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/errors.h"
37#include "main/glheader.h"
38#include "main/mtypes.h"
39#include "main/macros.h"
40#include "main/enums.h"
41#include "main/context.h"
42#include "main/ffvertex_prog.h"
43#include "program/program.h"
44#include "program/prog_cache.h"
45#include "program/prog_instruction.h"
46#include "program/prog_parameter.h"
47#include "program/prog_print.h"
48#include "program/prog_statevars.h"
49#include "util/bitscan.h"
50
51#include "state_tracker/st_program.h"
52
53/** Max of number of lights and texture coord units */
54#define NUM_UNITS MAX2(MAX_TEXTURE_COORD_UNITS, MAX_LIGHTS)
55
56struct state_key {
57   GLbitfield varying_vp_inputs;
58
59   unsigned fragprog_inputs_read:12;
60
61   unsigned light_color_material_mask:12;
62   unsigned light_global_enabled:1;
63   unsigned light_local_viewer:1;
64   unsigned light_twoside:1;
65   unsigned material_shininess_is_zero:1;
66   unsigned need_eye_coords:1;
67   unsigned normalize:1;
68   unsigned rescale_normals:1;
69
70   unsigned fog_distance_mode:2;
71   unsigned separate_specular:1;
72   unsigned point_attenuated:1;
73
74   struct {
75      unsigned char light_enabled:1;
76      unsigned char light_eyepos3_is_zero:1;
77      unsigned char light_spotcutoff_is_180:1;
78      unsigned char light_attenuated:1;
79      unsigned char texmat_enabled:1;
80      unsigned char coord_replace:1;
81      unsigned char texgen_enabled:1;
82      unsigned char texgen_mode0:4;
83      unsigned char texgen_mode1:4;
84      unsigned char texgen_mode2:4;
85      unsigned char texgen_mode3:4;
86   } unit[NUM_UNITS];
87};
88
89
90#define TXG_NONE           0
91#define TXG_OBJ_LINEAR     1
92#define TXG_EYE_LINEAR     2
93#define TXG_SPHERE_MAP     3
94#define TXG_REFLECTION_MAP 4
95#define TXG_NORMAL_MAP     5
96
97static GLuint translate_texgen( GLboolean enabled, GLenum mode )
98{
99   if (!enabled)
100      return TXG_NONE;
101
102   switch (mode) {
103   case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
104   case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
105   case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
106   case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
107   case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
108   default: return TXG_NONE;
109   }
110}
111
112#define FDM_EYE_RADIAL    0
113#define FDM_EYE_PLANE     1
114#define FDM_EYE_PLANE_ABS 2
115#define FDM_FROM_ARRAY    3
116
117static GLuint translate_fog_distance_mode(GLenum source, GLenum mode)
118{
119   if (source == GL_FRAGMENT_DEPTH_EXT) {
120      switch (mode) {
121      case GL_EYE_RADIAL_NV:
122         return FDM_EYE_RADIAL;
123      case GL_EYE_PLANE:
124         return FDM_EYE_PLANE;
125      default: /* shouldn't happen; fall through to a sensible default */
126      case GL_EYE_PLANE_ABSOLUTE_NV:
127         return FDM_EYE_PLANE_ABS;
128      }
129   } else {
130      return FDM_FROM_ARRAY;
131   }
132}
133
134static GLboolean check_active_shininess( struct gl_context *ctx,
135                                         const struct state_key *key,
136                                         GLuint side )
137{
138   GLuint attr = MAT_ATTRIB_FRONT_SHININESS + side;
139
140   if ((key->varying_vp_inputs & VERT_BIT_COLOR0) &&
141       (key->light_color_material_mask & (1 << attr)))
142      return GL_TRUE;
143
144   if (key->varying_vp_inputs & VERT_BIT_MAT(attr))
145      return GL_TRUE;
146
147   if (ctx->Light.Material.Attrib[attr][0] != 0.0F)
148      return GL_TRUE;
149
150   return GL_FALSE;
151}
152
153
154static void make_state_key( struct gl_context *ctx, struct state_key *key )
155{
156   const struct gl_program *fp = ctx->FragmentProgram._Current;
157   GLbitfield mask;
158
159   memset(key, 0, sizeof(struct state_key));
160
161   if (_mesa_hw_select_enabled(ctx)) {
162      /* GL_SELECT mode only need position calculation.
163       * glBegin/End use VERT_BIT_SELECT_RESULT_OFFSET for multi name stack in one draw.
164       * glDrawArrays may also be called without user shader, fallback to FF one.
165       */
166      key->varying_vp_inputs = ctx->VertexProgram._VaryingInputs &
167         (VERT_BIT_POS | VERT_BIT_SELECT_RESULT_OFFSET);
168      return;
169   }
170
171   /* This now relies on texenvprogram.c being active:
172    */
173   assert(fp);
174
175   key->need_eye_coords = ctx->_NeedEyeCoords;
176
177   key->fragprog_inputs_read = fp->info.inputs_read;
178   key->varying_vp_inputs = ctx->VertexProgram._VaryingInputs;
179
180   if (ctx->RenderMode == GL_FEEDBACK) {
181      /* make sure the vertprog emits color and tex0 */
182      key->fragprog_inputs_read |= (VARYING_BIT_COL0 | VARYING_BIT_TEX0);
183   }
184
185   if (ctx->Light.Enabled) {
186      key->light_global_enabled = 1;
187
188      if (ctx->Light.Model.LocalViewer)
189	 key->light_local_viewer = 1;
190
191      if (ctx->Light.Model.TwoSide)
192	 key->light_twoside = 1;
193
194      if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)
195         key->separate_specular = 1;
196
197      if (ctx->Light.ColorMaterialEnabled) {
198	 key->light_color_material_mask = ctx->Light._ColorMaterialBitmask;
199      }
200
201      mask = ctx->Light._EnabledLights;
202      while (mask) {
203         const int i = u_bit_scan(&mask);
204         struct gl_light_uniforms *lu = &ctx->Light.LightSource[i];
205
206         key->unit[i].light_enabled = 1;
207
208         if (lu->EyePosition[3] == 0.0F)
209            key->unit[i].light_eyepos3_is_zero = 1;
210
211         if (lu->SpotCutoff == 180.0F)
212            key->unit[i].light_spotcutoff_is_180 = 1;
213
214         if (lu->ConstantAttenuation != 1.0F ||
215             lu->LinearAttenuation != 0.0F ||
216             lu->QuadraticAttenuation != 0.0F)
217            key->unit[i].light_attenuated = 1;
218      }
219
220      if (check_active_shininess(ctx, key, 0)) {
221         key->material_shininess_is_zero = 0;
222      }
223      else if (key->light_twoside &&
224               check_active_shininess(ctx, key, 1)) {
225         key->material_shininess_is_zero = 0;
226      }
227      else {
228         key->material_shininess_is_zero = 1;
229      }
230   }
231
232   if (ctx->Transform.Normalize)
233      key->normalize = 1;
234
235   if (ctx->Transform.RescaleNormals)
236      key->rescale_normals = 1;
237
238   /* Only distinguish fog parameters if we actually need */
239   if (key->fragprog_inputs_read & VARYING_BIT_FOGC)
240      key->fog_distance_mode =
241         translate_fog_distance_mode(ctx->Fog.FogCoordinateSource,
242                                     ctx->Fog.FogDistanceMode);
243
244   if (ctx->Point._Attenuated)
245      key->point_attenuated = 1;
246
247   mask = ctx->Texture._EnabledCoordUnits | ctx->Texture._TexGenEnabled
248      | ctx->Texture._TexMatEnabled | ctx->Point.CoordReplace;
249   while (mask) {
250      const int i = u_bit_scan(&mask);
251      struct gl_fixedfunc_texture_unit *texUnit =
252         &ctx->Texture.FixedFuncUnit[i];
253
254      if (ctx->Point.PointSprite)
255	 if (ctx->Point.CoordReplace & (1u << i))
256	    key->unit[i].coord_replace = 1;
257
258      if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
259	 key->unit[i].texmat_enabled = 1;
260
261      if (texUnit->TexGenEnabled) {
262	 key->unit[i].texgen_enabled = 1;
263
264	 key->unit[i].texgen_mode0 =
265	    translate_texgen( texUnit->TexGenEnabled & (1<<0),
266			      texUnit->GenS.Mode );
267	 key->unit[i].texgen_mode1 =
268	    translate_texgen( texUnit->TexGenEnabled & (1<<1),
269			      texUnit->GenT.Mode );
270	 key->unit[i].texgen_mode2 =
271	    translate_texgen( texUnit->TexGenEnabled & (1<<2),
272			      texUnit->GenR.Mode );
273	 key->unit[i].texgen_mode3 =
274	    translate_texgen( texUnit->TexGenEnabled & (1<<3),
275			      texUnit->GenQ.Mode );
276      }
277   }
278}
279
280
281
282/* Very useful debugging tool - produces annotated listing of
283 * generated program with line/function references for each
284 * instruction back into this file:
285 */
286#define DISASSEM 0
287
288
289/* Use uregs to represent registers internally, translate to Mesa's
290 * expected formats on emit.
