src/mesa/tnl/t_vp_build.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.5
   4  *
   5  * Copyright (C) 2006  Tungsten Graphics   All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * TUNGSTEN GRAPHICS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  21  * WHETHER IN
  22  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  23  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  */
  25
  26 /**
  27  * \file t_vp_build.c
  28  * Create a vertex program to execute the current fixed function T&L pipeline.
  29  * \author Keith Whitwell
  30  */
  31
  32
  33 #include "glheader.h"
  34 #include "macros.h"
  35 #include "enums.h"
  36 #include "t_context.h" /* NOTE: very light dependency on this */
  37 #include "t_vp_build.h"
  38
  39 #include "shader/program.h"
  40 #include "shader/program_instruction.h"
  41
  42 struct state_key {
  43    unsigned light_global_enabled:1;
  44    unsigned light_local_viewer:1;
  45    unsigned light_twoside:1;
  46    unsigned light_color_material:1;
  47    unsigned light_color_material_mask:12;
  48    unsigned light_material_mask:12;
  49
  50    unsigned normalize:1;
  51    unsigned rescale_normals:1;
  52    unsigned fog_source_is_depth:1;
  53    unsigned tnl_do_vertex_fog:1;
  54    unsigned separate_specular:1;
  55    unsigned fog_mode:2;
  56    unsigned point_attenuated:1;
  57    unsigned texture_enabled_global:1;
  58    unsigned fragprog_inputs_read:12;
  59
  60    struct {
  61       unsigned light_enabled:1;
  62       unsigned light_eyepos3_is_zero:1;
  63       unsigned light_spotcutoff_is_180:1;
  64       unsigned light_attenuated:1;
  65       unsigned texunit_really_enabled:1;
  66       unsigned texmat_enabled:1;
  67       unsigned texgen_enabled:4;
  68       unsigned texgen_mode0:4;
  69       unsigned texgen_mode1:4;
  70       unsigned texgen_mode2:4;
  71       unsigned texgen_mode3:4;
  72    } unit[8];
  73 };
  74
  75
  76
  77 #define FOG_NONE   0
  78 #define FOG_LINEAR 1
  79 #define FOG_EXP    2
  80 #define FOG_EXP2   3
  81
  82 static GLuint translate_fog_mode( GLenum mode )
  83 {
  84    switch (mode) {
  85    case GL_LINEAR: return FOG_LINEAR;
  86    case GL_EXP: return FOG_EXP;
  87    case GL_EXP2: return FOG_EXP2;
  88    default: return FOG_NONE;
  89    }
  90 }
  91
  92 #define TXG_NONE           0
  93 #define TXG_OBJ_LINEAR     1
  94 #define TXG_EYE_LINEAR     2
  95 #define TXG_SPHERE_MAP     3
  96 #define TXG_REFLECTION_MAP 4
  97 #define TXG_NORMAL_MAP     5
  98
  99 static GLuint translate_texgen( GLboolean enabled, GLenum mode )
 100 {
 101    if (!enabled)
 102       return TXG_NONE;
 103
 104    switch (mode) {
 105    case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
 106    case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
 107    case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
 108    case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
 109    case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
 110    default: return TXG_NONE;
 111    }
 112 }
 113
 114 static struct state_key *make_state_key( GLcontext *ctx )
 115 {
 116    TNLcontext *tnl = TNL_CONTEXT(ctx);
 117    struct vertex_buffer *VB = &tnl->vb;
 118    const struct gl_fragment_program *fp = ctx->FragmentProgram._Current;
 119    struct state_key *key = CALLOC_STRUCT(state_key);
 120    GLuint i;
 121
 122    /* This now relies on texenvprogram.c being active:
 123     */
 124    assert(fp);
 125
 126    key->fragprog_inputs_read = fp->Base.InputsRead;
 127
 128    key->separate_specular = (ctx->Light.Model.ColorControl ==
 129                              GL_SEPARATE_SPECULAR_COLOR);
 130
 131    if (ctx->Light.Enabled) {
 132       key->light_global_enabled = 1;
 133
 134       if (ctx->Light.Model.LocalViewer)
 135          key->light_local_viewer = 1;
 136
 137       if (ctx->Light.Model.TwoSide)
 138          key->light_twoside = 1;
 139
 140       if (ctx->Light.ColorMaterialEnabled) {
 141          key->light_color_material = 1;
 142          key->light_color_material_mask = ctx->Light.ColorMaterialBitmask;
 143       }
 144
 145       for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++)
 146          if (VB->AttribPtr[i]->stride)
 147             key->light_material_mask |= 1<<(i-_TNL_ATTRIB_MAT_FRONT_AMBIENT);
 148
 149       for (i = 0; i < MAX_LIGHTS; i++) {
 150          struct gl_light *light = &ctx->Light.Light[i];
 151
 152          if (light->Enabled) {
 153             key->unit[i].light_enabled = 1;
 154
 155             if (light->EyePosition[3] == 0.0)
 156                key->unit[i].light_eyepos3_is_zero = 1;
 157
 158             if (light->SpotCutoff == 180.0)
 159                key->unit[i].light_spotcutoff_is_180 = 1;
 160
 161             if (light->ConstantAttenuation != 1.0 ||
 162                 light->LinearAttenuation != 0.0 ||
 163                 light->QuadraticAttenuation != 0.0)
 164                key->unit[i].light_attenuated = 1;
 165          }
 166       }
 167    }
 168
 169    if (ctx->Transform.Normalize)
 170       key->normalize = 1;
 171
 172    if (ctx->Transform.RescaleNormals)
 173       key->rescale_normals = 1;
 174
 175    key->fog_mode = translate_fog_mode(fp->FogOption);
 176
 177    if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT)
 178       key->fog_source_is_depth = 1;
 179
 180    if (tnl->_DoVertexFog)
 181       key->tnl_do_vertex_fog = 1;
 182
 183    if (ctx->Point._Attenuated)
 184       key->point_attenuated = 1;
 185
 186    if (ctx->Texture._TexGenEnabled ||
 187        ctx->Texture._TexMatEnabled ||
 188        ctx->Texture._EnabledUnits)
 189       key->texture_enabled_global = 1;
