Subversion Repositories Kolibri OS

Rev

Rev 4358 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed

Rev Author Line No. Line
4358 Serge 1
 
2
 * Mesa 3-D graphics library
3
 *
4
 * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
5
 *
6
 * Permission is hereby granted, free of charge, to any person obtaining a
7
 * copy of this software and associated documentation files (the "Software"),
8
 * to deal in the Software without restriction, including without limitation
9
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10
 * and/or sell copies of the Software, and to permit persons to whom the
11
 * Software is furnished to do so, subject to the following conditions:
12
 *
13
 * The above copyright notice and this permission notice shall be included
14
 * in all copies or substantial portions of the Software.
15
 *
16
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22
 * OTHER DEALINGS IN THE SOFTWARE.
23
 */
24
25
 
26
  * - insert PREFETCH instructions to avoid cache-misses !
27
  * - some more optimizations are possible...
28
  * - for 40-50% more performance in the SSE-functions, the
29
  *   data (trans-matrix, src_vert, dst_vert) needs to be 16byte aligned !
30
  */
31
32
 
33
#include "assyntax.h"
34
#include "matypes.h"
35
#include "xform_args.h"
36
37
 
38
39
 
40
#define D(i) 	REGOFF(i * 4, EDI)
41
#define M(i) 	REGOFF(i * 4, EDX)
42
43
 
44
 
45
GLOBL GLNAME(_mesa_sse_transform_points3_general)
4632 Serge 46
HIDDEN(_mesa_sse_transform_points3_general)
47
GLNAME( _mesa_sse_transform_points3_general ):
48
4358 Serge 49
 
50
    PUSH_L    ( ESI )
51
    PUSH_L    ( EDI )
52
53
 
54
    MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
55
56
 
57
    MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
58
59
 
60
    JE        ( LLBL(K_GTPGR_finish) )		/* yes -> nothing to do. */
61
62
 
63
    OR_L( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
64
65
 
66
    MOV_L( CONST(4), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
67
68
 
69
    MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
70
71
 
72
    ADD_L( EDI, ECX ) 				/* count += dest ptr */
73
74
 
75
 
76
    MOVAPS    ( REGOFF(0, EDX), XMM0 )	/* m0  | m1  | m2  | m3 */
77
    MOVAPS    ( REGOFF(16, EDX), XMM1 )	/* m4  | m5  | m6  | m7 */
78
    MOVAPS    ( REGOFF(32, EDX), XMM2 )	/* m8  | m9  | m10 | m11 */
79
    MOVAPS    ( REGOFF(48, EDX), XMM3 )	/* m12 | m13 | m14 | m15 */
80
81
 
82
 
83
LLBL(K_GTPGR_top):
84
    MOVSS     ( REGOFF(0, ESI), XMM4 )		/*    |    |    | ox */
85
    SHUFPS    ( CONST(0x0), XMM4, XMM4 )	/* ox | ox | ox | ox */
86
    MOVSS     ( REGOFF(4, ESI), XMM5 )		/*    |    |    | oy */
87
    SHUFPS    ( CONST(0x0), XMM5, XMM5 )	/* oy | oy | oy | oy */
88
    MOVSS     ( REGOFF(8, ESI), XMM6 )		/*    |    |    | oz */
89
    SHUFPS    ( CONST(0x0), XMM6, XMM6 )	/* oz | oz | oz | oz */
90
91
 
92
    MULPS     ( XMM1, XMM5 )		/* m7*oy  | m6*oy  | m5*oy | m4*oy */
93
    MULPS     ( XMM2, XMM6 )		/* m11*oz | m10*oz | m9*oz | m8*oz */
94
95
 
96
    ADDPS     ( XMM6, XMM4 )
97
    ADDPS     ( XMM3, XMM4 )
98
99
 
100
101
 
102
    ADD_L     ( CONST(16), EDI )
103
    ADD_L     ( EAX, ESI )
104
    CMP_L     ( ECX, EDI )
105
    JNE       ( LLBL(K_GTPGR_top) )
106
107
 
108
    POP_L     ( EDI )
109
    POP_L     ( ESI )
110
    RET
111
#undef FRAME_OFFSET
112
113
 
114
 
115
GLOBL GLNAME(_mesa_sse_transform_points3_identity)
4632 Serge 116
HIDDEN(_mesa_sse_transform_points3_identity)
117
GLNAME( _mesa_sse_transform_points3_identity ):
118
4358 Serge 119
 
