Merge branch 'glsl-to-tgsi'
[mesa.git] / src / mesa / main / ff_fragment_shader.cpp
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2009 VMware, Inc. 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
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 extern "C" {
30 #include "glheader.h"
31 #include "imports.h"
32 #include "mtypes.h"
33 #include "program/program.h"
34 #include "program/prog_parameter.h"
35 #include "program/prog_cache.h"
36 #include "program/prog_instruction.h"
37 #include "program/prog_print.h"
38 #include "program/prog_statevars.h"
39 #include "program/programopt.h"
40 #include "texenvprogram.h"
41 }
42
43 /*
44 * Note on texture units:
45 *
46 * The number of texture units supported by fixed-function fragment
47 * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
48 * That's because there's a one-to-one correspondence between texture
49 * coordinates and samplers in fixed-function processing.
50 *
51 * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
52 * sets of texcoords, so is fixed-function fragment processing.
53 *
54 * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
55 */
56
57
58 struct texenvprog_cache_item
59 {
60 GLuint hash;
61 void *key;
62 struct gl_fragment_program *data;
63 struct texenvprog_cache_item *next;
64 };
65
66 static GLboolean
67 texenv_doing_secondary_color(struct gl_context *ctx)
68 {
69 if (ctx->Light.Enabled &&
70 (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR))
71 return GL_TRUE;
72
73 if (ctx->Fog.ColorSumEnabled)
74 return GL_TRUE;
75
76 return GL_FALSE;
77 }
78
79 /**
80 * Up to nine instructions per tex unit, plus fog, specular color.
81 */
82 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12)
83
84 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
85
86 struct mode_opt {
87 #ifdef __GNUC__
88 __extension__ GLubyte Source:4; /**< SRC_x */
89 __extension__ GLubyte Operand:3; /**< OPR_x */
90 #else
91 GLubyte Source; /**< SRC_x */
92 GLubyte Operand; /**< OPR_x */
93 #endif
94 };
95
96 struct state_key {
97 GLuint nr_enabled_units:8;
98 GLuint enabled_units:8;
99 GLuint separate_specular:1;
100 GLuint fog_enabled:1;
101 GLuint fog_mode:2; /**< FOG_x */
102 GLuint inputs_available:12;
103 GLuint num_draw_buffers:4;
104
105 /* NOTE: This array of structs must be last! (see "keySize" below) */
106 struct {
107 GLuint enabled:1;
108 GLuint source_index:3; /**< TEXTURE_x_INDEX */
109 GLuint shadow:1;
110 GLuint ScaleShiftRGB:2;
111 GLuint ScaleShiftA:2;
112
113 GLuint NumArgsRGB:3; /**< up to MAX_COMBINER_TERMS */
114 GLuint ModeRGB:5; /**< MODE_x */
115
116 GLuint NumArgsA:3; /**< up to MAX_COMBINER_TERMS */
117 GLuint ModeA:5; /**< MODE_x */
118
119 GLuint texture_cyl_wrap:1; /**< For gallium test/debug only */
120
121 struct mode_opt OptRGB[MAX_COMBINER_TERMS];
122 struct mode_opt OptA[MAX_COMBINER_TERMS];
123 } unit[MAX_TEXTURE_UNITS];
124 };
125
126 #define FOG_LINEAR 0
127 #define FOG_EXP 1
128 #define FOG_EXP2 2
129 #define FOG_UNKNOWN 3
130
131 static GLuint translate_fog_mode( GLenum mode )
132 {
133 switch (mode) {
134 case GL_LINEAR: return FOG_LINEAR;
135 case GL_EXP: return FOG_EXP;
136 case GL_EXP2: return FOG_EXP2;
137 default: return FOG_UNKNOWN;
138 }
139 }
140
141 #define OPR_SRC_COLOR 0
142 #define OPR_ONE_MINUS_SRC_COLOR 1
143 #define OPR_SRC_ALPHA 2
144 #define OPR_ONE_MINUS_SRC_ALPHA 3
145 #define OPR_ZERO 4
146 #define OPR_ONE 5
147 #define OPR_UNKNOWN 7
148
149 static GLuint translate_operand( GLenum operand )
150 {
151 switch (operand) {
152 case GL_SRC_COLOR: return OPR_SRC_COLOR;
153 case GL_ONE_MINUS_SRC_COLOR: return OPR_ONE_MINUS_SRC_COLOR;
154 case GL_SRC_ALPHA: return OPR_SRC_ALPHA;
155 case GL_ONE_MINUS_SRC_ALPHA: return OPR_ONE_MINUS_SRC_ALPHA;
156 case GL_ZERO: return OPR_ZERO;
157 case GL_ONE: return OPR_ONE;
158 default:
159 assert(0);
160 return OPR_UNKNOWN;
161 }
162 }
163
164 #define SRC_TEXTURE 0
165 #define SRC_TEXTURE0 1
166 #define SRC_TEXTURE1 2
167 #define SRC_TEXTURE2 3
168 #define SRC_TEXTURE3 4
169 #define SRC_TEXTURE4 5
170 #define SRC_TEXTURE5 6
171 #define SRC_TEXTURE6 7
172 #define SRC_TEXTURE7 8
173 #define SRC_CONSTANT 9
174 #define SRC_PRIMARY_COLOR 10
175 #define SRC_PREVIOUS 11
176 #define SRC_ZERO 12
177 #define SRC_UNKNOWN 15
178
179 static GLuint translate_source( GLenum src )
180 {
181 switch (src) {
182 case GL_TEXTURE: return SRC_TEXTURE;
183 case GL_TEXTURE0:
184 case GL_TEXTURE1:
185 case GL_TEXTURE2:
186 case GL_TEXTURE3:
187 case GL_TEXTURE4:
188 case GL_TEXTURE5:
189 case GL_TEXTURE6:
190 case GL_TEXTURE7: return SRC_TEXTURE0 + (src - GL_TEXTURE0);
191 case GL_CONSTANT: return SRC_CONSTANT;
192 case GL_PRIMARY_COLOR: return SRC_PRIMARY_COLOR;
193 case GL_PREVIOUS: return SRC_PREVIOUS;
194 case GL_ZERO:
195 return SRC_ZERO;
196 default:
197 assert(0);
198 return SRC_UNKNOWN;
199 }
200 }
201
202 #define MODE_REPLACE 0 /* r = a0 */
203 #define MODE_MODULATE 1 /* r = a0 * a1 */
204 #define MODE_ADD 2 /* r = a0 + a1 */
205 #define MODE_ADD_SIGNED 3 /* r = a0 + a1 - 0.5 */
206 #define MODE_INTERPOLATE 4 /* r = a0 * a2 + a1 * (1 - a2) */
207 #define MODE_SUBTRACT 5 /* r = a0 - a1 */
208 #define MODE_DOT3_RGB 6 /* r = a0 . a1 */
209 #define MODE_DOT3_RGB_EXT 7 /* r = a0 . a1 */
210 #define MODE_DOT3_RGBA 8 /* r = a0 . a1 */
