src/mesa/main/texenvprogram.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  * Copyright 2009 VMware, Inc.  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
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 #include "glheader.h"
  30 #include "macros.h"
  31 #include "enums.h"
  32 #include "shader/program.h"
  33 #include "shader/prog_parameter.h"
  34 #include "shader/prog_cache.h"
  35 #include "shader/prog_instruction.h"
  36 #include "shader/prog_print.h"
  37 #include "shader/prog_statevars.h"
  38 #include "shader/programopt.h"
  39 #include "texenvprogram.h"
  40
  41
  42 /*
  43  * Note on texture units:
  44  *
  45  * The number of texture units supported by fixed-function fragment
  46  * processing is MAX_TEXTURE_COORD_UNITS, not MAX_TEXTURE_IMAGE_UNITS.
  47  * That's because there's a one-to-one correspondence between texture
  48  * coordinates and samplers in fixed-function processing.
  49  *
  50  * Since fixed-function vertex processing is limited to MAX_TEXTURE_COORD_UNITS
  51  * sets of texcoords, so is fixed-function fragment processing.
  52  *
  53  * We can safely use ctx->Const.MaxTextureUnits for loop bounds.
  54  */
  55
  56
  57 struct texenvprog_cache_item
  58 {
  59    GLuint hash;
  60    void *key;
  61    struct gl_fragment_program *data;
  62    struct texenvprog_cache_item *next;
  63 };
  64
  65
  66 /**
  67  * Up to nine instructions per tex unit, plus fog, specular color.
  68  */
  69 #define MAX_INSTRUCTIONS ((MAX_TEXTURE_COORD_UNITS * 9) + 12)
  70
  71 #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM)
  72
  73 struct mode_opt {
  74    GLuint Source:4;
  75    GLuint Operand:3;
  76 };
  77
  78 struct state_key {
  79    GLuint nr_enabled_units:8;
  80    GLuint enabled_units:8;
  81    GLuint separate_specular:1;
  82    GLuint fog_enabled:1;
  83    GLuint fog_mode:2;
  84    GLuint inputs_available:12;
  85
  86    struct {
  87       GLuint enabled:1;
  88       GLuint source_index:3;   /* one of TEXTURE_1D/2D/3D/CUBE/RECT_INDEX */
  89       GLuint shadow:1;
  90       GLuint ScaleShiftRGB:2;
  91       GLuint ScaleShiftA:2;
  92
  93       GLuint NumArgsRGB:3;
  94       GLuint ModeRGB:5;
  95       struct mode_opt OptRGB[MAX_COMBINER_TERMS];
  96
  97       GLuint NumArgsA:3;
  98       GLuint ModeA:5;
  99       struct mode_opt OptA[MAX_COMBINER_TERMS];
 100    } unit[8];
 101 };
 102
 103 #define FOG_LINEAR  0
 104 #define FOG_EXP     1
 105 #define FOG_EXP2    2
 106 #define FOG_UNKNOWN 3
 107
 108 static GLuint translate_fog_mode( GLenum mode )
 109 {
 110    switch (mode) {
 111    case GL_LINEAR: return FOG_LINEAR;
 112    case GL_EXP: return FOG_EXP;
 113    case GL_EXP2: return FOG_EXP2;
 114    default: return FOG_UNKNOWN;
 115    }
 116 }
 117
 118 #define OPR_SRC_COLOR           0
 119 #define OPR_ONE_MINUS_SRC_COLOR 1
 120 #define OPR_SRC_ALPHA           2
 121 #define OPR_ONE_MINUS_SRC_ALPHA 3
 122 #define OPR_ZERO                4
 123 #define OPR_ONE                 5
 124 #define OPR_UNKNOWN             7
 125
 126 static GLuint translate_operand( GLenum operand )
 127 {
 128    switch (operand) {
 129    case GL_SRC_COLOR: return OPR_SRC_COLOR;
 130    case GL_ONE_MINUS_SRC_COLOR: return OPR_ONE_MINUS_SRC_COLOR;
 131    case GL_SRC_ALPHA: return OPR_SRC_ALPHA;
 132    case GL_ONE_MINUS_SRC_ALPHA: return OPR_ONE_MINUS_SRC_ALPHA;
 133    case GL_ZERO: return OPR_ZERO;
 134    case GL_ONE: return OPR_ONE;
 135    default:
 136       assert(0);
 137       return OPR_UNKNOWN;
 138    }
 139 }
 140
 141 #define SRC_TEXTURE  0
 142 #define SRC_TEXTURE0 1
 143 #define SRC_TEXTURE1 2
 144 #define SRC_TEXTURE2 3
 145 #define SRC_TEXTURE3 4
 146 #define SRC_TEXTURE4 5
 147 #define SRC_TEXTURE5 6
 148 #define SRC_TEXTURE6 7
 149 #define SRC_TEXTURE7 8
 150 #define SRC_CONSTANT 9
 151 #define SRC_PRIMARY_COLOR 10
 152 #define SRC_PREVIOUS 11
 153 #define SRC_ZERO     12
 154 #define SRC_UNKNOWN  15
 155
 156 static GLuint translate_source( GLenum src )
 157 {
 158    switch (src) {
 159    case GL_TEXTURE: return SRC_TEXTURE;
 160    case GL_TEXTURE0:
 161    case GL_TEXTURE1:
 162    case GL_TEXTURE2:
 163    case GL_TEXTURE3:
 164    case GL_TEXTURE4:
 165    case GL_TEXTURE5:
 166    case GL_TEXTURE6:
 167    case GL_TEXTURE7: return SRC_TEXTURE0 + (src - GL_TEXTURE0);
 168    case GL_CONSTANT: return SRC_CONSTANT;
 169    case GL_PRIMARY_COLOR: return SRC_PRIMARY_COLOR;
 170    case GL_PREVIOUS: return SRC_PREVIOUS;
 171    case GL_ZERO:
 172       return SRC_ZERO;
 173    default:
 174       assert(0);
 175       return SRC_UNKNOWN;
 176    }
 177 }
 178
 179 #define MODE_REPLACE                     0  /* r = a0 */
 180 #define MODE_MODULATE                    1  /* r = a0 * a1 */
 181 #define MODE_ADD                         2  /* r = a0 + a1 */
 182 #define MODE_ADD_SIGNED                  3  /* r = a0 + a1 - 0.5 */
 183 #define MODE_INTERPOLATE                 4  /* r = a0 * a2 + a1 * (1 - a2) */
 184 #define MODE_SUBTRACT                    5  /* r = a0 - a1 */
 185 #define MODE_DOT3_RGB                    6  /* r = a0 . a1 */
 186 #define MODE_DOT3_RGB_EXT                7  /* r = a0 . a1 */
 187 #define MODE_DOT3_RGBA                   8  /* r = a0 . a1 */
 188 #define MODE_DOT3_RGBA_EXT               9  /* r = a0 . a1 */
 189 #define MODE_MODULATE_ADD_ATI           10  /* r = a0 * a2 + a1 */
 190 #define MODE_MODULATE_SIGNED_ADD_ATI    11  /* r = a0 * a2 + a1 - 0.5 */
 191 #define MODE_MODULATE_SUBTRACT_ATI      12  /* r = a0 * a2 - a1 */
 192 #define MODE_ADD_PRODUCTS               13  /* r = a0 * a1 + a2 * a3 */
 193 #define MODE_ADD_PRODUCTS_SIGNED        14  /* r = a0 * a1 + a2 * a3 - 0.5 */
 194 #define MODE_BUMP_ENVMAP_ATI            15  /* special */
 195 #define MODE_UNKNOWN                    16
 196
 197 /**
