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 "shader/program.h"
  37 #include "shader/prog_instruction.h"
  38 #include "shader/prog_parameter.h"
  39 #include "shader/prog_print.h"
  40 #include "shader/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;
 122    struct state_key *key = CALLOC_STRUCT(state_key);
 123    GLuint i;
 124
 125    if (ctx->Shader.CurrentProgram &&
 126        ctx->Shader.CurrentProgram->LinkStatus &&
 127        ctx->Shader.CurrentProgram->FragmentProgram)
 128       fp = ctx->Shader.CurrentProgram->FragmentProgram;
 129    else
 130       fp = ctx->FragmentProgram._Current;
 131
 132    /* This now relies on texenvprogram.c being active:
 133     */
 134    assert(fp);
 135
 136    key->fragprog_inputs_read = fp->Base.InputsRead;
 137
 138    if (ctx->RenderMode == GL_FEEDBACK) {
 139       /* make sure the vertprog emits color and tex0 */
 140       key->fragprog_inputs_read |= (FRAG_BIT_COL0 | FRAG_BIT_TEX0);
 141    }
 142
 143    key->separate_specular = (ctx->Light.Model.ColorControl ==
 144                              GL_SEPARATE_SPECULAR_COLOR);
 145
 146    if (ctx->Light.Enabled) {
 147       key->light_global_enabled = 1;
 148
 149       if (ctx->Light.Model.LocalViewer)
 150          key->light_local_viewer = 1;
 151
 152       if (ctx->Light.Model.TwoSide)
 153          key->light_twoside = 1;
 154
 155       if (ctx->Light.ColorMaterialEnabled) {
 156          key->light_color_material = 1;
 157          key->light_color_material_mask = ctx->Light.ColorMaterialBitmask;
 158       }
 159
 160       for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++)
 161          if (VB->AttribPtr[i] && VB->AttribPtr[i]->stride)
 162             key->light_material_mask |= 1<<(i-_TNL_ATTRIB_MAT_FRONT_AMBIENT);
 163
 164       for (i = 0; i < MAX_LIGHTS; i++) {
 165          struct gl_light *light = &ctx->Light.Light[i];
 166
 167          if (light->Enabled) {
 168             key->unit[i].light_enabled = 1;
 169
 170             if (light->EyePosition[3] == 0.0)
 171                key->unit[i].light_eyepos3_is_zero = 1;
 172
 173             if (light->SpotCutoff == 180.0)
 174                key->unit[i].light_spotcutoff_is_180 = 1;
 175
 176             if (light->ConstantAttenuation != 1.0 ||
 177                 light->LinearAttenuation != 0.0 ||
 178                 light->QuadraticAttenuation != 0.0)
 179                key->unit[i].light_attenuated = 1;
 180          }
 181       }
 182    }
 183
 184    if (ctx->Transform.Normalize)
 185       key->normalize = 1;
 186
 187    if (ctx->Transform.RescaleNormals)
 188       key->rescale_normals = 1;
 189
 190    key->fog_mode = translate_fog_mode(fp->FogOption);
 191
 192    if (ctx->Fog.FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT)
 193       key->fog_source_is_depth = 1;
 194
 195    if (tnl->_DoVertexFog)
 196       key->tnl_do_vertex_fog = 1;
 197
 198    if (ctx->Point._Attenuated)
 199       key->point_attenuated = 1;
 200
 201    if (ctx->Texture._TexGenEnabled ||
 202        ctx->Texture._TexMatEnabled ||
 203        ctx->Texture._EnabledUnits)
 204       key->texture_enabled_global = 1;
 205
 206    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 207       struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
 208
 209       if (texUnit->_ReallyEnabled)
 210          key->unit[i].texunit_really_enabled = 1;
 211
 212       if (ctx->Texture._TexMatEnabled & ENABLE_TEXMAT(i))
 213          key->unit[i].texmat_enabled = 1;
 214
 215       if (texUnit->TexGenEnabled) {
 216          key->unit[i].texgen_enabled = 1;
 217
 218          key->unit[i].texgen_mode0 =
 219             translate_texgen( texUnit->TexGenEnabled & (1<<0),
 220                               texUnit->GenModeS );
 221          key->unit[i].texgen_mode1 =
 222             translate_texgen( texUnit->TexGenEnabled & (1<<1),
 223                               texUnit->GenModeT );
 224          key->unit[i].texgen_mode2 =
 225             translate_texgen( texUnit->TexGenEnabled & (1<<2),
 226                               texUnit->GenModeR );
 227          key->unit[i].texgen_mode3 =
 228             translate_texgen( texUnit->TexGenEnabled & (1<<3),
 229                               texUnit->GenModeQ );
 230       }
 231    }
 232
 233    return key;
 234 }
 235
 236
 237
 238 /* Very useful debugging tool - produces annotated listing of
 239  * generated program with line/function references for each
 240  * instruction back into this file:
 241  */
 242 #define DISASSEM (MESA_VERBOSE&VERBOSE_DISASSEM)
 243
 244 /* Should be tunable by the driver - do we want to do matrix
 245  * multiplications with DP4's or with MUL/MAD's?  SSE works better
 246  * with the latter, drivers may differ.
 247  */
 248 #define PREFER_DP4 0
 249
 250 #define MAX_INSN 256
 251
 252 /* Use uregs to represent registers internally, translate to Mesa's
 253  * expected formats on emit.
 254  *
 255  * NOTE: These are passed by value extensively in this file rather
 256  * than as usual by pointer reference.  If this disturbs you, try
 257  * remembering they are just 32bits in size.
 258  *
 259  * GCC is smart enough to deal with these dword-sized structures in
 260  * much the same way as if I had defined them as dwords and was using
 261  * macros to access and set the fields.  This is much nicer and easier
 262  * to evolve.