211 #define MODE_DOT3_RGBA_EXT 9 /* r = a0 . a1 */
212 #define MODE_MODULATE_ADD_ATI 10 /* r = a0 * a2 + a1 */
213 #define MODE_MODULATE_SIGNED_ADD_ATI 11 /* r = a0 * a2 + a1 - 0.5 */
214 #define MODE_MODULATE_SUBTRACT_ATI 12 /* r = a0 * a2 - a1 */
215 #define MODE_ADD_PRODUCTS 13 /* r = a0 * a1 + a2 * a3 */
216 #define MODE_ADD_PRODUCTS_SIGNED 14 /* r = a0 * a1 + a2 * a3 - 0.5 */
217 #define MODE_BUMP_ENVMAP_ATI 15 /* special */
218 #define MODE_UNKNOWN 16
219
220 /**
221 * Translate GL combiner state into a MODE_x value
222 */
223 static GLuint translate_mode( GLenum envMode, GLenum mode )
224 {
225 switch (mode) {
226 case GL_REPLACE: return MODE_REPLACE;
227 case GL_MODULATE: return MODE_MODULATE;
228 case GL_ADD:
229 if (envMode == GL_COMBINE4_NV)
230 return MODE_ADD_PRODUCTS;
231 else
232 return MODE_ADD;
233 case GL_ADD_SIGNED:
234 if (envMode == GL_COMBINE4_NV)
235 return MODE_ADD_PRODUCTS_SIGNED;
236 else
237 return MODE_ADD_SIGNED;
238 case GL_INTERPOLATE: return MODE_INTERPOLATE;
239 case GL_SUBTRACT: return MODE_SUBTRACT;
240 case GL_DOT3_RGB: return MODE_DOT3_RGB;
241 case GL_DOT3_RGB_EXT: return MODE_DOT3_RGB_EXT;
242 case GL_DOT3_RGBA: return MODE_DOT3_RGBA;
243 case GL_DOT3_RGBA_EXT: return MODE_DOT3_RGBA_EXT;
244 case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI;
245 case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI;
246 case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI;
247 case GL_BUMP_ENVMAP_ATI: return MODE_BUMP_ENVMAP_ATI;
248 default:
249 assert(0);
250 return MODE_UNKNOWN;
251 }
252 }
253
254
255 /**
256 * Do we need to clamp the results of the given texture env/combine mode?
257 * If the inputs to the mode are in [0,1] we don't always have to clamp
258 * the results.
259 */
260 static GLboolean
261 need_saturate( GLuint mode )
262 {
263 switch (mode) {
264 case MODE_REPLACE:
265 case MODE_MODULATE:
266 case MODE_INTERPOLATE:
267 return GL_FALSE;
268 case MODE_ADD:
269 case MODE_ADD_SIGNED:
270 case MODE_SUBTRACT:
271 case MODE_DOT3_RGB:
272 case MODE_DOT3_RGB_EXT:
273 case MODE_DOT3_RGBA:
274 case MODE_DOT3_RGBA_EXT:
275 case MODE_MODULATE_ADD_ATI:
276 case MODE_MODULATE_SIGNED_ADD_ATI:
277 case MODE_MODULATE_SUBTRACT_ATI:
278 case MODE_ADD_PRODUCTS:
279 case MODE_ADD_PRODUCTS_SIGNED:
280 case MODE_BUMP_ENVMAP_ATI:
281 return GL_TRUE;
282 default:
283 assert(0);
284 return GL_FALSE;
285 }
286 }
287
288
289
290 /**
291 * Translate TEXTURE_x_BIT to TEXTURE_x_INDEX.
292 */
293 static GLuint translate_tex_src_bit( GLbitfield bit )
294 {
295 ASSERT(bit);
296 return _mesa_ffs(bit) - 1;
297 }
298
299
300 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
301 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
302
303 /**
304 * Identify all possible varying inputs. The fragment program will
305 * never reference non-varying inputs, but will track them via state
306 * constants instead.
307 *
308 * This function figures out all the inputs that the fragment program
309 * has access to. The bitmask is later reduced to just those which
310 * are actually referenced.
311 */
312 static GLbitfield get_fp_input_mask( struct gl_context *ctx )
313 {
314 /* _NEW_PROGRAM */
315 const GLboolean vertexShader =
316 (ctx->Shader.CurrentVertexProgram &&
317 ctx->Shader.CurrentVertexProgram->LinkStatus &&
318 ctx->Shader.CurrentVertexProgram->VertexProgram);
319 const GLboolean vertexProgram = ctx->VertexProgram._Enabled;
320 GLbitfield fp_inputs = 0x0;
321
322 if (ctx->VertexProgram._Overriden) {
323 /* Somebody's messing with the vertex program and we don't have
324 * a clue what's happening. Assume that it could be producing
325 * all possible outputs.
326 */
327 fp_inputs = ~0;
328 }
329 else if (ctx->RenderMode == GL_FEEDBACK) {
330 /* _NEW_RENDERMODE */
331 fp_inputs = (FRAG_BIT_COL0 | FRAG_BIT_TEX0);
332 }
333 else if (!(vertexProgram || vertexShader)) {
334 /* Fixed function vertex logic */
335 /* _NEW_ARRAY */
336 GLbitfield varying_inputs = ctx->varying_vp_inputs;
337
338 /* These get generated in the setup routine regardless of the
339 * vertex program:
340 */
341 /* _NEW_POINT */
342 if (ctx->Point.PointSprite)
343 varying_inputs |= FRAG_BITS_TEX_ANY;
344
345 /* First look at what values may be computed by the generated
346 * vertex program:
347 */
348 /* _NEW_LIGHT */
349 if (ctx->Light.Enabled) {
350 fp_inputs |= FRAG_BIT_COL0;
351
352 if (texenv_doing_secondary_color(ctx))
353 fp_inputs |= FRAG_BIT_COL1;
354 }
355
356 /* _NEW_TEXTURE */
357 fp_inputs |= (ctx->Texture._TexGenEnabled |
358 ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0;
359
360 /* Then look at what might be varying as a result of enabled
361 * arrays, etc:
362 */
363 if (varying_inputs & VERT_BIT_COLOR0)
364 fp_inputs |= FRAG_BIT_COL0;
365 if (varying_inputs & VERT_BIT_COLOR1)
366 fp_inputs |= FRAG_BIT_COL1;
367
368 fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0)
369 << FRAG_ATTRIB_TEX0);
370
371 }
372 else {
373 /* calculate from vp->outputs */
374 struct gl_vertex_program *vprog;
375 GLbitfield64 vp_outputs;
376
377 /* Choose GLSL vertex shader over ARB vertex program. Need this
378 * since vertex shader state validation comes after fragment state
379 * validation (see additional comments in state.c).