120
    PUSH_L    ( ESI )
121
    PUSH_L    ( EDI )
122
123
 
124
    MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
125
126
 
127
128
 
129
    JZ( LLBL(K_GTPIR_finish) ) 			/* count was zero; go to finish */
130
131
 
132
    OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
133
134
 
135
    MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
136
137
 
138
    MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
139
140
 
141
    ADD_L( EDI, ECX ) 				/* count += dest ptr */
142
143
 
144
    JE( LLBL(K_GTPIR_finish) )
145
146
 
147
 
148
LLBL(K_GTPIR_top):
149
    MOVLPS    ( S(0), XMM0 )
150
    MOVLPS    ( XMM0, D(0) )
151
    MOVSS     ( S(2), XMM0 )
152
    MOVSS     ( XMM0, D(2) )
153
154
 
155
    ADD_L     ( CONST(16), EDI )
156
    ADD_L     ( EAX, ESI )
157
    CMP_L     ( ECX, EDI )
158
    JNE       ( LLBL(K_GTPIR_top) )
159
160
 
161
    POP_L     ( EDI )
162
    POP_L     ( ESI )
163
    RET
164
#undef FRAME_OFFSET
165
166
 
167
 
168
 
169
 
170
GLOBL GLNAME(_mesa_sse_transform_points3_3d_no_rot)
4632 Serge 171
HIDDEN(_mesa_sse_transform_points3_3d_no_rot)
172
GLNAME(_mesa_sse_transform_points3_3d_no_rot):
173
4358 Serge 174
 
175
    PUSH_L( ESI )
176
    PUSH_L( EDI )
177
178
 
179
    MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
180
181
 
182
 
183
    MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
184
185
 
186
    JZ( LLBL(K_GTP3DNRR_finish) ) 		/* count was zero; go to finish */
187
188
 
189
    OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
190
191
 
192
    MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
193
194
 
195
    MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
196
197
 
198
    ADD_L( EDI, ECX ) 				/* count += dest ptr */
199
200
 
201
202
 
203
    MOVSS    ( M(0), XMM1 )			/* - | - |  -  | m0  */
204
    MOVSS    ( M(5), XMM2 )			/* - | - |  -  | m5  */
205
    UNPCKLPS ( XMM2, XMM1 )			/* - | - | m5  | m0  */
206
    MOVLPS   ( M(12), XMM2 )			/* - | - | m13 | m12 */
207
    MOVSS    ( M(10), XMM3 )			/* - | - |  -  | m10 */
208
    MOVSS    ( M(14), XMM4 )			/* - | - |  -  | m14 */
209
210
 
211
LLBL(K_GTP3DNRR_top):
212
213
 
214
    MULPS    ( XMM1, XMM0 )			/* - | - | s1*m5 | s0*m0 */
215
    ADDPS    ( XMM2, XMM0 )			/* - | - | +m13  | +m12 */
216
    MOVLPS   ( XMM0, D(0) )			/* -> D(1) | -> D(0) */
217
218
 
219
    MULSS    ( XMM3, XMM0 )			/* sz*m10 */
220
    ADDSS    ( XMM4, XMM0 )			/* +m14 */
221
    MOVSS    ( XMM0, D(2) )			/* -> D(2) */
222
223
 
224
    ADD_L    ( CONST(16), EDI )
225
    ADD_L    ( EAX, ESI )
226
    CMP_L    ( ECX, EDI )
227
    JNE      ( LLBL(K_GTP3DNRR_top) )
228
229
 
230
    POP_L    ( EDI )
231
    POP_L    ( ESI )
232
    RET
233
#undef FRAME_OFFSET
234
235
 
236
 
237
 
238
GLOBL GLNAME(_mesa_sse_transform_points3_perspective)
4632 Serge 239
HIDDEN(_mesa_sse_transform_points3_perspective)
240
GLNAME(_mesa_sse_transform_points3_perspective):
241
4358 Serge 242
 
243
    PUSH_L   ( ESI )
244
    PUSH_L   ( EDI )
245
246
 
247
    MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
248
249
 
250
    MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
251
252
 
253
    JZ( LLBL(K_GTP3PR_finish) )			/* count was zero; go to finish */
254
255
 
256
    OR_L( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
257
258
 
259
    MOV_L( CONST(4), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
260
261
 
262
    MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
263
264
 
265
    ADD_L( EDI, ECX ) 				/* count += dest ptr */
266
267
 
268
    MOVSS    ( M(0), XMM1 )			/* -  | -  |  -  | m0  */
269
    MOVSS    ( M(5), XMM2 )			/* -  | -  |  -  | m5  */
270
    UNPCKLPS ( XMM2, XMM1 )			/* -  | -  | m5  | m0  */
271
    MOVLPS   ( M(8), XMM2 )			/* -  | -  | m9  | m8  */
272
    MOVSS    ( M(10), XMM3 )			/* m10 */
273
    MOVSS    ( M(14), XMM4 )			/* m14 */
274
    XORPS    ( XMM6, XMM6 )			/* 0 */
275
276
 
