src/mesa/drivers/dri/i965/brw_vs_tnl.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.3
   4  *
   5  * Copyright (C) 2005  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/prog_parameter.h"
  37 #include "shader/prog_print.h"
  38 #include "brw_vs.h"
  39 #include "brw_state.h"
  40
  41
  42 struct state_key {
  43    unsigned light_global_enabled:1;
  44    unsigned light_local_viewer:1;
  45    unsigned light_twoside:1;
  46    unsigned light_color_material:1;
  47    unsigned light_color_material_mask:12;
  48    unsigned light_material_mask:12;
  49    unsigned normalize:1;
  50    unsigned rescale_normals:1;
  51    unsigned fog_source_is_depth:1;
  52    unsigned tnl_do_vertex_fog:1;
  53    unsigned separate_specular:1;
  54    unsigned fog_option:2;
  55    unsigned point_attenuated:1;
  56    unsigned texture_enabled_global:1;
  57    unsigned fragprog_inputs_read:12;
  58
  59    struct {
  60       unsigned light_enabled:1;
  61       unsigned light_eyepos3_is_zero:1;
  62       unsigned light_spotcutoff_is_180:1;
  63       unsigned light_attenuated:1;
  64       unsigned texunit_really_enabled:1;
  65       unsigned texmat_enabled:1;
  66       unsigned texgen_enabled:4;
  67       unsigned texgen_mode0:4;
  68       unsigned texgen_mode1:4;
  69       unsigned texgen_mode2:4;
  70       unsigned texgen_mode3:4;
  71    } unit[8];
  72 };
  73
  74
  75
  76 #define FOG_NONE   0
  77 #define FOG_LINEAR 1
  78 #define FOG_EXP    2
  79 #define FOG_EXP2   3
  80
  81 static GLuint translate_fog_mode( GLenum mode )
  82 {
  83    switch (mode) {
  84    case GL_LINEAR: return FOG_LINEAR;
  85    case GL_EXP: return FOG_EXP;
  86    case GL_EXP2: return FOG_EXP2;
  87    default: return FOG_NONE;
  88    }
  89 }
  90
  91 #define TXG_NONE           0
  92 #define TXG_OBJ_LINEAR     1
  93 #define TXG_EYE_LINEAR     2
  94 #define TXG_SPHERE_MAP     3
  95 #define TXG_REFLECTION_MAP 4
  96 #define TXG_NORMAL_MAP     5
  97
  98 static GLuint translate_texgen( GLboolean enabled, GLenum mode )
  99 {
 100    if (!enabled)
 101       return TXG_NONE;
 102
 103    switch (mode) {
 104    case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
 105    case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
 106    case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
 107    case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
 108    case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
 109    default: return TXG_NONE;
 110    }
 111 }
 112
 113 static void make_state_key( GLcontext *ctx, struct state_key *key )
 114 {
 115    struct brw_context *brw = brw_context(ctx);
 116    const struct gl_fragment_program *fp = brw->fragment_program;
 117    GLuint i;
 118
 119    /* This now relies on texenvprogram.c being active:
 120     */
 121    assert(fp);
 122
 123    memset(key, 0, sizeof(*key));
 124
 125    /* BRW_NEW_FRAGMENT_PROGRAM */
 126    key->fragprog_inputs_read = fp->Base.InputsRead;
 127
 128    /* _NEW_LIGHT */
 129    key->separate_specular = (brw->attribs.Light->Model.ColorControl ==
 130                              GL_SEPARATE_SPECULAR_COLOR);
 131
 132    /* _NEW_LIGHT */
 133    if (brw->attribs.Light->Enabled) {
 134       key->light_global_enabled = 1;
 135
 136       if (brw->attribs.Light->Model.LocalViewer)
 137          key->light_local_viewer = 1;
 138
 139       if (brw->attribs.Light->Model.TwoSide)
 140          key->light_twoside = 1;
 141
 142       if (brw->attribs.Light->ColorMaterialEnabled) {
 143          key->light_color_material = 1;
 144          key->light_color_material_mask = brw->attribs.Light->ColorMaterialBitmask;
 145       }
 146
 147       /* BRW_NEW_INPUT_VARYING */
 148
 149       /* For these programs, material values are stuffed into the
 150        * generic slots:
 151        */
 152       for (i = 0 ; i < MAT_ATTRIB_MAX ; i++)
 153          if (brw->vb.info.varying & (1<<(VERT_ATTRIB_GENERIC0 + i)))
 154             key->light_material_mask |= 1<<i;
 155
 156       for (i = 0; i < MAX_LIGHTS; i++) {
 157          struct gl_light *light = &brw->attribs.Light->Light[i];
 158
 159          if (light->Enabled) {
 160             key->unit[i].light_enabled = 1;
 161
 162             if (light->EyePosition[3] == 0.0)
 163                key->unit[i].light_eyepos3_is_zero = 1;
 164
 165             if (light->SpotCutoff == 180.0)
 166                key->unit[i].light_spotcutoff_is_180 = 1;
 167
 168             if (light->ConstantAttenuation != 1.0 ||
 169                 light->LinearAttenuation != 0.0 ||
 170                 light->QuadraticAttenuation != 0.0)
 171                key->unit[i].light_attenuated = 1;
 172          }
 173       }
 174    }
 175
 176    /* _NEW_TRANSFORM */
 177    if (brw->attribs.Transform->Normalize)
 178       key->normalize = 1;
 179
 180    if (brw->attribs.Transform->RescaleNormals)
 181       key->rescale_normals = 1;
 182
 183    /* BRW_NEW_FRAGMENT_PROGRAM */
 184    key->fog_option = translate_fog_mode(fp->FogOption);
 185    if (key->fog_option)
 186       key->fragprog_inputs_read |= FRAG_BIT_FOGC;
 187
 188    /* _NEW_FOG */
 189    if (brw->attribs.Fog->FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT)
 190       key->fog_source_is_depth = 1;
 191
 192    /* _NEW_HINT, ??? */
 193    if (1)
 194       key->tnl_do_vertex_fog = 1;
 195
 196    /* _NEW_POINT */
 197    if (brw->attribs.Point->_Attenuated)
 198       key->point_attenuated = 1;
 199
 200    /* _NEW_TEXTURE */
 201    if (brw->attribs.Texture->_TexGenEnabled ||
 202        brw->attribs.Texture->_TexMatEnabled ||
 203        brw->attribs.Texture->_EnabledUnits)
 204       key->texture_enabled_global = 1;
 205
 206    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 207       struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[i];
 208
 209       if (texUnit->_ReallyEnabled)
 210          key->unit[i].texunit_really_enabled = 1;
 211
 212       if (brw->attribs.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
 234
 235
 236 /* Very useful debugging tool - produces annotated listing of
 237  * generated program with line/function references for each
 238  * instruction back into this file:
 239  */
 240 #define DISASSEM 0
 241
 242 /* Should be tunable by the driver - do we want to do matrix
 243  * multiplications with DP4's or with MUL/MAD's?  SSE works better
 244  * with the latter, drivers may differ.
