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 "program.h"
  37 #include "prog_instruction.h"
  38 #include "prog_parameter.h"
  39 #include "prog_print.h"
  40 #include "prog_statevars.h"
  41 #include "t_context.h" /* NOTE: very light dependency on this */
  42 #include "t_vp_build.h"
  43
  44
  45 struct state_key {
  46    unsigned light_global_enabled:1;
  47    unsigned light_local_viewer:1;
  48    unsigned light_twoside:1;
  49    unsigned light_color_material:1;
  50    unsigned light_color_material_mask:12;
  51    unsigned light_material_mask:12;
  52
  53    unsigned normalize:1;
  54    unsigned rescale_normals:1;
  55    unsigned fog_source_is_depth:1;
  56    unsigned tnl_do_vertex_fog:1;
  57    unsigned separate_specular:1;
  58    unsigned fog_mode:2;
  59    unsigned point_attenuated:1;
  60    unsigned texture_enabled_global:1;
  61    unsigned fragprog_inputs_read:12;
  62
  63    struct {
  64       unsigned light_enabled:1;
  65       unsigned light_eyepos3_is_zero:1;
  66       unsigned light_spotcutoff_is_180:1;
  67       unsigned light_attenuated:1;
  68       unsigned texunit_really_enabled:1;
  69       unsigned texmat_enabled:1;
  70       unsigned texgen_enabled:4;
  71       unsigned texgen_mode0:4;
  72       unsigned texgen_mode1:4;
  73       unsigned texgen_mode2:4;
  74       unsigned texgen_mode3:4;
  75    } unit[8];
  76 };
  77
  78
  79
  80 #define FOG_NONE   0
  81 #define FOG_LINEAR 1
  82 #define FOG_EXP    2
  83 #define FOG_EXP2   3
  84
  85 static GLuint translate_fog_mode( GLenum mode )
  86 {
  87    switch (mode) {
  88    case GL_LINEAR: return FOG_LINEAR;
  89    case GL_EXP: return FOG_EXP;
  90    case GL_EXP2: return FOG_EXP2;
  91    default: return FOG_NONE;
  92    }
  93 }
  94
  95 #define TXG_NONE           0
  96 #define TXG_OBJ_LINEAR     1
  97 #define TXG_EYE_LINEAR     2
  98 #define TXG_SPHERE_MAP     3
  99 #define TXG_REFLECTION_MAP 4
 100 #define TXG_NORMAL_MAP     5
 101
 102 static GLuint translate_texgen( GLboolean enabled, GLenum mode )
 103 {
 104    if (!enabled)
 105       return TXG_NONE;
 106
 107    switch (mode) {
 108    case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
 109    case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
 110    case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
 111    case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
 112    case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
 113    default: return TXG_NONE;
 114    }
 115 }
 116
 117 static struct state_key *make_state_key( GLcontext *ctx )
 118 {
 119    TNLcontext *tnl = TNL_CONTEXT(ctx);
 120    struct vertex_buffer *VB = &tnl->vb;
 121    const struct gl_fragment_program *fp = ctx->FragmentProgram._Current;
 122    struct state_key *key = CALLOC_STRUCT(state_key);
 123    GLuint i;
 124
 125    /* This now relies on texenvprogram.c being active:
 126     */
 127    assert(fp);
 128
 129    key->fragprog_inputs_read = fp->Base.InputsRead;
 130
 131    key->separate_specular = (ctx->Light.Model.ColorControl ==
 132                              GL_SEPARATE_SPECULAR_COLOR);
 133
 134    if (ctx->Light.Enabled) {
 135       key->light_global_enabled = 1;
 136
 137       if (ctx->Light.Model.LocalViewer)
 138          key->light_local_viewer = 1;
 139
 140       if (ctx->Light.Model.TwoSide)
 141          key->light_twoside = 1;
 142
 143       if (ctx->Light.ColorMaterialEnabled) {
 144          key->light_color_material = 1;
 145          key->light_color_material_mask = ctx->Light.ColorMaterialBitmask;
 146       }
 147
 148       for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++)
 149          if (VB->AttribPtr[i]->stride)
 150             key->light_material_mask |= 1<<(i-_TNL_ATTRIB_MAT_FRONT_AMBIENT);
 151
 152       for (i = 0; i < MAX_LIGHTS; i++) {
 153          struct gl_light *light = &ctx->Light.Light[i];
 154
 155          if (light->Enabled) {
 156             key->unit[i].light_enabled = 1;
 157
 158             if (light->EyePosition[3] == 0.0)
 159                key->unit[i].light_eyepos3_is_zero = 1;
 160
 161             if (light->SpotCutoff == 180.0)
 162                key->unit[i].light_spotcutoff_is_180 = 1;
 163
 164             if (light->ConstantAttenuation != 1.0 ||
 165                 light->LinearAttenuation != 0.0 ||
 166                 light->QuadraticAttenuation != 0.0)
 167                key->unit[i].light_attenuated = 1;
 168          }
 169       }
 170    }
 171
 172    if (ctx->Transform.Normalize)
 173       key->normalize = 1;
 174
 175    if (ctx->Transform.RescaleNormals)
 176       key->rescale_normals = 1;
 177
 178    key->fog_mode = translate_fog_mode(fp->FogOption);
 179
 180    if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT)
 181       key->fog_source_is_depth = 1;
 182
 183    if (tnl->_DoVertexFog)
 184       key->tnl_do_vertex_fog = 1;
 185
 186    if (ctx->Point._Attenuated)
 187       key->point_attenuated = 1;
 188
 189    if (ctx->Texture._TexGenEnabled ||
 190        ctx->Texture._TexMatEnabled ||
 191        ctx->Texture._EnabledUnits)
 192       key->texture_enabled_global = 1;
 193
 194    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 195       struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
 196
 197       if (texUnit->_ReallyEnabled)
 198          key->unit[i].texunit_really_enabled = 1;
 199
 200       if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
 201          key->unit[i].texmat_enabled = 1;
 202
 203       if (texUnit->TexGenEnabled) {
 204          key->unit[i].texgen_enabled = 1;
 205
 206          key->unit[i].texgen_mode0 =
 207             translate_texgen( texUnit->TexGenEnabled & (1<<0),
 208                               texUnit->GenModeS );
 209          key->unit[i].texgen_mode1 =
 210             translate_texgen( texUnit->TexGenEnabled & (1<<1),
 211                               texUnit->GenModeT );
 212          key->unit[i].texgen_mode2 =
 213             translate_texgen( texUnit->TexGenEnabled & (1<<2),
 214                               texUnit->GenModeR );
 215          key->unit[i].texgen_mode3 =
 216             translate_texgen( texUnit->TexGenEnabled & (1<<3),
 217                               texUnit->GenModeQ );
 218       }
 219    }
 220
 221    return key;
 222 }
 223
 224
 225
 226 /* Very useful debugging tool - produces annotated listing of
 227  * generated program with line/function references for each
 228  * instruction back into this file:
 229  */
 230 #define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM)
 231
 232 /* Should be tunable by the driver - do we want to do matrix
 233  * multiplications with DP4's or with MUL/MAD's?  SSE works better
 234  * with the latter, drivers may differ.
