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/ffvertex_prog.h"
42 #include "program/program.h"
43 #include "program/prog_cache.h"
44 #include "program/prog_instruction.h"
45 #include "program/prog_parameter.h"
46 #include "program/prog_print.h"
47 #include "program/prog_statevars.h"
48 #include "util/bitscan.h"
49 
50 
51 /** Max of number of lights and texture coord units */
52 #define NUM_UNITS MAX2(MAX_TEXTURE_COORD_UNITS, MAX_LIGHTS)
53 
54 struct state_key {
55    GLbitfield varying_vp_inputs;
56 
57    unsigned fragprog_inputs_read:12;
58 
59    unsigned light_color_material_mask:12;
60    unsigned light_global_enabled:1;
61    unsigned light_local_viewer:1;
62    unsigned light_twoside:1;
63    unsigned material_shininess_is_zero:1;
64    unsigned need_eye_coords:1;
65    unsigned normalize:1;
66    unsigned rescale_normals:1;
67 
68    unsigned fog_distance_mode:2;
69    unsigned separate_specular:1;
70    unsigned point_attenuated:1;
71 
72    struct {
73       unsigned char light_enabled:1;
74       unsigned char light_eyepos3_is_zero:1;
75       unsigned char light_spotcutoff_is_180:1;
76       unsigned char light_attenuated:1;
77       unsigned char texmat_enabled:1;
78       unsigned char coord_replace:1;
79       unsigned char texgen_enabled:1;
80       unsigned char texgen_mode0:4;
81       unsigned char texgen_mode1:4;
82       unsigned char texgen_mode2:4;
83       unsigned char texgen_mode3:4;
84    } unit[NUM_UNITS];
85 };
86 
87 
88 #define TXG_NONE           0
89 #define TXG_OBJ_LINEAR     1
90 #define TXG_EYE_LINEAR     2
91 #define TXG_SPHERE_MAP     3
92 #define TXG_REFLECTION_MAP 4
93 #define TXG_NORMAL_MAP     5
94 
translate_texgen(GLboolean enabled,GLenum mode)95 static GLuint translate_texgen( GLboolean enabled, GLenum mode )
96 {
97    if (!enabled)
98       return TXG_NONE;
99 
100    switch (mode) {
101    case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
102    case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
103    case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
104    case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
105    case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
106    default: return TXG_NONE;
107    }
108 }
109 
110 #define FDM_EYE_RADIAL    0
111 #define FDM_EYE_PLANE     1
112 #define FDM_EYE_PLANE_ABS 2
113 #define FDM_FROM_ARRAY    3
114 
translate_fog_distance_mode(GLenum source,GLenum mode)115 static GLuint translate_fog_distance_mode(GLenum source, GLenum mode)
116 {
117    if (source == GL_FRAGMENT_DEPTH_EXT) {
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    } else {
128       return FDM_FROM_ARRAY;
129    }
130 }
131 
check_active_shininess(struct gl_context * ctx,const struct state_key * key,GLuint side)132 static GLboolean check_active_shininess( struct gl_context *ctx,
133                                          const struct state_key *key,
134                                          GLuint side )
135 {
136    GLuint attr = MAT_ATTRIB_FRONT_SHININESS + side;
137 
138    if ((key->varying_vp_inputs & VERT_BIT_COLOR0) &&
139        (key->light_color_material_mask & (1 << attr)))
140       return GL_TRUE;
141 
142    if (key->varying_vp_inputs & VERT_BIT_MAT(attr))
143       return GL_TRUE;
144 
145    if (ctx->Light.Material.Attrib[attr][0] != 0.0F)
146       return GL_TRUE;
147 
148    return GL_FALSE;
149 }
150 
151 
make_state_key(struct gl_context * ctx,struct state_key * key)152 static void make_state_key( struct gl_context *ctx, struct state_key *key )
153 {
154    const struct gl_program *fp = ctx->FragmentProgram._Current;
155    GLbitfield mask;
156 
157    memset(key, 0, sizeof(struct state_key));
158 
159    /* This now relies on texenvprogram.c being active:
160     */
161    assert(fp);
162 
163    key->need_eye_coords = ctx->_NeedEyeCoords;
164 
165    key->fragprog_inputs_read = fp->info.inputs_read;
166    key->varying_vp_inputs = ctx->varying_vp_inputs;
167 
168    if (ctx->RenderMode == GL_FEEDBACK) {
169       /* make sure the vertprog emits color and tex0 */
170       key->fragprog_inputs_read |= (VARYING_BIT_COL0 | VARYING_BIT_TEX0);
171    }
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.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)
183          key->separate_specular = 1;
184 
185       if (ctx->Light.ColorMaterialEnabled) {
186 	 key->light_color_material_mask = ctx->Light._ColorMaterialBitmask;
187       }
188 
189       mask = ctx->Light._EnabledLights;
190       while (mask) {
191          const int i = u_bit_scan(&mask);
192          struct gl_light *light = &ctx->Light.Light[i];
193 
194          key->unit[i].light_enabled = 1;
195 
196          if (light->EyePosition[3] == 0.0F)
197             key->unit[i].light_eyepos3_is_zero = 1;
198 
199          if (light->SpotCutoff == 180.0F)
200             key->unit[i].light_spotcutoff_is_180 = 1;
201 
202          if (light->ConstantAttenuation != 1.0F ||
203              light->LinearAttenuation != 0.0F ||
204              light->QuadraticAttenuation != 0.0F)
205             key->unit[i].light_attenuated = 1;
206       }
207 
208       if (check_active_shininess(ctx, key, 0)) {
209          key->material_shininess_is_zero = 0;
210       }
211       else if (key->light_twoside &&
212                check_active_shininess(ctx, key, 1)) {
213          key->material_shininess_is_zero = 0;
214       }
215       else {
216          key->material_shininess_is_zero = 1;
217       }
218    }
219 
220    if (ctx->Transform.Normalize)
221       key->normalize = 1;
222 
223    if (ctx->Transform.RescaleNormals)
224       key->rescale_normals = 1;
225 
226    /* Only distinguish fog parameters if we actually need */
227    if (key->fragprog_inputs_read & VARYING_BIT_FOGC)
228       key->fog_distance_mode =
229          translate_fog_distance_mode(ctx->Fog.FogCoordinateSource,
230                                      ctx->Fog.FogDistanceMode);
231 
232    if (ctx->Point._Attenuated)
233       key->point_attenuated = 1;
234 
235    mask = ctx->Texture._EnabledCoordUnits | ctx->Texture._TexGenEnabled
236       | ctx->Texture._TexMatEnabled | ctx->Point.CoordReplace;
237    while (mask) {
238       const int i = u_bit_scan(&mask);
239       struct gl_fixedfunc_texture_unit *texUnit =
240          &ctx->Texture.FixedFuncUnit[i];
241 
242       if (ctx->Point.PointSprite)
243 	 if (ctx->Point.CoordReplace & (1u << i))
244 	    key->unit[i].coord_replace = 1;
245 
246       if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
247 	 key->unit[i].texmat_enabled = 1;
248 
249       if (texUnit->TexGenEnabled) {
250 	 key->unit[i].texgen_enabled = 1;
251 
252 	 key->unit[i].texgen_mode0 =
253 	    translate_texgen( texUnit->TexGenEnabled & (1<<0),
254 			      texUnit->GenS.Mode );
255 	 key->unit[i].texgen_mode1 =
256 	    translate_texgen( texUnit->TexGenEnabled & (1<<1),
257 			      texUnit->GenT.Mode );
258 	 key->unit[i].texgen_mode2 =
259 	    translate_texgen( texUnit->TexGenEnabled & (1<<2),
260 			      texUnit->GenR.Mode );
261 	 key->unit[i].texgen_mode3 =
262 	    translate_texgen( texUnit->TexGenEnabled & (1<<3),
263 			      texUnit->GenQ.Mode );
264       }
265    }
266 }
267 
268 
269 
270 /* Very useful debugging tool - produces annotated listing of
271  * generated program with line/function references for each
272  * instruction back into this file:
273  */
274 #define DISASSEM 0
275 
276 
277 /* Use uregs to represent registers internally, translate to Mesa's
278  * expected formats on emit.