380 */
381 if (vertexShader)
382 vprog = ctx->Shader.CurrentVertexProgram->VertexProgram;
383 else
384 vprog = ctx->VertexProgram.Current;
385
386 vp_outputs = vprog->Base.OutputsWritten;
387
388 /* These get generated in the setup routine regardless of the
389 * vertex program:
390 */
391 /* _NEW_POINT */
392 if (ctx->Point.PointSprite)
393 vp_outputs |= FRAG_BITS_TEX_ANY;
394
395 if (vp_outputs & (1 << VERT_RESULT_COL0))
396 fp_inputs |= FRAG_BIT_COL0;
397 if (vp_outputs & (1 << VERT_RESULT_COL1))
398 fp_inputs |= FRAG_BIT_COL1;
399
400 fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY) >> VERT_RESULT_TEX0)
401 << FRAG_ATTRIB_TEX0);
402 }
403
404 return fp_inputs;
405 }
406
407
408 /**
409 * Examine current texture environment state and generate a unique
410 * key to identify it.
411 */
412 static GLuint make_state_key( struct gl_context *ctx, struct state_key *key )
413 {
414 GLuint i, j;
415 GLbitfield inputs_referenced = FRAG_BIT_COL0;
416 const GLbitfield inputs_available = get_fp_input_mask( ctx );
417 GLuint keySize;
418
419 memset(key, 0, sizeof(*key));
420
421 /* _NEW_TEXTURE */
422 for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
423 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
424 const struct gl_texture_object *texObj = texUnit->_Current;
425 const struct gl_tex_env_combine_state *comb = texUnit->_CurrentCombine;
426 GLenum format;
427
428 if (!texUnit->_ReallyEnabled || !texUnit->Enabled)
429 continue;
430
431 format = texObj->Image[0][texObj->BaseLevel]->_BaseFormat;
432
433 key->unit[i].enabled = 1;
434 key->enabled_units |= (1<<i);
435 key->nr_enabled_units = i + 1;
436 inputs_referenced |= FRAG_BIT_TEX(i);
437
438 key->unit[i].source_index =
439 translate_tex_src_bit(texUnit->_ReallyEnabled);
440
441 key->unit[i].shadow =
442 ((texObj->Sampler.CompareMode == GL_COMPARE_R_TO_TEXTURE) &&
443 ((format == GL_DEPTH_COMPONENT) ||
444 (format == GL_DEPTH_STENCIL_EXT)));
445
446 key->unit[i].NumArgsRGB = comb->_NumArgsRGB;
447 key->unit[i].NumArgsA = comb->_NumArgsA;
448
449 key->unit[i].ModeRGB =
450 translate_mode(texUnit->EnvMode, comb->ModeRGB);
451 key->unit[i].ModeA =
452 translate_mode(texUnit->EnvMode, comb->ModeA);
453
454 key->unit[i].ScaleShiftRGB = comb->ScaleShiftRGB;
455 key->unit[i].ScaleShiftA = comb->ScaleShiftA;
456
457 for (j = 0; j < MAX_COMBINER_TERMS; j++) {
458 key->unit[i].OptRGB[j].Operand = translate_operand(comb->OperandRGB[j]);
459 key->unit[i].OptA[j].Operand = translate_operand(comb->OperandA[j]);
460 key->unit[i].OptRGB[j].Source = translate_source(comb->SourceRGB[j]);
461 key->unit[i].OptA[j].Source = translate_source(comb->SourceA[j]);
462 }
463
464 if (key->unit[i].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
465 /* requires some special translation */
466 key->unit[i].NumArgsRGB = 2;
467 key->unit[i].ScaleShiftRGB = 0;
468 key->unit[i].OptRGB[0].Operand = OPR_SRC_COLOR;
469 key->unit[i].OptRGB[0].Source = SRC_TEXTURE;
470 key->unit[i].OptRGB[1].Operand = OPR_SRC_COLOR;
471 key->unit[i].OptRGB[1].Source = texUnit->BumpTarget - GL_TEXTURE0 + SRC_TEXTURE0;
472 }
473
474 /* this is a back-door for enabling cylindrical texture wrap mode */
475 if (texObj->Priority == 0.125)
476 key->unit[i].texture_cyl_wrap = 1;
477 }
478
479 /* _NEW_LIGHT | _NEW_FOG */
480 if (texenv_doing_secondary_color(ctx)) {
481 key->separate_specular = 1;
482 inputs_referenced |= FRAG_BIT_COL1;
483 }
484
485 /* _NEW_FOG */
486 if (ctx->Fog.Enabled) {
487 key->fog_enabled = 1;
488 key->fog_mode = translate_fog_mode(ctx->Fog.Mode);
489 inputs_referenced |= FRAG_BIT_FOGC; /* maybe */
490 }
491
492 /* _NEW_BUFFERS */
493 key->num_draw_buffers = ctx->DrawBuffer->_NumColorDrawBuffers;
494
495 key->inputs_available = (inputs_available & inputs_referenced);
496
497 /* compute size of state key, ignoring unused texture units */
498 keySize = sizeof(*key) - sizeof(key->unit)
499 + key->nr_enabled_units * sizeof(key->unit[0]);
500
501 return keySize;
502 }
503
504
505 /**
506 * Use uregs to represent registers internally, translate to Mesa's
507 * expected formats on emit.
508 *
509 * NOTE: These are passed by value extensively in this file rather
510 * than as usual by pointer reference. If this disturbs you, try
511 * remembering they are just 32bits in size.
512 *
513 * GCC is smart enough to deal with these dword-sized structures in
514 * much the same way as if I had defined them as dwords and was using
515 * macros to access and set the fields. This is much nicer and easier
516 * to evolve.
517 */
518 struct ureg {
519 GLuint file:4;
520 GLuint idx:8;
521 GLuint negatebase:1;
522 GLuint swz:12;
523 GLuint pad:7;
524 };
525
526 static const struct ureg undef = {
527 PROGRAM_UNDEFINED,
528 255,
529 0,
530 0,
531 0
532 };
533
534
535 /** State used to build the fragment program:
536 */
537 struct texenv_fragment_program {
538 struct gl_fragment_program *program;
539 struct state_key *state;
540
541 GLbitfield alu_temps; /**< Track texture indirections, see spec. */
542 GLbitfield temps_output; /**< Track texture indirections, see spec. */
543 GLbitfield temp_in_use; /**< Tracks temporary regs which are in use. */
544 GLboolean error;
545
546 struct ureg src_texture[MAX_TEXTURE_COORD_UNITS];
547 /* Reg containing each texture unit's sampled texture color,
548 * else undef.
549 */
550
551 struct ureg texcoord_tex[MAX_TEXTURE_COORD_UNITS];
552 /* Reg containing texcoord for a texture unit,
553 * needed for bump mapping, else undef.
554 */
555
556 struct ureg src_previous; /**< Reg containing color from previous
557 * stage. May need to be decl'd.