 198  * Translate GL combiner state into a MODE_x value
 199  */
 200 static GLuint translate_mode( GLenum envMode, GLenum mode )
 201 {
 202    switch (mode) {
 203    case GL_REPLACE: return MODE_REPLACE;
 204    case GL_MODULATE: return MODE_MODULATE;
 205    case GL_ADD:
 206       if (envMode == GL_COMBINE4_NV)
 207          return MODE_ADD_PRODUCTS;
 208       else
 209          return MODE_ADD;
 210    case GL_ADD_SIGNED:
 211       if (envMode == GL_COMBINE4_NV)
 212          return MODE_ADD_PRODUCTS_SIGNED;
 213       else
 214          return MODE_ADD_SIGNED;
 215    case GL_INTERPOLATE: return MODE_INTERPOLATE;
 216    case GL_SUBTRACT: return MODE_SUBTRACT;
 217    case GL_DOT3_RGB: return MODE_DOT3_RGB;
 218    case GL_DOT3_RGB_EXT: return MODE_DOT3_RGB_EXT;
 219    case GL_DOT3_RGBA: return MODE_DOT3_RGBA;
 220    case GL_DOT3_RGBA_EXT: return MODE_DOT3_RGBA_EXT;
 221    case GL_MODULATE_ADD_ATI: return MODE_MODULATE_ADD_ATI;
 222    case GL_MODULATE_SIGNED_ADD_ATI: return MODE_MODULATE_SIGNED_ADD_ATI;
 223    case GL_MODULATE_SUBTRACT_ATI: return MODE_MODULATE_SUBTRACT_ATI;
 224    case GL_BUMP_ENVMAP_ATI: return MODE_BUMP_ENVMAP_ATI;
 225    default:
 226       assert(0);
 227       return MODE_UNKNOWN;
 228    }
 229 }
 230
 231 #define TEXTURE_UNKNOWN_INDEX 7
 232 static GLuint translate_tex_src_bit( GLbitfield bit )
 233 {
 234    /* make sure number of switch cases is correct */
 235    assert(NUM_TEXTURE_TARGETS == 7);
 236    switch (bit) {
 237    case TEXTURE_1D_BIT:   return TEXTURE_1D_INDEX;
 238    case TEXTURE_2D_BIT:   return TEXTURE_2D_INDEX;
 239    case TEXTURE_RECT_BIT: return TEXTURE_RECT_INDEX;
 240    case TEXTURE_3D_BIT:   return TEXTURE_3D_INDEX;
 241    case TEXTURE_CUBE_BIT: return TEXTURE_CUBE_INDEX;
 242    case TEXTURE_1D_ARRAY_BIT: return TEXTURE_1D_ARRAY_INDEX;
 243    case TEXTURE_2D_ARRAY_BIT: return TEXTURE_2D_ARRAY_INDEX;
 244    default:
 245       assert(0);
 246       return TEXTURE_UNKNOWN_INDEX;
 247    }
 248 }
 249
 250 #define VERT_BIT_TEX_ANY    (0xff << VERT_ATTRIB_TEX0)
 251 #define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0)
 252
 253 /**
 254  * Identify all possible varying inputs.  The fragment program will
 255  * never reference non-varying inputs, but will track them via state
 256  * constants instead.
 257  *
 258  * This function figures out all the inputs that the fragment program
 259  * has access to.  The bitmask is later reduced to just those which
 260  * are actually referenced.
 261  */
 262 static GLbitfield get_fp_input_mask( GLcontext *ctx )
 263 {
 264    const GLboolean vertexShader = (ctx->Shader.CurrentProgram &&
 265                                    ctx->Shader.CurrentProgram->VertexProgram);
 266    const GLboolean vertexProgram = ctx->VertexProgram._Enabled;
 267    GLbitfield fp_inputs = 0x0;
 268
 269    if (ctx->VertexProgram._Overriden) {
 270       /* Somebody's messing with the vertex program and we don't have
 271        * a clue what's happening.  Assume that it could be producing
 272        * all possible outputs.
 273        */
 274       fp_inputs = ~0;
 275    }
 276    else if (ctx->RenderMode == GL_FEEDBACK) {
 277       fp_inputs = (FRAG_BIT_COL0 | FRAG_BIT_TEX0);
 278    }
 279    else if (!(vertexProgram || vertexShader) ||
 280             !ctx->VertexProgram._Current) {
 281       /* Fixed function vertex logic */
 282       GLbitfield varying_inputs = ctx->varying_vp_inputs;
 283
 284       /* These get generated in the setup routine regardless of the
 285        * vertex program:
 286        */
 287       if (ctx->Point.PointSprite)
 288          varying_inputs |= FRAG_BITS_TEX_ANY;
 289
 290       /* First look at what values may be computed by the generated
 291        * vertex program:
 292        */
 293       if (ctx->Light.Enabled) {
 294          fp_inputs |= FRAG_BIT_COL0;
 295
 296          if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
 297             fp_inputs |= FRAG_BIT_COL1;
 298       }
 299
 300       fp_inputs |= (ctx->Texture._TexGenEnabled |
 301                     ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0;
 302
 303       /* Then look at what might be varying as a result of enabled
 304        * arrays, etc:
 305        */
 306       if (varying_inputs & VERT_BIT_COLOR0) fp_inputs |= FRAG_BIT_COL0;
 307       if (varying_inputs & VERT_BIT_COLOR1) fp_inputs |= FRAG_BIT_COL1;
 308
 309       fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0)
 310                     << FRAG_ATTRIB_TEX0);
 311
 312    }
 313    else {
 314       /* calculate from vp->outputs */
 315       struct gl_vertex_program *vprog;
 316       GLbitfield vp_outputs;
 317
 318       /* Choose GLSL vertex shader over ARB vertex program.  Need this
 319        * since vertex shader state validation comes after fragment state
 320        * validation (see additional comments in state.c).
 321        */
 322       if (vertexShader)
 323          vprog = ctx->Shader.CurrentProgram->VertexProgram;
 324       else
 325          vprog = ctx->VertexProgram._Current;
 326
 327       vp_outputs = vprog->Base.OutputsWritten;
 328
 329       /* These get generated in the setup routine regardless of the
 330        * vertex program:
 331        */
 332       if (ctx->Point.PointSprite)
 333          vp_outputs |= FRAG_BITS_TEX_ANY;
 334
 335       if (vp_outputs & (1 << VERT_RESULT_COL0)) fp_inputs |= FRAG_BIT_COL0;
 336       if (vp_outputs & (1 << VERT_RESULT_COL1)) fp_inputs |= FRAG_BIT_COL1;
 337
 338       fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY) >> VERT_RESULT_TEX0)
 339                     << FRAG_ATTRIB_TEX0);
 340    }
 341
 342    return fp_inputs;
 343 }
 344
 345
 346 /**
 347  * Examine current texture environment state and generate a unique
 348  * key to identify it.
