Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1901 | serge | 1 | |
2 | * Mesa 3-D graphics library |
||
3 | * Version: 3.5 |
||
4 | * |
||
5 | * Copyright (C) 1999-2001 Brian Paul All Rights Reserved. |
||
6 | * |
||
7 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
8 | * copy of this software and associated documentation files (the "Software"), |
||
9 | * to deal in the Software without restriction, including without limitation |
||
10 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
11 | * and/or sell copies of the Software, and to permit persons to whom the |
||
12 | * Software is furnished to do so, subject to the following conditions: |
||
13 | * |
||
14 | * The above copyright notice and this permission notice shall be included |
||
15 | * in all copies or substantial portions of the Software. |
||
16 | * |
||
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS |
||
18 | * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
20 | * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN |
||
21 | * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
||
22 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
||
23 | */ |
||
24 | |||
25 | |||
26 | #include "assyntax.h" |
||
27 | #include "matypes.h" |
||
28 | #include "xform_args.h" |
||
29 | |||
30 | |||
31 | |||
32 | |||
33 | |||
34 | |||
35 | |||
36 | GLOBL GLNAME( _mesa_3dnow_transform_points4_general ) |
||
37 | HIDDEN(_mesa_3dnow_transform_points4_general) |
||
38 | GLNAME( _mesa_3dnow_transform_points4_general ): |
||
39 | |||
40 | |||
41 | |||
42 | |||
43 | MOV_L ( ARG_MATRIX, ESI ) |
||
44 | MOV_L ( ARG_SOURCE, EAX ) |
||
45 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
46 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
47 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
48 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
49 | |||
50 | |||
51 | |||
52 | |||
53 | MOV_L ( ESI, ECX ) |
||
54 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
55 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
56 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
57 | |||
58 | |||
59 | JZ ( LLBL( G3TPGR_2 ) ) |
||
60 | |||
61 | |||
62 | |||
63 | |||
64 | LLBL( G3TPGR_1 ): |
||
65 | |||
66 | |||
67 | |||
68 | |||
69 | MOVQ ( REGOFF(8, EAX), MM4 ) /* x3 | x2 */ |
||
70 | |||
71 | |||
72 | PREFETCH ( REGIND(EAX) ) |
||
73 | |||
74 | |||
75 | MOVQ ( MM4, MM6 ) /* x3 | x2 */ |
||
76 | |||
77 | |||
78 | PUNPCKHDQ ( MM2, MM2 ) /* x1 | x1 */ |
||
79 | |||
80 | |||
81 | ADD_L ( CONST(16), EDX ) /* next r */ |
||
82 | |||
83 | |||
84 | MOVQ ( MM2, MM3 ) /* x1 | x1 */ |
||
85 | |||
86 | |||
87 | PUNPCKLDQ ( MM4, MM4 ) /* x2 | x2 */ |
||
88 | |||
89 | |||
90 | MOVQ ( MM4, MM5 ) /* x2 | x2 */ |
||
91 | |||
92 | |||
93 | PUNPCKHDQ ( MM6, MM6 ) /* x3 | x3 */ |
||
94 | |||
95 | |||
96 | MOVQ ( MM6, MM7 ) /* x3 | x3 */ |
||
97 | |||
98 | |||
99 | PFADD ( MM0, MM2 ) |
||
100 | |||
