 if (!render || iwidth == 0) // || data == 0 || op_pointer == 0)
699 //#define OP_DEBUG_BMP
700 //#ifdef OP_DEBUG_BMP
701 // 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",
702 // 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"));
705 // int32 leftMargin = xpos, rightMargin = (xpos + (phraseWidthToPixels[depth] * iwidth)) - 1;
706 int32 startPos = xpos, endPos = xpos +
707 (!flagREFLECT ? (phraseWidthToPixels[depth] * iwidth) - 1
708 : -((phraseWidthToPixels[depth] * iwidth) + 1));
709 uint32 clippedWidth = 0, phraseClippedWidth = 0, dataClippedWidth = 0;//, phrasePixel = 0;
710 bool in24BPPMode = (((GET16(tom_ram_8, 0x0028) >> 1) & 0x03) == 1 ? true : false); // VMODE
711 // Not sure if this is Jaguar Two only location or what...
712 // From the docs, it is... If we want to limit here we should think of something else.
713 // int32 limit = GET16(tom_ram_8, 0x0008); // LIMIT
715 int32 lbufWidth = (!in24BPPMode ? limit - 1 : (limit / 2) - 1); // Zero based limit...
717 // If the image is completely to the left or right of the line buffer, then bail.
718 //If in REFLECT mode, then these values are swapped! !!! FIX !!! [DONE]
719 //There are four possibilities:
720 // 1. image sits on left edge and no REFLECT; starts out of bounds but ends in bounds.
721 // 2. image sits on left edge and REFLECT; starts in bounds but ends out of bounds.
722 // 3. image sits on right edge and REFLECT; starts out of bounds but ends in bounds.
723 // 4. image sits on right edge and no REFLECT; starts in bounds but ends out of bounds.
724 //Numbers 2 & 4 can be caught by checking the LBUF clip while in the inner loop,
725 // numbers 1 & 3 are of concern.
726 // This *indirectly* handles only cases 2 & 4! And is WRONG is REFLECT is set...!
727 // if (rightMargin < 0 || leftMargin > lbufWidth)
729 // It might be easier to swap these (if REFLECTed) and just use XPOS down below...
730 // That way, you could simply set XPOS to leftMargin if !REFLECT and to rightMargin otherwise.
731 // Still have to be careful with the DATA and IWIDTH values though...
733 // if ((!flagREFLECT && (rightMargin < 0 || leftMargin > lbufWidth))
734 // || (flagREFLECT && (leftMargin < 0 || rightMargin > lbufWidth)))
736 if ((!flagREFLECT && (endPos < 0 || startPos > lbufWidth))
737 || (flagREFLECT && (startPos < 0 || endPos > lbufWidth)))
740 // Otherwise, find the clip limits and clip the phrase as well...
741 // NOTE: I'm fudging here by letting the actual blit overstep the bounds of the
742 // line buffer, but it shouldn't matter since there are two unused line
743 // buffers below and nothing above and I'll at most write 8 bytes outside
744 // the line buffer... I could use a fractional clip begin/end value, but
745 // this makes the blit a *lot* more hairy. I might fix this in the future
746 // if it becomes necessary. (JLH)
747 // Probably wouldn't be *that* hairy. Just use a delta that tells the inner loop
748 // which pixel in the phrase is being written, and quit when either end of phrases
749 // is reached or line buffer extents are surpassed.
751 //This stuff is probably wrong as well... !!! FIX !!!
752 //The strange thing is that it seems to work, but that's no guarantee that it's bulletproof!
753 //Yup. Seems that JagMania doesn't work correctly with this...
754 //Dunno if this is the problem, but Atari Karts is showing *some* of the road now...
759 clippedWidth = 0 - leftMargin,
760 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth],
761 leftMargin = 0 - (clippedWidth % phraseWidthToPixels[depth]);
764 if (rightMargin > lbufWidth)
765 clippedWidth = rightMargin - lbufWidth,
766 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];//,
767 // rightMargin = lbufWidth + (clippedWidth % phraseWidthToPixels[depth]);
768 // rightMargin = lbufWidth;
771 WriteLog("We're about to encounter a divide by zero error!\n");
772 // NOTE: We're just using endPos to figure out how much, if any, to clip by.