 235  */
 236 #define PREFER_DP4 0
 237
 238 #define MAX_INSN 256
 239
 240 /* Use uregs to represent registers internally, translate to Mesa's
 241  * expected formats on emit.
 242  *
 243  * NOTE: These are passed by value extensively in this file rather
 244  * than as usual by pointer reference.  If this disturbs you, try
 245  * remembering they are just 32bits in size.
 246  *
 247  * GCC is smart enough to deal with these dword-sized structures in
 248  * much the same way as if I had defined them as dwords and was using
 249  * macros to access and set the fields.  This is much nicer and easier
 250  * to evolve.
 251  */
 252 struct ureg {
 253    GLuint file:4;
 254    GLint idx:8;      /* relative addressing may be negative */
 255    GLuint negate:1;
 256    GLuint swz:12;
 257    GLuint pad:7;
 258 };
 259
 260
 261 struct tnl_program {
 262    const struct state_key *state;
 263    struct gl_vertex_program *program;
 264
 265    GLuint temp_in_use;
 266    GLuint temp_reserved;
 267
 268    struct ureg eye_position;
 269    struct ureg eye_position_normalized;
 270    struct ureg eye_normal;
 271    struct ureg identity;
 272
 273    GLuint materials;
 274    GLuint color_materials;
 275 };
 276
 277
 278 static const struct ureg undef = {
 279    PROGRAM_UNDEFINED,
 280    ~0,
 281    0,
 282    0,
 283    0
 284 };
 285
 286 /* Local shorthand:
 287  */
 288 #define X    SWIZZLE_X
 289 #define Y    SWIZZLE_Y
 290 #define Z    SWIZZLE_Z
 291 #define W    SWIZZLE_W
 292
 293
 294 /* Construct a ureg:
 295  */
 296 static struct ureg make_ureg(GLuint file, GLint idx)
 297 {
 298    struct ureg reg;
 299    reg.file = file;
 300    reg.idx = idx;
 301    reg.negate = 0;
 302    reg.swz = SWIZZLE_NOOP;
 303    reg.pad = 0;
 304    return reg;
 305 }
 306
 307
 308
 309 static struct ureg negate( struct ureg reg )
 310 {
 311    reg.negate ^= 1;
 312    return reg;
 313 }
 314
 315
 316 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 317 {
 318    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 319                            GET_SWZ(reg.swz, y),
 320                            GET_SWZ(reg.swz, z),
 321                            GET_SWZ(reg.swz, w));
 322
 323    return reg;
 324 }
 325
 326 static struct ureg swizzle1( struct ureg reg, int x )
 327 {
 328    return swizzle(reg, x, x, x, x);
 329 }
 330
 331 static struct ureg get_temp( struct tnl_program *p )
 332 {
 333    int bit = _mesa_ffs( ~p->temp_in_use );
 334    if (!bit) {
 335       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
 336       _mesa_exit(1);
 337    }
 338
 339    if ((GLuint) bit > p->program->Base.NumTemporaries)
 340       p->program->Base.NumTemporaries = bit;
 341
 342    p->temp_in_use |= 1<<(bit-1);
 343    return make_ureg(PROGRAM_TEMPORARY, bit-1);
 344 }
 345
 346 static struct ureg reserve_temp( struct tnl_program *p )
 347 {
 348    struct ureg temp = get_temp( p );
 349    p->temp_reserved |= 1<<temp.idx;
 350    return temp;
 351 }
 352
 353 static void release_temp( struct tnl_program *p, struct ureg reg )
 354 {
 355    if (reg.file == PROGRAM_TEMPORARY) {
 356       p->temp_in_use &= ~(1<<reg.idx);
 357       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
 358    }
 359 }
 360
 361 static void release_temps( struct tnl_program *p )
 362 {
 363    p->temp_in_use = p->temp_reserved;
 364 }
 365
 366
 367
 368 static struct ureg register_input( struct tnl_program *p, GLuint input )
 369 {
 370    p->program->Base.InputsRead |= (1<<input);
 371    return make_ureg(PROGRAM_INPUT, input);
 372 }
 373
 374 static struct ureg register_output( struct tnl_program *p, GLuint output )
 375 {
 376    p->program->Base.OutputsWritten |= (1<<output);
 377    return make_ureg(PROGRAM_OUTPUT, output);
 378 }
 379
 380 static struct ureg register_const4f( struct tnl_program *p,
 381                               GLfloat s0,
 382                               GLfloat s1,
 383                               GLfloat s2,
 384                               GLfloat s3)
 385 {
 386    GLfloat values[4];
 387    GLint idx;
 388    GLuint swizzle;
 389    values[0] = s0;
 390    values[1] = s1;
 391    values[2] = s2;
 392    values[3] = s3;
 393    idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
 394                                      &swizzle );
 395    ASSERT(swizzle == SWIZZLE_NOOP);
 396    return make_ureg(PROGRAM_STATE_VAR, idx);
 397 }
 398
 399 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
 400 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
 401 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
 402 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
 403
 404 static GLboolean is_undef( struct ureg reg )
 405 {
 406    return reg.file == PROGRAM_UNDEFINED;
 407 }
 408
 409 static struct ureg get_identity_param( struct tnl_program *p )
 410 {
 411    if (is_undef(p->identity))