279  *
280  * NOTE: These are passed by value extensively in this file rather
281  * than as usual by pointer reference.  If this disturbs you, try
282  * remembering they are just 32bits in size.
283  *
284  * GCC is smart enough to deal with these dword-sized structures in
285  * much the same way as if I had defined them as dwords and was using
286  * macros to access and set the fields.  This is much nicer and easier
287  * to evolve.
288  */
289 struct ureg {
290    GLuint file:4;
291    GLint idx:9;      /* relative addressing may be negative */
292                      /* sizeof(idx) should == sizeof(prog_src_reg::Index) */
293    GLuint negate:1;
294    GLuint swz:12;
295    GLuint pad:6;
296 };
297 
298 
299 struct tnl_program {
300    const struct state_key *state;
301    struct gl_program *program;
302    GLuint max_inst;  /** number of instructions allocated for program */
303    GLboolean mvp_with_dp4;
304 
305    GLuint temp_in_use;
306    GLuint temp_reserved;
307 
308    struct ureg eye_position;
309    struct ureg eye_position_z;
310    struct ureg eye_position_normalized;
311    struct ureg transformed_normal;
312    struct ureg identity;
313 
314    GLuint materials;
315    GLuint color_materials;
316 };
317 
318 
319 static const struct ureg undef = {
320    PROGRAM_UNDEFINED,
321    0,
322    0,
323    0,
324    0
325 };
326 
327 /* Local shorthand:
328  */
329 #define X    SWIZZLE_X
330 #define Y    SWIZZLE_Y
331 #define Z    SWIZZLE_Z
332 #define W    SWIZZLE_W
333 
334 
335 /* Construct a ureg:
336  */
make_ureg(GLuint file,GLint idx)337 static struct ureg make_ureg(GLuint file, GLint idx)
338 {
339    struct ureg reg;
340    reg.file = file;
341    reg.idx = idx;
342    reg.negate = 0;
343    reg.swz = SWIZZLE_NOOP;
344    reg.pad = 0;
345    return reg;
346 }
347 
348 
negate(struct ureg reg)349 static struct ureg negate( struct ureg reg )
350 {
351    reg.negate ^= 1;
352    return reg;
353 }
354 
355 
swizzle(struct ureg reg,int x,int y,int z,int w)356 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
357 {
358    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
359 			   GET_SWZ(reg.swz, y),
360 			   GET_SWZ(reg.swz, z),
361 			   GET_SWZ(reg.swz, w));
362    return reg;
363 }
364 
365 
swizzle1(struct ureg reg,int x)366 static struct ureg swizzle1( struct ureg reg, int x )
367 {
368    return swizzle(reg, x, x, x, x);
369 }
370 
371 
get_temp(struct tnl_program * p)372 static struct ureg get_temp( struct tnl_program *p )
373 {
374    int bit = ffs( ~p->temp_in_use );
375    if (!bit) {
376       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
377       exit(1);
378    }
379 
380    if ((GLuint) bit > p->program->arb.NumTemporaries)
381       p->program->arb.NumTemporaries = bit;
382 
383    p->temp_in_use |= 1<<(bit-1);
384    return make_ureg(PROGRAM_TEMPORARY, bit-1);
385 }
386 
387 
reserve_temp(struct tnl_program * p)388 static struct ureg reserve_temp( struct tnl_program *p )
389 {
390    struct ureg temp = get_temp( p );
391    p->temp_reserved |= 1<<temp.idx;
392    return temp;
393 }
394 
395 
release_temp(struct tnl_program * p,struct ureg reg)396 static void release_temp( struct tnl_program *p, struct ureg reg )
397 {
398    if (reg.file == PROGRAM_TEMPORARY) {
399       p->temp_in_use &= ~(1<<reg.idx);
400       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
401    }
402 }
403 
release_temps(struct tnl_program * p)404 static void release_temps( struct tnl_program *p )
405 {
406    p->temp_in_use = p->temp_reserved;
407 }
408 
409 
register_param5(struct tnl_program * p,GLint s0,GLint s1,GLint s2,GLint s3,GLint s4)410 static struct ureg register_param5(struct tnl_program *p,
411 				   GLint s0,
412 				   GLint s1,
413 				   GLint s2,
414 				   GLint s3,
415                                    GLint s4)
416 {
417    gl_state_index16 tokens[STATE_LENGTH];
418    GLint idx;
419    tokens[0] = s0;
420    tokens[1] = s1;
421    tokens[2] = s2;
422    tokens[3] = s3;
423    tokens[4] = s4;
424    idx = _mesa_add_state_reference(p->program->Parameters, tokens );
425    return make_ureg(PROGRAM_STATE_VAR, idx);
426 }
427 
428 
429 #define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
430 #define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
431 #define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
432 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
433 
434 
435 
436 /**
437  * \param input  one of VERT_ATTRIB_x tokens.
438  */
register_input(struct tnl_program * p,GLuint input)439 static struct ureg register_input( struct tnl_program *p, GLuint input )
440 {
441    assert(input < VERT_ATTRIB_MAX);
442 
443    if (p->state->varying_vp_inputs & VERT_BIT(input)) {
444       p->program->info.inputs_read |= VERT_BIT(input);
445       return make_ureg(PROGRAM_INPUT, input);
446    }
447    else {
448       return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, input );
449    }
450 }
451 
452 
453 /**
454  * \param input  one of VARYING_SLOT_x tokens.
