ffvertex_prog.c revision af69d88d
1/**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
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 VMWARE 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/**
29 * \file ffvertex_prog.c
30 *
31 * Create a vertex program to execute the current fixed function T&L pipeline.
32 * \author Keith Whitwell
33 */
34
35
36#include "main/glheader.h"
37#include "main/mtypes.h"
38#include "main/macros.h"
39#include "main/enums.h"
40#include "main/ffvertex_prog.h"
41#include "program/program.h"
42#include "program/prog_cache.h"
43#include "program/prog_instruction.h"
44#include "program/prog_parameter.h"
45#include "program/prog_print.h"
46#include "program/prog_statevars.h"
47
48
49/** Max of number of lights and texture coord units */
50#define NUM_UNITS MAX2(MAX_TEXTURE_COORD_UNITS, MAX_LIGHTS)
51
52struct state_key {
53   unsigned light_color_material_mask:12;
54   unsigned light_global_enabled:1;
55   unsigned light_local_viewer:1;
56   unsigned light_twoside:1;
57   unsigned material_shininess_is_zero:1;
58   unsigned need_eye_coords:1;
59   unsigned normalize:1;
60   unsigned rescale_normals:1;
61
62   unsigned fog_source_is_depth:1;
63   unsigned fog_distance_mode:2;
64   unsigned separate_specular:1;
65   unsigned point_attenuated:1;
66   unsigned point_array:1;
67   unsigned texture_enabled_global:1;
68   unsigned fragprog_inputs_read:12;
69
70   GLbitfield64 varying_vp_inputs;
71
72   struct {
73      unsigned light_enabled:1;
74      unsigned light_eyepos3_is_zero:1;
75      unsigned light_spotcutoff_is_180:1;
76      unsigned light_attenuated:1;
77      unsigned texunit_really_enabled:1;
78      unsigned texmat_enabled:1;
79      unsigned coord_replace:1;
80      unsigned texgen_enabled:4;
81      unsigned texgen_mode0:4;
82      unsigned texgen_mode1:4;
83      unsigned texgen_mode2:4;
84      unsigned texgen_mode3:4;
85   } unit[NUM_UNITS];
86};
87
88
89#define TXG_NONE           0
90#define TXG_OBJ_LINEAR     1
91#define TXG_EYE_LINEAR     2
92#define TXG_SPHERE_MAP     3
93#define TXG_REFLECTION_MAP 4
94#define TXG_NORMAL_MAP     5
95
96static GLuint translate_texgen( GLboolean enabled, GLenum mode )
97{
98   if (!enabled)
99      return TXG_NONE;
100
101   switch (mode) {
102   case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
103   case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
104   case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
105   case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
106   case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
107   default: return TXG_NONE;
108   }
109}
110
111#define FDM_EYE_RADIAL    0
112#define FDM_EYE_PLANE     1
113#define FDM_EYE_PLANE_ABS 2
114
115static GLuint translate_fog_distance_mode( GLenum mode )
116{
117   switch (mode) {
118   case GL_EYE_RADIAL_NV:
119      return FDM_EYE_RADIAL;
120   case GL_EYE_PLANE:
121      return FDM_EYE_PLANE;
122   default: /* shouldn't happen; fall through to a sensible default */
123   case GL_EYE_PLANE_ABSOLUTE_NV:
124      return FDM_EYE_PLANE_ABS;
125   }
126}
127
128static GLboolean check_active_shininess( struct gl_context *ctx,
129                                         const struct state_key *key,
130                                         GLuint side )
131{
132   GLuint attr = MAT_ATTRIB_FRONT_SHININESS + side;
133
134   if ((key->varying_vp_inputs & VERT_BIT_COLOR0) &&
135       (key->light_color_material_mask & (1 << attr)))
136      return GL_TRUE;
137
138   if (key->varying_vp_inputs & VERT_ATTRIB_GENERIC(attr))
139      return GL_TRUE;
140
141   if (ctx->Light.Material.Attrib[attr][0] != 0.0F)
142      return GL_TRUE;
143
144   return GL_FALSE;
145}
146
147
148static void make_state_key( struct gl_context *ctx, struct state_key *key )
149{
150   const struct gl_fragment_program *fp;
151   GLuint i;
152
153   memset(key, 0, sizeof(struct state_key));
154   fp = ctx->FragmentProgram._Current;
155
156   /* This now relies on texenvprogram.c being active:
157    */
158   assert(fp);
159
160   key->need_eye_coords = ctx->_NeedEyeCoords;
161
162   key->fragprog_inputs_read = fp->Base.InputsRead;
163   key->varying_vp_inputs = ctx->varying_vp_inputs;
164
165   if (ctx->RenderMode == GL_FEEDBACK) {
166      /* make sure the vertprog emits color and tex0 */
167      key->fragprog_inputs_read |= (VARYING_BIT_COL0 | VARYING_BIT_TEX0);
168   }
169
170   key->separate_specular = (ctx->Light.Model.ColorControl ==
171			     GL_SEPARATE_SPECULAR_COLOR);
172
173   if (ctx->Light.Enabled) {
174      key->light_global_enabled = 1;
175
176      if (ctx->Light.Model.LocalViewer)
177	 key->light_local_viewer = 1;
178
179      if (ctx->Light.Model.TwoSide)
180	 key->light_twoside = 1;
181
182      if (ctx->Light.ColorMaterialEnabled) {
183	 key->light_color_material_mask = ctx->Light._ColorMaterialBitmask;
184      }
185
186      for (i = 0; i < MAX_LIGHTS; i++) {
187	 struct gl_light *light = &ctx->Light.Light[i];
188
189	 if (light->Enabled) {
190	    key->unit[i].light_enabled = 1;
191
192	    if (light->EyePosition[3] == 0.0)
193	       key->unit[i].light_eyepos3_is_zero = 1;
194
195	    if (light->SpotCutoff == 180.0)
196	       key->unit[i].light_spotcutoff_is_180 = 1;
197
198	    if (light->ConstantAttenuation != 1.0 ||
199		light->LinearAttenuation != 0.0 ||
200		light->QuadraticAttenuation != 0.0)
201	       key->unit[i].light_attenuated = 1;
202	 }
203      }
204
205      if (check_active_shininess(ctx, key, 0)) {
206         key->material_shininess_is_zero = 0;
207      }
208      else if (key->light_twoside &&
209               check_active_shininess(ctx, key, 1)) {
210         key->material_shininess_is_zero = 0;
211      }
212      else {
213         key->material_shininess_is_zero = 1;
214      }
215   }
216
217   if (ctx->Transform.Normalize)
218      key->normalize = 1;
219
220   if (ctx->Transform.RescaleNormals)