773 // ALSO: There may be another case where we start out of bounds and end out of bounds...!
774 if (startPos < 0) // Case #1: Begin out, end in, L to R
775 clippedWidth = 0 - startPos,
776 dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth],
777 startPos = 0 - (clippedWidth % phraseWidthToPixels[depth]);
779 if (endPos < 0) // Case #2: Begin in, end out, R to L
780 clippedWidth = 0 - endPos,
781 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];
783 if (endPos > lbufWidth) // Case #3: Begin in, end out, L to R
784 clippedWidth = endPos - lbufWidth,
785 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];
787 if (startPos > lbufWidth) // Case #4: Begin out, end in, R to L
788 clippedWidth = startPos - lbufWidth,
789 dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth],
790 startPos = lbufWidth + (clippedWidth % phraseWidthToPixels[depth]);
792 // If the image is sitting on the line buffer left or right edge, we need to compensate
793 // by decreasing the image phrase width accordingly.
794 iwidth -= phraseClippedWidth;
796 // Also, if we're clipping the phrase we need to make sure we're in the correct part of
798 // data += phraseClippedWidth * (pitch << 3);
799 data += dataClippedWidth * (pitch << 3);
801 // NOTE: When the bitmap is in REFLECT mode, the XPOS marks the *right* side of the
802 // bitmap! This makes clipping & etc. MUCH, much easier...!
803 // uint32 lbufAddress = 0x1800 + (!in24BPPMode ? leftMargin * 2 : leftMargin * 4);
804 uint32 lbufAddress = 0x1800 + (!in24BPPMode ? startPos * 2 : startPos * 4);
805 uint8 * currentLineBuffer = &tom_ram_8[lbufAddress];
809 // Hmm. We check above for 24 BPP mode, but don't do anything about it below...
810 // If we *were* in 24 BPP mode, how would you convert CRY to RGB24? Seems to me
811 // that if you're in CRY mode then you wouldn't be able to use 24 BPP bitmaps
814 if (depth == 0) // 1 BPP
816 // The LSB of flags is OPFLAG_REFLECT, so sign extend it and or 2 into it.
817 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
819 // Fetch 1st phrase...
820 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
821 //Note that firstPix should only be honored *if* we start with the 1st phrase of the bitmap
822 //i.e., we didn't clip on the margin...
823 pixels <<= firstPix; // Skip first N pixels (N=firstPix)...
824 int i = firstPix; // Start counter at right spot...
830 uint8 bit = pixels >> 63;
831 if (flagTRANS && bit == 0)
836 //Optimize: Set palleteRAM16 to beginning of palette RAM + index*2 and use only [bit] as index...
837 //Won't optimize RMW case though...
838 // This is the *only* correct use of endian-dependent code
839 // (i.e., mem-to-mem direct copying)!
840 *(uint16 *)currentLineBuffer = paletteRAM16[index | bit];
843 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bit) << 1]),
844 *(currentLineBuffer + 1) =
845 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bit) << 1) + 1]);
848 currentLineBuffer += lbufDelta;
852 // Fetch next phrase...
853 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
854 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
857 else if (depth == 1) // 2 BPP
860 WriteLog("OP: Fixed bitmap @ 2 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
861 index &= 0xFC; // Top six bits form CLUT index
862 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
863 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
868 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
869 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
871 for(int i=0; i<32; i++)
873 uint8 bits = pixels >> 62;
874 // Seems to me that both of these are in the same endian, so we could cast it as
875 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
876 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
877 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
878 // No, it isn't because we read the memory in an endian safe way--this *won't* work...