 349  */
 350 static void make_state_key( GLcontext *ctx,  struct state_key *key )
 351 {
 352    GLuint i, j;
 353    GLbitfield inputs_referenced = FRAG_BIT_COL0;
 354    GLbitfield inputs_available = get_fp_input_mask( ctx );
 355
 356    memset(key, 0, sizeof(*key));
 357
 358    for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
 359       const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i];
 360       GLenum format;
 361
 362       if (!texUnit->_ReallyEnabled || !texUnit->Enabled)
 363          continue;
 364
 365       format = texUnit->_Current->Image[0][texUnit->_Current->BaseLevel]->_BaseFormat;
 366
 367       key->unit[i].enabled = 1;
 368       key->enabled_units |= (1<<i);
 369       key->nr_enabled_units = i+1;
 370       inputs_referenced |= FRAG_BIT_TEX(i);
 371
 372       key->unit[i].source_index =
 373          translate_tex_src_bit(texUnit->_ReallyEnabled);
 374       key->unit[i].shadow = ((texUnit->_Current->CompareMode == GL_COMPARE_R_TO_TEXTURE) &&
 375                              ((format == GL_DEPTH_COMPONENT) ||
 376                               (format == GL_DEPTH_STENCIL_EXT)));
 377
 378       key->unit[i].NumArgsRGB = texUnit->_CurrentCombine->_NumArgsRGB;
 379       key->unit[i].NumArgsA = texUnit->_CurrentCombine->_NumArgsA;
 380
 381       key->unit[i].ModeRGB =
 382          translate_mode(texUnit->EnvMode, texUnit->_CurrentCombine->ModeRGB);
 383       key->unit[i].ModeA =
 384          translate_mode(texUnit->EnvMode, texUnit->_CurrentCombine->ModeA);
 385
 386       key->unit[i].ScaleShiftRGB = texUnit->_CurrentCombine->ScaleShiftRGB;
 387       key->unit[i].ScaleShiftA = texUnit->_CurrentCombine->ScaleShiftA;
 388
 389       for (j = 0; j < MAX_COMBINER_TERMS; j++) {
 390          key->unit[i].OptRGB[j].Operand =
 391             translate_operand(texUnit->_CurrentCombine->OperandRGB[j]);
 392          key->unit[i].OptA[j].Operand =
 393             translate_operand(texUnit->_CurrentCombine->OperandA[j]);
 394          key->unit[i].OptRGB[j].Source =
 395             translate_source(texUnit->_CurrentCombine->SourceRGB[j]);
 396          key->unit[i].OptA[j].Source =
 397             translate_source(texUnit->_CurrentCombine->SourceA[j]);
 398       }
 399
 400       if (key->unit[i].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
 401          /* requires some special translation */
 402          key->unit[i].NumArgsRGB = 2;
 403          key->unit[i].ScaleShiftRGB = 0;
 404          key->unit[i].OptRGB[0].Operand = OPR_SRC_COLOR;
 405          key->unit[i].OptRGB[0].Source = SRC_TEXTURE;
 406          key->unit[i].OptRGB[1].Operand = OPR_SRC_COLOR;
 407          key->unit[i].OptRGB[1].Source = texUnit->BumpTarget - GL_TEXTURE0 + SRC_TEXTURE0;
 408        }
 409    }
 410
 411    if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
 412       key->separate_specular = 1;
 413       inputs_referenced |= FRAG_BIT_COL1;
 414    }
 415
 416    if (ctx->Fog.Enabled) {
 417       key->fog_enabled = 1;
 418       key->fog_mode = translate_fog_mode(ctx->Fog.Mode);
 419       inputs_referenced |= FRAG_BIT_FOGC; /* maybe */
 420    }
 421
 422    key->inputs_available = (inputs_available & inputs_referenced);
 423 }
 424
 425 /**
 426  * Use uregs to represent registers internally, translate to Mesa's
 427  * expected formats on emit.
 428  *
 429  * NOTE: These are passed by value extensively in this file rather
 430  * than as usual by pointer reference.  If this disturbs you, try
 431  * remembering they are just 32bits in size.
 432  *
 433  * GCC is smart enough to deal with these dword-sized structures in
 434  * much the same way as if I had defined them as dwords and was using
 435  * macros to access and set the fields.  This is much nicer and easier
 436  * to evolve.
 437  */
 438 struct ureg {
 439    GLuint file:4;
 440    GLuint idx:8;
 441    GLuint negatebase:1;
 442    GLuint abs:1;
 443    GLuint negateabs:1;
 444    GLuint swz:12;
 445    GLuint pad:5;
 446 };
 447
 448 static const struct ureg undef = {
 449    PROGRAM_UNDEFINED,
 450    ~0,
 451    0,
 452    0,
 453    0,
 454    0,
 455    0
 456 };
 457
 458
 459 /** State used to build the fragment program:
 460  */
 461 struct texenv_fragment_program {
 462    struct gl_fragment_program *program;
 463    GLcontext *ctx;
 464    struct state_key *state;
 465
 466    GLbitfield alu_temps;        /**< Track texture indirections, see spec. */
 467    GLbitfield temps_output;     /**< Track texture indirections, see spec. */
 468    GLbitfield temp_in_use;      /**< Tracks temporary regs which are in use. */
 469    GLboolean error;
 470
 471    struct ureg src_texture[MAX_TEXTURE_COORD_UNITS];
 472    /* Reg containing each texture unit's sampled texture color,
 473     * else undef.
 474     */
 475
 476    struct ureg texcoord_tex[MAX_TEXTURE_COORD_UNITS];
 477    /* Reg containing texcoord for a texture unit,
 478     * needed for bump mapping, else undef.
 479     */
 480
 481    struct ureg src_previous;    /**< Reg containing color from previous
 482                                  * stage.  May need to be decl'd.