 263  */
 264 struct ureg {
 265    GLuint file:4;
 266    GLint idx:8;      /* relative addressing may be negative */
 267    GLuint negate:1;
 268    GLuint swz:12;
 269    GLuint pad:7;
 270 };
 271
 272
 273 struct tnl_program {
 274    const struct state_key *state;
 275    struct gl_vertex_program *program;
 276
 277    GLuint temp_in_use;
 278    GLuint temp_reserved;
 279
 280    struct ureg eye_position;
 281    struct ureg eye_position_normalized;
 282    struct ureg eye_normal;
 283    struct ureg identity;
 284
 285    GLuint materials;
 286    GLuint color_materials;
 287 };
 288
 289
 290 static const struct ureg undef = {
 291    PROGRAM_UNDEFINED,
 292    ~0,
 293    0,
 294    0,
 295    0
 296 };
 297
 298 /* Local shorthand:
 299  */
 300 #define X    SWIZZLE_X
 301 #define Y    SWIZZLE_Y
 302 #define Z    SWIZZLE_Z
 303 #define W    SWIZZLE_W
 304
 305
 306 /* Construct a ureg:
 307  */
 308 static struct ureg make_ureg(GLuint file, GLint idx)
 309 {
 310    struct ureg reg;
 311    reg.file = file;
 312    reg.idx = idx;
 313    reg.negate = 0;
 314    reg.swz = SWIZZLE_NOOP;
 315    reg.pad = 0;
 316    return reg;
 317 }
 318
 319
 320
 321 static struct ureg negate( struct ureg reg )
 322 {
 323    reg.negate ^= 1;
 324    return reg;
 325 }
 326
 327
 328 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 329 {
 330    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 331                            GET_SWZ(reg.swz, y),
 332                            GET_SWZ(reg.swz, z),
 333                            GET_SWZ(reg.swz, w));
 334
 335    return reg;
 336 }
 337
 338 static struct ureg swizzle1( struct ureg reg, int x )
 339 {
 340    return swizzle(reg, x, x, x, x);
 341 }
 342
 343 static struct ureg get_temp( struct tnl_program *p )
 344 {
 345    int bit = _mesa_ffs( ~p->temp_in_use );
 346    if (!bit) {
 347       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
 348       _mesa_exit(1);
 349    }
 350
 351    if ((GLuint) bit > p->program->Base.NumTemporaries)
 352       p->program->Base.NumTemporaries = bit;
 353
 354    p->temp_in_use |= 1<<(bit-1);
 355    return make_ureg(PROGRAM_TEMPORARY, bit-1);
 356 }
 357
 358 static struct ureg reserve_temp( struct tnl_program *p )
 359 {
 360    struct ureg temp = get_temp( p );
 361    p->temp_reserved |= 1<<temp.idx;
 362    return temp;
 363 }
 364
 365 static void release_temp( struct tnl_program *p, struct ureg reg )
 366 {
 367    if (reg.file == PROGRAM_TEMPORARY) {
 368       p->temp_in_use &= ~(1<<reg.idx);
 369       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
 370    }
 371 }
 372
 373 static void release_temps( struct tnl_program *p )
 374 {
 375    p->temp_in_use = p->temp_reserved;
 376 }
 377
 378
 379
 380 static struct ureg register_input( struct tnl_program *p, GLuint input )
 381 {
 382    p->program->Base.InputsRead |= (1<<input);
 383    return make_ureg(PROGRAM_INPUT, input);
 384 }
 385
 386 static struct ureg register_output( struct tnl_program *p, GLuint output )
 387 {
 388    p->program->Base.OutputsWritten |= (1<<output);
 389    return make_ureg(PROGRAM_OUTPUT, output);
 390 }
 391
 392 static struct ureg register_const4f( struct tnl_program *p,
 393                               GLfloat s0,
 394                               GLfloat s1,
 395                               GLfloat s2,
 396                               GLfloat s3)
 397 {
 398    GLfloat values[4];
 399    GLint idx;
 400    GLuint swizzle;
 401    values[0] = s0;
 402    values[1] = s1;
 403    values[2] = s2;
 404    values[3] = s3;
 405    idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
 406                                      &swizzle );
 407    ASSERT(swizzle == SWIZZLE_NOOP);
 408    return make_ureg(PROGRAM_STATE_VAR, idx);
 409 }
 410
 411 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
 412 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
 413 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
 414 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
 415
 416 static GLboolean is_undef( struct ureg reg )
 417 {
 418    return reg.file == PROGRAM_UNDEFINED;
 419 }
 420
 421 static struct ureg get_identity_param( struct tnl_program *p )
 422 {
 423    if (is_undef(p->identity))
 424       p->identity = register_const4f(p, 0,0,0,1);
 425
 426    return p->identity;
 427 }
 428
 429 static struct ureg register_param5(struct tnl_program *p,
 430                                    GLint s0,
 431                                    GLint s1,
 432                                    GLint s2,
 433                                    GLint s3,
 434                                    GLint s4)
 435 {
 436    gl_state_index tokens[STATE_LENGTH];
 437    GLint idx;
 438    tokens[0] = s0;
 439    tokens[1] = s1;
 440    tokens[2] = s2;
 441    tokens[3] = s3;
 442    tokens[4] = s4;
 443    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 444    return make_ureg(PROGRAM_STATE_VAR, idx);