 190
 191    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 192       struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
 193
 194       if (texUnit->_ReallyEnabled)
 195          key->unit[i].texunit_really_enabled = 1;
 196
 197       if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
 198          key->unit[i].texmat_enabled = 1;
 199
 200       if (texUnit->TexGenEnabled) {
 201          key->unit[i].texgen_enabled = 1;
 202
 203          key->unit[i].texgen_mode0 =
 204             translate_texgen( texUnit->TexGenEnabled & (1<<0),
 205                               texUnit->GenModeS );
 206          key->unit[i].texgen_mode1 =
 207             translate_texgen( texUnit->TexGenEnabled & (1<<1),
 208                               texUnit->GenModeT );
 209          key->unit[i].texgen_mode2 =
 210             translate_texgen( texUnit->TexGenEnabled & (1<<2),
 211                               texUnit->GenModeR );
 212          key->unit[i].texgen_mode3 =
 213             translate_texgen( texUnit->TexGenEnabled & (1<<3),
 214                               texUnit->GenModeQ );
 215       }
 216    }
 217
 218    return key;
 219 }
 220
 221
 222
 223 /* Very useful debugging tool - produces annotated listing of
 224  * generated program with line/function references for each
 225  * instruction back into this file:
 226  */
 227 #define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM)
 228
 229 /* Should be tunable by the driver - do we want to do matrix
 230  * multiplications with DP4's or with MUL/MAD's?  SSE works better
 231  * with the latter, drivers may differ.
 232  */
 233 #define PREFER_DP4 0
 234
 235 #define MAX_INSN 256
 236
 237 /* Use uregs to represent registers internally, translate to Mesa's
 238  * expected formats on emit.
 239  *
 240  * NOTE: These are passed by value extensively in this file rather
 241  * than as usual by pointer reference.  If this disturbs you, try
 242  * remembering they are just 32bits in size.
 243  *
 244  * GCC is smart enough to deal with these dword-sized structures in
 245  * much the same way as if I had defined them as dwords and was using
 246  * macros to access and set the fields.  This is much nicer and easier
 247  * to evolve.
 248  */
 249 struct ureg {
 250    GLuint file:4;
 251    GLint idx:8;      /* relative addressing may be negative */
 252    GLuint negate:1;
 253    GLuint swz:12;
 254    GLuint pad:7;
 255 };
 256
 257
 258 struct tnl_program {
 259    const struct state_key *state;
 260    struct gl_vertex_program *program;
 261
 262    GLuint temp_in_use;
 263    GLuint temp_reserved;
 264
 265    struct ureg eye_position;
 266    struct ureg eye_position_normalized;
 267    struct ureg eye_normal;
 268    struct ureg identity;
 269
 270    GLuint materials;
 271    GLuint color_materials;
 272 };
 273
 274
 275 static const struct ureg undef = {
 276    PROGRAM_UNDEFINED,
 277    ~0,
 278    0,
 279    0,
 280    0
 281 };
 282
 283 /* Local shorthand:
 284  */
 285 #define X    SWIZZLE_X
 286 #define Y    SWIZZLE_Y
 287 #define Z    SWIZZLE_Z
 288 #define W    SWIZZLE_W
 289
 290
 291 /* Construct a ureg:
 292  */
 293 static struct ureg make_ureg(GLuint file, GLint idx)
 294 {
 295    struct ureg reg;
 296    reg.file = file;
 297    reg.idx = idx;
 298    reg.negate = 0;
 299    reg.swz = SWIZZLE_NOOP;
 300    reg.pad = 0;
 301    return reg;
 302 }
 303
 304
 305
 306 static struct ureg negate( struct ureg reg )
 307 {
 308    reg.negate ^= 1;
 309    return reg;
 310 }
 311
 312
 313 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 314 {
 315    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 316                            GET_SWZ(reg.swz, y),
 317                            GET_SWZ(reg.swz, z),
 318                            GET_SWZ(reg.swz, w));
 319
 320    return reg;
 321 }
 322
 323 static struct ureg swizzle1( struct ureg reg, int x )
 324 {
 325    return swizzle(reg, x, x, x, x);
 326 }
 327
 328 static struct ureg get_temp( struct tnl_program *p )
 329 {
 330    int bit = _mesa_ffs( ~p->temp_in_use );
 331    if (!bit) {
 332       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
 333       _mesa_exit(1);
 334    }
 335
 336    if ((GLuint) bit > p->program->Base.NumTemporaries)
 337       p->program->Base.NumTemporaries = bit;
 338
 339    p->temp_in_use |= 1<<(bit-1);
 340    return make_ureg(PROGRAM_TEMPORARY, bit-1);
 341 }
 342
 343 static struct ureg reserve_temp( struct tnl_program *p )
 344 {
 345    struct ureg temp = get_temp( p );
 346    p->temp_reserved |= 1<<temp.idx;
 347    return temp;
 348 }
 349
 350 static void release_temp( struct tnl_program *p, struct ureg reg )
 351 {
 352    if (reg.file == PROGRAM_TEMPORARY) {
 353       p->temp_in_use &= ~(1<<reg.idx);
 354       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
 355    }
 356 }
 357
 358 static void release_temps( struct tnl_program *p )
 359 {
 360    p->temp_in_use = p->temp_reserved;
 361 }
 362
 363
 364
 365 static struct ureg register_input( struct tnl_program *p, GLuint input )
 366 {
 367    p->program->Base.InputsRead |= (1<<input);
 368    return make_ureg(PROGRAM_INPUT, input);
 369 }
 370
 371 static struct ureg register_output( struct tnl_program *p, GLuint output )
 372 {
 373    p->program->Base.OutputsWritten |= (1<<output);
 374    return make_ureg(PROGRAM_OUTPUT, output);
 375 }
 376
 377 static struct ureg register_const4f( struct tnl_program *p,
 378                               GLfloat s0,
 379                               GLfloat s1,
 380                               GLfloat s2,
 381                               GLfloat s3)