291 *
292 * NOTE: These are passed by value extensively in this file rather
293 * than as usual by pointer reference.  If this disturbs you, try
294 * remembering they are just 32bits in size.
295 *
296 * GCC is smart enough to deal with these dword-sized structures in
297 * much the same way as if I had defined them as dwords and was using
298 * macros to access and set the fields.  This is much nicer and easier
299 * to evolve.
300 */
301struct ureg {
302   GLuint file:4;
303   GLint idx:9;      /* relative addressing may be negative */
304                     /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
305   GLuint negate:1;
306   GLuint swz:12;
307   GLuint pad:6;
308};
309
310
311struct tnl_program {
312   const struct state_key *state;
313   struct gl_program *program;
314   struct gl_program_parameter_list *state_params;
315   GLuint max_inst;  /** number of instructions allocated for program */
316   GLboolean mvp_with_dp4;
317
318   GLuint temp_in_use;
319   GLuint temp_reserved;
320
321   struct ureg eye_position;
322   struct ureg eye_position_z;
323   struct ureg eye_position_normalized;
324   struct ureg transformed_normal;
325   struct ureg identity;
326
327   GLuint materials;
328   GLuint color_materials;
329};
330
331
332static const struct ureg undef = {
333   PROGRAM_UNDEFINED,
334   0,
335   0,
336   0,
337   0
338};
339
340/* Local shorthand:
341 */
342#define X    SWIZZLE_X
343#define Y    SWIZZLE_Y
344#define Z    SWIZZLE_Z
345#define W    SWIZZLE_W
346
347
348/* Construct a ureg:
349 */
350static struct ureg make_ureg(GLuint file, GLint idx)
351{
352   struct ureg reg;
353   reg.file = file;
354   reg.idx = idx;
355   reg.negate = 0;
356   reg.swz = SWIZZLE_NOOP;
357   reg.pad = 0;
358   return reg;
359}
360
361
362static struct ureg negate( struct ureg reg )
363{
364   reg.negate ^= 1;
365   return reg;
366}
367
368
369static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
370{
371   reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
372			   GET_SWZ(reg.swz, y),
373			   GET_SWZ(reg.swz, z),
374			   GET_SWZ(reg.swz, w));
375   return reg;
376}
377
378
379static struct ureg swizzle1( struct ureg reg, int x )
380{
381   return swizzle(reg, x, x, x, x);
382}
383
384
385static struct ureg get_temp( struct tnl_program *p )
386{
387   int bit = ffs( ~p->temp_in_use );
388   if (!bit) {
389      _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
390      exit(1);
391   }
392
393   if ((GLuint) bit > p->program->arb.NumTemporaries)
394      p->program->arb.NumTemporaries = bit;
395
396   p->temp_in_use |= 1<<(bit-1);
397   return make_ureg(PROGRAM_TEMPORARY, bit-1);
398}
399
400
401static struct ureg reserve_temp( struct tnl_program *p )
402{
403   struct ureg temp = get_temp( p );
404   p->temp_reserved |= 1<<temp.idx;
405   return temp;
406}
407
408
409static void release_temp( struct tnl_program *p, struct ureg reg )
410{
411   if (reg.file == PROGRAM_TEMPORARY) {
412      p->temp_in_use &= ~(1<<reg.idx);
413      p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
414   }
415}
416
417static void release_temps( struct tnl_program *p )
418{
419   p->temp_in_use = p->temp_reserved;
420}
421
422
423static struct ureg register_param4(struct tnl_program *p,
424				   GLint s0,
425				   GLint s1,
426				   GLint s2,
427				   GLint s3)
428{
429   gl_state_index16 tokens[STATE_LENGTH];
430   GLint idx;
431   tokens[0] = s0;
432   tokens[1] = s1;
433   tokens[2] = s2;
434   tokens[3] = s3;
435   idx = _mesa_add_state_reference(p->state_params, tokens);
436   return make_ureg(PROGRAM_STATE_VAR, idx);
437}
438
439
440#define register_param1(p,s0)          register_param4(p,s0,0,0,0)
441#define register_param2(p,s0,s1)       register_param4(p,s0,s1,0,0)
442#define register_param3(p,s0,s1,s2)    register_param4(p,s0,s1,s2,0)
443
444
445
446/**
447 * \param input  one of VERT_ATTRIB_x tokens.
448 */
449static struct ureg register_input( struct tnl_program *p, GLuint input )
450{
451   assert(input < VERT_ATTRIB_MAX);
452
453   if (p->state->varying_vp_inputs & VERT_BIT(input)) {
454      p->program->info.inputs_read |= (uint64_t)VERT_BIT(input);
455      return make_ureg(PROGRAM_INPUT, input);
456   }
457   else {
458      return register_param2(p, STATE_CURRENT_ATTRIB, input);
459   }
460}
461
462
463/**
464 * \param input  one of VARYING_SLOT_x tokens.
