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_param6( struct tnl_program *p,
 418                                    GLint s0,
 419                                    GLint s1,
 420                                    GLint s2,
 421                                    GLint s3,
 422                                    GLint s4,
 423                                    GLint s5)
 424 {
 425    GLint tokens[6];
 426    GLint idx;
 427    tokens[0] = s0;
 428    tokens[1] = s1;
 429    tokens[2] = s2;
 430    tokens[3] = s3;
 431    tokens[4] = s4;
 432    tokens[5] = s5;
 433    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 434    return make_ureg(PROGRAM_STATE_VAR, idx);
 435 }
 436
 437
 438 #define register_param1(p,s0)          register_param6(p,s0,0,0,0,0,0)
 439 #define register_param2(p,s0,s1)       register_param6(p,s0,s1,0,0,0,0)
 440 #define register_param3(p,s0,s1,s2)    register_param6(p,s0,s1,s2,0,0,0)
 441 #define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0)
 442
 443
 444 static void register_matrix_param6( struct tnl_program *p,
 445                                     GLint s0,
 446                                     GLint s1,
 447                                     GLint s2,
 448                                     GLint s3,
 449                                     GLint s4,
 450                                     GLint s5,
 451                                     struct ureg *matrix )
 452 {
 453    GLint i;
 454
 455    /* This is a bit sad as the support is there to pull the whole
 456     * matrix out in one go:
 457     */
 458    for (i = 0; i <= s4 - s3; i++)
 459       matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 );
 460 }
 461
 462
 463 static void emit_arg( struct prog_src_register *src,
 464                       struct ureg reg )
 465 {
 466    src->File = reg.file;
 467    src->Index = reg.idx;
 468    src->Swizzle = reg.swz;
 469    src->NegateBase = reg.negate ? NEGATE_XYZW : 0;
 470    src->Abs = 0;
 471    src->NegateAbs = 0;
 472    src->RelAddr = 0;
 473 }
 474
 475 static void emit_dst( struct prog_dst_register *dst,
 476                       struct ureg reg, GLuint mask )
 477 {
 478    dst->File = reg.file;
 479    dst->Index = reg.idx;
 480    /* allow zero as a shorthand for xyzw */
 481    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
 482    dst->CondMask = COND_TR;
 483    dst->CondSwizzle = 0;
 484    dst->CondSrc = 0;
 485    dst->pad = 0;
 486 }
 487
 488 static void debug_insn( struct prog_instruction *inst, const char *fn,
 489                         GLuint line )
 490 {
 491    if (DISASSEM) {
 492       static const char *last_fn;
 493
 494       if (fn != last_fn) {
 495          last_fn = fn;
 496          _mesa_printf("%s:\n", fn);
 497       }
 498
 499       _mesa_printf("%d:\t", line);
 500       _mesa_print_instruction(inst);
 501    }
 502 }
 503
 504
 505 static void emit_op3fn(struct tnl_program *p,
 506                        GLuint op,
 507                        struct ureg dest,
 508                        GLuint mask,
 509                        struct ureg src0,
 510                        struct ureg src1,
 511                        struct ureg src2,
 512                        const char *fn,
 513                        GLuint line)
 514 {
 515    GLuint nr = p->program->Base.NumInstructions++;
 516    struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 517
 518    if (p->program->Base.NumInstructions > MAX_INSN) {
 519       _mesa_problem(0, "Out of instructions in emit_op3fn\n");
 520       return;
 521    }
 522
 523    inst->Opcode = (enum prog_opcode) op;
 524    inst->StringPos = 0;
 525    inst->Data = 0;
 526
 527    emit_arg( &inst->SrcReg[0], src0 );
 528    emit_arg( &inst->SrcReg[1], src1 );
 529    emit_arg( &inst->SrcReg[2], src2 );
 530
 531    emit_dst( &inst->DstReg, dest, mask );
 532
 533    debug_insn(inst, fn, line);
 534 }
 535
 536
 537 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 538    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 539
 540 #define emit_op2(p, op, dst, mask, src0, src1) \
 541     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 542
 543 #define emit_op1(p, op, dst, mask, src0) \
 544     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 545
 546
 547 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 548 {
 549    if (reg.file == PROGRAM_TEMPORARY &&
 550        !(p->temp_reserved & (1<<reg.idx)))
 551       return reg;
 552    else {
 553       struct ureg temp = get_temp(p);
 554       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 555       return temp;
 556    }
 557 }
 558
 559
 560 /* Currently no tracking performed of input/output/register size or
 561  * active elements.  Could be used to reduce these operations, as
 562  * could the matrix type.