101 | |||
102 | PFADD ( MM1, MM3 ) |
||
103 | |||
104 | |||
105 | PFADD ( MM4, MM6 ) |
||
106 | |||
107 | |||
108 | PFADD ( MM2, MM6 ) |
||
109 | |||
110 | |||
111 | MOVQ ( MM6, REGOFF(-16, EDX) ) |
||
112 | |||
113 | |||
114 | |||
115 | |||
116 | JNZ ( LLBL( G3TPGR_1 ) ) /* cnt > 0 ? -> process next vertex */ |
||
117 | |||
118 | |||
119 | |||
120 | |||
121 | POP_L ( EDI ) |
||
122 | POP_L ( ESI ) |
||
123 | RET |
||
124 | |||
125 | |||
126 | |||
127 | |||
128 | |||
129 | GLOBL GLNAME( _mesa_3dnow_transform_points4_perspective ) |
||
130 | HIDDEN(_mesa_3dnow_transform_points4_perspective) |
||
131 | GLNAME( _mesa_3dnow_transform_points4_perspective ): |
||
132 | |||
133 | |||
134 | |||
135 | |||
136 | MOV_L ( ARG_MATRIX, ESI ) |
||
137 | MOV_L ( ARG_SOURCE, EAX ) |
||
138 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
139 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
140 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
141 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
142 | |||
143 | |||
144 | |||
145 | |||
146 | MOV_L ( ESI, ECX ) |
||
147 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
148 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
149 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
150 | |||
151 | |||
152 | JZ ( LLBL( G3TPPR_2 ) ) |
||
153 | |||
154 | |||
155 | PREFETCHW ( REGIND(EDX) ) |
||
156 | |||
157 | |||
158 | PUNPCKLDQ ( REGOFF(20, ECX), MM0 ) /* m11 | m00 */ |
||
159 | |||
160 | |||
161 | PUNPCKLDQ ( REGOFF(56, ECX), MM1 ) /* m32 | m22 */ |
||
162 | |||
163 | |||
164 | PXOR ( MM7, MM7 ) /* 0 | 0 */ |
||
165 | |||
166 | |||
167 | LLBL( G3TPPR_1 ): |
||
168 | |||
169 | |||
170 | |||
171 | |||
172 | MOVQ ( REGOFF(8, EAX), MM5 ) /* x3 | x2 */ |
||
173 | MOVD ( REGOFF(8, EAX), MM3 ) /* | x2 */ |
||
174 | |||
175 | |||
176 | PREFETCH ( REGOFF(32, EAX) ) /* hopefully stride is zero */ |
||
177 | |||
178 | |||
179 | PFMUL ( MM0, MM4 ) /* x1*m11 | x0*m00 */ |
||
180 | |||
181 | |||
182 | ADD_L ( CONST(16), EDX ) /* next r */ |
||
183 | |||
184 | |||
185 | PFSUBR ( MM7, MM3 ) /* | -x2 */ |
||
186 | |||
187 | |||
188 | PFADD ( MM4, MM5 ) /* x1*m11+x2*m21 | x0*m00+x2*m20 */ |
||
189 | |||
190 | |||
191 | MOVQ ( MM5, REGOFF(-16, EDX) ) /* write r0, r1 */ |
||
192 | |||
193 | |||
194 | DEC_L ( ESI ) /* decrement vertex counter */ |
||
195 | |||
196 | |||
197 | |||
198 | |||
199 | |||
200 | |||
201 | POP_L ( EDI ) |
||
202 | POP_L ( ESI ) |
||
203 | RET |
||
204 | |||
205 | |||
206 | |||
207 | |||
208 | |||
209 | GLOBL GLNAME( _mesa_3dnow_transform_points4_3d ) |
||
210 | HIDDEN(_mesa_3dnow_transform_points4_3d) |
||
211 | GLNAME( _mesa_3dnow_transform_points4_3d ): |
||
212 | |||
213 | |||
214 | |||
215 | |||
216 | MOV_L ( ARG_MATRIX, ESI ) |
||
217 | MOV_L ( ARG_SOURCE, EAX ) |
||
218 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
219 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