 245  */
 246 #define PREFER_DP4 1
 247
 248
 249 /* Use uregs to represent registers internally, translate to Mesa's
 250  * expected formats on emit.
 251  *
 252  * NOTE: These are passed by value extensively in this file rather
 253  * than as usual by pointer reference.  If this disturbs you, try
 254  * remembering they are just 32bits in size.
 255  *
 256  * GCC is smart enough to deal with these dword-sized structures in
 257  * much the same way as if I had defined them as dwords and was using
 258  * macros to access and set the fields.  This is much nicer and easier
 259  * to evolve.
 260  */
 261 struct ureg {
 262    GLuint file:4;
 263    GLint idx:8;      /* relative addressing may be negative */
 264    GLuint negate:1;
 265    GLuint swz:12;
 266    GLuint pad:7;
 267 };
 268
 269
 270 struct tnl_program {
 271    const struct state_key *state;
 272    struct gl_vertex_program *program;
 273
 274    GLuint nr_instructions;
 275    GLuint temp_in_use;
 276    GLuint temp_reserved;
 277
 278    struct ureg eye_position;
 279    struct ureg eye_position_normalized;
 280    struct ureg eye_normal;
 281    struct ureg identity;
 282
 283    GLuint materials;
 284    GLuint color_materials;
 285 };
 286
 287
 288 const static struct ureg undef = {
 289    PROGRAM_UNDEFINED,
 290    ~0,
 291    0,
 292    0,
 293    0
 294 };
 295
 296 /* Local shorthand:
 297  */
 298 #define X    SWIZZLE_X
 299 #define Y    SWIZZLE_Y
 300 #define Z    SWIZZLE_Z
 301 #define W    SWIZZLE_W
 302
 303
 304 /* Construct a ureg:
 305  */
 306 static struct ureg make_ureg(GLuint file, GLint idx)
 307 {
 308    struct ureg reg;
 309    reg.file = file;
 310    reg.idx = idx;
 311    reg.negate = 0;
 312    reg.swz = SWIZZLE_NOOP;
 313    reg.pad = 0;
 314    return reg;
 315 }
 316
 317
 318
 319 static struct ureg ureg_negate( struct ureg reg )
 320 {
 321    reg.negate ^= 1;
 322    return reg;
 323 }
 324
 325
 326 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 327 {
 328    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 329                            GET_SWZ(reg.swz, y),
 330                            GET_SWZ(reg.swz, z),
 331                            GET_SWZ(reg.swz, w));
 332
 333    return reg;
 334 }
 335
 336 static struct ureg swizzle1( struct ureg reg, int x )
 337 {
 338    return swizzle(reg, x, x, x, x);
 339 }
 340
 341 static struct ureg get_temp( struct tnl_program *p )
 342 {
 343    int bit = ffs( ~p->temp_in_use );
 344    if (!bit) {
 345       fprintf(stderr, "%s: out of temporaries\n", __FILE__);
 346       assert(0);
 347    }
 348
 349    if (bit > p->program->Base.NumTemporaries)
 350       p->program->Base.NumTemporaries = bit;
 351
 352    p->temp_in_use |= 1<<(bit-1);
 353    return make_ureg(PROGRAM_TEMPORARY, bit-1);
 354 }
 355
 356 static struct ureg reserve_temp( struct tnl_program *p )
 357 {
 358    struct ureg temp = get_temp( p );
 359    p->temp_reserved |= 1<<temp.idx;
 360    return temp;
 361 }
 362
 363 static void release_temp( struct tnl_program *p, struct ureg reg )
 364 {
 365    if (reg.file == PROGRAM_TEMPORARY) {
 366       p->temp_in_use &= ~(1<<reg.idx);
 367       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
 368    }
 369 }
 370
 371 static void release_temps( struct tnl_program *p )
 372 {
 373    p->temp_in_use = p->temp_reserved;
 374 }
 375
 376
 377
 378 static struct ureg register_input( struct tnl_program *p, GLuint input )
 379 {
 380    assert(input < 32);
 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    /* XXX what about swizzle? */
 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, /* matrix name */
 456                                     GLint s1, /* texture matrix number */
 457                                     GLint s2, /* first row */
 458                                     GLint s3, /* last row */
 459                                     GLint s4, /* modifier */
 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->RelAddr = 0;
 479    src->NegateBase = reg.negate;
 480    src->Abs = 0;
 481    src->NegateAbs = 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 = 0;
 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
 526    if (nr >= p->nr_instructions) {
 527       p->program->Base.Instructions =
 528          _mesa_realloc(p->program->Base.Instructions,
 529                        sizeof(struct prog_instruction) * p->nr_instructions,
 530                        sizeof(struct prog_instruction) * (p->nr_instructions *= 2));
 531    }
 532
 533    {
 534       struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 535       memset(inst, 0, sizeof(*inst));
 536       inst->Opcode = op;
 537       inst->StringPos = 0;
 538       inst->Data = 0;
 539
 540       emit_arg( &inst->SrcReg[0], src0 );
 541       emit_arg( &inst->SrcReg[1], src1 );
 542       emit_arg( &inst->SrcReg[2], src2 );
 543
 544       emit_dst( &inst->DstReg, dest, mask );
 545
 546       debug_insn(inst, fn, line);
 547    }
 548 }
 549
 550
 551
 552 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 553    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 554
 555 #define emit_op2(p, op, dst, mask, src0, src1) \
 556     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 557
 558 #define emit_op1(p, op, dst, mask, src0) \
 559     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 560
 561
 562 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 563 {
 564    if (reg.file == PROGRAM_TEMPORARY &&
 565        !(p->temp_reserved & (1<<reg.idx)))
 566       return reg;
 567    else {
 568       struct ureg temp = get_temp(p);
 569       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 570       return temp;
 571    }
 572 }
 573
 574
 575 /* Currently no tracking performed of input/output/register size or
 576  * active elements.  Could be used to reduce these operations, as
 577  * could the matrix type.
