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 "brw_vs.h"
  37 #include "brw_state.h"
  38
  39 #include "shader/program.h"
  40 #include "shader/program_instruction.h"
  41 #include "shader/arbprogparse.h"
  42
  43 struct state_key {
  44    unsigned light_global_enabled:1;
  45    unsigned light_local_viewer:1;
  46    unsigned light_twoside:1;
  47    unsigned light_color_material:1;
  48    unsigned light_color_material_mask:12;
  49    unsigned light_material_mask:12;
  50    unsigned normalize:1;
  51    unsigned rescale_normals:1;
  52    unsigned fog_source_is_depth:1;
  53    unsigned tnl_do_vertex_fog:1;
  54    unsigned separate_specular:1;
  55    unsigned fog_option:2;
  56    unsigned point_attenuated:1;
  57    unsigned texture_enabled_global:1;
  58    unsigned fragprog_inputs_read:12;
  59
  60    struct {
  61       unsigned light_enabled:1;
  62       unsigned light_eyepos3_is_zero:1;
  63       unsigned light_spotcutoff_is_180:1;
  64       unsigned light_attenuated:1;
  65       unsigned texunit_really_enabled:1;
  66       unsigned texmat_enabled:1;
  67       unsigned texgen_enabled:4;
  68       unsigned texgen_mode0:4;
  69       unsigned texgen_mode1:4;
  70       unsigned texgen_mode2:4;
  71       unsigned texgen_mode3:4;
  72    } unit[8];
  73 };
  74
  75
  76
  77 #define FOG_NONE   0
  78 #define FOG_LINEAR 1
  79 #define FOG_EXP    2
  80 #define FOG_EXP2   3
  81
  82 static GLuint translate_fog_mode( GLenum mode )
  83 {
  84    switch (mode) {
  85    case GL_LINEAR: return FOG_LINEAR;
  86    case GL_EXP: return FOG_EXP;
  87    case GL_EXP2: return FOG_EXP2;
  88    default: return FOG_NONE;
  89    }
  90 }
  91
  92 #define TXG_NONE           0
  93 #define TXG_OBJ_LINEAR     1
  94 #define TXG_EYE_LINEAR     2
  95 #define TXG_SPHERE_MAP     3
  96 #define TXG_REFLECTION_MAP 4
  97 #define TXG_NORMAL_MAP     5
  98
  99 static GLuint translate_texgen( GLboolean enabled, GLenum mode )
 100 {
 101    if (!enabled)
 102       return TXG_NONE;
 103
 104    switch (mode) {
 105    case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR;
 106    case GL_EYE_LINEAR: return TXG_EYE_LINEAR;
 107    case GL_SPHERE_MAP: return TXG_SPHERE_MAP;
 108    case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP;
 109    case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP;
 110    default: return TXG_NONE;
 111    }
 112 }
 113
 114 static void make_state_key( GLcontext *ctx, struct state_key *key )
 115 {
 116    struct brw_context *brw = brw_context(ctx);
 117    const struct gl_fragment_program *fp = brw->fragment_program;
 118    GLuint i;
 119
 120    /* This now relies on texenvprogram.c being active:
 121     */
 122    assert(fp);
 123
 124    memset(key, 0, sizeof(*key));
 125
 126    /* BRW_NEW_FRAGMENT_PROGRAM */
 127    key->fragprog_inputs_read = fp->Base.InputsRead;
 128
 129    /* _NEW_LIGHT */
 130    key->separate_specular = (brw->attribs.Light->Model.ColorControl ==
 131                              GL_SEPARATE_SPECULAR_COLOR);
 132
 133    /* _NEW_LIGHT */
 134    if (brw->attribs.Light->Enabled) {
 135       key->light_global_enabled = 1;
 136
 137       if (brw->attribs.Light->Model.LocalViewer)
 138          key->light_local_viewer = 1;
 139
 140       if (brw->attribs.Light->Model.TwoSide)
 141          key->light_twoside = 1;
 142
 143       if (brw->attribs.Light->ColorMaterialEnabled) {
 144          key->light_color_material = 1;
 145          key->light_color_material_mask = brw->attribs.Light->ColorMaterialBitmask;
 146       }
 147
 148       /* BRW_NEW_INPUT_VARYING */
 149       for (i = BRW_ATTRIB_MAT_FRONT_AMBIENT ; i < BRW_ATTRIB_MAX ; i++)
 150          if (brw->vb.info.varying & ((GLuint64EXT)1<<i))
 151             key->light_material_mask |= 1<<(i-BRW_ATTRIB_MAT_FRONT_AMBIENT);
 152
 153       for (i = 0; i < MAX_LIGHTS; i++) {
 154          struct gl_light *light = &brw->attribs.Light->Light[i];
 155
 156          if (light->Enabled) {
 157             key->unit[i].light_enabled = 1;
 158
 159             if (light->EyePosition[3] == 0.0)
 160                key->unit[i].light_eyepos3_is_zero = 1;
 161
 162             if (light->SpotCutoff == 180.0)
 163                key->unit[i].light_spotcutoff_is_180 = 1;
 164
 165             if (light->ConstantAttenuation != 1.0 ||
 166                 light->LinearAttenuation != 0.0 ||
 167                 light->QuadraticAttenuation != 0.0)
 168                key->unit[i].light_attenuated = 1;
 169          }
 170       }
 171    }
 172
 173    /* _NEW_TRANSFORM */
 174    if (brw->attribs.Transform->Normalize)
 175       key->normalize = 1;
 176
 177    if (brw->attribs.Transform->RescaleNormals)
 178       key->rescale_normals = 1;
 179
 180    /* BRW_NEW_FRAGMENT_PROGRAM */
 181    key->fog_option = translate_fog_mode(fp->FogOption);
 182    if (key->fog_option)
 183       key->fragprog_inputs_read |= FRAG_BIT_FOGC;
 184
 185    /* _NEW_FOG */
 186    if (brw->attribs.Fog->FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT)
 187       key->fog_source_is_depth = 1;
 188
 189    /* _NEW_HINT, ??? */
 190    if (1)
 191       key->tnl_do_vertex_fog = 1;
 192
 193    /* _NEW_POINT */
 194    if (brw->attribs.Point->_Attenuated)
 195       key->point_attenuated = 1;
 196
 197    /* _NEW_TEXTURE */
 198    if (brw->attribs.Texture->_TexGenEnabled ||
 199        brw->attribs.Texture->_TexMatEnabled ||
 200        brw->attribs.Texture->_EnabledUnits)
 201       key->texture_enabled_global = 1;
 202
 203    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 204       struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[i];
 205
 206       if (texUnit->_ReallyEnabled)
 207          key->unit[i].texunit_really_enabled = 1;
 208
 209       if (brw->attribs.Texture->_TexMatEnabled & ENABLE_TEXMAT(i))
 210          key->unit[i].texmat_enabled = 1;
 211
 212       if (texUnit->TexGenEnabled) {
 213          key->unit[i].texgen_enabled = 1;
 214
 215          key->unit[i].texgen_mode0 =
 216             translate_texgen( texUnit->TexGenEnabled & (1<<0),
 217                               texUnit->GenModeS );
 218          key->unit[i].texgen_mode1 =
 219             translate_texgen( texUnit->TexGenEnabled & (1<<1),
 220                               texUnit->GenModeT );
 221          key->unit[i].texgen_mode2 =
 222             translate_texgen( texUnit->TexGenEnabled & (1<<2),
 223                               texUnit->GenModeR );
 224          key->unit[i].texgen_mode3 =
 225             translate_texgen( texUnit->TexGenEnabled & (1<<3),
 226                               texUnit->GenModeQ );
 227       }
 228    }
 229 }
 230
 231
 232
 233 /* Very useful debugging tool - produces annotated listing of
 234  * generated program with line/function references for each
 235  * instruction back into this file:
 236  */
 237 #define DISASSEM 0
 238
 239 /* Should be tunable by the driver - do we want to do matrix
 240  * multiplications with DP4's or with MUL/MAD's?  SSE works better
 241  * with the latter, drivers may differ.