221      key->rescale_normals = 1;
222
223   if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) {
224      key->fog_source_is_depth = 1;
225      key->fog_distance_mode = translate_fog_distance_mode(ctx->Fog.FogDistanceMode);
226   }
227
228   if (ctx->Point._Attenuated)
229      key->point_attenuated = 1;
230
231   if (ctx->Array.VAO->VertexAttrib[VERT_ATTRIB_POINT_SIZE].Enabled)
232      key->point_array = 1;
233
234   if (ctx->Texture._TexGenEnabled ||
235       ctx->Texture._TexMatEnabled ||
236       ctx->Texture._MaxEnabledTexImageUnit != -1)
237      key->texture_enabled_global = 1;
238
239   for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
240      struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
241
242      if (texUnit->_Current)
243	 key->unit[i].texunit_really_enabled = 1;
244
245      if (ctx->Point.PointSprite)
246	 if (ctx->Point.CoordReplace[i])
247	    key->unit[i].coord_replace = 1;
248
249      if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
250	 key->unit[i].texmat_enabled = 1;
251
252      if (texUnit->TexGenEnabled) {
253	 key->unit[i].texgen_enabled = 1;
254
255	 key->unit[i].texgen_mode0 =
256	    translate_texgen( texUnit->TexGenEnabled & (1<<0),
257			      texUnit->GenS.Mode );
258	 key->unit[i].texgen_mode1 =
259	    translate_texgen( texUnit->TexGenEnabled & (1<<1),
260			      texUnit->GenT.Mode );
261	 key->unit[i].texgen_mode2 =
262	    translate_texgen( texUnit->TexGenEnabled & (1<<2),
263			      texUnit->GenR.Mode );
264	 key->unit[i].texgen_mode3 =
265	    translate_texgen( texUnit->TexGenEnabled & (1<<3),
266			      texUnit->GenQ.Mode );
267      }
268   }
269}
270
271
272
273/* Very useful debugging tool - produces annotated listing of
274 * generated program with line/function references for each
275 * instruction back into this file:
276 */
277#define DISASSEM 0
278
279
280/* Use uregs to represent registers internally, translate to Mesa's
281 * expected formats on emit.
282 *
283 * NOTE: These are passed by value extensively in this file rather
284 * than as usual by pointer reference.  If this disturbs you, try
285 * remembering they are just 32bits in size.
286 *
287 * GCC is smart enough to deal with these dword-sized structures in
288 * much the same way as if I had defined them as dwords and was using
289 * macros to access and set the fields.  This is much nicer and easier
290 * to evolve.
291 */
292struct ureg {
293   GLuint file:4;
294   GLint idx:9;      /* relative addressing may be negative */
295                     /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
296   GLuint negate:1;
297   GLuint swz:12;
298   GLuint pad:6;
299};
300
301
302struct tnl_program {
303   const struct state_key *state;
304   struct gl_vertex_program *program;
305   GLint max_inst;  /** number of instructions allocated for program */
306   GLboolean mvp_with_dp4;
307
308   GLuint temp_in_use;
309   GLuint temp_reserved;
310
311   struct ureg eye_position;
312   struct ureg eye_position_z;
313   struct ureg eye_position_normalized;
314   struct ureg transformed_normal;
315   struct ureg identity;
316
317   GLuint materials;
318   GLuint color_materials;
319};
320
321
322static const struct ureg undef = {
323   PROGRAM_UNDEFINED,
324   0,
325   0,
326   0,
327   0
328};
329
330/* Local shorthand:
331 */
332#define X    SWIZZLE_X
333#define Y    SWIZZLE_Y
334#define Z    SWIZZLE_Z
335#define W    SWIZZLE_W
336
337
338/* Construct a ureg:
339 */
340static struct ureg make_ureg(GLuint file, GLint idx)
341{
342   struct ureg reg;
343   reg.file = file;
344   reg.idx = idx;
345   reg.negate = 0;
346   reg.swz = SWIZZLE_NOOP;
347   reg.pad = 0;
348   return reg;
349}
350
351
352
353static struct ureg negate( struct ureg reg )
354{
355   reg.negate ^= 1;
356   return reg;
357}
358
359
360static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
361{
362   reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
363			   GET_SWZ(reg.swz, y),
364			   GET_SWZ(reg.swz, z),
365			   GET_SWZ(reg.swz, w));
366   return reg;
367}
368
369
370static struct ureg swizzle1( struct ureg reg, int x )
371{
372   return swizzle(reg, x, x, x, x);
373}
374
375
376static struct ureg get_temp( struct tnl_program *p )
377{
378   int bit = ffs( ~p->temp_in_use );
379   if (!bit) {
380      _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
381      exit(1);
382   }
383
384   if ((GLuint) bit > p->program->Base.NumTemporaries)
385      p->program->Base.NumTemporaries = bit;
386
387   p->temp_in_use |= 1<<(bit-1);
388   return make_ureg(PROGRAM_TEMPORARY, bit-1);
389}
390
391
392static struct ureg reserve_temp( struct tnl_program *p )
393{
394   struct ureg temp = get_temp( p );
395   p->temp_reserved |= 1<<temp.idx;
396   return temp;
397}
398
399
400static void release_temp( struct tnl_program *p, struct ureg reg )
401{
402   if (reg.file == PROGRAM_TEMPORARY) {
403      p->temp_in_use &= ~(1<<reg.idx);
404      p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
405   }
406}
407
408static void release_temps( struct tnl_program *p )
409{
410   p->temp_in_use = p->temp_reserved;
411}
412
413
414static struct ureg register_param5(struct tnl_program *p,
415				   GLint s0,
416				   GLint s1,
417				   GLint s2,
418				   GLint s3,
419                                   GLint s4)
420{
421   gl_state_index tokens[STATE_LENGTH];
422   GLint idx;
423   tokens[0] = s0;
424   tokens[1] = s1;
425   tokens[2] = s2;
426   tokens[3] = s3;
427   tokens[4] = s4;
428   idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
429   return make_ureg(PROGRAM_STATE_VAR, idx);
430}
431
432
433#define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
434#define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
435#define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
436#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
437
438
439
440/**
441 * \param input  one of VERT_ATTRIB_x tokens.