 578  */
 579 static void emit_matrix_transform_vec4( struct tnl_program *p,
 580                                         struct ureg dest,
 581                                         const struct ureg *mat,
 582                                         struct ureg src)
 583 {
 584    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 585    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 586    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 587    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 588 }
 589
 590 /* This version is much easier to implement if writemasks are not
 591  * supported natively on the target or (like SSE), the target doesn't
 592  * have a clean/obvious dotproduct implementation.
 593  */
 594 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 595                                                   struct ureg dest,
 596                                                   const struct ureg *mat,
 597                                                   struct ureg src)
 598 {
 599    struct ureg tmp;
 600
 601    if (dest.file != PROGRAM_TEMPORARY)
 602       tmp = get_temp(p);
 603    else
 604       tmp = dest;
 605
 606    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 607    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 608    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 609    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 610
 611    if (dest.file != PROGRAM_TEMPORARY)
 612       release_temp(p, tmp);
 613 }
 614
 615 static void emit_matrix_transform_vec3( struct tnl_program *p,
 616                                         struct ureg dest,
 617                                         const struct ureg *mat,
 618                                         struct ureg src)
 619 {
 620    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 621    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 622    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 623 }
 624
 625
 626 static void emit_normalize_vec3( struct tnl_program *p,
 627                                  struct ureg dest,
 628                                  struct ureg src )
 629 {
 630    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_W, src, src);
 631    emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_W, swizzle1(dest,W));
 632    emit_op2(p, OPCODE_MUL, dest, WRITEMASK_XYZ, src, swizzle1(dest,W));
 633 }
 634
 635 static void emit_passthrough( struct tnl_program *p,
 636                               GLuint input,
 637                               GLuint output )
 638 {
 639    struct ureg out = register_output(p, output);
 640    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 641 }
 642
 643 static struct ureg get_eye_position( struct tnl_program *p )
 644 {
 645    if (is_undef(p->eye_position)) {
 646       struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 647       struct ureg modelview[4];
 648
 649       p->eye_position = reserve_temp(p);
 650
 651       if (PREFER_DP4) {
 652          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 653                                  0, modelview );
 654
 655          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 656       }
 657       else {
 658          register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 659                                  STATE_MATRIX_TRANSPOSE, modelview );
 660
 661          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 662       }
 663    }
 664
 665    return p->eye_position;
 666 }
 667
 668
 669 #if 0
 670 static struct ureg get_eye_z( struct tnl_program *p )
 671 {
 672    if (!is_undef(p->eye_position)) {
 673       return swizzle1(p->eye_position, Z);
 674    }
 675    else if (!is_undef(p->eye_z)) {
 676       struct ureg pos = register_input( p, BRW_ATTRIB_POS );
 677       struct ureg modelview2;
 678
 679       p->eye_z = reserve_temp(p);
 680
 681       register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 2, 1,
 682                               STATE_MATRIX, &modelview2 );
 683
 684       emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 685       emit_op2(p, OPCODE_DP4, p->eye_z, WRITEMASK_Z, pos, modelview2);
 686    }
 687
 688    return swizzle1(p->eye_z, Z)
 689 }
 690 #endif
 691
 692
 693
 694 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 695 {
 696    if (is_undef(p->eye_position_normalized)) {
 697       struct ureg eye = get_eye_position(p);
 698       p->eye_position_normalized = reserve_temp(p);
 699       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 700    }
 701
 702    return p->eye_position_normalized;
 703 }
 704
 705
 706 static struct ureg get_eye_normal( struct tnl_program *p )
 707 {
 708    if (is_undef(p->eye_normal)) {
 709       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
 710       struct ureg mvinv[3];
 711
 712       register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
 713                               STATE_MATRIX_INVTRANS, mvinv );
 714
 715       p->eye_normal = reserve_temp(p);
 716
 717       /* Transform to eye space:
 718        */
 719       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 720
 721       /* Normalize/Rescale:
 722        */
 723       if (p->state->normalize) {
 724          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 725       }
 726       else if (p->state->rescale_normals) {
 727          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 728                                                STATE_NORMAL_SCALE);
 729
 730          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal,
 731                    swizzle1(rescale, X));
 732       }
 733    }
 734
 735    return p->eye_normal;
 736 }
 737
 738
 739
 740 static void build_hpos( struct tnl_program *p )
 741 {
 742    struct ureg pos = register_input( p, VERT_ATTRIB_POS );
 743    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 744    struct ureg mvp[4];
 745
 746    if (PREFER_DP4) {
 747       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 748                               0, mvp );
 749       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 750    }
 751    else {
 752       register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3,
 753                               STATE_MATRIX_TRANSPOSE, mvp );
 754       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 755    }
 756 }
 757
 758
 759 static GLuint material_attrib( GLuint side, GLuint property )
 760 {
 761    return (property - STATE_AMBIENT) * 2 + side;
 762 }
 763
 764 /* Get a bitmask of which material values vary on a per-vertex basis.