 242  */
 243 #define PREFER_DP4 1
 244
 245
 246 /* Use uregs to represent registers internally, translate to Mesa's
 247  * expected formats on emit.
 248  *
 249  * NOTE: These are passed by value extensively in this file rather
 250  * than as usual by pointer reference.  If this disturbs you, try
 251  * remembering they are just 32bits in size.
 252  *
 253  * GCC is smart enough to deal with these dword-sized structures in
 254  * much the same way as if I had defined them as dwords and was using
 255  * macros to access and set the fields.  This is much nicer and easier
 256  * to evolve.
 257  */
 258 struct ureg {
 259    GLuint file:4;
 260    GLint idx:8;      /* relative addressing may be negative */
 261    GLuint negate:1;
 262    GLuint swz:12;
 263    GLuint pad:7;
 264 };
 265
 266
 267 struct tnl_program {
 268    const struct state_key *state;
 269    struct gl_vertex_program *program;
 270
 271    GLuint nr_instructions;
 272    GLuint temp_in_use;
 273    GLuint temp_reserved;
 274
 275    struct ureg eye_position;
 276    struct ureg eye_position_normalized;
 277    struct ureg eye_normal;
 278    struct ureg identity;
 279
 280    GLuint materials;
 281    GLuint color_materials;
 282 };
 283
 284
 285 const static struct ureg undef = {
 286    PROGRAM_UNDEFINED,
 287    ~0,
 288    0,
 289    0,
 290    0
 291 };
 292
 293 /* Local shorthand:
 294  */
 295 #define X    SWIZZLE_X
 296 #define Y    SWIZZLE_Y
 297 #define Z    SWIZZLE_Z
 298 #define W    SWIZZLE_W
 299
 300
 301 /* Construct a ureg:
 302  */
 303 static struct ureg make_ureg(GLuint file, GLint idx)
 304 {
 305    struct ureg reg;
 306    reg.file = file;
 307    reg.idx = idx;
 308    reg.negate = 0;
 309    reg.swz = SWIZZLE_NOOP;
 310    reg.pad = 0;
 311    return reg;
 312 }
 313
 314
 315
 316 static struct ureg ureg_negate( struct ureg reg )
 317 {
 318    reg.negate ^= 1;
 319    return reg;
 320 }
 321
 322
 323 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 324 {
 325    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 326                            GET_SWZ(reg.swz, y),
 327                            GET_SWZ(reg.swz, z),
 328                            GET_SWZ(reg.swz, w));
 329
 330    return reg;
 331 }
 332
 333 static struct ureg swizzle1( struct ureg reg, int x )
 334 {
 335    return swizzle(reg, x, x, x, x);
 336 }
 337
 338 static struct ureg get_temp( struct tnl_program *p )
 339 {
 340    int bit = ffs( ~p->temp_in_use );
 341    if (!bit) {
 342       fprintf(stderr, "%s: out of temporaries\n", __FILE__);
 343       assert(0);
 344    }
 345
 346    if (bit > p->program->Base.NumTemporaries)
 347       p->program->Base.NumTemporaries = bit;
 348
 349    p->temp_in_use |= 1<<(bit-1);
 350    return make_ureg(PROGRAM_TEMPORARY, bit-1);
 351 }
 352
 353 static struct ureg reserve_temp( struct tnl_program *p )
 354 {
 355    struct ureg temp = get_temp( p );
 356    p->temp_reserved |= 1<<temp.idx;
 357    return temp;
 358 }
 359
 360 static void release_temp( struct tnl_program *p, struct ureg reg )
 361 {
 362    if (reg.file == PROGRAM_TEMPORARY) {
 363       p->temp_in_use &= ~(1<<reg.idx);
 364       p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */
 365    }
 366 }
 367
 368 static void release_temps( struct tnl_program *p )
 369 {
 370    p->temp_in_use = p->temp_reserved;
 371 }
 372
 373
 374
 375 static struct ureg register_input( struct tnl_program *p, GLuint input )
 376 {
 377    GLuint orig_input = input;
 378    /* Cram the material flags into the generic range.  We'll translate
 379     * them back later.
