src/mesa/program/program.c

   1 /*
   2  * Mesa 3-D graphics library
   3  *
   4  * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included
  14  * in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  17  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 /**
  26  * \file program.c
  27  * Vertex and fragment program support functions.
  28  * \author Brian Paul
  29  */
  30
  31
  32 #include "main/glheader.h"
  33 #include "main/context.h"
  34 #include "main/hash.h"
  35 #include "main/macros.h"
  36 #include "program.h"
  37 #include "prog_cache.h"
  38 #include "prog_parameter.h"
  39 #include "prog_instruction.h"
  40
  41
  42 /**
  43  * A pointer to this dummy program is put into the hash table when
  44  * glGenPrograms is called.
  45  */
  46 struct gl_program _mesa_DummyProgram;
  47
  48
  49 /**
  50  * Init context's vertex/fragment program state
  51  */
  52 void
  53 _mesa_init_program(struct gl_context *ctx)
  54 {
  55    /*
  56     * If this assertion fails, we need to increase the field
  57     * size for register indexes (see INST_INDEX_BITS).
  58     */
  59    ASSERT(ctx->Const.VertexProgram.MaxUniformComponents / 4
  60           <= (1 << INST_INDEX_BITS));
  61    ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents / 4
  62           <= (1 << INST_INDEX_BITS));
  63
  64    ASSERT(ctx->Const.VertexProgram.MaxTemps <= (1 << INST_INDEX_BITS));
  65    ASSERT(ctx->Const.VertexProgram.MaxLocalParams <= (1 << INST_INDEX_BITS));
  66    ASSERT(ctx->Const.FragmentProgram.MaxTemps <= (1 << INST_INDEX_BITS));
  67    ASSERT(ctx->Const.FragmentProgram.MaxLocalParams <= (1 << INST_INDEX_BITS));
  68
  69    ASSERT(ctx->Const.VertexProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS);
  70    ASSERT(ctx->Const.FragmentProgram.MaxUniformComponents <= 4 * MAX_UNIFORMS);
  71
  72    ASSERT(ctx->Const.VertexProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS));
  73    ASSERT(ctx->Const.FragmentProgram.MaxAddressOffset <= (1 << INST_INDEX_BITS));
  74
  75    /* If this fails, increase prog_instruction::TexSrcUnit size */
  76    STATIC_ASSERT(MAX_TEXTURE_UNITS <= (1 << 5));
  77
  78    /* If this fails, increase prog_instruction::TexSrcTarget size */
  79    STATIC_ASSERT(NUM_TEXTURE_TARGETS <= (1 << 4));
  80
  81    ctx->Program.ErrorPos = -1;
  82    ctx->Program.ErrorString = _mesa_strdup("");
  83
  84    ctx->VertexProgram.Enabled = GL_FALSE;
  85    ctx->VertexProgram.PointSizeEnabled =
  86       (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE;
  87    ctx->VertexProgram.TwoSideEnabled = GL_FALSE;
  88    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
  89                             ctx->Shared->DefaultVertexProgram);
  90    assert(ctx->VertexProgram.Current);
  91    ctx->VertexProgram.Cache = _mesa_new_program_cache();
  92
  93    ctx->FragmentProgram.Enabled = GL_FALSE;
  94    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
  95                             ctx->Shared->DefaultFragmentProgram);
  96    assert(ctx->FragmentProgram.Current);
  97    ctx->FragmentProgram.Cache = _mesa_new_program_cache();
  98
  99    ctx->GeometryProgram.Enabled = GL_FALSE;
 100    /* right now by default we don't have a geometry program */
 101    _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current,
 102                             NULL);
 103    ctx->GeometryProgram.Cache = _mesa_new_program_cache();
 104
 105    /* XXX probably move this stuff */
 106    ctx->ATIFragmentShader.Enabled = GL_FALSE;
 107    ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
 108    assert(ctx->ATIFragmentShader.Current);
 109    ctx->ATIFragmentShader.Current->RefCount++;
 110 }
 111
 112
 113 /**
 114  * Free a context's vertex/fragment program state
 115  */
 116 void
 117 _mesa_free_program_data(struct gl_context *ctx)
 118 {
 119    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL);
 120    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 121    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL);
 122    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 123    _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, NULL);
 124    _mesa_delete_program_cache(ctx, ctx->GeometryProgram.Cache);
 125
 126    /* XXX probably move this stuff */
 127    if (ctx->ATIFragmentShader.Current) {
 128       ctx->ATIFragmentShader.Current->RefCount--;
 129       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 130          free(ctx->ATIFragmentShader.Current);
 131       }
 132    }
 133
 134    free((void *) ctx->Program.ErrorString);
 135 }
 136
 137
 138 /**
 139  * Update the default program objects in the given context to reference those
 140  * specified in the shared state and release those referencing the old
 141  * shared state.