879 if (flagTRANS && bits == 0)
884 *(uint16 *)currentLineBuffer = paletteRAM16[index | bits];
887 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]),
888 *(currentLineBuffer + 1) =
889 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]);
892 currentLineBuffer += lbufDelta;
897 else if (depth == 2) // 4 BPP
900 WriteLog("OP: Fixed bitmap @ 4 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
901 index &= 0xF0; // Top four bits form CLUT index
902 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
903 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
908 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
909 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
911 for(int i=0; i<16; i++)
913 uint8 bits = pixels >> 60;
914 // Seems to me that both of these are in the same endian, so we could cast it as
915 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
916 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
917 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
918 // No, it isn't because we read the memory in an endian safe way--this *won't* work...
919 if (flagTRANS && bits == 0)
924 *(uint16 *)currentLineBuffer = paletteRAM16[index | bits];
927 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]),
928 *(currentLineBuffer + 1) =
929 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]);
932 currentLineBuffer += lbufDelta;
937 else if (depth == 3) // 8 BPP
940 WriteLog("OP: Fixed bitmap @ 8 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
941 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
942 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
947 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
948 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
950 for(int i=0; i<8; i++)
952 uint8 bits = pixels >> 56;
953 // Seems to me that both of these are in the same endian, so we could cast it as
954 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
955 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
956 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
957 // No, it isn't because we read the memory in an endian safe way--this *won't* work...
958 if (flagTRANS && bits == 0)
963 *(uint16 *)currentLineBuffer = paletteRAM16[bits];
966 BLEND_CR(*currentLineBuffer, paletteRAM[bits << 1]),
967 *(currentLineBuffer + 1) =
968 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[(bits << 1) + 1]);
971 currentLineBuffer += lbufDelta;
976 else if (depth == 4) // 16 BPP
979 WriteLog("OP: Fixed bitmap @ 16 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
980 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
981 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
986 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
987 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
989 for(int i=0; i<4; i++)
991 uint8 bitsHi = pixels >> 56, bitsLo = pixels >> 48;
992 // Seems to me that both of these are in the same endian, so we could cast it as
993 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
994 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
995 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
996 // No, it isn't because we read the memory in an endian safe way--it *won't* work...
997 if (flagTRANS && (bitsLo | bitsHi) == 0)
1002 *currentLineBuffer = bitsHi,
1003 *(currentLineBuffer + 1) = bitsLo;
1005 *currentLineBuffer =
1006 BLEND_CR(*currentLineBuffer, bitsHi),
1007 *(currentLineBuffer + 1) =
1008 BLEND_Y(*(currentLineBuffer + 1), bitsLo);
1011 currentLineBuffer += lbufDelta;
1016 else if (depth == 5) // 24 BPP
1018 //Looks like Iron Soldier is the only game that uses 24BPP mode...
1019 //There *might* be others...
1020 //WriteLog("OP: Writing 24 BPP bitmap!\n");
1022 WriteLog("OP: Fixed bitmap @ 24 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
1023 // Not sure, but I think RMW only works with 16 BPP and below, and only in CRY mode...
1024 // The LSB is OPFLAG_REFLECT, so sign extend it and or 4 into it.
1025 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 4) | 0x04;
1030 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1031 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
1033 for(int i=0; i<2; i++)
1035 uint8 bits3 = pixels >> 56, bits2 = pixels >> 48,
1036 bits1 = pixels >> 40, bits0 = pixels >> 32;
1037 // Seems to me that both of these are in the same endian, so we could cast it as
1038 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
1039 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
1040 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
1041 // No, it isn't because we read the memory in an endian safe way--it *won't* work...
1042 if (flagTRANS && (bits3 | bits2 | bits1 | bits0) == 0)
1045 *currentLineBuffer = bits3,
1046 *(currentLineBuffer + 1) = bits2,
1047 *(currentLineBuffer + 2) = bits1,
1048 *(currentLineBuffer + 3) = bits0;
1050 currentLineBuffer += lbufDelta;
1058 // Store scaled bitmap in line buffer
1060 void OPProcessScaledBitmap(int scanline, uint64 p0, uint64 p1, uint64 p2, bool render)
1062 // Need to make sure that when writing that it stays within the line buffer...