 445 }
 446
 447
 448 #define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
 449 #define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
 450 #define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
 451 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
 452
 453
 454 static void register_matrix_param5( struct tnl_program *p,
 455                                     GLint s0, /* modelview, projection, etc */
 456                                     GLint s1, /* texture matrix number */
 457                                     GLint s2, /* first row */
 458                                     GLint s3, /* last row */
 459                                     GLint s4, /* inverse, transpose, etc */
 460                                     struct ureg *matrix )
 461 {
 462    GLint i;
 463
 464    /* This is a bit sad as the support is there to pull the whole
 465     * matrix out in one go:
 466     */
 467    for (i = 0; i <= s3 - s2; i++)
 468       matrix[i] = register_param5( p, s0, s1, i, i, s4 );
 469 }
 470
 471
 472 static void emit_arg( struct prog_src_register *src,
 473                       struct ureg reg )
 474 {
 475    src->File = reg.file;
 476    src->Index = reg.idx;
 477    src->Swizzle = reg.swz;
 478    src->NegateBase = reg.negate ? NEGATE_XYZW : 0;
 479    src->Abs = 0;
 480    src->NegateAbs = 0;
 481    src->RelAddr = 0;
 482 }
 483
 484 static void emit_dst( struct prog_dst_register *dst,
 485                       struct ureg reg, GLuint mask )
 486 {
 487    dst->File = reg.file;
 488    dst->Index = reg.idx;
 489    /* allow zero as a shorthand for xyzw */
 490    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
 491    dst->CondMask = COND_TR;
 492    dst->CondSwizzle = 0;
 493    dst->CondSrc = 0;
 494    dst->pad = 0;
 495 }
 496
 497 static void debug_insn( struct prog_instruction *inst, const char *fn,
 498                         GLuint line )
 499 {
 500    if (DISASSEM) {
 501       static const char *last_fn;
 502
 503       if (fn != last_fn) {
 504          last_fn = fn;
 505          _mesa_printf("%s:\n", fn);
 506       }
 507
 508       _mesa_printf("%d:\t", line);
 509       _mesa_print_instruction(inst);
 510    }
 511 }
 512
 513
 514 static void emit_op3fn(struct tnl_program *p,
 515                        GLuint op,
 516                        struct ureg dest,
 517                        GLuint mask,
 518                        struct ureg src0,
 519                        struct ureg src1,
 520                        struct ureg src2,
 521                        const char *fn,
 522                        GLuint line)
 523 {
 524    GLuint nr = p->program->Base.NumInstructions++;
 525    struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 526
 527    if (p->program->Base.NumInstructions > MAX_INSN) {
 528       _mesa_problem(0, "Out of instructions in emit_op3fn\n");
 529       return;
 530    }
 531
 532    inst->Opcode = (enum prog_opcode) op;
 533    inst->StringPos = 0;
 534    inst->Data = 0;
 535
 536    emit_arg( &inst->SrcReg[0], src0 );
 537    emit_arg( &inst->SrcReg[1], src1 );
 538    emit_arg( &inst->SrcReg[2], src2 );
 539
 540    emit_dst( &inst->DstReg, dest, mask );
 541
 542    debug_insn(inst, fn, line);
 543 }
 544
 545
 546 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 547    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 548
 549 #define emit_op2(p, op, dst, mask, src0, src1) \
 550     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 551
 552 #define emit_op1(p, op, dst, mask, src0) \
 553     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 554
 555
 556 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 557 {
 558    if (reg.file == PROGRAM_TEMPORARY &&
 559        !(p->temp_reserved & (1<<reg.idx)))
 560       return reg;
 561    else {
 562       struct ureg temp = get_temp(p);
 563       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 564       return temp;
 565    }
 566 }
 567
 568
 569 /* Currently no tracking performed of input/output/register size or
 570  * active elements.  Could be used to reduce these operations, as
 571  * could the matrix type.
 572  */
 573 static void emit_matrix_transform_vec4( struct tnl_program *p,
 574                                         struct ureg dest,
 575                                         const struct ureg *mat,
 576                                         struct ureg src)
 577 {
 578    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 579    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 580    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 581    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 582 }
 583
 584 /* This version is much easier to implement if writemasks are not
 585  * supported natively on the target or (like SSE), the target doesn't
 586  * have a clean/obvious dotproduct implementation.