 380     */
 381    if (input >= BRW_ATTRIB_MAT_FRONT_AMBIENT)
 382       input -= BRW_ATTRIB_MAT_FRONT_AMBIENT - BRW_ATTRIB_GENERIC0;
 383
 384    assert(input < 32);
 385
 386    p->program->Base.InputsRead |= (1<<input);
 387    return make_ureg(PROGRAM_INPUT, orig_input);
 388 }
 389
 390 static struct ureg register_output( struct tnl_program *p, GLuint output )
 391 {
 392    p->program->Base.OutputsWritten |= (1<<output);
 393    return make_ureg(PROGRAM_OUTPUT, output);
 394 }
 395
 396 static struct ureg register_const4f( struct tnl_program *p,
 397                               GLfloat s0,
 398                               GLfloat s1,
 399                               GLfloat s2,
 400                               GLfloat s3)
 401 {
 402    GLfloat values[4];
 403    GLint idx;
 404    values[0] = s0;
 405    values[1] = s1;
 406    values[2] = s2;
 407    values[3] = s3;
 408    idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4 );
 409    return make_ureg(PROGRAM_STATE_VAR, idx);
 410 }
 411
 412 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
 413 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
 414 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
 415 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
 416
 417 static GLboolean is_undef( struct ureg reg )
 418 {
 419    return reg.file == PROGRAM_UNDEFINED;
 420 }
 421
 422 static struct ureg get_identity_param( struct tnl_program *p )
 423 {
 424    if (is_undef(p->identity))
 425       p->identity = register_const4f(p, 0,0,0,1);
 426
 427    return p->identity;
 428 }
 429
 430 static struct ureg register_param6( struct tnl_program *p,
 431                                    GLint s0,
 432                                    GLint s1,
 433                                    GLint s2,
 434                                    GLint s3,
 435                                    GLint s4,
 436                                    GLint s5)
 437 {
 438    GLint tokens[6];
 439    GLint idx;
 440    tokens[0] = s0;
 441    tokens[1] = s1;
 442    tokens[2] = s2;
 443    tokens[3] = s3;
 444    tokens[4] = s4;
 445    tokens[5] = s5;
 446    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 447    return make_ureg(PROGRAM_STATE_VAR, idx);
 448 }
 449
 450
 451 #define register_param1(p,s0)          register_param6(p,s0,0,0,0,0,0)
 452 #define register_param2(p,s0,s1)       register_param6(p,s0,s1,0,0,0,0)
 453 #define register_param3(p,s0,s1,s2)    register_param6(p,s0,s1,s2,0,0,0)
 454 #define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0)
 455
 456
 457 static void register_matrix_param6( struct tnl_program *p,
 458                                     GLint s0,
 459                                     GLint s1,
 460                                     GLint s2,
 461                                     GLint s3,
 462                                     GLint s4,
 463                                     GLint s5,
 464                                     struct ureg *matrix )
 465 {
 466    GLint i;
 467
 468    /* This is a bit sad as the support is there to pull the whole
 469     * matrix out in one go:
 470     */
 471    for (i = 0; i <= s4 - s3; i++)
 472       matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 );
 473 }
 474
 475
 476 static void emit_arg( struct prog_src_register *src,
 477                       struct ureg reg )
 478 {
 479    src->File = reg.file;
 480    src->Index = reg.idx;
 481    src->Swizzle = reg.swz;
 482    src->RelAddr = 0;
 483    src->NegateBase = reg.negate;
 484    src->Abs = 0;
 485    src->NegateAbs = 0;
 486 }
 487
 488 static void emit_dst( struct prog_dst_register *dst,
 489                       struct ureg reg, GLuint mask )
 490 {
 491    dst->File = reg.file;
 492    dst->Index = reg.idx;
 493    /* allow zero as a shorthand for xyzw */
 494    dst->WriteMask = mask ? mask : WRITEMASK_XYZW;
 495    dst->CondMask = 0;
 496    dst->CondSwizzle = 0;
 497    dst->CondSrc = 0;
 498    dst->pad = 0;
 499 }
 500
 501 static void debug_insn( struct prog_instruction *inst, const char *fn,
 502                         GLuint line )
 503 {
 504    if (DISASSEM) {
 505       static const char *last_fn;
 506
 507       if (fn != last_fn) {
 508          last_fn = fn;
 509          _mesa_printf("%s:\n", fn);
 510       }
 511
 512       _mesa_printf("%d:\t", line);
 513       _mesa_print_instruction(inst);
 514    }
 515 }
 516
 517
 518 static void emit_op3fn(struct tnl_program *p,
 519                        GLuint op,
 520                        struct ureg dest,
 521                        GLuint mask,
 522                        struct ureg src0,
 523                        struct ureg src1,
 524                        struct ureg src2,
 525                        const char *fn,
 526                        GLuint line)
 527 {
 528    GLuint nr = p->program->Base.NumInstructions++;
 529
 530    if (nr >= p->nr_instructions) {
 531       p->program->Base.Instructions =
 532          _mesa_realloc(p->program->Base.Instructions,
 533                        sizeof(struct prog_instruction) * p->nr_instructions,
 534                        sizeof(struct prog_instruction) * (p->nr_instructions *= 2));
 535    }
 536
 537    {
 538       struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 539       memset(inst, 0, sizeof(*inst));
 540       inst->Opcode = op;
 541       inst->StringPos = 0;
 542       inst->Data = 0;
 543
 544       emit_arg( &inst->SrcReg[0], src0 );
 545       emit_arg( &inst->SrcReg[1], src1 );
 546       emit_arg( &inst->SrcReg[2], src2 );
 547
 548       emit_dst( &inst->DstReg, dest, mask );
 549
 550       debug_insn(inst, fn, line);
 551    }
 552 }
 553
 554
 555
 556 #define emit_op3(p, op, dst, mask, src0, src1, src2) \
 557    emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__)
 558
 559 #define emit_op2(p, op, dst, mask, src0, src1) \
 560     emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__)
 561
 562 #define emit_op1(p, op, dst, mask, src0) \
 563     emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__)
 564
 565
 566 static struct ureg make_temp( struct tnl_program *p, struct ureg reg )
 567 {
 568    if (reg.file == PROGRAM_TEMPORARY &&
 569        !(p->temp_reserved & (1<<reg.idx)))
 570       return reg;
 571    else {
 572       struct ureg temp = get_temp(p);
 573       emit_op1(p, OPCODE_MOV, temp, 0, reg);
 574       return temp;
 575    }
 576 }
 577
 578
 579 /* Currently no tracking performed of input/output/register size or
 580  * active elements.  Could be used to reduce these operations, as
 581  * could the matrix type.
 582  */
 583 static void emit_matrix_transform_vec4( struct tnl_program *p,
 584                                         struct ureg dest,
 585                                         const struct ureg *mat,
 586                                         struct ureg src)
 587 {
 588    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]);
 589    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]);
 590    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]);
 591    emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]);
 592 }
 593
 594 /* This version is much easier to implement if writemasks are not
 595  * supported natively on the target or (like SSE), the target doesn't
 596  * have a clean/obvious dotproduct implementation.