 765  */
 766 static void set_material_flags( struct tnl_program *p )
 767 {
 768    p->color_materials = 0;
 769    p->materials = 0;
 770
 771    if (p->state->light_color_material) {
 772       p->materials =
 773          p->color_materials = p->state->light_color_material_mask;
 774    }
 775
 776    p->materials |= p->state->light_material_mask;
 777 }
 778
 779
 780 static struct ureg get_material( struct tnl_program *p, GLuint side,
 781                                  GLuint property )
 782 {
 783    GLuint attrib = material_attrib(side, property);
 784
 785    if (p->color_materials & (1<<attrib))
 786       return register_input(p, VERT_ATTRIB_COLOR0);
 787    else if (p->materials & (1<<attrib))
 788       return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT );
 789    else
 790       return register_param3( p, STATE_MATERIAL, side, property );
 791 }
 792
 793 #define SCENE_COLOR_BITS(side) ((MAT_BIT_FRONT_EMISSION | \
 794                                  MAT_BIT_FRONT_AMBIENT | \
 795                                  MAT_BIT_FRONT_DIFFUSE) << (side))
 796
 797 /* Either return a precalculated constant value or emit code to
 798  * calculate these values dynamically in the case where material calls
 799  * are present between begin/end pairs.
 800  *
 801  * Probably want to shift this to the program compilation phase - if
 802  * we always emitted the calculation here, a smart compiler could
 803  * detect that it was constant (given a certain set of inputs), and
 804  * lift it out of the main loop.  That way the programs created here
 805  * would be independent of the vertex_buffer details.
 806  */
 807 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 808 {
 809    if (p->materials & SCENE_COLOR_BITS(side)) {
 810       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 811       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 812       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 813       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 814       struct ureg tmp = make_temp(p, material_diffuse);
 815       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 816                material_ambient, material_emission);
 817       return tmp;
 818    }
 819    else
 820       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 821 }
 822
 823
 824 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 825                                   GLuint side, GLuint property )
 826 {
 827    GLuint attrib = material_attrib(side, property);
 828    if (p->materials & (1<<attrib)) {
 829       struct ureg light_value =
 830          register_param3(p, STATE_LIGHT, light, property);
 831       struct ureg material_value = get_material(p, side, property);
 832       struct ureg tmp = get_temp(p);
 833       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 834       return tmp;
 835    }
 836    else
 837       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 838 }
 839
 840 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 841                                                 GLuint i,
 842                                                 struct ureg VPpli,
 843                                                 struct ureg dist )
 844 {
 845    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 846                                              STATE_ATTENUATION);
 847    struct ureg att = get_temp(p);
 848
 849    /* Calculate spot attenuation:
 850     */
 851    if (!p->state->unit[i].light_spotcutoff_is_180) {
 852       struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL,
 853                                                   STATE_SPOT_DIR_NORMALIZED, i);
 854       struct ureg spot = get_temp(p);
 855       struct ureg slt = get_temp(p);
 856
 857       emit_op2(p, OPCODE_DP3, spot, 0, ureg_negate(VPpli), spot_dir_norm);
 858       emit_op2(p, OPCODE_SLT, slt, 0, spot, swizzle1(spot_dir_norm,W));
 859       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 860       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 861
 862       release_temp(p, spot);
 863       release_temp(p, slt);
 864    }
 865
 866    /* Calculate distance attenuation:
 867     */
 868    if (p->state->unit[i].light_attenuated) {
 869
 870       /* 1/d,d,d,1/d */
 871       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 872       /* 1,d,d*d,1/d */
 873       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 874       /* 1/dist-atten */
 875       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 876
 877       if (!p->state->unit[i].light_spotcutoff_is_180) {
 878          /* dist-atten */
 879          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 880          /* spot-atten * dist-atten */
 881          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 882       } else {
 883          /* dist-atten */
 884          emit_op1(p, OPCODE_RCP, att, 0, dist);
 885       }
 886    }
 887
 888    return att;
 889 }
 890
 891
 892
 893
 894
 895 /* Need to add some addtional parameters to allow lighting in object
 896  * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
 897  * space lighting.
 898  */
 899 static void build_lighting( struct tnl_program *p )
 900 {
 901    const GLboolean twoside = p->state->light_twoside;
 902    const GLboolean separate = p->state->separate_specular;
 903    GLuint nr_lights = 0, count = 0;
 904    struct ureg normal = get_eye_normal(p);
 905    struct ureg lit = get_temp(p);
 906    struct ureg dots = get_temp(p);
 907    struct ureg _col0 = undef, _col1 = undef;
 908    struct ureg _bfc0 = undef, _bfc1 = undef;
 909    GLuint i;
 910
 911    for (i = 0; i < MAX_LIGHTS; i++)
 912       if (p->state->unit[i].light_enabled)
 913          nr_lights++;
 914
 915    set_material_flags(p);
 916
 917    {
 918       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 919       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 920       release_temp(p, shininess);
 921
 922       _col0 = make_temp(p, get_scenecolor(p, 0));
 923       if (separate)
 924          _col1 = make_temp(p, get_identity_param(p));
 925       else
 926          _col1 = _col0;
 927
 928    }
 929
 930    if (twoside) {
 931       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 932       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 933                ureg_negate(swizzle1(shininess,X)));
 934       release_temp(p, shininess);
 935
 936       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 937       if (separate)
 938          _bfc1 = make_temp(p, get_identity_param(p));
 939       else
 940          _bfc1 = _bfc0;
 941    }
 942
 943
 944    /* If no lights, still need to emit the scenecolor.