 142  */
 143 void
 144 _mesa_update_default_objects_program(struct gl_context *ctx)
 145 {
 146    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
 147                             ctx->Shared->DefaultVertexProgram);
 148    assert(ctx->VertexProgram.Current);
 149
 150    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
 151                             ctx->Shared->DefaultFragmentProgram);
 152    assert(ctx->FragmentProgram.Current);
 153
 154    _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current,
 155                       ctx->Shared->DefaultGeometryProgram);
 156
 157    /* XXX probably move this stuff */
 158    if (ctx->ATIFragmentShader.Current) {
 159       ctx->ATIFragmentShader.Current->RefCount--;
 160       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 161          free(ctx->ATIFragmentShader.Current);
 162       }
 163    }
 164    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 165    assert(ctx->ATIFragmentShader.Current);
 166    ctx->ATIFragmentShader.Current->RefCount++;
 167 }
 168
 169
 170 /**
 171  * Set the vertex/fragment program error state (position and error string).
 172  * This is generally called from within the parsers.
 173  */
 174 void
 175 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 176 {
 177    ctx->Program.ErrorPos = pos;
 178    free((void *) ctx->Program.ErrorString);
 179    if (!string)
 180       string = "";
 181    ctx->Program.ErrorString = _mesa_strdup(string);
 182 }
 183
 184
 185 /**
 186  * Find the line number and column for 'pos' within 'string'.
 187  * Return a copy of the line which contains 'pos'.  Free the line with
 188  * free().
 189  * \param string  the program string
 190  * \param pos     the position within the string
 191  * \param line    returns the line number corresponding to 'pos'.
 192  * \param col     returns the column number corresponding to 'pos'.
 193  * \return copy of the line containing 'pos'.
 194  */
 195 const GLubyte *
 196 _mesa_find_line_column(const GLubyte *string, const GLubyte *pos,
 197                        GLint *line, GLint *col)
 198 {
 199    const GLubyte *lineStart = string;
 200    const GLubyte *p = string;
 201    GLubyte *s;
 202    int len;
 203
 204    *line = 1;
 205
 206    while (p != pos) {
 207       if (*p == (GLubyte) '\n') {
 208          (*line)++;
 209          lineStart = p + 1;
 210       }
 211       p++;
 212    }
 213
 214    *col = (pos - lineStart) + 1;
 215
 216    /* return copy of this line */
 217    while (*p != 0 && *p != '\n')
 218       p++;
 219    len = p - lineStart;
 220    s = malloc(len + 1);
 221    memcpy(s, lineStart, len);
 222    s[len] = 0;
 223
 224    return s;
 225 }
 226
 227
 228 /**
 229  * Initialize a new vertex/fragment program object.
 230  */
 231 static struct gl_program *
 232 _mesa_init_program_struct( struct gl_context *ctx, struct gl_program *prog,
 233                            GLenum target, GLuint id)
 234 {
 235    (void) ctx;
 236    if (prog) {
 237       GLuint i;
 238       memset(prog, 0, sizeof(*prog));
 239       prog->Id = id;
 240       prog->Target = target;
 241       prog->RefCount = 1;
 242       prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 243
 244       /* default mapping from samplers to texture units */
 245       for (i = 0; i < MAX_SAMPLERS; i++)
 246          prog->SamplerUnits[i] = i;
 247    }
 248
 249    return prog;
 250 }
 251
 252
 253 /**
 254  * Initialize a new fragment program object.
 255  */
 256 struct gl_program *
 257 _mesa_init_fragment_program( struct gl_context *ctx, struct gl_fragment_program *prog,
 258                              GLenum target, GLuint id)
 259 {
 260    if (prog)
 261       return _mesa_init_program_struct( ctx, &prog->Base, target, id );
 262    else
 263       return NULL;
 264 }
 265
 266
 267 /**
 268  * Initialize a new vertex program object.
 269  */
 270 struct gl_program *
 271 _mesa_init_vertex_program( struct gl_context *ctx, struct gl_vertex_program *prog,
 272                            GLenum target, GLuint id)
 273 {
 274    if (prog)
 275       return _mesa_init_program_struct( ctx, &prog->Base, target, id );
 276    else
 277       return NULL;
 278 }
 279
 280
 281 /**
 282  * Initialize a new geometry program object.