 597  */
 598 static void emit_transpose_matrix_transform_vec4( struct tnl_program *p,
 599                                                   struct ureg dest,
 600                                                   const struct ureg *mat,
 601                                                   struct ureg src)
 602 {
 603    struct ureg tmp;
 604
 605    if (dest.file != PROGRAM_TEMPORARY)
 606       tmp = get_temp(p);
 607    else
 608       tmp = dest;
 609
 610    emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
 611    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
 612    emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
 613    emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
 614
 615    if (dest.file != PROGRAM_TEMPORARY)
 616       release_temp(p, tmp);
 617 }
 618
 619 static void emit_matrix_transform_vec3( struct tnl_program *p,
 620                                         struct ureg dest,
 621                                         const struct ureg *mat,
 622                                         struct ureg src)
 623 {
 624    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]);
 625    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]);
 626    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]);
 627 }
 628
 629
 630 static void emit_normalize_vec3( struct tnl_program *p,
 631                                  struct ureg dest,
 632                                  struct ureg src )
 633 {
 634    emit_op2(p, OPCODE_DP3, dest, WRITEMASK_W, src, src);
 635    emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_W, swizzle1(dest,W));
 636    emit_op2(p, OPCODE_MUL, dest, WRITEMASK_XYZ, src, swizzle1(dest,W));
 637 }
 638
 639 static void emit_passthrough( struct tnl_program *p,
 640                               GLuint input,
 641                               GLuint output )
 642 {
 643    struct ureg out = register_output(p, output);
 644    emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input));
 645 }
 646
 647 static struct ureg get_eye_position( struct tnl_program *p )
 648 {
 649    if (is_undef(p->eye_position)) {
 650       struct ureg pos = register_input( p, BRW_ATTRIB_POS );
 651       struct ureg modelview[4];
 652
 653       p->eye_position = reserve_temp(p);
 654
 655       if (PREFER_DP4) {
 656          register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3,
 657                                  STATE_MATRIX, modelview );
 658
 659          emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 660       }
 661       else {
 662          register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3,
 663                                  STATE_MATRIX_TRANSPOSE, modelview );
 664
 665          emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 666       }
 667    }
 668
 669    return p->eye_position;
 670 }
 671
 672
 673 #if 0
 674 static struct ureg get_eye_z( struct tnl_program *p )
 675 {
 676    if (!is_undef(p->eye_position)) {
 677       return swizzle1(p->eye_position, Z);
 678    }
 679    else if (!is_undef(p->eye_z)) {
 680       struct ureg pos = register_input( p, BRW_ATTRIB_POS );
 681       struct ureg modelview2;
 682
 683       p->eye_z = reserve_temp(p);
 684
 685       register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 2, 1,
 686                               STATE_MATRIX, &modelview2 );
 687
 688       emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
 689       emit_op2(p, OPCODE_DP4, p->eye_z, WRITEMASK_Z, pos, modelview2);
 690    }
 691
 692    return swizzle1(p->eye_z, Z)
 693 }
 694 #endif
 695
 696
 697
 698 static struct ureg get_eye_position_normalized( struct tnl_program *p )
 699 {
 700    if (is_undef(p->eye_position_normalized)) {
 701       struct ureg eye = get_eye_position(p);
 702       p->eye_position_normalized = reserve_temp(p);
 703       emit_normalize_vec3(p, p->eye_position_normalized, eye);
 704    }
 705
 706    return p->eye_position_normalized;
 707 }
 708
 709
 710 static struct ureg get_eye_normal( struct tnl_program *p )
 711 {
 712    if (is_undef(p->eye_normal)) {
 713       struct ureg normal = register_input(p, BRW_ATTRIB_NORMAL );
 714       struct ureg mvinv[3];
 715
 716       register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2,
 717                               STATE_MATRIX_INVTRANS, mvinv );
 718
 719       p->eye_normal = reserve_temp(p);
 720
 721       /* Transform to eye space:
 722        */
 723       emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal );
 724
 725       /* Normalize/Rescale:
 726        */
 727       if (p->state->normalize) {
 728          emit_normalize_vec3( p, p->eye_normal, p->eye_normal );
 729       }
 730       else if (p->state->rescale_normals) {
 731          struct ureg rescale = register_param2(p, STATE_INTERNAL,
 732                                                STATE_NORMAL_SCALE);
 733
 734          emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal,
 735                    swizzle1(rescale, X));
 736       }
 737    }
 738
 739    return p->eye_normal;
 740 }
 741
 742
 743
 744 static void build_hpos( struct tnl_program *p )
 745 {
 746    struct ureg pos = register_input( p, BRW_ATTRIB_POS );
 747    struct ureg hpos = register_output( p, VERT_RESULT_HPOS );
 748    struct ureg mvp[4];
 749
 750    if (PREFER_DP4) {
 751       register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3,
 752                               STATE_MATRIX, mvp );
 753       emit_matrix_transform_vec4( p, hpos, mvp, pos );
 754    }
 755    else {
 756       register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3,
 757                               STATE_MATRIX_TRANSPOSE, mvp );
 758       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
 759    }
 760 }
 761
 762
 763 static GLuint material_attrib( GLuint side, GLuint property )
 764 {
 765    return (property - STATE_AMBIENT) * 2 + side;
 766 }
 767
 768 /* Get a bitmask of which material values vary on a per-vertex basis.
 769  */
 770 static void set_material_flags( struct tnl_program *p )
 771 {
 772    p->color_materials = 0;
 773    p->materials = 0;
 774
 775    if (p->state->light_color_material) {
 776       p->materials =
 777          p->color_materials = p->state->light_color_material_mask;
 778    }
 779
 780    p->materials |= p->state->light_material_mask;
 781 }
 782
 783
 784 static struct ureg get_material( struct tnl_program *p, GLuint side,
 785                                  GLuint property )
 786 {
 787    GLuint attrib = material_attrib(side, property);
 788
 789    if (p->color_materials & (1<<attrib))
 790       return register_input(p, BRW_ATTRIB_COLOR0);
 791    else if (p->materials & (1<<attrib))
 792       return register_input( p, attrib + BRW_ATTRIB_MAT_FRONT_AMBIENT );
 793    else
 794       return register_param3( p, STATE_MATERIAL, side, property );
 795 }
 796
 797 #define SCENE_COLOR_BITS(side) ((MAT_BIT_FRONT_EMISSION | \
 798                                  MAT_BIT_FRONT_AMBIENT | \
 799                                  MAT_BIT_FRONT_DIFFUSE) << (side))
 800
 801 /* Either return a precalculated constant value or emit code to
 802  * calculate these values dynamically in the case where material calls
 803  * are present between begin/end pairs.
 804  *
 805  * Probably want to shift this to the program compilation phase - if
 806  * we always emitted the calculation here, a smart compiler could
 807  * detect that it was constant (given a certain set of inputs), and
 808  * lift it out of the main loop.  That way the programs created here
 809  * would be independent of the vertex_buffer details.
 810  */
 811 static struct ureg get_scenecolor( struct tnl_program *p, GLuint side )
 812 {
 813    if (p->materials & SCENE_COLOR_BITS(side)) {
 814       struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT);
 815       struct ureg material_emission = get_material(p, side, STATE_EMISSION);
 816       struct ureg material_ambient = get_material(p, side, STATE_AMBIENT);
 817       struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE);
 818       struct ureg tmp = make_temp(p, material_diffuse);
 819       emit_op3(p, OPCODE_MAD, tmp,  WRITEMASK_XYZ, lm_ambient,
 820                material_ambient, material_emission);
 821       return tmp;
 822    }
 823    else
 824       return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side );
 825 }
 826
 827
 828 static struct ureg get_lightprod( struct tnl_program *p, GLuint light,
 829                                   GLuint side, GLuint property )
 830 {
 831    GLuint attrib = material_attrib(side, property);
 832    if (p->materials & (1<<attrib)) {
 833       struct ureg light_value =
 834          register_param3(p, STATE_LIGHT, light, property);
 835       struct ureg material_value = get_material(p, side, property);
 836       struct ureg tmp = get_temp(p);
 837       emit_op2(p, OPCODE_MUL, tmp,  0, light_value, material_value);
 838       return tmp;
 839    }
 840    else
 841       return register_param4(p, STATE_LIGHTPROD, light, side, property);
 842 }
 843
 844 static struct ureg calculate_light_attenuation( struct tnl_program *p,
 845                                                 GLuint i,
 846                                                 struct ureg VPpli,
 847                                                 struct ureg dist )
 848 {
 849    struct ureg attenuation = register_param3(p, STATE_LIGHT, i,
 850                                              STATE_ATTENUATION);
 851    struct ureg att = get_temp(p);
 852
 853    /* Calculate spot attenuation:
 854     */
 855    if (!p->state->unit[i].light_spotcutoff_is_180) {
 856       struct ureg spot_dir = register_param3(p, STATE_LIGHT, i,
 857                                              STATE_SPOT_DIRECTION);
 858       struct ureg spot = get_temp(p);
 859       struct ureg slt = get_temp(p);
 860
 861       emit_normalize_vec3( p, spot, spot_dir ); /* XXX: precompute! */
 862       emit_op2(p, OPCODE_DP3, spot, 0, ureg_negate(VPpli), spot);
 863       emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir,W), spot);
 864       emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W));
 865       emit_op2(p, OPCODE_MUL, att, 0, slt, spot);
 866
 867       release_temp(p, spot);
 868       release_temp(p, slt);
 869    }
 870
 871    /* Calculate distance attenuation:
 872     */
 873    if (p->state->unit[i].light_attenuated) {
 874
 875       /* 1/d,d,d,1/d */
 876       emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist);
 877       /* 1,d,d*d,1/d */
 878       emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y));
 879       /* 1/dist-atten */
 880       emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist);
 881
 882       if (!p->state->unit[i].light_spotcutoff_is_180) {
 883          /* dist-atten */
 884          emit_op1(p, OPCODE_RCP, dist, 0, dist);
 885          /* spot-atten * dist-atten */
 886          emit_op2(p, OPCODE_MUL, att, 0, dist, att);
 887       } else {
 888          /* dist-atten */
 889          emit_op1(p, OPCODE_RCP, att, 0, dist);
 890       }
 891    }
 892
 893    return att;
 894 }
 895
 896
 897
 898
 899
 900 /* Need to add some addtional parameters to allow lighting in object
 901  * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye
 902  * space lighting.