465 */
466static struct ureg register_output( struct tnl_program *p, GLuint output )
467{
468   p->program->info.outputs_written |= BITFIELD64_BIT(output);
469   return make_ureg(PROGRAM_OUTPUT, output);
470}
471
472
473static struct ureg register_const4f( struct tnl_program *p,
474			      GLfloat s0,
475			      GLfloat s1,
476			      GLfloat s2,
477			      GLfloat s3)
478{
479   gl_constant_value values[4];
480   GLint idx;
481   GLuint swizzle;
482   values[0].f = s0;
483   values[1].f = s1;
484   values[2].f = s2;
485   values[3].f = s3;
486   idx = _mesa_add_unnamed_constant(p->program->Parameters, values, 4,
487                                    &swizzle );
488   assert(swizzle == SWIZZLE_NOOP);
489   return make_ureg(PROGRAM_CONSTANT, idx);
490}
491
492#define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
493#define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
494#define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
495#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
496
497static GLboolean is_undef( struct ureg reg )
498{
499   return reg.file == PROGRAM_UNDEFINED;
500}
501
502
503static struct ureg get_identity_param( struct tnl_program *p )
504{
505   if (is_undef(p->identity))
506      p->identity = register_const4f(p, 0,0,0,1);
507
508   return p->identity;
509}
510
511static void register_matrix_param5( struct tnl_program *p,
512				    GLint s0, /* modelview, projection, etc */
513				    GLint s1, /* texture matrix number */
514				    GLint s2, /* first row */
515				    GLint s3, /* last row */
516				    struct ureg *matrix )
517{
518   GLint i;
519
520   /* This is a bit sad as the support is there to pull the whole
521    * matrix out in one go:
522    */
523   for (i = 0; i <= s3 - s2; i++)
524      matrix[i] = register_param4(p, s0, s1, i, i);
525}
526
527
528static void emit_arg( struct prog_src_register *src,
529		      struct ureg reg )
530{
531   src->File = reg.file;
532   src->Index = reg.idx;
533   src->Swizzle = reg.swz;
534   src->Negate = reg.negate ? NEGATE_XYZW : NEGATE_NONE;
535   src->RelAddr = 0;
536   /* Check that bitfield sizes aren't exceeded */
537   assert(src->Index == reg.idx);
538}
539
540
541static void emit_dst( struct prog_dst_register *dst,
542		      struct ureg reg, GLuint mask )
543{
544   dst->File = reg.file;
545   dst->Index = reg.idx;
546   /* allow zero as a shorthand for xyzw */
547   dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
548   /* Check that bitfield sizes aren't exceeded */
549   assert(dst->Index == reg.idx);
550}
551
552
553static void debug_insn( struct prog_instruction *inst, const char *fn,
554			GLuint line )
555{
556   if (DISASSEM) {
557      static const char *last_fn;
558
559      if (fn != last_fn) {
560	 last_fn = fn;
561	 printf("%s:\n", fn);
562      }
563
564      printf("%d:\t", line);
565      _mesa_print_instruction(inst);
566   }
567}
568
569
570static void emit_op3fn(struct tnl_program *p,
571                       enum prog_opcode op,
572		       struct ureg dest,
573		       GLuint mask,
574		       struct ureg src0,
575		       struct ureg src1,
576		       struct ureg src2,
577		       const char *fn,
578		       GLuint line)
579{
580   GLuint nr;
581   struct prog_instruction *inst;
582
583   assert(p->program->arb.NumInstructions <= p->max_inst);
584
585   if (p->program->arb.NumInstructions == p->max_inst) {
586      /* need to extend the program's instruction array */
587      struct prog_instruction *newInst;
588
589      /* double the size */
590      p->max_inst *= 2;
591
592      newInst =
593         rzalloc_array(p->program, struct prog_instruction, p->max_inst);
594      if (!newInst) {
595         _mesa_error(NULL, GL_OUT_OF_MEMORY, "vertex program build");
596         return;
597      }
598
599      _mesa_copy_instructions(newInst, p->program->arb.Instructions,
600                              p->program->arb.NumInstructions);
601
602      ralloc_free(p->program->arb.Instructions);
603
604      p->program->arb.Instructions = newInst;
605   }
606
607   nr = p->program->arb.NumInstructions++;
608
609   inst = &p->program->arb.Instructions[nr];
610   inst->Opcode = (enum prog_opcode) op;
611
612   emit_arg( &inst->SrcReg[0], src0 );
613   emit_arg( &inst->SrcReg[1], src1 );
614   emit_arg( &inst->SrcReg[2], src2 );
615
616   emit_dst( &inst->DstReg, dest, mask );
617
618   debug_insn(inst, fn, line);
619}
620
621
622#define emit_op3(p, op, dst, mask, src0, src1, src2) \
623   emit_op3fn(p, op, dst, mask, src0, src1, src2, __func__, __LINE__)
624
625#define emit_op2(p, op, dst, mask, src0, src1) \
626    emit_op3fn(p, op, dst, mask, src0, src1, undef, __func__, __LINE__)
627
628#define emit_op1(p, op, dst, mask, src0) \
629    emit_op3fn(p, op, dst, mask, src0, undef, undef, __func__, __LINE__)
630
631
632static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
633{
634   if (reg.file == PROGRAM_TEMPORARY &&
635       !(p->temp_reserved & (1<<reg.idx)))
636      return reg;
637   else {
638      struct ureg temp = get_temp(p);
639      emit_op1(p, OPCODE_MOV, temp, 0, reg);
640      return temp;
641   }
642}
643
644
645/* Currently no tracking performed of input/output/register size or
646 * active elements.  Could be used to reduce these operations, as
647 * could the matrix type.
648 */
649static void emit_matrix_transform_vec4( struct tnl_program *p,
650					struct ureg dest,
651					const struct ureg *mat,
652					struct ureg src)
653{
654   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
655   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
656   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
657   emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
658}
659
660
661/* This version is much easier to implement if writemasks are not
662 * supported natively on the target or (like SSE), the target doesn't
663 * have a clean/obvious dotproduct implementation.
