mesa: avoid generating constant vertex attributes in fixedfunc programs
[mesa.git] / src / mesa / main / texenvprogram.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * 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
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "glheader.h"
29 #include "macros.h"
30 #include "enums.h"
31 #include "shader/prog_parameter.h"
32 #include "shader/prog_cache.h"
33 #include "shader/prog_instruction.h"
34 #include "shader/prog_print.h"
35 #include "shader/prog_statevars.h"
36 #include "shader/programopt.h"
37 #include "texenvprogram.h"
38
39 /**
40 * Up to nine instructions per tex unit, plus fog, specular color.
41 */
42 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_UNITS * 9) + 12)
43
44 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
45
46 struct mode_opt {
47 GLubyte Source:4;
48 GLubyte Operand:3;
49 };
50
51 struct state_key {
52 GLuint nr_enabled_units:8;
53 GLuint enabled_units:8;
54 GLuint separate_specular:1;
55 GLuint fog_enabled:1;
56 GLuint fog_mode:2;
57 GLuint inputs_available:12;
58
59 struct {
60 GLuint enabled:1;
61 GLuint source_index:3; /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
62 GLuint ScaleShiftRGB:2;
63 GLuint ScaleShiftA:2;
64
65 GLuint NumArgsRGB:2;
66 GLuint ModeRGB:4;
67 GLuint NumArgsA:2;
68 GLuint ModeA:4;
69
70 struct mode_opt OptRGB[3];
71 struct mode_opt OptA[3];
72 } unit[8];
73 };
74
75 #define FOG_LINEAR 0
76 #define FOG_EXP 1
77 #define FOG_EXP2 2
78 #define FOG_UNKNOWN 3
79
80 static GLuint translate_fog_mode( GLenum mode )
81 {
82 switch (mode) {
83 case GL_LINEAR: return FOG_LINEAR;
84 case GL_EXP: return FOG_EXP;
85 case GL_EXP2: return FOG_EXP2;
86 default: return FOG_UNKNOWN;
87 }
88 }
89
90 #define OPR_SRC_COLOR 0
91 #define OPR_ONE_MINUS_SRC_COLOR 1
92 #define OPR_SRC_ALPHA 2
93 #define OPR_ONE_MINUS_SRC_ALPHA 3
94 #define OPR_ZERO 4
95 #define OPR_ONE 5
96 #define OPR_UNKNOWN 7
97
98 static GLuint translate_operand( GLenum operand )
99 {
100 switch (operand) {
101 case GL_SRC_COLOR: return OPR_SRC_COLOR;
102 case GL_ONE_MINUS_SRC_COLOR: return OPR_ONE_MINUS_SRC_COLOR;
103 case GL_SRC_ALPHA: return OPR_SRC_ALPHA;
104 case GL_ONE_MINUS_SRC_ALPHA: return OPR_ONE_MINUS_SRC_ALPHA;
105 case GL_ZERO: return OPR_ZERO;
106 case GL_ONE: return OPR_ONE;
107 default: return OPR_UNKNOWN;
108 }
109 }
110
111 #define SRC_TEXTURE 0
112 #define SRC_TEXTURE0 1
113 #define SRC_TEXTURE1 2
114 #define SRC_TEXTURE2 3
115 #define SRC_TEXTURE3 4
116 #define SRC_TEXTURE4 5
117 #define SRC_TEXTURE5 6
118 #define SRC_TEXTURE6 7
119 #define SRC_TEXTURE7 8
120 #define SRC_CONSTANT 9
121 #define SRC_PRIMARY_COLOR 10
122 #define SRC_PREVIOUS 11
123 #define SRC_UNKNOWN 15
124
125 static GLuint translate_source( GLenum src )
126 {
127 switch (src) {
128 case GL_TEXTURE: return SRC_TEXTURE;
129 case GL_TEXTURE0:
130 case GL_TEXTURE1:
131 case GL_TEXTURE2:
132 case GL_TEXTURE3:
133 case GL_TEXTURE4:
134 case GL_TEXTURE5:
135 case GL_TEXTURE6:
136 case GL_TEXTURE7: return SRC_TEXTURE0 + (src - GL_TEXTURE0);
137 case GL_CONSTANT: return SRC_CONSTANT;
138 case GL_PRIMARY_COLOR: return SRC_PRIMARY_COLOR;
139 case GL_PREVIOUS: return SRC_PREVIOUS;
140 default: return SRC_UNKNOWN;
141 }
142 }
143
144 #define MODE_REPLACE 0
145 #define MODE_MODULATE 1
146 #define MODE_ADD 2
147 #define MODE_ADD_SIGNED 3
148 #define MODE_INTERPOLATE 4
149 #define MODE_SUBTRACT 5
150 #define MODE_DOT3_RGB 6
151 #define MODE_DOT3_RGB_EXT 7
152 #define MODE_DOT3_RGBA 8
153 #define MODE_DOT3_RGBA_EXT 9
154 #define MODE_MODULATE_ADD_ATI 10
155 #define MODE_MODULATE_SIGNED_ADD_ATI 11
156 #define MODE_MODULATE_SUBTRACT_ATI 12
157 #define MODE_UNKNOWN 15
158
159 static GLuint translate_mode( GLenum mode )
160 {
161 switch (mode) {
162 case GL_REPLACE: return MODE_REPLACE;
163 case GL_MODULATE: return MODE_MODULATE;
164 case GL_ADD: return MODE_ADD;
165 case GL_ADD_SIGNED: return MODE_ADD_SIGNED;
166 case GL_INTERPOLATE: return MODE_INTERPOLATE;
167 case GL_SUBTRACT: return MODE_SUBTRACT;
168 case GL_DOT3_RGB: return MODE_DOT3_RGB;
169 case GL_DOT3_RGB_EXT: return MODE_DOT3_RGB_EXT;
170 case GL_DOT3_RGBA: return MODE_DOT3_RGBA;
171 case GL_DOT3_RGBA_EXT: return MODE_DOT3_RGBA_EXT;
172 case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI;
173 case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI;
