mesa: handle vertex program enabled case also in texenvprogram.c
[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 (!ctx->VertexProgram._Enabled) {
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 = ctx->VertexProgram._Current->Base.OutputsWritten;
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) >> VERT_RESULT_TEX0)
241 << FRAG_ATTRIB_TEX0);
242 }
243
244 return fp_inputs;
245 }
246
247
248 /**
249 * Examine current texture environment state and generate a unique
250 * key to identify it.
251 */
252 static void make_state_key( GLcontext *ctx, struct state_key *key )
253 {
254 GLuint i, j;
255 GLuint inputs_referenced = FRAG_BIT_COL0;
256 GLuint inputs_available = get_fp_input_mask( ctx );
257
258 memset(key, 0, sizeof(*key));
259
260 for (i=0;i<MAX_TEXTURE_UNITS;i++) {
261 const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
262
263 if (!texUnit->_ReallyEnabled)
264 continue;
265
266 key->unit[i].enabled = 1;
267 key->enabled_units |= (1<<i);
268 key->nr_enabled_units = i+1;
269 inputs_referenced |= FRAG_BIT_TEX(i);
270
271 key->unit[i].source_index =
272 translate_tex_src_bit(texUnit->_ReallyEnabled);
273
274 key->unit[i].NumArgsRGB = texUnit->_CurrentCombine->_NumArgsRGB;
275 key->unit[i].NumArgsA = texUnit->_CurrentCombine->_NumArgsA;
276
277 key->unit[i].ModeRGB =
278 translate_mode(texUnit->_CurrentCombine->ModeRGB);
279 key->unit[i].ModeA =
280 translate_mode(texUnit->_CurrentCombine->ModeA);
281
282 key->unit[i].ScaleShiftRGB = texUnit->_CurrentCombine->ScaleShiftRGB;
283 key->unit[i].ScaleShiftA = texUnit->_CurrentCombine->ScaleShiftA;
284
285 for (j=0;j<3;j++) {
286 key->unit[i].OptRGB[j].Operand =
287 translate_operand(texUnit->_CurrentCombine->OperandRGB[j]);
288 key->unit[i].OptA[j].Operand =
289 translate_operand(texUnit->_CurrentCombine->OperandA[j]);
290 key->unit[i].OptRGB[j].Source =
291 translate_source(texUnit->_CurrentCombine->SourceRGB[j]);
292 key->unit[i].OptA[j].Source =
293 translate_source(texUnit->_CurrentCombine->SourceA[j]);
294 }
295 }
296
297 if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
298 key->separate_specular = 1;
299 inputs_referenced |= FRAG_BIT_COL1;
300 }
301
302 if (ctx->Fog.Enabled) {
303 key->fog_enabled = 1;
304 key->fog_mode = translate_fog_mode(ctx->Fog.Mode);
305 inputs_referenced |= FRAG_BIT_FOGC; /* maybe */
306 }
307
308 key->inputs_available = (inputs_available & inputs_referenced);
309 }
310
311 /* Use uregs to represent registers internally, translate to Mesa's
312 * expected formats on emit.
313 *
314 * NOTE: These are passed by value extensively in this file rather
315 * than as usual by pointer reference. If this disturbs you, try
316 * remembering they are just 32bits in size.
317 *
318 * GCC is smart enough to deal with these dword-sized structures in
319 * much the same way as if I had defined them as dwords and was using
320 * macros to access and set the fields. This is much nicer and easier
321 * to evolve.
322 */
323 struct ureg {
324 GLuint file:4;
325 GLuint idx:8;
326 GLuint negatebase:1;
327 GLuint abs:1;
328 GLuint negateabs:1;
329 GLuint swz:12;
330 GLuint pad:5;
331 };
332
333 static const struct ureg undef = {
334 PROGRAM_UNDEFINED,
335 ~0,
336 0,
337 0,
338 0,
339 0,
340 0
341 };
342
343
344 /* State used to build the fragment program:
345 */
346 struct texenv_fragment_program {
347 struct gl_fragment_program *program;
348 GLcontext *ctx;
349 struct state_key *state;
350
351 GLbitfield alu_temps; /* Track texture indirections, see spec. */
352 GLbitfield temps_output; /* Track texture indirections, see spec. */
353 GLbitfield temp_in_use; /* Tracks temporary regs which are in use. */
354 GLboolean error;
355
356 struct ureg src_texture[MAX_TEXTURE_UNITS];
357 /* Reg containing each texture unit's sampled texture color,
358 * else undef.