558 */
559
560 GLuint last_tex_stage; /**< Number of last enabled texture unit */
561
562 struct ureg half;
563 struct ureg one;
564 struct ureg zero;
565 };
566
567
568
569 static struct ureg make_ureg(GLuint file, GLuint idx)
570 {
571 struct ureg reg;
572 reg.file = file;
573 reg.idx = idx;
574 reg.negatebase = 0;
575 reg.swz = SWIZZLE_NOOP;
576 reg.pad = 0;
577 return reg;
578 }
579
580 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
581 {
582 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
583 GET_SWZ(reg.swz, y),
584 GET_SWZ(reg.swz, z),
585 GET_SWZ(reg.swz, w));
586
587 return reg;
588 }
589
590 static struct ureg swizzle1( struct ureg reg, int x )
591 {
592 return swizzle(reg, x, x, x, x);
593 }
594
595 static struct ureg negate( struct ureg reg )
596 {
597 reg.negatebase ^= 1;
598 return reg;
599 }
600
601 static GLboolean is_undef( struct ureg reg )
602 {
603 return reg.file == PROGRAM_UNDEFINED;
604 }
605
606
607 static struct ureg get_temp( struct texenv_fragment_program *p )
608 {
609 GLint bit;
610
611 /* First try and reuse temps which have been used already:
612 */
613 bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps );
614
615 /* Then any unused temporary:
616 */
617 if (!bit)
618 bit = _mesa_ffs( ~p->temp_in_use );
619
620 if (!bit) {
621 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
622 exit(1);
623 }
624
625 if ((GLuint) bit > p->program->Base.NumTemporaries)
626 p->program->Base.NumTemporaries = bit;
627
628 p->temp_in_use |= 1<<(bit-1);
629 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
630 }
631
632 static struct ureg get_tex_temp( struct texenv_fragment_program *p )
633 {
634 int bit;
635
636 /* First try to find available temp not previously used (to avoid
637 * starting a new texture indirection). According to the spec, the
638 * ~p->temps_output isn't necessary, but will keep it there for
639 * now:
640 */
641 bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output );
642
643 /* Then any unused temporary:
644 */
645 if (!bit)
646 bit = _mesa_ffs( ~p->temp_in_use );
647
648 if (!bit) {
649 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
650 exit(1);
651 }
652
653 if ((GLuint) bit > p->program->Base.NumTemporaries)
654 p->program->Base.NumTemporaries = bit;
655
656 p->temp_in_use |= 1<<(bit-1);
657 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
658 }
659
660
661 /** Mark a temp reg as being no longer allocatable. */
662 static void reserve_temp( struct texenv_fragment_program *p, struct ureg r )
663 {
664 if (r.file == PROGRAM_TEMPORARY)
665 p->temps_output |= (1 << r.idx);
666 }
667
668
669 static void release_temps(struct gl_context *ctx, struct texenv_fragment_program *p )
670 {
671 GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps;
672
673 /* KW: To support tex_env_crossbar, don't release the registers in
674 * temps_output.
675 */
676 if (max_temp >= sizeof(int) * 8)
677 p->temp_in_use = p->temps_output;
678 else
679 p->temp_in_use = ~((1<<max_temp)-1) | p->temps_output;
680 }
681
682
683 static struct ureg register_param5( struct texenv_fragment_program *p,
684 GLint s0,
685 GLint s1,
686 GLint s2,
687 GLint s3,
688 GLint s4)
689 {
690 int tokens[STATE_LENGTH];
691 GLuint idx;
692 tokens[0] = s0;
693 tokens[1] = s1;
694 tokens[2] = s2;
695 tokens[3] = s3;
696 tokens[4] = s4;
697 idx = _mesa_add_state_reference(p->program->Base.Parameters,
698 (gl_state_index *)tokens);
699 return make_ureg(PROGRAM_STATE_VAR, idx);
700 }
701
702
703 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
704 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
705 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
706 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
707
708 static GLuint frag_to_vert_attrib( GLuint attrib )
709 {
710 switch (attrib) {
711 case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0;
712 case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1;
713 default:
714 assert(attrib >= FRAG_ATTRIB_TEX0);
715 assert(attrib <= FRAG_ATTRIB_TEX7);
716 return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0;
717 }
718 }
719
720
721 static struct ureg register_input( struct texenv_fragment_program *p, GLuint input )
722 {
723 if (p->state->inputs_available & (1<<input)) {
724 p->program->Base.InputsRead |= (1 << input);
725 return make_ureg(PROGRAM_INPUT, input);
726 }
727 else {
728 GLuint idx = frag_to_vert_attrib( input );
729 return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB_MAYBE_VP_CLAMPED, idx );
730 }
731 }
732
733
734 static void emit_arg( struct prog_src_register *reg,
735 struct ureg ureg )
736 {
737 reg->File = ureg.file;
738 reg->Index = ureg.idx;
739 reg->Swizzle = ureg.swz;
740 reg->Negate = ureg.negatebase ? NEGATE_XYZW : NEGATE_NONE;
741 reg->Abs = GL_FALSE;
742 }
743
744 static void emit_dst( struct prog_dst_register *dst,
745 struct ureg ureg, GLuint mask )
746 {
747 dst->File = ureg.file;
748 dst->Index = ureg.idx;
749 dst->WriteMask = mask;
750 dst->CondMask = COND_TR; /* always pass cond test */
751 dst->CondSwizzle = SWIZZLE_NOOP;
752 }
753
754 static struct prog_instruction *
755 emit_op(struct texenv_fragment_program *p,
756 enum prog_opcode op,
757 struct ureg dest,
758 GLuint mask,
759 GLboolean saturate,
760 struct ureg src0,
761 struct ureg src1,
762 struct ureg src2 )
763 {
764 const GLuint nr = p->program->Base.NumInstructions++;
765 struct prog_instruction *inst = &p->program->Base.Instructions[nr];
766
767 assert(nr < MAX_INSTRUCTIONS);
768
769 _mesa_init_instructions(inst, 1);
770 inst->Opcode = op;
771
772 emit_arg( &inst->SrcReg[0], src0 );
773 emit_arg( &inst->SrcReg[1], src1 );
774 emit_arg( &inst->SrcReg[2], src2 );
775
776 inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF;
777
778 emit_dst( &inst->DstReg, dest, mask );
779
780 #if 0
781 /* Accounting for indirection tracking:
782 */
783 if (dest.file == PROGRAM_TEMPORARY)
784 p->temps_output |= 1 << dest.idx;
785 #endif
786
787 return inst;
788 }
789
790
791 static struct ureg emit_arith( struct texenv_fragment_program *p,
792 enum prog_opcode op,
793 struct ureg dest,
794 GLuint mask,
795 GLboolean saturate,
796 struct ureg src0,
797 struct ureg src1,
798 struct ureg src2 )
799 {
800 emit_op(p, op, dest, mask, saturate, src0, src1, src2);
801
802 /* Accounting for indirection tracking:
803 */
804 if (src0.file == PROGRAM_TEMPORARY)
805 p->alu_temps |= 1 << src0.idx;
806
807 if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY)
808 p->alu_temps |= 1 << src1.idx;
809
810 if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY)
811 p->alu_temps |= 1 << src2.idx;
812
813 if (dest.file == PROGRAM_TEMPORARY)
814 p->alu_temps |= 1 << dest.idx;
815
816 p->program->Base.NumAluInstructions++;
817 return dest;
818 }
819
820 static struct ureg emit_texld( struct texenv_fragment_program *p,
821 enum prog_opcode op,
822 struct ureg dest,
823 GLuint destmask,
824 GLuint tex_unit,
825 GLuint tex_idx,
826 GLuint tex_shadow,
827 struct ureg coord )
828 {
829 struct prog_instruction *inst = emit_op( p, op,
830 dest, destmask,
831 GL_FALSE, /* don't saturate? */
832 coord, /* arg 0? */
833 undef,
834 undef);
835
836 inst->TexSrcTarget = tex_idx;
837 inst->TexSrcUnit = tex_unit;
838 inst->TexShadow = tex_shadow;
839
840 p->program->Base.NumTexInstructions++;
841
842 /* Accounting for indirection tracking:
843 */
844 reserve_temp(p, dest);
845
846 #if 0
847 /* Is this a texture indirection?