 412       p->identity = register_const4f(p, 0,0,0,1);
 413
 414    return p->identity;
 415 }
 416
 417 static struct ureg register_param5(struct tnl_program *p,
 418                                    GLint s0,
 419                                    GLint s1,
 420                                    GLint s2,
 421                                    GLint s3,
 422                                    GLint s4)
 423 {
 424    gl_state_index tokens[STATE_LENGTH];
 425    GLint idx;
 426    tokens[0] = s0;
 427    tokens[1] = s1;
 428    tokens[2] = s2;
 429    tokens[3] = s3;
 430    tokens[4] = s4;
 431    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 432    return make_ureg(PROGRAM_STATE_VAR, idx);
 433 }
 434
 435
 436 #define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
 437 #define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
 438 #define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
 439 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
 440
 441
 442 static void register_matrix_param5( struct tnl_program *p,
 443                                     GLint s0, /* modelview, projection, etc */
 444                                     GLint s1, /* texture matrix number */
 445                                     GLint s2, /* first row */
 446                                     GLint s3, /* last row */
 447                                     GLint s4, /* inverse, transpose, etc */
 448                                     struct ureg *matrix )
 449 {
 450    GLint i;
 451
 452    /* This is a bit sad as the support is there to pull the whole
 453     * matrix out in one go:
 454     */
 455    for (i = 0; i <= s3 - s2; i++)
 456       matrix[i] = register_param5( p, s0, s1, i, i, s4 );
 457 }
 458
 459
 460 static void emit_arg( struct prog_src_register *src,
 461                       struct ureg reg )
 462 {
 463    src->File = reg.file;
 464    src->Index = reg.idx;
 465    src->Swizzle = reg.swz;
 466    src->NegateBase = reg.negate ? NEGATE_XYZW : 0;
 467    src->Abs = 0;
 468    src->NegateAbs = 0;
 469    src->RelAddr = 0;
 470 }
 471
 472 static void emit_dst( struct prog_dst_register *dst,
 473                       struct ureg reg, GLuint mask )
 474 {
 475    dst->File = reg.file;
 476    dst->Index = reg.idx;
 477    /* allow zero as a shorthand for xyzw */
 478    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
 479    dst->CondMask = COND_TR;
 480    dst->CondSwizzle = 0;
 481    dst->CondSrc = 0;
 482    dst->pad = 0;
 483 }
 484
 485 static void debug_insn( struct prog_instruction *inst, const char *fn,
 486                         GLuint line )
 487 {
 488    if (DISASSEM) {
 489       static const char *last_fn;
 490
 491       if (fn != last_fn) {
 492          last_fn = fn;
 493          _mesa_printf("%s:\n", fn);
 494       }
 495
 496       _mesa_printf("%d:\t", line);
 497       _mesa_print_instruction(inst);
 498    }
 499 }
 500
 501
 502 static void emit_op3fn(struct tnl_program *p,
 503                        GLuint op,
 504                        struct ureg dest,
 505                        GLuint mask,
 506                        struct ureg src0,
 507                        struct ureg src1,
 508                        struct ureg src2,
 509                        const char *fn,
 510                        GLuint line)
 511 {
 512    GLuint nr = p->program->Base.NumInstructions++;
 513    struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 514
 515    if (p->program->Base.NumInstructions > MAX_INSN) {
 516       _mesa_problem(0, "Out of instructions in emit_op3fn\n");
 517       return;
 518    }
 519
 520    inst->Opcode = (enum prog_opcode) op;
 521    inst->StringPos = 0;
 522    inst->Data = 0;
 523
 524    emit_arg( &inst->SrcReg[0], src0 );
 525    emit_arg( &inst->SrcReg[1], src1 );
 526    emit_arg( &inst->SrcReg[2], src2 );
 527
 528    emit_dst( &inst->DstReg, dest, mask );
 529
 530    debug_insn(inst, fn, line);
 531 }
 532
 533
 534 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 535    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 536
 537 #define emit_op2(p, op, dst, mask, src0, src1) \
 538     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 539
 540 #define emit_op1(p, op, dst, mask, src0) \
 541     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 542
 543
 544 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 545 {
 546    if (reg.file == PROGRAM_TEMPORARY &&
 547        !(p->temp_reserved & (1<<reg.idx)))
 548       return reg;
 549    else {
 550       struct ureg temp = get_temp(p);
 551       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 552       return temp;
 553    }
 554 }
 555
 556
 557 /* Currently no tracking performed of input/output/register size or
 558  * active elements.  Could be used to reduce these operations, as
 559  * could the matrix type.