359 */
360
361 struct ureg src_previous; /* Reg containing color from previous
362 * stage. May need to be decl'd.
363 */
364
365 GLuint last_tex_stage; /* Number of last enabled texture unit */
366
367 struct ureg half;
368 struct ureg one;
369 struct ureg zero;
370 };
371
372
373
374 static struct ureg make_ureg(GLuint file, GLuint idx)
375 {
376 struct ureg reg;
377 reg.file = file;
378 reg.idx = idx;
379 reg.negatebase = 0;
380 reg.abs = 0;
381 reg.negateabs = 0;
382 reg.swz = SWIZZLE_NOOP;
383 reg.pad = 0;
384 return reg;
385 }
386
387 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
388 {
389 reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
390 GET_SWZ(reg.swz, y),
391 GET_SWZ(reg.swz, z),
392 GET_SWZ(reg.swz, w));
393
394 return reg;
395 }
396
397 static struct ureg swizzle1( struct ureg reg, int x )
398 {
399 return swizzle(reg, x, x, x, x);
400 }
401
402 static struct ureg negate( struct ureg reg )
403 {
404 reg.negatebase ^= 1;
405 return reg;
406 }
407
408 static GLboolean is_undef( struct ureg reg )
409 {
410 return reg.file == PROGRAM_UNDEFINED;
411 }
412
413
414 static struct ureg get_temp( struct texenv_fragment_program *p )
415 {
416 GLint bit;
417
418 /* First try and reuse temps which have been used already:
419 */
420 bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps );
421
422 /* Then any unused temporary:
423 */
424 if (!bit)
425 bit = _mesa_ffs( ~p->temp_in_use );
426
427 if (!bit) {
428 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
429 _mesa_exit(1);
430 }
431
432 if ((GLuint) bit > p->program->Base.NumTemporaries)
433 p->program->Base.NumTemporaries = bit;
434
435 p->temp_in_use |= 1<<(bit-1);
436 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
437 }
438
439 static struct ureg get_tex_temp( struct texenv_fragment_program *p )
440 {
441 int bit;
442
443 /* First try to find available temp not previously used (to avoid
444 * starting a new texture indirection). According to the spec, the
445 * ~p->temps_output isn't necessary, but will keep it there for
446 * now:
447 */
448 bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output );
449
450 /* Then any unused temporary:
451 */
452 if (!bit)
453 bit = _mesa_ffs( ~p->temp_in_use );
454
455 if (!bit) {
456 _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
457 _mesa_exit(1);
458 }
459
460 if ((GLuint) bit > p->program->Base.NumTemporaries)
461 p->program->Base.NumTemporaries = bit;
462
463 p->temp_in_use |= 1<<(bit-1);
464 return make_ureg(PROGRAM_TEMPORARY, (bit-1));
465 }
466
467
468 /** Mark a temp reg as being no longer allocatable. */
469 static void reserve_temp( struct texenv_fragment_program *p, struct ureg r )
470 {
471 if (r.file == PROGRAM_TEMPORARY)
472 p->temps_output |= (1 << r.idx);
473 }
474
475
476 static void release_temps(GLcontext *ctx, struct texenv_fragment_program *p )
477 {
478 GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps;
479
480 /* KW: To support tex_env_crossbar, don't release the registers in
481 * temps_output.