 587  */
 588 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 589                                                   struct ureg dest,
 590                                                   const struct ureg *mat,
 591                                                   struct ureg src)
 592 {
 593    struct ureg tmp;
 594
 595    if (dest.file != PROGRAM_TEMPORARY)
 596       tmp = get_temp(p);
 597    else
 598       tmp = dest;
 599
 600    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 601    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 602    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 603    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 604
 605    if (dest.file != PROGRAM_TEMPORARY)
 606       release_temp(p, tmp);
 607 }
 608
 609 static void emit_matrix_transform_vec3( struct tnl_program *p,
 610                                         struct ureg dest,
 611                                         const struct ureg *mat,
 612                                         struct ureg src)
 613 {
 614    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 615    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 616    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 617 }
 618
 619
 620 static void emit_normalize_vec3( struct tnl_program *p,
 621                                  struct ureg dest,
 622                                  struct ureg src )
 623 {
 624    struct ureg tmp = get_temp(p);
 625    emit_op2(p, OPCODE_DP3, tmp, 0, src, src);
 626    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
 627    emit_op2(p, OPCODE_MUL, dest, 0, src, tmp);
 628    release_temp(p, tmp);
 629 }
 630
 631 static void emit_passthrough( struct tnl_program *p,
 632                               GLuint input,
 633                               GLuint output )
 634 {
 635    struct ureg out = register_output(p, output);
 636    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 637 }
 638
 639 static struct ureg get_eye_position( struct tnl_program *p )
 640 {
 641    if (is_undef(p->eye_position)) {
 642       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 643       struct ureg modelview[4];
 644
 645       p->eye_position = reserve_temp(p);
 646
 647       if (PREFER_DP4) {
 648          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 649                                  0, modelview );
 650
 651          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 652       }
 653       else {
 654          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 655                                  STATE_MATRIX_TRANSPOSE, modelview );
 656
 657          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 658       }
 659    }
 660
 661    return p->eye_position;
 662 }
 663
 664
 665 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 666 {
 667    if (is_undef(p->eye_position_normalized)) {
 668       struct ureg eye = get_eye_position(p);
 669       p->eye_position_normalized = reserve_temp(p);
 670       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 671    }
 672
 673    return p->eye_position_normalized;
 674 }
 675
 676
 677 static struct ureg get_eye_normal( struct tnl_program *p )
 678 {
 679    if (is_undef(p->eye_normal)) {
 680       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
 681       struct ureg mvinv[3];
 682
 683       register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
 684                               STATE_MATRIX_INVTRANS, mvinv );
 685
 686       p->eye_normal = reserve_temp(p);
 687
 688       /* Transform to eye space:
 689        */
 690       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 691
 692       /* Normalize/Rescale:
 693        */
 694       if (p->state->normalize) {
 695          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 696       }
 697       else if (p->state->rescale_normals) {
 698          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 699                                                STATE_NORMAL_SCALE);
 700
 701          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal,
 702                    swizzle1(rescale, X));
 703       }
 704    }
 705
 706    return p->eye_normal;
 707 }
 708
 709
 710
 711 static void build_hpos( struct tnl_program *p )
 712 {
 713    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 714    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 715    struct ureg mvp[4];
 716
 717    if (PREFER_DP4) {
 718       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 719                               0, mvp );
 720       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 721    }
 722    else {
 723       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 724                               STATE_MATRIX_TRANSPOSE, mvp );
 725       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 726    }
 727 }
 728
 729
 730 static GLuint material_attrib( GLuint side, GLuint property )
 731 {
 732    return ((property - STATE_AMBIENT) * 2 +
 733            side);
 734 }
 735
 736 /* Get a bitmask of which material values vary on a per-vertex basis.
 737  */
 738 static void set_material_flags( struct tnl_program *p )
 739 {
 740    p->color_materials = 0;
 741    p->materials = 0;
 742
 743    if (p->state->light_color_material) {
 744       p->materials =
 745          p->color_materials = p->state->light_color_material_mask;
 746    }
 747
 748    p->materials |= p->state->light_material_mask;
 749 }
 750
 751
 752 static struct ureg get_material( struct tnl_program *p, GLuint side,
 753                                  GLuint property )
 754 {
 755    GLuint attrib = material_attrib(side, property);
 756
 757    if (p->color_materials & (1<<attrib))
 758       return register_input(p, VERT_ATTRIB_COLOR0);
 759    else if (p->materials & (1<<attrib))
 760       return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
 761    else
 762       return register_param3( p, STATE_MATERIAL, side, property );
 763 }
 764
 765 #define SCENE_COLOR_BITS(side) (( MAT_BIT_FRONT_EMISSION | \
 766                                    MAT_BIT_FRONT_AMBIENT | \
 767                                    MAT_BIT_FRONT_DIFFUSE) << (side))
 768
 769 /* Either return a precalculated constant value or emit code to
 770  * calculate these values dynamically in the case where material calls
 771  * are present between begin/end pairs.
 772  *
 773  * Probably want to shift this to the program compilation phase - if
 774  * we always emitted the calculation here, a smart compiler could
 775  * detect that it was constant (given a certain set of inputs), and
 776  * lift it out of the main loop.  That way the programs created here
 777  * would be independent of the vertex_buffer details.
 778  */
 779 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 780 {
 781    if (p->materials & SCENE_COLOR_BITS(side)) {
 782       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 783       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 784       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 785       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 786       struct ureg tmp = make_temp(p, material_diffuse);
 787       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 788                material_ambient, material_emission);
 789       return tmp;
 790    }
 791    else
 792       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 793 }
 794
 795
 796 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 797                                   GLuint side, GLuint property )
 798 {
 799    GLuint attrib = material_attrib(side, property);
 800    if (p->materials & (1<<attrib)) {
 801       struct ureg light_value =
 802          register_param3(p, STATE_LIGHT, light, property);
 803       struct ureg material_value = get_material(p, side, property);
 804       struct ureg tmp = get_temp(p);
 805       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 806       return tmp;
 807    }
 808    else
 809       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 810 }
 811
 812 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 813                                                 GLuint i,
 814                                                 struct ureg VPpli,
 815                                                 struct ureg dist )
 816 {
 817    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 818                                              STATE_ATTENUATION);
 819    struct ureg att = get_temp(p);
 820
 821    /* Calculate spot attenuation:
 822     */
 823    if (!p->state->unit[i].light_spotcutoff_is_180) {
 824       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
 825                                                   STATE_SPOT_DIR_NORMALIZED, i);
 826       struct ureg spot = get_temp(p);
 827       struct ureg slt = get_temp(p);
 828
 829       emit_op2(p, OPCODE_DP3, spot, 0, negate(VPpli), spot_dir_norm);
 830       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot);
 831       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 832       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 833
 834       release_temp(p, spot);
 835       release_temp(p, slt);
 836    }
 837
 838    /* Calculate distance attenuation:
 839     */
 840    if (p->state->unit[i].light_attenuated) {
 841
 842       /* 1/d,d,d,1/d */
 843       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 844       /* 1,d,d*d,1/d */
 845       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 846       /* 1/dist-atten */
 847       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 848
 849       if (!p->state->unit[i].light_spotcutoff_is_180) {
 850          /* dist-atten */
 851          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 852          /* spot-atten * dist-atten */
 853          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 854       } else {
 855          /* dist-atten */
 856          emit_op1(p, OPCODE_RCP, att, 0, dist);
 857       }
 858    }
 859
 860    return att;
 861 }
 862
 863
 864
 865
 866
 867 /* Need to add some addtional parameters to allow lighting in object
 868  * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
 869  * space lighting.