277
LLBL(K_GTP3PR_top):
278
    MOVLPS   ( S(0), XMM0 )			/* oy | ox */
279
    MULPS    ( XMM1, XMM0 )			/* oy*m5 | ox*m0 */
280
    MOVSS    ( S(2), XMM5 )			/* oz */
281
    SHUFPS   ( CONST(0x0), XMM5, XMM5 )		/* oz | oz */
282
    MULPS    ( XMM2, XMM5 )			/* oz*m9 | oz*m8 */
283
    ADDPS    ( XMM5, XMM0 )			/* +oy*m5 | +ox*m0 */
284
    MOVLPS   ( XMM0, D(0) )			/* ->D(1) | ->D(0) */
285
286
 
287
    MULSS    ( XMM3, XMM0 )			/* oz*m10 */
288
    ADDSS    ( XMM4, XMM0 )			/* +m14 */
289
    MOVSS    ( XMM0, D(2) )			/* ->D(2) */
290
291
 
292
    MOVSS    ( XMM6, XMM5 )			/* 0 */
293
    SUBPS    ( XMM0, XMM5 )			/* -oz */
294
    MOVSS    ( XMM5, D(3) )			/* ->D(3) */
295
296
 
297
    ADD_L( CONST(16), EDI )
298
    ADD_L( EAX, ESI )
299
    CMP_L( ECX, EDI )
300
    JNE( LLBL(K_GTP3PR_top) )
301
302
 
303
    POP_L    ( EDI )
304
    POP_L    ( ESI )
305
    RET
306
#undef FRAME_OFFSET
307
308
 
309
 
310
 
311
GLOBL GLNAME(_mesa_sse_transform_points3_2d)
4632 Serge 312
HIDDEN(_mesa_sse_transform_points3_2d)
313
GLNAME(_mesa_sse_transform_points3_2d):
314
4358 Serge 315
 
316
    PUSH_L( ESI )
317
    PUSH_L( EDI )
318
319
 
320
    MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
321
322
 
323
    MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
324
325
 
326
    JZ( LLBL(K_GTP3P2DR_finish) ) 		/* count was zero; go to finish */
327
328
 
329
    OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
330
331
 
332
    MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
333
334
 
335
    MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
336
337
 
338
    ADD_L( EDI, ECX ) 				/* count += dest ptr */
339
340
 
341
    MOVLPS( M(0), XMM0 )			/* m1  | m0 */
342
    MOVLPS( M(4), XMM1 )			/* m5  | m4 */
343
    MOVLPS( M(12), XMM2 )			/* m13 | m12 */
344
345
 
346
LLBL(K_GTP3P2DR_top):
347
    MOVSS    ( S(0), XMM3 )			/* ox */
348
    SHUFPS   ( CONST(0x0), XMM3, XMM3 )		/* ox | ox */
349
    MULPS    ( XMM0, XMM3 )			/* ox*m1 | ox*m0 */
350
    MOVSS    ( S(1), XMM4 )			/* oy */
351
    SHUFPS   ( CONST(0x0), XMM4, XMM4 )		/* oy | oy */
352
    MULPS    ( XMM1, XMM4 )			/* oy*m5 | oy*m4 */
353
354
 
355
    ADDPS    ( XMM2, XMM3 )
356
    MOVLPS   ( XMM3, D(0) )
357
358
 
359
    MOVSS    ( XMM3, D(2) )
360
361
 
362
    ADD_L    ( CONST(16), EDI )
363
    ADD_L    ( EAX, ESI )
364
    CMP_L    ( ECX, EDI )
365
    JNE      ( LLBL(K_GTP3P2DR_top) )
366
367
 
368
    POP_L    ( EDI )
369
    POP_L    ( ESI )
370
    RET
371
#undef FRAME_OFFSET
372
373
 
374
 
375
 
376
GLOBL GLNAME(_mesa_sse_transform_points3_2d_no_rot)
4632 Serge 377
HIDDEN(_mesa_sse_transform_points3_2d_no_rot)
378
GLNAME(_mesa_sse_transform_points3_2d_no_rot):
379
4358 Serge 380
 