 563  */
 564 static void emit_matrix_transform_vec4( struct tnl_program *p,
 565                                         struct ureg dest,
 566                                         const struct ureg *mat,
 567                                         struct ureg src)
 568 {
 569    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 570    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 571    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 572    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 573 }
 574
 575 /* This version is much easier to implement if writemasks are not
 576  * supported natively on the target or (like SSE), the target doesn't
 577  * have a clean/obvious dotproduct implementation.
 578  */
 579 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 580                                                   struct ureg dest,
 581                                                   const struct ureg *mat,
 582                                                   struct ureg src)
 583 {
 584    struct ureg tmp;
 585
 586    if (dest.file != PROGRAM_TEMPORARY)
 587       tmp = get_temp(p);
 588    else
 589       tmp = dest;
 590
 591    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 592    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 593    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 594    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 595
 596    if (dest.file != PROGRAM_TEMPORARY)
 597       release_temp(p, tmp);
 598 }
 599
 600 static void emit_matrix_transform_vec3( struct tnl_program *p,
 601                                         struct ureg dest,
 602                                         const struct ureg *mat,
 603                                         struct ureg src)
 604 {
 605    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 606    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 607    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 608 }
 609
 610
 611 static void emit_normalize_vec3( struct tnl_program *p,
 612                                  struct ureg dest,
 613                                  struct ureg src )
 614 {
 615    struct ureg tmp = get_temp(p);
 616    emit_op2(p, OPCODE_DP3, tmp, 0, src, src);
 617    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
 618    emit_op2(p, OPCODE_MUL, dest, 0, src, tmp);
 619    release_temp(p, tmp);
 620 }
 621
 622 static void emit_passthrough( struct tnl_program *p,
 623                               GLuint input,
 624                               GLuint output )
 625 {
 626    struct ureg out = register_output(p, output);
 627    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 628 }
 629
 630 static struct ureg get_eye_position( struct tnl_program *p )
 631 {
 632    if (is_undef(p->eye_position)) {
 633       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 634       struct ureg modelview[4];
 635
 636       p->eye_position = reserve_temp(p);
 637
 638       if (PREFER_DP4) {
 639          register_matrix_param6( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 640                                  0, 0, modelview );
 641
 642          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 643       }
 644       else {
 645          register_matrix_param6( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 646                                  STATE_MATRIX_TRANSPOSE, 0, modelview );
 647
 648          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 649       }
 650    }
 651
 652    return p->eye_position;
 653 }
 654
 655
 656 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 657 {
 658    if (is_undef(p->eye_position_normalized)) {
 659       struct ureg eye = get_eye_position(p);
 660       p->eye_position_normalized = reserve_temp(p);
 661       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 662    }
 663
 664    return p->eye_position_normalized;
 665 }
 666
 667
 668 static struct ureg get_eye_normal( struct tnl_program *p )
 669 {
 670    if (is_undef(p->eye_normal)) {
 671       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
 672       struct ureg mvinv[3];
 673
 674       register_matrix_param6( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
 675                               STATE_MATRIX_INVTRANS, 0, mvinv );
 676
 677       p->eye_normal = reserve_temp(p);
 678
 679       /* Transform to eye space:
 680        */
 681       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 682
 683       /* Normalize/Rescale:
 684        */
 685       if (p->state->normalize) {
 686          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 687       }
 688       else if (p->state->rescale_normals) {
 689          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 690                                                STATE_NORMAL_SCALE);
 691
 692          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, normal,
 693                    swizzle1(rescale, X));
 694       }
 695    }
 696
 697    return p->eye_normal;
 698 }
 699
 700
 701
 702 static void build_hpos( struct tnl_program *p )
 703 {
 704    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 705    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 706    struct ureg mvp[4];
 707
 708    if (PREFER_DP4) {
 709       register_matrix_param6( p, STATE_MVP_MATRIX, 0, 0, 3,
 710                               0, 0, mvp );
 711       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 712    }
 713    else {
 714       register_matrix_param6( p, STATE_MVP_MATRIX, 0, 0, 3,
 715                               STATE_MATRIX_TRANSPOSE, 0, mvp );
 716       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 717    }
 718 }
 719
 720
 721 static GLuint material_attrib( GLuint side, GLuint property )
 722 {
 723    return ((property - STATE_AMBIENT) * 2 +
 724            side);
 725 }
 726
 727 /* Get a bitmask of which material values vary on a per-vertex basis.