174 case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI;
175 default: return MODE_UNKNOWN;
176 }
177 }
178
179 #define TEXTURE_UNKNOWN_INDEX 7
180 static GLuint translate_tex_src_bit( GLbitfield bit )
181 {
182 switch (bit) {
183 case TEXTURE_1D_BIT: return TEXTURE_1D_INDEX;
184 case TEXTURE_2D_BIT: return TEXTURE_2D_INDEX;
185 case TEXTURE_RECT_BIT: return TEXTURE_RECT_INDEX;
186 case TEXTURE_3D_BIT: return TEXTURE_3D_INDEX;
187 case TEXTURE_CUBE_BIT: return TEXTURE_CUBE_INDEX;
188 default: return TEXTURE_UNKNOWN_INDEX;
189 }
190 }
191
192 #define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0)
193 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
194
195 /* Identify all possible varying inputs. The fragment program will
196 * never reference non-varying inputs, but will track them via state
197 * constants instead.
198 *
199 * This function figures out all the inputs that the fragment program
200 * has access to. The bitmask is later reduced to just those which
201 * are actually referenced.
202 */
203 static GLuint get_fp_input_mask( GLcontext *ctx )
204 {
205 GLuint fp_inputs = 0;
206
207 if (1) {
208 GLuint varying_inputs = ctx->varying_vp_inputs;
209
210 /* First look at what values may be computed by the generated
211 * vertex program:
212 */
213 if (ctx->Light.Enabled) {
214 fp_inputs |= FRAG_BIT_COL0;
215
216 if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
217 fp_inputs |= FRAG_BIT_COL1;
218 }
219
220 fp_inputs |= (ctx->Texture._TexGenEnabled |
221 ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0;
222
223 /* Then look at what might be varying as a result of enabled
224 * arrays, etc:
225 */
226 if (varying_inputs & VERT_BIT_COLOR0) fp_inputs |= FRAG_BIT_COL0;
227 if (varying_inputs & VERT_BIT_COLOR1) fp_inputs |= FRAG_BIT_COL1;
228
229 fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0)
230 << FRAG_ATTRIB_TEX0);
231
232 }
233 else {
234 /* calculate from vp->outputs */
235 GLuint vp_outputs = 0;
236
237 if (vp_outputs & (1 << VERT_RESULT_COL0)) fp_inputs |= FRAG_BIT_COL0;
238 if (vp_outputs & (1 << VERT_RESULT_COL1)) fp_inputs |= FRAG_BIT_COL1;
239
240 fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY)
241 << VERT_RESULT_TEX0)
242 >> FRAG_ATTRIB_TEX0);
243 }
244
245 return fp_inputs;
246 }
247
248
249 /**
250 * Examine current texture environment state and generate a unique
251 * key to identify it.
252 */
253 static void make_state_key( GLcontext *ctx, struct state_key *key )
254 {
255 GLuint i, j;
256 GLuint inputs_referenced = FRAG_BIT_COL0;
257 GLuint inputs_available = get_fp_input_mask( ctx );
258
259 memset(key, 0, sizeof(*key));
260
261 for (i=0;i<MAX_TEXTURE_UNITS;i++) {
262 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
263
264 if (!texUnit->_ReallyEnabled)
265 continue;
266
267 key->unit[i].enabled = 1;
268 key->enabled_units |= (1<<i);
269 key->nr_enabled_units = i+1;
270 inputs_referenced |= FRAG_BIT_TEX(i);
271
272 key->unit[i].source_index =
273 translate_tex_src_bit(texUnit->_ReallyEnabled);
274
275 key->unit[i].NumArgsRGB = texUnit->_CurrentCombine->_NumArgsRGB;
276 key->unit[i].NumArgsA = texUnit->_CurrentCombine->_NumArgsA;
277
278 key->unit[i].ModeRGB =
279 translate_mode(texUnit->_CurrentCombine->ModeRGB);
280 key->unit[i].ModeA =
281 translate_mode(texUnit->_CurrentCombine->ModeA);
282
283 key->unit[i].ScaleShiftRGB = texUnit->_CurrentCombine->ScaleShiftRGB;
284 key->unit[i].ScaleShiftA = texUnit->_CurrentCombine->ScaleShiftA;
285
286 for (j=0;j<3;j++) {
287 key->unit[i].OptRGB[j].Operand =
288 translate_operand(texUnit->_CurrentCombine->OperandRGB[j]);
289 key->unit[i].OptA[j].Operand =
290 translate_operand(texUnit->_CurrentCombine->OperandA[j]);
291 key->unit[i].OptRGB[j].Source =
292 translate_source(texUnit->_CurrentCombine->SourceRGB[j]);
293 key->unit[i].OptA[j].Source =
294 translate_source(texUnit->_CurrentCombine->SourceA[j]);
295 }
296 }
297
298 if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
299 key->separate_specular = 1;
300 inputs_referenced |= FRAG_BIT_COL1;
301 }
302
303 if (ctx->Fog.Enabled) {
304 key->fog_enabled = 1;
305 key->fog_mode = translate_fog_mode(ctx->Fog.Mode);
306 inputs_referenced |= FRAG_BIT_FOGC; /* maybe */
307 }
308
309 key->inputs_available = (inputs_available & inputs_referenced);
310 }
311
312 /* Use uregs to represent registers internally, translate to Mesa's
313 * expected formats on emit.