220 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
221 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
222 | |||
223 | |||
224 | |||
225 | |||
226 | MOV_L ( ESI, ECX ) |
||
227 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
228 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
229 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
230 | |||
231 | |||
232 | JZ ( LLBL( G3TP3R_2 ) ) |
||
233 | |||
234 | |||
235 | PUNPCKLDQ ( REGOFF(24, ECX), MM6 ) /* m6 | m2 */ |
||
236 | |||
237 | |||
238 | PUNPCKLDQ ( REGOFF(56, ECX), MM7 ) /* m14 | m10 */ |
||
239 | |||
240 | |||
241 | LLBL( G3TP3R_1 ): |
||
242 | |||
243 | |||
244 | PREFETCH ( REGOFF(32, EAX) ) /* hopefully array is tightly packed */ |
||
245 | |||
246 | |||
247 | MOVQ ( REGOFF(8, EAX), MM3 ) /* x3 | x2 */ |
||
248 | |||
249 | |||
250 | MOVQ ( MM3, MM4 ) /* x3 | x2 */ |
||
251 | |||
252 | |||
253 | MOVQ ( MM4, MM5 ) /* x3 | x2 */ |
||
254 | |||
255 | |||
256 | PUNPCKHDQ ( MM1, MM1 ) /* x1 | x1 */ |
||
257 | |||
258 | |||
259 | PUNPCKLDQ ( MM3, MM3 ) /* x2 | x2 */ |
||
260 | |||
261 | |||
262 | PUNPCKHDQ ( MM4, MM4 ) /* x3 | x3 */ |
||
263 | |||
264 | |||
265 | PFADD ( MM0, MM1 ) /* x0*m1+x1*m5 | x0*m0+x1*m4 */ |
||
266 | |||
267 | |||
268 | ADD_L ( CONST(16), EDX ) /* next r */ |
||
269 | |||
270 | |||
271 | PFADD ( MM1, MM3 ) /* x0*m1+..+x2*m9 | x0*m0+...+x2*m8 */ |
||
272 | |||
273 | |||
274 | PFADD ( MM3, MM4 ) /* r1 | r0 */ |
||
275 | |||
276 | |||
277 | MOVD ( REGOFF(12, EAX), MM0 ) /* | x3 */ |
||
278 | |||
279 | |||
280 | PFACC ( MM0, MM5 ) /* r3 | r2 */ |
||
281 | |||
282 | |||
283 | MOVQ ( MM5, REGOFF(-8, EDX) ) /* write r2, r3 */ |
||
284 | |||
285 | |||
286 | JNZ ( LLBL( G3TP3R_1 ) ) /* cnt > 0 ? -> process next vertex */ |
||
287 | |||
288 | |||
289 | |||
290 | |||
291 | POP_L ( EDI ) |
||
292 | POP_L ( ESI ) |
||
293 | RET |
||
294 | |||
295 | |||
296 | |||
297 | |||
298 | |||
299 | GLOBL GLNAME( _mesa_3dnow_transform_points4_3d_no_rot ) |
||
300 | HIDDEN(_mesa_3dnow_transform_points4_3d_no_rot) |
||
301 | GLNAME( _mesa_3dnow_transform_points4_3d_no_rot ): |
||
302 | |||
303 | |||
304 | MOV_L ( ARG_DEST, ECX ) |
||
305 | MOV_L ( ARG_MATRIX, ESI ) |
||
306 | MOV_L ( ARG_SOURCE, EAX ) |
||
307 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
308 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
309 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
310 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
311 | |||
312 | |||
313 | |||
314 | |||
315 | MOV_L ( ESI, ECX ) |
||
316 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
317 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
318 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
319 | |||
320 | |||
321 | JZ ( LLBL( G3TP3NRR_2 ) ) |
||
322 | |||
323 | |||
324 | PUNPCKLDQ ( REGOFF(20, ECX), MM0 ) /* m11 | m00 */ |
||
325 | |||
326 | |||
327 | PUNPCKLDQ ( REGOFF(56, ECX), MM2 ) /* m32 | m22 */ |
||
328 | |||