 382 {
 383    GLfloat values[4];
 384    GLint idx;
 385    values[0] = s0;
 386    values[1] = s1;
 387    values[2] = s2;
 388    values[3] = s3;
 389    idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4 );
 390    return make_ureg(PROGRAM_STATE_VAR, idx);
 391 }
 392
 393 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
 394 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
 395 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
 396 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
 397
 398 static GLboolean is_undef( struct ureg reg )
 399 {
 400    return reg.file == PROGRAM_UNDEFINED;
 401 }
 402
 403 static struct ureg get_identity_param( struct tnl_program *p )
 404 {
 405    if (is_undef(p->identity))
 406       p->identity = register_const4f(p, 0,0,0,1);
 407
 408    return p->identity;
 409 }
 410
 411 static struct ureg register_param6( struct tnl_program *p,
 412                                    GLint s0,
 413                                    GLint s1,
 414                                    GLint s2,
 415                                    GLint s3,
 416                                    GLint s4,
 417                                    GLint s5)
 418 {
 419    GLint tokens[6];
 420    GLint idx;
 421    tokens[0] = s0;
 422    tokens[1] = s1;
 423    tokens[2] = s2;
 424    tokens[3] = s3;
 425    tokens[4] = s4;
 426    tokens[5] = s5;
 427    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 428    return make_ureg(PROGRAM_STATE_VAR, idx);
 429 }
 430
 431
 432 #define register_param1(p,s0)          register_param6(p,s0,0,0,0,0,0)
 433 #define register_param2(p,s0,s1)       register_param6(p,s0,s1,0,0,0,0)
 434 #define register_param3(p,s0,s1,s2)    register_param6(p,s0,s1,s2,0,0,0)
 435 #define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0)
 436
 437
 438 static void register_matrix_param6( struct tnl_program *p,
 439                                     GLint s0,
 440                                     GLint s1,
 441                                     GLint s2,
 442                                     GLint s3,
 443                                     GLint s4,
 444                                     GLint s5,
 445                                     struct ureg *matrix )
 446 {
 447    GLint i;
 448
 449    /* This is a bit sad as the support is there to pull the whole
 450     * matrix out in one go:
 451     */
 452    for (i = 0; i <= s4 - s3; i++)
 453       matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 );
 454 }
 455
 456
 457 static void emit_arg( struct prog_src_register *src,
 458                       struct ureg reg )
 459 {
 460    src->File = reg.file;
 461    src->Index = reg.idx;
 462    src->Swizzle = reg.swz;
 463    src->NegateBase = reg.negate ? NEGATE_XYZW : 0;
 464    src->Abs = 0;
 465    src->NegateAbs = 0;
 466    src->RelAddr = 0;
 467 }
 468
 469 static void emit_dst( struct prog_dst_register *dst,
 470                       struct ureg reg, GLuint mask )
 471 {
 472    dst->File = reg.file;
 473    dst->Index = reg.idx;
 474    /* allow zero as a shorthand for xyzw */
 475    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
 476    dst->CondMask = COND_TR;
 477    dst->CondSwizzle = 0;
 478    dst->CondSrc = 0;
 479    dst->pad = 0;
 480 }
 481
 482 static void debug_insn( struct prog_instruction *inst, const char *fn,
 483                         GLuint line )
 484 {
 485    if (DISASSEM) {
 486       static const char *last_fn;
 487
 488       if (fn != last_fn) {
 489          last_fn = fn;
 490          _mesa_printf("%s:\n", fn);
 491       }
 492
 493       _mesa_printf("%d:\t", line);
 494       _mesa_print_instruction(inst);
 495    }
 496 }
 497
 498
 499 static void emit_op3fn(struct tnl_program *p,
 500                        GLuint op,
 501                        struct ureg dest,
 502                        GLuint mask,
 503                        struct ureg src0,
 504                        struct ureg src1,
 505                        struct ureg src2,
 506                        const char *fn,
 507                        GLuint line)
 508 {
 509    GLuint nr = p->program->Base.NumInstructions++;
 510    struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 511
 512    if (p->program->Base.NumInstructions > MAX_INSN) {
 513       _mesa_problem(0, "Out of instructions in emit_op3fn\n");
 514       return;
 515    }
 516
 517    inst->Opcode = (enum prog_opcode) op;
 518    inst->StringPos = 0;
 519    inst->Data = 0;
 520
 521    emit_arg( &inst->SrcReg[0], src0 );
 522    emit_arg( &inst->SrcReg[1], src1 );
 523    emit_arg( &inst->SrcReg[2], src2 );
 524
 525    emit_dst( &inst->DstReg, dest, mask );
 526
 527    debug_insn(inst, fn, line);
 528 }
 529
 530
 531 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 532    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 533
 534 #define emit_op2(p, op, dst, mask, src0, src1) \
 535     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 536
 537 #define emit_op1(p, op, dst, mask, src0) \
 538     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 539
 540
 541 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 542 {
 543    if (reg.file == PROGRAM_TEMPORARY &&
 544        !(p->temp_reserved & (1<<reg.idx)))
 545       return reg;
 546    else {
 547       struct ureg temp = get_temp(p);
 548       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 549       return temp;
 550    }
 551 }
 552
 553
 554 /* Currently no tracking performed of input/output/register size or
 555  * active elements.  Could be used to reduce these operations, as
 556  * could the matrix type.
