Double-buffer generated instructions and only notify driver when the
[mesa.git] / src / mesa / tnl / t_vp_build.c
1 /*
2 * Mesa 3-D graphics library
3 * Version: 6.3
4 *
5 * Copyright (C) 2005 Tungsten Graphics 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 * TUNGSTEN GRAPHICS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
21 * WHETHER IN
22 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
23 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
24 */
25
26 /**
27 * \file t_vp_build.c
28 * Create a vertex program to execute the current fixed function T&L pipeline.
29 * \author Keith Whitwell
30 */
31
32
33 #include <strings.h>
34
35 #include "glheader.h"
36 #include "macros.h"
37 #include "enums.h"
38 #include "t_context.h"
39 #include "t_vp_build.h"
40
41 #include "shader/program.h"
42 #include "shader/nvvertprog.h"
43 #include "shader/arbvertparse.h"
44
45
46 /* Very useful debugging tool - produces annotated listing of
47 * generated program with line/function references for each
48 * instruction back into this file:
49 */
50 #define DISASSEM 0
51
52 /* Use uregs to represent registers internally, translate to Mesa's
53 * expected formats on emit.
54 *
55 * NOTE: These are passed by value extensively in this file rather
56 * than as usual by pointer reference. If this disturbs you, try
57 * remembering they are just 32bits in size.
58 *
59 * GCC is smart enough to deal with these dword-sized structures in
60 * much the same way as if I had defined them as dwords and was using
61 * macros to access and set the fields. This is much nicer and easier
62 * to evolve.
63 */
64 struct ureg {
65 GLuint file:4;
66 GLuint idx:8;
67 GLuint negate:1;
68 GLuint swz:12;
69 GLuint pad:7;
70 };
71
72
73 struct tnl_program {
74 GLcontext *ctx;
75 struct vertex_program *program;
76
77 GLuint temp_flag;
78 GLuint temp_reserved;
79
80 struct ureg eye_position;
81 struct ureg eye_position_normalized;
82 struct ureg eye_normal;
83 struct ureg identity;
84
85 GLuint materials;
86 GLuint color_materials;
87 };
88
89
90 const static struct ureg undef = {
91 ~0,
92 ~0,
93 0,
94 0,
95 0
96 };
97
98 /* Local shorthand:
99 */
100 #define X SWIZZLE_X
101 #define Y SWIZZLE_Y
102 #define Z SWIZZLE_Z
103 #define W SWIZZLE_W
104
105
106 /* Construct a ureg:
107 */
108 static struct ureg make_ureg(GLuint file, GLuint idx)
109 {
110 struct ureg reg;
111 reg.file = file;
112 reg.idx = idx;
113 reg.negate = 0;
114 reg.swz = SWIZZLE_NOOP;
115 reg.pad = 0;
116 return reg;
117 }
118
119
120
121 static struct ureg negate( struct ureg reg )
122 {
123 reg.negate ^= 1;
124 return reg;
125 }
126
127
128 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
129 {
130 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
131 GET_SWZ(reg.swz, y),
132 GET_SWZ(reg.swz, z),
133 GET_SWZ(reg.swz, w));
134
135 return reg;
136 }
137
138 static struct ureg swizzle1( struct ureg reg, int x )
139 {
140 return swizzle(reg, x, x, x, x);
141 }
142
143 static struct ureg get_temp( struct tnl_program *p )
144 {
145 int bit = ffs( ~p->temp_flag );
146 if (!bit) {
147 fprintf(stderr, "%s: out of temporaries\n", __FILE__);
148 exit(1);
149 }
150
151 p->temp_flag |= 1<<(bit-1);
152 return make_ureg(PROGRAM_TEMPORARY, bit-1);
153 }
154
155 static struct ureg reserve_temp( struct tnl_program *p )
156 {
157 struct ureg temp = get_temp( p );
158 p->temp_reserved |= 1<<temp.idx;
159 return temp;
160 }
161
162 static void release_temp( struct tnl_program *p, struct ureg reg )
163 {
164 if (reg.file == PROGRAM_TEMPORARY) {
165 p->temp_flag &= ~(1<<reg.idx);
166 p->temp_flag |= p->temp_reserved; /* can't release reserved temps */
167 }
168 }
169
170 static void release_temps( struct tnl_program *p )
171 {
172 p->temp_flag = p->temp_reserved;
173 }
174
175
176
177 static struct ureg register_input( struct tnl_program *p, GLuint input )
178 {
179 p->program->InputsRead |= (1<<input);
180 return make_ureg(PROGRAM_INPUT, input);
181 }
182
183 static struct ureg register_output( struct tnl_program *p, GLuint output )
184 {
185 p->program->OutputsWritten |= (1<<output);
186 return make_ureg(PROGRAM_OUTPUT, output);
187 }
188
189 static struct ureg register_const4f( struct tnl_program *p,