 870  */
 871 static void build_lighting( struct tnl_program *p )
 872 {
 873    const GLboolean twoside = p->state->light_twoside;
 874    const GLboolean separate = p->state->separate_specular;
 875    GLuint nr_lights = 0, count = 0;
 876    struct ureg normal = get_eye_normal(p);
 877    struct ureg lit = get_temp(p);
 878    struct ureg dots = get_temp(p);
 879    struct ureg _col0 = undef, _col1 = undef;
 880    struct ureg _bfc0 = undef, _bfc1 = undef;
 881    GLuint i;
 882
 883    for (i = 0; i < MAX_LIGHTS; i++)
 884       if (p->state->unit[i].light_enabled)
 885          nr_lights++;
 886
 887    set_material_flags(p);
 888
 889    {
 890       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 891       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 892       release_temp(p, shininess);
 893
 894       _col0 = make_temp(p, get_scenecolor(p, 0));
 895       if (separate)
 896          _col1 = make_temp(p, get_identity_param(p));
 897       else
 898          _col1 = _col0;
 899
 900    }
 901
 902    if (twoside) {
 903       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 904       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 905                negate(swizzle1(shininess,X)));
 906       release_temp(p, shininess);
 907
 908       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 909       if (separate)
 910          _bfc1 = make_temp(p, get_identity_param(p));
 911       else
 912          _bfc1 = _bfc0;
 913    }
 914
 915
 916    /* If no lights, still need to emit the scenecolor.
 917     */
 918       {
 919          struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 920          emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 921       }
 922
 923       if (separate) {
 924          struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 925          emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 926       }
 927
 928       if (twoside) {
 929          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 930          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 931       }
 932
 933       if (twoside && separate) {
 934          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 935          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 936       }
 937
 938    if (nr_lights == 0) {
 939       release_temps(p);
 940       return;
 941    }
 942
 943
 944    for (i = 0; i < MAX_LIGHTS; i++) {
 945       if (p->state->unit[i].light_enabled) {
 946          struct ureg half = undef;
 947          struct ureg att = undef, VPpli = undef;
 948
 949          count++;
 950
 951          if (p->state->unit[i].light_eyepos3_is_zero) {
 952             /* Can used precomputed constants in this case.
 953              * Attenuation never applies to infinite lights.
 954              */
 955             VPpli = register_param3(p, STATE_LIGHT, i,
 956                                     STATE_POSITION_NORMALIZED);
 957             if (p->state->light_local_viewer) {
 958                 struct ureg eye_hat = get_eye_position_normalized(p);
 959                 half = get_temp(p);
 960                 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 961                 emit_normalize_vec3(p, half, half);
 962             } else {
 963                 half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
 964             }
 965          }
 966          else {
 967             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
 968                                                STATE_POSITION);
 969             struct ureg V = get_eye_position(p);
 970             struct ureg dist = get_temp(p);
 971
 972             VPpli = get_temp(p);
 973             half = get_temp(p);
 974
 975             /* Calulate VPpli vector
 976              */
 977             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
 978
 979             /* Normalize VPpli.  The dist value also used in
 980              * attenuation below.
 981              */
 982             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
 983             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
 984             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
 985
 986
 987             /* Calculate  attenuation:
 988              */
 989             if (!p->state->unit[i].light_spotcutoff_is_180 ||
 990                 p->state->unit[i].light_attenuated) {
 991                att = calculate_light_attenuation(p, i, VPpli, dist);
 992             }
 993
 994
 995             /* Calculate viewer direction, or use infinite viewer:
 996              */
 997             if (p->state->light_local_viewer) {
 998                struct ureg eye_hat = get_eye_position_normalized(p);
 999                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1000             }
1001             else {
1002                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1003                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1004             }
1005
1006             emit_normalize_vec3(p, half, half);
1007
1008             release_temp(p, dist);
1009          }
1010
1011          /* Calculate dot products:
1012           */
1013          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1014          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1015
1016
1017          /* Front face lighting:
1018           */
1019          {
1020             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1021             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1022             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1023             struct ureg res0, res1;
1024             GLuint mask0, mask1;
1025
1026             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1027
1028             if (!is_undef(att))
1029                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1030
1031
1032             if (count == nr_lights) {
1033                if (separate) {
1034                   mask0 = WRITEMASK_XYZ;
1035                   mask1 = WRITEMASK_XYZ;
1036                   res0 = register_output( p, VERT_RESULT_COL0 );
1037                   res1 = register_output( p, VERT_RESULT_COL1 );
1038                }
1039                else {
1040                   mask0 = 0;
1041                   mask1 = WRITEMASK_XYZ;
1042                   res0 = _col0;
1043                   res1 = register_output( p, VERT_RESULT_COL0 );
1044                }
1045             } else {
1046                mask0 = 0;
1047                mask1 = 0;
1048                res0 = _col0;
1049                res1 = _col1;
1050             }
1051