455  */
register_output(struct tnl_program * p,GLuint output)456 static struct ureg register_output( struct tnl_program *p, GLuint output )
457 {
458    p->program->info.outputs_written |= BITFIELD64_BIT(output);
459    return make_ureg(PROGRAM_OUTPUT, output);
460 }
461 
462 
register_const4f(struct tnl_program * p,GLfloat s0,GLfloat s1,GLfloat s2,GLfloat s3)463 static struct ureg register_const4f( struct tnl_program *p,
464 			      GLfloat s0,
465 			      GLfloat s1,
466 			      GLfloat s2,
467 			      GLfloat s3)
468 {
469    gl_constant_value values[4];
470    GLint idx;
471    GLuint swizzle;
472    values[0].f = s0;
473    values[1].f = s1;
474    values[2].f = s2;
475    values[3].f = s3;
476    idx = _mesa_add_unnamed_constant(p->program->Parameters, values, 4,
477                                     &swizzle );
478    assert(swizzle == SWIZZLE_NOOP);
479    return make_ureg(PROGRAM_CONSTANT, idx);
480 }
481 
482 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
483 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
484 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
485 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
486 
is_undef(struct ureg reg)487 static GLboolean is_undef( struct ureg reg )
488 {
489    return reg.file == PROGRAM_UNDEFINED;
490 }
491 
492 
get_identity_param(struct tnl_program * p)493 static struct ureg get_identity_param( struct tnl_program *p )
494 {
495    if (is_undef(p->identity))
496       p->identity = register_const4f(p, 0,0,0,1);
497 
498    return p->identity;
499 }
500 
register_matrix_param5(struct tnl_program * p,GLint s0,GLint s1,GLint s2,GLint s3,GLint s4,struct ureg * matrix)501 static void register_matrix_param5( struct tnl_program *p,
502 				    GLint s0, /* modelview, projection, etc */
503 				    GLint s1, /* texture matrix number */
504 				    GLint s2, /* first row */
505 				    GLint s3, /* last row */
506 				    GLint s4, /* inverse, transpose, etc */
507 				    struct ureg *matrix )
508 {
509    GLint i;
510 
511    /* This is a bit sad as the support is there to pull the whole
512     * matrix out in one go:
513     */
514    for (i = 0; i <= s3 - s2; i++)
515       matrix[i] = register_param5( p, s0, s1, i, i, s4 );
516 }
517 
518 
emit_arg(struct prog_src_register * src,struct ureg reg)519 static void emit_arg( struct prog_src_register *src,
520 		      struct ureg reg )
521 {
522    src->File = reg.file;
523    src->Index = reg.idx;
524    src->Swizzle = reg.swz;
525    src->Negate = reg.negate ? NEGATE_XYZW : NEGATE_NONE;
526    src->RelAddr = 0;
527    /* Check that bitfield sizes aren't exceeded */
528    assert(src->Index == reg.idx);
529 }
530 
531 
emit_dst(struct prog_dst_register * dst,struct ureg reg,GLuint mask)532 static void emit_dst( struct prog_dst_register *dst,
533 		      struct ureg reg, GLuint mask )
534 {
535    dst->File = reg.file;
536    dst->Index = reg.idx;
537    /* allow zero as a shorthand for xyzw */
538    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
539    /* Check that bitfield sizes aren't exceeded */
540    assert(dst->Index == reg.idx);
541 }
542 
543 
debug_insn(struct prog_instruction * inst,const char * fn,GLuint line)544 static void debug_insn( struct prog_instruction *inst, const char *fn,
545 			GLuint line )
546 {
547    if (DISASSEM) {
548       static const char *last_fn;
549 
550       if (fn != last_fn) {
551 	 last_fn = fn;
552 	 printf("%s:\n", fn);
553       }
554 
555       printf("%d:\t", line);
556       _mesa_print_instruction(inst);
557    }
558 }
559 
560 
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)561 static void emit_op3fn(struct tnl_program *p,
562                        enum prog_opcode op,
563 		       struct ureg dest,
564 		       GLuint mask,
565 		       struct ureg src0,
566 		       struct ureg src1,
567 		       struct ureg src2,
568 		       const char *fn,
569 		       GLuint line)
570 {
571    GLuint nr;
572    struct prog_instruction *inst;
573 
574    assert(p->program->arb.NumInstructions <= p->max_inst);
575 
576    if (p->program->arb.NumInstructions == p->max_inst) {
577       /* need to extend the program's instruction array */
578       struct prog_instruction *newInst;
579 
580       /* double the size */
581       p->max_inst *= 2;
582 
583       newInst =
584          rzalloc_array(p->program, struct prog_instruction, p->max_inst);
585       if (!newInst) {
586          _mesa_error(NULL, GL_OUT_OF_MEMORY, "vertex program build");
587          return;
588       }
589 
590       _mesa_copy_instructions(newInst, p->program->arb.Instructions,
591                               p->program->arb.NumInstructions);
592 
593       ralloc_free(p->program->arb.Instructions);
594 
595       p->program->arb.Instructions = newInst;
596    }
597 
598    nr = p->program->arb.NumInstructions++;
599 
600    inst = &p->program->arb.Instructions[nr];
601    inst->Opcode = (enum prog_opcode) op;
602 
603    emit_arg( &inst->SrcReg[0], src0 );
604    emit_arg( &inst->SrcReg[1], src1 );
605    emit_arg( &inst->SrcReg[2], src2 );
606 
607    emit_dst( &inst->DstReg, dest, mask );
608 
609    debug_insn(inst, fn, line);
610 }
611 
612 
613 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
614    emit_op3fn(p, op, dst, mask, src0, src1, src2, __func__, __LINE__)
615 
616 #define emit_op2(p, op, dst, mask, src0, src1) \
617     emit_op3fn(p, op, dst, mask, src0, src1, undef, __func__, __LINE__)
618 
619 #define emit_op1(p, op, dst, mask, src0) \
620     emit_op3fn(p, op, dst, mask, src0, undef, undef, __func__, __LINE__)
621 
622 
make_temp(struct tnl_program * p,struct ureg reg)623 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
624 {
625    if (reg.file == PROGRAM_TEMPORARY &&
626        !(p->temp_reserved & (1<<reg.idx)))
627       return reg;
628    else {
629       struct ureg temp = get_temp(p);
630       emit_op1(p, OPCODE_MOV, temp, 0, reg);
631       return temp;
632    }
633 }
634 
635 
636 /* Currently no tracking performed of input/output/register size or
637  * active elements.  Could be used to reduce these operations, as
638  * could the matrix type.
639  */
emit_matrix_transform_vec4(struct tnl_program * p,struct ureg dest,const struct ureg * mat,struct ureg src)640 static void emit_matrix_transform_vec4( struct tnl_program *p,
641 					struct ureg dest,
642 					const struct ureg *mat,
643 					struct ureg src)
644 {
645    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
646    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
647    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
648    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
649 }
650 
651 
652 /* This version is much easier to implement if writemasks are not
653  * supported natively on the target or (like SSE), the target doesn't
654  * have a clean/obvious dotproduct implementation.