190 GLfloat s0,
191 GLfloat s1,
192 GLfloat s2,
193 GLfloat s3)
194 {
195 GLfloat values[4];
196 GLuint idx;
197 values[0] = s0;
198 values[1] = s1;
199 values[2] = s2;
200 values[3] = s3;
201 idx = _mesa_add_unnamed_constant( p->program->Parameters, values );
202 return make_ureg(PROGRAM_STATE_VAR, idx);
203 }
204
205 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
206 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
207 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
208
209 static GLboolean is_undef( struct ureg reg )
210 {
211 return reg.file == 0xf;
212 }
213
214 static struct ureg get_identity_param( struct tnl_program *p )
215 {
216 if (is_undef(p->identity))
217 p->identity = register_const4f(p, 0,0,0,1);
218
219 return p->identity;
220 }
221
222 static struct ureg register_param6( struct tnl_program *p,
223 GLint s0,
224 GLint s1,
225 GLint s2,
226 GLint s3,
227 GLint s4,
228 GLint s5)
229 {
230 GLint tokens[6];
231 GLuint idx;
232 tokens[0] = s0;
233 tokens[1] = s1;
234 tokens[2] = s2;
235 tokens[3] = s3;
236 tokens[4] = s4;
237 tokens[5] = s5;
238 idx = _mesa_add_state_reference( p->program->Parameters, tokens );
239 return make_ureg(PROGRAM_STATE_VAR, idx);
240 }
241
242
243 #define register_param1(p,s0) register_param6(p,s0,0,0,0,0,0)
244 #define register_param2(p,s0,s1) register_param6(p,s0,s1,0,0,0,0)
245 #define register_param3(p,s0,s1,s2) register_param6(p,s0,s1,s2,0,0,0)
246 #define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0)
247
248
249 static void register_matrix_param6( struct tnl_program *p,
250 GLint s0,
251 GLint s1,
252 GLint s2,
253 GLint s3,
254 GLint s4,
255 GLint s5,
256 struct ureg *matrix )
257 {
258 GLuint i;
259
260 /* This is a bit sad as the support is there to pull the whole
261 * matrix out in one go:
262 */
263 for (i = 0; i <= s4 - s3; i++)
264 matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 );
265 }
266
267
268 static void emit_arg( struct vp_src_register *src,
269 struct ureg reg )
270 {
271 src->File = reg.file;
272 src->Index = reg.idx;
273 src->Swizzle = reg.swz;
274 src->Negate = reg.negate;
275 src->RelAddr = 0;
276 src->pad = 0;
277 }
278
279 static void emit_dst( struct vp_dst_register *dst,
280 struct ureg reg, GLuint mask )
281 {
282 dst->File = reg.file;
283 dst->Index = reg.idx;
284 /* allow zero as a shorthand for xyzw */
285 dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
286 dst->pad = 0;
287 }
288
289 static void debug_insn( struct vp_instruction *inst, const char *fn,
290 GLuint line )
291 {
292 #if DISASSEM
293 static const char *last_fn;
294
295 if (fn != last_fn) {
296 last_fn = fn;
297 _mesa_printf("%s:\n", fn);
298 }
299
300 _mesa_printf("%d:\t", line);
301 _mesa_debug_vp_inst(1, inst);
302 #endif
303 }
304
305
306 static void emit_op3fn(struct tnl_program *p,
307 GLuint op,
308 struct ureg dest,
309 GLuint mask,
310 struct ureg src0,
311 struct ureg src1,
312 struct ureg src2,
313 const char *fn,
314 GLuint line)
315 {
316 GLuint nr = p->program->Base.NumInstructions++;
317 struct vp_instruction *inst = &p->program->Instructions[nr];
318
319 inst->Opcode = op;
320 inst->StringPos = 0;
321 inst->Data = 0;
322
323 emit_arg( &inst->SrcReg[0], src0 );
324 emit_arg( &inst->SrcReg[1], src1 );
325 emit_arg( &inst->SrcReg[2], src2 );
326
327 emit_dst( &inst->DstReg, dest, mask );
328
329 debug_insn(inst, fn, line);
330 }
331
332
333
334 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
335 emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
336
337 #define emit_op2(p, op, dst, mask, src0, src1) \
338 emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
339
340 #define emit_op1(p, op, dst, mask, src0) \
341 emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
342
343
344 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
345 {
346 if (reg.file == PROGRAM_TEMPORARY &&
347 !(p->temp_reserved & (1<<reg.idx)))
348 return reg;
349 else {
350 struct ureg temp = get_temp(p);
351 emit_op1(p, VP_OPCODE_MOV, temp, 0, reg);
352 return temp;
353 }
354 }
355
356
357 /* Currently no tracking performed of input/output/register size or
358 * active elements. Could be used to reduce these operations, as
359 * could the matrix type.