848 */
849 if ((coord.file == PROGRAM_TEMPORARY &&
850 (p->temps_output & (1<<coord.idx))) ||
851 (dest.file == PROGRAM_TEMPORARY &&
852 (p->alu_temps & (1<<dest.idx)))) {
853 p->program->Base.NumTexIndirections++;
854 p->temps_output = 1<<coord.idx;
855 p->alu_temps = 0;
856 assert(0); /* KW: texture env crossbar */
857 }
858 #endif
859
860 return dest;
861 }
862
863
864 static struct ureg register_const4f( struct texenv_fragment_program *p,
865 GLfloat s0,
866 GLfloat s1,
867 GLfloat s2,
868 GLfloat s3)
869 {
870 GLfloat values[4];
871 GLuint idx, swizzle;
872 struct ureg r;
873 values[0] = s0;
874 values[1] = s1;
875 values[2] = s2;
876 values[3] = s3;
877 idx = _mesa_add_unnamed_constant( p->program->Base.Parameters,
878 (gl_constant_value *) values, 4,
879 &swizzle );
880 r = make_ureg(PROGRAM_CONSTANT, idx);
881 r.swz = swizzle;
882 return r;
883 }
884
885 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
886 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
887 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
888 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
889
890
891 static struct ureg get_one( struct texenv_fragment_program *p )
892 {
893 if (is_undef(p->one))
894 p->one = register_scalar_const(p, 1.0);
895 return p->one;
896 }
897
898 static struct ureg get_half( struct texenv_fragment_program *p )
899 {
900 if (is_undef(p->half))
901 p->half = register_scalar_const(p, 0.5);
902 return p->half;
903 }
904
905 static struct ureg get_zero( struct texenv_fragment_program *p )
906 {
907 if (is_undef(p->zero))
908 p->zero = register_scalar_const(p, 0.0);
909 return p->zero;
910 }
911
912
913 static void program_error( struct texenv_fragment_program *p, const char *msg )
914 {
915 _mesa_problem(NULL, "%s", msg);
916 p->error = 1;
917 }
918
919 static struct ureg get_source( struct texenv_fragment_program *p,
920 GLuint src, GLuint unit )
921 {
922 switch (src) {
923 case SRC_TEXTURE:
924 assert(!is_undef(p->src_texture[unit]));
925 return p->src_texture[unit];
926
927 case SRC_TEXTURE0:
928 case SRC_TEXTURE1:
929 case SRC_TEXTURE2:
930 case SRC_TEXTURE3:
931 case SRC_TEXTURE4:
932 case SRC_TEXTURE5:
933 case SRC_TEXTURE6:
934 case SRC_TEXTURE7:
935 assert(!is_undef(p->src_texture[src - SRC_TEXTURE0]));
936 return p->src_texture[src - SRC_TEXTURE0];
937
938 case SRC_CONSTANT:
939 return register_param2(p, STATE_TEXENV_COLOR, unit);
940
941 case SRC_PRIMARY_COLOR:
942 return register_input(p, FRAG_ATTRIB_COL0);
943
944 case SRC_ZERO:
945 return get_zero(p);
946
947 case SRC_PREVIOUS:
948 if (is_undef(p->src_previous))
949 return register_input(p, FRAG_ATTRIB_COL0);
950 else
951 return p->src_previous;
952
953 default:
954 assert(0);
955 return undef;
956 }
957 }
958
959 static struct ureg emit_combine_source( struct texenv_fragment_program *p,
960 GLuint mask,
961 GLuint unit,
962 GLuint source,
963 GLuint operand )
964 {
965 struct ureg arg, src, one;
966
967 src = get_source(p, source, unit);
968
969 switch (operand) {
970 case OPR_ONE_MINUS_SRC_COLOR:
971 /* Get unused tmp,
972 * Emit tmp = 1.0 - arg.xyzw
973 */
974 arg = get_temp( p );
975 one = get_one( p );
976 return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef);
977
978 case OPR_SRC_ALPHA:
979 if (mask == WRITEMASK_W)
980 return src;
981 else
982 return swizzle1( src, SWIZZLE_W );
983 case OPR_ONE_MINUS_SRC_ALPHA:
984 /* Get unused tmp,
985 * Emit tmp = 1.0 - arg.wwww
986 */
987 arg = get_temp(p);
988 one = get_one(p);
989 return emit_arith(p, OPCODE_SUB, arg, mask, 0,
990 one, swizzle1(src, SWIZZLE_W), undef);
991 case OPR_ZERO:
992 return get_zero(p);
993 case OPR_ONE:
994 return get_one(p);
995 case OPR_SRC_COLOR:
996 return src;
997 default:
998 assert(0);
999 return src;
1000 }
1001 }
1002
1003 /**
1004 * Check if the RGB and Alpha sources and operands match for the given
1005 * texture unit's combinder state. When the RGB and A sources and
1006 * operands match, we can emit fewer instructions.