1052             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1053             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1054             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1055
1056             release_temp(p, ambient);
1057             release_temp(p, diffuse);
1058             release_temp(p, specular);
1059          }
1060
1061          /* Back face lighting:
1062           */
1063          if (twoside) {
1064             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1065             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1066             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1067             struct ureg res0, res1;
1068             GLuint mask0, mask1;
1069
1070             emit_op1(p, OPCODE_LIT, lit, 0, negate(swizzle(dots,X,Y,W,Z)));
1071
1072             if (!is_undef(att))
1073                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1074
1075             if (count == nr_lights) {
1076                if (separate) {
1077                   mask0 = WRITEMASK_XYZ;
1078                   mask1 = WRITEMASK_XYZ;
1079                   res0 = register_output( p, VERT_RESULT_BFC0 );
1080                   res1 = register_output( p, VERT_RESULT_BFC1 );
1081                }
1082                else {
1083                   mask0 = 0;
1084                   mask1 = WRITEMASK_XYZ;
1085                   res0 = _bfc0;
1086                   res1 = register_output( p, VERT_RESULT_BFC0 );
1087                }
1088             } else {
1089                res0 = _bfc0;
1090                res1 = _bfc1;
1091                mask0 = 0;
1092                mask1 = 0;
1093             }
1094
1095             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1096             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1097             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1098
1099             release_temp(p, ambient);
1100             release_temp(p, diffuse);
1101             release_temp(p, specular);
1102          }
1103
1104          release_temp(p, half);
1105          release_temp(p, VPpli);
1106          release_temp(p, att);
1107       }
1108    }
1109
1110    release_temps( p );
1111 }
1112
1113
1114 static void build_fog( struct tnl_program *p )
1115 {
1116    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1117    struct ureg input;
1118
1119    if (p->state->fog_source_is_depth) {
1120       input = swizzle1(get_eye_position(p), Z);
1121    }
1122    else {
1123       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1124    }
1125
1126    if (p->state->fog_mode && p->state->tnl_do_vertex_fog) {
1127       struct ureg params = register_param2(p, STATE_INTERNAL,
1128                                            STATE_FOG_PARAMS_OPTIMIZED);
1129       struct ureg tmp = get_temp(p);
1130       GLboolean useabs = (p->state->fog_mode != FOG_EXP2);
1131
1132       if (useabs) {
1133          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1134       }
1135
1136       switch (p->state->fog_mode) {
1137       case FOG_LINEAR: {
1138          struct ureg id = get_identity_param(p);
1139          emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input,
1140                         swizzle1(params,X), swizzle1(params,Y));
1141          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1142          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1143          break;
1144       }
1145       case FOG_EXP:
1146          emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input,
1147                         swizzle1(params,Z));
1148          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1149          break;
1150       case FOG_EXP2:
1151          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
1152          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1153          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, negate(tmp));
1154          break;
1155       }
1156
1157       release_temp(p, tmp);
1158    }
1159    else {
1160       /* results = incoming fog coords (compute fog per-fragment later)
1161        *
1162        * KW:  Is it really necessary to do anything in this case?
1163        * BP: Yes, we always need to compute the absolute value, unless
1164        * we want to push that down into the fragment program...
1165        */
1166       GLboolean useabs = GL_TRUE;
1167       emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, WRITEMASK_X, input);
1168    }
1169 }
1170
1171 static void build_reflect_texgen( struct tnl_program *p,
1172                                   struct ureg dest,
1173                                   GLuint writemask )
1174 {
1175    struct ureg normal = get_eye_normal(p);
1176    struct ureg eye_hat = get_eye_position_normalized(p);
1177    struct ureg tmp = get_temp(p);
1178
1179    /* n.u */
1180    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1181    /* 2n.u */
1182    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1183    /* (-2n.u)n + u */
1184    emit_op3(p, OPCODE_MAD, dest, writemask, negate(tmp), normal, eye_hat);
1185
1186    release_temp(p, tmp);
1187 }
1188
1189 static void build_sphere_texgen( struct tnl_program *p,
1190                                  struct ureg dest,
1191                                  GLuint writemask )
1192 {
1193    struct ureg normal = get_eye_normal(p);
1194    struct ureg eye_hat = get_eye_position_normalized(p);
1195    struct ureg tmp = get_temp(p);
1196    struct ureg half = register_scalar_const(p, .5);
1197    struct ureg r = get_temp(p);
1198    struct ureg inv_m = get_temp(p);
1199    struct ureg id = get_identity_param(p);
1200
1201    /* Could share the above calculations, but it would be
1202     * a fairly odd state for someone to set (both sphere and
1203     * reflection active for different texture coordinate
1204     * components.  Of course - if two texture units enable
1205     * reflect and/or sphere, things start to tilt in favour
1206     * of seperating this out:
1207     */
1208
1209    /* n.u */
1210    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1211    /* 2n.u */
1212    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1213    /* (-2n.u)n + u */
1214    emit_op3(p, OPCODE_MAD, r, 0, negate(tmp), normal, eye_hat);
1215    /* r + 0,0,1 */
1216    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1217    /* rx^2 + ry^2 + (rz+1)^2 */
1218    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1219    /* 2/m */
1220    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1221    /* 1/m */
1222    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1223    /* r/m + 1/2 */
1224    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1225
1226    release_temp(p, tmp);
1227    release_temp(p, r);
1228    release_temp(p, inv_m);