 560  */
 561 static void emit_matrix_transform_vec4( struct tnl_program *p,
 562                                         struct ureg dest,
 563                                         const struct ureg *mat,
 564                                         struct ureg src)
 565 {
 566    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 567    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 568    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 569    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 570 }
 571
 572 /* This version is much easier to implement if writemasks are not
 573  * supported natively on the target or (like SSE), the target doesn't
 574  * have a clean/obvious dotproduct implementation.
 575  */
 576 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 577                                                   struct ureg dest,
 578                                                   const struct ureg *mat,
 579                                                   struct ureg src)
 580 {
 581    struct ureg tmp;
 582
 583    if (dest.file != PROGRAM_TEMPORARY)
 584       tmp = get_temp(p);
 585    else
 586       tmp = dest;
 587
 588    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 589    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 590    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 591    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 592
 593    if (dest.file != PROGRAM_TEMPORARY)
 594       release_temp(p, tmp);
 595 }
 596
 597 static void emit_matrix_transform_vec3( struct tnl_program *p,
 598                                         struct ureg dest,
 599                                         const struct ureg *mat,
 600                                         struct ureg src)
 601 {
 602    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 603    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 604    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 605 }
 606
 607
 608 static void emit_normalize_vec3( struct tnl_program *p,
 609                                  struct ureg dest,
 610                                  struct ureg src )
 611 {
 612    struct ureg tmp = get_temp(p);
 613    emit_op2(p, OPCODE_DP3, tmp, 0, src, src);
 614    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
 615    emit_op2(p, OPCODE_MUL, dest, 0, src, tmp);
 616    release_temp(p, tmp);
 617 }
 618
 619 static void emit_passthrough( struct tnl_program *p,
 620                               GLuint input,
 621                               GLuint output )
 622 {
 623    struct ureg out = register_output(p, output);
 624    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 625 }
 626
 627 static struct ureg get_eye_position( struct tnl_program *p )
 628 {
 629    if (is_undef(p->eye_position)) {
 630       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 631       struct ureg modelview[4];
 632
 633       p->eye_position = reserve_temp(p);
 634
 635       if (PREFER_DP4) {
 636          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 637                                  0, modelview );
 638
 639          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 640       }
 641       else {
 642          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 643                                  STATE_MATRIX_TRANSPOSE, modelview );
 644
 645          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 646       }
 647    }
 648
 649    return p->eye_position;
 650 }
 651
 652
 653 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 654 {
 655    if (is_undef(p->eye_position_normalized)) {
 656       struct ureg eye = get_eye_position(p);
 657       p->eye_position_normalized = reserve_temp(p);
 658       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 659    }
 660
 661    return p->eye_position_normalized;
 662 }
 663
 664
 665 static struct ureg get_eye_normal( struct tnl_program *p )
 666 {
 667    if (is_undef(p->eye_normal)) {
 668       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
 669       struct ureg mvinv[3];
 670
 671       register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
 672                               STATE_MATRIX_INVTRANS, mvinv );
 673
 674       p->eye_normal = reserve_temp(p);
 675
 676       /* Transform to eye space:
 677        */
 678       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 679
 680       /* Normalize/Rescale:
 681        */
 682       if (p->state->normalize) {
 683          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 684       }
 685       else if (p->state->rescale_normals) {
 686          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 687                                                STATE_NORMAL_SCALE);
 688
 689          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, normal,
 690                    swizzle1(rescale, X));
 691       }
 692    }
 693
 694    return p->eye_normal;
 695 }
 696
 697
 698
 699 static void build_hpos( struct tnl_program *p )
 700 {
 701    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 702    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 703    struct ureg mvp[4];
 704
 705    if (PREFER_DP4) {
 706       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 707                               0, mvp );
 708       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 709    }
 710    else {
 711       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 712                               STATE_MATRIX_TRANSPOSE, mvp );
 713       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 714    }
 715 }
 716
 717
 718 static GLuint material_attrib( GLuint side, GLuint property )
 719 {
 720    return ((property - STATE_AMBIENT) * 2 +
 721            side);
 722 }
 723
 724 /* Get a bitmask of which material values vary on a per-vertex basis.
 725  */
 726 static void set_material_flags( struct tnl_program *p )
 727 {
 728    p->color_materials = 0;
 729    p->materials = 0;
 730
 731    if (p->state->light_color_material) {
 732       p->materials =
 733          p->color_materials = p->state->light_color_material_mask;
 734    }
 735
 736    p->materials |= p->state->light_material_mask;
 737 }
 738
 739
 740 static struct ureg get_material( struct tnl_program *p, GLuint side,
 741                                  GLuint property )
 742 {
 743    GLuint attrib = material_attrib(side, property);
 744
 745    if (p->color_materials & (1<<attrib))
 746       return register_input(p, VERT_ATTRIB_COLOR0);
 747    else if (p->materials & (1<<attrib))
 748       return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
 749    else
 750       return register_param3( p, STATE_MATERIAL, side, property );
 751 }
 752
 753 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
 754                                    MAT_BIT_FRONT_AMBIENT | \
 755                                    MAT_BIT_FRONT_DIFFUSE) << (side))
 756
 757 /* Either return a precalculated constant value or emit code to
 758  * calculate these values dynamically in the case where material calls
 759  * are present between begin/end pairs.
 760  *
 761  * Probably want to shift this to the program compilation phase - if
 762  * we always emitted the calculation here, a smart compiler could
 763  * detect that it was constant (given a certain set of inputs), and
 764  * lift it out of the main loop.  That way the programs created here
 765  * would be independent of the vertex_buffer details.