664 */
665static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
666						  struct ureg dest,
667						  const struct ureg *mat,
668						  struct ureg src)
669{
670   struct ureg tmp;
671
672   if (dest.file != PROGRAM_TEMPORARY)
673      tmp = get_temp(p);
674   else
675      tmp = dest;
676
677   emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
678   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
679   emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
680   emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
681
682   if (dest.file != PROGRAM_TEMPORARY)
683      release_temp(p, tmp);
684}
685
686
687static void emit_matrix_transform_vec3( struct tnl_program *p,
688					struct ureg dest,
689					const struct ureg *mat,
690					struct ureg src)
691{
692   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
693   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
694   emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
695}
696
697
698static void emit_normalize_vec3( struct tnl_program *p,
699				 struct ureg dest,
700				 struct ureg src )
701{
702   struct ureg tmp = get_temp(p);
703   emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src);
704   emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
705   emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X));
706   release_temp(p, tmp);
707}
708
709
710static void emit_passthrough( struct tnl_program *p,
711			      GLuint input,
712			      GLuint output )
713{
714   struct ureg out = register_output(p, output);
715   emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
716}
717
718
719static struct ureg get_eye_position( struct tnl_program *p )
720{
721   if (is_undef(p->eye_position)) {
722      struct ureg pos = register_input( p, VERT_ATTRIB_POS );
723      struct ureg modelview[4];
724
725      p->eye_position = reserve_temp(p);
726
727      if (p->mvp_with_dp4) {
728	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
729                                 modelview );
730
731	 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
732      }
733      else {
734	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX_TRANSPOSE, 0, 0, 3,
735				 modelview );
736
737	 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
738      }
739   }
740
741   return p->eye_position;
742}
743
744
745static struct ureg get_eye_position_z( struct tnl_program *p )
746{
747   if (!is_undef(p->eye_position))
748      return swizzle1(p->eye_position, Z);
749
750   if (is_undef(p->eye_position_z)) {
751      struct ureg pos = register_input( p, VERT_ATTRIB_POS );
752      struct ureg modelview[4];
753
754      p->eye_position_z = reserve_temp(p);
755
756      register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
757                              modelview );
758
759      emit_op2(p, OPCODE_DP4, p->eye_position_z, 0, pos, modelview[2]);
760   }
761
762   return p->eye_position_z;
763}
764
765
766static struct ureg get_eye_position_normalized( struct tnl_program *p )
767{
768   if (is_undef(p->eye_position_normalized)) {
769      struct ureg eye = get_eye_position(p);
770      p->eye_position_normalized = reserve_temp(p);
771      emit_normalize_vec3(p, p->eye_position_normalized, eye);
772   }
773
774   return p->eye_position_normalized;
775}
776
777
778static struct ureg get_transformed_normal( struct tnl_program *p )
779{
780   if (is_undef(p->transformed_normal) &&
781       !p->state->need_eye_coords &&
782       !p->state->normalize &&
783       !(p->state->need_eye_coords == p->state->rescale_normals))
784   {
785      p->transformed_normal = register_input(p, VERT_ATTRIB_NORMAL );
786   }
787   else if (is_undef(p->transformed_normal))
788   {
789      struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
790      struct ureg mvinv[3];
791      struct ureg transformed_normal = reserve_temp(p);
792
793      if (p->state->need_eye_coords) {
794         register_matrix_param5( p, STATE_MODELVIEW_MATRIX_INVTRANS, 0, 0, 2,
795                                 mvinv );
796
797         /* Transform to eye space:
798          */
799         emit_matrix_transform_vec3( p, transformed_normal, mvinv, normal );
800         normal = transformed_normal;
801      }
802
803      /* Normalize/Rescale:
804       */
805      if (p->state->normalize) {
806	 emit_normalize_vec3( p, transformed_normal, normal );
807         normal = transformed_normal;
808      }
809      else if (p->state->need_eye_coords == p->state->rescale_normals) {
810         /* This is already adjusted for eye/non-eye rendering:
811          */
812	 struct ureg rescale = register_param1(p, 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			      mvp );
835      emit_matrix_transform_vec4( p, hpos, mvp, pos );
836   }
837   else {
838      register_matrix_param5( p, STATE_MVP_MATRIX_TRANSPOSE, 0, 0, 3,
839			      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   switch (property) {
848   case STATE_AMBIENT:
849      return MAT_ATTRIB_FRONT_AMBIENT + side;
850   case STATE_DIFFUSE:
851      return MAT_ATTRIB_FRONT_DIFFUSE + side;
852   case STATE_SPECULAR:
853      return MAT_ATTRIB_FRONT_SPECULAR + side;
854   case STATE_EMISSION:
855      return MAT_ATTRIB_FRONT_EMISSION + side;
856   case STATE_SHININESS:
857      return MAT_ATTRIB_FRONT_SHININESS + side;
858   default:
859      unreachable("invalid value");
860   }
861}
862
863
864/**
865 * Get a bitmask of which material values vary on a per-vertex basis.
866 */
867static void set_material_flags( struct tnl_program *p )
868{
869   p->color_materials = 0;
870   p->materials = 0;
871
872   if (p->state->varying_vp_inputs & VERT_BIT_COLOR0) {
873      p->materials =
874	 p->color_materials = p->state->light_color_material_mask;
875   }
876
877   p->materials |= ((p->state->varying_vp_inputs & VERT_BIT_MAT_ALL)
878                    >> VERT_ATTRIB_MAT(0));
879}
880
881
882static struct ureg get_material( struct tnl_program *p, GLuint side,
883				 GLuint property )
884{
885   GLuint attrib = material_attrib(side, property);
886
887   if (p->color_materials & (1<<attrib))
888      return register_input(p, VERT_ATTRIB_COLOR0);
889   else if (p->materials & (1<<attrib)) {
890      /* Put material values in the GENERIC slots -- they are not used
891       * for anything in fixed function mode.
892       */
893      return register_input( p, VERT_ATTRIB_MAT(attrib) );
894   }
895   else
896      return register_param2(p, STATE_MATERIAL, attrib);
897}
898
899#define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
900				   MAT_BIT_FRONT_AMBIENT | \
901				   MAT_BIT_FRONT_DIFFUSE) << (side))
902
903
904/**
905 * Either return a precalculated constant value or emit code to
906 * calculate these values dynamically in the case where material calls
907 * are present between begin/end pairs.
908 *
909 * Probably want to shift this to the program compilation phase - if
910 * we always emitted the calculation here, a smart compiler could
911 * detect that it was constant (given a certain set of inputs), and
912 * lift it out of the main loop.  That way the programs created here
913 * would be independent of the vertex_buffer details.
914 */
915static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
916{
917   if (p->materials & SCENE_COLOR_BITS(side)) {
918      struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
919      struct ureg material_emission = get_material(p, side, STATE_EMISSION);
920      struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
921      struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
922      struct ureg tmp = make_temp(p, material_diffuse);
923      emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
924	       material_ambient, material_emission);
925      return tmp;
926   }
927   else
928      return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
929}
930
931
932static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
933				  GLuint side, GLuint property, bool *is_state_light )
934{
935   GLuint attrib = material_attrib(side, property);
936   if (p->materials & (1<<attrib)) {
937      struct ureg light_value =
938	 register_param3(p, STATE_LIGHT, light, property);
939    *is_state_light = true;
940    return light_value;
941   }
942   else {
943      *is_state_light = false;
944      return register_param3(p, STATE_LIGHTPROD, light, attrib);
945   }
946}
947
948
949static struct ureg calculate_light_attenuation( struct tnl_program *p,
950						GLuint i,
951						struct ureg VPpli,
952						struct ureg dist )
953{
954   struct ureg attenuation = undef;
955   struct ureg att = undef;
956
957   /* Calculate spot attenuation:
958    */
959   if (!p->state->unit[i].light_spotcutoff_is_180) {
960      struct ureg spot_dir_norm = register_param2(p, STATE_LIGHT_SPOT_DIR_NORMALIZED, i);
961      struct ureg spot = get_temp(p);
962      struct ureg slt = get_temp(p);
963
964      attenuation = register_param3(p, STATE_LIGHT, i, STATE_ATTENUATION);
965      att = get_temp(p);
966
967      emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
968      emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
969      emit_op1(p, OPCODE_ABS, spot, 0, spot);
970      emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
971      emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
972
973      release_temp(p, spot);
974      release_temp(p, slt);
975   }
976
977   /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
978    *
979    * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
980    */
981   if (p->state->unit[i].light_attenuated && !is_undef(dist)) {
982      if (is_undef(att))
983         att = get_temp(p);
984
985      if (is_undef(attenuation))
986         attenuation = register_param3(p, STATE_LIGHT, i, STATE_ATTENUATION);
987
988      /* 1/d,d,d,1/d */
989      emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
990      /* 1,d,d*d,1/d */
991      emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
992      /* 1/dist-atten */
993      emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
994
995      if (!p->state->unit[i].light_spotcutoff_is_180) {
996	 /* dist-atten */
997	 emit_op1(p, OPCODE_RCP, dist, 0, dist);
998	 /* spot-atten * dist-atten */
999	 emit_op2(p, OPCODE_MUL, att, 0, dist, att);
1000      }
1001      else {
1002	 /* dist-atten */
1003	 emit_op1(p, OPCODE_RCP, att, 0, dist);
1004      }
1005   }
1006
1007   return att;
1008}
1009
1010
1011/**
1012 * Compute:
1013 *   lit.y = MAX(0, dots.x)
1014 *   lit.z = SLT(0, dots.x)
1015 */
1016static void emit_degenerate_lit( struct tnl_program *p,
1017                                 struct ureg lit,
1018                                 struct ureg dots )
1019{
1020   struct ureg id = get_identity_param(p);  /* id = {0,0,0,1} */
1021
1022   /* Note that lit.x & lit.w will not be examined.  Note also that
1023    * dots.xyzw == dots.xxxx.