 903  */
 904 static void build_lighting( struct tnl_program *p )
 905 {
 906    const GLboolean twoside = p->state->light_twoside;
 907    const GLboolean separate = p->state->separate_specular;
 908    GLuint nr_lights = 0, count = 0;
 909    struct ureg normal = get_eye_normal(p);
 910    struct ureg lit = get_temp(p);
 911    struct ureg dots = get_temp(p);
 912    struct ureg _col0 = undef, _col1 = undef;
 913    struct ureg _bfc0 = undef, _bfc1 = undef;
 914    GLuint i;
 915
 916    for (i = 0; i < MAX_LIGHTS; i++)
 917       if (p->state->unit[i].light_enabled)
 918          nr_lights++;
 919
 920    set_material_flags(p);
 921
 922    {
 923       struct ureg shininess = get_material(p, 0, STATE_SHININESS);
 924       emit_op1(p, OPCODE_MOV, dots,  WRITEMASK_W, swizzle1(shininess,X));
 925       release_temp(p, shininess);
 926
 927       _col0 = make_temp(p, get_scenecolor(p, 0));
 928       if (separate)
 929          _col1 = make_temp(p, get_identity_param(p));
 930       else
 931          _col1 = _col0;
 932
 933    }
 934
 935    if (twoside) {
 936       struct ureg shininess = get_material(p, 1, STATE_SHININESS);
 937       emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z,
 938                ureg_negate(swizzle1(shininess,X)));
 939       release_temp(p, shininess);
 940
 941       _bfc0 = make_temp(p, get_scenecolor(p, 1));
 942       if (separate)
 943          _bfc1 = make_temp(p, get_identity_param(p));
 944       else
 945          _bfc1 = _bfc0;
 946    }
 947
 948
 949    /* If no lights, still need to emit the scenecolor.
 950     */
 951    /* KW: changed to do this always - v1.17 "Fix lighting alpha result"?
 952     */
 953    if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
 954    {
 955       struct ureg res0 = register_output( p, VERT_RESULT_COL0 );
 956       emit_op1(p, OPCODE_MOV, res0, 0, _col0);
 957
 958       if (twoside) {
 959          struct ureg res0 = register_output( p, VERT_RESULT_BFC0 );
 960          emit_op1(p, OPCODE_MOV, res0, 0, _bfc0);
 961       }
 962    }
 963
 964    if (separate && (p->state->fragprog_inputs_read & FRAG_BIT_COL1)) {
 965
 966       struct ureg res1 = register_output( p, VERT_RESULT_COL1 );
 967       emit_op1(p, OPCODE_MOV, res1, 0, _col1);
 968
 969       if (twoside) {
 970          struct ureg res1 = register_output( p, VERT_RESULT_BFC1 );
 971          emit_op1(p, OPCODE_MOV, res1, 0, _bfc1);
 972       }
 973    }
 974
 975    if (nr_lights == 0) {
 976       release_temps(p);
 977       return;
 978    }
 979
 980
 981    for (i = 0; i < MAX_LIGHTS; i++) {
 982       if (p->state->unit[i].light_enabled) {
 983          struct ureg half = undef;
 984          struct ureg att = undef, VPpli = undef;
 985
 986          count++;
 987
 988          if (p->state->unit[i].light_eyepos3_is_zero) {
 989             /* Can used precomputed constants in this case.
 990              * Attenuation never applies to infinite lights.
 991              */
 992             VPpli = register_param3(p, STATE_LIGHT, i,
 993                                     STATE_POSITION_NORMALIZED);
 994             half = register_param3(p, STATE_LIGHT, i, STATE_HALF);
 995          }
 996          else {
 997             struct ureg Ppli = register_param3(p, STATE_LIGHT, i,
 998                                                STATE_POSITION);
 999             struct ureg V = get_eye_position(p);
1000             struct ureg dist = get_temp(p);
1001
1002             VPpli = get_temp(p);
1003             half = get_temp(p);
1004
1005             /* Calulate VPpli vector
1006              */
1007             emit_op2(p, OPCODE_SUB, VPpli, 0, Ppli, V);
1008
1009             /* Normalize VPpli.  The dist value also used in
1010              * attenuation below.