442 */
443static struct ureg register_input( struct tnl_program *p, GLuint input )
444{
445   assert(input < VERT_ATTRIB_MAX);
446
447   if (p->state->varying_vp_inputs & VERT_BIT(input)) {
448      p->program->Base.InputsRead |= VERT_BIT(input);
449      return make_ureg(PROGRAM_INPUT, input);
450   }
451   else {
452      return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, input );
453   }
454}
455
456
457/**
458 * \param input  one of VARYING_SLOT_x tokens.
459 */
460static struct ureg register_output( struct tnl_program *p, GLuint output )
461{
462   p->program->Base.OutputsWritten |= BITFIELD64_BIT(output);
463   return make_ureg(PROGRAM_OUTPUT, output);
464}
465
466
467static struct ureg register_const4f( struct tnl_program *p,
468			      GLfloat s0,
469			      GLfloat s1,
470			      GLfloat s2,
471			      GLfloat s3)
472{
473   gl_constant_value values[4];
474   GLint idx;
475   GLuint swizzle;
476   values[0].f = s0;
477   values[1].f = s1;
478   values[2].f = s2;
479   values[3].f = s3;
480   idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
481                                     &swizzle );
482   ASSERT(swizzle == SWIZZLE_NOOP);
483   return make_ureg(PROGRAM_CONSTANT, idx);
484}
485
486#define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
487#define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
488#define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
489#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
490
491static GLboolean is_undef( struct ureg reg )
492{
493   return reg.file == PROGRAM_UNDEFINED;
494}
495
496
497static struct ureg get_identity_param( struct tnl_program *p )
498{
499   if (is_undef(p->identity))
500      p->identity = register_const4f(p, 0,0,0,1);
501
502   return p->identity;
503}
504
505static void register_matrix_param5( struct tnl_program *p,
506				    GLint s0, /* modelview, projection, etc */
507				    GLint s1, /* texture matrix number */
508				    GLint s2, /* first row */
509				    GLint s3, /* last row */
510				    GLint s4, /* inverse, transpose, etc */
511				    struct ureg *matrix )
512{
513   GLint i;
514
515   /* This is a bit sad as the support is there to pull the whole
516    * matrix out in one go:
517    */
518   for (i = 0; i <= s3 - s2; i++)
519      matrix[i] = register_param5( p, s0, s1, i, i, s4 );
520}
521
522
523static void emit_arg( struct prog_src_register *src,
524		      struct ureg reg )
525{
526   src->File = reg.file;
527   src->Index = reg.idx;
528   src->Swizzle = reg.swz;
529   src->Negate = reg.negate ? NEGATE_XYZW : NEGATE_NONE;
530   src->Abs = 0;
531   src->RelAddr = 0;
532   /* Check that bitfield sizes aren't exceeded */
533   ASSERT(src->Index == reg.idx);
534}
535
536
537static void emit_dst( struct prog_dst_register *dst,
538		      struct ureg reg, GLuint mask )
539{
540   dst->File = reg.file;
541   dst->Index = reg.idx;
542   /* allow zero as a shorthand for xyzw */
543   dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
544   dst->CondMask = COND_TR;  /* always pass cond test */
545   dst->CondSwizzle = SWIZZLE_NOOP;
546   /* Check that bitfield sizes aren't exceeded */
547   ASSERT(dst->Index == reg.idx);
548}
549
550
551static void debug_insn( struct prog_instruction *inst, const char *fn,
552			GLuint line )
553{
554   if (DISASSEM) {
555      static const char *last_fn;
556
557      if (fn != last_fn) {
558	 last_fn = fn;
559	 printf("%s:\n", fn);
560      }
561
562      printf("%d:\t", line);
563      _mesa_print_instruction(inst);
564   }
565}
566
567
568static void emit_op3fn(struct tnl_program *p,
569                       enum prog_opcode op,
570		       struct ureg dest,
571		       GLuint mask,
572		       struct ureg src0,
573		       struct ureg src1,
574		       struct ureg src2,
575		       const char *fn,
576		       GLuint line)
577{
578   GLuint nr;
579   struct prog_instruction *inst;
580
581   assert((GLint) p->program->Base.NumInstructions <= p->max_inst);
582
583   if (p->program->Base.NumInstructions == p->max_inst) {
584      /* need to extend the program's instruction array */
585      struct prog_instruction *newInst;
586
587      /* double the size */
588      p->max_inst *= 2;
589
590      newInst = _mesa_alloc_instructions(p->max_inst);
591      if (!newInst) {
592         _mesa_error(NULL, GL_OUT_OF_MEMORY, "vertex program build");
593         return;
594      }
595
596      _mesa_copy_instructions(newInst,
597                              p->program->Base.Instructions,
598                              p->program->Base.NumInstructions);
599
600      _mesa_free_instructions(p->program->Base.Instructions,
601                              p->program->Base.NumInstructions);
602
603      p->program->Base.Instructions = newInst;
604   }
605
606   nr = p->program->Base.NumInstructions++;
607
608   inst = &p->program->Base.Instructions[nr];
609   inst->Opcode = (enum prog_opcode) op;
610
611   emit_arg( &inst->SrcReg[0], src0 );
612   emit_arg( &inst->SrcReg[1], src1 );
613   emit_arg( &inst->SrcReg[2], src2 );
614
615   emit_dst( &inst->DstReg, dest, mask );
616
617   debug_insn(inst, fn, line);
618}
619
620
621#define emit_op3(p, op, dst, mask, src0, src1, src2) \
622   emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
623
624#define emit_op2(p, op, dst, mask, src0, src1) \
625    emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
626
627#define emit_op1(p, op, dst, mask, src0) \
628    emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
629
630
631static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
632{
633   if (reg.file == PROGRAM_TEMPORARY &&
634       !(p->temp_reserved & (1<<reg.idx)))
635      return reg;
636   else {
637      struct ureg temp = get_temp(p);
638      emit_op1(p, OPCODE_MOV, temp, 0, reg);
639      return temp;
640   }
641}
642
643
644/* Currently no tracking performed of input/output/register size or
645 * active elements.  Could be used to reduce these operations, as
646 * could the matrix type.
647 */
648static void emit_matrix_transform_vec4( struct tnl_program *p,
649					struct ureg dest,
650					const struct ureg *mat,
651					struct ureg src)
652{
653   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
654   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
655   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
656   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
657}
658
659
660/* This version is much easier to implement if writemasks are not
661 * supported natively on the target or (like SSE), the target doesn't
662 * have a clean/obvious dotproduct implementation.