1024    */
1025
1026   /* MAX lit, id, dots;
1027    */
1028   emit_op2(p, OPCODE_MAX, lit, WRITEMASK_XYZW, id, dots);
1029
1030   /* result[2] = (in > 0 ? 1 : 0)
1031    * SLT lit.z, id.z, dots;   # lit.z = (0 < dots.z) ? 1 : 0
1032    */
1033   emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, swizzle1(id,Z), dots);
1034}
1035
1036
1037/* Need to add some addtional parameters to allow lighting in object
1038 * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1039 * space lighting.
1040 */
1041static void build_lighting( struct tnl_program *p )
1042{
1043   const GLboolean twoside = p->state->light_twoside;
1044   const GLboolean separate = p->state->separate_specular;
1045   GLuint nr_lights = 0, count = 0;
1046   struct ureg normal = get_transformed_normal(p);
1047   struct ureg lit = get_temp(p);
1048   struct ureg dots = get_temp(p);
1049   struct ureg _col0 = undef, _col1 = undef;
1050   struct ureg _bfc0 = undef, _bfc1 = undef;
1051   GLuint i;
1052
1053   /*
1054    * NOTE:
1055    * dots.x = dot(normal, VPpli)
1056    * dots.y = dot(normal, halfAngle)
1057    * dots.z = back.shininess
1058    * dots.w = front.shininess
1059    */
1060
1061   for (i = 0; i < MAX_LIGHTS; i++)
1062      if (p->state->unit[i].light_enabled)
1063	 nr_lights++;
1064
1065   set_material_flags(p);
1066
1067   {
1068      if (!p->state->material_shininess_is_zero) {
1069         struct ureg shininess = get_material(p, 0, STATE_SHININESS);
1070         emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
1071         release_temp(p, shininess);
1072      }
1073
1074      _col0 = make_temp(p, get_scenecolor(p, 0));
1075      if (separate)
1076	 _col1 = make_temp(p, get_identity_param(p));
1077      else
1078	 _col1 = _col0;
1079   }
1080
1081   if (twoside) {
1082      if (!p->state->material_shininess_is_zero) {
1083         /* Note that we negate the back-face specular exponent here.
1084          * The negation will be un-done later in the back-face code below.
1085          */
1086         struct ureg shininess = get_material(p, 1, STATE_SHININESS);
1087         emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
1088                  negate(swizzle1(shininess,X)));
1089         release_temp(p, shininess);
1090      }
1091
1092      _bfc0 = make_temp(p, get_scenecolor(p, 1));
1093      if (separate)
1094	 _bfc1 = make_temp(p, get_identity_param(p));
1095      else
1096	 _bfc1 = _bfc0;
1097   }
1098
1099   /* If no lights, still need to emit the scenecolor.
1100    */
1101   {
1102      struct ureg res0 = register_output( p, VARYING_SLOT_COL0 );
1103      emit_op1(p, OPCODE_MOV, res0, 0, _col0);
1104   }
1105
1106   if (separate) {
1107      struct ureg res1 = register_output( p, VARYING_SLOT_COL1 );
1108      emit_op1(p, OPCODE_MOV, res1, 0, _col1);
1109   }
1110
1111   if (twoside) {
1112      struct ureg res0 = register_output( p, VARYING_SLOT_BFC0 );
1113      emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
1114   }
1115
1116   if (twoside && separate) {
1117      struct ureg res1 = register_output( p, VARYING_SLOT_BFC1 );
1118      emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
1119   }
1120
1121   if (nr_lights == 0) {
1122      release_temps(p);
1123      return;
1124   }
1125
1126   /* Declare light products first to place them sequentially next to each
1127    * other for optimal constant uploads.
1128    */
1129   struct ureg lightprod_front[MAX_LIGHTS][3];
1130   struct ureg lightprod_back[MAX_LIGHTS][3];
1131   bool lightprod_front_is_state_light[MAX_LIGHTS][3];
1132   bool lightprod_back_is_state_light[MAX_LIGHTS][3];
1133
1134   for (i = 0; i < MAX_LIGHTS; i++) {
1135      if (p->state->unit[i].light_enabled) {
1136         lightprod_front[i][0] = get_lightprod(p, i, 0, STATE_AMBIENT,
1137                                               &lightprod_front_is_state_light[i][0]);
1138         if (twoside)
1139            lightprod_back[i][0] = get_lightprod(p, i, 1, STATE_AMBIENT,
1140                                                 &lightprod_back_is_state_light[i][0]);
1141
1142         lightprod_front[i][1] = get_lightprod(p, i, 0, STATE_DIFFUSE,
1143                                               &lightprod_front_is_state_light[i][1]);
1144         if (twoside)
1145            lightprod_back[i][1] = get_lightprod(p, i, 1, STATE_DIFFUSE,
1146                                                 &lightprod_back_is_state_light[i][1]);
1147
1148         lightprod_front[i][2] = get_lightprod(p, i, 0, STATE_SPECULAR,
1149                                               &lightprod_front_is_state_light[i][2]);
1150         if (twoside)
1151            lightprod_back[i][2] = get_lightprod(p, i, 1, STATE_SPECULAR,
1152                                                 &lightprod_back_is_state_light[i][2]);
1153      }
1154   }
1155
1156   /* Add more variables now that we'll use later, so that they are nicely
1157    * sorted in the parameter list.