 283  */
 284 struct gl_program *
 285 _mesa_init_geometry_program( struct gl_context *ctx, struct gl_geometry_program *prog,
 286                              GLenum target, GLuint id)
 287 {
 288    if (prog)
 289       return _mesa_init_program_struct( ctx, &prog->Base, target, id );
 290    else
 291       return NULL;
 292 }
 293
 294
 295 /**
 296  * Allocate and initialize a new fragment/vertex program object but
 297  * don't put it into the program hash table.  Called via
 298  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 299  * device driver function to implement OO deriviation with additional
 300  * types not understood by this function.
 301  *
 302  * \param ctx  context
 303  * \param id   program id/number
 304  * \param target  program target/type
 305  * \return  pointer to new program object
 306  */
 307 struct gl_program *
 308 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id)
 309 {
 310    struct gl_program *prog;
 311    switch (target) {
 312    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 313       prog = _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 314                                        target, id );
 315       break;
 316    case GL_FRAGMENT_PROGRAM_NV:
 317    case GL_FRAGMENT_PROGRAM_ARB:
 318       prog =_mesa_init_fragment_program(ctx,
 319                                          CALLOC_STRUCT(gl_fragment_program),
 320                                          target, id );
 321       break;
 322    case MESA_GEOMETRY_PROGRAM:
 323       prog = _mesa_init_geometry_program(ctx,
 324                                          CALLOC_STRUCT(gl_geometry_program),
 325                                          target, id);
 326       break;
 327    default:
 328       _mesa_problem(ctx, "bad target in _mesa_new_program");
 329       prog = NULL;
 330    }
 331    return prog;
 332 }
 333
 334
 335 /**
 336  * Delete a program and remove it from the hash table, ignoring the
 337  * reference count.
 338  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 339  * by a device driver function.
 340  */
 341 void
 342 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 343 {
 344    (void) ctx;
 345    ASSERT(prog);
 346    ASSERT(prog->RefCount==0);
 347
 348    if (prog == &_mesa_DummyProgram)
 349       return;
 350
 351    free(prog->String);
 352    free(prog->LocalParams);
 353
 354    if (prog->Instructions) {
 355       _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
 356    }
 357    if (prog->Parameters) {
 358       _mesa_free_parameter_list(prog->Parameters);
 359    }
 360
 361    free(prog);
 362 }
 363
 364
 365 /**
 366  * Return the gl_program object for a given ID.
 367  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 368  * casts elsewhere.
 369  */
 370 struct gl_program *
 371 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 372 {
 373    if (id)
 374       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 375    else
 376       return NULL;
 377 }
 378
 379
 380 /**
 381  * Reference counting for vertex/fragment programs
 382  * This is normally only called from the _mesa_reference_program() macro
 383  * when there's a real pointer change.
 384  */
 385 void
 386 _mesa_reference_program_(struct gl_context *ctx,
 387                          struct gl_program **ptr,
 388                          struct gl_program *prog)
 389 {
 390 #ifndef NDEBUG
 391    assert(ptr);
 392    if (*ptr && prog) {
 393       /* sanity check */
 394       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 395          ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB);
 396       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 397          ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 398                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 399       else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM)
 400          ASSERT(prog->Target == MESA_GEOMETRY_PROGRAM);
 401    }
 402 #endif
 403
 404    if (*ptr) {
 405       GLboolean deleteFlag;
 406
 407       /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/
 408 #if 0
 409       printf("Program %p ID=%u Target=%s  Refcount-- to %d\n",
 410              *ptr, (*ptr)->Id,
 411              ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 412               ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 413              (*ptr)->RefCount - 1);
 414 #endif
 415       ASSERT((*ptr)->RefCount > 0);
 416       (*ptr)->RefCount--;
 417
 418       deleteFlag = ((*ptr)->RefCount == 0);
 419       /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/
 420
 421       if (deleteFlag) {
 422          ASSERT(ctx);
 423          ctx->Driver.DeleteProgram(ctx, *ptr);
 424       }
 425
 426       *ptr = NULL;
 427    }
 428
 429    assert(!*ptr);
 430    if (prog) {
 431       /*_glthread_LOCK_MUTEX(prog->Mutex);*/
 432       prog->RefCount++;
 433 #if 0
 434       printf("Program %p ID=%u Target=%s  Refcount++ to %d\n",
 435              prog, prog->Id,
 436              (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 437               (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 438              prog->RefCount);
 439 #endif
 440       /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/
 441    }
 442
 443    *ptr = prog;
 444 }
 445
 446
 447 /**
 448  * Return a copy of a program.