 557  */
 558 static void emit_matrix_transform_vec4( struct tnl_program *p,
 559                                         struct ureg dest,
 560                                         const struct ureg *mat,
 561                                         struct ureg src)
 562 {
 563    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 564    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 565    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 566    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 567 }
 568
 569 /* This version is much easier to implement if writemasks are not
 570  * supported natively on the target or (like SSE), the target doesn't
 571  * have a clean/obvious dotproduct implementation.
 572  */
 573 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 574                                                   struct ureg dest,
 575                                                   const struct ureg *mat,
 576                                                   struct ureg src)
 577 {
 578    struct ureg tmp;
 579
 580    if (dest.file != PROGRAM_TEMPORARY)
 581       tmp = get_temp(p);
 582    else
 583       tmp = dest;
 584
 585    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 586    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 587    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 588    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 589
 590    if (dest.file != PROGRAM_TEMPORARY)
 591       release_temp(p, tmp);
 592 }
 593
 594 static void emit_matrix_transform_vec3( struct tnl_program *p,
 595                                         struct ureg dest,
 596                                         const struct ureg *mat,
 597                                         struct ureg src)
 598 {
 599    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 600    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 601    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 602 }
 603
 604
 605 static void emit_normalize_vec3( struct tnl_program *p,
 606                                  struct ureg dest,
 607                                  struct ureg src )
 608 {
 609    struct ureg tmp = get_temp(p);
 610    emit_op2(p, OPCODE_DP3, tmp, 0, src, src);
 611    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
 612    emit_op2(p, OPCODE_MUL, dest, 0, src, tmp);
 613    release_temp(p, tmp);
 614 }
 615
 616 static void emit_passthrough( struct tnl_program *p,
 617                               GLuint input,
 618                               GLuint output )
 619 {
 620    struct ureg out = register_output(p, output);
 621    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 622 }
 623
 624 static struct ureg get_eye_position( struct tnl_program *p )
 625 {
 626    if (is_undef(p->eye_position)) {
 627       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 628       struct ureg modelview[4];
 629
 630       p->eye_position = reserve_temp(p);
 631
 632       if (PREFER_DP4) {
 633          register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3,
 634                                  STATE_MATRIX, modelview );
 635
 636          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 637       }
 638       else {
 639          register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3,
 640                                  STATE_MATRIX_TRANSPOSE, modelview );
 641
 642          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 643       }
 644    }
 645
 646    return p->eye_position;
 647 }
 648
 649
 650 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 651 {
 652    if (is_undef(p->eye_position_normalized)) {
 653       struct ureg eye = get_eye_position(p);
 654       p->eye_position_normalized = reserve_temp(p);
 655       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 656    }
 657
 658    return p->eye_position_normalized;
 659 }
 660
 661
 662 static struct ureg get_eye_normal( struct tnl_program *p )
 663 {
 664    if (is_undef(p->eye_normal)) {
 665       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
 666       struct ureg mvinv[3];
 667
 668       register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2,
 669                               STATE_MATRIX_INVTRANS, mvinv );
 670
 671       p->eye_normal = reserve_temp(p);
 672
 673       /* Transform to eye space:
 674        */
 675       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 676
 677       /* Normalize/Rescale:
 678        */
 679       if (p->state->normalize) {
 680          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 681       }
 682       else if (p->state->rescale_normals) {
 683          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 684                                                STATE_NORMAL_SCALE);
 685
 686          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, normal,
 687                    swizzle1(rescale, X));
 688       }
 689    }
 690
 691    return p->eye_normal;
 692 }
 693
 694
 695
 696 static void build_hpos( struct tnl_program *p )
 697 {
 698    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 699    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 700    struct ureg mvp[4];
 701
 702    if (PREFER_DP4) {
 703       register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3,
 704                               STATE_MATRIX, mvp );
 705       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 706    }
 707    else {
 708       register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3,
 709                               STATE_MATRIX_TRANSPOSE, mvp );
 710       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 711    }
 712 }
 713
 714
 715 static GLuint material_attrib( GLuint side, GLuint property )
 716 {
 717    return ((property - STATE_AMBIENT) * 2 +
 718            side);
 719 }
 720
 721 /* Get a bitmask of which material values vary on a per-vertex basis.
 722  */
 723 static void set_material_flags( struct tnl_program *p )
 724 {
 725    p->color_materials = 0;
 726    p->materials = 0;
 727
 728    if (p->state->light_color_material) {
 729       p->materials =
 730          p->color_materials = p->state->light_color_material_mask;
 731    }
 732
 733    p->materials |= p->state->light_material_mask;
 734 }
 735
 736
 737 static struct ureg get_material( struct tnl_program *p, GLuint side,
 738                                  GLuint property )
 739 {
 740    GLuint attrib = material_attrib(side, property);
 741
 742    if (p->color_materials & (1<<attrib))
 743       return register_input(p, VERT_ATTRIB_COLOR0);
 744    else if (p->materials & (1<<attrib))
 745       return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
 746    else
 747       return register_param3( p, STATE_MATERIAL, side, property );
 748 }
 749
 750 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
 751                                    MAT_BIT_FRONT_AMBIENT | \
 752                                    MAT_BIT_FRONT_DIFFUSE) << (side))
 753
 754 /* Either return a precalculated constant value or emit code to
 755  * calculate these values dynamically in the case where material calls
 756  * are present between begin/end pairs.
 757  *
 758  * Probably want to shift this to the program compilation phase - if
 759  * we always emitted the calculation here, a smart compiler could
 760  * detect that it was constant (given a certain set of inputs), and
 761  * lift it out of the main loop.  That way the programs created here
 762  * would be independent of the vertex_buffer details.