 945     */
 946    /* KW: changed to do this always - v1.17 "Fix lighting alpha result"?
 947     */
 948    if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
 949    {
 950       struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 951       emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 952
 953       if (twoside) {
 954          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 955          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 956       }
 957    }
 958
 959    if (separate && (p->state->fragprog_inputs_read & FRAG_BIT_COL1)) {
 960
 961       struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 962       emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 963
 964       if (twoside) {
 965          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 966          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 967       }
 968    }
 969
 970    if (nr_lights == 0) {
 971       release_temps(p);
 972       return;
 973    }
 974
 975
 976    for (i = 0; i < MAX_LIGHTS; i++) {
 977       if (p->state->unit[i].light_enabled) {
 978          struct ureg half = undef;
 979          struct ureg att = undef, VPpli = undef;
 980
 981          count++;
 982
 983          if (p->state->unit[i].light_eyepos3_is_zero) {
 984             /* Can used precomputed constants in this case.
 985              * Attenuation never applies to infinite lights.
 986              */
 987             VPpli = register_param3(p, STATE_LIGHT, i,
 988                                     STATE_POSITION_NORMALIZED);
 989             if (p->state->light_local_viewer) {
 990                 struct ureg eye_hat = get_eye_position_normalized(p);
 991                 half = get_temp(p);
 992                 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
 993                 emit_normalize_vec3(p, half, half);
 994             } else {
 995                 half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
 996             }
 997          }
 998          else {
 999             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
1000                                                STATE_POSITION);
1001             struct ureg V = get_eye_position(p);
1002             struct ureg dist = get_temp(p);
1003
1004             VPpli = get_temp(p);
1005             half = get_temp(p);
1006
1007             /* Calulate VPpli vector
1008              */
1009             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
1010
1011             /* Normalize VPpli.  The dist value also used in
1012              * attenuation below.
1013              */
1014             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
1015             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
1016             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
1017
1018
1019             /* Calculate  attenuation:
1020              */
1021             if (!p->state->unit[i].light_spotcutoff_is_180 ||
1022                 p->state->unit[i].light_attenuated) {
1023                att = calculate_light_attenuation(p, i, VPpli, dist);
1024             }
1025
1026
1027             /* Calculate viewer direction, or use infinite viewer:
1028              */
1029             if (p->state->light_local_viewer) {
1030                struct ureg eye_hat = get_eye_position_normalized(p);
1031                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1032             }
1033             else {
1034                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1035                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1036             }
1037
1038             emit_normalize_vec3(p, half, half);
1039
1040             release_temp(p, dist);
1041          }
1042
1043          /* Calculate dot products:
1044           */
1045          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1046          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1047
1048
1049          /* Front face lighting:
1050           */
1051          {
1052             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1053             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1054             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1055             struct ureg res0, res1;
1056             GLuint mask0, mask1;
1057
1058             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1059
1060             if (!is_undef(att))
1061                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1062
1063
1064             mask0 = 0;
1065             mask1 = 0;
1066             res0 = _col0;
1067             res1 = _col1;
1068
1069             if (count == nr_lights) {
1070                if (separate) {
1071                   mask0 = WRITEMASK_XYZ;
1072                   mask1 = WRITEMASK_XYZ;
1073
1074                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1075                      res0 = register_output( p, VERT_RESULT_COL0 );
1076
1077                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1078                      res1 = register_output( p, VERT_RESULT_COL1 );
1079                }
1080                else {
1081                   mask1 = WRITEMASK_XYZ;
1082
1083                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1084                      res1 = register_output( p, VERT_RESULT_COL0 );
1085                }
1086             }
1087
1088             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1089             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1090             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1091
1092             release_temp(p, ambient);
1093             release_temp(p, diffuse);
1094             release_temp(p, specular);
1095          }
1096
1097          /* Back face lighting:
1098           */
1099          if (twoside) {
1100             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1101             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1102             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1103             struct ureg res0, res1;
1104             GLuint mask0, mask1;
1105
1106             emit_op1(p, OPCODE_LIT, lit, 0, ureg_negate(swizzle(dots,X,Y,W,Z)));
1107
1108             if (!is_undef(att))
1109                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1110
1111             mask0 = 0;
1112             mask1 = 0;
1113             res0 = _bfc0;
1114             res1 = _bfc1;
1115
1116             if (count == nr_lights) {
1117                if (separate) {
1118                   mask0 = WRITEMASK_XYZ;
1119                   mask1 = WRITEMASK_XYZ;
1120                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1121                      res0 = register_output( p, VERT_RESULT_BFC0 );
1122
1123                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1124                      res1 = register_output( p, VERT_RESULT_BFC1 );
1125                }
1126                else {
1127                   mask1 = WRITEMASK_XYZ;
1128
1129                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1130                      res1 = register_output( p, VERT_RESULT_BFC0 );
1131                }
1132             }
1133
1134             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1135             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1136             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1137
1138             release_temp(p, ambient);
1139             release_temp(p, diffuse);
1140             release_temp(p, specular);
1141          }
1142
1143          release_temp(p, half);
1144          release_temp(p, VPpli);
1145          release_temp(p, att);
1146       }
1147    }
1148
1149    release_temps( p );
1150 }
1151
1152
1153 static void build_fog( struct tnl_program *p )
1154 {
1155    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1156    struct ureg input;
1157    GLuint useabs = p->state->fog_source_is_depth && p->state->fog_option &&
1158                    (p->state->fog_option != FOG_EXP2);
1159
1160    if (p->state->fog_source_is_depth) {
1161       input = swizzle1(get_eye_position(p), Z);
1162    }
1163    else {
1164       input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X);
1165    }
1166
1167    if (p->state->fog_option &&
1168        p->state->tnl_do_vertex_fog) {
1169       struct ureg params = register_param2(p, STATE_INTERNAL,
1170                                            STATE_FOG_PARAMS_OPTIMIZED);
1171       struct ureg tmp = get_temp(p);
1172       struct ureg id = get_identity_param(p);
1173
1174       emit_op1(p, OPCODE_MOV, fog, 0, id);
1175
1176       if (useabs) {
1177          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1178       }
1179
1180       switch (p->state->fog_option) {
1181       case FOG_LINEAR: {
1182          emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input,
1183                         swizzle1(params,X), swizzle1(params,Y));
1184          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1185          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1186          break;
1187       }
1188       case FOG_EXP:
1189          emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input,
1190                         swizzle1(params,Z));
1191          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
1192          break;
1193       case FOG_EXP2:
1194          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W));
1195          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1196          emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp));
1197          break;
1198       }
1199
1200       release_temp(p, tmp);
1201    }
1202    else {
1203       /* results = incoming fog coords (compute fog per-fragment later)
1204        *
1205        * KW:  Is it really necessary to do anything in this case?