482 */
483 if (max_temp >= sizeof(int) * 8)
484 p->temp_in_use = p->temps_output;
485 else
486 p->temp_in_use = ~((1<<max_temp)-1) | p->temps_output;
487 }
488
489
490 static struct ureg register_param5( struct texenv_fragment_program *p,
491 GLint s0,
492 GLint s1,
493 GLint s2,
494 GLint s3,
495 GLint s4)
496 {
497 gl_state_index tokens[STATE_LENGTH];
498 GLuint idx;
499 tokens[0] = s0;
500 tokens[1] = s1;
501 tokens[2] = s2;
502 tokens[3] = s3;
503 tokens[4] = s4;
504 idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
505 return make_ureg(PROGRAM_STATE_VAR, idx);
506 }
507
508
509 #define register_param1(p,s0) register_param5(p,s0,0,0,0,0)
510 #define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0)
511 #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0)
512 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
513
514 static GLuint frag_to_vert_attrib( GLuint attrib )
515 {
516 switch (attrib) {
517 case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0;
518 case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1;
519 default:
520 assert(attrib >= FRAG_ATTRIB_TEX0);
521 assert(attrib <= FRAG_ATTRIB_TEX7);
522 return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0;
523 }
524 }
525
526
527 static struct ureg register_input( struct texenv_fragment_program *p, GLuint input )
528 {
529 if (p->state->inputs_available & (1<<input)) {
530 p->program->Base.InputsRead |= (1 << input);
531 return make_ureg(PROGRAM_INPUT, input);
532 }
533 else {
534 GLuint idx = frag_to_vert_attrib( input );
535 return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, idx );
536 }
537 }
538
539
540 static void emit_arg( struct prog_src_register *reg,
541 struct ureg ureg )
542 {
543 reg->File = ureg.file;
544 reg->Index = ureg.idx;
545 reg->Swizzle = ureg.swz;
546 reg->NegateBase = ureg.negatebase ? 0xf : 0x0;
547 reg->Abs = ureg.abs;
548 reg->NegateAbs = ureg.negateabs;
549 }
550
551 static void emit_dst( struct prog_dst_register *dst,
552 struct ureg ureg, GLuint mask )
553 {
554 dst->File = ureg.file;
555 dst->Index = ureg.idx;
556 dst->WriteMask = mask;
557 dst->CondMask = COND_TR; /* always pass cond test */
558 dst->CondSwizzle = SWIZZLE_NOOP;
559 }
560
561 static struct prog_instruction *
562 emit_op(struct texenv_fragment_program *p,
563 enum prog_opcode op,
564 struct ureg dest,
565 GLuint mask,
566 GLboolean saturate,
567 struct ureg src0,
568 struct ureg src1,
569 struct ureg src2 )
570 {
571 GLuint nr = p->program->Base.NumInstructions++;
572 struct prog_instruction *inst = &p->program->Base.Instructions[nr];
573
574 assert(nr < MAX_INSTRUCTIONS);
575
576 _mesa_init_instructions(inst, 1);
577 inst->Opcode = op;
578
579 emit_arg( &inst->SrcReg[0], src0 );
580 emit_arg( &inst->SrcReg[1], src1 );
581 emit_arg( &inst->SrcReg[2], src2 );
582
583 inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF;
584
585 emit_dst( &inst->DstReg, dest, mask );
586
587 #if 0
588 /* Accounting for indirection tracking:
589 */
590 if (dest.file == PROGRAM_TEMPORARY)
591 p->temps_output |= 1 << dest.idx;
592 #endif
593
594 return inst;
595 }
596
597
598 static struct ureg emit_arith( struct texenv_fragment_program *p,
599 enum prog_opcode op,
600 struct ureg dest,
601 GLuint mask,
602 GLboolean saturate,
603 struct ureg src0,
604 struct ureg src1,
605 struct ureg src2 )
606 {
607 emit_op(p, op, dest, mask, saturate, src0, src1, src2);
608
609 /* Accounting for indirection tracking:
610 */
611 if (src0.file == PROGRAM_TEMPORARY)
612 p->alu_temps |= 1 << src0.idx;
613
614 if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY)
615 p->alu_temps |= 1 << src1.idx;
616
617 if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY)
618 p->alu_temps |= 1 << src2.idx;
619
620 if (dest.file == PROGRAM_TEMPORARY)
621 p->alu_temps |= 1 << dest.idx;
622
623 p->program->Base.NumAluInstructions++;
624 return dest;
625 }
626
627 static struct ureg emit_texld( struct texenv_fragment_program *p,
628 enum prog_opcode op,
629 struct ureg dest,
630 GLuint destmask,
631 GLuint tex_unit,
632 GLuint tex_idx,
633 struct ureg coord )
634 {
635 struct prog_instruction *inst = emit_op( p, op,
636 dest, destmask,
637 GL_FALSE, /* don't saturate? */
638 coord, /* arg 0? */
639 undef,
640 undef);
641
642 inst->TexSrcTarget = tex_idx;
643 inst->TexSrcUnit = tex_unit;
644
645 p->program->Base.NumTexInstructions++;
646
647 /* Accounting for indirection tracking:
648 */
649 reserve_temp(p, dest);
650
651 /* Is this a texture indirection?