1011              */
1012             emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli);
1013             emit_op1(p, OPCODE_RSQ, dist, 0, dist);
1014             emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist);
1015
1016
1017             /* Calculate  attenuation:
1018              */
1019             if (!p->state->unit[i].light_spotcutoff_is_180 ||
1020                 p->state->unit[i].light_attenuated) {
1021                att = calculate_light_attenuation(p, i, VPpli, dist);
1022             }
1023
1024
1025             /* Calculate viewer direction, or use infinite viewer:
1026              */
1027             if (p->state->light_local_viewer) {
1028                struct ureg eye_hat = get_eye_position_normalized(p);
1029                emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
1030             }
1031             else {
1032                struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z);
1033                emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir);
1034             }
1035
1036             emit_normalize_vec3(p, half, half);
1037
1038             release_temp(p, dist);
1039          }
1040
1041          /* Calculate dot products:
1042           */
1043          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli);
1044          emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half);
1045
1046
1047          /* Front face lighting:
1048           */
1049          {
1050             struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT);
1051             struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE);
1052             struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR);
1053             struct ureg res0, res1;
1054             GLuint mask0, mask1;
1055
1056             emit_op1(p, OPCODE_LIT, lit, 0, dots);
1057
1058             if (!is_undef(att))
1059                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1060
1061
1062             mask0 = 0;
1063             mask1 = 0;
1064             res0 = _col0;
1065             res1 = _col1;
1066
1067             if (count == nr_lights) {
1068                if (separate) {
1069                   mask0 = WRITEMASK_XYZ;
1070                   mask1 = WRITEMASK_XYZ;
1071
1072                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1073                      res0 = register_output( p, VERT_RESULT_COL0 );
1074
1075                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1076                      res1 = register_output( p, VERT_RESULT_COL1 );
1077                }
1078                else {
1079                   mask1 = WRITEMASK_XYZ;
1080
1081                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1082                      res1 = register_output( p, VERT_RESULT_COL0 );
1083                }
1084             }
1085
1086             emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0);
1087             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0);
1088             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1);
1089
1090             release_temp(p, ambient);
1091             release_temp(p, diffuse);
1092             release_temp(p, specular);
1093          }
1094
1095          /* Back face lighting:
1096           */
1097          if (twoside) {
1098             struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT);
1099             struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE);
1100             struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR);
1101             struct ureg res0, res1;
1102             GLuint mask0, mask1;
1103
1104             emit_op1(p, OPCODE_LIT, lit, 0, ureg_negate(swizzle(dots,X,Y,W,Z)));
1105
1106             if (!is_undef(att))
1107                emit_op2(p, OPCODE_MUL, lit, 0, lit, att);
1108
1109             mask0 = 0;
1110             mask1 = 0;
1111             res0 = _bfc0;
1112             res1 = _bfc1;
1113
1114             if (count == nr_lights) {
1115                if (separate) {
1116                   mask0 = WRITEMASK_XYZ;
1117                   mask1 = WRITEMASK_XYZ;
1118                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1119                      res0 = register_output( p, VERT_RESULT_BFC0 );
1120
1121                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1122                      res1 = register_output( p, VERT_RESULT_BFC1 );
1123                }
1124                else {
1125                   mask1 = WRITEMASK_XYZ;
1126
1127                   if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1128                      res1 = register_output( p, VERT_RESULT_BFC0 );
1129                }
1130             }
1131
1132             emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0);
1133             emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0);
1134             emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1);
1135
1136             release_temp(p, ambient);
1137             release_temp(p, diffuse);
1138             release_temp(p, specular);
1139          }
1140
1141          release_temp(p, half);
1142          release_temp(p, VPpli);
1143          release_temp(p, att);
1144       }
1145    }
1146
1147    release_temps( p );
1148 }
1149
1150
1151 static void build_fog( struct tnl_program *p )
1152 {
1153    struct ureg fog = register_output(p, VERT_RESULT_FOGC);
1154    struct ureg input;
1155
1156    if (p->state->fog_source_is_depth) {
1157       input = swizzle1(get_eye_position(p), Z);
1158    }
1159    else {
1160       input = swizzle1(register_input(p, BRW_ATTRIB_FOG), X);
1161    }
1162
1163    if (p->state->fog_option &&
1164        p->state->tnl_do_vertex_fog) {
1165       struct ureg params = register_param1(p, STATE_FOG_PARAMS);
1166       struct ureg tmp = get_temp(p);
1167       struct ureg id = get_identity_param(p);
1168
1169       emit_op1(p, OPCODE_MOV, fog, 0, id);
1170
1171       switch (p->state->fog_option) {
1172       case FOG_LINEAR: {
1173          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1174          emit_op2(p, OPCODE_SUB, tmp, 0, swizzle1(params,Z), tmp);
1175          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,W));
1176          emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */
1177          emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W));
1178          break;
1179       }
1180       case FOG_EXP:
1181          emit_op1(p, OPCODE_ABS, tmp, 0, input);
1182          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, swizzle1(params,X));
1183          emit_op2(p, OPCODE_POW, fog, WRITEMASK_X,
1184                   register_const1f(p, M_E), ureg_negate(tmp));
1185          break;
1186       case FOG_EXP2:
1187          emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,X));
1188          emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp);
1189          emit_op2(p, OPCODE_POW, fog, WRITEMASK_X,
1190                   register_const1f(p, M_E), ureg_negate(tmp));
1191          break;
1192       }
1193
1194       release_temp(p, tmp);
1195    }
1196    else {
1197       /* results = incoming fog coords (compute fog per-fragment later)
1198        *
1199        * KW:  Is it really necessary to do anything in this case?
1200        */
1201       emit_op1(p, OPCODE_MOV, fog, 0, input);
1202    }
1203 }
1204
1205 static void build_reflect_texgen( struct tnl_program *p,
1206                                   struct ureg dest,
1207                                   GLuint writemask )
1208 {
1209    struct ureg normal = get_eye_normal(p);
1210    struct ureg eye_hat = get_eye_position_normalized(p);
1211    struct ureg tmp = get_temp(p);
1212
1213    /* n.u */
1214    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1215    /* 2n.u */
1216    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1217    /* (-2n.u)n + u */
1218    emit_op3(p, OPCODE_MAD, dest, writemask, ureg_negate(tmp), normal, eye_hat);