1063 // LBUF ($F01800 - $F01D9E) 360 x 32-bit RAM
1064 uint8 depth = (p1 >> 12) & 0x07; // Color depth of image
1065 int32 xpos = ((int16)((p1 << 4) & 0xFFFF)) >> 4;// Image xpos in LBUF
1066 uint32 iwidth = (p1 >> 28) & 0x3FF; // Image width in *phrases*
1067 uint32 data = (p0 >> 40) & 0xFFFFF8; // Pixel data address
1068 //#ifdef OP_DEBUG_BMP
1069 // Prolly should use this... Though not sure exactly how.
1070 //Use the upper bits as an offset into the phrase depending on the BPP. That's how!
1071 uint32 firstPix = (p1 >> 49) & 0x3F;
1072 //This is WEIRD! I'm sure I saw Atari Karts request 8 BPP FIRSTPIX! What happened???
1074 WriteLog("OP: FIRSTPIX != 0!\n");
1076 // We can ignore the RELEASE (high order) bit for now--probably forever...!
1077 // uint8 flags = (p1 >> 45) & 0x0F; // REFLECT, RMW, TRANS, RELEASE
1078 //Optimize: break these out to their own BOOL values
1079 uint8 flags = (p1 >> 45) & 0x07; // REFLECT (0), RMW (1), TRANS (2)
1080 bool flagREFLECT = (flags & OPFLAG_REFLECT ? true : false),
1081 flagRMW = (flags & OPFLAG_RMW ? true : false),
1082 flagTRANS = (flags & OPFLAG_TRANS ? true : false);
1083 uint8 index = (p1 >> 37) & 0xFE; // CLUT index offset (upper pix, 1-4 bpp)
1084 uint32 pitch = (p1 >> 15) & 0x07; // Phrase pitch
1086 // int16 scanlineWidth = tom_getVideoModeWidth();
1087 uint8 * tom_ram_8 = tom_get_ram_pointer();
1088 uint8 * paletteRAM = &tom_ram_8[0x400];
1089 // This is OK as long as it's used correctly: For 16-bit RAM to RAM direct copies--NOT
1090 // for use when using endian-corrected data (i.e., any of the *_word_read functions!)
1091 uint16 * paletteRAM16 = (uint16 *)paletteRAM;
1093 uint8 hscale = p2 & 0xFF;
1094 uint8 horizontalRemainder = hscale; // Not sure if it starts full, but seems reasonable
1095 int32 scaledWidthInPixels = (iwidth * phraseWidthToPixels[depth] * hscale) >> 5;
1096 uint32 scaledPhrasePixels = (phraseWidthToPixels[depth] * hscale) >> 5;
1098 // WriteLog("bitmap %ix? %ibpp at %i,? firstpix=? data=0x%.8x pitch %i hflipped=%s dwidth=? (linked to ?) RMW=%s Tranparent=%s\n",
1099 // iwidth, op_bitmap_bit_depth[bitdepth], xpos, ptr, pitch, (flags&OPFLAG_REFLECT ? "yes" : "no"), (flags&OPFLAG_RMW ? "yes" : "no"), (flags&OPFLAG_TRANS ? "yes" : "no"));
1101 //Looks like an hscale of zero means don't draw!
1102 if (!render || iwidth == 0 || hscale == 0)
1105 //#define OP_DEBUG_BMP
1106 //#ifdef OP_DEBUG_BMP
1107 // 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",
1108 // 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"));
1111 int32 startPos = xpos, endPos = xpos +
1112 (!flagREFLECT ? scaledWidthInPixels - 1 : -(scaledWidthInPixels + 1));
1113 uint32 clippedWidth = 0, phraseClippedWidth = 0, dataClippedWidth = 0;
1114 bool in24BPPMode = (((GET16(tom_ram_8, 0x0028) >> 1) & 0x03) == 1 ? true : false); // VMODE
1115 // Not sure if this is Jaguar Two only location or what...