360 */
361 static void emit_matrix_transform_vec4( struct tnl_program *p,
362 struct ureg dest,
363 const struct ureg *mat,
364 struct ureg src)
365 {
366 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
367 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
368 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
369 emit_op2(p, VP_OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
370 }
371
372 /* This version is much easier to implement if writemasks are not
373 * supported natively on the target or (like SSE), the target doesn't
374 * have a clean/obvious dotproduct implementation.
375 */
376 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
377 struct ureg dest,
378 const struct ureg *mat,
379 struct ureg src)
380 {
381 struct ureg tmp;
382
383 if (dest.file != PROGRAM_TEMPORARY)
384 tmp = get_temp(p);
385 else
386 tmp = dest;
387
388 emit_op2(p, VP_OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
389 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
390 emit_op3(p, VP_OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
391 emit_op3(p, VP_OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
392
393 if (dest.file != PROGRAM_TEMPORARY)
394 release_temp(p, tmp);
395 }
396
397 static void emit_matrix_transform_vec3( struct tnl_program *p,
398 struct ureg dest,
399 const struct ureg *mat,
400 struct ureg src)
401 {
402 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
403 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
404 emit_op2(p, VP_OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
405 }
406
407
408 static void emit_normalize_vec3( struct tnl_program *p,
409 struct ureg dest,
410 struct ureg src )
411 {
412 struct ureg tmp = get_temp(p);
413 emit_op2(p, VP_OPCODE_DP3, tmp, 0, src, src);
414 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp);
415 emit_op2(p, VP_OPCODE_MUL, dest, 0, src, tmp);
416 release_temp(p, tmp);
417 }
418
419 static struct ureg get_eye_position( struct tnl_program *p )
420 {
421 if (is_undef(p->eye_position)) {
422 struct ureg pos = register_input( p, VERT_ATTRIB_POS );
423 struct ureg modelview[4];
424
425 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3,
426 STATE_MATRIX_TRANSPOSE, modelview );
427 p->eye_position = reserve_temp(p);
428
429 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
430 }
431
432 return p->eye_position;
433 }
434
435
436 static struct ureg get_eye_position_normalized( struct tnl_program *p )
437 {
438 if (is_undef(p->eye_position_normalized)) {
439 struct ureg eye = get_eye_position(p);
440 p->eye_position_normalized = reserve_temp(p);
441 emit_normalize_vec3(p, p->eye_position_normalized, eye);
442 }
443
444 return p->eye_position_normalized;
445 }
446
447
448 static struct ureg get_eye_normal( struct tnl_program *p )
449 {
450 if (is_undef(p->eye_normal)) {
451 struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
452 struct ureg mvinv[3];
453
454 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2,
455 STATE_MATRIX_INVTRANS, mvinv );
456
457 p->eye_normal = reserve_temp(p);
458
459 /* Transform to eye space:
460 */
461 emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
462
463 /* Normalize/Rescale:
464 */
465 if (p->ctx->Transform.Normalize) {
466 emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
467 }
468 else if (p->ctx->Transform.RescaleNormals) {
469 struct ureg rescale = register_param2(p, STATE_INTERNAL,
470 STATE_NORMAL_SCALE);
471
472 emit_op2( p, VP_OPCODE_MUL, p->eye_normal, 0, normal,
473 swizzle1(rescale, X));
474 }
475 }
476
477 return p->eye_normal;
478 }
479
480
481
482 static void build_hpos( struct tnl_program *p )
483 {
484 struct ureg pos = register_input( p, VERT_ATTRIB_POS );
485 struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
486 struct ureg mvp[4];
487
488 register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3,
489 STATE_MATRIX_TRANSPOSE, mvp );
490 emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
491 }
492
493
494 static GLuint material_attrib( GLuint side, GLuint property )
495 {
496 return (_TNL_ATTRIB_MAT_FRONT_AMBIENT +
497 (property - STATE_AMBIENT) * 2 +
498 side);
499 }
500
501 static void set_material_flags( struct tnl_program *p )
502 {
503 GLcontext *ctx = p->ctx;
504 TNLcontext *tnl = TNL_CONTEXT(ctx);
505 GLuint i;
506
507 p->color_materials = 0;
508 p->materials = 0;
509
510 if (ctx->Light.ColorMaterialEnabled) {
511 p->materials =
512 p->color_materials =
513 ctx->Light.ColorMaterialBitmask << _TNL_ATTRIB_MAT_FRONT_AMBIENT;
514 }
515
516 for (i = _TNL_ATTRIB_MAT_FRONT_AMBIENT ; i < _TNL_ATTRIB_INDEX ; i++)
517 if (tnl->vb.AttribPtr[i]->stride)
518 p->materials |= 1<<i;
519 }
520
521
522 static struct ureg get_material( struct tnl_program *p, GLuint side,
523 GLuint property )
524 {
525 GLuint attrib = material_attrib(side, property);
526
527 if (p->color_materials & (1<<attrib))
528 return register_input(p, VERT_ATTRIB_COLOR0);
529 else if (p->materials & (1<<attrib))
530 return register_input( p, attrib );
531 else
532 return register_param3( p, STATE_MATERIAL, side, property );
533 }
534
535 #define SCENE_COLOR_BITS(side) (( _TNL_BIT_MAT_FRONT_EMISSION | \
536 _TNL_BIT_MAT_FRONT_AMBIENT | \
537 _TNL_BIT_MAT_FRONT_DIFFUSE) << (side))
538
539 /* Either return a precalculated constant value or emit code to
540 * calculate these values dynamically in the case where material calls
541 * are present between begin/end pairs.