 483                                  */
 484
 485    GLuint last_tex_stage;       /**< Number of last enabled texture unit */
 486
 487    struct ureg half;
 488    struct ureg one;
 489    struct ureg zero;
 490 };
 491
 492
 493
 494 static struct ureg make_ureg(GLuint file, GLuint idx)
 495 {
 496    struct ureg reg;
 497    reg.file = file;
 498    reg.idx = idx;
 499    reg.negatebase = 0;
 500    reg.abs = 0;
 501    reg.negateabs = 0;
 502    reg.swz = SWIZZLE_NOOP;
 503    reg.pad = 0;
 504    return reg;
 505 }
 506
 507 static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w )
 508 {
 509    reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x),
 510                            GET_SWZ(reg.swz, y),
 511                            GET_SWZ(reg.swz, z),
 512                            GET_SWZ(reg.swz, w));
 513
 514    return reg;
 515 }
 516
 517 static struct ureg swizzle1( struct ureg reg, int x )
 518 {
 519    return swizzle(reg, x, x, x, x);
 520 }
 521
 522 static struct ureg negate( struct ureg reg )
 523 {
 524    reg.negatebase ^= 1;
 525    return reg;
 526 }
 527
 528 static GLboolean is_undef( struct ureg reg )
 529 {
 530    return reg.file == PROGRAM_UNDEFINED;
 531 }
 532
 533
 534 static struct ureg get_temp( struct texenv_fragment_program *p )
 535 {
 536    GLint bit;
 537
 538    /* First try and reuse temps which have been used already:
 539     */
 540    bit = _mesa_ffs( ~p->temp_in_use & p->alu_temps );
 541
 542    /* Then any unused temporary:
 543     */
 544    if (!bit)
 545       bit = _mesa_ffs( ~p->temp_in_use );
 546
 547    if (!bit) {
 548       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
 549       _mesa_exit(1);
 550    }
 551
 552    if ((GLuint) bit > p->program->Base.NumTemporaries)
 553       p->program->Base.NumTemporaries = bit;
 554
 555    p->temp_in_use |= 1<<(bit-1);
 556    return make_ureg(PROGRAM_TEMPORARY, (bit-1));
 557 }
 558
 559 static struct ureg get_tex_temp( struct texenv_fragment_program *p )
 560 {
 561    int bit;
 562
 563    /* First try to find available temp not previously used (to avoid
 564     * starting a new texture indirection).  According to the spec, the
 565     * ~p->temps_output isn't necessary, but will keep it there for
 566     * now:
 567     */
 568    bit = _mesa_ffs( ~p->temp_in_use & ~p->alu_temps & ~p->temps_output );
 569
 570    /* Then any unused temporary:
 571     */
 572    if (!bit)
 573       bit = _mesa_ffs( ~p->temp_in_use );
 574
 575    if (!bit) {
 576       _mesa_problem(NULL, "%s: out of temporaries\n", __FILE__);
 577       _mesa_exit(1);
 578    }
 579
 580    if ((GLuint) bit > p->program->Base.NumTemporaries)
 581       p->program->Base.NumTemporaries = bit;
 582
 583    p->temp_in_use |= 1<<(bit-1);
 584    return make_ureg(PROGRAM_TEMPORARY, (bit-1));
 585 }
 586
 587
 588 /** Mark a temp reg as being no longer allocatable. */
 589 static void reserve_temp( struct texenv_fragment_program *p, struct ureg r )
 590 {
 591    if (r.file == PROGRAM_TEMPORARY)
 592       p->temps_output |= (1 << r.idx);
 593 }
 594
 595
 596 static void release_temps(GLcontext *ctx, struct texenv_fragment_program *p )
 597 {
 598    GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps;
 599
 600    /* KW: To support tex_env_crossbar, don't release the registers in
 601     * temps_output.
 602     */
 603    if (max_temp >= sizeof(int) * 8)
 604       p->temp_in_use = p->temps_output;
 605    else
 606       p->temp_in_use = ~((1<<max_temp)-1) | p->temps_output;
 607 }
 608
 609
 610 static struct ureg register_param5( struct texenv_fragment_program *p,
 611                                     GLint s0,
 612                                     GLint s1,
 613                                     GLint s2,
 614                                     GLint s3,
 615                                     GLint s4)
 616 {
 617    gl_state_index tokens[STATE_LENGTH];
 618    GLuint idx;
 619    tokens[0] = s0;
 620    tokens[1] = s1;
 621    tokens[2] = s2;
 622    tokens[3] = s3;
 623    tokens[4] = s4;
 624    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
 625    return make_ureg(PROGRAM_STATE_VAR, idx);
 626 }
 627
 628
 629 #define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
 630 #define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
 631 #define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
 632 #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
 633
 634 static GLuint frag_to_vert_attrib( GLuint attrib )
 635 {
 636    switch (attrib) {
 637    case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0;
 638    case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1;
 639    default:
 640       assert(attrib >= FRAG_ATTRIB_TEX0);
 641       assert(attrib <= FRAG_ATTRIB_TEX7);
 642       return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0;
 643    }
 644 }
 645
 646
 647 static struct ureg register_input( struct texenv_fragment_program *p, GLuint input )
 648 {
 649    if (p->state->inputs_available & (1<<input)) {
 650       p->program->Base.InputsRead |= (1 << input);
 651       return make_ureg(PROGRAM_INPUT, input);
 652    }
 653    else {
 654       GLuint idx = frag_to_vert_attrib( input );
 655       return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, idx );
 656    }
 657 }
 658
 659
 660 static void emit_arg( struct prog_src_register *reg,
 661                       struct ureg ureg )
 662 {
 663    reg->File = ureg.file;
 664    reg->Index = ureg.idx;
 665    reg->Swizzle = ureg.swz;
 666    reg->NegateBase = ureg.negatebase ? 0xf : 0x0;
 667    reg->Abs = ureg.abs;
 668    reg->NegateAbs = ureg.negateabs;
 669 }
 670
 671 static void emit_dst( struct prog_dst_register *dst,
 672                       struct ureg ureg, GLuint mask )
 673 {
 674    dst->File = ureg.file;
 675    dst->Index = ureg.idx;
 676    dst->WriteMask = mask;
 677    dst->CondMask = COND_TR;  /* always pass cond test */
 678    dst->CondSwizzle = SWIZZLE_NOOP;
 679 }
 680
 681 static struct prog_instruction *
 682 emit_op(struct texenv_fragment_program *p,
 683         enum prog_opcode op,
 684         struct ureg dest,
 685         GLuint mask,
 686         GLboolean saturate,
 687         struct ureg src0,
 688         struct ureg src1,
 689         struct ureg src2 )
 690 {
 691    GLuint nr = p->program->Base.NumInstructions++;
 692    struct prog_instruction *inst = &p->program->Base.Instructions[nr];
 693
 694    assert(nr < MAX_INSTRUCTIONS);
 695
 696    _mesa_init_instructions(inst, 1);
 697    inst->Opcode = op;
 698
 699    emit_arg( &inst->SrcReg[0], src0 );
 700    emit_arg( &inst->SrcReg[1], src1 );
 701    emit_arg( &inst->SrcReg[2], src2 );
 702
 703    inst->SaturateMode = saturate ? SATURATE_ZERO_ONE : SATURATE_OFF;
 704
 705    emit_dst( &inst->DstReg, dest, mask );
 706
 707 #if 0
 708    /* Accounting for indirection tracking:
 709     */
 710    if (dest.file == PROGRAM_TEMPORARY)
 711       p->temps_output |= 1 << dest.idx;
 712 #endif
 713
 714    return inst;
 715 }
 716
 717
 718 static struct ureg emit_arith( struct texenv_fragment_program *p,
 719                                enum prog_opcode op,
 720                                struct ureg dest,
 721                                GLuint mask,
 722                                GLboolean saturate,
 723                                struct ureg src0,
 724                                struct ureg src1,
 725                                struct ureg src2 )
 726 {
 727    emit_op(p, op, dest, mask, saturate, src0, src1, src2);
 728
 729    /* Accounting for indirection tracking:
 730     */
 731    if (src0.file == PROGRAM_TEMPORARY)
 732       p->alu_temps |= 1 << src0.idx;
 733
 734    if (!is_undef(src1) && src1.file == PROGRAM_TEMPORARY)
 735       p->alu_temps |= 1 << src1.idx;
 736
 737    if (!is_undef(src2) && src2.file == PROGRAM_TEMPORARY)
 738       p->alu_temps |= 1 << src2.idx;
 739
 740    if (dest.file == PROGRAM_TEMPORARY)
 741       p->alu_temps |= 1 << dest.idx;
 742
 743    p->program->Base.NumAluInstructions++;
 744    return dest;
 745 }
 746
 747 static struct ureg emit_texld( struct texenv_fragment_program *p,
 748                                enum prog_opcode op,
 749                                struct ureg dest,
 750                                GLuint destmask,
 751                                GLuint tex_unit,
 752                                GLuint tex_idx,
 753                                GLuint tex_shadow,
 754                                struct ureg coord )
 755 {
 756    struct prog_instruction *inst = emit_op( p, op,
 757                                           dest, destmask,
 758                                           GL_FALSE,     /* don't saturate? */
 759                                           coord,        /* arg 0? */
 760                                           undef,
 761                                           undef);
 762
 763    inst->TexSrcTarget = tex_idx;
 764    inst->TexSrcUnit = tex_unit;
 765    inst->TexShadow = tex_shadow;
 766
 767    p->program->Base.NumTexInstructions++;
 768
 769    /* Accounting for indirection tracking:
 770     */
 771    reserve_temp(p, dest);
 772
 773 #if 0
 774    /* Is this a texture indirection?