314 *
315 * NOTE: These are passed by value extensively in this file rather
316 * than as usual by pointer reference. If this disturbs you, try
317 * remembering they are just 32bits in size.
318 *
319 * GCC is smart enough to deal with these dword-sized structures in
320 * much the same way as if I had defined them as dwords and was using
321 * macros to access and set the fields. This is much nicer and easier
322 * to evolve.
323 */
324 struct ureg {
325 GLuint file:4;
326 GLuint idx:8;
327 GLuint negatebase:1;
328 GLuint abs:1;
329 GLuint negateabs:1;
330 GLuint swz:12;
331 GLuint pad:5;
332 };
333
334 static const struct ureg undef = {
335 PROGRAM_UNDEFINED,
336 ~0,
337 0,
338 0,
339 0,
340 0,
341 0
342 };
343
344
345 /* State used to build the fragment program:
346 */
347 struct texenv_fragment_program {
348 struct gl_fragment_program *program;
349 GLcontext *ctx;
350 struct state_key *state;
351
352 GLbitfield alu_temps; /* Track texture indirections, see spec. */
353 GLbitfield temps_output; /* Track texture indirections, see spec. */
354 GLbitfield temp_in_use; /* Tracks temporary regs which are in use. */
355 GLboolean error;
356
357 struct ureg src_texture[MAX_TEXTURE_UNITS];
358 /* Reg containing each texture unit's sampled texture color,
359 * else undef.
360 */
361
362 struct ureg src_previous; /* Reg containing color from previous
363 * stage. May need to be decl'd.
364 */
365
366 GLuint last_tex_stage; /* Number of last enabled texture unit */
367
368 struct ureg half;
369 struct ureg one;
370 struct ureg zero;
371 };
372
373
374
375 static struct ureg make_ureg(GLuint file, GLuint idx)
376 {
377 struct ureg reg;
378 reg.file = file;
379 reg.idx = idx;
380 reg.negatebase = 0;
381 reg.abs = 0;
382 reg.negateabs = 0;
383 reg.swz = SWIZZLE_NOOP;
384 reg.pad = 0;
385 return reg;
386 }
387
388 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
389 {
390 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
391 GET_SWZ(reg.swz, y),
392 GET_SWZ(reg.swz, z),
393 GET_SWZ(reg.swz, w));
394
395 return reg;
396 }
397
398 static struct ureg swizzle1( struct ureg reg, int x )
399 {
400 return swizzle(reg, x, x, x, x);
401 }
402
403 static struct ureg negate( struct ureg reg )
404 {
405 reg.negatebase ^= 1;
406 return reg;
407 }
408
409 static GLboolean is_undef( struct ureg reg )
410 {
411 return reg.file == PROGRAM_UNDEFINED;
412 }
413
414
415 static struct ureg get_temp( struct texenv_fragment_program *p )
416 {
417 GLint bit;
418
419 /* First try and reuse temps which have been used already:
420 */
421 bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps );
422
423 /* Then any unused temporary:
424 */
425 if (!bit)
426 bit = _mesa_ffs( ~p->temp_in_use );
427
428 if (!bit) {
429 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
430 _mesa_exit(1);
431 }
432
433 if ((GLuint) bit > p->program->Base.NumTemporaries)
434 p->program->Base.NumTemporaries = bit;
435
436 p->temp_in_use |= 1<<(bit-1);
437 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
438 }
439
440 static struct ureg get_tex_temp( struct texenv_fragment_program *p )
441 {
442 int bit;
443
444 /* First try to find available temp not previously used (to avoid
445 * starting a new texture indirection). According to the spec, the
446 * ~p->temps_output isn't necessary, but will keep it there for
447 * now:
448 */
449 bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output );
450
451 /* Then any unused temporary:
452 */
453 if (!bit)
454 bit = _mesa_ffs( ~p->temp_in_use );
455
456 if (!bit) {
457 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
458 _mesa_exit(1);
459 }
460
461 if ((GLuint) bit > p->program->Base.NumTemporaries)
462 p->program->Base.NumTemporaries = bit;
463
464 p->temp_in_use |= 1<<(bit-1);
465 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
466 }
467
468
469 /** Mark a temp reg as being no longer allocatable. */
470 static void reserve_temp( struct texenv_fragment_program *p, struct ureg r )
471 {
472 if (r.file == PROGRAM_TEMPORARY)
473 p->temps_output |= (1 << r.idx);
474 }
475
476
477 static void release_temps(GLcontext *ctx, struct texenv_fragment_program *p )
478 {
479 GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps;
480
481 /* KW: To support tex_env_crossbar, don't release the registers in
482 * temps_output.