 728  */
 729 static void set_material_flags( struct tnl_program *p )
 730 {
 731    p->color_materials = 0;
 732    p->materials = 0;
 733
 734    if (p->state->light_color_material) {
 735       p->materials =
 736          p->color_materials = p->state->light_color_material_mask;
 737    }
 738
 739    p->materials |= p->state->light_material_mask;
 740 }
 741
 742
 743 static struct ureg get_material( struct tnl_program *p, GLuint side,
 744                                  GLuint property )
 745 {
 746    GLuint attrib = material_attrib(side, property);
 747
 748    if (p->color_materials & (1<<attrib))
 749       return register_input(p, VERT_ATTRIB_COLOR0);
 750    else if (p->materials & (1<<attrib))
 751       return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
 752    else
 753       return register_param3( p, STATE_MATERIAL, side, property );
 754 }
 755
 756 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
 757                                    MAT_BIT_FRONT_AMBIENT | \
 758                                    MAT_BIT_FRONT_DIFFUSE) << (side))
 759
 760 /* Either return a precalculated constant value or emit code to
 761  * calculate these values dynamically in the case where material calls
 762  * are present between begin/end pairs.
 763  *
 764  * Probably want to shift this to the program compilation phase - if
 765  * we always emitted the calculation here, a smart compiler could
 766  * detect that it was constant (given a certain set of inputs), and
 767  * lift it out of the main loop.  That way the programs created here
 768  * would be independent of the vertex_buffer details.
 769  */
 770 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 771 {
 772    if (p->materials & SCENE_COLOR_BITS(side)) {
 773       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 774       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 775       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 776       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 777       struct ureg tmp = make_temp(p, material_diffuse);
 778       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 779                material_ambient, material_emission);
 780       return tmp;
 781    }
 782    else
 783       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 784 }
 785
 786
 787 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 788                                   GLuint side, GLuint property )
 789 {
 790    GLuint attrib = material_attrib(side, property);
 791    if (p->materials & (1<<attrib)) {
 792       struct ureg light_value =
 793          register_param3(p, STATE_LIGHT, light, property);
 794       struct ureg material_value = get_material(p, side, property);
 795       struct ureg tmp = get_temp(p);
 796       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 797       return tmp;
 798    }
 799    else
 800       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 801 }
 802
 803 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 804                                                 GLuint i,
 805                                                 struct ureg VPpli,
 806                                                 struct ureg dist )
 807 {
 808    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 809                                              STATE_ATTENUATION);
 810    struct ureg att = get_temp(p);
 811
 812    /* Calculate spot attenuation:
 813     */
 814    if (!p->state->unit[i].light_spotcutoff_is_180) {
 815       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
 816                                                   STATE_SPOT_DIR_NORMALIZED, i);
 817       struct ureg spot = get_temp(p);
 818       struct ureg slt = get_temp(p);
 819
 820       emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
 821       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
 822       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 823       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 824
 825       release_temp(p, spot);
 826       release_temp(p, slt);
 827    }
 828
 829    /* Calculate distance attenuation:
 830     */
 831    if (p->state->unit[i].light_attenuated) {
 832
 833       /* 1/d,d,d,1/d */
 834       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 835       /* 1,d,d*d,1/d */
 836       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 837       /* 1/dist-atten */
 838       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 839
 840       if (!p->state->unit[i].light_spotcutoff_is_180) {
 841          /* dist-atten */
 842          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 843          /* spot-atten * dist-atten */
 844          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 845       } else {
 846          /* dist-atten */
 847          emit_op1(p, OPCODE_RCP, att, 0, dist);
 848       }
 849    }
 850
 851    return att;
 852 }
 853
 854
 855
 856
 857
 858 /* Need to add some addtional parameters to allow lighting in object
 859  * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye
 860  * space lighting.