 766  */
 767 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 768 {
 769    if (p->materials & SCENE_COLOR_BITS(side)) {
 770       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 771       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 772       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 773       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 774       struct ureg tmp = make_temp(p, material_diffuse);
 775       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 776                material_ambient, material_emission);
 777       return tmp;
 778    }
 779    else
 780       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 781 }
 782
 783
 784 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 785                                   GLuint side, GLuint property )
 786 {
 787    GLuint attrib = material_attrib(side, property);
 788    if (p->materials & (1<<attrib)) {
 789       struct ureg light_value =
 790          register_param3(p, STATE_LIGHT, light, property);
 791       struct ureg material_value = get_material(p, side, property);
 792       struct ureg tmp = get_temp(p);
 793       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 794       return tmp;
 795    }
 796    else
 797       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 798 }
 799
 800 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 801                                                 GLuint i,
 802                                                 struct ureg VPpli,
 803                                                 struct ureg dist )
 804 {
 805    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 806                                              STATE_ATTENUATION);
 807    struct ureg att = get_temp(p);
 808
 809    /* Calculate spot attenuation:
 810     */
 811    if (!p->state->unit[i].light_spotcutoff_is_180) {
 812       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
 813                                                   STATE_SPOT_DIR_NORMALIZED, i);
 814       struct ureg spot = get_temp(p);
 815       struct ureg slt = get_temp(p);
 816
 817       emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
 818       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
 819       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 820       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 821
 822       release_temp(p, spot);
 823       release_temp(p, slt);
 824    }
 825
 826    /* Calculate distance attenuation:
 827     */
 828    if (p->state->unit[i].light_attenuated) {
 829
 830       /* 1/d,d,d,1/d */
 831       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 832       /* 1,d,d*d,1/d */
 833       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 834       /* 1/dist-atten */
 835       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 836
 837       if (!p->state->unit[i].light_spotcutoff_is_180) {
 838          /* dist-atten */
 839          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 840          /* spot-atten * dist-atten */
 841          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 842       } else {
 843          /* dist-atten */
 844          emit_op1(p, OPCODE_RCP, att, 0, dist);
 845       }
 846    }
 847
 848    return att;
 849 }
 850
 851
 852
 853
 854
 855 /* Need to add some addtional parameters to allow lighting in object
 856  * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
 857  * space lighting.
 858  */
 859 static void build_lighting( struct tnl_program *p )
 860 {
 861    const GLboolean twoside = p->state->light_twoside;
 862    const GLboolean separate = p->state->separate_specular;
 863    GLuint nr_lights = 0, count = 0;
 864    struct ureg normal = get_eye_normal(p);
 865    struct ureg lit = get_temp(p);
 866    struct ureg dots = get_temp(p);
 867    struct ureg _col0 = undef, _col1 = undef;
 868    struct ureg _bfc0 = undef, _bfc1 = undef;
 869    GLuint i;
 870
 871    for (i = 0; i < MAX_LIGHTS; i++)
 872       if (p->state->unit[i].light_enabled)
 873          nr_lights++;
 874
 875    set_material_flags(p);
 876
 877    {
 878       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 879       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 880       release_temp(p, shininess);
 881
 882       _col0 = make_temp(p, get_scenecolor(p, 0));
 883       if (separate)
 884          _col1 = make_temp(p, get_identity_param(p));
 885       else
 886          _col1 = _col0;
 887
 888    }
 889
 890    if (twoside) {
 891       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 892       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 893                negate(swizzle1(shininess,X)));
 894       release_temp(p, shininess);
 895
 896       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 897       if (separate)
 898          _bfc1 = make_temp(p, get_identity_param(p));
 899       else
 900          _bfc1 = _bfc0;
 901    }
 902
 903
 904    /* If no lights, still need to emit the scenecolor.
 905     */
 906       {
 907          struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 908          emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 909       }
 910
 911       if (separate) {
 912          struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 913          emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 914       }
 915
 916       if (twoside) {
 917          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 918          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 919       }
 920
 921       if (twoside && separate) {
 922          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 923          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 924       }
 925
 926    if (nr_lights == 0) {
 927       release_temps(p);
 928       return;
 929    }
 930
 931
 932    for (i = 0; i < MAX_LIGHTS; i++) {
 933       if (p->state->unit[i].light_enabled) {
 934          struct ureg half = undef;
 935          struct ureg att = undef, VPpli = undef;
 936
 937          count++;
 938
 939          if (p->state->unit[i].light_eyepos3_is_zero) {
 940             /* Can used precomputed constants in this case.
 941              * Attenuation never applies to infinite lights.
 942              */
 943             VPpli = register_param3(p, STATE_LIGHT, i,
 944                                     STATE_POSITION_NORMALIZED);
 945             if (p->state->light_local_viewer) {
 946                 struct ureg eye_hat = get_eye_position_normalized(p);
 947                 half = get_temp(p);
 948                 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 949                 emit_normalize_vec3(p, half, half);
 950             } else {
 951                 half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
 952             }
 953          }
 954          else {
 955             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
 956                                                STATE_POSITION);
 957             struct ureg V = get_eye_position(p);
 958             struct ureg dist = get_temp(p);
 959
 960             VPpli = get_temp(p);
 961             half = get_temp(p);
 962
 963             /* Calulate VPpli vector
 964              */
 965             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
 966
 967             /* Normalize VPpli.  The dist value also used in
 968              * attenuation below.