1007 */
1008 static GLboolean args_match( const struct state_key *key, GLuint unit )
1009 {
1010 GLuint i, numArgs = key->unit[unit].NumArgsRGB;
1011
1012 for (i = 0; i < numArgs; i++) {
1013 if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
1014 return GL_FALSE;
1015
1016 switch (key->unit[unit].OptA[i].Operand) {
1017 case OPR_SRC_ALPHA:
1018 switch (key->unit[unit].OptRGB[i].Operand) {
1019 case OPR_SRC_COLOR:
1020 case OPR_SRC_ALPHA:
1021 break;
1022 default:
1023 return GL_FALSE;
1024 }
1025 break;
1026 case OPR_ONE_MINUS_SRC_ALPHA:
1027 switch (key->unit[unit].OptRGB[i].Operand) {
1028 case OPR_ONE_MINUS_SRC_COLOR:
1029 case OPR_ONE_MINUS_SRC_ALPHA:
1030 break;
1031 default:
1032 return GL_FALSE;
1033 }
1034 break;
1035 default:
1036 return GL_FALSE; /* impossible */
1037 }
1038 }
1039
1040 return GL_TRUE;
1041 }
1042
1043 static struct ureg emit_combine( struct texenv_fragment_program *p,
1044 struct ureg dest,
1045 GLuint mask,
1046 GLboolean saturate,
1047 GLuint unit,
1048 GLuint nr,
1049 GLuint mode,
1050 const struct mode_opt *opt)
1051 {
1052 struct ureg src[MAX_COMBINER_TERMS];
1053 struct ureg tmp, half;
1054 GLuint i;
1055
1056 assert(nr <= MAX_COMBINER_TERMS);
1057
1058 for (i = 0; i < nr; i++)
1059 src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand );
1060
1061 switch (mode) {
1062 case MODE_REPLACE:
1063 if (mask == WRITEMASK_XYZW && !saturate)
1064 return src[0];
1065 else
1066 return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef );
1067 case MODE_MODULATE:
1068 return emit_arith( p, OPCODE_MUL, dest, mask, saturate,
1069 src[0], src[1], undef );
1070 case MODE_ADD:
1071 return emit_arith( p, OPCODE_ADD, dest, mask, saturate,
1072 src[0], src[1], undef );
1073 case MODE_ADD_SIGNED:
1074 /* tmp = arg0 + arg1
1075 * result = tmp - .5
1076 */
1077 half = get_half(p);
1078 tmp = get_temp( p );
1079 emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef );
1080 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef );
1081 return dest;
1082 case MODE_INTERPOLATE:
1083 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
1084 */
1085 return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] );
1086
1087 case MODE_SUBTRACT:
1088 return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef );
1089
1090 case MODE_DOT3_RGBA:
1091 case MODE_DOT3_RGBA_EXT:
1092 case MODE_DOT3_RGB_EXT:
1093 case MODE_DOT3_RGB: {
1094 struct ureg tmp0 = get_temp( p );
1095 struct ureg tmp1 = get_temp( p );
1096 struct ureg neg1 = register_scalar_const(p, -1);
1097 struct ureg two = register_scalar_const(p, 2);
1098
1099 /* tmp0 = 2*src0 - 1
1100 * tmp1 = 2*src1 - 1
1101 *
1102 * dst = tmp0 dot3 tmp1
1103 */
1104 emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0,
1105 two, src[0], neg1);
1106
1107 if (memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
1108 tmp1 = tmp0;
1109 else
1110 emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0,
1111 two, src[1], neg1);
1112 emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
1113 return dest;
1114 }
1115 case MODE_MODULATE_ADD_ATI:
1116 /* Arg0 * Arg2 + Arg1 */
1117 return emit_arith( p, OPCODE_MAD, dest, mask, saturate,
1118 src[0], src[2], src[1] );
1119 case MODE_MODULATE_SIGNED_ADD_ATI: {
1120 /* Arg0 * Arg2 + Arg1 - 0.5 */
1121 struct ureg tmp0 = get_temp(p);
1122 half = get_half(p);
1123 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
1124 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1125 return dest;
1126 }
1127 case MODE_MODULATE_SUBTRACT_ATI:
1128 /* Arg0 * Arg2 - Arg1 */
1129 emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
1130 return dest;
1131 case MODE_ADD_PRODUCTS:
1132 /* Arg0 * Arg1 + Arg2 * Arg3 */
1133 {
1134 struct ureg tmp0 = get_temp(p);
1135 emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1136 emit_arith( p, OPCODE_MAD, dest, mask, saturate, src[2], src[3], tmp0 );
1137 }
1138 return dest;
1139 case MODE_ADD_PRODUCTS_SIGNED:
1140 /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */
1141 {
1142 struct ureg tmp0 = get_temp(p);
1143 half = get_half(p);
1144 emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1145 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[2], src[3], tmp0 );
1146 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1147 }
1148 return dest;
1149 case MODE_BUMP_ENVMAP_ATI:
1150 /* special - not handled here */
1151 assert(0);
1152 return src[0];
1153 default:
1154 assert(0);
1155 return src[0];
1156 }
1157 }
1158
1159
1160 /**
1161 * Generate instructions for one texture unit's env/combiner mode.
1162 */
1163 static struct ureg
1164 emit_texenv(struct texenv_fragment_program *p, GLuint unit)
1165 {
1166 const struct state_key *key = p->state;
1167 GLboolean rgb_saturate, alpha_saturate;
1168 GLuint rgb_shift, alpha_shift;
1169 struct ureg out, dest;
1170
1171 if (!key->unit[unit].enabled) {
1172 return get_source(p, SRC_PREVIOUS, 0);
1173 }
1174 if (key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1175 /* this isn't really a env stage delivering a color and handled elsewhere */
1176 return get_source(p, SRC_PREVIOUS, 0);
1177 }
1178
1179 switch (key->unit[unit].ModeRGB) {
1180 case MODE_DOT3_RGB_EXT:
1181 alpha_shift = key->unit[unit].ScaleShiftA;
1182 rgb_shift = 0;
1183 break;
1184 case MODE_DOT3_RGBA_EXT:
1185 alpha_shift = 0;
1186 rgb_shift = 0;
1187 break;
1188 default:
1189 rgb_shift = key->unit[unit].ScaleShiftRGB;
1190 alpha_shift = key->unit[unit].ScaleShiftA;
1191 break;
1192 }
1193
1194 /* If we'll do rgb/alpha shifting don't saturate in emit_combine().
1195 * We don't want to clamp twice.
1196 */
1197 if (rgb_shift)
1198 rgb_saturate = GL_FALSE; /* saturate after rgb shift */
1199 else if (need_saturate(key->unit[unit].ModeRGB))
1200 rgb_saturate = GL_TRUE;
1201 else
1202 rgb_saturate = GL_FALSE;
1203
1204 if (alpha_shift)
1205 alpha_saturate = GL_FALSE; /* saturate after alpha shift */
1206 else if (need_saturate(key->unit[unit].ModeA))
1207 alpha_saturate = GL_TRUE;
1208 else
1209 alpha_saturate = GL_FALSE;
1210
1211 /* If this is the very last calculation (and various other conditions
1212 * are met), emit directly to the color output register. Otherwise,
1213 * emit to a temporary register.