 861  */
 862 static void build_lighting( struct tnl_program *p )
 863 {
 864    const GLboolean twoside = p->state->light_twoside;
 865    const GLboolean separate = p->state->separate_specular;
 866    GLuint nr_lights = 0, count = 0;
 867    struct ureg normal = get_eye_normal(p);
 868    struct ureg lit = get_temp(p);
 869    struct ureg dots = get_temp(p);
 870    struct ureg _col0 = undef, _col1 = undef;
 871    struct ureg _bfc0 = undef, _bfc1 = undef;
 872    GLuint i;
 873
 874    for (i = 0; i < MAX_LIGHTS; i++)
 875       if (p->state->unit[i].light_enabled)
 876          nr_lights++;
 877
 878    set_material_flags(p);
 879
 880    {
 881       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 882       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 883       release_temp(p, shininess);
 884
 885       _col0 = make_temp(p, get_scenecolor(p, 0));
 886       if (separate)
 887          _col1 = make_temp(p, get_identity_param(p));
 888       else
 889          _col1 = _col0;
 890
 891    }
 892
 893    if (twoside) {
 894       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 895       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 896                negate(swizzle1(shininess,X)));
 897       release_temp(p, shininess);
 898
 899       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 900       if (separate)
 901          _bfc1 = make_temp(p, get_identity_param(p));
 902       else
 903          _bfc1 = _bfc0;
 904    }
 905
 906
 907    /* If no lights, still need to emit the scenecolor.
 908     */
 909       {
 910          struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 911          emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 912       }
 913
 914       if (separate) {
 915          struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 916          emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 917       }
 918
 919       if (twoside) {
 920          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 921          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 922       }
 923
 924       if (twoside && separate) {
 925          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 926          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 927       }
 928
 929    if (nr_lights == 0) {
 930       release_temps(p);
 931       return;
 932    }
 933
 934
 935    for (i = 0; i < MAX_LIGHTS; i++) {
 936       if (p->state->unit[i].light_enabled) {
 937          struct ureg half = undef;
 938          struct ureg att = undef, VPpli = undef;
 939
 940          count++;
 941
 942          if (p->state->unit[i].light_eyepos3_is_zero) {
 943             /* Can used precomputed constants in this case.
 944              * Attenuation never applies to infinite lights.
 945              */
 946             VPpli = register_param3(p, STATE_LIGHT, i,
 947                                     STATE_POSITION_NORMALIZED);
 948             half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
 949          }
 950          else {
 951             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
 952                                                STATE_POSITION);
 953             struct ureg V = get_eye_position(p);
 954             struct ureg dist = get_temp(p);
 955
 956             VPpli = get_temp(p);
 957             half = get_temp(p);
 958
 959             /* Calulate VPpli vector
 960              */
 961             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
 962
 963             /* Normalize VPpli.  The dist value also used in
 964              * attenuation below.