542 *
543 * Probably want to shift this to the program compilation phase - if
544 * we always emitted the calculation here, a smart compiler could
545 * detect that it was constant (given a certain set of inputs), and
546 * lift it out of the main loop. That way the programs created here
547 * would be independent of the vertex_buffer details.
548 */
549 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
550 {
551 if (p->materials & SCENE_COLOR_BITS(side)) {
552 struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
553 struct ureg material_emission = get_material(p, side, STATE_EMISSION);
554 struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
555 struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
556 struct ureg tmp = make_temp(p, material_diffuse);
557 emit_op3(p, VP_OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
558 material_ambient, material_emission);
559 return tmp;
560 }
561 else
562 return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
563 }
564
565
566 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
567 GLuint side, GLuint property )
568 {
569 GLuint attrib = material_attrib(side, property);
570 if (p->materials & (1<<attrib)) {
571 struct ureg light_value =
572 register_param3(p, STATE_LIGHT, light, property);
573 struct ureg material_value = get_material(p, side, property);
574 struct ureg tmp = get_temp(p);
575 emit_op2(p, VP_OPCODE_MUL, tmp, 0, light_value, material_value);
576 return tmp;
577 }
578 else
579 return register_param4(p, STATE_LIGHTPROD, light, side, property);
580 }
581
582 static struct ureg calculate_light_attenuation( struct tnl_program *p,
583 GLuint i,
584 struct gl_light *light,
585 struct ureg VPpli,
586 struct ureg dist )
587 {
588 struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
589 STATE_ATTENUATION);
590 struct ureg att = get_temp(p);
591
592 /* Calculate spot attenuation:
593 */
594 if (light->SpotCutoff != 180.0F) {
595 struct ureg spot_dir = register_param3(p, STATE_LIGHT, i,
596 STATE_SPOT_DIRECTION);
597 struct ureg spot = get_temp(p);
598 struct ureg slt = get_temp(p);
599
600 emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */
601 emit_op2(p, VP_OPCODE_DP3, spot, 0, negate(VPpli), spot_dir);
602 emit_op2(p, VP_OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot);
603 emit_op2(p, VP_OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
604 emit_op2(p, VP_OPCODE_MUL, att, 0, slt, spot);
605
606 release_temp(p, spot);
607 release_temp(p, slt);
608 }
609
610 /* Calculate distance attenuation:
611 */
612 if (light->ConstantAttenuation != 1.0 ||
613 light->LinearAttenuation != 1.0 ||
614 light->QuadraticAttenuation != 1.0) {
615
616 /* 1/d,d,d,1/d */
617 emit_op1(p, VP_OPCODE_RCP, dist, WRITEMASK_YZ, dist);
618 /* 1,d,d*d,1/d */
619 emit_op2(p, VP_OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
620 /* 1/dist-atten */
621 emit_op2(p, VP_OPCODE_DP3, dist, 0, attenuation, dist);
622
623 if (light->SpotCutoff != 180.0F) {
624 /* dist-atten */
625 emit_op1(p, VP_OPCODE_RCP, dist, 0, dist);
626 /* spot-atten * dist-atten */
627 emit_op2(p, VP_OPCODE_MUL, att, 0, dist, att);
628 } else {
629 /* dist-atten */
630 emit_op1(p, VP_OPCODE_RCP, att, 0, dist);
631 }
632 }
633
634 return att;
635 }
636
637
638
639
640
641 /* Need to add some addtional parameters to allow lighting in object
642 * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye
643 * space lighting.