1206        */
1207       emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, 0, input);
1208    }
1209 }
1210
1211 static void build_reflect_texgen( struct tnl_program *p,
1212                                   struct ureg dest,
1213                                   GLuint writemask )
1214 {
1215    struct ureg normal = get_eye_normal(p);
1216    struct ureg eye_hat = get_eye_position_normalized(p);
1217    struct ureg tmp = get_temp(p);
1218
1219    /* n.u */
1220    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1221    /* 2n.u */
1222    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1223    /* (-2n.u)n + u */
1224    emit_op3(p, OPCODE_MAD, dest, writemask, ureg_negate(tmp), normal, eye_hat);
1225
1226    release_temp(p, tmp);
1227 }
1228
1229 static void build_sphere_texgen( struct tnl_program *p,
1230                                  struct ureg dest,
1231                                  GLuint writemask )
1232 {
1233    struct ureg normal = get_eye_normal(p);
1234    struct ureg eye_hat = get_eye_position_normalized(p);
1235    struct ureg tmp = get_temp(p);
1236    struct ureg half = register_scalar_const(p, .5);
1237    struct ureg r = get_temp(p);
1238    struct ureg inv_m = get_temp(p);
1239    struct ureg id = get_identity_param(p);
1240
1241    /* Could share the above calculations, but it would be
1242     * a fairly odd state for someone to set (both sphere and
1243     * reflection active for different texture coordinate
1244     * components.  Of course - if two texture units enable
1245     * reflect and/or sphere, things start to tilt in favour
1246     * of seperating this out:
1247     */
1248
1249    /* n.u */
1250    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1251    /* 2n.u */
1252    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1253    /* (-2n.u)n + u */
1254    emit_op3(p, OPCODE_MAD, r, 0, ureg_negate(tmp), normal, eye_hat);
1255    /* r + 0,0,1 */
1256    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1257    /* rx^2 + ry^2 + (rz+1)^2 */
1258    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1259    /* 2/m */
1260    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1261    /* 1/m */
1262    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1263    /* r/m + 1/2 */
1264    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1265
1266    release_temp(p, tmp);
1267    release_temp(p, r);
1268    release_temp(p, inv_m);
1269 }
1270
1271
1272 static void build_texture_transform( struct tnl_program *p )
1273 {
1274    GLuint i, j;
1275
1276    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1277
1278       if (!(p->state->fragprog_inputs_read & (FRAG_BIT_TEX0<<i)))
1279          continue;
1280
1281       if (p->state->unit[i].texgen_enabled ||
1282           p->state->unit[i].texmat_enabled) {
1283
1284          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1285          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1286          struct ureg out_texgen = undef;
1287
1288          if (p->state->unit[i].texgen_enabled) {
1289             GLuint copy_mask = 0;
1290             GLuint sphere_mask = 0;
1291             GLuint reflect_mask = 0;
1292             GLuint normal_mask = 0;
1293             GLuint modes[4];
1294
1295             if (texmat_enabled)
1296                out_texgen = get_temp(p);
1297             else
1298                out_texgen = out;
1299
1300             modes[0] = p->state->unit[i].texgen_mode0;
1301             modes[1] = p->state->unit[i].texgen_mode1;
1302             modes[2] = p->state->unit[i].texgen_mode2;
1303             modes[3] = p->state->unit[i].texgen_mode3;
1304
1305             for (j = 0; j < 4; j++) {
1306                switch (modes[j]) {
1307                case TXG_OBJ_LINEAR: {
1308                   struct ureg obj = register_input(p, VERT_ATTRIB_POS);
1309                   struct ureg plane =
1310                      register_param3(p, STATE_TEXGEN, i,
1311                                      STATE_TEXGEN_OBJECT_S + j);
1312
1313                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1314                            obj, plane );
1315                   break;
1316                }
1317                case TXG_EYE_LINEAR: {
1318                   struct ureg eye = get_eye_position(p);
1319                   struct ureg plane =
1320                      register_param3(p, STATE_TEXGEN, i,
1321                                      STATE_TEXGEN_EYE_S + j);
1322
1323                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1324                            eye, plane );
1325                   break;
1326                }
1327                case TXG_SPHERE_MAP:
1328                   sphere_mask |= WRITEMASK_X << j;
1329                   break;
1330                case TXG_REFLECTION_MAP:
1331                   reflect_mask |= WRITEMASK_X << j;
1332                   break;
1333                case TXG_NORMAL_MAP:
1334                   normal_mask |= WRITEMASK_X << j;
1335                   break;
1336                case TXG_NONE:
1337                   copy_mask |= WRITEMASK_X << j;
1338                }
1339
1340             }
1341
1342
1343             if (sphere_mask) {
1344                build_sphere_texgen(p, out_texgen, sphere_mask);
1345             }
1346
1347             if (reflect_mask) {
1348                build_reflect_texgen(p, out_texgen, reflect_mask);
1349             }
1350
1351             if (normal_mask) {
1352                struct ureg normal = get_eye_normal(p);
1353                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1354             }
1355
1356             if (copy_mask) {
1357                struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i);
1358                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1359             }
1360          }
1361
1362          if (texmat_enabled) {
1363             struct ureg texmat[4];
1364             struct ureg in = (!is_undef(out_texgen) ?