 775     */
 776    if ((coord.file == PROGRAM_TEMPORARY &&
 777         (p->temps_output & (1<<coord.idx))) ||
 778        (dest.file == PROGRAM_TEMPORARY &&
 779         (p->alu_temps & (1<<dest.idx)))) {
 780       p->program->Base.NumTexIndirections++;
 781       p->temps_output = 1<<coord.idx;
 782       p->alu_temps = 0;
 783       assert(0);                /* KW: texture env crossbar */
 784    }
 785 #endif
 786
 787    return dest;
 788 }
 789
 790
 791 static struct ureg register_const4f( struct texenv_fragment_program *p,
 792                                      GLfloat s0,
 793                                      GLfloat s1,
 794                                      GLfloat s2,
 795                                      GLfloat s3)
 796 {
 797    GLfloat values[4];
 798    GLuint idx, swizzle;
 799    struct ureg r;
 800    values[0] = s0;
 801    values[1] = s1;
 802    values[2] = s2;
 803    values[3] = s3;
 804    idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
 805                                      &swizzle );
 806    r = make_ureg(PROGRAM_CONSTANT, idx);
 807    r.swz = swizzle;
 808    return r;
 809 }
 810
 811 #define register_scalar_const(p, s0)    register_const4f(p, s0, s0, s0, s0)
 812 #define register_const1f(p, s0)         register_const4f(p, s0, 0, 0, 1)
 813 #define register_const2f(p, s0, s1)     register_const4f(p, s0, s1, 0, 1)
 814 #define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1)
 815
 816
 817 static struct ureg get_one( struct texenv_fragment_program *p )
 818 {
 819    if (is_undef(p->one))
 820       p->one = register_scalar_const(p, 1.0);
 821    return p->one;
 822 }
 823
 824 static struct ureg get_half( struct texenv_fragment_program *p )
 825 {
 826    if (is_undef(p->half))
 827       p->half = register_scalar_const(p, 0.5);
 828    return p->half;
 829 }
 830
 831 static struct ureg get_zero( struct texenv_fragment_program *p )
 832 {
 833    if (is_undef(p->zero))
 834       p->zero = register_scalar_const(p, 0.0);
 835    return p->zero;
 836 }
 837
 838
 839 static void program_error( struct texenv_fragment_program *p, const char *msg )
 840 {
 841    _mesa_problem(NULL, msg);
 842    p->error = 1;
 843 }
 844
 845 static struct ureg get_source( struct texenv_fragment_program *p,
 846                                GLuint src, GLuint unit )
 847 {
 848    switch (src) {
 849    case SRC_TEXTURE:
 850       assert(!is_undef(p->src_texture[unit]));
 851       return p->src_texture[unit];
 852
 853    case SRC_TEXTURE0:
 854    case SRC_TEXTURE1:
 855    case SRC_TEXTURE2:
 856    case SRC_TEXTURE3:
 857    case SRC_TEXTURE4:
 858    case SRC_TEXTURE5:
 859    case SRC_TEXTURE6:
 860    case SRC_TEXTURE7:
 861       assert(!is_undef(p->src_texture[src - SRC_TEXTURE0]));
 862       return p->src_texture[src - SRC_TEXTURE0];
 863
 864    case SRC_CONSTANT:
 865       return register_param2(p, STATE_TEXENV_COLOR, unit);
 866
 867    case SRC_PRIMARY_COLOR:
 868       return register_input(p, FRAG_ATTRIB_COL0);
 869
 870    case SRC_ZERO:
 871       return get_zero(p);
 872
 873    case SRC_PREVIOUS:
 874       if (is_undef(p->src_previous))
 875          return register_input(p, FRAG_ATTRIB_COL0);
 876       else
 877          return p->src_previous;
 878
 879    default:
 880       assert(0);
 881    }
 882 }
 883
 884 static struct ureg emit_combine_source( struct texenv_fragment_program *p,
 885                                         GLuint mask,
 886                                         GLuint unit,
 887                                         GLuint source,
 888                                         GLuint operand )
 889 {
 890    struct ureg arg, src, one;
 891
 892    src = get_source(p, source, unit);
 893
 894    switch (operand) {
 895    case OPR_ONE_MINUS_SRC_COLOR:
 896       /* Get unused tmp,
 897        * Emit tmp = 1.0 - arg.xyzw
 898        */
 899       arg = get_temp( p );
 900       one = get_one( p );
 901       return emit_arith( p, OPCODE_SUB, arg, mask, 0, one, src, undef);
 902
 903    case OPR_SRC_ALPHA:
 904       if (mask == WRITEMASK_W)
 905          return src;
 906       else
 907          return swizzle1( src, SWIZZLE_W );
 908    case OPR_ONE_MINUS_SRC_ALPHA:
 909       /* Get unused tmp,
 910        * Emit tmp = 1.0 - arg.wwww
 911        */
 912       arg = get_temp(p);
 913       one = get_one(p);
 914       return emit_arith(p, OPCODE_SUB, arg, mask, 0,
 915                         one, swizzle1(src, SWIZZLE_W), undef);
 916    case OPR_ZERO:
 917       return get_zero(p);
 918    case OPR_ONE:
 919       return get_one(p);
 920    case OPR_SRC_COLOR:
 921       return src;
 922    default:
 923       assert(0);
 924       return src;
 925    }
 926 }
 927
 928 static GLboolean args_match( struct state_key *key, GLuint unit )
 929 {
 930    GLuint i, nr = key->unit[unit].NumArgsRGB;
 931
 932    for (i = 0 ; i < nr ; i++) {
 933       if (key->unit[unit].OptA[i].Source != key->unit[unit].OptRGB[i].Source)
 934          return GL_FALSE;
 935
 936       switch(key->unit[unit].OptA[i].Operand) {
 937       case OPR_SRC_ALPHA:
 938          switch(key->unit[unit].OptRGB[i].Operand) {
 939          case OPR_SRC_COLOR:
 940          case OPR_SRC_ALPHA:
 941             break;
 942          default:
 943             return GL_FALSE;
 944          }
 945          break;
 946       case OPR_ONE_MINUS_SRC_ALPHA:
 947          switch(key->unit[unit].OptRGB[i].Operand) {
 948          case OPR_ONE_MINUS_SRC_COLOR:
 949          case OPR_ONE_MINUS_SRC_ALPHA:
 950             break;
 951          default:
 952             return GL_FALSE;
 953          }
 954          break;
 955       default:
 956          return GL_FALSE;       /* impossible */
 957       }
 958    }
 959
 960    return GL_TRUE;
 961 }
 962
 963 static struct ureg emit_combine( struct texenv_fragment_program *p,
 964                                  struct ureg dest,
 965                                  GLuint mask,
 966                                  GLboolean saturate,
 967                                  GLuint unit,
 968                                  GLuint nr,
 969                                  GLuint mode,
 970                                  const struct mode_opt *opt)
 971 {
 972    struct ureg src[MAX_COMBINER_TERMS];
 973    struct ureg tmp, half;
 974    GLuint i;
 975
 976    assert(nr <= MAX_COMBINER_TERMS);
 977
 978    tmp = undef; /* silence warning (bug 5318) */
 979
 980    for (i = 0; i < nr; i++)
 981       src[i] = emit_combine_source( p, mask, unit, opt[i].Source, opt[i].Operand );
 982
 983    switch (mode) {
 984    case MODE_REPLACE:
 985       if (mask == WRITEMASK_XYZW && !saturate)
 986          return src[0];
 987       else
 988          return emit_arith( p, OPCODE_MOV, dest, mask, saturate, src[0], undef, undef );
 989    case MODE_MODULATE:
 990       return emit_arith( p, OPCODE_MUL, dest, mask, saturate,
 991                          src[0], src[1], undef );
 992    case MODE_ADD:
 993       return emit_arith( p, OPCODE_ADD, dest, mask, saturate,
 994                          src[0], src[1], undef );
 995    case MODE_ADD_SIGNED:
 996       /* tmp = arg0 + arg1
 997        * result = tmp - .5
 998        */
 999       half = get_half(p);
1000       tmp = get_temp( p );
1001       emit_arith( p, OPCODE_ADD, tmp, mask, 0, src[0], src[1], undef );
1002       emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp, half, undef );
1003       return dest;
1004    case MODE_INTERPOLATE:
1005       /* Arg0 * (Arg2) + Arg1 * (1-Arg2) -- note arguments are reordered:
1006        */
1007       return emit_arith( p, OPCODE_LRP, dest, mask, saturate, src[2], src[0], src[1] );
1008
1009    case MODE_SUBTRACT:
1010       return emit_arith( p, OPCODE_SUB, dest, mask, saturate, src[0], src[1], undef );
1011
1012    case MODE_DOT3_RGBA:
1013    case MODE_DOT3_RGBA_EXT:
1014    case MODE_DOT3_RGB_EXT:
1015    case MODE_DOT3_RGB: {
1016       struct ureg tmp0 = get_temp( p );
1017       struct ureg tmp1 = get_temp( p );
1018       struct ureg neg1 = register_scalar_const(p, -1);
1019       struct ureg two  = register_scalar_const(p, 2);
1020
1021       /* tmp0 = 2*src0 - 1
1022        * tmp1 = 2*src1 - 1
1023        *
1024        * dst = tmp0 dot3 tmp1
1025        */
1026       emit_arith( p, OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0,
1027                   two, src[0], neg1);
1028
1029       if (_mesa_memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
1030          tmp1 = tmp0;
1031       else
1032          emit_arith( p, OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0,
1033                      two, src[1], neg1);
1034       emit_arith( p, OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
1035       return dest;
1036    }
1037    case MODE_MODULATE_ADD_ATI:
1038       /* Arg0 * Arg2 + Arg1 */
1039       return emit_arith( p, OPCODE_MAD, dest, mask, saturate,
1040                          src[0], src[2], src[1] );
1041    case MODE_MODULATE_SIGNED_ADD_ATI: {
1042       /* Arg0 * Arg2 + Arg1 - 0.5 */
1043       struct ureg tmp0 = get_temp(p);
1044       half = get_half(p);
1045       emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
1046       emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1047       return dest;
1048    }
1049    case MODE_MODULATE_SUBTRACT_ATI:
1050       /* Arg0 * Arg2 - Arg1 */
1051       emit_arith( p, OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
1052       return dest;
1053    case MODE_ADD_PRODUCTS:
1054       /* Arg0 * Arg1 + Arg2 * Arg3 */
1055       {
1056          struct ureg tmp0 = get_temp(p);
1057          emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1058          emit_arith( p, OPCODE_MAD, dest, mask, saturate, src[2], src[3], tmp0 );
1059       }
1060       return dest;
1061    case MODE_ADD_PRODUCTS_SIGNED:
1062       /* Arg0 * Arg1 + Arg2 * Arg3 - 0.5 */
1063       {
1064          struct ureg tmp0 = get_temp(p);
1065          half = get_half(p);
1066          emit_arith( p, OPCODE_MUL, tmp0, mask, 0, src[0], src[1], undef );
1067          emit_arith( p, OPCODE_MAD, tmp0, mask, 0, src[2], src[3], tmp0 );
1068          emit_arith( p, OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
1069       }
1070       return dest;
1071    case MODE_BUMP_ENVMAP_ATI:
1072       /* special - not handled here */
1073       assert(0);
1074       return src[0];
1075    default:
1076       assert(0);
1077       return src[0];
1078    }
1079 }
1080
1081
1082 /**
1083  * Generate instructions for one texture unit's env/combiner mode.