663 */
664static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
665						  struct ureg dest,
666						  const struct ureg *mat,
667						  struct ureg src)
668{
669   struct ureg tmp;
670
671   if (dest.file != PROGRAM_TEMPORARY)
672      tmp = get_temp(p);
673   else
674      tmp = dest;
675
676   emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
677   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
678   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
679   emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
680
681   if (dest.file != PROGRAM_TEMPORARY)
682      release_temp(p, tmp);
683}
684
685
686static void emit_matrix_transform_vec3( struct tnl_program *p,
687					struct ureg dest,
688					const struct ureg *mat,
689					struct ureg src)
690{
691   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
692   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
693   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
694}
695
696
697static void emit_normalize_vec3( struct tnl_program *p,
698				 struct ureg dest,
699				 struct ureg src )
700{
701   struct ureg tmp = get_temp(p);
702   emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src);
703   emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
704   emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X));
705   release_temp(p, tmp);
706}
707
708
709static void emit_passthrough( struct tnl_program *p,
710			      GLuint input,
711			      GLuint output )
712{
713   struct ureg out = register_output(p, output);
714   emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
715}
716
717
718static struct ureg get_eye_position( struct tnl_program *p )
719{
720   if (is_undef(p->eye_position)) {
721      struct ureg pos = register_input( p, VERT_ATTRIB_POS );
722      struct ureg modelview[4];
723
724      p->eye_position = reserve_temp(p);
725
726      if (p->mvp_with_dp4) {
727	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
728                                 0, modelview );
729
730	 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
731      }
732      else {
733	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
734				 STATE_MATRIX_TRANSPOSE, modelview );
735
736	 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
737      }
738   }
739
740   return p->eye_position;
741}
742
743
744static struct ureg get_eye_position_z( struct tnl_program *p )
745{
746   if (!is_undef(p->eye_position))
747      return swizzle1(p->eye_position, Z);
748
749   if (is_undef(p->eye_position_z)) {
750      struct ureg pos = register_input( p, VERT_ATTRIB_POS );
751      struct ureg modelview[4];
752
753      p->eye_position_z = reserve_temp(p);
754
755      register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
756                              0, modelview );
757
758      emit_op2(p, OPCODE_DP4, p->eye_position_z, 0, pos, modelview[2]);
759   }
760
761   return p->eye_position_z;
762}
763
764
765static struct ureg get_eye_position_normalized( struct tnl_program *p )
766{
767   if (is_undef(p->eye_position_normalized)) {
768      struct ureg eye = get_eye_position(p);
769      p->eye_position_normalized = reserve_temp(p);
770      emit_normalize_vec3(p, p->eye_position_normalized, eye);
771   }
772
773   return p->eye_position_normalized;
774}
775
776
777static struct ureg get_transformed_normal( struct tnl_program *p )
778{
779   if (is_undef(p->transformed_normal) &&
780       !p->state->need_eye_coords &&
781       !p->state->normalize &&
782       !(p->state->need_eye_coords == p->state->rescale_normals))
783   {
784      p->transformed_normal = register_input(p, VERT_ATTRIB_NORMAL );
785   }
786   else if (is_undef(p->transformed_normal))
787   {
788      struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
789      struct ureg mvinv[3];
790      struct ureg transformed_normal = reserve_temp(p);
791
792      if (p->state->need_eye_coords) {
793         register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
794                                 STATE_MATRIX_INVTRANS, mvinv );
795
796         /* Transform to eye space:
797          */
798         emit_matrix_transform_vec3( p, transformed_normal, mvinv, normal );
799         normal = transformed_normal;
800      }
801
802      /* Normalize/Rescale:
803       */
804      if (p->state->normalize) {
805	 emit_normalize_vec3( p, transformed_normal, normal );
806         normal = transformed_normal;
807      }
808      else if (p->state->need_eye_coords == p->state->rescale_normals) {
809         /* This is already adjusted for eye/non-eye rendering:
810          */
811	 struct ureg rescale = register_param2(p, STATE_INTERNAL,
812                                               STATE_NORMAL_SCALE);
813
814	 emit_op2( p, OPCODE_MUL, transformed_normal, 0, normal, rescale );
815         normal = transformed_normal;
816      }
817
818      assert(normal.file == PROGRAM_TEMPORARY);
819      p->transformed_normal = normal;
820   }
821
822   return p->transformed_normal;
823}
824
825
826static void build_hpos( struct tnl_program *p )
827{
828   struct ureg pos = register_input( p, VERT_ATTRIB_POS );
829   struct ureg hpos = register_output( p, VARYING_SLOT_POS );
830   struct ureg mvp[4];
831
832   if (p->mvp_with_dp4) {
833      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
834			      0, mvp );
835      emit_matrix_transform_vec4( p, hpos, mvp, pos );
836   }
837   else {
838      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
839			      STATE_MATRIX_TRANSPOSE, mvp );
840      emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
841   }
842}
843
844
845static GLuint material_attrib( GLuint side, GLuint property )
846{
847   return (property - STATE_AMBIENT) * 2 + side;
848}
849
850
851/**
852 * Get a bitmask of which material values vary on a per-vertex basis.
853 */
854static void set_material_flags( struct tnl_program *p )
855{
856   p->color_materials = 0;
857   p->materials = 0;
858
859   if (p->state->varying_vp_inputs & VERT_BIT_COLOR0) {
860      p->materials =
861	 p->color_materials = p->state->light_color_material_mask;
862   }
863
864   p->materials |= (p->state->varying_vp_inputs >> VERT_ATTRIB_GENERIC0);
865}
866
867
868static struct ureg get_material( struct tnl_program *p, GLuint side,
869				 GLuint property )
870{
871   GLuint attrib = material_attrib(side, property);
872
873   if (p->color_materials & (1<<attrib))
874      return register_input(p, VERT_ATTRIB_COLOR0);
875   else if (p->materials & (1<<attrib)) {
876      /* Put material values in the GENERIC slots -- they are not used
877       * for anything in fixed function mode.