1229 }
1230
1231
1232 static void build_texture_transform( struct tnl_program *p )
1233 {
1234    GLuint i, j;
1235
1236    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1237
1238       if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
1239          continue;
1240
1241       if (p->state->unit[i].texgen_enabled ||
1242           p->state->unit[i].texmat_enabled) {
1243
1244          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1245          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1246          struct ureg out_texgen = undef;
1247
1248          if (p->state->unit[i].texgen_enabled) {
1249             GLuint copy_mask = 0;
1250             GLuint sphere_mask = 0;
1251             GLuint reflect_mask = 0;
1252             GLuint normal_mask = 0;
1253             GLuint modes[4];
1254
1255             if (texmat_enabled)
1256                out_texgen = get_temp(p);
1257             else
1258                out_texgen = out;
1259
1260             modes[0] = p->state->unit[i].texgen_mode0;
1261             modes[1] = p->state->unit[i].texgen_mode1;
1262             modes[2] = p->state->unit[i].texgen_mode2;
1263             modes[3] = p->state->unit[i].texgen_mode3;
1264
1265             for (j = 0; j < 4; j++) {
1266                switch (modes[j]) {
1267                case TXG_OBJ_LINEAR: {
1268                   struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1269                   struct ureg plane =
1270                      register_param3(p, STATE_TEXGEN, i,
1271                                      STATE_TEXGEN_OBJECT_S + j);
1272
1273                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1274                            obj, plane );
1275                   break;
1276                }
1277                case TXG_EYE_LINEAR: {
1278                   struct ureg eye = get_eye_position(p);
1279                   struct ureg plane =
1280                      register_param3(p, STATE_TEXGEN, i,
1281                                      STATE_TEXGEN_EYE_S + j);
1282
1283                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1284                            eye, plane );
1285                   break;
1286                }
1287                case TXG_SPHERE_MAP:
1288                   sphere_mask |= WRITEMASK_X << j;
1289                   break;
1290                case TXG_REFLECTION_MAP:
1291                   reflect_mask |= WRITEMASK_X << j;
1292                   break;
1293                case TXG_NORMAL_MAP:
1294                   normal_mask |= WRITEMASK_X << j;
1295                   break;
1296                case TXG_NONE:
1297                   copy_mask |= WRITEMASK_X << j;
1298                }
1299
1300             }
1301
1302
1303             if (sphere_mask) {
1304                build_sphere_texgen(p, out_texgen, sphere_mask);
1305             }
1306
1307             if (reflect_mask) {
1308                build_reflect_texgen(p, out_texgen, reflect_mask);
1309             }
1310
1311             if (normal_mask) {
1312                struct ureg normal = get_eye_normal(p);
1313                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1314             }
1315
1316             if (copy_mask) {
1317                struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1318                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1319             }
1320          }
1321
1322          if (texmat_enabled) {
1323             struct ureg texmat[4];
1324             struct ureg in = (!is_undef(out_texgen) ?
1325                               out_texgen :
1326                               register_input(p, VERT_ATTRIB_TEX0+i));
1327             if (PREFER_DP4) {
1328                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1329                                        0, texmat );
1330                emit_matrix_transform_vec4( p, out, texmat, in );
1331             }
1332             else {
1333                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1334                                        STATE_MATRIX_TRANSPOSE, texmat );
1335                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1336             }
1337          }
1338
1339          release_temps(p);
1340       }
1341       else {
1342          emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1343       }
1344    }
1345 }
1346
1347
1348 static void build_pointsize( struct tnl_program *p )
1349 {
1350    struct ureg eye = get_eye_position(p);
1351    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1352    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1353    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1354    struct ureg ut = get_temp(p);
1355
1356    /* dist = |eyez| */
1357    emit_op1(p, OPCODE_ABS, ut, WRITEMASK_Y, swizzle1(eye, Z));
1358    /* p1 + dist * (p2 + dist * p3); */
1359    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1360                 swizzle1(state_attenuation, Z), swizzle1(state_attenuation, Y));
1361    emit_op3(p, OPCODE_MAD, ut, WRITEMASK_X, swizzle1(ut, Y),
1362                 ut, swizzle1(state_attenuation, X));
1363
1364    /* 1 / sqrt(factor) */
1365    emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1366
1367 #if 1
1368    /* out = pointSize / sqrt(factor) */
1369    emit_op2(p, OPCODE_MUL, out, WRITEMASK_X, ut, state_size);
1370 #else
1371    /* not sure, might make sense to do clamping here,
1372       but it's not done in t_vb_points neither */
1373    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1374    emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1375    emit_op2(p, OPCODE_MIN, out, WRITEMASK_X, ut, swizzle1(state_size, Z));
1376 #endif
1377
1378    release_temp(p, ut);
1379 }
1380
1381 /**
1382  * Emit constant point size.
1383  */
1384 static void constant_pointsize( struct tnl_program *p )
1385 {
1386    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1387    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1388    emit_op1(p, OPCODE_MOV, out, WRITEMASK_X, state_size);
1389 }
1390
1391 static void build_tnl_program( struct tnl_program *p )
1392 {   /* Emit the program, starting with modelviewproject:
1393     */
1394    build_hpos(p);
1395
1396    /* Lighting calculations:
1397     */
1398    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1399       if (p->state->light_global_enabled)
1400          build_lighting(p);
1401       else {
1402          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1403             emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);
1404
1405          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1406             emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