483 */
484 if (max_temp >= sizeof(int) * 8)
485 p->temp_in_use = p->temps_output;
486 else
487 p->temp_in_use = ~((1<<max_temp)-1) | p->temps_output;
488 }
489
490
491 static struct ureg register_param5( struct texenv_fragment_program *p,
492 GLint s0,
493 GLint s1,
494 GLint s2,
495 GLint s3,
496 GLint s4)
497 {
498 gl_state_index tokens[STATE_LENGTH];
499 GLuint idx;
500 tokens[0] = s0;
501 tokens[1] = s1;
502 tokens[2] = s2;
503 tokens[3] = s3;
504 tokens[4] = s4;
505 idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
506 return make_ureg(PROGRAM_STATE_VAR, idx);
507 }
508
509
510 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
511 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
512 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
513 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
514
515 static GLuint frag_to_vert_attrib( GLuint attrib )
516 {
517 switch (attrib) {
518 case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0;
519 case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1;
520 default:
521 assert(attrib >= FRAG_ATTRIB_TEX0);
522 assert(attrib <= FRAG_ATTRIB_TEX7);
523 return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0;
524 }
525 }
526
527
528 static struct ureg register_input( struct texenv_fragment_program *p, GLuint input )
529 {
530 if (p->state->inputs_available & (1<<input)) {
531 p->program->Base.InputsRead |= (1 << input);
532 return make_ureg(PROGRAM_INPUT, input);
533 }
534 else {
535 GLuint idx = frag_to_vert_attrib( input );
536 return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, idx );
537 }
538 }
539
540
541 static void emit_arg( struct prog_src_register *reg,
542 struct ureg ureg )
543 {
544 reg->File = ureg.file;
545 reg->Index = ureg.idx;
546 reg->Swizzle = ureg.swz;
547 reg->NegateBase = ureg.negatebase ? 0xf : 0x0;
548 reg->Abs = ureg.abs;
549 reg->NegateAbs = ureg.negateabs;
550 }
551
552 static void emit_dst( struct prog_dst_register *dst,
553 struct ureg ureg, GLuint mask )
554 {
555 dst->File = ureg.file;
556 dst->Index = ureg.idx;
557 dst->WriteMask = mask;
558 dst->CondMask = COND_TR; /* always pass cond test */
559 dst->CondSwizzle = SWIZZLE_NOOP;
560 }
561
562 static struct prog_instruction *
563 emit_op(struct texenv_fragment_program *p,
564 enum prog_opcode op,
565 struct ureg dest,
566 GLuint mask,
567 GLboolean saturate,
568 struct ureg src0,
569 struct ureg src1,
570 struct ureg src2 )
571 {
572 GLuint nr = p->program->Base.NumInstructions++;
573 struct prog_instruction *inst = &p->program->Base.Instructions[nr];
574
575 assert(nr < MAX_INSTRUCTIONS);
576
577 _mesa_init_instructions(inst, 1);
578 inst->Opcode = op;
579
580 emit_arg( &inst->SrcReg[0], src0 );
581 emit_arg( &inst->SrcReg[1], src1 );
582 emit_arg( &inst->SrcReg[2], src2 );
583
584 inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF;
585
586 emit_dst( &inst->DstReg, dest, mask );
587
588 #if 0
589 /* Accounting for indirection tracking:
590 */
591 if (dest.file == PROGRAM_TEMPORARY)
592 p->temps_output |= 1 << dest.idx;
593 #endif
594
595 return inst;
596 }
597
598
599 static struct ureg emit_arith( struct texenv_fragment_program *p,
600 enum prog_opcode op,
601 struct ureg dest,
602 GLuint mask,
603 GLboolean saturate,
604 struct ureg src0,
605 struct ureg src1,
606 struct ureg src2 )
607 {
608 emit_op(p, op, dest, mask, saturate, src0, src1, src2);
609
610 /* Accounting for indirection tracking:
611 */
612 if (src0.file == PROGRAM_TEMPORARY)
613 p->alu_temps |= 1 << src0.idx;
614
615 if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY)
616 p->alu_temps |= 1 << src1.idx;
617
618 if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY)
619 p->alu_temps |= 1 << src2.idx;
620
621 if (dest.file == PROGRAM_TEMPORARY)
622 p->alu_temps |= 1 << dest.idx;
623
624 p->program->Base.NumAluInstructions++;
625 return dest;
626 }
627
628 static struct ureg emit_texld( struct texenv_fragment_program *p,
629 enum prog_opcode op,
630 struct ureg dest,
631 GLuint destmask,
632 GLuint tex_unit,
633 GLuint tex_idx,
634 struct ureg coord )
635 {
636 struct prog_instruction *inst = emit_op( p, op,
637 dest, destmask,
638 GL_FALSE, /* don't saturate? */
639 coord, /* arg 0? */
640 undef,
641 undef);
642
643 inst->TexSrcTarget = tex_idx;
644 inst->TexSrcUnit = tex_unit;
645
646 p->program->Base.NumTexInstructions++;
647
648 /* Accounting for indirection tracking:
649 */
650 reserve_temp(p, dest);
651
652 /* Is this a texture indirection?