652 */
653 if ((coord.file == PROGRAM_TEMPORARY &&
654 (p->temps_output & (1<<coord.idx))) ||
655 (dest.file == PROGRAM_TEMPORARY &&
656 (p->alu_temps & (1<<dest.idx)))) {
657 p->program->Base.NumTexIndirections++;
658 p->temps_output = 1<<coord.idx;
659 p->alu_temps = 0;
660 assert(0); /* KW: texture env crossbar */
661 }
662
663 return dest;
664 }
665
666
667 static struct ureg register_const4f( struct texenv_fragment_program *p,
668 GLfloat s0,
669 GLfloat s1,
670 GLfloat s2,
671 GLfloat s3)
672 {
673 GLfloat values[4];
674 GLuint idx, swizzle;
675 struct ureg r;
676 values[0] = s0;
677 values[1] = s1;
678 values[2] = s2;
679 values[3] = s3;
680 idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
681 &swizzle );
682 r = make_ureg(PROGRAM_CONSTANT, idx);
683 r.swz = swizzle;
684 return r;
685 }
686
687 #define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0)
688 #define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1)
689 #define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1)
690 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
691
692
693 static struct ureg get_one( struct texenv_fragment_program *p )
694 {
695 if (is_undef(p->one))
696 p->one = register_scalar_const(p, 1.0);
697 return p->one;
698 }
699
700 static struct ureg get_half( struct texenv_fragment_program *p )
701 {
702 if (is_undef(p->half))
703 p->half = register_scalar_const(p, 0.5);
704 return p->half;
705 }
706
707 static struct ureg get_zero( struct texenv_fragment_program *p )
708 {
709 if (is_undef(p->zero))
710 p->zero = register_scalar_const(p, 0.0);
711 return p->zero;
712 }
713
714
715 static void program_error( struct texenv_fragment_program *p, const char *msg )
716 {
717 _mesa_problem(NULL, msg);
718 p->error = 1;
719 }
720
721 static struct ureg get_source( struct texenv_fragment_program *p,
722 GLuint src, GLuint unit )
723 {
724 switch (src) {
725 case SRC_TEXTURE:
726 assert(!is_undef(p->src_texture[unit]));
727 return p->src_texture[unit];
728
729 case SRC_TEXTURE0:
730 case SRC_TEXTURE1:
731 case SRC_TEXTURE2:
732 case SRC_TEXTURE3:
733 case SRC_TEXTURE4:
734 case SRC_TEXTURE5:
735 case SRC_TEXTURE6:
736 case SRC_TEXTURE7:
737 assert(!is_undef(p->src_texture[src - SRC_TEXTURE0]));
738 return p->src_texture[src - SRC_TEXTURE0];
739
740 case SRC_CONSTANT:
741 return register_param2(p, STATE_TEXENV_COLOR, unit);
742
743 case SRC_PRIMARY_COLOR:
744 return register_input(p, FRAG_ATTRIB_COL0);
745
746 case SRC_PREVIOUS:
747 default:
748 if (is_undef(p->src_previous))
749 return register_input(p, FRAG_ATTRIB_COL0);
750 else
751 return p->src_previous;
752 }
753 }
754
755 static struct ureg emit_combine_source( struct texenv_fragment_program *p,
756 GLuint mask,
757 GLuint unit,
758 GLuint source,
759 GLuint operand )
760 {
761 struct ureg arg, src, one;
762
763 src = get_source(p, source, unit);
764
765 switch (operand) {
766 case OPR_ONE_MINUS_SRC_COLOR:
767 /* Get unused tmp,
768 * Emit tmp = 1.0 - arg.xyzw
769 */
770 arg = get_temp( p );
771 one = get_one( p );
772 return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef);
773
774 case OPR_SRC_ALPHA:
775 if (mask == WRITEMASK_W)
776 return src;
777 else
778 return swizzle1( src, SWIZZLE_W );
779 case OPR_ONE_MINUS_SRC_ALPHA:
780 /* Get unused tmp,
781 * Emit tmp = 1.0 - arg.wwww
782 */
783 arg = get_temp(p);
784 one = get_one(p);
785 return emit_arith(p, OPCODE_SUB, arg, mask, 0,
786 one, swizzle1(src, SWIZZLE_W), undef);
787 case OPR_ZERO:
788 return get_zero(p);
789 case OPR_ONE:
790 return get_one(p);
791 case OPR_SRC_COLOR:
792 default:
793 return src;
794 }
795 }
796