1116 // From the docs, it is... If we want to limit here we should think of something else.
1117 // int32 limit = GET16(tom_ram_8, 0x0008); // LIMIT
1119 int32 lbufWidth = (!in24BPPMode ? limit - 1 : (limit / 2) - 1); // Zero based limit...
1121 // If the image is completely to the left or right of the line buffer, then bail.
1122 //If in REFLECT mode, then these values are swapped! !!! FIX !!! [DONE]
1123 //There are four possibilities:
1124 // 1. image sits on left edge and no REFLECT; starts out of bounds but ends in bounds.
1125 // 2. image sits on left edge and REFLECT; starts in bounds but ends out of bounds.
1126 // 3. image sits on right edge and REFLECT; starts out of bounds but ends in bounds.
1127 // 4. image sits on right edge and no REFLECT; starts in bounds but ends out of bounds.
1128 //Numbers 2 & 4 can be caught by checking the LBUF clip while in the inner loop,
1129 // numbers 1 & 3 are of concern.
1130 // This *indirectly* handles only cases 2 & 4! And is WRONG is REFLECT is set...!
1131 // if (rightMargin < 0 || leftMargin > lbufWidth)
1133 // It might be easier to swap these (if REFLECTed) and just use XPOS down below...
1134 // That way, you could simply set XPOS to leftMargin if !REFLECT and to rightMargin otherwise.
1135 // Still have to be careful with the DATA and IWIDTH values though...
1137 if ((!flagREFLECT && (endPos < 0 || startPos > lbufWidth))
1138 || (flagREFLECT && (startPos < 0 || endPos > lbufWidth)))
1141 // Otherwise, find the clip limits and clip the phrase as well...
1142 // NOTE: I'm fudging here by letting the actual blit overstep the bounds of the
1143 // line buffer, but it shouldn't matter since there are two unused line
1144 // buffers below and nothing above and I'll at most write 40 bytes outside
1145 // the line buffer... I could use a fractional clip begin/end value, but
1146 // this makes the blit a *lot* more hairy. I might fix this in the future
1147 // if it becomes necessary. (JLH)
1148 // Probably wouldn't be *that* hairy. Just use a delta that tells the inner loop
1149 // which pixel in the phrase is being written, and quit when either end of phrases
1150 // is reached or line buffer extents are surpassed.
1152 //This stuff is probably wrong as well... !!! FIX !!!
1153 //The strange thing is that it seems to work, but that's no guarantee that it's bulletproof!
1154 //Yup. Seems that JagMania doesn't work correctly with this...
1155 //Dunno if this is the problem, but Atari Karts is showing *some* of the road now...
1156 //Actually, it is! Or, it was. It doesn't seem to be clipping here, so the problem lies
1157 //elsewhere! Hmm. Putting the scaling code into the 1/2/8 BPP cases seems to draw the ground
1158 // a bit more accurately... Strange!
1159 //It's probably a case of the REFLECT flag being set and the background being written
1160 //from the right side of the screen...
1161 //But no, it isn't... At least if the diagnostics are telling the truth!
1163 // NOTE: We're just using endPos to figure out how much, if any, to clip by.
1164 // ALSO: There may be another case where we start out of bounds and end out of bounds...!
1166 //There's a problem here with scaledPhrasePixels in that it can be forced to zero when
1167 //the scaling factor is small. So fix it already! !!! FIX !!!