878       */
879      return register_input( p, attrib + VERT_ATTRIB_GENERIC0 );
880   }
881   else
882      return register_param3( p, STATE_MATERIAL, side, property );
883}
884
885#define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
886				   MAT_BIT_FRONT_AMBIENT | \
887				   MAT_BIT_FRONT_DIFFUSE) << (side))
888
889
890/**
891 * Either return a precalculated constant value or emit code to
892 * calculate these values dynamically in the case where material calls
893 * are present between begin/end pairs.
894 *
895 * Probably want to shift this to the program compilation phase - if
896 * we always emitted the calculation here, a smart compiler could
897 * detect that it was constant (given a certain set of inputs), and
898 * lift it out of the main loop.  That way the programs created here
899 * would be independent of the vertex_buffer details.
900 */
901static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
902{
903   if (p->materials & SCENE_COLOR_BITS(side)) {
904      struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
905      struct ureg material_emission = get_material(p, side, STATE_EMISSION);
906      struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
907      struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
908      struct ureg tmp = make_temp(p, material_diffuse);
909      emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
910	       material_ambient, material_emission);
911      return tmp;
912   }
913   else
914      return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
915}
916
917
918static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
919				  GLuint side, GLuint property )
920{
921   GLuint attrib = material_attrib(side, property);
922   if (p->materials & (1<<attrib)) {
923      struct ureg light_value =
924	 register_param3(p, STATE_LIGHT, light, property);
925      struct ureg material_value = get_material(p, side, property);
926      struct ureg tmp = get_temp(p);
927      emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value);
928      return tmp;
929   }
930   else
931      return register_param4(p, STATE_LIGHTPROD, light, side, property);
932}
933
934
935static struct ureg calculate_light_attenuation( struct tnl_program *p,
936						GLuint i,
937						struct ureg VPpli,
938						struct ureg dist )
939{
940   struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
941					     STATE_ATTENUATION);
942   struct ureg att = undef;
943
944   /* Calculate spot attenuation:
945    */
946   if (!p->state->unit[i].light_spotcutoff_is_180) {
947      struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
948						  STATE_LIGHT_SPOT_DIR_NORMALIZED, i);
949      struct ureg spot = get_temp(p);
950      struct ureg slt = get_temp(p);
951
952      att = get_temp(p);
953
954      emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
955      emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
956      emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
957      emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
958
959      release_temp(p, spot);
960      release_temp(p, slt);
961   }
962
963   /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
964    *
965    * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
966    */
967   if (p->state->unit[i].light_attenuated && !is_undef(dist)) {
968      if (is_undef(att))
969         att = get_temp(p);
970      /* 1/d,d,d,1/d */
971      emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
972      /* 1,d,d*d,1/d */
973      emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
974      /* 1/dist-atten */
975      emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
976
977      if (!p->state->unit[i].light_spotcutoff_is_180) {
978	 /* dist-atten */
979	 emit_op1(p, OPCODE_RCP, dist, 0, dist);
980	 /* spot-atten * dist-atten */
981	 emit_op2(p, OPCODE_MUL, att, 0, dist, att);
982      }
983      else {
984	 /* dist-atten */
985	 emit_op1(p, OPCODE_RCP, att, 0, dist);
986      }
987   }
988
989   return att;
990}
991
992
993/**
994 * Compute:
995 *   lit.y = MAX(0, dots.x)
996 *   lit.z = SLT(0, dots.x)
997 */
998static void emit_degenerate_lit( struct tnl_program *p,
999                                 struct ureg lit,
1000                                 struct ureg dots )
1001{
1002   struct ureg id = get_identity_param(p);  /* id = {0,0,0,1} */
1003
1004   /* Note that lit.x & lit.w will not be examined.  Note also that
1005    * dots.xyzw == dots.xxxx.
1006    */
1007
1008   /* MAX lit, id, dots;
1009    */
1010   emit_op2(p, OPCODE_MAX, lit, WRITEMASK_XYZW, id, dots);
1011
1012   /* result[2] = (in > 0 ? 1 : 0)
1013    * SLT lit.z, id.z, dots;   # lit.z = (0 < dots.z) ? 1 : 0
1014    */
1015   emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, swizzle1(id,Z), dots);
1016}
1017
1018
1019/* Need to add some addtional parameters to allow lighting in object
1020 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1021 * space lighting.
1022 */
1023static void build_lighting( struct tnl_program *p )
1024{
1025   const GLboolean twoside = p->state->light_twoside;
1026   const GLboolean separate = p->state->separate_specular;
1027   GLuint nr_lights = 0, count = 0;
1028   struct ureg normal = get_transformed_normal(p);
1029   struct ureg lit = get_temp(p);
1030   struct ureg dots = get_temp(p);
1031   struct ureg _col0 = undef, _col1 = undef;
1032   struct ureg _bfc0 = undef, _bfc1 = undef;
1033   GLuint i;
1034
1035   /*
1036    * NOTE:
1037    * dots.x = dot(normal, VPpli)
1038    * dots.y = dot(normal, halfAngle)
1039    * dots.z = back.shininess
1040    * dots.w = front.shininess
1041    */
1042
1043   for (i = 0; i < MAX_LIGHTS; i++)
1044      if (p->state->unit[i].light_enabled)
1045	 nr_lights++;
1046
1047   set_material_flags(p);
1048
1049   {
1050      if (!p->state->material_shininess_is_zero) {
1051         struct ureg shininess = get_material(p, 0, STATE_SHININESS);
1052         emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
1053         release_temp(p, shininess);
1054      }
1055
1056      _col0 = make_temp(p, get_scenecolor(p, 0));
1057      if (separate)
1058	 _col1 = make_temp(p, get_identity_param(p));
1059      else
1060	 _col1 = _col0;
1061   }
1062
1063   if (twoside) {
1064      if (!p->state->material_shininess_is_zero) {
1065         /* Note that we negate the back-face specular exponent here.
1066          * The negation will be un-done later in the back-face code below.
1067          */
1068         struct ureg shininess = get_material(p, 1, STATE_SHININESS);
1069         emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
1070                  negate(swizzle1(shininess,X)));
1071         release_temp(p, shininess);
1072      }
1073
1074      _bfc0 = make_temp(p, get_scenecolor(p, 1));
1075      if (separate)
1076	 _bfc1 = make_temp(p, get_identity_param(p));
1077      else
1078	 _bfc1 = _bfc0;
1079   }
1080
1081   /* If no lights, still need to emit the scenecolor.