 969              */
 970             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
 971             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
 972             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
 973
 974
 975             /* Calculate  attenuation:
 976              */
 977             if (!p->state->unit[i].light_spotcutoff_is_180 ||
 978                 p->state->unit[i].light_attenuated) {
 979                att = calculate_light_attenuation(p, i, VPpli, dist);
 980             }
 981
 982
 983             /* Calculate viewer direction, or use infinite viewer:
 984              */
 985             if (p->state->light_local_viewer) {
 986                struct ureg eye_hat = get_eye_position_normalized(p);
 987                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 988             }
 989             else {
 990                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
 991                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
 992             }
 993
 994             emit_normalize_vec3(p, half, half);
 995
 996             release_temp(p, dist);
 997          }
 998
 999          /* Calculate dot products:
1000           */
1001          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1002          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1003
1004
1005          /* Front face lighting:
1006           */
1007          {
1008             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1009             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1010             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1011             struct ureg res0, res1;
1012             GLuint mask0, mask1;
1013
1014             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1015
1016             if (!is_undef(att))
1017                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1018
1019
1020             if (count == nr_lights) {
1021                if (separate) {
1022                   mask0 = WRITEMASK_XYZ;
1023                   mask1 = WRITEMASK_XYZ;
1024                   res0 = register_output( p, VERT_RESULT_COL0 );
1025                   res1 = register_output( p, VERT_RESULT_COL1 );
1026                }
1027                else {
1028                   mask0 = 0;
1029                   mask1 = WRITEMASK_XYZ;
1030                   res0 = _col0;
1031                   res1 = register_output( p, VERT_RESULT_COL0 );
1032                }
1033             } else {
1034                mask0 = 0;
1035                mask1 = 0;
1036                res0 = _col0;
1037                res1 = _col1;
1038             }
1039
1040             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1041             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1042             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1043
1044             release_temp(p, ambient);
1045             release_temp(p, diffuse);
1046             release_temp(p, specular);
1047          }
1048
1049          /* Back face lighting:
1050           */
1051          if (twoside) {
1052             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1053             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1054             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1055             struct ureg res0, res1;
1056             GLuint mask0, mask1;
1057
1058             emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
1059
1060             if (!is_undef(att))
1061                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1062
1063             if (count == nr_lights) {
1064                if (separate) {
1065                   mask0 = WRITEMASK_XYZ;
1066                   mask1 = WRITEMASK_XYZ;
1067                   res0 = register_output( p, VERT_RESULT_BFC0 );
1068                   res1 = register_output( p, VERT_RESULT_BFC1 );
1069                }
1070                else {
1071                   mask0 = 0;
1072                   mask1 = WRITEMASK_XYZ;
1073                   res0 = _bfc0;
1074                   res1 = register_output( p, VERT_RESULT_BFC0 );
1075                }
1076             } else {
1077                res0 = _bfc0;
1078                res1 = _bfc1;
1079                mask0 = 0;
1080                mask1 = 0;
1081             }
1082
1083             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1084             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1085             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1086
1087             release_temp(p, ambient);
1088             release_temp(p, diffuse);
1089             release_temp(p, specular);
1090          }
1091
1092          release_temp(p, half);
1093          release_temp(p, VPpli);
1094          release_temp(p, att);
1095       }
1096    }
1097
1098    release_temps( p );
1099 }
1100
1101
1102 static void build_fog( struct tnl_program *p )
1103 {
1104    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1105    struct ureg input;
1106
1107    if (p->state->fog_source_is_depth) {
1108       input = swizzle1(get_eye_position(p), Z);
1109    }
1110    else {
1111       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1112    }
1113
1114    if (p->state->tnl_do_vertex_fog) {
1115       struct ureg params = register_param2(p, STATE_INTERNAL,
1116                                            STATE_FOG_PARAMS_OPTIMIZED);
1117       struct ureg tmp = get_temp(p);
1118
1119       switch (p->state->fog_mode) {
1120       case FOG_LINEAR: {
1121          struct ureg id = get_identity_param(p);
1122          emit_op3(p, OPCODE_MAD, tmp, 0, input, swizzle1(params,X), swizzle1(params,Y));
1123          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1124          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1125          break;
1126       }
1127       case FOG_EXP:
1128          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1129          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,Z));
1130          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1131          break;
1132       case FOG_EXP2:
1133          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
1134          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1135          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1136          break;
1137       }
1138
1139       release_temp(p, tmp);
1140    }
1141    else {
1142       /* results = incoming fog coords (compute fog per-fragment later)
1143        *
1144        * KW:  Is it really necessary to do anything in this case?