1084  */
1085 static struct ureg
1086 emit_texenv(struct texenv_fragment_program *p, GLuint unit)
1087 {
1088    struct state_key *key = p->state;
1089    GLboolean saturate = (unit < p->last_tex_stage);
1090    GLuint rgb_shift, alpha_shift;
1091    struct ureg out, shift;
1092    struct ureg dest;
1093
1094    if (!key->unit[unit].enabled) {
1095       return get_source(p, SRC_PREVIOUS, 0);
1096    }
1097    if (key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1098       /* this isn't really a env stage delivering a color and handled elsewhere */
1099       return get_source(p, SRC_PREVIOUS, 0);
1100    }
1101
1102    switch (key->unit[unit].ModeRGB) {
1103    case MODE_DOT3_RGB_EXT:
1104       alpha_shift = key->unit[unit].ScaleShiftA;
1105       rgb_shift = 0;
1106       break;
1107    case MODE_DOT3_RGBA_EXT:
1108       alpha_shift = 0;
1109       rgb_shift = 0;
1110       break;
1111    default:
1112       rgb_shift = key->unit[unit].ScaleShiftRGB;
1113       alpha_shift = key->unit[unit].ScaleShiftA;
1114       break;
1115    }
1116
1117    /* If this is the very last calculation, emit direct to output reg:
1118     */
1119    if (key->separate_specular ||
1120        unit != p->last_tex_stage ||
1121        alpha_shift ||
1122        rgb_shift)
1123       dest = get_temp( p );
1124    else
1125       dest = make_ureg(PROGRAM_OUTPUT, FRAG_RESULT_COLOR);
1126
1127    /* Emit the RGB and A combine ops
1128     */
1129    if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
1130        args_match(key, unit)) {
1131       out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
1132                           unit,
1133                           key->unit[unit].NumArgsRGB,
1134                           key->unit[unit].ModeRGB,
1135                           key->unit[unit].OptRGB);
1136    }
1137    else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
1138             key->unit[unit].ModeRGB == MODE_DOT3_RGBA) {
1139
1140       out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
1141                           unit,
1142                           key->unit[unit].NumArgsRGB,
1143                           key->unit[unit].ModeRGB,
1144                           key->unit[unit].OptRGB);
1145    }
1146    else {
1147       /* Need to do something to stop from re-emitting identical
1148        * argument calculations here:
1149        */
1150       out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
1151                           unit,
1152                           key->unit[unit].NumArgsRGB,
1153                           key->unit[unit].ModeRGB,
1154                           key->unit[unit].OptRGB);
1155       out = emit_combine( p, dest, WRITEMASK_W, saturate,
1156                           unit,
1157                           key->unit[unit].NumArgsA,
1158                           key->unit[unit].ModeA,
1159                           key->unit[unit].OptA);
1160    }
1161
1162    /* Deal with the final shift:
1163     */
1164    if (alpha_shift || rgb_shift) {
1165       if (rgb_shift == alpha_shift) {
1166          shift = register_scalar_const(p, (GLfloat)(1<<rgb_shift));
1167       }
1168       else {
1169          shift = register_const4f(p,
1170                                   (GLfloat)(1<<rgb_shift),
1171                                   (GLfloat)(1<<rgb_shift),
1172                                   (GLfloat)(1<<rgb_shift),
1173                                   (GLfloat)(1<<alpha_shift));
1174       }
1175       return emit_arith( p, OPCODE_MUL, dest, WRITEMASK_XYZW,
1176                          saturate, out, shift, undef );
1177    }
1178    else
1179       return out;
1180 }
1181
1182
1183 /**
1184  * Generate instruction for getting a texture source term.
1185  */
1186 static void load_texture( struct texenv_fragment_program *p, GLuint unit )
1187 {
1188    if (is_undef(p->src_texture[unit])) {
1189       GLuint texTarget = p->state->unit[unit].source_index;
1190       struct ureg texcoord;
1191       struct ureg tmp = get_tex_temp( p );
1192
1193       if (is_undef(p->texcoord_tex[unit])) {
1194          texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
1195       }
1196       else {
1197          /* might want to reuse this reg for tex output actually */
1198          texcoord = p->texcoord_tex[unit];
1199       }
1200
1201       if (texTarget == TEXTURE_UNKNOWN_INDEX)
1202          program_error(p, "TexSrcBit");
1203
1204       /* TODO: Use D0_MASK_XY where possible.
1205        */
1206       if (p->state->unit[unit].enabled) {
1207          GLboolean shadow = GL_FALSE;
1208
1209          if (p->state->unit[unit].shadow) {
1210             p->program->Base.ShadowSamplers |= 1 << unit;
1211             shadow = GL_TRUE;
1212          }
1213
1214          p->src_texture[unit] = emit_texld( p, OPCODE_TXP,
1215                                             tmp, WRITEMASK_XYZW,
1216                                             unit, texTarget, shadow,
1217                                             texcoord );
1218
1219          p->program->Base.SamplersUsed |= (1 << unit);
1220          /* This identity mapping should already be in place
1221           * (see _mesa_init_program_struct()) but let's be safe.
1222           */
1223          p->program->Base.SamplerUnits[unit] = unit;
1224       }
1225       else
1226          p->src_texture[unit] = get_zero(p);
1227    }
1228 }
1229
1230 static GLboolean load_texenv_source( struct texenv_fragment_program *p,
1231                                      GLuint src, GLuint unit )
1232 {
1233    switch (src) {
1234    case SRC_TEXTURE:
1235       load_texture(p, unit);
1236       break;
1237
1238    case SRC_TEXTURE0:
1239    case SRC_TEXTURE1:
1240    case SRC_TEXTURE2:
1241    case SRC_TEXTURE3:
1242    case SRC_TEXTURE4:
1243    case SRC_TEXTURE5:
1244    case SRC_TEXTURE6:
1245    case SRC_TEXTURE7:
1246       load_texture(p, src - SRC_TEXTURE0);
1247       break;
1248
1249    default:
1250       /* not a texture src - do nothing */
1251       break;
1252    }
1253
1254    return GL_TRUE;
1255 }
1256
1257
1258 /**
1259  * Generate instructions for loading all texture source terms.
1260  */
1261 static GLboolean
1262 load_texunit_sources( struct texenv_fragment_program *p, int unit )
1263 {
1264    struct state_key *key = p->state;
1265    GLuint i;
1266
1267    for (i = 0; i < key->unit[unit].NumArgsRGB; i++) {
1268       load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit );
1269    }
1270
1271    for (i = 0; i < key->unit[unit].NumArgsA; i++) {
1272       load_texenv_source( p, key->unit[unit].OptA[i].Source, unit );
1273    }
1274
1275    return GL_TRUE;
1276 }
1277
1278 /**
1279  * Generate instructions for loading bump map textures.
1280  */
1281 static GLboolean
1282 load_texunit_bumpmap( struct texenv_fragment_program *p, int unit )
1283 {
1284    struct state_key *key = p->state;
1285    GLuint bumpedUnitNr = key->unit[unit].OptRGB[1].Source - SRC_TEXTURE0;
1286    struct ureg texcDst, bumpMapRes;
1287    struct ureg constdudvcolor = register_const4f(p, 0.0, 0.0, 0.0, 1.0);
1288    struct ureg texcSrc = register_input(p, FRAG_ATTRIB_TEX0 + bumpedUnitNr);
1289    struct ureg rotMat0 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_0, unit );
1290    struct ureg rotMat1 = register_param3( p, STATE_INTERNAL, STATE_ROT_MATRIX_1, unit );
1291
1292    load_texenv_source( p, unit + SRC_TEXTURE0, unit );
1293
1294    bumpMapRes = get_source(p, key->unit[unit].OptRGB[0].Source, unit);
1295    texcDst = get_tex_temp( p );
1296    p->texcoord_tex[bumpedUnitNr] = texcDst;
1297
1298    /* apply rot matrix and add coords to be available in next phase */
1299    /* dest = (Arg0.xxxx * rotMat0 + Arg1) + (Arg0.yyyy * rotMat1) */
1300    /* note only 2 coords are affected the rest are left unchanged (mul by 0) */
1301    emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1302                swizzle1(bumpMapRes, SWIZZLE_X), rotMat0, texcSrc );
1303    emit_arith( p, OPCODE_MAD, texcDst, WRITEMASK_XYZW, 0,
1304                swizzle1(bumpMapRes, SWIZZLE_Y), rotMat1, texcDst );
1305
1306    /* move 0,0,0,1 into bumpmap src if someone (crossbar) is foolish
1307       enough to access this later, should optimize away */
1308    emit_arith( p, OPCODE_MOV, bumpMapRes, WRITEMASK_XYZW, 0, constdudvcolor, undef, undef );
1309
1310    return GL_TRUE;
1311 }
1312
1313 /**
1314  * Generate a new fragment program which implements the context's
1315  * current texture env/combine mode.