1082    */
1083   {
1084      struct ureg res0 = register_output( p, VARYING_SLOT_COL0 );
1085      emit_op1(p, OPCODE_MOV, res0, 0, _col0);
1086   }
1087
1088   if (separate) {
1089      struct ureg res1 = register_output( p, VARYING_SLOT_COL1 );
1090      emit_op1(p, OPCODE_MOV, res1, 0, _col1);
1091   }
1092
1093   if (twoside) {
1094      struct ureg res0 = register_output( p, VARYING_SLOT_BFC0 );
1095      emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
1096   }
1097
1098   if (twoside && separate) {
1099      struct ureg res1 = register_output( p, VARYING_SLOT_BFC1 );
1100      emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
1101   }
1102
1103   if (nr_lights == 0) {
1104      release_temps(p);
1105      return;
1106   }
1107
1108   for (i = 0; i < MAX_LIGHTS; i++) {
1109      if (p->state->unit[i].light_enabled) {
1110	 struct ureg half = undef;
1111	 struct ureg att = undef, VPpli = undef;
1112	 struct ureg dist = undef;
1113
1114	 count++;
1115         if (p->state->unit[i].light_eyepos3_is_zero) {
1116             VPpli = register_param3(p, STATE_INTERNAL,
1117                                     STATE_LIGHT_POSITION_NORMALIZED, i);
1118         } else {
1119            struct ureg Ppli = register_param3(p, STATE_INTERNAL,
1120                                               STATE_LIGHT_POSITION, i);
1121            struct ureg V = get_eye_position(p);
1122
1123            VPpli = get_temp(p);
1124            dist = get_temp(p);
1125
1126            /* Calculate VPpli vector
1127             */
1128            emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
1129
1130            /* Normalize VPpli.  The dist value also used in
1131             * attenuation below.
1132             */
1133            emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
1134            emit_op1(p, OPCODE_RSQ, dist, 0, dist);
1135            emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
1136         }
1137
1138         /* Calculate attenuation:
1139          */
1140         att = calculate_light_attenuation(p, i, VPpli, dist);
1141         release_temp(p, dist);
1142
1143	 /* Calculate viewer direction, or use infinite viewer:
1144	  */
1145         if (!p->state->material_shininess_is_zero) {
1146            if (p->state->light_local_viewer) {
1147               struct ureg eye_hat = get_eye_position_normalized(p);
1148               half = get_temp(p);
1149               emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1150               emit_normalize_vec3(p, half, half);
1151            } else if (p->state->unit[i].light_eyepos3_is_zero) {
1152               half = register_param3(p, STATE_INTERNAL,
1153                                      STATE_LIGHT_HALF_VECTOR, i);
1154            } else {
1155               struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1156               half = get_temp(p);
1157               emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1158               emit_normalize_vec3(p, half, half);
1159            }
1160	 }
1161
1162	 /* Calculate dot products:
1163	  */
1164         if (p->state->material_shininess_is_zero) {
1165            emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli);
1166         }
1167         else {
1168            emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1169            emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1170         }
1171
1172	 /* Front face lighting:
1173	  */
1174	 {
1175	    struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1176	    struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1177	    struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1178	    struct ureg res0, res1;
1179	    GLuint mask0, mask1;
1180
1181	    if (count == nr_lights) {
1182	       if (separate) {
1183		  mask0 = WRITEMASK_XYZ;
1184		  mask1 = WRITEMASK_XYZ;
1185		  res0 = register_output( p, VARYING_SLOT_COL0 );
1186		  res1 = register_output( p, VARYING_SLOT_COL1 );
1187	       }
1188	       else {
1189		  mask0 = 0;
1190		  mask1 = WRITEMASK_XYZ;
1191		  res0 = _col0;
1192		  res1 = register_output( p, VARYING_SLOT_COL0 );
1193	       }
1194	    }
1195            else {
1196	       mask0 = 0;
1197	       mask1 = 0;
1198	       res0 = _col0;
1199	       res1 = _col1;
1200	    }
1201
1202	    if (!is_undef(att)) {
1203               /* light is attenuated by distance */
1204               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1205               emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1206               emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1207            }
1208            else if (!p->state->material_shininess_is_zero) {
1209               /* there's a non-zero specular term */
1210               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1211               emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1212            }
1213            else {
1214               /* no attenutation, no specular */
1215               emit_degenerate_lit(p, lit, dots);
1216               emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1217            }
1218
1219	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1220	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1221
1222	    release_temp(p, ambient);
1223	    release_temp(p, diffuse);
1224	    release_temp(p, specular);
1225	 }
1226
1227	 /* Back face lighting:
1228	  */
1229	 if (twoside) {
1230	    struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1231	    struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1232	    struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1233	    struct ureg res0, res1;
1234	    GLuint mask0, mask1;
1235
1236	    if (count == nr_lights) {
1237	       if (separate) {
1238		  mask0 = WRITEMASK_XYZ;
1239		  mask1 = WRITEMASK_XYZ;
1240		  res0 = register_output( p, VARYING_SLOT_BFC0 );
1241		  res1 = register_output( p, VARYING_SLOT_BFC1 );
1242	       }
1243	       else {
1244		  mask0 = 0;
1245		  mask1 = WRITEMASK_XYZ;
1246		  res0 = _bfc0;
1247		  res1 = register_output( p, VARYING_SLOT_BFC0 );
1248	       }
1249	    }
1250            else {
1251	       res0 = _bfc0;
1252	       res1 = _bfc1;
1253	       mask0 = 0;
1254	       mask1 = 0;
1255	    }
1256
1257            /* For the back face we need to negate the X and Y component
1258             * dot products.  dots.Z has the negated back-face specular
1259             * exponent.  We swizzle that into the W position.  This
1260             * negation makes the back-face specular term positive again.
1261             */
1262            dots = negate(swizzle(dots,X,Y,W,Z));
1263
1264	    if (!is_undef(att)) {
1265               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1266	       emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1267               emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1268            }
1269            else if (!p->state->material_shininess_is_zero) {
1270               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1271               emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); /**/
1272            }
1273            else {
1274               emit_degenerate_lit(p, lit, dots);
1275               emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
1276            }
1277
1278	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1279	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1280            /* restore dots to its original state for subsequent lights
1281             * by negating and swizzling again.