655  */
emit_transpose_matrix_transform_vec4(struct tnl_program * p,struct ureg dest,const struct ureg * mat,struct ureg src)656 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
657 						  struct ureg dest,
658 						  const struct ureg *mat,
659 						  struct ureg src)
660 {
661    struct ureg tmp;
662 
663    if (dest.file != PROGRAM_TEMPORARY)
664       tmp = get_temp(p);
665    else
666       tmp = dest;
667 
668    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
669    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
670    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
671    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
672 
673    if (dest.file != PROGRAM_TEMPORARY)
674       release_temp(p, tmp);
675 }
676 
677 
emit_matrix_transform_vec3(struct tnl_program * p,struct ureg dest,const struct ureg * mat,struct ureg src)678 static void emit_matrix_transform_vec3( struct tnl_program *p,
679 					struct ureg dest,
680 					const struct ureg *mat,
681 					struct ureg src)
682 {
683    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
684    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
685    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
686 }
687 
688 
emit_normalize_vec3(struct tnl_program * p,struct ureg dest,struct ureg src)689 static void emit_normalize_vec3( struct tnl_program *p,
690 				 struct ureg dest,
691 				 struct ureg src )
692 {
693    struct ureg tmp = get_temp(p);
694    emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, src, src);
695    emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
696    emit_op2(p, OPCODE_MUL, dest, 0, src, swizzle1(tmp, X));
697    release_temp(p, tmp);
698 }
699 
700 
emit_passthrough(struct tnl_program * p,GLuint input,GLuint output)701 static void emit_passthrough( struct tnl_program *p,
702 			      GLuint input,
703 			      GLuint output )
704 {
705    struct ureg out = register_output(p, output);
706    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
707 }
708 
709 
get_eye_position(struct tnl_program * p)710 static struct ureg get_eye_position( struct tnl_program *p )
711 {
712    if (is_undef(p->eye_position)) {
713       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
714       struct ureg modelview[4];
715 
716       p->eye_position = reserve_temp(p);
717 
718       if (p->mvp_with_dp4) {
719 	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
720                                  0, modelview );
721 
722 	 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
723       }
724       else {
725 	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
726 				 STATE_MATRIX_TRANSPOSE, modelview );
727 
728 	 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
729       }
730    }
731 
732    return p->eye_position;
733 }
734 
735 
get_eye_position_z(struct tnl_program * p)736 static struct ureg get_eye_position_z( struct tnl_program *p )
737 {
738    if (!is_undef(p->eye_position))
739       return swizzle1(p->eye_position, Z);
740 
741    if (is_undef(p->eye_position_z)) {
742       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
743       struct ureg modelview[4];
744 
745       p->eye_position_z = reserve_temp(p);
746 
747       register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
748                               0, modelview );
749 
750       emit_op2(p, OPCODE_DP4, p->eye_position_z, 0, pos, modelview[2]);
751    }
752 
753    return p->eye_position_z;
754 }
755 
756 
get_eye_position_normalized(struct tnl_program * p)757 static struct ureg get_eye_position_normalized( struct tnl_program *p )
758 {
759    if (is_undef(p->eye_position_normalized)) {
760       struct ureg eye = get_eye_position(p);
761       p->eye_position_normalized = reserve_temp(p);
762       emit_normalize_vec3(p, p->eye_position_normalized, eye);
763    }
764 
765    return p->eye_position_normalized;
766 }
767 
768 
get_transformed_normal(struct tnl_program * p)769 static struct ureg get_transformed_normal( struct tnl_program *p )
770 {
771    if (is_undef(p->transformed_normal) &&
772        !p->state->need_eye_coords &&
773        !p->state->normalize &&
774        !(p->state->need_eye_coords == p->state->rescale_normals))
775    {
776       p->transformed_normal = register_input(p, VERT_ATTRIB_NORMAL );
777    }
778    else if (is_undef(p->transformed_normal))
779    {
780       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
781       struct ureg mvinv[3];
782       struct ureg transformed_normal = reserve_temp(p);
783 
784       if (p->state->need_eye_coords) {
785          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
786                                  STATE_MATRIX_INVTRANS, mvinv );
787 
788          /* Transform to eye space:
789           */
790          emit_matrix_transform_vec3( p, transformed_normal, mvinv, normal );
791          normal = transformed_normal;
792       }
793 
794       /* Normalize/Rescale:
795        */
796       if (p->state->normalize) {
797 	 emit_normalize_vec3( p, transformed_normal, normal );
798          normal = transformed_normal;
799       }
800       else if (p->state->need_eye_coords == p->state->rescale_normals) {
801          /* This is already adjusted for eye/non-eye rendering:
802           */
803 	 struct ureg rescale = register_param2(p, STATE_INTERNAL,
804                                                STATE_NORMAL_SCALE);
805 
806 	 emit_op2( p, OPCODE_MUL, transformed_normal, 0, normal, rescale );
807          normal = transformed_normal;
808       }
809 
810       assert(normal.file == PROGRAM_TEMPORARY);
811       p->transformed_normal = normal;
812    }
813 
814    return p->transformed_normal;
815 }
816 
817 
build_hpos(struct tnl_program * p)818 static void build_hpos( struct tnl_program *p )
819 {
820    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
821    struct ureg hpos = register_output( p, VARYING_SLOT_POS );
822    struct ureg mvp[4];
823 
824    if (p->mvp_with_dp4) {
825       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
826 			      0, mvp );
827       emit_matrix_transform_vec4( p, hpos, mvp, pos );
828    }
829    else {
830       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
831 			      STATE_MATRIX_TRANSPOSE, mvp );
832       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
833    }
834 }
835 
836 
material_attrib(GLuint side,GLuint property)837 static GLuint material_attrib( GLuint side, GLuint property )
838 {
839    return (property - STATE_AMBIENT) * 2 + side;
840 }
841 
842 
843 /**
844  * Get a bitmask of which material values vary on a per-vertex basis.
845  */
set_material_flags(struct tnl_program * p)846 static void set_material_flags( struct tnl_program *p )
847 {
848    p->color_materials = 0;
849    p->materials = 0;
850 
851    if (p->state->varying_vp_inputs & VERT_BIT_COLOR0) {
852       p->materials =
853 	 p->color_materials = p->state->light_color_material_mask;
854    }
855 
856    p->materials |= ((p->state->varying_vp_inputs & VERT_BIT_MAT_ALL)
857                     >> VERT_ATTRIB_MAT(0));
858 }
859 
860 
get_material(struct tnl_program * p,GLuint side,GLuint property)861 static struct ureg get_material( struct tnl_program *p, GLuint side,
862 				 GLuint property )
863 {
864    GLuint attrib = material_attrib(side, property);
865 
866    if (p->color_materials & (1<<attrib))
867       return register_input(p, VERT_ATTRIB_COLOR0);
868    else if (p->materials & (1<<attrib)) {
869       /* Put material values in the GENERIC slots -- they are not used
870        * for anything in fixed function mode.
871        */
872       return register_input( p, VERT_ATTRIB_MAT(attrib) );
873    }
874    else
875       return register_param3( p, STATE_MATERIAL, side, property );
876 }
877 
878 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
879 				   MAT_BIT_FRONT_AMBIENT | \
880 				   MAT_BIT_FRONT_DIFFUSE) << (side))
881 
882 
883 /**
884  * Either return a precalculated constant value or emit code to
885  * calculate these values dynamically in the case where material calls
886  * are present between begin/end pairs.
887  *
888  * Probably want to shift this to the program compilation phase - if
889  * we always emitted the calculation here, a smart compiler could
890  * detect that it was constant (given a certain set of inputs), and
891  * lift it out of the main loop.  That way the programs created here
892  * would be independent of the vertex_buffer details.