1365                               out_texgen :
1366                               register_input(p, VERT_ATTRIB_TEX0+i));
1367             if (PREFER_DP4) {
1368                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1369                                        0, texmat );
1370                emit_matrix_transform_vec4( p, out, texmat, in );
1371             }
1372             else {
1373                register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
1374                                        STATE_MATRIX_TRANSPOSE, texmat );
1375                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1376             }
1377          }
1378
1379          release_temps(p);
1380       }
1381       else {
1382          emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1383       }
1384    }
1385 }
1386
1387
1388 /* Seems like it could be tighter:
1389  */
1390 static void build_pointsize( struct tnl_program *p )
1391 {
1392    struct ureg eye = get_eye_position(p);
1393    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1394    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1395    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1396    struct ureg ut = get_temp(p);
1397
1398    /* 1, Z, Z * Z, 1 */
1399    emit_op1(p, OPCODE_MOV, ut, WRITEMASK_XW, swizzle1(get_identity_param(p), W));
1400    emit_op1(p, OPCODE_ABS, ut, WRITEMASK_YZ, swizzle1(eye, Z));
1401    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_Z, ut, ut);
1402
1403
1404    /* p1 +  p2 * dist + p3 * dist * dist, 0 */
1405    emit_op2(p, OPCODE_DP3, ut, WRITEMASK_X, ut, state_attenuation);
1406
1407    /* 1 / sqrt(factor) */
1408    emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1409
1410    /* ut = pointSize / factor */
1411    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1412
1413    /* Clamp to min/max - state_size.[yz]
1414     */
1415    emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1416    emit_op2(p, OPCODE_MIN, out, 0, swizzle1(ut, X), swizzle1(state_size, Z));
1417
1418    release_temp(p, ut);
1419 }
1420
1421 static void build_tnl_program( struct tnl_program *p )
1422 {
1423    /* Emit the program, starting with modelviewproject:
1424     */
1425    build_hpos(p);
1426
1427    /* Lighting calculations:
1428     */
1429    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1430       if (p->state->light_global_enabled)
1431          build_lighting(p);
1432       else {
1433          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1434             emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0);
1435
1436          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1437             emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1);
1438       }
1439    }
1440
1441    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1442        p->state->fog_option != FOG_NONE)
1443       build_fog(p);
1444
1445    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1446       build_texture_transform(p);
1447
1448    if (p->state->point_attenuated)
1449       build_pointsize(p);
1450
1451    /* Finish up:
1452     */
1453    emit_op1(p, OPCODE_END, undef, 0, undef);
1454
1455    /* Disassemble:
1456     */
1457    if (DISASSEM) {
1458       _mesa_printf ("\n");
1459    }
1460 }
1461
1462
1463 static void build_new_tnl_program( const struct state_key *key,
1464                                    struct gl_vertex_program *program,
1465                                    GLuint max_temps)
1466 {
1467    struct tnl_program p;
1468
1469    _mesa_memset(&p, 0, sizeof(p));
1470    p.state = key;
1471    p.program = program;
1472    p.eye_position = undef;
1473    p.eye_position_normalized = undef;
1474    p.eye_normal = undef;
1475    p.identity = undef;
1476    p.temp_in_use = 0;
1477    p.nr_instructions = 16;
1478
1479    if (max_temps >= sizeof(int) * 8)
1480       p.temp_reserved = 0;
1481    else
1482       p.temp_reserved = ~((1<<max_temps)-1);
1483
1484    p.program->Base.Instructions =
1485       _mesa_malloc(sizeof(struct prog_instruction) * p.nr_instructions);
1486    p.program->Base.String = 0;
1487    p.program->Base.NumInstructions =
1488    p.program->Base.NumTemporaries =
1489    p.program->Base.NumParameters =
1490    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1491    p.program->Base.Parameters = _mesa_new_parameter_list();
1492    p.program->Base.InputsRead = 0;
1493    p.program->Base.OutputsWritten = 0;
1494
1495    build_tnl_program( &p );
1496 }
1497
1498 static void *search_cache( struct brw_tnl_cache *cache,
1499                            GLuint hash,
1500                            const void *key,
1501                            GLuint keysize)
1502 {
1503    struct brw_tnl_cache_item *c;
1504
1505    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1506       if (c->hash == hash && memcmp(c->key, key, keysize) == 0)
1507          return c->data;
1508    }
1509
1510    return NULL;
1511 }
1512
1513 static void rehash( struct brw_tnl_cache *cache )
1514 {
1515    struct brw_tnl_cache_item **items;
1516    struct brw_tnl_cache_item *c, *next;
1517    GLuint size, i;
1518
1519    size = cache->size * 3;