1282             */
1283            dots = negate(swizzle(dots,X,Y,W,Z));
1284
1285	    release_temp(p, ambient);
1286	    release_temp(p, diffuse);
1287	    release_temp(p, specular);
1288	 }
1289
1290	 release_temp(p, half);
1291	 release_temp(p, VPpli);
1292	 release_temp(p, att);
1293      }
1294   }
1295
1296   release_temps( p );
1297}
1298
1299
1300static void build_fog( struct tnl_program *p )
1301{
1302   struct ureg fog = register_output(p, VARYING_SLOT_FOGC);
1303   struct ureg input;
1304
1305   if (p->state->fog_source_is_depth) {
1306
1307      switch (p->state->fog_distance_mode) {
1308      case FDM_EYE_RADIAL: /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1309         input = get_eye_position(p);
1310         emit_op2(p, OPCODE_DP3, fog, WRITEMASK_X, input, input);
1311         emit_op1(p, OPCODE_RSQ, fog, WRITEMASK_X, fog);
1312         emit_op1(p, OPCODE_RCP, fog, WRITEMASK_X, fog);
1313         break;
1314      case FDM_EYE_PLANE: /* Z = Ze */
1315         input = get_eye_position_z(p);
1316         emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1317         break;
1318      case FDM_EYE_PLANE_ABS: /* Z = abs(Ze) */
1319         input = get_eye_position_z(p);
1320         emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1321         break;
1322      default:
1323         assert(!"Bad fog mode in build_fog()");
1324         break;
1325      }
1326
1327   }
1328   else {
1329      input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1330      emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1331   }
1332
1333   emit_op1(p, OPCODE_MOV, fog, WRITEMASK_YZW, get_identity_param(p));
1334}
1335
1336
1337static void build_reflect_texgen( struct tnl_program *p,
1338				  struct ureg dest,
1339				  GLuint writemask )
1340{
1341   struct ureg normal = get_transformed_normal(p);
1342   struct ureg eye_hat = get_eye_position_normalized(p);
1343   struct ureg tmp = get_temp(p);
1344
1345   /* n.u */
1346   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1347   /* 2n.u */
1348   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1349   /* (-2n.u)n + u */
1350   emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1351
1352   release_temp(p, tmp);
1353}
1354
1355
1356static void build_sphere_texgen( struct tnl_program *p,
1357				 struct ureg dest,
1358				 GLuint writemask )
1359{
1360   struct ureg normal = get_transformed_normal(p);
1361   struct ureg eye_hat = get_eye_position_normalized(p);
1362   struct ureg tmp = get_temp(p);
1363   struct ureg half = register_scalar_const(p, .5);
1364   struct ureg r = get_temp(p);
1365   struct ureg inv_m = get_temp(p);
1366   struct ureg id = get_identity_param(p);
1367
1368   /* Could share the above calculations, but it would be
1369    * a fairly odd state for someone to set (both sphere and
1370    * reflection active for different texture coordinate
1371    * components.  Of course - if two texture units enable
1372    * reflect and/or sphere, things start to tilt in favour
1373    * of seperating this out:
1374    */
1375
1376   /* n.u */
1377   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1378   /* 2n.u */
1379   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1380   /* (-2n.u)n + u */
1381   emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1382   /* r + 0,0,1 */
1383   emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1384   /* rx^2 + ry^2 + (rz+1)^2 */
1385   emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1386   /* 2/m */
1387   emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1388   /* 1/m */
1389   emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1390   /* r/m + 1/2 */
1391   emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1392
1393   release_temp(p, tmp);
1394   release_temp(p, r);
1395   release_temp(p, inv_m);
1396}
1397
1398
1399static void build_texture_transform( struct tnl_program *p )
1400{
1401   GLuint i, j;
1402
1403   for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
1404
1405      if (!(p->state->fragprog_inputs_read & VARYING_BIT_TEX(i)))
1406	 continue;
1407
1408      if (p->state->unit[i].coord_replace)
1409  	 continue;
1410
1411      if (p->state->unit[i].texgen_enabled ||
1412	  p->state->unit[i].texmat_enabled) {
1413
1414	 GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1415	 struct ureg out = register_output(p, VARYING_SLOT_TEX0 + i);
1416	 struct ureg out_texgen = undef;
1417
1418	 if (p->state->unit[i].texgen_enabled) {
1419	    GLuint copy_mask = 0;
1420	    GLuint sphere_mask = 0;
1421	    GLuint reflect_mask = 0;
1422	    GLuint normal_mask = 0;
1423	    GLuint modes[4];
1424
1425	    if (texmat_enabled)
1426	       out_texgen = get_temp(p);
1427	    else
1428	       out_texgen = out;
1429
1430	    modes[0] = p->state->unit[i].texgen_mode0;
1431	    modes[1] = p->state->unit[i].texgen_mode1;
1432	    modes[2] = p->state->unit[i].texgen_mode2;
1433	    modes[3] = p->state->unit[i].texgen_mode3;
1434
1435	    for (j = 0; j < 4; j++) {
1436	       switch (modes[j]) {
1437	       case TXG_OBJ_LINEAR: {
1438		  struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1439		  struct ureg plane =
1440		     register_param3(p, STATE_TEXGEN, i,
1441				     STATE_TEXGEN_OBJECT_S + j);
1442
1443		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1444			   obj, plane );
1445		  break;
1446	       }
1447	       case TXG_EYE_LINEAR: {
1448		  struct ureg eye = get_eye_position(p);
1449		  struct ureg plane =
1450		     register_param3(p, STATE_TEXGEN, i,
1451				     STATE_TEXGEN_EYE_S + j);
1452
1453		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1454			   eye, plane );
1455		  break;
1456	       }
1457	       case TXG_SPHERE_MAP:
1458		  sphere_mask |= WRITEMASK_X << j;
1459		  break;
1460	       case TXG_REFLECTION_MAP:
1461		  reflect_mask |= WRITEMASK_X << j;
1462		  break;
1463	       case TXG_NORMAL_MAP:
1464		  normal_mask |= WRITEMASK_X << j;
1465		  break;
1466	       case TXG_NONE:
1467		  copy_mask |= WRITEMASK_X << j;
1468	       }
1469	    }
1470
1471	    if (sphere_mask) {
1472	       build_sphere_texgen(p, out_texgen, sphere_mask);
1473	    }
1474
1475	    if (reflect_mask) {
1476	       build_reflect_texgen(p, out_texgen, reflect_mask);
1477	    }
1478
1479	    if (normal_mask) {
1480	       struct ureg normal = get_transformed_normal(p);
1481	       emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1482	    }
1483
1484	    if (copy_mask) {
1485	       struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1486	       emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1487	    }
1488	 }
1489
1490	 if (texmat_enabled) {
1491	    struct ureg texmat[4];
1492	    struct ureg in = (!is_undef(out_texgen) ?