381
	PUSH_L( ESI )
382
	PUSH_L( EDI )
383
384
 
385
	MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
386
387
 
388
	MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
389
390
 
391
	JZ( LLBL(K_GTP3P2DNRR_finish) ) 	/* count was zero; go to finish */
392
393
 
394
	OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
395
396
 
397
	MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
398
399
 
400
	MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
401
402
 
403
	ADD_L( EDI, ECX ) 			/* count += dest ptr */
404
405
 
406
	MOVSS    ( M(0), XMM1 )			/* m0 */
407
	MOVSS    ( M(5), XMM2 )			/* m5 */
408
	UNPCKLPS ( XMM2, XMM1 )			/* m5 | m0 */
409
	MOVLPS   ( M(12), XMM2 )		/* m13 | m12 */
410
411
 
412
LLBL(K_GTP3P2DNRR_top):
413
	MOVLPS( S(0), XMM0 )			/* oy | ox */
414
	MULPS( XMM1, XMM0 )			/* oy*m5 | ox*m0 */
415
	ADDPS( XMM2, XMM0 )			/* +m13 | +m12 */
416
	MOVLPS( XMM0, D(0) )			/* ->D(1) | ->D(0) */
417
418
 
419
	MOVSS( XMM0, D(2) )
420
421
 
422
	ADD_L( CONST(16), EDI )
423
	ADD_L( EAX, ESI )
424
	CMP_L( ECX, EDI )
425
	JNE( LLBL(K_GTP3P2DNRR_top) )
426
427
 
428
	POP_L( EDI )
429
	POP_L( ESI )
430
	RET
431
#undef FRAME_OFFSET
432
433
 
434
 
435
 
436
 
437
GLOBL GLNAME(_mesa_sse_transform_points3_3d)
4632 Serge 438
HIDDEN(_mesa_sse_transform_points3_3d)
439
GLNAME(_mesa_sse_transform_points3_3d):
440
4358 Serge 441
 
442
	PUSH_L( ESI )
443
	PUSH_L( EDI )
444
445
 
446
	MOV_L( REGOFF(OFFSET_DEST+8, ESP), EDI ) 	/* ptr to dest GLvector4f */
447
448
 
449
 
450
	MOV_L( REGOFF(V4F_COUNT, ESI), ECX ) 	/* source count */
451
452
 
453
	JZ( LLBL(K_GTP3P3DR_finish) ) 	/* count was zero; go to finish */
454
455
 
456
	OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) 	/* set dest flags */
457
458
 
459
	MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) ) 	/* set dest size */
460
461
 
462
	MOV_L( REGOFF(V4F_START, ESI), ESI ) 	/* ptr to first source vertex */
463
464
 
465
	ADD_L( EDI, ECX ) 			/* count += dest ptr */
466
467
 
468
 
469
	MOVAPS( M(0), XMM0 )			/* m2  | m1  | m0 */
470
	MOVAPS( M(4), XMM1 )			/* m6  | m5  | m4 */
471
	MOVAPS( M(8), XMM2 )			/* m10 | m9  | m8 */
472
	MOVAPS( M(12), XMM3 )			/* m14 | m13 | m12 */
473
474
 
475
LLBL(K_GTP3P3DR_top):
476
	MOVSS( S(0), XMM4 )
477
	SHUFPS( CONST(0x0), XMM4, XMM4 )	/* ox | ox | ox */
478
	MULPS( XMM0, XMM4 )			/* ox*m2 | ox*m1 | ox*m0 */
479
480
 
481
	SHUFPS( CONST(0x0), XMM5, XMM5 )	/* oy | oy | oy */
482
	MULPS( XMM1, XMM5 )			/* oy*m6 | oy*m5 | oy*m4 */
483
484
 
485
	SHUFPS( CONST(0x0), XMM6, XMM6 )	/* oz | oz | oz */
486
	MULPS( XMM2, XMM6 )			/* oz*m10 | oz*m9 | oz*m8 */
487
488
 
489
	ADDPS( XMM6, XMM4 )			/* + | + | + */
490
	ADDPS( XMM3, XMM4 )			/* + | + | + */
491
492
 
493
	UNPCKHPS( XMM4, XMM4 )
494
	MOVSS( XMM4, D(2) )
495
496
 
497
	ADD_L( CONST(16), EDI )
498
	ADD_L( EAX, ESI )
499
	CMP_L( ECX, EDI )
500
	JNE( LLBL(K_GTP3P3DR_top) )
501
502
 
503
	POP_L( EDI )
504
	POP_L( ESI )
505
	RET
506
#undef FRAME_OFFSET
507
#endif
508
509
 
510
	.section .note.GNU-stack,"",%progbits
511
#endif
512