893  */
get_scenecolor(struct tnl_program * p,GLuint side)894 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
895 {
896    if (p->materials & SCENE_COLOR_BITS(side)) {
897       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
898       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
899       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
900       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
901       struct ureg tmp = make_temp(p, material_diffuse);
902       emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient,
903 	       material_ambient, material_emission);
904       return tmp;
905    }
906    else
907       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
908 }
909 
910 
get_lightprod(struct tnl_program * p,GLuint light,GLuint side,GLuint property)911 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
912 				  GLuint side, GLuint property )
913 {
914    GLuint attrib = material_attrib(side, property);
915    if (p->materials & (1<<attrib)) {
916       struct ureg light_value =
917 	 register_param3(p, STATE_LIGHT, light, property);
918       struct ureg material_value = get_material(p, side, property);
919       struct ureg tmp = get_temp(p);
920       emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value);
921       return tmp;
922    }
923    else
924       return register_param4(p, STATE_LIGHTPROD, light, side, property);
925 }
926 
927 
calculate_light_attenuation(struct tnl_program * p,GLuint i,struct ureg VPpli,struct ureg dist)928 static struct ureg calculate_light_attenuation( struct tnl_program *p,
929 						GLuint i,
930 						struct ureg VPpli,
931 						struct ureg dist )
932 {
933    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
934 					     STATE_ATTENUATION);
935    struct ureg att = undef;
936 
937    /* Calculate spot attenuation:
938     */
939    if (!p->state->unit[i].light_spotcutoff_is_180) {
940       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
941 						  STATE_LIGHT_SPOT_DIR_NORMALIZED, i);
942       struct ureg spot = get_temp(p);
943       struct ureg slt = get_temp(p);
944 
945       att = get_temp(p);
946 
947       emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
948       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
949       emit_op1(p, OPCODE_ABS, spot, 0, spot);
950       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
951       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
952 
953       release_temp(p, spot);
954       release_temp(p, slt);
955    }
956 
957    /* Calculate distance attenuation(See formula (2.4) at glspec 2.1 page 62):
958     *
959     * Skip the calucation when _dist_ is undefined(light_eyepos3_is_zero)
960     */
961    if (p->state->unit[i].light_attenuated && !is_undef(dist)) {
962       if (is_undef(att))
963          att = get_temp(p);
964       /* 1/d,d,d,1/d */
965       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
966       /* 1,d,d*d,1/d */
967       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
968       /* 1/dist-atten */
969       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
970 
971       if (!p->state->unit[i].light_spotcutoff_is_180) {
972 	 /* dist-atten */
973 	 emit_op1(p, OPCODE_RCP, dist, 0, dist);
974 	 /* spot-atten * dist-atten */
975 	 emit_op2(p, OPCODE_MUL, att, 0, dist, att);
976       }
977       else {
978 	 /* dist-atten */
979 	 emit_op1(p, OPCODE_RCP, att, 0, dist);
980       }
981    }
982 
983    return att;
984 }
985 
986 
987 /**
988  * Compute:
989  *   lit.y = MAX(0, dots.x)
990  *   lit.z = SLT(0, dots.x)
991  */
emit_degenerate_lit(struct tnl_program * p,struct ureg lit,struct ureg dots)992 static void emit_degenerate_lit( struct tnl_program *p,
993                                  struct ureg lit,
994                                  struct ureg dots )
995 {
996    struct ureg id = get_identity_param(p);  /* id = {0,0,0,1} */
997 
998    /* Note that lit.x & lit.w will not be examined.  Note also that
999     * dots.xyzw == dots.xxxx.
1000     */
1001 
1002    /* MAX lit, id, dots;
1003     */
1004    emit_op2(p, OPCODE_MAX, lit, WRITEMASK_XYZW, id, dots);
1005 
1006    /* result[2] = (in > 0 ? 1 : 0)
1007     * SLT lit.z, id.z, dots;   # lit.z = (0 < dots.z) ? 1 : 0
1008     */
1009    emit_op2(p, OPCODE_SLT, lit, WRITEMASK_Z, swizzle1(id,Z), dots);
1010 }
1011 
1012 
1013 /* Need to add some addtional parameters to allow lighting in object
1014  * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
1015  * space lighting.
1016  */
build_lighting(struct tnl_program * p)1017 static void build_lighting( struct tnl_program *p )
1018 {
1019    const GLboolean twoside = p->state->light_twoside;
1020    const GLboolean separate = p->state->separate_specular;
1021    GLuint nr_lights = 0, count = 0;
1022    struct ureg normal = get_transformed_normal(p);
1023    struct ureg lit = get_temp(p);
1024    struct ureg dots = get_temp(p);
1025    struct ureg _col0 = undef, _col1 = undef;
1026    struct ureg _bfc0 = undef, _bfc1 = undef;
1027    GLuint i;
1028 
1029    /*
1030     * NOTE:
1031     * dots.x = dot(normal, VPpli)
1032     * dots.y = dot(normal, halfAngle)
1033     * dots.z = back.shininess
1034     * dots.w = front.shininess
1035     */
1036 
1037    for (i = 0; i < MAX_LIGHTS; i++)
1038       if (p->state->unit[i].light_enabled)
1039 	 nr_lights++;
1040 
1041    set_material_flags(p);
1042 
1043    {
1044       if (!p->state->material_shininess_is_zero) {
1045          struct ureg shininess = get_material(p, 0, STATE_SHININESS);
1046          emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X));
1047          release_temp(p, shininess);
1048       }
1049 
1050       _col0 = make_temp(p, get_scenecolor(p, 0));
1051       if (separate)
1052 	 _col1 = make_temp(p, get_identity_param(p));
1053       else
1054 	 _col1 = _col0;
1055    }
1056 
1057    if (twoside) {
1058       if (!p->state->material_shininess_is_zero) {
1059          /* Note that we negate the back-face specular exponent here.
1060           * The negation will be un-done later in the back-face code below.
1061           */
1062          struct ureg shininess = get_material(p, 1, STATE_SHININESS);
1063          emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
1064                   negate(swizzle1(shininess,X)));
1065          release_temp(p, shininess);
1066       }
1067 
1068       _bfc0 = make_temp(p, get_scenecolor(p, 1));
1069       if (separate)
1070 	 _bfc1 = make_temp(p, get_identity_param(p));
1071       else
1072 	 _bfc1 = _bfc0;
1073    }
1074 
1075    /* If no lights, still need to emit the scenecolor.