1520    items = (struct brw_tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1521    _mesa_memset(items, 0, size * sizeof(*items));
1522
1523    for (i = 0; i < cache->size; i++)
1524       for (c = cache->items[i]; c; c = next) {
1525          next = c->next;
1526          c->next = items[c->hash % size];
1527          items[c->hash % size] = c;
1528       }
1529
1530    FREE(cache->items);
1531    cache->items = items;
1532    cache->size = size;
1533 }
1534
1535 static void cache_item( struct brw_tnl_cache *cache,
1536                         GLuint hash,
1537                         const struct state_key *key,
1538                         void *data )
1539 {
1540    struct brw_tnl_cache_item *c = MALLOC(sizeof(*c));
1541    c->hash = hash;
1542
1543    c->key = malloc(sizeof(*key));
1544    memcpy(c->key, key, sizeof(*key));
1545
1546    c->data = data;
1547
1548    if (++cache->n_items > cache->size * 1.5)
1549       rehash(cache);
1550
1551    c->next = cache->items[hash % cache->size];
1552    cache->items[hash % cache->size] = c;
1553 }
1554
1555
1556 static GLuint hash_key( struct state_key *key )
1557 {
1558    GLuint *ikey = (GLuint *)key;
1559    GLuint hash = 0, i;
1560
1561    /* I'm sure this can be improved on, but speed is important:
1562     */
1563    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1564       hash += ikey[i];
1565
1566    return hash;
1567 }
1568
1569 static void update_tnl_program( struct brw_context *brw )
1570 {
1571    GLcontext *ctx = &brw->intel.ctx;
1572    struct state_key key;
1573    GLuint hash;
1574    struct gl_vertex_program *old = brw->tnl_program;
1575
1576    /* _NEW_PROGRAM */
1577    if (brw->attribs.VertexProgram->_Enabled)
1578       return;
1579
1580    /* Grab all the relevent state and put it in a single structure:
1581     */
1582    make_state_key(ctx, &key);
1583    hash = hash_key(&key);
1584
1585    /* Look for an already-prepared program for this state:
1586     */
1587    brw->tnl_program = (struct gl_vertex_program *)
1588       search_cache( &brw->tnl_program_cache, hash, &key, sizeof(key) );
1589
1590    /* OK, we'll have to build a new one:
1591     */
1592    if (!brw->tnl_program) {
1593       brw->tnl_program = (struct gl_vertex_program *)
1594          ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1595
1596       build_new_tnl_program( &key, brw->tnl_program,
1597 /*                           ctx->Const.MaxVertexProgramTemps  */
1598                              32
1599          );
1600
1601       if (ctx->Driver.ProgramStringNotify)
1602          ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1603                                           &brw->tnl_program->Base );
1604
1605       cache_item( &brw->tnl_program_cache,
1606                   hash, &key, brw->tnl_program );
1607    }
1608
1609    if (old != brw->tnl_program)
1610       brw->state.dirty.brw |= BRW_NEW_TNL_PROGRAM;
1611 }
1612
1613 /* Note: See brw_draw.c - the vertex program must not rely on
1614  * brw->primitive or brw->reduced_prim.
1615  */
1616 const struct brw_tracked_state brw_tnl_vertprog = {
1617    .dirty = {
1618       .mesa = (_NEW_PROGRAM |
1619                _NEW_LIGHT |
1620                _NEW_TRANSFORM |
1621                _NEW_FOG |
1622                _NEW_HINT |
1623                _NEW_POINT |
1624                _NEW_TEXTURE),
1625       .brw = (BRW_NEW_FRAGMENT_PROGRAM |
1626               BRW_NEW_INPUT_VARYING),
1627       .cache = 0
1628    },
1629    .update = update_tnl_program
1630 };
1631
1632
1633
1634
1635 static void update_active_vertprog( struct brw_context *brw )
1636 {
1637    const struct gl_vertex_program *prev = brw->vertex_program;
1638
1639    /* NEW_PROGRAM */
1640    if (brw->attribs.VertexProgram->_Enabled) {
1641       brw->vertex_program = brw->attribs.VertexProgram->Current;
1642    }
1643    else {
1644       /* BRW_NEW_TNL_PROGRAM */
1645       brw->vertex_program = brw->tnl_program;
1646    }
1647
1648    if (brw->vertex_program != prev)
1649       brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
1650 }
1651
1652
1653
1654 const struct brw_tracked_state brw_active_vertprog = {
1655    .dirty = {
1656       .mesa = _NEW_PROGRAM,
1657       .brw = BRW_NEW_TNL_PROGRAM,
1658       .cache = 0
1659    },
1660    .update = update_active_vertprog
1661 };
1662
1663
1664 void brw_ProgramCacheInit( GLcontext *ctx )
1665 {
1666    struct brw_context *brw = brw_context(ctx);
1667
1668    brw->tnl_program_cache.size = 17;
1669    brw->tnl_program_cache.n_items = 0;
1670    brw->tnl_program_cache.items = (struct brw_tnl_cache_item **)
1671       _mesa_calloc(brw->tnl_program_cache.size *
1672                    sizeof(struct brw_tnl_cache_item));
1673 }
1674
1675 void brw_ProgramCacheDestroy( GLcontext *ctx )
1676 {
1677    struct brw_context *brw = brw_context(ctx);
1678    struct brw_tnl_cache_item *c, *next;
1679    GLuint i;
1680
1681    for (i = 0; i < brw->tnl_program_cache.size; i++)
1682       for (c = brw->tnl_program_cache.items[i]; c; c = next) {
1683          next = c->next;
1684          FREE(c->key);
1685          FREE(c->data);
1686          FREE(c);
1687       }
1688
1689    FREE(brw->tnl_program_cache.items);
1690 }