797 static GLboolean args_match( struct state_key *key, GLuint unit )
798 {
799 GLuint i, nr = key->unit[unit].NumArgsRGB;
800
801 for (i = 0 ; i < nr ; i++) {
802 if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
803 return GL_FALSE;
804
805 switch(key->unit[unit].OptA[i].Operand) {
806 case OPR_SRC_ALPHA:
807 switch(key->unit[unit].OptRGB[i].Operand) {
808 case OPR_SRC_COLOR:
809 case OPR_SRC_ALPHA:
810 break;
811 default:
812 return GL_FALSE;
813 }
814 break;
815 case OPR_ONE_MINUS_SRC_ALPHA:
816 switch(key->unit[unit].OptRGB[i].Operand) {
817 case OPR_ONE_MINUS_SRC_COLOR:
818 case OPR_ONE_MINUS_SRC_ALPHA:
819 break;
820 default:
821 return GL_FALSE;
822 }
823 break;
824 default:
825 return GL_FALSE; /* impossible */
826 }
827 }
828
829 return GL_TRUE;
830 }
831
832 static struct ureg emit_combine( struct texenv_fragment_program *p,
833 struct ureg dest,
834 GLuint mask,
835 GLboolean saturate,
836 GLuint unit,
837 GLuint nr,
838 GLuint mode,
839 const struct mode_opt *opt)
840 {
841 struct ureg src[3];
842 struct ureg tmp, half;
843 GLuint i;
844
845 tmp = undef; /* silence warning (bug 5318) */
846
847 for (i = 0; i < nr; i++)
848 src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand );
849
850 switch (mode) {
851 case MODE_REPLACE:
852 if (mask == WRITEMASK_XYZW && !saturate)
853 return src[0];
854 else
855 return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef );
856 case MODE_MODULATE:
857 return emit_arith( p, OPCODE_MUL, dest, mask, saturate,
858 src[0], src[1], undef );
859 case MODE_ADD:
860 return emit_arith( p, OPCODE_ADD, dest, mask, saturate,
861 src[0], src[1], undef );
862 case MODE_ADD_SIGNED:
863 /* tmp = arg0 + arg1
864 * result = tmp - .5
865 */
866 half = get_half(p);
867 tmp = get_temp( p );
868 emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef );
869 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef );
870 return dest;
871 case MODE_INTERPOLATE:
872 /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
873 */
874 return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] );
875
876 case MODE_SUBTRACT:
877 return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef );
878
879 case MODE_DOT3_RGBA:
880 case MODE_DOT3_RGBA_EXT:
881 case MODE_DOT3_RGB_EXT:
882 case MODE_DOT3_RGB: {
883 struct ureg tmp0 = get_temp( p );
884 struct ureg tmp1 = get_temp( p );
885 struct ureg neg1 = register_scalar_const(p, -1);
886 struct ureg two = register_scalar_const(p, 2);
887
888 /* tmp0 = 2*src0 - 1
889 * tmp1 = 2*src1 - 1
890 *
891 * dst = tmp0 dot3 tmp1
892 */
893 emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0,
894 two, src[0], neg1);
895
896 if (_mesa_memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
897 tmp1 = tmp0;
898 else
899 emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0,
900 two, src[1], neg1);
901 emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
902 return dest;
903 }
904 case MODE_MODULATE_ADD_ATI:
905 /* Arg0 * Arg2 + Arg1 */
906 return emit_arith( p, OPCODE_MAD, dest, mask, saturate,
907 src[0], src[2], src[1] );
908 case MODE_MODULATE_SIGNED_ADD_ATI: {
909 /* Arg0 * Arg2 + Arg1 - 0.5 */
910 struct ureg tmp0 = get_temp(p);
911 half = get_half(p);
912 emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
913 emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
914 return dest;
915 }
916 case MODE_MODULATE_SUBTRACT_ATI:
917 /* Arg0 * Arg2 - Arg1 */
918 emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
919 return dest;
920 default:
921 return src[0];
922 }
923 }
924
925
926 /**
927 * Generate instructions for one texture unit's env/combiner mode.