1145        */
1146       emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1147    }
1148 }
1149
1150 static void build_reflect_texgen( struct tnl_program *p,
1151                                   struct ureg dest,
1152                                   GLuint writemask )
1153 {
1154    struct ureg normal = get_eye_normal(p);
1155    struct ureg eye_hat = get_eye_position_normalized(p);
1156    struct ureg tmp = get_temp(p);
1157
1158    /* n.u */
1159    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1160    /* 2n.u */
1161    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1162    /* (-2n.u)n + u */
1163    emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1164
1165    release_temp(p, tmp);
1166 }
1167
1168 static void build_sphere_texgen( struct tnl_program *p,
1169                                  struct ureg dest,
1170                                  GLuint writemask )
1171 {
1172    struct ureg normal = get_eye_normal(p);
1173    struct ureg eye_hat = get_eye_position_normalized(p);
1174    struct ureg tmp = get_temp(p);
1175    struct ureg half = register_scalar_const(p, .5);
1176    struct ureg r = get_temp(p);
1177    struct ureg inv_m = get_temp(p);
1178    struct ureg id = get_identity_param(p);
1179
1180    /* Could share the above calculations, but it would be
1181     * a fairly odd state for someone to set (both sphere and
1182     * reflection active for different texture coordinate
1183     * components.  Of course - if two texture units enable
1184     * reflect and/or sphere, things start to tilt in favour
1185     * of seperating this out:
1186     */
1187
1188    /* n.u */
1189    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1190    /* 2n.u */
1191    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1192    /* (-2n.u)n + u */
1193    emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1194    /* r + 0,0,1 */
1195    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1196    /* rx^2 + ry^2 + (rz+1)^2 */
1197    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1198    /* 2/m */
1199    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1200    /* 1/m */
1201    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1202    /* r/m + 1/2 */
1203    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1204
1205    release_temp(p, tmp);
1206    release_temp(p, r);
1207    release_temp(p, inv_m);
1208 }
1209
1210
1211 static void build_texture_transform( struct tnl_program *p )
1212 {
1213    GLuint i, j;
1214
1215    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1216
1217       if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
1218          continue;
1219
1220       if (p->state->unit[i].texgen_enabled ||
1221           p->state->unit[i].texmat_enabled) {
1222
1223          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1224          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1225          struct ureg out_texgen = undef;
1226
1227          if (p->state->unit[i].texgen_enabled) {
1228             GLuint copy_mask = 0;
1229             GLuint sphere_mask = 0;
1230             GLuint reflect_mask = 0;
1231             GLuint normal_mask = 0;
1232             GLuint modes[4];
1233
1234             if (texmat_enabled)
1235                out_texgen = get_temp(p);
1236             else
1237                out_texgen = out;
1238
1239             modes[0] = p->state->unit[i].texgen_mode0;
1240             modes[1] = p->state->unit[i].texgen_mode1;
1241             modes[2] = p->state->unit[i].texgen_mode2;
1242             modes[3] = p->state->unit[i].texgen_mode3;
1243
1244             for (j = 0; j < 4; j++) {
1245                switch (modes[j]) {
1246                case TXG_OBJ_LINEAR: {
1247                   struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1248                   struct ureg plane =
1249                      register_param3(p, STATE_TEXGEN, i,
1250                                      STATE_TEXGEN_OBJECT_S + j);
1251
1252                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1253                            obj, plane );
1254                   break;
1255                }
1256                case TXG_EYE_LINEAR: {
1257                   struct ureg eye = get_eye_position(p);
1258                   struct ureg plane =
1259                      register_param3(p, STATE_TEXGEN, i,
1260                                      STATE_TEXGEN_EYE_S + j);
1261
1262                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1263                            eye, plane );
1264                   break;
1265                }
1266                case TXG_SPHERE_MAP:
1267                   sphere_mask |= WRITEMASK_X << j;
1268                   break;
1269                case TXG_REFLECTION_MAP:
1270                   reflect_mask |= WRITEMASK_X << j;
1271                   break;
1272                case TXG_NORMAL_MAP:
1273                   normal_mask |= WRITEMASK_X << j;
1274                   break;
1275                case TXG_NONE:
1276                   copy_mask |= WRITEMASK_X << j;
1277                }
1278
1279             }
1280
1281
1282             if (sphere_mask) {
1283                build_sphere_texgen(p, out_texgen, sphere_mask);
1284             }
1285
1286             if (reflect_mask) {
1287                build_reflect_texgen(p, out_texgen, reflect_mask);
1288             }
1289
1290             if (normal_mask) {
1291                struct ureg normal = get_eye_normal(p);
1292                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1293             }
1294
1295             if (copy_mask) {
1296                struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1297                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1298             }
1299          }
1300
1301          if (texmat_enabled) {
1302             struct ureg texmat[4];
1303             struct ureg in = (!is_undef(out_texgen) ?
1304                               out_texgen :
1305                               register_input(p, VERT_ATTRIB_TEX0+i));
1306             if (PREFER_DP4) {
1307                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1308                                        0, texmat );
1309                emit_matrix_transform_vec4( p, out, texmat, in );
1310             }
1311             else {
1312                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1313                                        STATE_MATRIX_TRANSPOSE, texmat );
1314                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1315             }
1316          }
1317
1318          release_temps(p);
1319       }
1320       else {
1321          emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1322       }
1323    }
1324 }
1325
1326
1327 static void build_pointsize( struct tnl_program *p )
1328 {
1329    struct ureg eye = get_eye_position(p);
1330    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1331    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1332    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1333    struct ureg ut = get_temp(p);
1334
1335    /* dist = |eyez| */
1336    emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
1337    /* p1 + dist * (p2 + dist * p3); */
1338    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1339                 swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1340    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1341                 ut, swizzle1(state_attenuation, X));
1342
1343    /* 1 / sqrt(factor) */
1344    emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1345
1346 #if 1
1347    /* out = pointSize / sqrt(factor) */