 449  * XXX Problem here if the program object is actually OO-derivation
 450  * made by a device driver.
 451  */
 452 struct gl_program *
 453 _mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog)
 454 {
 455    struct gl_program *clone;
 456
 457    clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id);
 458    if (!clone)
 459       return NULL;
 460
 461    assert(clone->Target == prog->Target);
 462    assert(clone->RefCount == 1);
 463
 464    clone->String = (GLubyte *) _mesa_strdup((char *) prog->String);
 465    clone->Format = prog->Format;
 466    clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions);
 467    if (!clone->Instructions) {
 468       _mesa_reference_program(ctx, &clone, NULL);
 469       return NULL;
 470    }
 471    _mesa_copy_instructions(clone->Instructions, prog->Instructions,
 472                            prog->NumInstructions);
 473    clone->InputsRead = prog->InputsRead;
 474    clone->OutputsWritten = prog->OutputsWritten;
 475    clone->SamplersUsed = prog->SamplersUsed;
 476    clone->ShadowSamplers = prog->ShadowSamplers;
 477    memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed));
 478
 479    if (prog->Parameters)
 480       clone->Parameters = _mesa_clone_parameter_list(prog->Parameters);
 481    if (prog->LocalParams) {
 482       clone->LocalParams = malloc(MAX_PROGRAM_LOCAL_PARAMS *
 483                                   sizeof(float[4]));
 484       if (!clone->LocalParams) {
 485          _mesa_reference_program(ctx, &clone, NULL);
 486          return NULL;
 487       }
 488       memcpy(clone->LocalParams, prog->LocalParams,
 489              MAX_PROGRAM_LOCAL_PARAMS * sizeof(float[4]));
 490    }
 491    clone->IndirectRegisterFiles = prog->IndirectRegisterFiles;
 492    clone->NumInstructions = prog->NumInstructions;
 493    clone->NumTemporaries = prog->NumTemporaries;
 494    clone->NumParameters = prog->NumParameters;
 495    clone->NumAttributes = prog->NumAttributes;
 496    clone->NumAddressRegs = prog->NumAddressRegs;
 497    clone->NumNativeInstructions = prog->NumNativeInstructions;
 498    clone->NumNativeTemporaries = prog->NumNativeTemporaries;
 499    clone->NumNativeParameters = prog->NumNativeParameters;
 500    clone->NumNativeAttributes = prog->NumNativeAttributes;
 501    clone->NumNativeAddressRegs = prog->NumNativeAddressRegs;
 502    clone->NumAluInstructions = prog->NumAluInstructions;
 503    clone->NumTexInstructions = prog->NumTexInstructions;
 504    clone->NumTexIndirections = prog->NumTexIndirections;
 505    clone->NumNativeAluInstructions = prog->NumNativeAluInstructions;
 506    clone->NumNativeTexInstructions = prog->NumNativeTexInstructions;
 507    clone->NumNativeTexIndirections = prog->NumNativeTexIndirections;
 508
 509    switch (prog->Target) {
 510    case GL_VERTEX_PROGRAM_ARB:
 511       {
 512          const struct gl_vertex_program *vp = gl_vertex_program_const(prog);
 513          struct gl_vertex_program *vpc = gl_vertex_program(clone);
 514          vpc->IsPositionInvariant = vp->IsPositionInvariant;
 515       }
 516       break;
 517    case GL_FRAGMENT_PROGRAM_ARB:
 518       {
 519          const struct gl_fragment_program *fp = gl_fragment_program_const(prog);
 520          struct gl_fragment_program *fpc = gl_fragment_program(clone);
 521          fpc->UsesKill = fp->UsesKill;
 522          fpc->UsesDFdy = fp->UsesDFdy;
 523          fpc->OriginUpperLeft = fp->OriginUpperLeft;
 524          fpc->PixelCenterInteger = fp->PixelCenterInteger;
 525       }
 526       break;
 527    case MESA_GEOMETRY_PROGRAM:
 528       {
 529          const struct gl_geometry_program *gp = gl_geometry_program_const(prog);
 530          struct gl_geometry_program *gpc = gl_geometry_program(clone);
 531          gpc->VerticesOut = gp->VerticesOut;
 532          gpc->InputType = gp->InputType;
 533          gpc->OutputType = gp->OutputType;
 534       }
 535       break;
 536    default:
 537       _mesa_problem(NULL, "Unexpected target in _mesa_clone_program");
 538    }
 539
 540    return clone;
 541 }
 542
 543
 544 /**
 545  * Insert 'count' NOP instructions at 'start' in the given program.