653 */
654 if ((coord.file == PROGRAM_TEMPORARY &&
655 (p->temps_output & (1<<coord.idx))) ||
656 (dest.file == PROGRAM_TEMPORARY &&
657 (p->alu_temps & (1<<dest.idx)))) {
658 p->program->Base.NumTexIndirections++;
659 p->temps_output = 1<<coord.idx;
660 p->alu_temps = 0;
661 assert(0); /* KW: texture env crossbar */
662 }
663
664 return dest;
665 }
666
667
668 static struct ureg register_const4f( struct texenv_fragment_program *p,
669 GLfloat s0,
670 GLfloat s1,
671 GLfloat s2,
672 GLfloat s3)
673 {
674 GLfloat values[4];
675 GLuint idx, swizzle;
676 struct ureg r;
677 values[0] = s0;
678 values[1] = s1;
679 values[2] = s2;
680 values[3] = s3;
681 idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
682 &swizzle );
683 r = make_ureg(PROGRAM_CONSTANT, idx);
684 r.swz = swizzle;
685 return r;
686 }
687
688 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
689 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
690 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
691 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
692
693
694 static struct ureg get_one( struct texenv_fragment_program *p )
695 {
696 if (is_undef(p->one))
697 p->one = register_scalar_const(p, 1.0);
698 return p->one;
699 }
700
701 static struct ureg get_half( struct texenv_fragment_program *p )
702 {
703 if (is_undef(p->half))
704 p->half = register_scalar_const(p, 0.5);
705 return p->half;
706 }
707
708 static struct ureg get_zero( struct texenv_fragment_program *p )
709 {
710 if (is_undef(p->zero))
711 p->zero = register_scalar_const(p, 0.0);
712 return p->zero;
713 }
714
715
716 static void program_error( struct texenv_fragment_program *p, const char *msg )
717 {
718 _mesa_problem(NULL, msg);
719 p->error = 1;
720 }
721
722 static struct ureg get_source( struct texenv_fragment_program *p,
723 GLuint src, GLuint unit )
724 {
725 switch (src) {
726 case SRC_TEXTURE:
727 assert(!is_undef(p->src_texture[unit]));
728 return p->src_texture[unit];
729
730 case SRC_TEXTURE0:
731 case SRC_TEXTURE1:
732 case SRC_TEXTURE2:
733 case SRC_TEXTURE3:
734 case SRC_TEXTURE4:
735 case SRC_TEXTURE5:
736 case SRC_TEXTURE6:
737 case SRC_TEXTURE7:
738 assert(!is_undef(p->src_texture[src - SRC_TEXTURE0]));
739 return p->src_texture[src - SRC_TEXTURE0];
740
741 case SRC_CONSTANT:
742 return register_param2(p, STATE_TEXENV_COLOR, unit);
743
744 case SRC_PRIMARY_COLOR:
745 return register_input(p, FRAG_ATTRIB_COL0);
746
747 case SRC_PREVIOUS:
748 default:
749 if (is_undef(p->src_previous))
750 return register_input(p, FRAG_ATTRIB_COL0);
751 else
752 return p->src_previous;
753 }
754 }
755
756 static struct ureg emit_combine_source( struct texenv_fragment_program *p,
757 GLuint mask,
758 GLuint unit,
759 GLuint source,
760 GLuint operand )
761 {
762 struct ureg arg, src, one;
763
764 src = get_source(p, source, unit);
765
766 switch (operand) {
767 case OPR_ONE_MINUS_SRC_COLOR:
768 /* Get unused tmp,
769 * Emit tmp = 1.0 - arg.xyzw
770 */
771 arg = get_temp( p );
772 one = get_one( p );
773 return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef);
774
775 case OPR_SRC_ALPHA:
776 if (mask == WRITEMASK_W)
777 return src;
778 else
779 return swizzle1( src, SWIZZLE_W );
780 case OPR_ONE_MINUS_SRC_ALPHA:
781 /* Get unused tmp,
782 * Emit tmp = 1.0 - arg.wwww
783 */
784 arg = get_temp(p);
785 one = get_one(p);
786 return emit_arith(p, OPCODE_SUB, arg, mask, 0,
787 one, swizzle1(src, SWIZZLE_W), undef);
788 case OPR_ZERO:
789 return get_zero(p);
790 case OPR_ONE:
791 return get_one(p);
792 case OPR_SRC_COLOR:
793 default:
794 return src;
795 }
796 }
797
798 static GLboolean args_match( struct state_key *key, GLuint unit )
799 {
800 GLuint i, nr = key->unit[unit].NumArgsRGB;
801
802 for (i = 0 ; i < nr ; i++) {
803 if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
804 return GL_FALSE;
805
806 switch(key->unit[unit].OptA[i].Operand) {
807 case OPR_SRC_ALPHA:
808 switch(key->unit[unit].OptRGB[i].Operand) {
809 case OPR_SRC_COLOR:
810 case OPR_SRC_ALPHA:
811 break;
812 default:
813 return GL_FALSE;
814 }
815 break;
816 case OPR_ONE_MINUS_SRC_ALPHA:
817 switch(key->unit[unit].OptRGB[i].Operand) {
818 case OPR_ONE_MINUS_SRC_COLOR:
819 case OPR_ONE_MINUS_SRC_ALPHA:
820 break;
821 default:
822 return GL_FALSE;
823 }
824 break;
825 default:
826 return GL_FALSE; /* impossible */
827 }
828 }
829
830 return GL_TRUE;
831 }
832
833 static struct ureg emit_combine( struct texenv_fragment_program *p,
834 struct ureg dest,
835 GLuint mask,
836 GLboolean saturate,
837 GLuint unit,
838 GLuint nr,
839 GLuint mode,
840 const struct mode_opt *opt)
841 {
842 struct ureg src[3];
843 struct ureg tmp, half;
844 GLuint i;
845
846 tmp = undef; /* silence warning (bug 5318) */
847
848 for (i = 0; i < nr; i++)
849 src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand );
850
851 switch (mode) {
852 case MODE_REPLACE:
853 if (mask == WRITEMASK_XYZW && !saturate)