 965              */
 966             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
 967             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
 968             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
 969
 970
 971             /* Calculate  attenuation:
 972              */
 973             if (!p->state->unit[i].light_spotcutoff_is_180 ||
 974                 p->state->unit[i].light_attenuated) {
 975                att = calculate_light_attenuation(p, i, VPpli, dist);
 976             }
 977
 978
 979             /* Calculate viewer direction, or use infinite viewer:
 980              */
 981             if (p->state->light_local_viewer) {
 982                struct ureg eye_hat = get_eye_position_normalized(p);
 983                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 984             }
 985             else {
 986                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
 987                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
 988             }
 989
 990             emit_normalize_vec3(p, half, half);
 991
 992             release_temp(p, dist);
 993          }
 994
 995          /* Calculate dot products:
 996           */
 997          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
 998          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
 999
1000
1001          /* Front face lighting:
1002           */
1003          {
1004             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1005             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1006             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1007             struct ureg res0, res1;
1008             GLuint mask0, mask1;
1009
1010             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1011
1012             if (!is_undef(att))
1013                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1014
1015
1016             if (count == nr_lights) {
1017                if (separate) {
1018                   mask0 = WRITEMASK_XYZ;
1019                   mask1 = WRITEMASK_XYZ;
1020                   res0 = register_output( p, VERT_RESULT_COL0 );
1021                   res1 = register_output( p, VERT_RESULT_COL1 );
1022                }
1023                else {
1024                   mask0 = 0;
1025                   mask1 = WRITEMASK_XYZ;
1026                   res0 = _col0;
1027                   res1 = register_output( p, VERT_RESULT_COL0 );
1028                }
1029             } else {
1030                mask0 = 0;
1031                mask1 = 0;
1032                res0 = _col0;
1033                res1 = _col1;
1034             }
1035
1036             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1037             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1038             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1039
1040             release_temp(p, ambient);
1041             release_temp(p, diffuse);
1042             release_temp(p, specular);
1043          }
1044
1045          /* Back face lighting:
1046           */
1047          if (twoside) {
1048             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1049             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1050             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1051             struct ureg res0, res1;
1052             GLuint mask0, mask1;
1053
1054             emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
1055
1056             if (!is_undef(att))
1057                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1058
1059             if (count == nr_lights) {
1060                if (separate) {
1061                   mask0 = WRITEMASK_XYZ;
1062                   mask1 = WRITEMASK_XYZ;
1063                   res0 = register_output( p, VERT_RESULT_BFC0 );
1064                   res1 = register_output( p, VERT_RESULT_BFC1 );
1065                }
1066                else {
1067                   mask0 = 0;
1068                   mask1 = WRITEMASK_XYZ;
1069                   res0 = _bfc0;
1070                   res1 = register_output( p, VERT_RESULT_BFC0 );
1071                }
1072             } else {
1073                res0 = _bfc0;
1074                res1 = _bfc1;
1075                mask0 = 0;
1076                mask1 = 0;
1077             }
1078
1079             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1080             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1081             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1082
1083             release_temp(p, ambient);
1084             release_temp(p, diffuse);
1085             release_temp(p, specular);
1086          }
1087
1088          release_temp(p, half);
1089          release_temp(p, VPpli);
1090          release_temp(p, att);
1091       }
1092    }
1093
1094    release_temps( p );
1095 }
1096
1097
1098 static void build_fog( struct tnl_program *p )
1099 {
1100    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1101    struct ureg input;
1102
1103    if (p->state->fog_source_is_depth) {
1104       input = swizzle1(get_eye_position(p), Z);
1105    }
1106    else {
1107       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1108    }
1109
1110    if (p->state->tnl_do_vertex_fog) {
1111       struct ureg params = register_param2(p, STATE_INTERNAL,
1112                                            STATE_FOG_PARAMS_OPTIMIZED);
1113       struct ureg tmp = get_temp(p);
1114
1115       switch (p->state->fog_mode) {
1116       case FOG_LINEAR: {
1117          struct ureg id = get_identity_param(p);
1118          emit_op3(p, OPCODE_MAD, tmp, 0, input, swizzle1(params,X), swizzle1(params,Y));
1119          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1120          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1121          break;
1122       }
1123       case FOG_EXP:
1124          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1125          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,Z));
1126          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1127          break;
1128       case FOG_EXP2:
1129          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
1130          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1131          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1132          break;
1133       }
1134
1135       release_temp(p, tmp);
1136    }
1137    else {
1138       /* results = incoming fog coords (compute fog per-fragment later)
1139        *
1140        * KW:  Is it really necessary to do anything in this case?