 546  * Adjust branch targets accordingly.
 547  */
 548 GLboolean
 549 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 550 {
 551    const GLuint origLen = prog->NumInstructions;
 552    const GLuint newLen = origLen + count;
 553    struct prog_instruction *newInst;
 554    GLuint i;
 555
 556    /* adjust branches */
 557    for (i = 0; i < prog->NumInstructions; i++) {
 558       struct prog_instruction *inst = prog->Instructions + i;
 559       if (inst->BranchTarget > 0) {
 560          if ((GLuint)inst->BranchTarget >= start) {
 561             inst->BranchTarget += count;
 562          }
 563       }
 564    }
 565
 566    /* Alloc storage for new instructions */
 567    newInst = _mesa_alloc_instructions(newLen);
 568    if (!newInst) {
 569       return GL_FALSE;
 570    }
 571
 572    /* Copy 'start' instructions into new instruction buffer */
 573    _mesa_copy_instructions(newInst, prog->Instructions, start);
 574
 575    /* init the new instructions */
 576    _mesa_init_instructions(newInst + start, count);
 577
 578    /* Copy the remaining/tail instructions to new inst buffer */
 579    _mesa_copy_instructions(newInst + start + count,
 580                            prog->Instructions + start,
 581                            origLen - start);
 582
 583    /* free old instructions */
 584    _mesa_free_instructions(prog->Instructions, origLen);
 585
 586    /* install new instructions */
 587    prog->Instructions = newInst;
 588    prog->NumInstructions = newLen;
 589
 590    return GL_TRUE;
 591 }
 592
 593 /**
 594  * Delete 'count' instructions at 'start' in the given program.
 595  * Adjust branch targets accordingly.
 596  */
 597 GLboolean
 598 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
 599 {
 600    const GLuint origLen = prog->NumInstructions;
 601    const GLuint newLen = origLen - count;
 602    struct prog_instruction *newInst;
 603    GLuint i;
 604
 605    /* adjust branches */
 606    for (i = 0; i < prog->NumInstructions; i++) {
 607       struct prog_instruction *inst = prog->Instructions + i;
 608       if (inst->BranchTarget > 0) {
 609          if (inst->BranchTarget > (GLint) start) {
 610             inst->BranchTarget -= count;
 611          }
 612       }
 613    }
 614
 615    /* Alloc storage for new instructions */
 616    newInst = _mesa_alloc_instructions(newLen);
 617    if (!newInst) {
 618       return GL_FALSE;
 619    }
 620
 621    /* Copy 'start' instructions into new instruction buffer */
 622    _mesa_copy_instructions(newInst, prog->Instructions, start);
 623
 624    /* Copy the remaining/tail instructions to new inst buffer */
 625    _mesa_copy_instructions(newInst + start,
 626                            prog->Instructions + start + count,
 627                            newLen - start);
 628
 629    /* free old instructions */
 630    _mesa_free_instructions(prog->Instructions, origLen);
 631
 632    /* install new instructions */
 633    prog->Instructions = newInst;
 634    prog->NumInstructions = newLen;
 635
 636    return GL_TRUE;
 637 }
 638
 639
 640 /**
 641  * Search instructions for registers that match (oldFile, oldIndex),
 642  * replacing them with (newFile, newIndex).
 643  */
 644 static void
 645 replace_registers(struct prog_instruction *inst, GLuint numInst,
 646                   GLuint oldFile, GLuint oldIndex,
 647                   GLuint newFile, GLuint newIndex)
 648 {
 649    GLuint i, j;
 650    for (i = 0; i < numInst; i++) {
 651       /* src regs */
 652       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 653          if (inst[i].SrcReg[j].File == oldFile &&
 654              inst[i].SrcReg[j].Index == oldIndex) {
 655             inst[i].SrcReg[j].File = newFile;
 656             inst[i].SrcReg[j].Index = newIndex;
 657          }
 658       }
 659       /* dst reg */
 660       if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) {
 661          inst[i].DstReg.File = newFile;
 662          inst[i].DstReg.Index = newIndex;
 663       }
 664    }
 665 }
 666
 667
 668 /**
 669  * Search instructions for references to program parameters.  When found,
 670  * increment the parameter index by 'offset'.
 671  * Used when combining programs.