1214 */
1215 if (key->separate_specular ||
1216 unit != p->last_tex_stage ||
1217 alpha_shift ||
1218 key->num_draw_buffers != 1 ||
1219 rgb_shift)
1220 dest = get_temp( p );
1221 else
1222 dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR);
1223
1224 /* Emit the RGB and A combine ops
1225 */
1226 if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
1227 args_match(key, unit)) {
1228 out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
1229 unit,
1230 key->unit[unit].NumArgsRGB,
1231 key->unit[unit].ModeRGB,
1232 key->unit[unit].OptRGB);
1233 }
1234 else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
1235 key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
1236 out = emit_combine( p, dest, WRITEMASK_XYZW, rgb_saturate,
1237 unit,
1238 key->unit[unit].NumArgsRGB,
1239 key->unit[unit].ModeRGB,
1240 key->unit[unit].OptRGB);
1241 }
1242 else {
1243 /* Need to do something to stop from re-emitting identical
1244 * argument calculations here:
1245 */
1246 out = emit_combine( p, dest, WRITEMASK_XYZ, rgb_saturate,
1247 unit,
1248 key->unit[unit].NumArgsRGB,
1249 key->unit[unit].ModeRGB,
1250 key->unit[unit].OptRGB);
1251 out = emit_combine( p, dest, WRITEMASK_W, alpha_saturate,
1252 unit,
1253 key->unit[unit].NumArgsA,
1254 key->unit[unit].ModeA,
1255 key->unit[unit].OptA);
1256 }
1257
1258 /* Deal with the final shift:
1259 */
1260 if (alpha_shift || rgb_shift) {
1261 struct ureg shift;
1262 GLboolean saturate = GL_TRUE; /* always saturate at this point */
1263
1264 if (rgb_shift == alpha_shift) {
1265 shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
1266 }
1267 else {
1268 shift = register_const4f(p,
1269 (GLfloat)(1<<rgb_shift),
1270 (GLfloat)(1<<rgb_shift),
1271 (GLfloat)(1<<rgb_shift),
1272 (GLfloat)(1<<alpha_shift));
1273 }
1274 return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW,
1275 saturate, out, shift, undef );
1276 }
1277 else
1278 return out;
1279 }
1280
1281
1282 /**
1283 * Generate instruction for getting a texture source term.
1284 */
1285 static void load_texture( struct texenv_fragment_program *p, GLuint unit )
1286 {
1287 if (is_undef(p->src_texture[unit])) {
1288 const GLuint texTarget = p->state->unit[unit].source_index;
1289 struct ureg texcoord;
1290 struct ureg tmp = get_tex_temp( p );
1291
1292 if (is_undef(p->texcoord_tex[unit])) {
1293 texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
1294 }
1295 else {
1296 /* might want to reuse this reg for tex output actually */
1297 texcoord = p->texcoord_tex[unit];
1298 }
1299
1300 /* TODO: Use D0_MASK_XY where possible.
1301 */
1302 if (p->state->unit[unit].enabled) {
1303 GLboolean shadow = GL_FALSE;
1304
1305 if (p->state->unit[unit].shadow) {
1306 p->program->Base.ShadowSamplers |= 1 << unit;
1307 shadow = GL_TRUE;
1308 }
1309
1310 p->src_texture[unit] = emit_texld( p, OPCODE_TXP,
1311 tmp, WRITEMASK_XYZW,
1312 unit, texTarget, shadow,
1313 texcoord );
1314
1315 p->program->Base.SamplersUsed |= (1 << unit);
1316 /* This identity mapping should already be in place
1317 * (see _mesa_init_program_struct()) but let's be safe.
1318 */
1319 p->program->Base.SamplerUnits[unit] = unit;
1320 }
1321 else
1322 p->src_texture[unit] = get_zero(p);
1323
1324 if (p->state->unit[unit].texture_cyl_wrap) {
1325 /* set flag which is checked by Mesa->Gallium program translation */
1326 p->program->Base.InputFlags[0] |= PROG_PARAM_BIT_CYL_WRAP;
1327 }
1328
1329 }
1330 }
1331
1332 static GLboolean load_texenv_source( struct texenv_fragment_program *p,
1333 GLuint src, GLuint unit )
1334 {
1335 switch (src) {
1336 case SRC_TEXTURE:
1337 load_texture(p, unit);
1338 break;
1339
1340 case SRC_TEXTURE0:
1341 case SRC_TEXTURE1:
1342 case SRC_TEXTURE2:
1343 case SRC_TEXTURE3:
1344 case SRC_TEXTURE4:
1345 case SRC_TEXTURE5:
1346 case SRC_TEXTURE6:
1347 case SRC_TEXTURE7:
1348 load_texture(p, src - SRC_TEXTURE0);
1349 break;
1350
1351 default:
1352 /* not a texture src - do nothing */
1353 break;
1354 }
1355
1356 return GL_TRUE;
1357 }
1358
1359
1360 /**
1361 * Generate instructions for loading all texture source terms.
1362 */
1363 static GLboolean
1364 load_texunit_sources( struct texenv_fragment_program *p, GLuint unit )
1365 {
1366 const struct state_key *key = p->state;
1367 GLuint i;
1368
1369 for (i = 0; i < key->unit[unit].NumArgsRGB; i++) {
1370 load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit );
1371 }
1372
1373 for (i = 0; i < key->unit[unit].NumArgsA; i++) {
1374 load_texenv_source( p, key->unit[unit].OptA[i].Source, unit );
1375 }
1376
1377 return GL_TRUE;
1378 }
1379
1380 /**
1381 * Generate instructions for loading bump map textures.
1382 */
1383 static GLboolean
1384 load_texunit_bumpmap( struct texenv_fragment_program *p, GLuint unit )
1385 {
1386 const struct state_key *key = p->state;
1387 GLuint bumpedUnitNr = key->unit[unit].OptRGB[1].Source - SRC_TEXTURE0;
1388 struct ureg texcDst, bumpMapRes;
1389 struct ureg constdudvcolor = register_const4f(p, 0.0, 0.0, 0.0, 1.0);
1390 struct ureg texcSrc = register_input(p, FRAG_ATTRIB_TEX0 + bumpedUnitNr);
1391 struct ureg rotMat0 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_0, unit );
1392 struct ureg rotMat1 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_1, unit );
1393
1394 load_texenv_source( p, unit + SRC_TEXTURE0, unit );
1395
1396 bumpMapRes = get_source(p, key->unit[unit].OptRGB[0].Source, unit);
1397 texcDst = get_tex_temp( p );
1398 p->texcoord_tex[bumpedUnitNr] = texcDst;
1399
1400 /* Apply rot matrix and add coords to be available in next phase.
1401 * dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1)
1402 * note only 2 coords are affected the rest are left unchanged (mul by 0)
1403 */
1404 emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1405 swizzle1(bumpMapRes, SWIZZLE_X), rotMat0, texcSrc );
1406 emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1407 swizzle1(bumpMapRes, SWIZZLE_Y), rotMat1, texcDst );
1408
1409 /* Move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
1410 * enough to access this later, should optimize away.
1411 */
1412 emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0,
1413 constdudvcolor, undef, undef );
1414
1415 return GL_TRUE;
1416 }
1417
1418 /**
1419 * Generate a new fragment program which implements the context's
1420 * current texture env/combine mode.