1076     */
1077    {
1078       struct ureg res0 = register_output( p, VARYING_SLOT_COL0 );
1079       emit_op1(p, OPCODE_MOV, res0, 0, _col0);
1080    }
1081 
1082    if (separate) {
1083       struct ureg res1 = register_output( p, VARYING_SLOT_COL1 );
1084       emit_op1(p, OPCODE_MOV, res1, 0, _col1);
1085    }
1086 
1087    if (twoside) {
1088       struct ureg res0 = register_output( p, VARYING_SLOT_BFC0 );
1089       emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
1090    }
1091 
1092    if (twoside && separate) {
1093       struct ureg res1 = register_output( p, VARYING_SLOT_BFC1 );
1094       emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
1095    }
1096 
1097    if (nr_lights == 0) {
1098       release_temps(p);
1099       return;
1100    }
1101 
1102    for (i = 0; i < MAX_LIGHTS; i++) {
1103       if (p->state->unit[i].light_enabled) {
1104 	 struct ureg half = undef;
1105 	 struct ureg att = undef, VPpli = undef;
1106 	 struct ureg dist = undef;
1107 
1108 	 count++;
1109          if (p->state->unit[i].light_eyepos3_is_zero) {
1110              VPpli = register_param3(p, STATE_INTERNAL,
1111                                      STATE_LIGHT_POSITION_NORMALIZED, i);
1112          } else {
1113             struct ureg Ppli = register_param3(p, STATE_INTERNAL,
1114                                                STATE_LIGHT_POSITION, i);
1115             struct ureg V = get_eye_position(p);
1116 
1117             VPpli = get_temp(p);
1118             dist = get_temp(p);
1119 
1120             /* Calculate VPpli vector
1121              */
1122             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
1123 
1124             /* Normalize VPpli.  The dist value also used in
1125              * attenuation below.
1126              */
1127             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
1128             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
1129             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
1130          }
1131 
1132          /* Calculate attenuation:
1133           */
1134          att = calculate_light_attenuation(p, i, VPpli, dist);
1135          release_temp(p, dist);
1136 
1137 	 /* Calculate viewer direction, or use infinite viewer:
1138 	  */
1139          if (!p->state->material_shininess_is_zero) {
1140             if (p->state->light_local_viewer) {
1141                struct ureg eye_hat = get_eye_position_normalized(p);
1142                half = get_temp(p);
1143                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1144                emit_normalize_vec3(p, half, half);
1145             } else if (p->state->unit[i].light_eyepos3_is_zero) {
1146                half = register_param3(p, STATE_INTERNAL,
1147                                       STATE_LIGHT_HALF_VECTOR, i);
1148             } else {
1149                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1150                half = get_temp(p);
1151                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1152                emit_normalize_vec3(p, half, half);
1153             }
1154 	 }
1155 
1156 	 /* Calculate dot products:
1157 	  */
1158          if (p->state->material_shininess_is_zero) {
1159             emit_op2(p, OPCODE_DP3, dots, 0, normal, VPpli);
1160          }
1161          else {
1162             emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1163             emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1164          }
1165 
1166 	 /* Front face lighting:
1167 	  */
1168 	 {
1169 	    struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1170 	    struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1171 	    struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1172 	    struct ureg res0, res1;
1173 	    GLuint mask0, mask1;
1174 
1175 	    if (count == nr_lights) {
1176 	       if (separate) {
1177 		  mask0 = WRITEMASK_XYZ;
1178 		  mask1 = WRITEMASK_XYZ;
1179 		  res0 = register_output( p, VARYING_SLOT_COL0 );
1180 		  res1 = register_output( p, VARYING_SLOT_COL1 );
1181 	       }
1182 	       else {
1183 		  mask0 = 0;
1184 		  mask1 = WRITEMASK_XYZ;
1185 		  res0 = _col0;
1186 		  res1 = register_output( p, VARYING_SLOT_COL0 );
1187 	       }
1188 	    }
1189             else {
1190 	       mask0 = 0;
1191 	       mask1 = 0;
1192 	       res0 = _col0;
1193 	       res1 = _col1;
1194 	    }
1195 
1196 	    if (!is_undef(att)) {
1197                /* light is attenuated by distance */
1198                emit_op1(p, OPCODE_LIT, lit, 0, dots);
1199                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1200                emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1201             }
1202             else if (!p->state->material_shininess_is_zero) {
1203                /* there's a non-zero specular term */
1204                emit_op1(p, OPCODE_LIT, lit, 0, dots);
1205                emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1206             }
1207             else {
1208                /* no attenutation, no specular */
1209                emit_degenerate_lit(p, lit, dots);
1210                emit_op2(p, OPCODE_ADD, _col0, 0, ambient, _col0);
1211             }
1212 
1213 	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1214 	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1215 
1216 	    release_temp(p, ambient);
1217 	    release_temp(p, diffuse);
1218 	    release_temp(p, specular);
1219 	 }
1220 
1221 	 /* Back face lighting:
1222 	  */
1223 	 if (twoside) {
1224 	    struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1225 	    struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1226 	    struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1227 	    struct ureg res0, res1;
1228 	    GLuint mask0, mask1;
1229 
1230 	    if (count == nr_lights) {
1231 	       if (separate) {
1232 		  mask0 = WRITEMASK_XYZ;
1233 		  mask1 = WRITEMASK_XYZ;
1234 		  res0 = register_output( p, VARYING_SLOT_BFC0 );
1235 		  res1 = register_output( p, VARYING_SLOT_BFC1 );
1236 	       }
1237 	       else {
1238 		  mask0 = 0;
1239 		  mask1 = WRITEMASK_XYZ;
1240 		  res0 = _bfc0;
1241 		  res1 = register_output( p, VARYING_SLOT_BFC0 );
1242 	       }
1243 	    }
1244             else {
1245 	       res0 = _bfc0;
1246 	       res1 = _bfc1;
1247 	       mask0 = 0;
1248 	       mask1 = 0;
1249 	    }
1250 
1251             /* For the back face we need to negate the X and Y component
1252              * dot products.  dots.Z has the negated back-face specular
1253              * exponent.  We swizzle that into the W position.  This
1254              * negation makes the back-face specular term positive again.
1255              */
1256             dots = negate(swizzle(dots,X,Y,W,Z));
1257 
1258 	    if (!is_undef(att)) {
1259                emit_op1(p, OPCODE_LIT, lit, 0, dots);
1260 	       emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1261                emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1262             }
1263             else if (!p->state->material_shininess_is_zero) {
1264                emit_op1(p, OPCODE_LIT, lit, 0, dots);
1265                emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0); /**/
1266             }
1267             else {
1268                emit_degenerate_lit(p, lit, dots);
1269                emit_op2(p, OPCODE_ADD, _bfc0, 0, ambient, _bfc0);
1270             }
1271 
1272 	    emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1273 	    emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1274             /* restore dots to its original state for subsequent lights
1275              * by negating and swizzling again.