 672  */
 673 static void
 674 adjust_param_indexes(struct prog_instruction *inst, GLuint numInst,
 675                      GLuint offset)
 676 {
 677    GLuint i, j;
 678    for (i = 0; i < numInst; i++) {
 679       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 680          GLuint f = inst[i].SrcReg[j].File;
 681          if (f == PROGRAM_CONSTANT ||
 682              f == PROGRAM_UNIFORM ||
 683              f == PROGRAM_STATE_VAR) {
 684             inst[i].SrcReg[j].Index += offset;
 685          }
 686       }
 687    }
 688 }
 689
 690
 691 /**
 692  * Combine two programs into one.  Fix instructions so the outputs of
 693  * the first program go to the inputs of the second program.
 694  */
 695 struct gl_program *
 696 _mesa_combine_programs(struct gl_context *ctx,
 697                        const struct gl_program *progA,
 698                        const struct gl_program *progB)
 699 {
 700    struct prog_instruction *newInst;
 701    struct gl_program *newProg;
 702    const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
 703    const GLuint lenB = progB->NumInstructions;
 704    const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
 705    const GLuint newLength = lenA + lenB;
 706    GLboolean usedTemps[MAX_PROGRAM_TEMPS];
 707    GLuint firstTemp = 0;
 708    GLbitfield64 inputsB;
 709    GLuint i;
 710
 711    ASSERT(progA->Target == progB->Target);
 712
 713    newInst = _mesa_alloc_instructions(newLength);
 714    if (!newInst)
 715       return GL_FALSE;
 716
 717    _mesa_copy_instructions(newInst, progA->Instructions, lenA);
 718    _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
 719
 720    /* adjust branch / instruction addresses for B's instructions */
 721    for (i = 0; i < lenB; i++) {
 722       newInst[lenA + i].BranchTarget += lenA;
 723    }
 724
 725    newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
 726    newProg->Instructions = newInst;
 727    newProg->NumInstructions = newLength;
 728
 729    /* find used temp regs (we may need new temps below) */
 730    _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY,
 731                              usedTemps, MAX_PROGRAM_TEMPS);
 732
 733    if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
 734       const struct gl_fragment_program *fprogA, *fprogB;
 735       struct gl_fragment_program *newFprog;
 736       GLbitfield64 progB_inputsRead = progB->InputsRead;
 737       GLint progB_colorFile, progB_colorIndex;
 738
 739       fprogA = gl_fragment_program_const(progA);
 740       fprogB = gl_fragment_program_const(progB);
 741       newFprog = gl_fragment_program(newProg);
 742
 743       newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
 744       newFprog->UsesDFdy = fprogA->UsesDFdy || fprogB->UsesDFdy;
 745
 746       /* We'll do a search and replace for instances
 747        * of progB_colorFile/progB_colorIndex below...
 748        */
 749       progB_colorFile = PROGRAM_INPUT;
 750       progB_colorIndex = VARYING_SLOT_COL0;
 751
 752       /*
 753        * The fragment program may get color from a state var rather than
 754        * a fragment input (vertex output) if it's constant.
 755        * See the texenvprogram.c code.
 756        * So, search the program's parameter list now to see if the program
 757        * gets color from a state var instead of a conventional fragment
 758        * input register.
 759        */
 760       for (i = 0; i < progB->Parameters->NumParameters; i++) {
 761          struct gl_program_parameter *p = &progB->Parameters->Parameters[i];
 762          if (p->Type == PROGRAM_STATE_VAR &&
 763              p->StateIndexes[0] == STATE_INTERNAL &&
 764              p->StateIndexes[1] == STATE_CURRENT_ATTRIB &&
 765              (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) {
 766             progB_inputsRead |= VARYING_BIT_COL0;
 767             progB_colorFile = PROGRAM_STATE_VAR;
 768             progB_colorIndex = i;
 769             break;
 770          }
 771       }
 772
 773       /* Connect color outputs of fprogA to color inputs of fprogB, via a
 774        * new temporary register.