1158    */
1159   for (i = 0; i < MAX_LIGHTS; i++) {
1160      if (p->state->unit[i].light_enabled) {
1161         if (p->state->unit[i].light_eyepos3_is_zero)
1162            register_param2(p, STATE_LIGHT_POSITION_NORMALIZED, i);
1163         else
1164            register_param2(p, STATE_LIGHT_POSITION, i);
1165      }
1166   }
1167   for (i = 0; i < MAX_LIGHTS; i++) {
1168      if (p->state->unit[i].light_enabled &&
1169          (!p->state->unit[i].light_spotcutoff_is_180 ||
1170           (p->state->unit[i].light_attenuated &&
1171            !p->state->unit[i].light_eyepos3_is_zero)))
1172         register_param3(p, STATE_LIGHT, i, STATE_ATTENUATION);
1173   }
1174
1175   for (i = 0; i < MAX_LIGHTS; i++) {
1176      if (p->state->unit[i].light_enabled) {
1177	 struct ureg half = undef;
1178	 struct ureg att = undef, VPpli = undef;
1179	 struct ureg dist = undef;
1180
1181	 count++;
1182         if (p->state->unit[i].light_eyepos3_is_zero) {
1183             VPpli = register_param2(p, STATE_LIGHT_POSITION_NORMALIZED, i);
1184         } else {
1185            struct ureg Ppli = register_param2(p, STATE_LIGHT_POSITION, i);
1186            struct ureg V = get_eye_position(p);
1187
1188            VPpli = get_temp(p);
1189            dist = get_temp(p);
1190
1191            /* Calculate VPpli vector
1192             */
1193            emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
1194
1195            /* Normalize VPpli.  The dist value also used in
1196             * attenuation below.
1197             */
1198            emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
1199            emit_op1(p, OPCODE_RSQ, dist, 0, dist);
1200            emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
1201         }
1202
1203         /* Calculate attenuation:
1204          */
1205         att = calculate_light_attenuation(p, i, VPpli, dist);
1206         release_temp(p, dist);
1207
1208	 /* Calculate viewer direction, or use infinite viewer:
1209	  */
1210         if (!p->state->material_shininess_is_zero) {
1211            if (p->state->light_local_viewer) {
1212               struct ureg eye_hat = get_eye_position_normalized(p);
1213               half = get_temp(p);
1214               emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1215               emit_normalize_vec3(p, half, half);
1216            } else if (p->state->unit[i].light_eyepos3_is_zero) {
1217               half = register_param2(p, STATE_LIGHT_HALF_VECTOR, i);
1218            } else {
1219               struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1220               half = get_temp(p);
1221               emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1222               emit_normalize_vec3(p, half, half);
1223            }
1224	 }
1225
1226	 /* Calculate dot products:
1227	  */
1228         if (p->state->material_shininess_is_zero) {
1229            emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli);
1230         }
1231         else {
1232            emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1233            emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1234         }
1235
1236	 /* Front face lighting:
1237	  */
1238	 {
1239      /* Transform STATE_LIGHT into STATE_LIGHTPROD if needed. This isn't done in
1240       * get_lightprod to avoid using too many temps.
1241       */
1242      for (int j = 0; j < 3; j++) {
1243         if (lightprod_front_is_state_light[i][j]) {
1244            struct ureg material_value = get_material(p, 0, STATE_AMBIENT + j);
1245            struct ureg tmp = get_temp(p);
1246            emit_op2(p, OPCODE_MUL, tmp, 0, lightprod_front[i][j], material_value);
1247            lightprod_front[i][j] = tmp;
1248         }
1249      }
1250
1251	    struct ureg ambient = lightprod_front[i][0];
1252	    struct ureg diffuse = lightprod_front[i][1];
1253	    struct ureg specular = lightprod_front[i][2];
1254	    struct ureg res0, res1;
1255	    GLuint mask0, mask1;
1256
1257	    if (count == nr_lights) {
1258	       if (separate) {
1259		  mask0 = WRITEMASK_XYZ;
1260		  mask1 = WRITEMASK_XYZ;
1261		  res0 = register_output( p, VARYING_SLOT_COL0 );
1262		  res1 = register_output( p, VARYING_SLOT_COL1 );
1263	       }
1264	       else {
1265		  mask0 = 0;
1266		  mask1 = WRITEMASK_XYZ;
1267		  res0 = _col0;
1268		  res1 = register_output( p, VARYING_SLOT_COL0 );
1269	       }
1270	    }
1271            else {
1272	       mask0 = 0;
1273	       mask1 = 0;
1274	       res0 = _col0;
1275	       res1 = _col1;
1276	    }
1277
1278	    if (!is_undef(att)) {
1279               /* light is attenuated by distance */
1280               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1281               emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1282               emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1283            }
1284            else if (!p->state->material_shininess_is_zero) {
1285               /* there's a non-zero specular term */
1286               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1287               emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1288            }
1289            else {
1290               /* no attenutation, no specular */
1291               emit_degenerate_lit(p, lit, dots);
1292               emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1293            }
1294
1295	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1296	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1297
1298	    release_temp(p, ambient);
1299	    release_temp(p, diffuse);
1300	    release_temp(p, specular);
1301	 }
1302
1303	 /* Back face lighting:
1304	  */
1305	 if (twoside) {
1306      /* Transform STATE_LIGHT into STATE_LIGHTPROD if needed. This isn't done in
1307       * get_lightprod to avoid using too many temps.
1308       */
1309      for (int j = 0; j < 3; j++) {
1310         if (lightprod_back_is_state_light[i][j]) {
1311            struct ureg material_value = get_material(p, 1, STATE_AMBIENT + j);
1312            struct ureg tmp = get_temp(p);
1313            emit_op2(p, OPCODE_MUL, tmp, 1, lightprod_back[i][j], material_value);
1314            lightprod_back[i][j] = tmp;
1315         }
1316      }
1317
1318	    struct ureg ambient = lightprod_back[i][0];
1319	    struct ureg diffuse = lightprod_back[i][1];
1320	    struct ureg specular = lightprod_back[i][2];
1321	    struct ureg res0, res1;
1322	    GLuint mask0, mask1;
1323
1324	    if (count == nr_lights) {
1325	       if (separate) {
1326		  mask0 = WRITEMASK_XYZ;
1327		  mask1 = WRITEMASK_XYZ;
1328		  res0 = register_output( p, VARYING_SLOT_BFC0 );
1329		  res1 = register_output( p, VARYING_SLOT_BFC1 );
1330	       }
1331	       else {
1332		  mask0 = 0;
1333		  mask1 = WRITEMASK_XYZ;
1334		  res0 = _bfc0;
1335		  res1 = register_output( p, VARYING_SLOT_BFC0 );
1336	       }
1337	    }
1338            else {
1339	       res0 = _bfc0;
1340	       res1 = _bfc1;
1341	       mask0 = 0;
1342	       mask1 = 0;
1343	    }
1344
1345            /* For the back face we need to negate the X and Y component
1346             * dot products.  dots.Z has the negated back-face specular
1347             * exponent.  We swizzle that into the W position.  This
1348             * negation makes the back-face specular term positive again.