1407       }
1408    }
1409
1410    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1411        p->state->fog_mode != FOG_NONE)
1412       build_fog(p);
1413
1414    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1415       build_texture_transform(p);
1416
1417    if (p->state->point_attenuated)
1418       build_pointsize(p);
1419 #if 0
1420    else
1421       constant_pointsize(p);
1422 #endif
1423
1424    /* Finish up:
1425     */
1426    emit_op1(p, OPCODE_END, undef, 0, undef);
1427
1428    /* Disassemble:
1429     */
1430    if (DISASSEM) {
1431       _mesa_printf ("\n");
1432    }
1433 }
1434
1435
1436 static void
1437 create_new_program( const struct state_key *key,
1438                     struct gl_vertex_program *program,
1439                     GLuint max_temps)
1440 {
1441    struct tnl_program p;
1442
1443    _mesa_memset(&p, 0, sizeof(p));
1444    p.state = key;
1445    p.program = program;
1446    p.eye_position = undef;
1447    p.eye_position_normalized = undef;
1448    p.eye_normal = undef;
1449    p.identity = undef;
1450    p.temp_in_use = 0;
1451
1452    if (max_temps >= sizeof(int) * 8)
1453       p.temp_reserved = 0;
1454    else
1455       p.temp_reserved = ~((1<<max_temps)-1);
1456
1457    p.program->Base.Instructions = _mesa_alloc_instructions(MAX_INSN);
1458    p.program->Base.String = NULL;
1459    p.program->Base.NumInstructions =
1460    p.program->Base.NumTemporaries =
1461    p.program->Base.NumParameters =
1462    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1463    p.program->Base.Parameters = _mesa_new_parameter_list();
1464    p.program->Base.InputsRead = 0;
1465    p.program->Base.OutputsWritten = 0;
1466
1467    build_tnl_program( &p );
1468 }
1469
1470 static void *search_cache( struct tnl_cache *cache,
1471                            GLuint hash,
1472                            const void *key,
1473                            GLuint keysize)
1474 {
1475    struct tnl_cache_item *c;
1476
1477    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1478       if (c->hash == hash && _mesa_memcmp(c->key, key, keysize) == 0)
1479          return c->data;
1480    }
1481
1482    return NULL;
1483 }
1484
1485 static void rehash( struct tnl_cache *cache )
1486 {
1487    struct tnl_cache_item **items;
1488    struct tnl_cache_item *c, *next;
1489    GLuint size, i;
1490
1491    size = cache->size * 3;
1492    items = (struct tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1493    _mesa_memset(items, 0, size * sizeof(*items));
1494
1495    for (i = 0; i < cache->size; i++)
1496       for (c = cache->items[i]; c; c = next) {
1497          next = c->next;
1498          c->next = items[c->hash % size];
1499          items[c->hash % size] = c;
1500       }
1501
1502    FREE(cache->items);
1503    cache->items = items;
1504    cache->size = size;
1505 }
1506
1507 static void cache_item( struct tnl_cache *cache,
1508                         GLuint hash,
1509                         void *key,
1510                         void *data )
1511 {
1512    struct tnl_cache_item *c = (struct tnl_cache_item*) _mesa_malloc(sizeof(*c));
1513    c->hash = hash;
1514    c->key = key;
1515    c->data = data;
1516
1517    if (++cache->n_items > cache->size * 1.5)
1518       rehash(cache);
1519
1520    c->next = cache->items[hash % cache->size];
1521    cache->items[hash % cache->size] = c;
1522 }
1523
1524 static GLuint hash_key( struct state_key *key )
1525 {
1526    GLuint *ikey = (GLuint *)key;
1527    GLuint hash = 0, i;
1528
1529    /* I'm sure this can be improved on, but speed is important:
1530     */
1531    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1532       hash ^= ikey[i];
1533
1534    return hash;
1535 }
1536
1537 void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
1538 {
1539    TNLcontext *tnl = TNL_CONTEXT(ctx);
1540    struct state_key *key;
1541    GLuint hash;
1542    const struct gl_vertex_program *prev = ctx->VertexProgram._Current;
1543
1544    if (!ctx->VertexProgram._Current ||
1545        ctx->VertexProgram._Current == ctx->VertexProgram._TnlProgram) {
1546       /* Grab all the relevent state and put it in a single structure:
1547        */
1548       key = make_state_key(ctx);
1549       hash = hash_key(key);
1550
1551       /* Look for an already-prepared program for this state:
1552        */
1553       ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1554          search_cache( tnl->vp_cache, hash, key, sizeof(*key) );
1555
1556       /* OK, we'll have to build a new one:
1557        */
1558       if (!ctx->VertexProgram._TnlProgram) {
1559          if (0)
1560             _mesa_printf("Build new TNL program\n");
1561
1562          ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
1563             ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1564
1565          create_new_program( key, ctx->VertexProgram._TnlProgram,
1566                              ctx->Const.VertexProgram.MaxTemps );
1567
1568          if (ctx->Driver.ProgramStringNotify)
1569             ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1570                                        &ctx->VertexProgram._TnlProgram->Base );
1571
1572          cache_item(tnl->vp_cache, hash, key, ctx->VertexProgram._TnlProgram );
1573       }
1574       else {
1575          FREE(key);
1576          if (0)
1577             _mesa_printf("Found existing TNL program for key %x\n", hash);
1578       }
1579       ctx->VertexProgram._Current = ctx->VertexProgram._TnlProgram;
1580    }
1581
1582    /* Tell the driver about the change.  Could define a new target for
1583     * this?
1584     */
1585    if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) {
1586       ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
1587                             (struct gl_program *) ctx->VertexProgram._Current);
1588    }
1589 }
1590
1591 void _tnl_ProgramCacheInit( GLcontext *ctx )
1592 {
1593    TNLcontext *tnl = TNL_CONTEXT(ctx);
1594
1595    tnl->vp_cache = (struct tnl_cache *) MALLOC(sizeof(*tnl->vp_cache));
1596    tnl->vp_cache->size = 17;
1597    tnl->vp_cache->n_items = 0;
1598    tnl->vp_cache->items = (struct tnl_cache_item**)
1599       _mesa_calloc(tnl->vp_cache->size * sizeof(*tnl->vp_cache->items));
1600 }
1601
1602 void _tnl_ProgramCacheDestroy( GLcontext *ctx )
1603 {
1604    TNLcontext *tnl = TNL_CONTEXT(ctx);
1605    struct tnl_cache_item *c, *next;
1606    GLuint i;
1607
1608    for (i = 0; i < tnl->vp_cache->size; i++)
1609       for (c = tnl->vp_cache->items[i]; c; c = next) {
1610          next = c->next;
1611          FREE(c->key);
1612          FREE(c->data);
1613          FREE(c);
1614       }
1615
1616    FREE(tnl->vp_cache->items);
1617    FREE(tnl->vp_cache);
1618 }