854 return src[0];
855 else
856 return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef );
857 case MODE_MODULATE:
858 return emit_arith( p, OPCODE_MUL, dest, mask, saturate,
859 src[0], src[1], undef );
860 case MODE_ADD:
861 return emit_arith( p, OPCODE_ADD, dest, mask, saturate,
862 src[0], src[1], undef );
863 case MODE_ADD_SIGNED:
864 /* tmp = arg0 + arg1
865 * result = tmp - .5
866 */
867 half = get_half(p);
868 tmp = get_temp( p );
869 emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef );
870 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef );
871 return dest;
872 case MODE_INTERPOLATE:
873 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
874 */
875 return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] );
876
877 case MODE_SUBTRACT:
878 return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef );
879
880 case MODE_DOT3_RGBA:
881 case MODE_DOT3_RGBA_EXT:
882 case MODE_DOT3_RGB_EXT:
883 case MODE_DOT3_RGB: {
884 struct ureg tmp0 = get_temp( p );
885 struct ureg tmp1 = get_temp( p );
886 struct ureg neg1 = register_scalar_const(p, -1);
887 struct ureg two = register_scalar_const(p, 2);
888
889 /* tmp0 = 2*src0 - 1
890 * tmp1 = 2*src1 - 1
891 *
892 * dst = tmp0 dot3 tmp1
893 */
894 emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0,
895 two, src[0], neg1);
896
897 if (_mesa_memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
898 tmp1 = tmp0;
899 else
900 emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0,
901 two, src[1], neg1);
902 emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
903 return dest;
904 }
905 case MODE_MODULATE_ADD_ATI:
906 /* Arg0 * Arg2 + Arg1 */
907 return emit_arith( p, OPCODE_MAD, dest, mask, saturate,
908 src[0], src[2], src[1] );
909 case MODE_MODULATE_SIGNED_ADD_ATI: {
910 /* Arg0 * Arg2 + Arg1 - 0.5 */
911 struct ureg tmp0 = get_temp(p);
912 half = get_half(p);
913 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
914 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
915 return dest;
916 }
917 case MODE_MODULATE_SUBTRACT_ATI:
918 /* Arg0 * Arg2 - Arg1 */
919 emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
920 return dest;
921 default:
922 return src[0];
923 }
924 }
925
926
927 /**
928 * Generate instructions for one texture unit's env/combiner mode.
929 */
930 static struct ureg
931 emit_texenv(struct texenv_fragment_program *p, GLuint unit)
932 {
933 struct state_key *key = p->state;
934 GLboolean saturate = (unit < p->last_tex_stage);
935 GLuint rgb_shift, alpha_shift;
936 struct ureg out, shift;
937 struct ureg dest;
938
939 if (!key->unit[unit].enabled) {
940 return get_source(p, SRC_PREVIOUS, 0);
941 }
942
943 switch (key->unit[unit].ModeRGB) {
944 case MODE_DOT3_RGB_EXT:
945 alpha_shift = key->unit[unit].ScaleShiftA;
946 rgb_shift = 0;
947 break;
948 case MODE_DOT3_RGBA_EXT:
949 alpha_shift = 0;
950 rgb_shift = 0;
951 break;
952 default:
953 rgb_shift = key->unit[unit].ScaleShiftRGB;
954 alpha_shift = key->unit[unit].ScaleShiftA;
955 break;
956 }
957
958 /* If this is the very last calculation, emit direct to output reg:
959 */
960 if (key->separate_specular ||
961 unit != p->last_tex_stage ||
962 alpha_shift ||
963 rgb_shift)
964 dest = get_temp( p );
965 else
966 dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLR);
967
968 /* Emit the RGB and A combine ops
969 */
970 if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
971 args_match(key, unit)) {
972 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
973 unit,
974 key->unit[unit].NumArgsRGB,
975 key->unit[unit].ModeRGB,
976 key->unit[unit].OptRGB);
977 }
978 else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
979 key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
980
981 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
982 unit,
983 key->unit[unit].NumArgsRGB,
984 key->unit[unit].ModeRGB,
985 key->unit[unit].OptRGB);
986 }
987 else {
988 /* Need to do something to stop from re-emitting identical
989 * argument calculations here:
990 */
991 out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
992 unit,
993 key->unit[unit].NumArgsRGB,
994 key->unit[unit].ModeRGB,
995 key->unit[unit].OptRGB);
996 out = emit_combine( p, dest, WRITEMASK_W, saturate,
997 unit,
998 key->unit[unit].NumArgsA,
999 key->unit[unit].ModeA,
1000 key->unit[unit].OptA);
1001 }
1002
1003 /* Deal with the final shift:
1004 */
1005 if (alpha_shift || rgb_shift) {
1006 if (rgb_shift == alpha_shift) {
1007 shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
1008 }
1009 else {
1010 shift = register_const4f(p,
1011 (GLfloat)(1<<rgb_shift),
1012 (GLfloat)(1<<rgb_shift),
1013 (GLfloat)(1<<rgb_shift),
1014 (GLfloat)(1<<alpha_shift));
1015 }
1016 return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW,
1017 saturate, out, shift, undef );
1018 }
1019 else
1020 return out;
1021 }
1022
1023
1024 /**
1025 * Generate instruction for getting a texture source term.