644 */
645 static void build_lighting( struct tnl_program *p )
646 {
647 GLcontext *ctx = p->ctx;
648 const GLboolean twoside = ctx->Light.Model.TwoSide;
649 const GLboolean separate = (ctx->Light.Model.ColorControl ==
650 GL_SEPARATE_SPECULAR_COLOR);
651 GLuint nr_lights = 0, count = 0;
652 struct ureg normal = get_eye_normal(p);
653 struct ureg lit = get_temp(p);
654 struct ureg dots = get_temp(p);
655 struct ureg _col0 = undef, _col1 = undef;
656 struct ureg _bfc0 = undef, _bfc1 = undef;
657 GLuint i;
658
659 for (i = 0; i < MAX_LIGHTS; i++)
660 if (ctx->Light.Light[i].Enabled)
661 nr_lights++;
662
663 set_material_flags(p);
664
665 {
666 struct ureg shininess = get_material(p, 0, STATE_SHININESS);
667 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
668 release_temp(p, shininess);
669
670 _col0 = make_temp(p, get_scenecolor(p, 0));
671 if (separate)
672 _col1 = make_temp(p, get_identity_param(p));
673 else
674 _col1 = _col0;
675
676 }
677
678 if (twoside) {
679 struct ureg shininess = get_material(p, 1, STATE_SHININESS);
680 emit_op1(p, VP_OPCODE_MOV, dots, WRITEMASK_Z,
681 negate(swizzle1(shininess,X)));
682 release_temp(p, shininess);
683
684 _bfc0 = make_temp(p, get_scenecolor(p, 1));
685 if (separate)
686 _bfc1 = make_temp(p, get_identity_param(p));
687 else
688 _bfc1 = _bfc0;
689 }
690
691 for (i = 0; i < MAX_LIGHTS; i++) {
692 struct gl_light *light = &ctx->Light.Light[i];
693
694 if (light->Enabled) {
695 struct ureg half = undef;
696 struct ureg att = undef, VPpli = undef;
697
698 count++;
699
700 if (light->EyePosition[3] == 0) {
701 /* Can used precomputed constants in this case.
702 * Attenuation never applies to infinite lights.
703 */
704 VPpli = register_param3(p, STATE_LIGHT, i,
705 STATE_POSITION_NORMALIZED);
706 half = register_param3(p, STATE_LIGHT, i, STATE_HALF);
707 }
708 else {
709 struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
710 STATE_POSITION);
711 struct ureg V = get_eye_position(p);
712 struct ureg dist = get_temp(p);
713
714 VPpli = get_temp(p);
715 half = get_temp(p);
716
717 /* Calulate VPpli vector
718 */
719 emit_op2(p, VP_OPCODE_SUB, VPpli, 0, Ppli, V);
720
721 /* Normalize VPpli. The dist value also used in
722 * attenuation below.
723 */
724 emit_op2(p, VP_OPCODE_DP3, dist, 0, VPpli, VPpli);
725 emit_op1(p, VP_OPCODE_RSQ, dist, 0, dist);
726 emit_op2(p, VP_OPCODE_MUL, VPpli, 0, VPpli, dist);
727
728
729 /* Calculate attenuation:
730 */
731 if (light->SpotCutoff != 180.0 ||
732 light->ConstantAttenuation != 1.0 ||
733 light->LinearAttenuation != 1.0 ||
734 light->QuadraticAttenuation != 1.0) {
735 att = calculate_light_attenuation(p, i, light, VPpli, dist);
736 }
737
738
739 /* Calculate viewer direction, or use infinite viewer:
740 */
741 if (ctx->Light.Model.LocalViewer) {
742 struct ureg eye_hat = get_eye_position_normalized(p);
743 emit_op2(p, VP_OPCODE_SUB, half, 0, VPpli, eye_hat);
744 }
745 else {
746 struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
747 emit_op2(p, VP_OPCODE_ADD, half, 0, VPpli, z_dir);
748 }
749
750 emit_normalize_vec3(p, half, half);
751
752 release_temp(p, dist);
753 }
754
755 /* Calculate dot products:
756 */
757 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
758 emit_op2(p, VP_OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
759
760
761 /* Front face lighting:
762 */
763 {
764 struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
765 struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
766 struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
767 struct ureg res0, res1;
768