1349             */
1350            dots = negate(swizzle(dots,X,Y,W,Z));
1351
1352	    if (!is_undef(att)) {
1353               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1354	       emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1355               emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1356            }
1357            else if (!p->state->material_shininess_is_zero) {
1358               emit_op1(p, OPCODE_LIT, lit, 0, dots);
1359               emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); /**/
1360            }
1361            else {
1362               emit_degenerate_lit(p, lit, dots);
1363               emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
1364            }
1365
1366	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1367	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1368            /* restore dots to its original state for subsequent lights
1369             * by negating and swizzling again.
1370             */
1371            dots = negate(swizzle(dots,X,Y,W,Z));
1372
1373	    release_temp(p, ambient);
1374	    release_temp(p, diffuse);
1375	    release_temp(p, specular);
1376	 }
1377
1378	 release_temp(p, half);
1379	 release_temp(p, VPpli);
1380	 release_temp(p, att);
1381      }
1382   }
1383
1384   release_temps( p );
1385}
1386
1387
1388static void build_fog( struct tnl_program *p )
1389{
1390   struct ureg fog = register_output(p, VARYING_SLOT_FOGC);
1391   struct ureg input;
1392
1393   switch (p->state->fog_distance_mode) {
1394   case FDM_EYE_RADIAL: { /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1395      struct ureg tmp = get_temp(p);
1396      input = get_eye_position(p);
1397      emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, input, input);
1398      emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
1399      emit_op1(p, OPCODE_RCP, fog, WRITEMASK_X, tmp);
1400      break;
1401   }
1402   case FDM_EYE_PLANE: /* Z = Ze */
1403      input = get_eye_position_z(p);
1404      emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1405      break;
1406   case FDM_EYE_PLANE_ABS: /* Z = abs(Ze) */
1407      input = get_eye_position_z(p);
1408      emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1409      break;
1410   case FDM_FROM_ARRAY:
1411      input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1412      emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1413      break;
1414   default:
1415      assert(!"Bad fog mode in build_fog()");
1416      break;
1417   }
1418
1419   emit_op1(p, OPCODE_MOV, fog, WRITEMASK_YZW, get_identity_param(p));
1420}
1421
1422
1423static void build_reflect_texgen( struct tnl_program *p,
1424				  struct ureg dest,
1425				  GLuint writemask )
1426{
1427   struct ureg normal = get_transformed_normal(p);
1428   struct ureg eye_hat = get_eye_position_normalized(p);
1429   struct ureg tmp = get_temp(p);
1430
1431   /* n.u */
1432   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1433   /* 2n.u */
1434   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1435   /* (-2n.u)n + u */
1436   emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1437
1438   release_temp(p, tmp);
1439}
1440
1441
1442static void build_sphere_texgen( struct tnl_program *p,
1443				 struct ureg dest,
1444				 GLuint writemask )
1445{
1446   struct ureg normal = get_transformed_normal(p);
1447   struct ureg eye_hat = get_eye_position_normalized(p);
1448   struct ureg tmp = get_temp(p);
1449   struct ureg half = register_scalar_const(p, .5);
1450   struct ureg r = get_temp(p);
1451   struct ureg inv_m = get_temp(p);
1452   struct ureg id = get_identity_param(p);
1453
1454   /* Could share the above calculations, but it would be
1455    * a fairly odd state for someone to set (both sphere and
1456    * reflection active for different texture coordinate
1457    * components.  Of course - if two texture units enable
1458    * reflect and/or sphere, things start to tilt in favour
1459    * of seperating this out:
1460    */
1461
1462   /* n.u */
1463   emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1464   /* 2n.u */
1465   emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1466   /* (-2n.u)n + u */
1467   emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1468   /* r + 0,0,1 */
1469   emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1470   /* rx^2 + ry^2 + (rz+1)^2 */
1471   emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1472   /* 2/m */
1473   emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1474   /* 1/m */
1475   emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1476   /* r/m + 1/2 */
1477   emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1478
1479   release_temp(p, tmp);
1480   release_temp(p, r);
1481   release_temp(p, inv_m);
1482}
1483
1484
1485static void build_texture_transform( struct tnl_program *p )
1486{
1487   GLuint i, j;
1488
1489   for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
1490
1491      if (!(p->state->fragprog_inputs_read & VARYING_BIT_TEX(i)))
1492	 continue;
1493
1494      if (p->state->unit[i].coord_replace)
1495  	 continue;
1496
1497      if (p->state->unit[i].texgen_enabled ||
1498	  p->state->unit[i].texmat_enabled) {
1499
1500	 GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1501	 struct ureg out = register_output(p, VARYING_SLOT_TEX0 + i);
1502	 struct ureg out_texgen = undef;
1503
1504	 if (p->state->unit[i].texgen_enabled) {
1505	    GLuint copy_mask = 0;
1506	    GLuint sphere_mask = 0;
1507	    GLuint reflect_mask = 0;
1508	    GLuint normal_mask = 0;
1509	    GLuint modes[4];
1510
1511	    if (texmat_enabled)
1512	       out_texgen = get_temp(p);
1513	    else
1514	       out_texgen = out;
1515
1516	    modes[0] = p->state->unit[i].texgen_mode0;
1517	    modes[1] = p->state->unit[i].texgen_mode1;
1518	    modes[2] = p->state->unit[i].texgen_mode2;
1519	    modes[3] = p->state->unit[i].texgen_mode3;
1520
1521	    for (j = 0; j < 4; j++) {
1522	       switch (modes[j]) {
1523	       case TXG_OBJ_LINEAR: {
1524		  struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1525		  struct ureg plane =
1526		     register_param3(p, STATE_TEXGEN, i,
1527				     STATE_TEXGEN_OBJECT_S + j);
1528
1529		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1530			   obj, plane );
1531		  break;
1532	       }
1533	       case TXG_EYE_LINEAR: {
1534		  struct ureg eye = get_eye_position(p);
1535		  struct ureg plane =
1536		     register_param3(p, STATE_TEXGEN, i,
1537				     STATE_TEXGEN_EYE_S + j);
1538
1539		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1540			   eye, plane );
1541		  break;
1542	       }
1543	       case TXG_SPHERE_MAP:
1544		  sphere_mask |= WRITEMASK_X << j;
1545		  break;
1546	       case TXG_REFLECTION_MAP:
1547		  reflect_mask |= WRITEMASK_X << j;
1548		  break;
1549	       case TXG_NORMAL_MAP:
1550		  normal_mask |= WRITEMASK_X << j;
1551		  break;
1552	       case TXG_NONE:
1553		  copy_mask |= WRITEMASK_X << j;
1554	       }
1555	    }
1556
1557	    if (sphere_mask) {
1558	       build_sphere_texgen(p, out_texgen, sphere_mask);
1559	    }
1560
1561	    if (reflect_mask) {
1562	       build_reflect_texgen(p, out_texgen, reflect_mask);
1563	    }
1564
1565	    if (normal_mask) {
1566	       struct ureg normal = get_transformed_normal(p);
1567	       emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1568	    }
1569
1570	    if (copy_mask) {
1571	       struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1572	       emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1573	    }
1574	 }
1575
1576	 if (texmat_enabled) {
1577	    struct ureg texmat[4];
1578	    struct ureg in = (!is_undef(out_texgen) ?