1219
1220    release_temp(p, tmp);
1221 }
1222
1223 static void build_sphere_texgen( struct tnl_program *p,
1224                                  struct ureg dest,
1225                                  GLuint writemask )
1226 {
1227    struct ureg normal = get_eye_normal(p);
1228    struct ureg eye_hat = get_eye_position_normalized(p);
1229    struct ureg tmp = get_temp(p);
1230    struct ureg half = register_scalar_const(p, .5);
1231    struct ureg r = get_temp(p);
1232    struct ureg inv_m = get_temp(p);
1233    struct ureg id = get_identity_param(p);
1234
1235    /* Could share the above calculations, but it would be
1236     * a fairly odd state for someone to set (both sphere and
1237     * reflection active for different texture coordinate
1238     * components.  Of course - if two texture units enable
1239     * reflect and/or sphere, things start to tilt in favour
1240     * of seperating this out:
1241     */
1242
1243    /* n.u */
1244    emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat);
1245    /* 2n.u */
1246    emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp);
1247    /* (-2n.u)n + u */
1248    emit_op3(p, OPCODE_MAD, r, 0, ureg_negate(tmp), normal, eye_hat);
1249    /* r + 0,0,1 */
1250    emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z));
1251    /* rx^2 + ry^2 + (rz+1)^2 */
1252    emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp);
1253    /* 2/m */
1254    emit_op1(p, OPCODE_RSQ, tmp, 0, tmp);
1255    /* 1/m */
1256    emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half);
1257    /* r/m + 1/2 */
1258    emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half);
1259
1260    release_temp(p, tmp);
1261    release_temp(p, r);
1262    release_temp(p, inv_m);
1263 }
1264
1265
1266 static void build_texture_transform( struct tnl_program *p )
1267 {
1268    GLuint i, j;
1269
1270    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
1271
1272       if (!(p->state->fragprog_inputs_read & (FRAG_BIT_TEX0<<i)))
1273          continue;
1274
1275       if (p->state->unit[i].texgen_enabled ||
1276           p->state->unit[i].texmat_enabled) {
1277
1278          GLuint texmat_enabled = p->state->unit[i].texmat_enabled;
1279          struct ureg out = register_output(p, VERT_RESULT_TEX0 + i);
1280          struct ureg out_texgen = undef;
1281
1282          if (p->state->unit[i].texgen_enabled) {
1283             GLuint copy_mask = 0;
1284             GLuint sphere_mask = 0;
1285             GLuint reflect_mask = 0;
1286             GLuint normal_mask = 0;
1287             GLuint modes[4];
1288
1289             if (texmat_enabled)
1290                out_texgen = get_temp(p);
1291             else
1292                out_texgen = out;
1293
1294             modes[0] = p->state->unit[i].texgen_mode0;
1295             modes[1] = p->state->unit[i].texgen_mode1;
1296             modes[2] = p->state->unit[i].texgen_mode2;
1297             modes[3] = p->state->unit[i].texgen_mode3;
1298
1299             for (j = 0; j < 4; j++) {
1300                switch (modes[j]) {
1301                case TXG_OBJ_LINEAR: {
1302                   struct ureg obj = register_input(p, BRW_ATTRIB_POS);
1303                   struct ureg plane =
1304                      register_param3(p, STATE_TEXGEN, i,
1305                                      STATE_TEXGEN_OBJECT_S + j);
1306
1307                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1308                            obj, plane );
1309                   break;
1310                }
1311                case TXG_EYE_LINEAR: {
1312                   struct ureg eye = get_eye_position(p);
1313                   struct ureg plane =
1314                      register_param3(p, STATE_TEXGEN, i,
1315                                      STATE_TEXGEN_EYE_S + j);
1316
1317                   emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j,
1318                            eye, plane );
1319                   break;
1320                }
1321                case TXG_SPHERE_MAP:
1322                   sphere_mask |= WRITEMASK_X << j;
1323                   break;
1324                case TXG_REFLECTION_MAP:
1325                   reflect_mask |= WRITEMASK_X << j;
1326                   break;
1327                case TXG_NORMAL_MAP:
1328                   normal_mask |= WRITEMASK_X << j;
1329                   break;
1330                case TXG_NONE:
1331                   copy_mask |= WRITEMASK_X << j;
1332                }
1333
1334             }
1335
1336
1337             if (sphere_mask) {
1338                build_sphere_texgen(p, out_texgen, sphere_mask);
1339             }
1340
1341             if (reflect_mask) {
1342                build_reflect_texgen(p, out_texgen, reflect_mask);
1343             }
1344
1345             if (normal_mask) {
1346                struct ureg normal = get_eye_normal(p);
1347                emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal );
1348             }
1349
1350             if (copy_mask) {
1351                struct ureg in = register_input(p, BRW_ATTRIB_TEX0+i);
1352                emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in );
1353             }
1354          }
1355
1356          if (texmat_enabled) {
1357             struct ureg texmat[4];
1358             struct ureg in = (!is_undef(out_texgen) ?
1359                               out_texgen :
1360                               register_input(p, BRW_ATTRIB_TEX0+i));
1361             if (PREFER_DP4) {
1362                register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i,
1363                                        0, 3, STATE_MATRIX, texmat );
1364                emit_matrix_transform_vec4( p, out, texmat, in );
1365             }
1366             else {
1367                register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i,
1368                                        0, 3, STATE_MATRIX_TRANSPOSE, texmat );
1369                emit_transpose_matrix_transform_vec4( p, out, texmat, in );
1370             }
1371          }
1372
1373          release_temps(p);
1374       }
1375       else {
1376          emit_passthrough(p, BRW_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i);
1377       }
1378    }
1379 }
1380
1381
1382 /* Seems like it could be tighter:
1383  */
1384 static void build_pointsize( struct tnl_program *p )
1385 {
1386    struct ureg eye = get_eye_position(p);
1387    struct ureg state_size = register_param1(p, STATE_POINT_SIZE);
1388    struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION);
1389    struct ureg out = register_output(p, VERT_RESULT_PSIZ);
1390    struct ureg ut = get_temp(p);
1391
1392    /* 1, Z, Z * Z, 1 */
1393    emit_op1(p, OPCODE_MOV, ut, WRITEMASK_XW, swizzle1(get_identity_param(p), W));
1394    emit_op1(p, OPCODE_ABS, ut, WRITEMASK_YZ, swizzle1(eye, Z));
1395    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_Z, ut, ut);
1396
1397
1398    /* p1 +  p2 * dist + p3 * dist * dist, 0 */
1399    emit_op2(p, OPCODE_DP3, ut, WRITEMASK_X, ut, state_attenuation);
1400
1401    /* 1 / sqrt(factor) */
1402    emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut );
1403
1404    /* ut = pointSize / factor */
1405    emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size);
1406
1407    /* Clamp to min/max - state_size.[yz]
1408     */
1409    emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y));
1410    emit_op2(p, OPCODE_MIN, out, 0, swizzle1(ut, X), swizzle1(state_size, Z));
1411
1412    release_temp(p, ut);
1413 }
1414
1415 static void build_tnl_program( struct tnl_program *p )
1416 {
1417    /* Emit the program, starting with modelviewproject:
1418     */
1419    build_hpos(p);
1420
1421    /* Lighting calculations:
1422     */
1423    if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) {
1424       if (p->state->light_global_enabled)
1425          build_lighting(p);
1426       else {