 763  */
 764 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 765 {
 766    if (p->materials & SCENE_COLOR_BITS(side)) {
 767       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 768       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 769       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 770       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 771       struct ureg tmp = make_temp(p, material_diffuse);
 772       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 773                material_ambient, material_emission);
 774       return tmp;
 775    }
 776    else
 777       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 778 }
 779
 780
 781 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 782                                   GLuint side, GLuint property )
 783 {
 784    GLuint attrib = material_attrib(side, property);
 785    if (p->materials & (1<<attrib)) {
 786       struct ureg light_value =
 787          register_param3(p, STATE_LIGHT, light, property);
 788       struct ureg material_value = get_material(p, side, property);
 789       struct ureg tmp = get_temp(p);
 790       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 791       return tmp;
 792    }
 793    else
 794       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 795 }
 796
 797 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 798                                                 GLuint i,
 799                                                 struct ureg VPpli,
 800                                                 struct ureg dist )
 801 {
 802    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 803                                              STATE_ATTENUATION);
 804    struct ureg att = get_temp(p);
 805
 806    /* Calculate spot attenuation:
 807     */
 808    if (!p->state->unit[i].light_spotcutoff_is_180) {
 809       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
 810                                                   STATE_SPOT_DIR_NORMALIZED, i);
 811       struct ureg spot = get_temp(p);
 812       struct ureg slt = get_temp(p);
 813
 814       emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
 815       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
 816       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 817       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 818
 819       release_temp(p, spot);
 820       release_temp(p, slt);
 821    }
 822
 823    /* Calculate distance attenuation:
 824     */
 825    if (p->state->unit[i].light_attenuated) {
 826
 827       /* 1/d,d,d,1/d */
 828       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 829       /* 1,d,d*d,1/d */
 830       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 831       /* 1/dist-atten */
 832       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 833
 834       if (!p->state->unit[i].light_spotcutoff_is_180) {
 835          /* dist-atten */
 836          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 837          /* spot-atten * dist-atten */
 838          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 839       } else {
 840          /* dist-atten */
 841          emit_op1(p, OPCODE_RCP, att, 0, dist);
 842       }
 843    }
 844
 845    return att;
 846 }
 847
 848
 849
 850
 851
 852 /* Need to add some addtional parameters to allow lighting in object
 853  * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye
 854  * space lighting.
 855  */
 856 static void build_lighting( struct tnl_program *p )
 857 {
 858    const GLboolean twoside = p->state->light_twoside;
 859    const GLboolean separate = p->state->separate_specular;
 860    GLuint nr_lights = 0, count = 0;
 861    struct ureg normal = get_eye_normal(p);
 862    struct ureg lit = get_temp(p);
 863    struct ureg dots = get_temp(p);
 864    struct ureg _col0 = undef, _col1 = undef;
 865    struct ureg _bfc0 = undef, _bfc1 = undef;
 866    GLuint i;
 867
 868    for (i = 0; i < MAX_LIGHTS; i++)
 869       if (p->state->unit[i].light_enabled)
 870          nr_lights++;
 871
 872    set_material_flags(p);
 873
 874    {
 875       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 876       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 877       release_temp(p, shininess);
 878
 879       _col0 = make_temp(p, get_scenecolor(p, 0));
 880       if (separate)
 881          _col1 = make_temp(p, get_identity_param(p));
 882       else
 883          _col1 = _col0;
 884
 885    }
 886
 887    if (twoside) {
 888       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 889       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 890                negate(swizzle1(shininess,X)));
 891       release_temp(p, shininess);
 892
 893       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 894       if (separate)
 895          _bfc1 = make_temp(p, get_identity_param(p));
 896       else
 897          _bfc1 = _bfc0;
 898    }
 899
 900
 901    /* If no lights, still need to emit the scenecolor.
 902     */
 903       {
 904          struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 905          emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 906       }
 907
 908       if (separate) {
 909          struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 910          emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 911       }
 912
 913       if (twoside) {
 914          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 915          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 916       }
 917
 918       if (twoside && separate) {
 919          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 920          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 921       }
 922
 923    if (nr_lights == 0) {
 924       release_temps(p);
 925       return;
 926    }
 927
 928
 929    for (i = 0; i < MAX_LIGHTS; i++) {
 930       if (p->state->unit[i].light_enabled) {
 931          struct ureg half = undef;
 932          struct ureg att = undef, VPpli = undef;
 933
 934          count++;
 935
 936          if (p->state->unit[i].light_eyepos3_is_zero) {
 937             /* Can used precomputed constants in this case.
 938              * Attenuation never applies to infinite lights.
 939              */
 940             VPpli = register_param3(p, STATE_LIGHT, i,
 941                                     STATE_POSITION_NORMALIZED);
 942             if (p->state->light_local_viewer) {
 943                 struct ureg eye_hat = get_eye_position_normalized(p);
 944                 half = get_temp(p);
 945                 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 946                 emit_normalize_vec3(p, half, half);
 947             } else {
 948                 half = register_param3(p, STATE_LIGHT, i, STATE_HALF);
 949             }
 950          }
 951          else {
 952             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
 953                                                STATE_POSITION);
 954             struct ureg V = get_eye_position(p);
 955             struct ureg dist = get_temp(p);
 956
 957             VPpli = get_temp(p);
 958             half = get_temp(p);
 959
 960             /* Calulate VPpli vector
 961              */
 962             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
 963
 964             /* Normalize VPpli.  The dist value also used in
 965              * attenuation below.