329 | |||
330 | |||
331 | |||
332 | LLBL( G3TP3NRR_1 ): |
||
333 | |||
334 | |||
335 | |||
336 | |||
337 | MOVQ ( REGOFF(8, EAX), MM5 ) /* x3 | x2 */ |
||
338 | MOVD ( REGOFF(12, EAX), MM7 ) /* | x3 */ |
||
339 | |||
340 | |||
341 | PREFETCH ( REGOFF(32, EAX) ) /* hopefully stride is zero */ |
||
342 | |||
343 | |||
344 | PFMUL ( MM0, MM4 ) /* x1*m11 | x0*m00 */ |
||
345 | |||
346 | |||
347 | PFMUL ( MM2, MM5 ) /* x3*m32 | x2*m22 */ |
||
348 | |||
349 | |||
350 | PFACC ( MM7, MM5 ) /* x3 | x2*m22+x3*m32 */ |
||
351 | |||
352 | |||
353 | ADD_L ( CONST(16), EDX ) /* next r */ |
||
354 | |||
355 | |||
356 | MOVQ ( MM5, REGOFF(-8, EDX) ) /* write r2, r3 */ |
||
357 | |||
358 | |||
359 | JNZ ( LLBL( G3TP3NRR_1 ) ) /* cnt > 0 ? -> process next vertex */ |
||
360 | |||
361 | |||
362 | |||
363 | |||
364 | POP_L ( EDI ) |
||
365 | POP_L ( ESI ) |
||
366 | RET |
||
367 | |||
368 | |||
369 | |||
370 | |||
371 | |||
372 | GLOBL GLNAME( _mesa_3dnow_transform_points4_2d ) |
||
373 | HIDDEN(_mesa_3dnow_transform_points4_2d) |
||
374 | GLNAME( _mesa_3dnow_transform_points4_2d ): |
||
375 | |||
376 | |||
377 | |||
378 | |||
379 | MOV_L ( ARG_MATRIX, ESI ) |
||
380 | MOV_L ( ARG_SOURCE, EAX ) |
||
381 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
382 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
383 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
384 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
385 | |||
386 | |||
387 | |||
388 | |||
389 | MOV_L ( ESI, ECX ) |
||
390 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
391 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
392 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
393 | |||
394 | |||
395 | JZ ( LLBL( G3TP2R_2 ) ) |
||
396 | |||
397 | |||
398 | PUNPCKLDQ ( REGOFF(16, ECX), MM0 ) /* m10 | m00 */ |
||
399 | |||
400 | |||
401 | PUNPCKLDQ ( REGOFF(20, ECX), MM1 ) /* m11 | m01 */ |
||
402 | |||
403 | |||
404 | |||
405 | |||
406 | LLBL( G3TP2R_1 ): |
||
407 | |||
408 | |||
409 | |||
410 | |||
411 | MOVQ ( REGOFF(8, EAX), MM5 ) /* x3 | x2 */ |
||
412 | |||
413 | |||
414 | PREFETCH ( REGIND(EAX) ) |
||
415 | |||
416 | |||
417 | MOVQ ( MM5, MM6 ) /* x3 | x2 */ |
||
418 | |||
419 | |||
420 | PUNPCKHDQ ( MM6, MM6 ) /* x3 | x3 */ |
||
421 | |||
422 | |||
423 | ADD_L ( CONST(16), EDX ) /* next r */ |
||
424 | |||
425 | |||
426 | PFMUL ( MM2, MM6 ) /* x3*m31 | x3*m30 */ |
||
427 | |||
428 | |||
429 | MOVQ ( MM5, REGOFF(-8, EDX) ) /* write r2, r3 */ |
||
430 | |||
431 | |||
432 | |||
433 | |||
434 | JNZ ( LLBL( G3TP2R_1 ) ) /* cnt > 0 ? -> process next vertex */ |
||
435 | |||
436 | |||
437 | |||
438 | |||
439 | POP_L ( EDI ) |
||
440 | POP_L ( ESI ) |
||
441 | RET |
||
442 | |||
443 | |||
444 | |||
445 | |||
446 | |||
447 | GLOBL GLNAME( _mesa_3dnow_transform_points4_2d_no_rot ) |
||
448 | HIDDEN(_mesa_3dnow_transform_points4_2d_no_rot) |
||
449 | GLNAME( _mesa_3dnow_transform_points4_2d_no_rot ): |