 775        */
 776       if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) &&
 777           (progB_inputsRead & VARYING_BIT_COL0)) {
 778          GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS,
 779                                                   firstTemp);
 780          if (tempReg < 0) {
 781             _mesa_problem(ctx, "No free temp regs found in "
 782                           "_mesa_combine_programs(), using 31");
 783             tempReg = 31;
 784          }
 785          firstTemp = tempReg + 1;
 786
 787          /* replace writes to result.color[0] with tempReg */
 788          replace_registers(newInst, lenA,
 789                            PROGRAM_OUTPUT, FRAG_RESULT_COLOR,
 790                            PROGRAM_TEMPORARY, tempReg);
 791          /* replace reads from the input color with tempReg */
 792          replace_registers(newInst + lenA, lenB,
 793                            progB_colorFile, progB_colorIndex, /* search for */
 794                            PROGRAM_TEMPORARY, tempReg  /* replace with */ );
 795       }
 796
 797       /* compute combined program's InputsRead */
 798       inputsB = progB_inputsRead;
 799       if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) {
 800          inputsB &= ~(1 << VARYING_SLOT_COL0);
 801       }
 802       newProg->InputsRead = progA->InputsRead | inputsB;
 803       newProg->OutputsWritten = progB->OutputsWritten;
 804       newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
 805    }
 806    else {
 807       /* vertex program */
 808       assert(0);      /* XXX todo */
 809    }
 810
 811    /*
 812     * Merge parameters (uniforms, constants, etc)
 813     */
 814    newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
 815                                                        progB->Parameters);
 816
 817    adjust_param_indexes(newInst + lenA, lenB, numParamsA);
 818
 819
 820    return newProg;
 821 }
 822
 823
 824 /**
 825  * Populate the 'used' array with flags indicating which registers (TEMPs,
 826  * INPUTs, OUTPUTs, etc, are used by the given program.
 827  * \param file  type of register to scan for
 828  * \param used  returns true/false flags for in use / free
 829  * \param usedSize  size of the 'used' array
 830  */
 831 void
 832 _mesa_find_used_registers(const struct gl_program *prog,
 833                           gl_register_file file,
 834                           GLboolean used[], GLuint usedSize)
 835 {
 836    GLuint i, j;
 837
 838    memset(used, 0, usedSize);
 839
 840    for (i = 0; i < prog->NumInstructions; i++) {
 841       const struct prog_instruction *inst = prog->Instructions + i;
 842       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 843
 844       if (inst->DstReg.File == file) {
 845          ASSERT(inst->DstReg.Index < usedSize);
 846          if(inst->DstReg.Index < usedSize)
 847             used[inst->DstReg.Index] = GL_TRUE;
 848       }
 849
 850       for (j = 0; j < n; j++) {
 851          if (inst->SrcReg[j].File == file) {
 852             ASSERT(inst->SrcReg[j].Index < (GLint) usedSize);
 853             if (inst->SrcReg[j].Index < (GLint) usedSize)
 854                used[inst->SrcReg[j].Index] = GL_TRUE;
 855          }
 856       }
 857    }
 858 }
 859
 860
 861 /**
 862  * Scan the given 'used' register flag array for the first entry
 863  * that's >= firstReg.
 864  * \param used  vector of flags indicating registers in use (as returned
 865  *              by _mesa_find_used_registers())
 866  * \param usedSize  size of the 'used' array
 867  * \param firstReg  first register to start searching at
 868  * \return index of unused register, or -1 if none.
 869  */
 870 GLint
 871 _mesa_find_free_register(const GLboolean used[],
 872                          GLuint usedSize, GLuint firstReg)
 873 {
 874    GLuint i;
 875
 876    assert(firstReg < usedSize);
 877
 878    for (i = firstReg; i < usedSize; i++)
 879       if (!used[i])
 880          return i;
 881
 882    return -1;
 883 }
 884
 885
 886
 887 /**
 888  * Check if the given register index is valid (doesn't exceed implementation-
 889  * dependent limits).
 890  * \return GL_TRUE if OK, GL_FALSE if bad index
 891  */
 892 GLboolean
 893 _mesa_valid_register_index(const struct gl_context *ctx,
 894                            gl_shader_stage shaderType,
 895                            gl_register_file file, GLint index)
 896 {
 897    const struct gl_program_constants *c;
 898
 899    switch (shaderType) {
 900    case MESA_SHADER_VERTEX:
 901       c = &ctx->Const.VertexProgram;
 902       break;
 903    case MESA_SHADER_FRAGMENT:
 904       c = &ctx->Const.FragmentProgram;
 905       break;
 906    case MESA_SHADER_GEOMETRY:
 907       c = &ctx->Const.GeometryProgram;
 908       break;
 909    default:
 910       _mesa_problem(ctx,
 911                     "unexpected shader type in _mesa_valid_register_index()");
 912       return GL_FALSE;
 913    }
 914
 915    switch (file) {
 916    case PROGRAM_UNDEFINED:
 917       return GL_TRUE;  /* XXX or maybe false? */
 918
 919    case PROGRAM_TEMPORARY:
 920       return index >= 0 && index < (GLint) c->MaxTemps;
 921
 922    case PROGRAM_UNIFORM:
 923    case PROGRAM_STATE_VAR:
 924       /* aka constant buffer */
 925       return index >= 0 && index < (GLint) c->MaxUniformComponents / 4;
 926
 927    case PROGRAM_CONSTANT:
 928       /* constant buffer w/ possible relative negative addressing */
 929       return (index > (int) c->MaxUniformComponents / -4 &&
 930               index < (int) c->MaxUniformComponents / 4);
 931
 932    case PROGRAM_INPUT:
 933       if (index < 0)
 934          return GL_FALSE;
 935
 936       switch (shaderType) {
 937       case MESA_SHADER_VERTEX:
 938          return index < VERT_ATTRIB_GENERIC0 + (GLint) c->MaxAttribs;
 939       case MESA_SHADER_FRAGMENT:
 940          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 941       case MESA_SHADER_GEOMETRY:
 942          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 943       default:
 944          return GL_FALSE;
 945       }
 946
 947    case PROGRAM_OUTPUT:
 948       if (index < 0)
 949          return GL_FALSE;
 950
 951       switch (shaderType) {
 952       case MESA_SHADER_VERTEX:
 953          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 954       case MESA_SHADER_FRAGMENT:
 955          return index < FRAG_RESULT_DATA0 + (GLint) ctx->Const.MaxDrawBuffers;
 956       case MESA_SHADER_GEOMETRY:
 957          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 958       default:
 959          return GL_FALSE;
 960       }
 961
 962    case PROGRAM_ADDRESS:
 963       return index >= 0 && index < (GLint) c->MaxAddressRegs;
 964
 965    default:
 966       _mesa_problem(ctx,
 967                     "unexpected register file in _mesa_valid_register_index()");
 968       return GL_FALSE;
 969    }
 970 }
 971
 972
 973
 974 /**
 975  * "Post-process" a GPU program.  This is intended to be used for debugging.
 976  * Example actions include no-op'ing instructions or changing instruction
 977  * behaviour.
 978  */
 979 void
 980 _mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog)
 981 {
 982    static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 };
 983    GLuint i;
 984    GLuint whiteSwizzle;
 985    GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters,
 986                                                  (gl_constant_value *) white,
 987                                                  4, &whiteSwizzle);
 988
 989    (void) whiteIndex;
 990
 991    for (i = 0; i < prog->NumInstructions; i++) {
 992       struct prog_instruction *inst = prog->Instructions + i;
 993       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 994
 995       (void) n;
 996
 997       if (_mesa_is_tex_instruction(inst->Opcode)) {
 998 #if 0
 999          /* replace TEX/TXP/TXB with MOV */
1000          inst->Opcode = OPCODE_MOV;
1001          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1002          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1003          inst->SrcReg[0].Negate = NEGATE_NONE;
1004 #endif
1005
1006 #if 0
1007          /* disable shadow texture mode */
1008          inst->TexShadow = 0;
1009 #endif
1010       }
1011
1012       if (inst->Opcode == OPCODE_TXP) {
1013 #if 0
1014          inst->Opcode = OPCODE_MOV;
1015          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1016          inst->SrcReg[0].File = PROGRAM_CONSTANT;
1017          inst->SrcReg[0].Index = whiteIndex;
1018          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1019          inst->SrcReg[0].Negate = NEGATE_NONE;
1020 #endif
1021 #if 0
1022          inst->TexShadow = 0;
1023 #endif
1024 #if 0
1025          inst->Opcode = OPCODE_TEX;
1026          inst->TexShadow = 0;
1027 #endif
1028       }
1029
1030    }
1031 }
1032
1033 /* Gets the minimum number of shader invocations per fragment.
1034  * This function is useful to determine if we need to do per
1035  * sample shading or per fragment shading.
1036  */
1037 GLint
1038 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
1039                                        const struct gl_fragment_program *prog)
1040 {
1041    /* From ARB_sample_shading specification:
1042     * "Using gl_SampleID in a fragment shader causes the entire shader
1043     *  to be evaluated per-sample."
1044     *
1045     * "Using gl_SamplePosition in a fragment shader causes the entire
1046     *  shader to be evaluated per-sample."
1047     *
1048     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
1049     *  has no effect."
1050     */
1051    if (ctx->Multisample.Enabled) {
1052       /* The ARB_gpu_shader5 specification says:
1053        *
1054        * "Use of the "sample" qualifier on a fragment shader input
1055        *  forces per-sample shading"
1056        */
1057       if (prog->IsSample)
1058          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1059
1060       if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
1061                                          SYSTEM_BIT_SAMPLE_POS))
1062          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1063       else if (ctx->Multisample.SampleShading)
1064          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
1065                           ctx->DrawBuffer->Visual.samples), 1);
1066       else
1067          return 1;
1068    }
1069    return 1;
1070 }