1141        */
1142       emit_op1(p, OPCODE_MOV, fog, WRITEMASK_X, input);
1143    }
1144 }
1145
1146 static void build_reflect_texgen( struct tnl_program *p,
1147                                   struct ureg dest,
1148                                   GLuint writemask )
1149 {
1150    struct ureg normal = get_eye_normal(p);
1151    struct ureg eye_hat = get_eye_position_normalized(p);
1152    struct ureg tmp = get_temp(p);
1153
1154    /* n.u */
1155    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1156    /* 2n.u */
1157    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1158    /* (-2n.u)n + u */
1159    emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1160
1161    release_temp(p, tmp);
1162 }
1163
1164 static void build_sphere_texgen( struct tnl_program *p,
1165                                  struct ureg dest,
1166                                  GLuint writemask )
1167 {
1168    struct ureg normal = get_eye_normal(p);
1169    struct ureg eye_hat = get_eye_position_normalized(p);
1170    struct ureg tmp = get_temp(p);
1171    struct ureg half = register_scalar_const(p, .5);
1172    struct ureg r = get_temp(p);
1173    struct ureg inv_m = get_temp(p);
1174    struct ureg id = get_identity_param(p);
1175
1176    /* Could share the above calculations, but it would be
1177     * a fairly odd state for someone to set (both sphere and
1178     * reflection active for different texture coordinate
1179     * components.  Of course - if two texture units enable
1180     * reflect and/or sphere, things start to tilt in favour
1181     * of seperating this out:
1182     */
1183
1184    /* n.u */
1185    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1186    /* 2n.u */
1187    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1188    /* (-2n.u)n + u */
1189    emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1190    /* r + 0,0,1 */
1191    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1192    /* rx^2 + ry^2 + (rz+1)^2 */
1193    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1194    /* 2/m */
1195    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1196    /* 1/m */
1197    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1198    /* r/m + 1/2 */
1199    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1200
1201    release_temp(p, tmp);
1202    release_temp(p, r);
1203    release_temp(p, inv_m);
1204 }
1205
1206
1207 static void build_texture_transform( struct tnl_program *p )
1208 {
1209    GLuint i, j;
1210
1211    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1212
1213       if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
1214          continue;
1215
1216       if (p->state->unit[i].texgen_enabled ||
1217           p->state->unit[i].texmat_enabled) {
1218
1219          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1220          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1221          struct ureg out_texgen = undef;
1222
1223          if (p->state->unit[i].texgen_enabled) {
1224             GLuint copy_mask = 0;
1225             GLuint sphere_mask = 0;
1226             GLuint reflect_mask = 0;
1227             GLuint normal_mask = 0;
1228             GLuint modes[4];
1229
1230             if (texmat_enabled)
1231                out_texgen = get_temp(p);
1232             else
1233                out_texgen = out;
1234
1235             modes[0] = p->state->unit[i].texgen_mode0;
1236             modes[1] = p->state->unit[i].texgen_mode1;
1237             modes[2] = p->state->unit[i].texgen_mode2;
1238             modes[3] = p->state->unit[i].texgen_mode3;
1239
1240             for (j = 0; j < 4; j++) {
1241                switch (modes[j]) {
1242                case TXG_OBJ_LINEAR: {
1243                   struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1244                   struct ureg plane =
1245                      register_param3(p, STATE_TEXGEN, i,
1246                                      STATE_TEXGEN_OBJECT_S + j);
1247
1248                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1249                            obj, plane );
1250                   break;
1251                }
1252                case TXG_EYE_LINEAR: {
1253                   struct ureg eye = get_eye_position(p);
1254                   struct ureg plane =
1255                      register_param3(p, STATE_TEXGEN, i,
1256                                      STATE_TEXGEN_EYE_S + j);
1257
1258                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1259                            eye, plane );
1260                   break;
1261                }
1262                case TXG_SPHERE_MAP:
1263                   sphere_mask |= WRITEMASK_X << j;
1264                   break;
1265                case TXG_REFLECTION_MAP:
1266                   reflect_mask |= WRITEMASK_X << j;
1267                   break;
1268                case TXG_NORMAL_MAP:
1269                   normal_mask |= WRITEMASK_X << j;
1270                   break;
1271                case TXG_NONE:
1272                   copy_mask |= WRITEMASK_X << j;
1273                }
1274
1275             }
1276
1277
1278             if (sphere_mask) {
1279                build_sphere_texgen(p, out_texgen, sphere_mask);
1280             }
1281
1282             if (reflect_mask) {
1283                build_reflect_texgen(p, out_texgen, reflect_mask);
1284             }
1285
1286             if (normal_mask) {
1287                struct ureg normal = get_eye_normal(p);
1288                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1289             }
1290
1291             if (copy_mask) {
1292                struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1293                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1294             }
1295          }
1296
1297          if (texmat_enabled) {
1298             struct ureg texmat[4];
1299             struct ureg in = (!is_undef(out_texgen) ?