1026 */
1027 static void load_texture( struct texenv_fragment_program *p, GLuint unit )
1028 {
1029 if (is_undef(p->src_texture[unit])) {
1030 GLuint dim = p->state->unit[unit].source_index;
1031 struct ureg texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
1032 struct ureg tmp = get_tex_temp( p );
1033
1034 if (dim == TEXTURE_UNKNOWN_INDEX)
1035 program_error(p, "TexSrcBit");
1036
1037 /* TODO: Use D0_MASK_XY where possible.
1038 */
1039 if (p->state->unit[unit].enabled) {
1040 p->src_texture[unit] = emit_texld( p, OPCODE_TXP,
1041 tmp, WRITEMASK_XYZW,
1042 unit, dim, texcoord );
1043 p->program->Base.SamplersUsed |= (1 << unit);
1044 /* This identity mapping should already be in place
1045 * (see _mesa_init_program_struct()) but let's be safe.
1046 */
1047 p->program->Base.SamplerUnits[unit] = unit;
1048 }
1049 else
1050 p->src_texture[unit] = get_zero(p);
1051 }
1052 }
1053
1054 static GLboolean load_texenv_source( struct texenv_fragment_program *p,
1055 GLuint src, GLuint unit )
1056 {
1057 switch (src) {
1058 case SRC_TEXTURE:
1059 load_texture(p, unit);
1060 break;
1061
1062 case SRC_TEXTURE0:
1063 case SRC_TEXTURE1:
1064 case SRC_TEXTURE2:
1065 case SRC_TEXTURE3:
1066 case SRC_TEXTURE4:
1067 case SRC_TEXTURE5:
1068 case SRC_TEXTURE6:
1069 case SRC_TEXTURE7:
1070 load_texture(p, src - SRC_TEXTURE0);
1071 break;
1072
1073 default:
1074 break;
1075 }
1076
1077 return GL_TRUE;
1078 }
1079
1080
1081 /**
1082 * Generate instructions for loading all texture source terms.
1083 */
1084 static GLboolean
1085 load_texunit_sources( struct texenv_fragment_program *p, int unit )
1086 {
1087 struct state_key *key = p->state;
1088 GLuint i;
1089
1090 for (i = 0; i < key->unit[unit].NumArgsRGB; i++) {
1091 load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit);
1092 }
1093
1094 for (i = 0; i < key->unit[unit].NumArgsA; i++) {
1095 load_texenv_source( p, key->unit[unit].OptA[i].Source, unit );
1096 }
1097
1098 return GL_TRUE;
1099 }
1100
1101
1102 /**
1103 * Generate a new fragment program which implements the context's
1104 * current texture env/combine mode.
1105 */
1106 static void
1107 create_new_program(GLcontext *ctx, struct state_key *key,
1108 struct gl_fragment_program *program)
1109 {
1110 struct prog_instruction instBuffer[MAX_INSTRUCTIONS];
1111 struct texenv_fragment_program p;
1112 GLuint unit;
1113 struct ureg cf, out;
1114
1115 _mesa_memset(&p, 0, sizeof(p));
1116 p.ctx = ctx;
1117 p.state = key;
1118 p.program = program;
1119
1120 /* During code generation, use locally-allocated instruction buffer,
1121 * then alloc dynamic storage below.
1122 */
1123 p.program->Base.Instructions = instBuffer;
1124 p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
1125 p.program->Base.NumTexIndirections = 1; /* correct? */
1126 p.program->Base.NumTexInstructions = 0;
1127 p.program->Base.NumAluInstructions = 0;
1128 p.program->Base.String = NULL;
1129 p.program->Base.NumInstructions =
1130 p.program->Base.NumTemporaries =
1131 p.program->Base.NumParameters =
1132 p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1133 p.program->Base.Parameters = _mesa_new_parameter_list();
1134
1135 p.program->Base.InputsRead = 0;
1136 p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLR;
1137
1138 for (unit = 0; unit < MAX_TEXTURE_UNITS; unit++)
1139 p.src_texture[unit] = undef;
1140
1141 p.src_previous = undef;
1142 p.half = undef;
1143 p.zero = undef;
1144 p.one = undef;
1145
1146 p.last_tex_stage = 0;
1147 release_temps(ctx, &p);
1148
1149 if (key->enabled_units) {
1150 /* First pass - to support texture_env_crossbar, first identify
1151 * all referenced texture sources and emit texld instructions
1152 * for each:
1153 */
1154 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1155 if (key->unit[unit].enabled) {
1156 load_texunit_sources( &p, unit );
1157 p.last_tex_stage = unit;
1158 }
1159
1160 /* Second pass - emit combine instructions to build final color:
1161 */
1162 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
1163 if (key->enabled_units & (1<<unit)) {
1164 p.src_previous = emit_texenv( &p, unit );
1165 reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
1166 release_temps(ctx, &p); /* release all temps */
1167 }
1168 }
1169
1170 cf = get_source( &p, SRC_PREVIOUS, 0 );
1171 out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLR );
1172
1173 if (key->separate_specular) {
1174 /* Emit specular add.