928 */
929 static struct ureg
930 emit_texenv(struct texenv_fragment_program *p, GLuint unit)
931 {
932 struct state_key *key = p->state;
933 GLboolean saturate = (unit < p->last_tex_stage);
934 GLuint rgb_shift, alpha_shift;
935 struct ureg out, shift;
936 struct ureg dest;
937
938 if (!key->unit[unit].enabled) {
939 return get_source(p, SRC_PREVIOUS, 0);
940 }
941
942 switch (key->unit[unit].ModeRGB) {
943 case MODE_DOT3_RGB_EXT:
944 alpha_shift = key->unit[unit].ScaleShiftA;
945 rgb_shift = 0;
946 break;
947 case MODE_DOT3_RGBA_EXT:
948 alpha_shift = 0;
949 rgb_shift = 0;
950 break;
951 default:
952 rgb_shift = key->unit[unit].ScaleShiftRGB;
953 alpha_shift = key->unit[unit].ScaleShiftA;
954 break;
955 }
956
957 /* If this is the very last calculation, emit direct to output reg:
958 */
959 if (key->separate_specular ||
960 unit != p->last_tex_stage ||
961 alpha_shift ||
962 rgb_shift)
963 dest = get_temp( p );
964 else
965 dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLR);
966
967 /* Emit the RGB and A combine ops
968 */
969 if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
970 args_match(key, unit)) {
971 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
972 unit,
973 key->unit[unit].NumArgsRGB,
974 key->unit[unit].ModeRGB,
975 key->unit[unit].OptRGB);
976 }
977 else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
978 key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
979
980 out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
981 unit,
982 key->unit[unit].NumArgsRGB,
983 key->unit[unit].ModeRGB,
984 key->unit[unit].OptRGB);
985 }
986 else {
987 /* Need to do something to stop from re-emitting identical
988 * argument calculations here:
989 */
990 out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
991 unit,
992 key->unit[unit].NumArgsRGB,
993 key->unit[unit].ModeRGB,
994 key->unit[unit].OptRGB);
995 out = emit_combine( p, dest, WRITEMASK_W, saturate,
996 unit,
997 key->unit[unit].NumArgsA,
998 key->unit[unit].ModeA,
999 key->unit[unit].OptA);
1000 }
1001
1002 /* Deal with the final shift:
1003 */
1004 if (alpha_shift || rgb_shift) {
1005 if (rgb_shift == alpha_shift) {
1006 shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
1007 }
1008 else {
1009 shift = register_const4f(p,
1010 (GLfloat)(1<<rgb_shift),
1011 (GLfloat)(1<<rgb_shift),
1012 (GLfloat)(1<<rgb_shift),
1013 (GLfloat)(1<<alpha_shift));
1014 }
1015 return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW,
1016 saturate, out, shift, undef );
1017 }
1018 else
1019 return out;
1020 }
1021
1022
1023 /**
1024 * Generate instruction for getting a texture source term.
1025 */
1026 static void load_texture( struct texenv_fragment_program *p, GLuint unit )
1027 {
1028 if (is_undef(p->src_texture[unit])) {
1029 GLuint dim = p->state->unit[unit].source_index;
1030 struct ureg texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
1031 struct ureg tmp = get_tex_temp( p );
1032
1033 if (dim == TEXTURE_UNKNOWN_INDEX)
1034 program_error(p, "TexSrcBit");
1035
1036 /* TODO: Use D0_MASK_XY where possible.
1037 */
1038 if (p->state->unit[unit].enabled) {
1039 p->src_texture[unit] = emit_texld( p, OPCODE_TXP,
1040 tmp, WRITEMASK_XYZW,
1041 unit, dim, texcoord );
1042 p->program->Base.SamplersUsed |= (1 << unit);
1043 /* This identity mapping should already be in place
1044 * (see _mesa_init_program_struct()) but let's be safe.