1168 /*if (scaledPhrasePixels == 0)
1170 WriteLog("OP: [Scaled] We're about to encounter a divide by zero error!\n");
1171 DumpScaledObject(p0, p1, p2);
1173 //NOTE: I'm almost 100% sure that this is wrong... And it is! :-p
1174 if (startPos < 0) // Case #1: Begin out, end in, L to R
1175 /* clippedWidth = 0 - startPos,
1176 dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth],
1177 startPos = 0 - (clippedWidth % phraseWidthToPixels[depth]);*/
1178 clippedWidth = 0 - startPos,
1179 dataClippedWidth = phraseClippedWidth = clippedWidth / scaledPhrasePixels,
1180 startPos = 0 - (clippedWidth % scaledPhrasePixels);
1182 if (endPos < 0) // Case #2: Begin in, end out, R to L
1183 /* clippedWidth = 0 - endPos,
1184 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];*/
1185 clippedWidth = 0 - endPos,
1186 phraseClippedWidth = clippedWidth / scaledPhrasePixels;
1188 if (endPos > lbufWidth) // Case #3: Begin in, end out, L to R
1189 /* clippedWidth = endPos - lbufWidth,
1190 phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth];*/
1191 clippedWidth = endPos - lbufWidth,
1192 phraseClippedWidth = clippedWidth / scaledPhrasePixels;
1194 if (startPos > lbufWidth) // Case #4: Begin out, end in, R to L
1195 /* clippedWidth = startPos - lbufWidth,
1196 dataClippedWidth = phraseClippedWidth = clippedWidth / phraseWidthToPixels[depth],
1197 startPos = lbufWidth + (clippedWidth % phraseWidthToPixels[depth]);*/
1198 clippedWidth = startPos - lbufWidth,
1199 dataClippedWidth = phraseClippedWidth = clippedWidth / scaledPhrasePixels,
1200 startPos = lbufWidth + (clippedWidth % scaledPhrasePixels);
1202 extern int op_start_log;
1203 if (op_start_log && clippedWidth != 0)
1204 WriteLog("OP: Clipped line. SP=%i, EP=%i, clip=%u, iwidth=%u, hscale=%02X\n", startPos, endPos, clippedWidth, iwidth, hscale);
1205 if (op_start_log && startPos == 13)
1207 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);
1208 DumpScaledObject(p0, p1, p2);
1210 // If the image is sitting on the line buffer left or right edge, we need to compensate
1211 // by decreasing the image phrase width accordingly.
1212 iwidth -= phraseClippedWidth;
1214 // Also, if we're clipping the phrase we need to make sure we're in the correct part of
1216 // data += phraseClippedWidth * (pitch << 3);
1217 data += dataClippedWidth * (pitch << 3);
1219 // NOTE: When the bitmap is in REFLECT mode, the XPOS marks the *right* side of the
1220 // bitmap! This makes clipping & etc. MUCH, much easier...!
1221 // uint32 lbufAddress = 0x1800 + (!in24BPPMode ? leftMargin * 2 : leftMargin * 4);
1222 uint32 lbufAddress = 0x1800 + (!in24BPPMode ? startPos * 2 : startPos * 4);
1223 uint8 * currentLineBuffer = &tom_ram_8[lbufAddress];
1227 // Hmm. We check above for 24 BPP mode, but don't do anything about it below...
1228 // If we *were* in 24 BPP mode, how would you convert CRY to RGB24? Seems to me
1229 // that if you're in CRY mode then you wouldn't be able to use 24 BPP bitmaps
1232 if (depth == 0) // 1 BPP
1235 WriteLog("OP: Scaled bitmap @ 1 BPP requesting FIRSTPIX!\n");
1236 // The LSB of flags is OPFLAG_REFLECT, so sign extend it and or 2 into it.
1237 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
1240 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1242 while ((int32)iwidth > 0)
1244 uint8 bits = pixels >> 63;
1246 if (flagTRANS && bits == 0)
1251 // This is the *only* correct use of endian-dependent code
1252 // (i.e., mem-to-mem direct copying)!