1276              */
1277             dots = negate(swizzle(dots,X,Y,W,Z));
1278 
1279 	    release_temp(p, ambient);
1280 	    release_temp(p, diffuse);
1281 	    release_temp(p, specular);
1282 	 }
1283 
1284 	 release_temp(p, half);
1285 	 release_temp(p, VPpli);
1286 	 release_temp(p, att);
1287       }
1288    }
1289 
1290    release_temps( p );
1291 }
1292 
1293 
build_fog(struct tnl_program * p)1294 static void build_fog( struct tnl_program *p )
1295 {
1296    struct ureg fog = register_output(p, VARYING_SLOT_FOGC);
1297    struct ureg input;
1298 
1299    switch (p->state->fog_distance_mode) {
1300    case FDM_EYE_RADIAL: { /* Z = sqrt(Xe*Xe + Ye*Ye + Ze*Ze) */
1301       struct ureg tmp = get_temp(p);
1302       input = get_eye_position(p);
1303       emit_op2(p, OPCODE_DP3, tmp, WRITEMASK_X, input, input);
1304       emit_op1(p, OPCODE_RSQ, tmp, WRITEMASK_X, tmp);
1305       emit_op1(p, OPCODE_RCP, fog, WRITEMASK_X, tmp);
1306       break;
1307    }
1308    case FDM_EYE_PLANE: /* Z = Ze */
1309       input = get_eye_position_z(p);
1310       emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1311       break;
1312    case FDM_EYE_PLANE_ABS: /* Z = abs(Ze) */
1313       input = get_eye_position_z(p);
1314       emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1315       break;
1316    case FDM_FROM_ARRAY:
1317       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1318       emit_op1(p, OPCODE_ABS, fog, WRITEMASK_X, input);
1319       break;
1320    default:
1321       assert(!"Bad fog mode in build_fog()");
1322       break;
1323    }
1324 
1325    emit_op1(p, OPCODE_MOV, fog, WRITEMASK_YZW, get_identity_param(p));
1326 }
1327 
1328 
build_reflect_texgen(struct tnl_program * p,struct ureg dest,GLuint writemask)1329 static void build_reflect_texgen( struct tnl_program *p,
1330 				  struct ureg dest,
1331 				  GLuint writemask )
1332 {
1333    struct ureg normal = get_transformed_normal(p);
1334    struct ureg eye_hat = get_eye_position_normalized(p);
1335    struct ureg tmp = get_temp(p);
1336 
1337    /* n.u */
1338    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1339    /* 2n.u */
1340    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1341    /* (-2n.u)n + u */
1342    emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1343 
1344    release_temp(p, tmp);
1345 }
1346 
1347 
build_sphere_texgen(struct tnl_program * p,struct ureg dest,GLuint writemask)1348 static void build_sphere_texgen( struct tnl_program *p,
1349 				 struct ureg dest,
1350 				 GLuint writemask )
1351 {
1352    struct ureg normal = get_transformed_normal(p);
1353    struct ureg eye_hat = get_eye_position_normalized(p);
1354    struct ureg tmp = get_temp(p);
1355    struct ureg half = register_scalar_const(p, .5);
1356    struct ureg r = get_temp(p);
1357    struct ureg inv_m = get_temp(p);
1358    struct ureg id = get_identity_param(p);
1359 
1360    /* Could share the above calculations, but it would be
1361     * a fairly odd state for someone to set (both sphere and
1362     * reflection active for different texture coordinate
1363     * components.  Of course - if two texture units enable
1364     * reflect and/or sphere, things start to tilt in favour
1365     * of seperating this out:
1366     */
1367 
1368    /* n.u */
1369    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1370    /* 2n.u */
1371    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1372    /* (-2n.u)n + u */
1373    emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1374    /* r + 0,0,1 */
1375    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1376    /* rx^2 + ry^2 + (rz+1)^2 */
1377    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1378    /* 2/m */
1379    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1380    /* 1/m */
1381    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1382    /* r/m + 1/2 */
1383    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1384 
1385    release_temp(p, tmp);
1386    release_temp(p, r);
1387    release_temp(p, inv_m);
1388 }
1389 
1390 
build_texture_transform(struct tnl_program * p)1391 static void build_texture_transform( struct tnl_program *p )
1392 {
1393    GLuint i, j;
1394 
1395    for (i = 0; i < MAX_TEXTURE_COORD_UNITS; i++) {
1396 
1397       if (!(p->state->fragprog_inputs_read & VARYING_BIT_TEX(i)))
1398 	 continue;
1399 
1400       if (p->state->unit[i].coord_replace)
1401   	 continue;
1402 
1403       if (p->state->unit[i].texgen_enabled ||
1404 	  p->state->unit[i].texmat_enabled) {
1405 
1406 	 GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1407 	 struct ureg out = register_output(p, VARYING_SLOT_TEX0 + i);
1408 	 struct ureg out_texgen = undef;
1409 
1410 	 if (p->state->unit[i].texgen_enabled) {
1411 	    GLuint copy_mask = 0;
1412 	    GLuint sphere_mask = 0;
1413 	    GLuint reflect_mask = 0;
1414 	    GLuint normal_mask = 0;
1415 	    GLuint modes[4];
1416 
1417 	    if (texmat_enabled)
1418 	       out_texgen = get_temp(p);
1419 	    else
1420 	       out_texgen = out;
1421 
1422 	    modes[0] = p->state->unit[i].texgen_mode0;
1423 	    modes[1] = p->state->unit[i].texgen_mode1;
1424 	    modes[2] = p->state->unit[i].texgen_mode2;
1425 	    modes[3] = p->state->unit[i].texgen_mode3;
1426 
1427 	    for (j = 0; j < 4; j++) {
1428 	       switch (modes[j]) {
1429 	       case TXG_OBJ_LINEAR: {
1430 		  struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1431 		  struct ureg plane =
1432 		     register_param3(p, STATE_TEXGEN, i,
1433 				     STATE_TEXGEN_OBJECT_S + j);
1434 
1435 		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1436 			   obj, plane );
1437 		  break;
1438 	       }
1439 	       case TXG_EYE_LINEAR: {
1440 		  struct ureg eye = get_eye_position(p);
1441 		  struct ureg plane =
1442 		     register_param3(p, STATE_TEXGEN, i,
1443 				     STATE_TEXGEN_EYE_S + j);
1444 
1445 		  emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1446 			   eye, plane );
1447 		  break;
1448 	       }
1449 	       case TXG_SPHERE_MAP:
1450 		  sphere_mask |= WRITEMASK_X << j;
1451 		  break;
1452 	       case TXG_REFLECTION_MAP:
1453 		  reflect_mask |= WRITEMASK_X << j;
1454 		  break;
1455 	       case TXG_NORMAL_MAP:
1456 		  normal_mask |= WRITEMASK_X << j;
1457 		  break;
1458 	       case TXG_NONE:
1459 		  copy_mask |= WRITEMASK_X << j;
1460 	       }
1461 	    }
1462 
1463 	    if (sphere_mask) {
1464 	       build_sphere_texgen(p, out_texgen, sphere_mask);
1465 	    }
1466 
1467 	    if (reflect_mask) {
1468 	       build_reflect_texgen(p, out_texgen, reflect_mask);
1469 	    }
1470 
1471 	    if (normal_mask) {
1472 	       struct ureg normal = get_transformed_normal(p);
1473 	       emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1474 	    }
1475 
1476 	    if (copy_mask) {
1477 	       struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1478 	       emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1479 	    }
1480 	 }
1481 
1482 	 if (texmat_enabled) {
1483 	    struct ureg texmat[4];
1484 	    struct ureg in = (!is_undef(out_texgen) ?