||
450 | |||
451 | |||
452 | |||
453 | |||
454 | MOV_L ( ARG_MATRIX, ESI ) |
||
455 | MOV_L ( ARG_SOURCE, EAX ) |
||
456 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
457 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
458 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
459 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
460 | |||
461 | |||
462 | |||
463 | |||
464 | MOV_L ( ESI, ECX ) |
||
465 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
466 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
467 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
468 | |||
469 | |||
470 | JZ ( LLBL( G3TP2NRR_3 ) ) |
||
471 | |||
472 | |||
473 | PUNPCKLDQ ( REGOFF(20, ECX), MM0 ) /* m11 | m00 */ |
||
474 | |||
475 | |||
476 | |||
477 | |||
478 | LLBL( G3TP2NRR_2 ): |
||
479 | |||
480 | |||
481 | |||
482 | |||
483 | MOVQ ( REGOFF(8, EAX), MM5 ) /* x3 | x2 */ |
||
484 | |||
485 | |||
486 | PREFETCH ( REGIND(EAX) ) |
||
487 | |||
488 | |||
489 | MOVQ ( MM5, MM6 ) /* x3 | x2 */ |
||
490 | |||
491 | |||
492 | PUNPCKHDQ ( MM6, MM6 ) /* x3 | x3 */ |
||
493 | |||
494 | |||
495 | PFADD ( MM4, MM6 ) /* x1*m11+x3*m31 | x0*m00+x3*m30 */ |
||
496 | |||
497 | |||
498 | MOVQ ( MM5, REGOFF(-8, EDX) ) /* write r2, r3 */ |
||
499 | |||
500 | |||
501 | |||
502 | |||
503 | |||
504 | |||
505 | |||
506 | |||
507 | POP_L ( EDI ) |
||
508 | POP_L ( ESI ) |
||
509 | RET |
||
510 | |||
511 | |||
512 | |||
513 | |||
514 | |||
515 | GLOBL GLNAME( _mesa_3dnow_transform_points4_identity ) |
||
516 | HIDDEN(_mesa_3dnow_transform_points4_identity) |
||
517 | GLNAME( _mesa_3dnow_transform_points4_identity ): |
||
518 | |||
519 | |||
520 | |||
521 | |||
522 | MOV_L ( ARG_MATRIX, ESI ) |
||
523 | MOV_L ( ARG_SOURCE, EAX ) |
||
524 | MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) ) |
||
525 | OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) ) |
||
526 | MOV_L ( REGOFF(V4F_COUNT, EAX), EDX ) |
||
527 | MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) ) |
||
528 | |||
529 | |||
530 | |||
531 | |||
532 | MOV_L ( ESI, ECX ) |
||
533 | MOV_L ( REGOFF(V4F_COUNT, EAX), ESI ) |
||
534 | MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI ) |
||
535 | MOV_L ( REGOFF(V4F_START, EAX), EAX ) |
||
536 | |||
537 | |||
538 | JZ ( LLBL( G3TPIR_2 ) ) |
||
539 | |||
540 | |||
541 | LLBL( G3TPIR_1 ): |
||
542 | |||
543 | |||
544 | |||
545 | |||
546 | MOVQ ( REGOFF(8, EAX), MM1 ) /* x3 | x2 */ |
||
547 | |||
548 | |||
549 | PREFETCH ( REGIND(EAX) ) |
||
550 | |||
551 | |||
552 | MOVQ ( MM0, REGOFF(-16, EDX) ) /* r1 | r0 */ |
||
553 | |||
554 | |||
555 | |||
556 | |||
557 | JNZ ( LLBL( G3TPIR_1 ) ) /* cnt > 0 ? -> process next vertex */ |
||
558 | |||
559 | |||
560 | |||
561 | |||
562 | POP_L ( EDI ) |
||
563 | POP_L ( ESI ) |
||
564 | RET |
||
565 | #endif |
||
566 | |||
567 | |||
568 | .section .note.GNU-stack,"",%progbits |
||
569 | #endif |
||
570 |