1253 *(uint16 *)currentLineBuffer = paletteRAM16[index | bits];
1255 *currentLineBuffer =
1256 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]),
1257 *(currentLineBuffer + 1) =
1258 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]);
1261 currentLineBuffer += lbufDelta;
1263 horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format
1264 while (horizontalRemainder & 0x80)
1266 horizontalRemainder += hscale;
1273 int phrasesToSkip = pixCount / 64, pixelShift = pixCount % 64;
1275 data += (pitch << 3) * phrasesToSkip;
1276 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1277 pixels <<= 1 * pixelShift;
1278 iwidth -= phrasesToSkip;
1279 pixCount = pixelShift;
1283 else if (depth == 1) // 2 BPP
1286 WriteLog("OP: Scaled bitmap @ 2 BPP requesting FIRSTPIX!\n");
1287 index &= 0xFC; // Top six bits form CLUT index
1288 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
1289 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
1292 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1294 while ((int32)iwidth > 0)
1296 uint8 bits = pixels >> 62;
1298 if (flagTRANS && bits == 0)
1303 // This is the *only* correct use of endian-dependent code
1304 // (i.e., mem-to-mem direct copying)!
1305 *(uint16 *)currentLineBuffer = paletteRAM16[index | bits];
1307 *currentLineBuffer =
1308 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]),
1309 *(currentLineBuffer + 1) =
1310 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]);
1313 currentLineBuffer += lbufDelta;
1315 horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format
1316 while (horizontalRemainder & 0x80)
1318 horizontalRemainder += hscale;
1325 int phrasesToSkip = pixCount / 32, pixelShift = pixCount % 32;
1327 data += (pitch << 3) * phrasesToSkip;
1328 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1329 pixels <<= 2 * pixelShift;
1330 iwidth -= phrasesToSkip;
1331 pixCount = pixelShift;
1335 else if (depth == 2) // 4 BPP
1338 WriteLog("OP: Scaled bitmap @ 4 BPP requesting FIRSTPIX!\n");
1339 index &= 0xF0; // Top four bits form CLUT index
1340 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
1341 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
1344 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1346 while ((int32)iwidth > 0)
1348 uint8 bits = pixels >> 60;
1350 if (flagTRANS && bits == 0)
1355 // This is the *only* correct use of endian-dependent code
1356 // (i.e., mem-to-mem direct copying)!
1357 *(uint16 *)currentLineBuffer = paletteRAM16[index | bits];
1359 *currentLineBuffer =
1360 BLEND_CR(*currentLineBuffer, paletteRAM[(index | bits) << 1]),
1361 *(currentLineBuffer + 1) =
1362 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[((index | bits) << 1) + 1]);
1365 currentLineBuffer += lbufDelta;
1367 horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format
1368 while (horizontalRemainder & 0x80)
1370 horizontalRemainder += hscale;
1377 int phrasesToSkip = pixCount / 16, pixelShift = pixCount % 16;
1379 data += (pitch << 3) * phrasesToSkip;
1380 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1381 pixels <<= 4 * pixelShift;
1382 iwidth -= phrasesToSkip;
1383 pixCount = pixelShift;
1387 else if (depth == 3) // 8 BPP
1390 WriteLog("OP: Scaled bitmap @ 8 BPP requesting FIRSTPIX! (fp=%u)\n", firstPix);
1391 // The LSB is OPFLAG_REFLECT, so sign extend it and or 2 into it.
1392 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
1395 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1397 while ((int32)iwidth > 0)
1399 uint8 bits = pixels >> 56;
1401 if (flagTRANS && bits == 0)
1406 // This is the *only* correct use of endian-dependent code
1407 // (i.e., mem-to-mem direct copying)!
1408 *(uint16 *)currentLineBuffer = paletteRAM16[bits];
1410 *currentLineBuffer =
1411 BLEND_CR(*currentLineBuffer, paletteRAM[bits << 1]),
1412 *(currentLineBuffer + 1) =
1413 BLEND_Y(*(currentLineBuffer + 1), paletteRAM[(bits << 1) + 1]);
1416 currentLineBuffer += lbufDelta;
1418 horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format
1419 while (horizontalRemainder & 0x80)
1421 horizontalRemainder += hscale;
1428 int phrasesToSkip = pixCount / 8, pixelShift = pixCount % 8;
1430 data += (pitch << 3) * phrasesToSkip;
1431 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1432 pixels <<= 8 * pixelShift;
1433 iwidth -= phrasesToSkip;
1434 pixCount = pixelShift;
1438 else if (depth == 4) // 16 BPP
1441 WriteLog("OP: Scaled bitmap @ 16 BPP requesting FIRSTPIX!\n");
1442 // The LSB is OPFLAG_REFLECT, so sign extend it and OR 2 into it.