1300                               out_texgen :
1301                               register_input(p, VERT_ATTRIB_TEX0+i));
1302             if (PREFER_DP4) {
1303                register_matrix_param6( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1304                                        0, 0, texmat );
1305                emit_matrix_transform_vec4( p, out, texmat, in );
1306             }
1307             else {
1308                register_matrix_param6( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1309                                        STATE_MATRIX_TRANSPOSE, 0, texmat );
1310                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1311             }
1312          }
1313
1314          release_temps(p);
1315       }
1316       else {
1317          emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1318       }
1319    }
1320 }
1321
1322
1323 static void build_pointsize( struct tnl_program *p )
1324 {
1325    struct ureg eye = get_eye_position(p);
1326    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1327    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1328    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1329    struct ureg ut = get_temp(p);
1330
1331    /* p1 + dist * (p2 + dist * p3); */
1332    emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
1333                 swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1334    emit_op3(p, OPCODE_MAD, ut, 0, negate(swizzle1(eye, Z)),
1335                 ut, swizzle1(state_attenuation, X));
1336
1337    /* 1 / sqrt(factor) */
1338    emit_op1(p, OPCODE_RSQ, ut, 0, ut );
1339
1340 #if 1
1341    /* out = pointSize / sqrt(factor) */
1342    emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1343 #else
1344    /* not sure, might make sense to do clamping here,
1345       but it's not done in t_vb_points neither */
1346    emit_op2(p, OPCODE_MUL, ut, 0, ut, state_size);
1347    emit_op2(p, OPCODE_MAX, ut, 0, ut, swizzle1(state_size, Y));
1348    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1349 #endif
1350
1351    release_temp(p, ut);
1352 }
1353
1354 static void build_tnl_program( struct tnl_program *p )
1355 {   /* Emit the program, starting with modelviewproject:
1356     */
1357    build_hpos(p);
1358
1359    /* Lighting calculations:
1360     */
1361    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1362       if (p->state->light_global_enabled)
1363          build_lighting(p);
1364       else {
1365          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1366             emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);
1367
1368          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1369             emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
1370       }
1371    }
1372
1373    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1374        p->state->fog_mode != FOG_NONE)
1375       build_fog(p);
1376
1377    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1378       build_texture_transform(p);
1379
1380    if (p->state->point_attenuated)
1381       build_pointsize(p);
1382
1383    /* Finish up:
1384     */
1385    emit_op1(p, OPCODE_END, undef, 0, undef);
1386
1387    /* Disassemble:
1388     */
1389    if (DISASSEM) {
1390       _mesa_printf ("\n");
1391    }
1392 }
1393
1394
1395 static void
1396 create_new_program( const struct state_key *key,
1397                     struct gl_vertex_program *program,
1398                     GLuint max_temps)
1399 {
1400    struct tnl_program p;
1401
1402    _mesa_memset(&p, 0, sizeof(p));
1403    p.state = key;
1404    p.program = program;
1405    p.eye_position = undef;
1406    p.eye_position_normalized = undef;
1407    p.eye_normal = undef;
1408    p.identity = undef;
1409    p.temp_in_use = 0;
1410
1411    if (max_temps >= sizeof(int) * 8)
1412       p.temp_reserved = 0;
1413    else
1414       p.temp_reserved = ~((1<<max_temps)-1);
1415
1416    p.program->Base.Instructions
1417       = (struct prog_instruction*) MALLOC(sizeof(struct prog_instruction) * MAX_INSN);
1418    p.program->Base.String = 0;
1419    p.program->Base.NumInstructions =
1420    p.program->Base.NumTemporaries =
1421    p.program->Base.NumParameters =
1422    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1423    p.program->Base.Parameters = _mesa_new_parameter_list();
1424    p.program->Base.InputsRead = 0;
1425    p.program->Base.OutputsWritten = 0;
1426
1427    build_tnl_program( &p );
1428 }
1429
1430 static void *search_cache( struct tnl_cache *cache,
1431                            GLuint hash,
1432                            const void *key,
1433                            GLuint keysize)
1434 {