769 emit_op1(p, VP_OPCODE_LIT, lit, 0, dots);
770
771 if (!is_undef(att))
772 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att);
773
774
775 if (count == nr_lights) {
776 if (separate) {
777 res0 = register_output( p, VERT_RESULT_COL0 );
778 res1 = register_output( p, VERT_RESULT_COL1 );
779 }
780 else {
781 res0 = _col0;
782 res1 = register_output( p, VERT_RESULT_COL0 );
783 }
784 } else {
785 res0 = _col0;
786 res1 = _col1;
787 }
788
789 emit_op3(p, VP_OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
790 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _col0);
791 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _col1);
792
793 release_temp(p, ambient);
794 release_temp(p, diffuse);
795 release_temp(p, specular);
796 }
797
798 /* Back face lighting:
799 */
800 if (twoside) {
801 struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
802 struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
803 struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
804 struct ureg res0, res1;
805
806 emit_op1(p, VP_OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
807
808 if (!is_undef(att))
809 emit_op2(p, VP_OPCODE_MUL, lit, 0, lit, att);
810
811 if (count == nr_lights) {
812 if (separate) {
813 res0 = register_output( p, VERT_RESULT_BFC0 );
814 res1 = register_output( p, VERT_RESULT_BFC1 );
815 }
816 else {
817 res0 = _bfc0;
818 res1 = register_output( p, VERT_RESULT_BFC0 );
819 }
820 } else {
821 res0 = _bfc0;
822 res1 = _bfc1;
823 }
824
825
826 emit_op3(p, VP_OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
827 emit_op3(p, VP_OPCODE_MAD, res0, 0, swizzle1(lit,Y), diffuse, _bfc0);
828 emit_op3(p, VP_OPCODE_MAD, res1, 0, swizzle1(lit,Z), specular, _bfc1);
829
830 release_temp(p, ambient);
831 release_temp(p, diffuse);
832 release_temp(p, specular);
833 }
834
835 release_temp(p, half);
836 release_temp(p, VPpli);
837 release_temp(p, att);
838 }
839 }
840
841 release_temps( p );
842 }
843
844
845 static void build_fog( struct tnl_program *p )
846 {
847 GLcontext *ctx = p->ctx;
848 TNLcontext *tnl = TNL_CONTEXT(ctx);
849 struct ureg fog = register_output(p, VERT_RESULT_FOGC);
850 struct ureg input;
851
852 if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) {
853 input = swizzle1(get_eye_position(p), Z);
854 }
855 else {
856 input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
857 }
858
859 if (tnl->_DoVertexFog) {
860 struct ureg params = register_param1(p, STATE_FOG_PARAMS);
861 struct ureg tmp = get_temp(p);
862
863 switch (ctx->Fog.Mode) {
864 case GL_LINEAR: {
865 struct ureg id = get_identity_param(p);
866 emit_op2(p, VP_OPCODE_SUB, tmp, 0, swizzle1(params,Z), input);
867 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W));
868 emit_op2(p, VP_OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
869 emit_op2(p, VP_OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
870 break;
871 }
872 case GL_EXP:
873 emit_op1(p, VP_OPCODE_ABS, tmp, 0, input);
874 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X));
875 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X,
876 register_const1f(p, M_E), negate(tmp));
877 break;
878 case GL_EXP2:
879 emit_op2(p, VP_OPCODE_MUL, tmp, 0, input, swizzle1(params,X));
880 emit_op2(p, VP_OPCODE_MUL, tmp, 0, tmp, tmp);
881 emit_op2(p, VP_OPCODE_POW, fog, WRITEMASK_X,
882 register_const1f(p, M_E), negate(tmp));
883 break;
884 }
885
886 release_temp(p, tmp);
887 }
888 else {
889 /* results = incoming fog coords (compute fog per-fragment later)
890 *
891 * KW: Is it really necessary to do anything in this case?