1045 */
1046 p->program->Base.SamplerUnits[unit] = unit;
1047 }
1048 else
1049 p->src_texture[unit] = get_zero(p);
1050 }
1051 }
1052
1053 static GLboolean load_texenv_source( struct texenv_fragment_program *p,
1054 GLuint src, GLuint unit )
1055 {
1056 switch (src) {
1057 case SRC_TEXTURE:
1058 load_texture(p, unit);
1059 break;
1060
1061 case SRC_TEXTURE0:
1062 case SRC_TEXTURE1:
1063 case SRC_TEXTURE2:
1064 case SRC_TEXTURE3:
1065 case SRC_TEXTURE4:
1066 case SRC_TEXTURE5:
1067 case SRC_TEXTURE6:
1068 case SRC_TEXTURE7:
1069 load_texture(p, src - SRC_TEXTURE0);
1070 break;
1071
1072 default:
1073 break;
1074 }
1075
1076 return GL_TRUE;
1077 }
1078
1079
1080 /**
1081 * Generate instructions for loading all texture source terms.
1082 */
1083 static GLboolean
1084 load_texunit_sources( struct texenv_fragment_program *p, int unit )
1085 {
1086 struct state_key *key = p->state;
1087 GLuint i;
1088
1089 for (i = 0; i < key->unit[unit].NumArgsRGB; i++) {
1090 load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit);
1091 }
1092
1093 for (i = 0; i < key->unit[unit].NumArgsA; i++) {
1094 load_texenv_source( p, key->unit[unit].OptA[i].Source, unit );
1095 }
1096
1097 return GL_TRUE;
1098 }
1099
1100
1101 /**
1102 * Generate a new fragment program which implements the context's
1103 * current texture env/combine mode.
1104 */
1105 static void
1106 create_new_program(GLcontext *ctx, struct state_key *key,
1107 struct gl_fragment_program *program)
1108 {
1109 struct prog_instruction instBuffer[MAX_INSTRUCTIONS];
1110 struct texenv_fragment_program p;
1111 GLuint unit;
1112 struct ureg cf, out;
1113
1114 _mesa_memset(&p, 0, sizeof(p));
1115 p.ctx = ctx;
1116 p.state = key;
1117 p.program = program;
1118
1119 /* During code generation, use locally-allocated instruction buffer,
1120 * then alloc dynamic storage below.
1121 */
1122 p.program->Base.Instructions = instBuffer;
1123 p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
1124 p.program->Base.NumTexIndirections = 1; /* correct? */
1125 p.program->Base.NumTexInstructions = 0;
1126 p.program->Base.NumAluInstructions = 0;
1127 p.program->Base.String = NULL;
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 p.program->Base.Parameters = _mesa_new_parameter_list();
1133
1134 p.program->Base.InputsRead = 0;
1135 p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLR;
1136
1137 for (unit = 0; unit < MAX_TEXTURE_UNITS; unit++)
1138 p.src_texture[unit] = undef;
1139
1140 p.src_previous = undef;
1141 p.half = undef;
1142 p.zero = undef;
1143 p.one = undef;
1144
1145 p.last_tex_stage = 0;
1146 release_temps(ctx, &p);
1147
1148 if (key->enabled_units) {
1149 /* First pass - to support texture_env_crossbar, first identify
1150 * all referenced texture sources and emit texld instructions
1151 * for each:
1152 */
1153 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1154 if (key->unit[unit].enabled) {
1155 load_texunit_sources( &p, unit );
1156 p.last_tex_stage = unit;
1157 }
1158
1159 /* Second pass - emit combine instructions to build final color:
1160 */
1161 for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
1162 if (key->enabled_units & (1<<unit)) {
1163 p.src_previous = emit_texenv( &p, unit );
1164 reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
1165 release_temps(ctx, &p); /* release all temps */
1166 }
1167 }
1168
1169 cf = get_source( &p, SRC_PREVIOUS, 0 );
1170 out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLR );
1171
1172 if (key->separate_specular) {
1173 /* Emit specular add.
1174 */
1175 struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
1176 emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
1177 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef );
1178 }
1179 else if (_mesa_memcmp(&cf, &out, sizeof(cf)) != 0) {
1180 /* Will wind up in here if no texture enabled or a couple of
1181 * other scenarios (GL_REPLACE for instance).
