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