1579			      out_texgen :
1580			      register_input(p, VERT_ATTRIB_TEX0+i));
1581	    if (p->mvp_with_dp4) {
1582	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1583				       texmat );
1584	       emit_matrix_transform_vec4( p, out, texmat, in );
1585	    }
1586	    else {
1587	       register_matrix_param5( p, STATE_TEXTURE_MATRIX_TRANSPOSE, i, 0, 3,
1588				       texmat );
1589	       emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1590	    }
1591	 }
1592
1593	 release_temps(p);
1594      }
1595      else {
1596	 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VARYING_SLOT_TEX0+i);
1597      }
1598   }
1599}
1600
1601
1602/**
1603 * Point size attenuation computation.
1604 */
1605static void build_atten_pointsize( struct tnl_program *p )
1606{
1607   struct ureg eye = get_eye_position_z(p);
1608   struct ureg state_size = register_param1(p, STATE_POINT_SIZE_CLAMPED);
1609   struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1610   struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1611   struct ureg ut = get_temp(p);
1612
1613   /* dist = |eyez| */
1614   emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
1615   /* p1 + dist * (p2 + dist * p3); */
1616   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1617		swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1618   emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1619		ut, swizzle1(state_attenuation, X));
1620
1621   /* 1 / sqrt(factor) */
1622   emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1623
1624#if 0
1625   /* out = pointSize / sqrt(factor) */
1626   emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1627#else
1628   /* this is a good place to clamp the point size since there's likely
1629    * no hardware registers to clamp point size at rasterization time.
1630    */
1631   emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1632   emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1633   emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1634#endif
1635
1636   release_temp(p, ut);
1637}
1638
1639
1640/**
1641 * Pass-though per-vertex point size, from user's point size array.
1642 */
1643static void build_array_pointsize( struct tnl_program *p )
1644{
1645   struct ureg in = register_input(p, VERT_ATTRIB_POINT_SIZE);
1646   struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1647   emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, in);
1648}
1649
1650
1651static void build_tnl_program( struct tnl_program *p )
1652{
1653   /* Emit the program, starting with the modelview, projection transforms:
1654    */
1655   build_hpos(p);
1656
1657   /* Lighting calculations:
1658    */
1659   if (p->state->fragprog_inputs_read & (VARYING_BIT_COL0|VARYING_BIT_COL1)) {
1660      if (p->state->light_global_enabled)
1661	 build_lighting(p);
1662      else {
1663	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL0)
1664	    emit_passthrough(p, VERT_ATTRIB_COLOR0, VARYING_SLOT_COL0);
1665
1666	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL1)
1667	    emit_passthrough(p, VERT_ATTRIB_COLOR1, VARYING_SLOT_COL1);
1668      }
1669   }
1670
1671   if (p->state->fragprog_inputs_read & VARYING_BIT_FOGC)
1672      build_fog(p);
1673
1674   if (p->state->fragprog_inputs_read & VARYING_BITS_TEX_ANY)
1675      build_texture_transform(p);
1676
1677   if (p->state->point_attenuated)
1678      build_atten_pointsize(p);
1679   else if (p->state->varying_vp_inputs & VERT_BIT_POINT_SIZE)
1680      build_array_pointsize(p);
1681
1682   if (p->state->varying_vp_inputs & VERT_BIT_SELECT_RESULT_OFFSET)
1683      emit_passthrough(p, VERT_ATTRIB_SELECT_RESULT_OFFSET, VARYING_SLOT_VAR0);
1684
1685   /* Finish up:
1686    */
1687   emit_op1(p, OPCODE_END, undef, 0, undef);
1688
1689   /* Disassemble:
1690    */
1691   if (DISASSEM) {
1692      printf ("\n");
1693   }
1694}
1695
1696
1697static void
1698create_new_program( const struct state_key *key,
1699                    struct gl_program *program,
1700                    GLboolean mvp_with_dp4,
1701                    GLuint max_temps)
1702{
1703   struct tnl_program p;
1704
1705   memset(&p, 0, sizeof(p));
1706   p.state = key;
1707   p.program = program;
1708   p.eye_position = undef;
1709   p.eye_position_z = undef;
1710   p.eye_position_normalized = undef;
1711   p.transformed_normal = undef;
1712   p.identity = undef;
1713   p.temp_in_use = 0;
1714   p.mvp_with_dp4 = mvp_with_dp4;
1715
1716   if (max_temps >= sizeof(int) * 8)
1717      p.temp_reserved = 0;
1718   else
1719      p.temp_reserved = ~((1<<max_temps)-1);
1720
1721   /* Start by allocating 32 instructions.
1722    * If we need more, we'll grow the instruction array as needed.
1723    */
1724   p.max_inst = 32;
1725   p.program->arb.Instructions =
1726      rzalloc_array(program, struct prog_instruction, p.max_inst);
1727   p.program->String = NULL;
1728   p.program->arb.NumInstructions =
1729   p.program->arb.NumTemporaries =
1730   p.program->arb.NumParameters =
1731   p.program->arb.NumAttributes = p.program->arb.NumAddressRegs = 0;
1732   p.program->Parameters = _mesa_new_parameter_list();
1733   p.program->info.inputs_read = 0;
1734   p.program->info.outputs_written = 0;
1735   p.state_params = _mesa_new_parameter_list();
1736
1737   build_tnl_program( &p );
1738
1739   _mesa_add_separate_state_parameters(p.program, p.state_params);
1740   _mesa_free_parameter_list(p.state_params);
1741}
1742
1743
1744/**
1745 * Return a vertex program which implements the current fixed-function
1746 * transform/lighting/texgen operations.
1747 */
1748struct gl_program *
1749_mesa_get_fixed_func_vertex_program(struct gl_context *ctx)
1750{
1751   struct gl_program *prog;
1752   struct state_key key;
1753
1754   /* We only update ctx->VertexProgram._VaryingInputs when in VP_MODE_FF _VPMode */
1755   assert(VP_MODE_FF == ctx->VertexProgram._VPMode);
1756
1757   /* Grab all the relevant state and put it in a single structure:
1758    */
1759   make_state_key(ctx, &key);
1760
1761   /* Look for an already-prepared program for this state:
1762    */
1763   prog = _mesa_search_program_cache(ctx->VertexProgram.Cache, &key,
1764                                     sizeof(key));
1765
1766   if (!prog) {
1767      /* OK, we'll have to build a new one */
1768      if (0)
1769         printf("Build new TNL program\n");
1770
1771      prog = ctx->Driver.NewProgram(ctx, MESA_SHADER_VERTEX, 0, true);
1772      if (!prog)
1773         return NULL;
1774
1775      create_new_program( &key, prog,
1776                          ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS,
1777                          ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps );
1778
1779      st_program_string_notify(ctx, GL_VERTEX_PROGRAM_ARB, prog);
1780
1781      _mesa_program_cache_insert(ctx, ctx->VertexProgram.Cache, &key,
1782                                 sizeof(key), prog);
1783   }
1784
1785   return prog;
1786}
1787