1182 */
1183 emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
1184 }
1185
1186 /* Finish up:
1187 */
1188 emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);
1189
1190 if (key->fog_enabled) {
1191 /* Pull fog mode from GLcontext, the value in the state key is
1192 * a reduced value and not what is expected in FogOption
1193 */
1194 p.program->FogOption = ctx->Fog.Mode;
1195 p.program->Base.InputsRead |= FRAG_BIT_FOGC; /* XXX new */
1196 } else
1197 p.program->FogOption = GL_NONE;
1198
1199 if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
1200 program_error(&p, "Exceeded max nr indirect texture lookups");
1201
1202 if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions)
1203 program_error(&p, "Exceeded max TEX instructions");
1204
1205 if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions)
1206 program_error(&p, "Exceeded max ALU instructions");
1207
1208 ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS);
1209
1210 /* Allocate final instruction array */
1211 p.program->Base.Instructions
1212 = _mesa_alloc_instructions(p.program->Base.NumInstructions);
1213 if (!p.program->Base.Instructions) {
1214 _mesa_error(ctx, GL_OUT_OF_MEMORY,
1215 "generating tex env program");
1216 return;
1217 }
1218 _mesa_copy_instructions(p.program->Base.Instructions, instBuffer,
1219 p.program->Base.NumInstructions);
1220
1221 if (p.program->FogOption) {
1222 _mesa_append_fog_code(ctx, p.program);
1223 p.program->FogOption = GL_NONE;
1224 }
1225
1226
1227 /* Notify driver the fragment program has (actually) changed.
1228 */
1229 if (ctx->Driver.ProgramStringNotify) {
1230 ctx->Driver.ProgramStringNotify( ctx, GL_FRAGMENT_PROGRAM_ARB,
1231 &p.program->Base );
1232 }
1233
1234 if (DISASSEM) {
1235 _mesa_print_program(&p.program->Base);
1236 _mesa_printf("\n");
1237 }
1238 }
1239
1240
1241 /**
1242 * Return a fragment program which implements the current
1243 * fixed-function texture, fog and color-sum operations.
1244 */
1245 struct gl_fragment_program *
1246 _mesa_get_fixed_func_fragment_program(GLcontext *ctx)
1247 {
1248 struct gl_fragment_program *prog;
1249 struct state_key key;
1250
1251 make_state_key(ctx, &key);
1252
1253 prog = (struct gl_fragment_program *)
1254 _mesa_search_program_cache(ctx->FragmentProgram.Cache,
1255 &key, sizeof(key));
1256
1257 if (!prog) {
1258 prog = (struct gl_fragment_program *)
1259 ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
1260
1261 create_new_program(ctx, &key, prog);
1262
1263 _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
1264 &key, sizeof(key), &prog->Base);
1265 }
1266
1267 return prog;
1268 }
1269
1270
1271
1272 /**
1273 * If _MaintainTexEnvProgram is set we'll generate a fragment program that
1274 * implements the current texture env/combine mode.
1275 * This function generates that program and puts it into effect.
1276 */
1277 void
1278 _mesa_UpdateTexEnvProgram( GLcontext *ctx )
1279 {
1280 const struct gl_fragment_program *prev = ctx->FragmentProgram._Current;
1281
1282 ASSERT(ctx->FragmentProgram._MaintainTexEnvProgram);
1283
1284 /* If a conventional fragment program/shader isn't in effect... */
1285 if (!ctx->FragmentProgram._Enabled &&
1286 (!ctx->Shader.CurrentProgram ||
1287 !ctx->Shader.CurrentProgram->FragmentProgram) ) {
1288
1289 ctx->FragmentProgram._Current
1290 = ctx->FragmentProgram._TexEnvProgram
1291 = _mesa_get_fixed_func_fragment_program(ctx);
1292 }
1293
1294 /* Tell the driver about the change. Could define a new target for
1295 * this?
1296 */
1297 if (ctx->FragmentProgram._Current != prev && ctx->Driver.BindProgram) {
1298 ctx->Driver.BindProgram(ctx, GL_FRAGMENT_PROGRAM_ARB,
1299 (struct gl_program *) ctx->FragmentProgram._Current);
1300 }
1301 }