1493			      out_texgen :
1494			      register_input(p, VERT_ATTRIB_TEX0+i));
1495	    if (p->mvp_with_dp4) {
1496	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1497				       0, texmat );
1498	       emit_matrix_transform_vec4( p, out, texmat, in );
1499	    }
1500	    else {
1501	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1502				       STATE_MATRIX_TRANSPOSE, texmat );
1503	       emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1504	    }
1505	 }
1506
1507	 release_temps(p);
1508      }
1509      else {
1510	 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VARYING_SLOT_TEX0+i);
1511      }
1512   }
1513}
1514
1515
1516/**
1517 * Point size attenuation computation.
1518 */
1519static void build_atten_pointsize( struct tnl_program *p )
1520{
1521   struct ureg eye = get_eye_position_z(p);
1522   struct ureg state_size = register_param2(p, STATE_INTERNAL, STATE_POINT_SIZE_CLAMPED);
1523   struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1524   struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1525   struct ureg ut = get_temp(p);
1526
1527   /* dist = |eyez| */
1528   emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
1529   /* p1 + dist * (p2 + dist * p3); */
1530   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1531		swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1532   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1533		ut, swizzle1(state_attenuation, X));
1534
1535   /* 1 / sqrt(factor) */
1536   emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1537
1538#if 0
1539   /* out = pointSize / sqrt(factor) */
1540   emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1541#else
1542   /* this is a good place to clamp the point size since there's likely
1543    * no hardware registers to clamp point size at rasterization time.
1544    */
1545   emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1546   emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1547   emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1548#endif
1549
1550   release_temp(p, ut);
1551}
1552
1553
1554/**
1555 * Pass-though per-vertex point size, from user's point size array.
1556 */
1557static void build_array_pointsize( struct tnl_program *p )
1558{
1559   struct ureg in = register_input(p, VERT_ATTRIB_POINT_SIZE);
1560   struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1561   emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, in);
1562}
1563
1564
1565static void build_tnl_program( struct tnl_program *p )
1566{
1567   /* Emit the program, starting with the modelview, projection transforms:
1568    */
1569   build_hpos(p);
1570
1571   /* Lighting calculations:
1572    */
1573   if (p->state->fragprog_inputs_read & (VARYING_BIT_COL0|VARYING_BIT_COL1)) {
1574      if (p->state->light_global_enabled)
1575	 build_lighting(p);
1576      else {
1577	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL0)
1578	    emit_passthrough(p, VERT_ATTRIB_COLOR0, VARYING_SLOT_COL0);
1579
1580	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL1)
1581	    emit_passthrough(p, VERT_ATTRIB_COLOR1, VARYING_SLOT_COL1);
1582      }
1583   }
1584
1585   if (p->state->fragprog_inputs_read & VARYING_BIT_FOGC)
1586      build_fog(p);
1587
1588   if (p->state->fragprog_inputs_read & VARYING_BITS_TEX_ANY)
1589      build_texture_transform(p);
1590
1591   if (p->state->point_attenuated)
1592      build_atten_pointsize(p);
1593   else if (p->state->point_array)
1594      build_array_pointsize(p);
1595
1596   /* Finish up:
1597    */
1598   emit_op1(p, OPCODE_END, undef, 0, undef);
1599
1600   /* Disassemble:
1601    */
1602   if (DISASSEM) {
1603      printf ("\n");
1604   }
1605}
1606
1607
1608static void
1609create_new_program( const struct state_key *key,
1610                    struct gl_vertex_program *program,
1611                    GLboolean mvp_with_dp4,
1612                    GLuint max_temps)
1613{
1614   struct tnl_program p;
1615
1616   memset(&p, 0, sizeof(p));
1617   p.state = key;
1618   p.program = program;
1619   p.eye_position = undef;
1620   p.eye_position_z = undef;
1621   p.eye_position_normalized = undef;
1622   p.transformed_normal = undef;
1623   p.identity = undef;
1624   p.temp_in_use = 0;
1625   p.mvp_with_dp4 = mvp_with_dp4;
1626
1627   if (max_temps >= sizeof(int) * 8)
1628      p.temp_reserved = 0;
1629   else
1630      p.temp_reserved = ~((1<<max_temps)-1);
1631
1632   /* Start by allocating 32 instructions.
1633    * If we need more, we'll grow the instruction array as needed.
1634    */
1635   p.max_inst = 32;
1636   p.program->Base.Instructions = _mesa_alloc_instructions(p.max_inst);
1637   p.program->Base.String = NULL;
1638   p.program->Base.NumInstructions =
1639   p.program->Base.NumTemporaries =
1640   p.program->Base.NumParameters =
1641   p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1642   p.program->Base.Parameters = _mesa_new_parameter_list();
1643   p.program->Base.InputsRead = 0;
1644   p.program->Base.OutputsWritten = 0;
1645
1646   build_tnl_program( &p );
1647}
1648
1649
1650/**
1651 * Return a vertex program which implements the current fixed-function
1652 * transform/lighting/texgen operations.
1653 */
1654struct gl_vertex_program *
1655_mesa_get_fixed_func_vertex_program(struct gl_context *ctx)
1656{
1657   struct gl_vertex_program *prog;
1658   struct state_key key;
1659
1660   /* Grab all the relevent state and put it in a single structure:
1661    */
1662   make_state_key(ctx, &key);
1663
1664   /* Look for an already-prepared program for this state:
1665    */
1666   prog = gl_vertex_program(
1667      _mesa_search_program_cache(ctx->VertexProgram.Cache, &key, sizeof(key)));
1668
1669   if (!prog) {
1670      /* OK, we'll have to build a new one */
1671      if (0)
1672         printf("Build new TNL program\n");
1673
1674      prog = gl_vertex_program(ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0));
1675      if (!prog)
1676         return NULL;
1677
1678      create_new_program( &key, prog,
1679                          ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS,
1680                          ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps );
1681
1682#if 0
1683      if (ctx->Driver.ProgramStringNotify)
1684         ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1685                                          &prog->Base );
1686#endif
1687      _mesa_program_cache_insert(ctx, ctx->VertexProgram.Cache,
1688                                 &key, sizeof(key), &prog->Base);
1689   }
1690
1691   return prog;
1692}
1693