 966              */
 967             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
 968             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
 969             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
 970
 971
 972             /* Calculate  attenuation:
 973              */
 974             if (!p->state->unit[i].light_spotcutoff_is_180 ||
 975                 p->state->unit[i].light_attenuated) {
 976                att = calculate_light_attenuation(p, i, VPpli, dist);
 977             }
 978
 979
 980             /* Calculate viewer direction, or use infinite viewer:
 981              */
 982             if (p->state->light_local_viewer) {
 983                struct ureg eye_hat = get_eye_position_normalized(p);
 984                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 985             }
 986             else {
 987                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
 988                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
 989             }
 990
 991             emit_normalize_vec3(p, half, half);
 992
 993             release_temp(p, dist);
 994          }
 995
 996          /* Calculate dot products:
 997           */
 998          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
 999          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1000
1001
1002          /* Front face lighting:
1003           */
1004          {
1005             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1006             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1007             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1008             struct ureg res0, res1;
1009             GLuint mask0, mask1;
1010
1011             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1012
1013             if (!is_undef(att))
1014                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1015
1016
1017             if (count == nr_lights) {
1018                if (separate) {
1019                   mask0 = WRITEMASK_XYZ;
1020                   mask1 = WRITEMASK_XYZ;
1021                   res0 = register_output( p, VERT_RESULT_COL0 );
1022                   res1 = register_output( p, VERT_RESULT_COL1 );
1023                }
1024                else {
1025                   mask0 = 0;
1026                   mask1 = WRITEMASK_XYZ;
1027                   res0 = _col0;
1028                   res1 = register_output( p, VERT_RESULT_COL0 );
1029                }
1030             } else {
1031                mask0 = 0;
1032                mask1 = 0;
1033                res0 = _col0;
1034                res1 = _col1;
1035             }
1036
1037             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1038             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1039             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1040
1041             release_temp(p, ambient);
1042             release_temp(p, diffuse);
1043             release_temp(p, specular);
1044          }
1045
1046          /* Back face lighting:
1047           */
1048          if (twoside) {
1049             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1050             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1051             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1052             struct ureg res0, res1;
1053             GLuint mask0, mask1;
1054
1055             emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
1056
1057             if (!is_undef(att))
1058                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1059
1060             if (count == nr_lights) {
1061                if (separate) {
1062                   mask0 = WRITEMASK_XYZ;
1063                   mask1 = WRITEMASK_XYZ;
1064                   res0 = register_output( p, VERT_RESULT_BFC0 );
1065                   res1 = register_output( p, VERT_RESULT_BFC1 );
1066                }
1067                else {
1068                   mask0 = 0;
1069                   mask1 = WRITEMASK_XYZ;
1070                   res0 = _bfc0;
1071                   res1 = register_output( p, VERT_RESULT_BFC0 );
1072                }
1073             } else {
1074                res0 = _bfc0;
1075                res1 = _bfc1;
1076                mask0 = 0;
1077                mask1 = 0;
1078             }
1079
1080             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1081             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1082             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1083
1084             release_temp(p, ambient);
1085             release_temp(p, diffuse);
1086             release_temp(p, specular);
1087          }
1088
1089          release_temp(p, half);
1090          release_temp(p, VPpli);
1091          release_temp(p, att);
1092       }
1093    }
1094
1095    release_temps( p );
1096 }
1097
1098
1099 static void build_fog( struct tnl_program *p )
1100 {
1101    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1102    struct ureg input;
1103
1104    if (p->state->fog_source_is_depth) {
1105       input = swizzle1(get_eye_position(p), Z);
1106    }
1107    else {
1108       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1109    }
1110
1111    if (p->state->tnl_do_vertex_fog) {
1112       struct ureg params = register_param2(p, STATE_INTERNAL,
1113                                            STATE_FOG_PARAMS_OPTIMIZED);
1114       struct ureg tmp = get_temp(p);
1115
1116       switch (p->state->fog_mode) {
1117       case FOG_LINEAR: {
1118          struct ureg id = get_identity_param(p);
1119          emit_op3(p, OPCODE_MAD, tmp, 0, input, swizzle1(params,X), swizzle1(params,Y));
1120          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1121          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1122          break;
1123       }
1124       case FOG_EXP:
1125          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1126          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,Z));
1127          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1128          break;
1129       case FOG_EXP2:
1130          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
1131          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1132          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1133          break;
1134       }
1135
1136       release_temp(p, tmp);
1137    }
1138    else {
1139       /* results = incoming fog coords (compute fog per-fragment later)
1140        *
1141        * KW:  Is it really necessary to do anything in this case?
1142        */
1143       emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1144    }
1145 }
1146
1147 static void build_reflect_texgen( struct tnl_program *p,
1148                                   struct ureg dest,
1149                                   GLuint writemask )
1150 {
1151    struct ureg normal = get_eye_normal(p);
1152    struct ureg eye_hat = get_eye_position_normalized(p);
1153    struct ureg tmp = get_temp(p);
1154
1155    /* n.u */
1156    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1157    /* 2n.u */
1158    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1159    /* (-2n.u)n + u */
1160    emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1161
1162    release_temp(p, tmp);