1427          if (p->state->fragprog_inputs_read & FRAG_BIT_COL0)
1428             emit_passthrough(p, BRW_ATTRIB_COLOR0, VERT_RESULT_COL0);
1429
1430          if (p->state->fragprog_inputs_read & FRAG_BIT_COL1)
1431             emit_passthrough(p, BRW_ATTRIB_COLOR1, VERT_RESULT_COL1);
1432       }
1433    }
1434
1435    if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) ||
1436        p->state->fog_option != FOG_NONE)
1437       build_fog(p);
1438
1439    if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY)
1440       build_texture_transform(p);
1441
1442    if (p->state->point_attenuated)
1443       build_pointsize(p);
1444
1445    /* Finish up:
1446     */
1447    emit_op1(p, OPCODE_END, undef, 0, undef);
1448
1449    /* Disassemble:
1450     */
1451    if (DISASSEM) {
1452       _mesa_printf ("\n");
1453    }
1454 }
1455
1456
1457 static void build_new_tnl_program( const struct state_key *key,
1458                                    struct gl_vertex_program *program,
1459                                    GLuint max_temps)
1460 {
1461    struct tnl_program p;
1462
1463    _mesa_memset(&p, 0, sizeof(p));
1464    p.state = key;
1465    p.program = program;
1466    p.eye_position = undef;
1467    p.eye_position_normalized = undef;
1468    p.eye_normal = undef;
1469    p.identity = undef;
1470    p.temp_in_use = 0;
1471    p.nr_instructions = 16;
1472
1473    if (max_temps >= sizeof(int) * 8)
1474       p.temp_reserved = 0;
1475    else
1476       p.temp_reserved = ~((1<<max_temps)-1);
1477
1478    p.program->Base.Instructions =
1479       _mesa_malloc(sizeof(struct prog_instruction) * p.nr_instructions);
1480    p.program->Base.String = 0;
1481    p.program->Base.NumInstructions =
1482    p.program->Base.NumTemporaries =
1483    p.program->Base.NumParameters =
1484    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1485    p.program->Base.Parameters = _mesa_new_parameter_list();
1486    p.program->Base.InputsRead = 0;
1487    p.program->Base.OutputsWritten = 0;
1488
1489    build_tnl_program( &p );
1490 }
1491
1492 static void *search_cache( struct brw_tnl_cache *cache,
1493                            GLuint hash,
1494                            const void *key,
1495                            GLuint keysize)
1496 {
1497    struct brw_tnl_cache_item *c;
1498
1499    for (c = cache->items[hash % cache->size]; c; c = c->next) {
1500       if (c->hash == hash && memcmp(c->key, key, keysize) == 0)
1501          return c->data;
1502    }
1503
1504    return NULL;
1505 }
1506
1507 static void rehash( struct brw_tnl_cache *cache )
1508 {
1509    struct brw_tnl_cache_item **items;
1510    struct brw_tnl_cache_item *c, *next;
1511    GLuint size, i;
1512
1513    size = cache->size * 3;
1514    items = (struct brw_tnl_cache_item**) _mesa_malloc(size * sizeof(*items));
1515    _mesa_memset(items, 0, size * sizeof(*items));
1516
1517    for (i = 0; i < cache->size; i++)
1518       for (c = cache->items[i]; c; c = next) {
1519          next = c->next;
1520          c->next = items[c->hash % size];
1521          items[c->hash % size] = c;
1522       }
1523
1524    FREE(cache->items);
1525    cache->items = items;
1526    cache->size = size;
1527 }
1528
1529 static void cache_item( struct brw_tnl_cache *cache,
1530                         GLuint hash,
1531                         const struct state_key *key,
1532                         void *data )
1533 {
1534    struct brw_tnl_cache_item *c = MALLOC(sizeof(*c));
1535    c->hash = hash;
1536
1537    c->key = malloc(sizeof(*key));
1538    memcpy(c->key, key, sizeof(*key));
1539
1540    c->data = data;
1541
1542    if (++cache->n_items > cache->size * 1.5)
1543       rehash(cache);
1544
1545    c->next = cache->items[hash % cache->size];
1546    cache->items[hash % cache->size] = c;
1547 }
1548
1549
1550 static GLuint hash_key( struct state_key *key )
1551 {
1552    GLuint *ikey = (GLuint *)key;
1553    GLuint hash = 0, i;
1554
1555    /* I'm sure this can be improved on, but speed is important:
1556     */
1557    for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
1558       hash += ikey[i];
1559
1560    return hash;
1561 }
1562
1563 static void update_tnl_program( struct brw_context *brw )
1564 {
1565    GLcontext *ctx = &brw->intel.ctx;
1566    struct state_key key;
1567    GLuint hash;
1568    struct gl_vertex_program *old = brw->tnl_program;
1569
1570    /* _NEW_PROGRAM */
1571    if (brw->attribs.VertexProgram->_Enabled)
1572       return;
1573
1574    /* Grab all the relevent state and put it in a single structure:
1575     */
1576    make_state_key(ctx, &key);
1577    hash = hash_key(&key);
1578
1579    /* Look for an already-prepared program for this state:
1580     */
1581    brw->tnl_program = (struct gl_vertex_program *)
1582       search_cache( &brw->tnl_program_cache, hash, &key, sizeof(key) );
1583
1584    /* OK, we'll have to build a new one:
1585     */
1586    if (!brw->tnl_program) {
1587       brw->tnl_program = (struct gl_vertex_program *)
1588          ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0);
1589
1590       build_new_tnl_program( &key, brw->tnl_program,
1591 /*                           ctx->Const.MaxVertexProgramTemps  */
1592                              32
1593          );
1594
1595       if (ctx->Driver.ProgramStringNotify)
1596          ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB,
1597                                           &brw->tnl_program->Base );
1598
1599       cache_item( &brw->tnl_program_cache,
1600                   hash, &key, brw->tnl_program );
1601    }
1602
1603    if (old != brw->tnl_program)
1604       brw->state.dirty.brw |= BRW_NEW_TNL_PROGRAM;
1605 }
1606
1607 /* Note: See brw_draw.c - the vertex program must not rely on
1608  * brw->primitive or brw->reduced_prim.
1609  */
1610 const struct brw_tracked_state brw_tnl_vertprog = {
1611    .dirty = {
1612       .mesa = (_NEW_PROGRAM |
1613                _NEW_LIGHT |
1614                _NEW_TRANSFORM |
1615                _NEW_FOG |
1616                _NEW_HINT |
1617                _NEW_POINT |
1618                _NEW_TEXTURE),
1619       .brw = (BRW_NEW_FRAGMENT_PROGRAM |
1620               BRW_NEW_INPUT_VARYING),
1621       .cache = 0
1622    },
1623    .update = update_tnl_program
1624 };
1625
1626
1627
1628
1629 static void update_active_vertprog( struct brw_context *brw )
1630 {
1631    const struct gl_vertex_program *prev = brw->vertex_program;
1632
1633    /* NEW_PROGRAM */
1634    if (brw->attribs.VertexProgram->_Enabled) {
1635       brw->vertex_program = brw->attribs.VertexProgram->Current;
1636    }
1637    else {
1638       /* BRW_NEW_TNL_PROGRAM */
1639       brw->vertex_program = brw->tnl_program;
1640    }
1641
1642    if (brw->vertex_program != prev)
1643       brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
1644 }
1645
1646
1647
1648 const struct brw_tracked_state brw_active_vertprog = {
1649    .dirty = {
1650       .mesa = _NEW_PROGRAM,
1651       .brw = BRW_NEW_TNL_PROGRAM,
1652       .cache = 0
1653    },
1654    .update = update_active_vertprog
1655 };
1656
1657
1658 void brw_ProgramCacheInit( GLcontext *ctx )
1659 {
1660    struct brw_context *brw = brw_context(ctx);
1661
1662    brw->tnl_program_cache.size = 17;
1663    brw->tnl_program_cache.n_items = 0;
1664    brw->tnl_program_cache.items = (struct brw_tnl_cache_item **)
1665       _mesa_calloc(brw->tnl_program_cache.size *
1666                    sizeof(struct brw_tnl_cache_item));
1667 }
1668
1669 void brw_ProgramCacheDestroy( GLcontext *ctx )
1670 {
1671    struct brw_context *brw = brw_context(ctx);
1672    struct brw_tnl_cache_item *c, *next;
1673    GLuint i;
1674
1675    for (i = 0; i < brw->tnl_program_cache.size; i++)
1676       for (c = brw->tnl_program_cache.items[i]; c; c = next) {
1677          next = c->next;
1678          FREE(c->key);
1679          FREE(c->data);
1680          FREE(c);
1681       }
1682
1683    FREE(brw->tnl_program_cache.items);
1684 }