1316  */
1317 static void
1318 create_new_program(GLcontext *ctx, struct state_key *key,
1319                    struct gl_fragment_program *program)
1320 {
1321    struct prog_instruction instBuffer[MAX_INSTRUCTIONS];
1322    struct texenv_fragment_program p;
1323    GLuint unit;
1324    struct ureg cf, out;
1325
1326    _mesa_memset(&p, 0, sizeof(p));
1327    p.ctx = ctx;
1328    p.state = key;
1329    p.program = program;
1330
1331    /* During code generation, use locally-allocated instruction buffer,
1332     * then alloc dynamic storage below.
1333     */
1334    p.program->Base.Instructions = instBuffer;
1335    p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
1336    p.program->Base.NumTexIndirections = 1;
1337    p.program->Base.NumTexInstructions = 0;
1338    p.program->Base.NumAluInstructions = 0;
1339    p.program->Base.String = NULL;
1340    p.program->Base.NumInstructions =
1341    p.program->Base.NumTemporaries =
1342    p.program->Base.NumParameters =
1343    p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
1344    p.program->Base.Parameters = _mesa_new_parameter_list();
1345
1346    p.program->Base.InputsRead = 0;
1347    p.program->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR;
1348
1349    for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++) {
1350       p.src_texture[unit] = undef;
1351       p.texcoord_tex[unit] = undef;
1352    }
1353
1354    p.src_previous = undef;
1355    p.half = undef;
1356    p.zero = undef;
1357    p.one = undef;
1358
1359    p.last_tex_stage = 0;
1360    release_temps(ctx, &p);
1361
1362    if (key->enabled_units) {
1363        GLboolean needbumpstage = GL_FALSE;
1364       /* Zeroth pass - bump map textures first */
1365       for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1366          if (key->unit[unit].enabled && key->unit[unit].ModeRGB == MODE_BUMP_ENVMAP_ATI) {
1367             needbumpstage = GL_TRUE;
1368             load_texunit_bumpmap( &p, unit );
1369          }
1370       if (needbumpstage)
1371          p.program->Base.NumTexIndirections++;
1372
1373       /* First pass - to support texture_env_crossbar, first identify
1374        * all referenced texture sources and emit texld instructions
1375        * for each:
1376        */
1377       for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
1378          if (key->unit[unit].enabled) {
1379             load_texunit_sources( &p, unit );
1380             p.last_tex_stage = unit;
1381          }
1382
1383       /* Second pass - emit combine instructions to build final color:
1384        */
1385       for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
1386          if (key->enabled_units & (1<<unit)) {
1387             p.src_previous = emit_texenv( &p, unit );
1388             reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */
1389             release_temps(ctx, &p);     /* release all temps */
1390          }
1391    }
1392
1393    cf = get_source( &p, SRC_PREVIOUS, 0 );
1394    out = make_ureg( PROGRAM_OUTPUT, FRAG_RESULT_COLOR );
1395
1396    if (key->separate_specular) {
1397       /* Emit specular add.
1398        */
1399       struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
1400       emit_arith( &p, OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
1401       emit_arith( &p, OPCODE_MOV, out, WRITEMASK_W, 0, cf, undef, undef );
1402    }
1403    else if (_mesa_memcmp(&cf, &out, sizeof(cf)) != 0) {
1404       /* Will wind up in here if no texture enabled or a couple of
1405        * other scenarios (GL_REPLACE for instance).
1406        */
1407       emit_arith( &p, OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
1408    }
1409
1410    /* Finish up:
1411     */
1412    emit_arith( &p, OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);
1413
1414    if (key->fog_enabled) {
1415       /* Pull fog mode from GLcontext, the value in the state key is
1416        * a reduced value and not what is expected in FogOption
1417        */
1418       p.program->FogOption = ctx->Fog.Mode;
1419       p.program->Base.InputsRead |= FRAG_BIT_FOGC; /* XXX new */
1420    } else
1421       p.program->FogOption = GL_NONE;
1422
1423    if (p.program->Base.NumTexIndirections > ctx->Const.FragmentProgram.MaxTexIndirections)
1424       program_error(&p, "Exceeded max nr indirect texture lookups");
1425
1426    if (p.program->Base.NumTexInstructions > ctx->Const.FragmentProgram.MaxTexInstructions)
1427       program_error(&p, "Exceeded max TEX instructions");
1428
1429    if (p.program->Base.NumAluInstructions > ctx->Const.FragmentProgram.MaxAluInstructions)
1430       program_error(&p, "Exceeded max ALU instructions");
1431
1432    ASSERT(p.program->Base.NumInstructions <= MAX_INSTRUCTIONS);
1433
1434    /* Allocate final instruction array */
1435    p.program->Base.Instructions
1436       = _mesa_alloc_instructions(p.program->Base.NumInstructions);
1437    if (!p.program->Base.Instructions) {
1438       _mesa_error(ctx, GL_OUT_OF_MEMORY,
1439                   "generating tex env program");
1440       return;
1441    }
1442    _mesa_copy_instructions(p.program->Base.Instructions, instBuffer,
1443                            p.program->Base.NumInstructions);
1444
1445    if (p.program->FogOption) {
1446       _mesa_append_fog_code(ctx, p.program);
1447       p.program->FogOption = GL_NONE;
1448    }
1449
1450
1451    /* Notify driver the fragment program has (actually) changed.
1452     */
1453    if (ctx->Driver.ProgramStringNotify) {
1454       ctx->Driver.ProgramStringNotify( ctx, GL_FRAGMENT_PROGRAM_ARB,
1455                                        &p.program->Base );
1456    }
1457
1458    if (DISASSEM) {
1459       _mesa_print_program(&p.program->Base);
1460       _mesa_printf("\n");
1461    }
1462 }
1463
1464
1465 /**
1466  * Return a fragment program which implements the current
1467  * fixed-function texture, fog and color-sum operations.
1468  */
1469 struct gl_fragment_program *
1470 _mesa_get_fixed_func_fragment_program(GLcontext *ctx)
1471 {
1472    struct gl_fragment_program *prog;
1473    struct state_key key;
1474
1475    make_state_key(ctx, &key);
1476
1477    prog = (struct gl_fragment_program *)
1478       _mesa_search_program_cache(ctx->FragmentProgram.Cache,
1479                                  &key, sizeof(key));
1480
1481    if (!prog) {
1482       prog = (struct gl_fragment_program *)
1483          ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
1484
1485       create_new_program(ctx, &key, prog);
1486
1487       _mesa_program_cache_insert(ctx, ctx->FragmentProgram.Cache,
1488                                  &key, sizeof(key), &prog->Base);
1489    }
1490
1491    return prog;
1492 }