1163 }
1164
1165 static void build_sphere_texgen( struct tnl_program *p,
1166                                  struct ureg dest,
1167                                  GLuint writemask )
1168 {
1169    struct ureg normal = get_eye_normal(p);
1170    struct ureg eye_hat = get_eye_position_normalized(p);
1171    struct ureg tmp = get_temp(p);
1172    struct ureg half = register_scalar_const(p, .5);
1173    struct ureg r = get_temp(p);
1174    struct ureg inv_m = get_temp(p);
1175    struct ureg id = get_identity_param(p);
1176
1177    /* Could share the above calculations, but it would be
1178     * a fairly odd state for someone to set (both sphere and
1179     * reflection active for different texture coordinate
1180     * components.  Of course - if two texture units enable
1181     * reflect and/or sphere, things start to tilt in favour
1182     * of seperating this out:
1183     */
1184
1185    /* n.u */
1186    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1187    /* 2n.u */
1188    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1189    /* (-2n.u)n + u */
1190    emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1191    /* r + 0,0,1 */
1192    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1193    /* rx^2 + ry^2 + (rz+1)^2 */
1194    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1195    /* 2/m */
1196    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1197    /* 1/m */
1198    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1199    /* r/m + 1/2 */
1200    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1201
1202    release_temp(p, tmp);
1203    release_temp(p, r);
1204    release_temp(p, inv_m);
1205 }
1206
1207
1208 static void build_texture_transform( struct tnl_program *p )
1209 {
1210    GLuint i, j;
1211
1212    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1213
1214       if (!(p->state->fragprog_inputs_read & (FRAG_BIT_TEX0<<i)))
1215          continue;
1216
1217       if (p->state->unit[i].texgen_enabled ||
1218           p->state->unit[i].texmat_enabled) {
1219
1220          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1221          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1222          struct ureg out_texgen = undef;
1223
1224          if (p->state->unit[i].texgen_enabled) {
1225             GLuint copy_mask = 0;
1226             GLuint sphere_mask = 0;
1227             GLuint reflect_mask = 0;
1228             GLuint normal_mask = 0;
1229             GLuint modes[4];
1230
1231             if (texmat_enabled)
1232                out_texgen = get_temp(p);
1233             else
1234                out_texgen = out;
1235
1236             modes[0] = p->state->unit[i].texgen_mode0;
1237             modes[1] = p->state->unit[i].texgen_mode1;
1238             modes[2] = p->state->unit[i].texgen_mode2;
1239             modes[3] = p->state->unit[i].texgen_mode3;
1240
1241             for (j = 0; j < 4; j++) {
1242                switch (modes[j]) {
1243                case TXG_OBJ_LINEAR: {
1244                   struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1245                   struct ureg plane =
1246                      register_param3(p, STATE_TEXGEN, i,
1247                                      STATE_TEXGEN_OBJECT_S + j);
1248
1249                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1250                            obj, plane );
1251                   break;
1252                }
1253                case TXG_EYE_LINEAR: {
1254                   struct ureg eye = get_eye_position(p);
1255                   struct ureg plane =
1256                      register_param3(p, STATE_TEXGEN, i,
1257                                      STATE_TEXGEN_EYE_S + j);
1258
1259                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1260                            eye, plane );
1261                   break;
1262                }
1263                case TXG_SPHERE_MAP:
1264                   sphere_mask |= WRITEMASK_X << j;
1265                   break;
1266                case TXG_REFLECTION_MAP:
1267                   reflect_mask |= WRITEMASK_X << j;
1268                   break;
1269                case TXG_NORMAL_MAP:
1270                   normal_mask |= WRITEMASK_X << j;
1271                   break;
1272                case TXG_NONE:
1273                   copy_mask |= WRITEMASK_X << j;
1274                }
1275
1276             }
1277
1278
1279             if (sphere_mask) {
1280                build_sphere_texgen(p, out_texgen, sphere_mask);
1281             }
1282
1283             if (reflect_mask) {
1284                build_reflect_texgen(p, out_texgen, reflect_mask);
1285             }
1286
1287             if (normal_mask) {
1288                struct ureg normal = get_eye_normal(p);
1289                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1290             }
1291
1292             if (copy_mask) {
1293                struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1294                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1295             }
1296          }
1297
1298          if (texmat_enabled) {
1299             struct ureg texmat[4];
1300             struct ureg in = (!is_undef(out_texgen) ?
1301                               out_texgen :
1302                               register_input(p, VERT_ATTRIB_TEX0+i));
1303             if (PREFER_DP4) {
1304                register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i,
1305                                        0, 3, STATE_MATRIX, texmat );
1306                emit_matrix_transform_vec4( p, out, texmat, in );
1307             }
1308             else {
1309                register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i,
1310                                        0, 3, STATE_MATRIX_TRANSPOSE, texmat );
1311                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1312             }
1313          }
1314
1315          release_temps(p);
1316       }
1317       else {
1318          emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1319       }
1320    }
1321 }
1322
1323
1324 static void build_pointsize( struct tnl_program *p )
1325 {
1326    struct ureg eye = get_eye_position(p);
1327    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1328    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1329    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1330    struct ureg ut = get_temp(p);
1331
1332    /* p1 + dist * (p2 + dist * p3); */
1333    emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
1334                 swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1335    emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
1336                 ut, swizzle1(state_attenuation, X));
1337
1338    /* 1 / sqrt(factor) */
1339    emit_op1(p, OPCODE_RSQ, ut, 0, ut );
1340
1341 #if 1
1342    /* out = pointSize / sqrt(factor) */
1343    emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1344 #else
1345    /* not sure, might make sense to do clamping here,
1346       but it's not done in t_vb_points neither */
1347    emit_op2(p, OPCODE_MUL, ut, 0, ut, state_size);
1348    emit_op2(p, OPCODE_MAX, ut, 0, ut, swizzle1(state_size, Y));
1349    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1350 #endif
1351
1352    release_temp(p, ut);