1421 */
1422 static void
1423 create_new_program(struct gl_context *ctx, struct state_key *key,
1424 struct gl_fragment_program *program)
1425 {
1426 struct prog_instruction instBuffer[MAX_INSTRUCTIONS];
1427 struct texenv_fragment_program p;
1428 GLuint unit;
1429 struct ureg cf, out;
1430 int i;
1431
1432 memset(&p, 0, sizeof(p));
1433 p.state = key;
1434 p.program = program;
1435
1436 /* During code generation, use locally-allocated instruction buffer,
1437 * then alloc dynamic storage below.
1438 */
1439 p.program->Base.Instructions = instBuffer;
1440 p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
1441 p.program->Base.String = NULL;
1442 p.program->Base.NumTexIndirections = 1; /* is this right? */
1443 p.program->Base.NumTexInstructions = 0;
1444 p.program->Base.NumAluInstructions = 0;
1445 p.program->Base.NumInstructions = 0;
1446 p.program->Base.NumTemporaries = 0;
1447 p.program->Base.NumParameters = 0;
1448 p.program->Base.NumAttributes = 0;
1449 p.program->Base.NumAddressRegs = 0;
1450 p.program->Base.Parameters = _mesa_new_parameter_list();
1451 p.program->Base.InputsRead = 0x0;
1452
1453 if (key->num_draw_buffers == 1)
1454 p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR;
1455 else {
1456 for (i = 0; i < key->num_draw_buffers; i++)
1457 p.program->Base.OutputsWritten |= (1 << (FRAG_RESULT_DATA0 + i));
1458 }
1459
1460 for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) {
1461 p.src_texture[unit] = undef;
1462 p.texcoord_tex[unit] = undef;
1463 }
1464
1465 p.src_previous = undef;
1466 p.half = undef;
1467 p.zero = undef;
1468 p.one = undef;
1469
1470 p.last_tex_stage = 0;
1471 release_temps(ctx, &p);
1472
1473 if (key->enabled_units && key->num_draw_buffers) {
1474 GLboolean needbumpstage = GL_FALSE;
1475
1476 /* Zeroth pass - bump map textures first */
1477 for (unit = 0; unit < key->nr_enabled_units; unit++)
1478 if (key->unit[unit].enabled &&
1479 key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1480 needbumpstage = GL_TRUE;
1481 load_texunit_bumpmap( &p, unit );
1482 }
1483 if (needbumpstage)
1484 p.program->Base.NumTexIndirections++;
1485
1486 /* First pass - to support texture_env_crossbar, first identify
1487 * all referenced texture sources and emit texld instructions
1488 * for each:
1489 */
1490 for (unit = 0; unit < key->nr_enabled_units; unit++)
1491 if (key->unit[unit].enabled) {
1492 load_texunit_sources( &p, unit );
1493 p.last_tex_stage = unit;
1494 }
1495
1496 /* Second pass - emit combine instructions to build final color:
1497 */
1498 for (unit = 0; unit < key->nr_enabled_units; unit++)
1499 if (key->unit[unit].enabled) {
1500 p.src_previous = emit_texenv( &p, unit );
1501 reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
1502 release_temps(ctx, &p); /* release all temps */
1503 }
1504 }
1505
1506 cf = get_source( &p, SRC_PREVIOUS, 0 );
1507
1508 for (i = 0; i < key->num_draw_buffers; i++) {
1509 if (key->num_draw_buffers == 1)
1510 out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLOR );
1511 else {
1512 out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_DATA0 + i );
1513 }
1514
1515 if (key->separate_specular) {
1516 /* Emit specular add.
1517 */
1518 struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
1519 emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
1520 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef );
1521 }
1522 else if (memcmp(&cf, &out, sizeof(cf)) != 0) {
1523 /* Will wind up in here if no texture enabled or a couple of
1524 * other scenarios (GL_REPLACE for instance).
1525 */
1526 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
1527 }
1528 }
1529 /* Finish up:
1530 */
1531 emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);
1532
1533 /* Allocate final instruction array. This has to be done before calling
1534 * _mesa_append_fog_code because that function frees the Base.Instructions.
1535 * At this point, Base.Instructions points to stack data, so it's a really
1536 * bad idea to free it.
1537 */
1538 p.program->Base.Instructions
1539 = _mesa_alloc_instructions(p.program->Base.NumInstructions);
1540 if (!p.program->Base.Instructions) {
1541 _mesa_error(ctx, GL_OUT_OF_MEMORY,
1542 "generating tex env program");
1543 return;
1544 }
1545 _mesa_copy_instructions(p.program->Base.Instructions, instBuffer,
1546 p.program->Base.NumInstructions);
1547
1548 /* Append fog code. This must be done before checking the program against
1549 * the limits becuase it will potentially add some instructions.
1550 */
1551 if (key->fog_enabled) {
1552 _mesa_append_fog_code(ctx, p.program, ctx->Fog.Mode, GL_FALSE);
1553 }
1554
1555 if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
1556 program_error(&p, "Exceeded max nr indirect texture lookups");
1557
1558 if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions)
1559 program_error(&p, "Exceeded max TEX instructions");
1560
1561 if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions)
1562 program_error(&p, "Exceeded max ALU instructions");
1563
1564 ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS);
1565
1566 /* Notify driver the fragment program has (actually) changed.
1567 */
1568 if (ctx->Driver.ProgramStringNotify) {
1569 GLboolean ok = ctx->Driver.ProgramStringNotify(ctx,
1570 GL_FRAGMENT_PROGRAM_ARB,
1571 &p.program->Base);
1572 /* Driver should be able to handle any texenv programs as long as
1573 * the driver correctly reported max number of texture units correctly,
1574 * etc.
1575 */
1576 ASSERT(ok);
1577 (void) ok; /* silence unused var warning */
1578 }
1579
1580 if (DISASSEM) {
1581 _mesa_print_program(&p.program->Base);
1582 printf("\n");
1583 }
1584 }
1585
1586 extern "C" {
1587
1588 /**
1589 * Return a fragment program which implements the current
1590 * fixed-function texture, fog and color-sum operations.
1591 */
1592 struct gl_fragment_program *
1593 _mesa_get_fixed_func_fragment_program(struct gl_context *ctx)
1594 {
1595 struct gl_fragment_program *prog;
1596 struct state_key key;
1597 GLuint keySize;
1598
1599 keySize = make_state_key(ctx, &key);
1600
1601 prog = (struct gl_fragment_program *)
1602 _mesa_search_program_cache(ctx->FragmentProgram.Cache,
1603 &key, keySize);
1604
1605 if (!prog) {
1606 prog = (struct gl_fragment_program *)
1607 ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
1608
1609 create_new_program(ctx, &key, prog);
1610
1611 _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
1612 &key, keySize, &prog->Base);
1613 }
1614
1615 return prog;
1616 }
1617
1618 }