1348    emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1349 #else
1350    /* not sure, might make sense to do clamping here,
1351       but it's not done in t_vb_points neither */
1352    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1353    emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1354    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1355 #endif
1356
1357    release_temp(p, ut);
1358 }
1359
1360 static void build_tnl_program( struct tnl_program *p )
1361 {   /* Emit the program, starting with modelviewproject:
1362     */
1363    build_hpos(p);
1364
1365    /* Lighting calculations:
1366     */
1367    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1368       if (p->state->light_global_enabled)
1369          build_lighting(p);
1370       else {
1371          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1372             emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);
1373
1374          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1375             emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
1376       }
1377    }
1378
1379    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1380        p->state->fog_mode != FOG_NONE)
1381       build_fog(p);
1382
1383    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1384       build_texture_transform(p);
1385
1386    if (p->state->point_attenuated)
1387       build_pointsize(p);
1388
1389    /* Finish up:
1390     */
1391    emit_op1(p, OPCODE_END, undef, 0, undef);
1392
1393    /* Disassemble:
1394     */
1395    if (DISASSEM) {
1396       _mesa_printf ("\n");
1397    }
1398 }
1399
1400
1401 static void
1402 create_new_program( const struct state_key *key,
1403                     struct gl_vertex_program *program,
1404                     GLuint max_temps)
1405 {
1406    struct tnl_program p;
1407
1408    _mesa_memset(&p, 0, sizeof(p));
1409    p.state = key;
1410    p.program = program;
1411    p.eye_position = undef;
1412    p.eye_position_normalized = undef;
1413    p.eye_normal = undef;
1414    p.identity = undef;
1415    p.temp_in_use = 0;
1416
1417    if (max_temps >= sizeof(int) * 8)
1418       p.temp_reserved = 0;
1419    else
1420       p.temp_reserved = ~((1<<max_temps)-1);
1421
1422    p.program->Base.Instructions = _mesa_alloc_instructions(MAX_INSN);
1423    p.program->Base.String = NULL;
1424    p.program->Base.NumInstructions =
1425    p.program->Base.NumTemporaries =
1426    p.program->Base.NumParameters =
1427    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1428    p.program->Base.Parameters = _mesa_new_parameter_list();
1429    p.program->Base.InputsRead = 0;
1430    p.program->Base.OutputsWritten = 0;
1431
1432    build_tnl_program( &p );
1433 }
1434
1435 static void *search_cache( struct tnl_cache *cache,
1436                            GLuint hash,
1437                            const void *key,
1438                            GLuint keysize)
1439 {
1440    struct tnl_cache_item *c;
1441
1442    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1443       if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0)
1444          return c->data;
1445    }
1446
1447    return NULL;
1448 }
1449
1450 static void rehash( struct tnl_cache *cache )
1451 {
1452    struct tnl_cache_item **items;
1453    struct tnl_cache_item *c, *next;
1454    GLuint size, i;
1455
1456    size = cache->size * 3;
1457    items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1458    _mesa_memset(items, 0, size * sizeof(*items));
1459
1460    for (i = 0; i < cache->size; i++)
1461       for (c = cache->items[i]; c; c = next) {
1462          next = c->next;
1463          c->next = items[c->hash % size];
1464          items[c->hash % size] = c;
1465       }
1466
1467    FREE(cache->items);
1468    cache->items = items;
1469    cache->size = size;
1470 }
1471
1472 static void cache_item( struct tnl_cache *cache,
1473                         GLuint hash,
1474                         void *key,
1475                         void *data )
1476 {
1477    struct tnl_cache_item *c = (struct tnl_cache_item*) _mesa_malloc(sizeof(*c));
1478    c->hash = hash;
1479    c->key = key;
1480    c->data = data;
1481
1482    if (++cache->n_items > cache->size * 1.5)
1483       rehash(cache);
1484
1485    c->next = cache->items[hash % cache->size];
1486    cache->items[hash % cache->size] = c;
1487 }
1488
1489 static GLuint hash_key( struct state_key *key )
1490 {
1491    GLuint *ikey = (GLuint *)key;
1492    GLuint hash = 0, i;
1493
1494    /* I'm sure this can be improved on, but speed is important:
1495     */
1496    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1497       hash ^= ikey[i];
1498
1499    return hash;
1500 }
1501
1502 void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
1503 {
1504    TNLcontext *tnl = TNL_CONTEXT(ctx);
1505    struct state_key *key;
1506    GLuint hash;
1507    const struct gl_vertex_program *prev = ctx->VertexProgram._Current;
1508
1509    if (!ctx->VertexProgram._Current) {
1510       /* Grab all the relevent state and put it in a single structure:
1511        */
1512       key = make_state_key(ctx);
1513       hash = hash_key(key);
1514
1515       /* Look for an already-prepared program for this state:
1516        */
1517       ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1518          search_cache( tnl->vp_cache, hash, key, sizeof(*key) );
1519
1520       /* OK, we'll have to build a new one:
1521        */
1522       if (!ctx->VertexProgram._TnlProgram) {
1523          if (0)
1524             _mesa_printf("Build new TNL program\n");
1525
1526          ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1527             ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1528
1529          create_new_program( key, ctx->VertexProgram._TnlProgram,
1530                              ctx->Const.VertexProgram.MaxTemps );
1531
1532          if (ctx->Driver.ProgramStringNotify)
1533             ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1534                                        &ctx->VertexProgram._TnlProgram->Base );
1535
1536          cache_item(tnl->vp_cache, hash, key, ctx->VertexProgram._TnlProgram );
1537       }
1538       else {
1539          FREE(key);
1540          if (0)
1541             _mesa_printf("Found existing TNL program for key %x\n", hash);
1542       }
1543       ctx->VertexProgram._Current = ctx->VertexProgram._TnlProgram;
1544    }
1545
1546    /* Tell the driver about the change.  Could define a new target for
1547     * this?
1548     */
1549    if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) {
1550       ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
1551                             (struct gl_program *) ctx->VertexProgram._Current);
1552    }
1553 }
1554
1555 void _tnl_ProgramCacheInit( GLcontext *ctx )
1556 {
1557    TNLcontext *tnl = TNL_CONTEXT(ctx);
1558
1559    tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache));
1560    tnl->vp_cache->size = 17;
1561    tnl->vp_cache->n_items = 0;
1562    tnl->vp_cache->items = (struct tnl_cache_item**)
1563       _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items));
1564 }
1565
1566 void _tnl_ProgramCacheDestroy( GLcontext *ctx )
1567 {
1568    TNLcontext *tnl = TNL_CONTEXT(ctx);
1569    struct tnl_cache_item *c, *next;
1570    GLuint i;
1571
1572    for (i = 0; i < tnl->vp_cache->size; i++)
1573       for (c = tnl->vp_cache->items[i]; c; c = next) {
1574          next = c->next;
1575          FREE(c->key);
1576          FREE(c->data);
1577          FREE(c);
1578       }
1579
1580    FREE(tnl->vp_cache->items);
1581    FREE(tnl->vp_cache);
1582 }