1353 }
1354
1355 static void build_tnl_program( struct tnl_program *p )
1356 {   /* Emit the program, starting with modelviewproject:
1357     */
1358    build_hpos(p);
1359
1360    /* Lighting calculations:
1361     */
1362    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1363       if (p->state->light_global_enabled)
1364          build_lighting(p);
1365       else {
1366          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1367             emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);
1368
1369          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1370             emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
1371       }
1372    }
1373
1374    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1375        p->state->fog_mode != FOG_NONE)
1376       build_fog(p);
1377
1378    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1379       build_texture_transform(p);
1380
1381    if (p->state->point_attenuated)
1382       build_pointsize(p);
1383
1384    /* Finish up:
1385     */
1386    emit_op1(p, OPCODE_END, undef, 0, undef);
1387
1388    /* Disassemble:
1389     */
1390    if (DISASSEM) {
1391       _mesa_printf ("\n");
1392    }
1393 }
1394
1395
1396 static void
1397 create_new_program( const struct state_key *key,
1398                     struct gl_vertex_program *program,
1399                     GLuint max_temps)
1400 {
1401    struct tnl_program p;
1402
1403    _mesa_memset(&p, 0, sizeof(p));
1404    p.state = key;
1405    p.program = program;
1406    p.eye_position = undef;
1407    p.eye_position_normalized = undef;
1408    p.eye_normal = undef;
1409    p.identity = undef;
1410    p.temp_in_use = 0;
1411
1412    if (max_temps >= sizeof(int) * 8)
1413       p.temp_reserved = 0;
1414    else
1415       p.temp_reserved = ~((1<<max_temps)-1);
1416
1417    p.program->Base.Instructions
1418       = (struct prog_instruction*) MALLOC(sizeof(struct prog_instruction) * MAX_INSN);
1419    p.program->Base.String = 0;
1420    p.program->Base.NumInstructions =
1421    p.program->Base.NumTemporaries =
1422    p.program->Base.NumParameters =
1423    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1424    p.program->Base.Parameters = _mesa_new_parameter_list();
1425    p.program->Base.InputsRead = 0;
1426    p.program->Base.OutputsWritten = 0;
1427
1428    build_tnl_program( &p );
1429 }
1430
1431 static void *search_cache( struct tnl_cache *cache,
1432                            GLuint hash,
1433                            const void *key,
1434                            GLuint keysize)
1435 {
1436    struct tnl_cache_item *c;
1437
1438    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1439       if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0)
1440          return c->data;
1441    }
1442
1443    return NULL;
1444 }
1445
1446 static void rehash( struct tnl_cache *cache )
1447 {
1448    struct tnl_cache_item **items;
1449    struct tnl_cache_item *c, *next;
1450    GLuint size, i;
1451
1452    size = cache->size * 3;
1453    items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1454    _mesa_memset(items, 0, size * sizeof(*items));
1455
1456    for (i = 0; i < cache->size; i++)
1457       for (c = cache->items[i]; c; c = next) {
1458          next = c->next;
1459          c->next = items[c->hash % size];
1460          items[c->hash % size] = c;
1461       }
1462
1463    FREE(cache->items);
1464    cache->items = items;
1465    cache->size = size;
1466 }
1467
1468 static void cache_item( struct tnl_cache *cache,
1469                         GLuint hash,
1470                         void *key,
1471                         void *data )
1472 {
1473    struct tnl_cache_item *c = (struct tnl_cache_item*) _mesa_malloc(sizeof(*c));
1474    c->hash = hash;
1475    c->key = key;
1476    c->data = data;
1477
1478    if (++cache->n_items > cache->size * 1.5)
1479       rehash(cache);
1480
1481    c->next = cache->items[hash % cache->size];
1482    cache->items[hash % cache->size] = c;
1483 }
1484
1485 static GLuint hash_key( struct state_key *key )
1486 {
1487    GLuint *ikey = (GLuint *)key;
1488    GLuint hash = 0, i;
1489
1490    /* I'm sure this can be improved on, but speed is important:
1491     */
1492    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1493       hash ^= ikey[i];
1494
1495    return hash;
1496 }
1497
1498 void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
1499 {
1500    TNLcontext *tnl = TNL_CONTEXT(ctx);
1501    struct state_key *key;
1502    GLuint hash;
1503    const struct gl_vertex_program *prev = ctx->VertexProgram._Current;
1504
1505    if (ctx->VertexProgram._Enabled == GL_FALSE) {
1506       /* Grab all the relevent state and put it in a single structure:
1507        */
1508       key = make_state_key(ctx);
1509       hash = hash_key(key);
1510
1511       /* Look for an already-prepared program for this state:
1512        */
1513       ctx->_TnlProgram = (struct gl_vertex_program *)
1514          search_cache( tnl->vp_cache, hash, key, sizeof(*key) );
1515
1516       /* OK, we'll have to build a new one:
1517        */
1518       if (!ctx->_TnlProgram) {
1519          if (0)
1520             _mesa_printf("Build new TNL program\n");
1521
1522          ctx->_TnlProgram = (struct gl_vertex_program *)
1523             ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1524
1525          create_new_program( key, ctx->_TnlProgram,
1526                              ctx->Const.VertexProgram.MaxTemps );
1527
1528          if (ctx->Driver.ProgramStringNotify)
1529             ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1530                                              &ctx->_TnlProgram->Base );
1531
1532          cache_item(tnl->vp_cache, hash, key, ctx->_TnlProgram );
1533       }
1534       else {
1535          FREE(key);
1536          if (0)
1537             _mesa_printf("Found existing TNL program for key %x\n", hash);
1538       }
1539       ctx->VertexProgram._Current = ctx->_TnlProgram;
1540    }
1541    else {
1542       ctx->VertexProgram._Current = ctx->VertexProgram.Current;
1543    }
1544
1545    /* Tell the driver about the change.  Could define a new target for
1546     * this?
1547     */
1548    if (ctx->VertexProgram._Current != prev &&
1549        ctx->Driver.BindProgram)
1550       ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
1551                             (struct gl_program *) ctx->VertexProgram._Current);
1552 }
1553
1554 void _tnl_ProgramCacheInit( GLcontext *ctx )
1555 {
1556    TNLcontext *tnl = TNL_CONTEXT(ctx);
1557
1558    tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache));
1559    tnl->vp_cache->size = 17;
1560    tnl->vp_cache->n_items = 0;
1561    tnl->vp_cache->items = (struct tnl_cache_item**)
1562       _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items));
1563 }
1564
1565 void _tnl_ProgramCacheDestroy( GLcontext *ctx )
1566 {
1567    TNLcontext *tnl = TNL_CONTEXT(ctx);
1568    struct tnl_cache_item *c, *next;
1569    GLuint i;
1570
1571    for (i = 0; i < tnl->vp_cache->size; i++)
1572       for (c = tnl->vp_cache->items[i]; c; c = next) {
1573          next = c->next;
1574          FREE(c->key);
1575          FREE(c->data);
1576          FREE(c);
1577       }
1578
1579    FREE(tnl->vp_cache->items);
1580    FREE(tnl->vp_cache);
1581 }