892 */
893 emit_op1(p, VP_OPCODE_MOV, fog, WRITEMASK_X, input);
894 }
895 }
896
897 static void build_reflect_texgen( struct tnl_program *p,
898 struct ureg dest,
899 GLuint writemask )
900 {
901 struct ureg normal = get_eye_normal(p);
902 struct ureg eye_hat = get_eye_position_normalized(p);
903 struct ureg tmp = get_temp(p);
904
905 /* n.u */
906 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat);
907 /* 2n.u */
908 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp);
909 /* (-2n.u)n + u */
910 emit_op3(p, VP_OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
911 }
912
913 static void build_sphere_texgen( struct tnl_program *p,
914 struct ureg dest,
915 GLuint writemask )
916 {
917 struct ureg normal = get_eye_normal(p);
918 struct ureg eye_hat = get_eye_position_normalized(p);
919 struct ureg tmp = get_temp(p);
920 struct ureg half = register_const1f(p, .5);
921 struct ureg r = get_temp(p);
922 struct ureg inv_m = get_temp(p);
923 struct ureg id = get_identity_param(p);
924
925 /* Could share the above calculations, but it would be
926 * a fairly odd state for someone to set (both sphere and
927 * reflection active for different texture coordinate
928 * components. Of course - if two texture units enable
929 * reflect and/or sphere, things start to tilt in favour
930 * of seperating this out:
931 */
932
933 /* n.u */
934 emit_op2(p, VP_OPCODE_DP3, tmp, 0, normal, eye_hat);
935 /* 2n.u */
936 emit_op2(p, VP_OPCODE_ADD, tmp, 0, tmp, tmp);
937 /* (-2n.u)n + u */
938 emit_op3(p, VP_OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
939 /* r + 0,0,1 */
940 emit_op2(p, VP_OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
941 /* rx^2 + ry^2 + (rz+1)^2 */
942 emit_op2(p, VP_OPCODE_DP3, tmp, 0, tmp, tmp);
943 /* 2/m */
944 emit_op1(p, VP_OPCODE_RSQ, tmp, 0, tmp);
945 /* 1/m */
946 emit_op2(p, VP_OPCODE_MUL, inv_m, 0, tmp, swizzle1(half,X));
947 /* r/m + 1/2 */
948 emit_op3(p, VP_OPCODE_MAD, dest, writemask, r, inv_m, swizzle1(half,X));
949
950 release_temp(p, tmp);
951 release_temp(p, r);
952 release_temp(p, inv_m);
953 }
954
955
956 static void build_texture_transform( struct tnl_program *p )
957 {
958 GLcontext *ctx = p->ctx;
959 GLuint i, j;
960
961 for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
962 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
963 GLuint texmat_enabled = ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i);
964 struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
965
966 if (texUnit->TexGenEnabled || texmat_enabled) {
967 struct ureg out_texgen = undef;
968
969 if (texUnit->TexGenEnabled) {
970 GLuint copy_mask = 0;
971 GLuint sphere_mask = 0;
972 GLuint reflect_mask = 0;
973 GLuint normal_mask = 0;
974 GLuint modes[4];
975
976 if (texmat_enabled)
977 out_texgen = get_temp(p);
978 else
979 out_texgen = out;
980
981 modes[0] = texUnit->GenModeS;
982 modes[1] = texUnit->GenModeT;
983 modes[2] = texUnit->GenModeR;
984 modes[3] = texUnit->GenModeQ;
985
986 for (j = 0; j < 4; j++) {
987 if (texUnit->TexGenEnabled & (1<<j)) {
988 switch (modes[j]) {
989 case GL_OBJECT_LINEAR: {
990 struct ureg obj = register_input(p, VERT_ATTRIB_POS);
991 struct ureg plane =
992 register_param3(p, STATE_TEXGEN, i,
993 STATE_TEXGEN_OBJECT_S + j);
994
995 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j,
996 obj, plane );
997 break;
998 }
999 case GL_EYE_LINEAR: {
1000 struct ureg eye = get_eye_position(p);
1001 struct ureg plane =
1002 register_param3(p, STATE_TEXGEN, i,
1003 STATE_TEXGEN_EYE_S + j);
1004
1005 emit_op2(p, VP_OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1006 eye, plane );
1007 break;
1008 }
1009 case GL_SPHERE_MAP:
1010 sphere_mask |= WRITEMASK_X << j;
1011 break;
1012 case GL_REFLECTION_MAP_NV:
1013 reflect_mask |= WRITEMASK_X << j;
1014 break;
1015 case GL_NORMAL_MAP_NV:
1016 normal_mask |= WRITEMASK_X << j;
1017 break;
1018 }
1019 }
1020 else
1021 copy_mask |= WRITEMASK_X << j;
1022 }
1023
1024
1025 if (sphere_mask) {
1026 build_sphere_texgen(p, out_texgen, sphere_mask);
1027 }
1028
1029 if (reflect_mask) {
1030 build_reflect_texgen(p, out_texgen, reflect_mask);
1031 }
1032
1033 if (normal_mask) {
1034 struct ureg normal = get_eye_normal(p);
1035 emit_op1(p, VP_OPCODE_MOV, out_texgen, normal_mask, normal );
1036 }
1037
1038 if (copy_mask) {
1039 struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1040 emit_op1(p, VP_OPCODE_MOV, out_texgen, copy_mask, in );
1041 }
1042 }
1043
1044 if (texmat_enabled) {
1045 struct ureg texmat[4];
1046 struct ureg in = (!is_undef(out_texgen) ?