1443 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 5) | 0x02;
1446 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1448 while ((int32)iwidth > 0)
1450 uint8 bitsHi = pixels >> 56, bitsLo = pixels >> 48;
1452 if (flagTRANS && (bitsLo | bitsHi) == 0)
1457 *currentLineBuffer = bitsHi,
1458 *(currentLineBuffer + 1) = bitsLo;
1460 *currentLineBuffer =
1461 BLEND_CR(*currentLineBuffer, bitsHi),
1462 *(currentLineBuffer + 1) =
1463 BLEND_Y(*(currentLineBuffer + 1), bitsLo);
1466 currentLineBuffer += lbufDelta;
1468 horizontalRemainder -= 0x20; // Subtract 1.0f in [3.5] fixed point format
1469 while (horizontalRemainder & 0x80)
1471 horizontalRemainder += hscale;
1478 int phrasesToSkip = pixCount / 4, pixelShift = pixCount % 4;
1480 data += (pitch << 3) * phrasesToSkip;
1481 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1482 pixels <<= 16 * pixelShift;
1484 iwidth -= phrasesToSkip;
1486 pixCount = pixelShift;
1490 else if (depth == 5) // 24 BPP
1492 //I'm not sure that you can scale a 24 BPP bitmap properly--the JTRM seem to indicate as much.
1493 WriteLog("OP: Writing 24 BPP scaled bitmap!\n");
1495 WriteLog("OP: Scaled bitmap @ 24 BPP requesting FIRSTPIX!\n");
1496 // Not sure, but I think RMW only works with 16 BPP and below, and only in CRY mode...
1497 // The LSB is OPFLAG_REFLECT, so sign extend it and or 4 into it.
1498 int32 lbufDelta = ((int8)((flags << 7) & 0xFF) >> 4) | 0x04;
1503 uint64 pixels = ((uint64)JaguarReadLong(data, OP) << 32) | JaguarReadLong(data + 4, OP);
1504 data += pitch << 3; // Multiply pitch * 8 (optimize: precompute this value)
1506 for(int i=0; i<2; i++)
1508 uint8 bits3 = pixels >> 56, bits2 = pixels >> 48,
1509 bits1 = pixels >> 40, bits0 = pixels >> 32;
1510 // Seems to me that both of these are in the same endian, so we could cast it as
1511 // uint16 * and do straight across copies (what about 24 bpp? Treat it differently...)
1512 // This only works for the palettized modes (1 - 8 BPP), since we actually have to
1513 // copy data from memory in 16 BPP mode (or does it? Isn't this the same as the CLUT case?)
1514 // No, it isn't because we read the memory in an endian safe way--it *won't* work...
1515 if (flagTRANS && (bits3 | bits2 | bits1 | bits0) == 0)
1518 *currentLineBuffer = bits3,
1519 *(currentLineBuffer + 1) = bits2,
1520 *(currentLineBuffer + 2) = bits1,
1521 *(currentLineBuffer + 3) = bits0;
1523 currentLineBuffer += lbufDelta;
1528 /*if (depth == 3 && startPos == 13)
1531 WriteLog("OP: Writing in the margins...\n");
1532 for(int i=0; i<100*2; i+=2)
1533 // for(int i=0; i<14*2; i+=2)
1534 tom_ram_8[0x1800 + i] = 0xFF,
1535 tom_ram_8[0x1800 + i + 1] = 0xFF;
1537 // uint32 lbufAddress = 0x1800 + (!in24BPPMode ? startPos * 2 : startPos * 4);
1538 // uint8 * currentLineBuffer = &tom_ram_8[lbufAddress];