1435    struct tnl_cache_item *c;
1436
1437    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1438       if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0)
1439          return c->data;
1440    }
1441
1442    return NULL;
1443 }
1444
1445 static void rehash( struct tnl_cache *cache )
1446 {
1447    struct tnl_cache_item **items;
1448    struct tnl_cache_item *c, *next;
1449    GLuint size, i;
1450
1451    size = cache->size * 3;
1452    items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1453    _mesa_memset(items, 0, size * sizeof(*items));
1454
1455    for (i = 0; i < cache->size; i++)
1456       for (c = cache->items[i]; c; c = next) {
1457          next = c->next;
1458          c->next = items[c->hash % size];
1459          items[c->hash % size] = c;
1460       }
1461
1462    FREE(cache->items);
1463    cache->items = items;
1464    cache->size = size;
1465 }
1466
1467 static void cache_item( struct tnl_cache *cache,
1468                         GLuint hash,
1469                         void *key,
1470                         void *data )
1471 {
1472    struct tnl_cache_item *c = (struct tnl_cache_item*) _mesa_malloc(sizeof(*c));
1473    c->hash = hash;
1474    c->key = key;
1475    c->data = data;
1476
1477    if (++cache->n_items > cache->size * 1.5)
1478       rehash(cache);
1479
1480    c->next = cache->items[hash % cache->size];
1481    cache->items[hash % cache->size] = c;
1482 }
1483
1484 static GLuint hash_key( struct state_key *key )
1485 {
1486    GLuint *ikey = (GLuint *)key;
1487    GLuint hash = 0, i;
1488
1489    /* I'm sure this can be improved on, but speed is important:
1490     */
1491    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1492       hash ^= ikey[i];
1493
1494    return hash;
1495 }
1496
1497 void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
1498 {
1499    TNLcontext *tnl = TNL_CONTEXT(ctx);
1500    struct state_key *key;
1501    GLuint hash;
1502    const struct gl_vertex_program *prev = ctx->VertexProgram._Current;
1503
1504    if (!ctx->VertexProgram._Current) {
1505       /* Grab all the relevent state and put it in a single structure:
1506        */
1507       key = make_state_key(ctx);
1508       hash = hash_key(key);
1509
1510       /* Look for an already-prepared program for this state:
1511        */
1512       ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1513          search_cache( tnl->vp_cache, hash, key, sizeof(*key) );
1514
1515       /* OK, we'll have to build a new one:
1516        */
1517       if (!ctx->VertexProgram._TnlProgram) {
1518          if (0)
1519             _mesa_printf("Build new TNL program\n");
1520
1521          ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1522             ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1523
1524          create_new_program( key, ctx->VertexProgram._TnlProgram,
1525                              ctx->Const.VertexProgram.MaxTemps );
1526
1527          if (ctx->Driver.ProgramStringNotify)
1528             ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1529                                        &ctx->VertexProgram._TnlProgram->Base );
1530
1531          cache_item(tnl->vp_cache, hash, key, ctx->VertexProgram._TnlProgram );
1532       }
1533       else {
1534          FREE(key);
1535          if (0)
1536             _mesa_printf("Found existing TNL program for key %x\n", hash);
1537       }
1538       ctx->VertexProgram._Current = ctx->VertexProgram._TnlProgram;
1539    }
1540
1541    /* Tell the driver about the change.  Could define a new target for
1542     * this?
1543     */
1544    if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) {
1545       ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
1546                             (struct gl_program *) ctx->VertexProgram._Current);
1547    }
1548 }
1549
1550 void _tnl_ProgramCacheInit( GLcontext *ctx )
1551 {
1552    TNLcontext *tnl = TNL_CONTEXT(ctx);
1553
1554    tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache));
1555    tnl->vp_cache->size = 17;
1556    tnl->vp_cache->n_items = 0;
1557    tnl->vp_cache->items = (struct tnl_cache_item**)
1558       _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items));
1559 }
1560
1561 void _tnl_ProgramCacheDestroy( GLcontext *ctx )
1562 {
1563    TNLcontext *tnl = TNL_CONTEXT(ctx);
1564    struct tnl_cache_item *c, *next;
1565    GLuint i;
1566
1567    for (i = 0; i < tnl->vp_cache->size; i++)
1568       for (c = tnl->vp_cache->items[i]; c; c = next) {
1569          next = c->next;
1570          FREE(c->key);
1571          FREE(c->data);
1572          FREE(c);
1573       }
1574
1575    FREE(tnl->vp_cache->items);
1576    FREE(tnl->vp_cache);
1577 }