1047 out_texgen :
1048 register_input(p, VERT_ATTRIB_TEX0+i));
1049 register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i,
1050 0, 3, 0, texmat );
1051 emit_matrix_transform_vec4( p, out, texmat, in );
1052 }
1053
1054 release_temps(p);
1055 }
1056 }
1057 }
1058
1059
1060 /* Seems like it could be tighter:
1061 */
1062 static void build_pointsize( struct tnl_program *p )
1063 {
1064 struct ureg eye = get_eye_position(p);
1065 struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1066 struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1067 struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1068 struct ureg ut = get_temp(p);
1069
1070 /* 1, -Z, Z * Z, 1 */
1071 emit_op1(p, VP_OPCODE_MOV, ut, 0, swizzle1(get_identity_param(p), W));
1072 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_YZ, ut, negate(swizzle1(eye, Z)));
1073 emit_op2(p, VP_OPCODE_MUL, ut, WRITEMASK_Z, ut, negate(swizzle1(eye, Z)));
1074
1075
1076 /* p1 + p2 * dist + p3 * dist * dist, 0 */
1077 emit_op2(p, VP_OPCODE_DP3, ut, 0, ut, state_attenuation);
1078
1079 /* 1 / factor */
1080 emit_op1(p, VP_OPCODE_RCP, ut, 0, ut );
1081
1082 /* out = pointSize / factor */
1083 emit_op2(p, VP_OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1084
1085 release_temp(p, ut);
1086 }
1087
1088
1089 static void build_passthrough( struct tnl_program *p, GLuint inputs )
1090 {
1091 }
1092
1093
1094
1095 void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
1096 {
1097 TNLcontext *tnl = TNL_CONTEXT(ctx);
1098 struct tnl_program p;
1099 GLuint db_NumInstructions;
1100 struct vp_instruction *db_Instructions;
1101
1102 if (ctx->VertexProgram._Enabled)
1103 return;
1104
1105 if (!ctx->_TnlProgram)
1106 ctx->_TnlProgram = (struct vertex_program *)
1107 ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1108
1109 memset(&p, 0, sizeof(p));
1110 p.ctx = ctx;
1111 p.program = ctx->_TnlProgram;
1112
1113 p.eye_position = undef;
1114 p.eye_position_normalized = undef;
1115 p.eye_normal = undef;
1116 p.identity = undef;
1117
1118 p.temp_flag = 0;
1119 p.temp_reserved = ~((1<<MAX_NV_VERTEX_PROGRAM_TEMPS)-1);
1120
1121 db_Instructions = p.program->Instructions;
1122 db_NumInstructions = p.program->Base.NumInstructions;
1123
1124 p.program->Instructions = MALLOC(sizeof(struct vp_instruction) * 100);
1125
1126 /* Initialize the arb_program struct */
1127 p.program->Base.String = 0;
1128 p.program->Base.NumInstructions =
1129 p.program->Base.NumTemporaries =
1130 p.program->Base.NumParameters =
1131 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1132
1133 if (p.program->Parameters)
1134 _mesa_free_parameters(p.program->Parameters);
1135 else
1136 p.program->Parameters = _mesa_new_parameter_list();
1137
1138 p.program->InputsRead = 0;
1139 p.program->OutputsWritten = 0;
1140
1141 /* Emit the program, starting with modelviewproject:
1142 */
1143 build_hpos(&p);
1144
1145 /* Lighting calculations:
1146 */
1147 if (ctx->Light.Enabled)
1148 build_lighting(&p);
1149
1150 if (ctx->Fog.Enabled)
1151 build_fog(&p);
1152
1153 if (ctx->Texture._TexGenEnabled || ctx->Texture._TexMatEnabled)
1154 build_texture_transform(&p);
1155
1156 if (ctx->Point._Attenuated)
1157 build_pointsize(&p);
1158
1159 /* Is there a need to copy inputs to outputs? The software
1160 * implementation might do this more efficiently by just assigning
1161 * the missing results to point at input arrays.
1162 */
1163 if (/* tnl->vp_copy_inputs && */
1164 (tnl->render_inputs & ~p.program->OutputsWritten)) {
1165 build_passthrough(&p, tnl->render_inputs);
1166 }
1167
1168
1169 /* Finish up:
1170 */
1171 emit_op1(&p, VP_OPCODE_END, undef, 0, undef);
1172
1173 /* Disassemble:
1174 */
1175 if (DISASSEM) {
1176 _mesa_printf ("\n");
1177 }
1178
1179
1180 /* Notify driver the fragment program has (actually) changed.
1181 */
1182 if (db_Instructions == NULL ||
1183 db_NumInstructions != p.program->Base.NumInstructions ||
1184 memcmp(db_Instructions, p.program->Instructions,
1185 db_NumInstructions * sizeof(*db_Instructions)) != 0) {
1186 _mesa_printf("new program string\n");
1187 ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1188 &p.program->Base );
1189 }
1190
1191 FREE(db_Instructions);
1192 }