1485 			      out_texgen :
1486 			      register_input(p, VERT_ATTRIB_TEX0+i));
1487 	    if (p->mvp_with_dp4) {
1488 	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1489 				       0, texmat );
1490 	       emit_matrix_transform_vec4( p, out, texmat, in );
1491 	    }
1492 	    else {
1493 	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1494 				       STATE_MATRIX_TRANSPOSE, texmat );
1495 	       emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1496 	    }
1497 	 }
1498 
1499 	 release_temps(p);
1500       }
1501       else {
1502 	 emit_passthrough(p, VERT_ATTRIB_TEX0+i, VARYING_SLOT_TEX0+i);
1503       }
1504    }
1505 }
1506 
1507 
1508 /**
1509  * Point size attenuation computation.
1510  */
build_atten_pointsize(struct tnl_program * p)1511 static void build_atten_pointsize( struct tnl_program *p )
1512 {
1513    struct ureg eye = get_eye_position_z(p);
1514    struct ureg state_size = register_param2(p, STATE_INTERNAL, STATE_POINT_SIZE_CLAMPED);
1515    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1516    struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1517    struct ureg ut = get_temp(p);
1518 
1519    /* dist = |eyez| */
1520    emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
1521    /* p1 + dist * (p2 + dist * p3); */
1522    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1523 		swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1524    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1525 		ut, swizzle1(state_attenuation, X));
1526 
1527    /* 1 / sqrt(factor) */
1528    emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1529 
1530 #if 0
1531    /* out = pointSize / sqrt(factor) */
1532    emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1533 #else
1534    /* this is a good place to clamp the point size since there's likely
1535     * no hardware registers to clamp point size at rasterization time.
1536     */
1537    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1538    emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1539    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1540 #endif
1541 
1542    release_temp(p, ut);
1543 }
1544 
1545 
1546 /**
1547  * Pass-though per-vertex point size, from user's point size array.
1548  */
build_array_pointsize(struct tnl_program * p)1549 static void build_array_pointsize( struct tnl_program *p )
1550 {
1551    struct ureg in = register_input(p, VERT_ATTRIB_POINT_SIZE);
1552    struct ureg out = register_output(p, VARYING_SLOT_PSIZ);
1553    emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, in);
1554 }
1555 
1556 
build_tnl_program(struct tnl_program * p)1557 static void build_tnl_program( struct tnl_program *p )
1558 {
1559    /* Emit the program, starting with the modelview, projection transforms:
1560     */
1561    build_hpos(p);
1562 
1563    /* Lighting calculations:
1564     */
1565    if (p->state->fragprog_inputs_read & (VARYING_BIT_COL0|VARYING_BIT_COL1)) {
1566       if (p->state->light_global_enabled)
1567 	 build_lighting(p);
1568       else {
1569 	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL0)
1570 	    emit_passthrough(p, VERT_ATTRIB_COLOR0, VARYING_SLOT_COL0);
1571 
1572 	 if (p->state->fragprog_inputs_read & VARYING_BIT_COL1)
1573 	    emit_passthrough(p, VERT_ATTRIB_COLOR1, VARYING_SLOT_COL1);
1574       }
1575    }
1576 
1577    if (p->state->fragprog_inputs_read & VARYING_BIT_FOGC)
1578       build_fog(p);
1579 
1580    if (p->state->fragprog_inputs_read & VARYING_BITS_TEX_ANY)
1581       build_texture_transform(p);
1582 
1583    if (p->state->point_attenuated)
1584       build_atten_pointsize(p);
1585    else if (p->state->varying_vp_inputs & VERT_BIT_POINT_SIZE)
1586       build_array_pointsize(p);
1587 
1588    /* Finish up:
1589     */
1590    emit_op1(p, OPCODE_END, undef, 0, undef);
1591 
1592    /* Disassemble:
1593     */
1594    if (DISASSEM) {
1595       printf ("\n");
1596    }
1597 }
1598 
1599 
1600 static void
create_new_program(const struct state_key * key,struct gl_program * program,GLboolean mvp_with_dp4,GLuint max_temps)1601 create_new_program( const struct state_key *key,
1602                     struct gl_program *program,
1603                     GLboolean mvp_with_dp4,
1604                     GLuint max_temps)
1605 {
1606    struct tnl_program p;
1607 
1608    memset(&p, 0, sizeof(p));
1609    p.state = key;
1610    p.program = program;
1611    p.eye_position = undef;
1612    p.eye_position_z = undef;
1613    p.eye_position_normalized = undef;
1614    p.transformed_normal = undef;
1615    p.identity = undef;
1616    p.temp_in_use = 0;
1617    p.mvp_with_dp4 = mvp_with_dp4;
1618 
1619    if (max_temps >= sizeof(int) * 8)
1620       p.temp_reserved = 0;
1621    else
1622       p.temp_reserved = ~((1<<max_temps)-1);
1623 
1624    /* Start by allocating 32 instructions.
1625     * If we need more, we'll grow the instruction array as needed.
1626     */
1627    p.max_inst = 32;
1628    p.program->arb.Instructions =
1629       rzalloc_array(program, struct prog_instruction, p.max_inst);
1630    p.program->String = NULL;
1631    p.program->arb.NumInstructions =
1632    p.program->arb.NumTemporaries =
1633    p.program->arb.NumParameters =
1634    p.program->arb.NumAttributes = p.program->arb.NumAddressRegs = 0;
1635    p.program->Parameters = _mesa_new_parameter_list();
1636    p.program->info.inputs_read = 0;
1637    p.program->info.outputs_written = 0;
1638 
1639    build_tnl_program( &p );
1640 }
1641 
1642 
1643 /**
1644  * Return a vertex program which implements the current fixed-function
1645  * transform/lighting/texgen operations.
1646  */
1647 struct gl_program *
_mesa_get_fixed_func_vertex_program(struct gl_context * ctx)1648 _mesa_get_fixed_func_vertex_program(struct gl_context *ctx)
1649 {
1650    struct gl_program *prog;
1651    struct state_key key;
1652 
1653    /* We only update ctx->varying_vp_inputs when in VP_MODE_FF _VPMode */
1654    assert(VP_MODE_FF == ctx->VertexProgram._VPMode);
1655 
1656    /* Grab all the relevant state and put it in a single structure:
1657     */
1658    make_state_key(ctx, &key);
1659 
1660    /* Look for an already-prepared program for this state:
1661     */
1662    prog = _mesa_search_program_cache(ctx->VertexProgram.Cache, &key,
1663                                      sizeof(key));
1664 
1665    if (!prog) {
1666       /* OK, we'll have to build a new one */
1667       if (0)
1668          printf("Build new TNL program\n");
1669 
1670       prog = ctx->Driver.NewProgram(ctx, MESA_SHADER_VERTEX, 0, true);
1671       if (!prog)
1672          return NULL;
1673 
1674       create_new_program( &key, prog,
1675                           ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].OptimizeForAOS,
1676                           ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps );
1677 
1678       if (ctx->Driver.ProgramStringNotify)
1679          ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB, prog);
1680 
1681       _mesa_program_cache_insert(ctx, ctx->VertexProgram.Cache, &key,
1682                                  sizeof(key), prog);
1683    }
1684 
1685    return prog;
1686 }
1687