1175 */
1176 struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
1177 emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
1178 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef );
1179 }
1180 else if (_mesa_memcmp(&cf, &out, sizeof(cf)) != 0) {
1181 /* Will wind up in here if no texture enabled or a couple of
1182 * other scenarios (GL_REPLACE for instance).
1183 */
1184 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
1185 }
1186
1187 /* Finish up:
1188 */
1189 emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);
1190
1191 if (key->fog_enabled) {
1192 /* Pull fog mode from GLcontext, the value in the state key is
1193 * a reduced value and not what is expected in FogOption
1194 */
1195 p.program->FogOption = ctx->Fog.Mode;
1196 p.program->Base.InputsRead |= FRAG_BIT_FOGC; /* XXX new */
1197 } else
1198 p.program->FogOption = GL_NONE;
1199
1200 if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
1201 program_error(&p, "Exceeded max nr indirect texture lookups");
1202
1203 if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions)
1204 program_error(&p, "Exceeded max TEX instructions");
1205
1206 if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions)
1207 program_error(&p, "Exceeded max ALU instructions");
1208
1209 ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS);
1210
1211 /* Allocate final instruction array */
1212 p.program->Base.Instructions
1213 = _mesa_alloc_instructions(p.program->Base.NumInstructions);
1214 if (!p.program->Base.Instructions) {
1215 _mesa_error(ctx, GL_OUT_OF_MEMORY,
1216 "generating tex env program");
1217 return;
1218 }
1219 _mesa_copy_instructions(p.program->Base.Instructions, instBuffer,
1220 p.program->Base.NumInstructions);
1221
1222 if (p.program->FogOption) {
1223 _mesa_append_fog_code(ctx, p.program);
1224 p.program->FogOption = GL_NONE;
1225 }
1226
1227
1228 /* Notify driver the fragment program has (actually) changed.
1229 */
1230 if (ctx->Driver.ProgramStringNotify) {
1231 ctx->Driver.ProgramStringNotify( ctx, GL_FRAGMENT_PROGRAM_ARB,
1232 &p.program->Base );
1233 }
1234
1235 if (DISASSEM) {
1236 _mesa_print_program(&p.program->Base);
1237 _mesa_printf("\n");
1238 }
1239 }
1240
1241
1242 /**
1243 * Return a fragment program which implements the current
1244 * fixed-function texture, fog and color-sum operations.
1245 */
1246 struct gl_fragment_program *
1247 _mesa_get_fixed_func_fragment_program(GLcontext *ctx)
1248 {
1249 struct gl_fragment_program *prog;
1250 struct state_key key;
1251
1252 make_state_key(ctx, &key);
1253
1254 prog = (struct gl_fragment_program *)
1255 _mesa_search_program_cache(ctx->FragmentProgram.Cache,
1256 &key, sizeof(key));
1257
1258 if (!prog) {
1259 prog = (struct gl_fragment_program *)
1260 ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
1261
1262 create_new_program(ctx, &key, prog);
1263
1264 _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
1265 &key, sizeof(key), &prog->Base);
1266 }
1267
1268 return prog;
1269 }
1270
1271
1272
1273 /**
1274 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1275 * implements the current texture env/combine mode.
1276 * This function generates that program and puts it into effect.
1277 */
1278 void
1279 _mesa_UpdateTexEnvProgram( GLcontext *ctx )
1280 {
1281 const struct gl_fragment_program *prev = ctx->FragmentProgram._Current;
1282
1283 ASSERT(ctx->FragmentProgram._MaintainTexEnvProgram);
1284
1285 /* If a conventional fragment program/shader isn't in effect... */
1286 if (!ctx->FragmentProgram._Enabled &&
1287 (!ctx->Shader.CurrentProgram ||
1288 !ctx->Shader.CurrentProgram->FragmentProgram) ) {
1289
1290 ctx->FragmentProgram._Current
1291 = ctx->FragmentProgram._TexEnvProgram
1292 = _mesa_get_fixed_func_fragment_program(ctx);
1293 }
1294
1295 /* Tell the driver about the change. Could define a new target for
1296 * this?
1297 */
1298 if (ctx->FragmentProgram._Current != prev && ctx->Driver.BindProgram) {
1299 ctx->Driver.BindProgram(ctx, GL_FRAGMENT_PROGRAM_ARB,
1300 (struct gl_program *) ctx->FragmentProgram._Current);
1301 }
1302 }