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
 353    if (prog->Instructions) {
 354       _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
 355    }
 356    if (prog->Parameters) {
 357       _mesa_free_parameter_list(prog->Parameters);
 358    }
 359
 360    free(prog);
 361 }
 362
 363
 364 /**
 365  * Return the gl_program object for a given ID.
 366  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 367  * casts elsewhere.
 368  */
 369 struct gl_program *
 370 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 371 {
 372    if (id)
 373       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 374    else
 375       return NULL;
 376 }
 377
 378
 379 /**
 380  * Reference counting for vertex/fragment programs
 381  * This is normally only called from the _mesa_reference_program() macro
 382  * when there's a real pointer change.
 383  */
 384 void
 385 _mesa_reference_program_(struct gl_context *ctx,
 386                          struct gl_program **ptr,
 387                          struct gl_program *prog)
 388 {
 389 #ifndef NDEBUG
 390    assert(ptr);
 391    if (*ptr && prog) {
 392       /* sanity check */
 393       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 394          ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB);
 395       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 396          ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 397                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 398       else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM)
 399          ASSERT(prog->Target == MESA_GEOMETRY_PROGRAM);
 400    }
 401 #endif
 402
 403    if (*ptr) {
 404       GLboolean deleteFlag;
 405
 406       /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/
 407 #if 0
 408       printf("Program %p ID=%u Target=%s  Refcount-- to %d\n",
 409              *ptr, (*ptr)->Id,
 410              ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 411               ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 412              (*ptr)->RefCount - 1);
 413 #endif
 414       ASSERT((*ptr)->RefCount > 0);
 415       (*ptr)->RefCount--;
 416
 417       deleteFlag = ((*ptr)->RefCount == 0);
 418       /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/
 419
 420       if (deleteFlag) {
 421          ASSERT(ctx);
 422          ctx->Driver.DeleteProgram(ctx, *ptr);
 423       }
 424
 425       *ptr = NULL;
 426    }
 427
 428    assert(!*ptr);
 429    if (prog) {
 430       /*_glthread_LOCK_MUTEX(prog->Mutex);*/
 431       prog->RefCount++;
 432 #if 0
 433       printf("Program %p ID=%u Target=%s  Refcount++ to %d\n",
 434              prog, prog->Id,
 435              (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 436               (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 437              prog->RefCount);
 438 #endif
 439       /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/
 440    }
 441
 442    *ptr = prog;
 443 }
 444
 445
 446 /**
 447  * Return a copy of a program.
 448  * XXX Problem here if the program object is actually OO-derivation
 449  * made by a device driver.
 450  */
 451 struct gl_program *
 452 _mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog)
 453 {
 454    struct gl_program *clone;
 455
 456    clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id);
 457    if (!clone)
 458       return NULL;
 459
 460    assert(clone->Target == prog->Target);
 461    assert(clone->RefCount == 1);
 462
 463    clone->String = (GLubyte *) _mesa_strdup((char *) prog->String);
 464    clone->Format = prog->Format;
 465    clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions);
 466    if (!clone->Instructions) {
 467       _mesa_reference_program(ctx, &clone, NULL);
 468       return NULL;
 469    }
 470    _mesa_copy_instructions(clone->Instructions, prog->Instructions,
 471                            prog->NumInstructions);
 472    clone->InputsRead = prog->InputsRead;
 473    clone->OutputsWritten = prog->OutputsWritten;
 474    clone->SamplersUsed = prog->SamplersUsed;
 475    clone->ShadowSamplers = prog->ShadowSamplers;
 476    memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed));
 477
 478    if (prog->Parameters)
 479       clone->Parameters = _mesa_clone_parameter_list(prog->Parameters);
 480    memcpy(clone->LocalParams, prog->LocalParams, sizeof(clone->LocalParams));
 481    clone->IndirectRegisterFiles = prog->IndirectRegisterFiles;
 482    clone->NumInstructions = prog->NumInstructions;
 483    clone->NumTemporaries = prog->NumTemporaries;
 484    clone->NumParameters = prog->NumParameters;
 485    clone->NumAttributes = prog->NumAttributes;
 486    clone->NumAddressRegs = prog->NumAddressRegs;
 487    clone->NumNativeInstructions = prog->NumNativeInstructions;
 488    clone->NumNativeTemporaries = prog->NumNativeTemporaries;
 489    clone->NumNativeParameters = prog->NumNativeParameters;
 490    clone->NumNativeAttributes = prog->NumNativeAttributes;
 491    clone->NumNativeAddressRegs = prog->NumNativeAddressRegs;
 492    clone->NumAluInstructions = prog->NumAluInstructions;
 493    clone->NumTexInstructions = prog->NumTexInstructions;
 494    clone->NumTexIndirections = prog->NumTexIndirections;
 495    clone->NumNativeAluInstructions = prog->NumNativeAluInstructions;
 496    clone->NumNativeTexInstructions = prog->NumNativeTexInstructions;
 497    clone->NumNativeTexIndirections = prog->NumNativeTexIndirections;
 498
 499    switch (prog->Target) {
 500    case GL_VERTEX_PROGRAM_ARB:
 501       {
 502          const struct gl_vertex_program *vp = gl_vertex_program_const(prog);
 503          struct gl_vertex_program *vpc = gl_vertex_program(clone);
 504          vpc->IsPositionInvariant = vp->IsPositionInvariant;
 505       }
 506       break;
 507    case GL_FRAGMENT_PROGRAM_ARB:
 508       {
 509          const struct gl_fragment_program *fp = gl_fragment_program_const(prog);
 510          struct gl_fragment_program *fpc = gl_fragment_program(clone);
 511          fpc->UsesKill = fp->UsesKill;
 512          fpc->UsesDFdy = fp->UsesDFdy;
 513          fpc->OriginUpperLeft = fp->OriginUpperLeft;
 514          fpc->PixelCenterInteger = fp->PixelCenterInteger;
 515       }
 516       break;
 517    case MESA_GEOMETRY_PROGRAM:
 518       {
 519          const struct gl_geometry_program *gp = gl_geometry_program_const(prog);
 520          struct gl_geometry_program *gpc = gl_geometry_program(clone);
 521          gpc->VerticesOut = gp->VerticesOut;
 522          gpc->InputType = gp->InputType;
 523          gpc->OutputType = gp->OutputType;
 524       }
 525       break;
 526    default:
 527       _mesa_problem(NULL, "Unexpected target in _mesa_clone_program");
 528    }
 529
 530    return clone;
 531 }
 532
 533
 534 /**
 535  * Insert 'count' NOP instructions at 'start' in the given program.
 536  * Adjust branch targets accordingly.
 537  */
 538 GLboolean
 539 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 540 {
 541    const GLuint origLen = prog->NumInstructions;
 542    const GLuint newLen = origLen + count;
 543    struct prog_instruction *newInst;
 544    GLuint i;
 545
 546    /* adjust branches */
 547    for (i = 0; i < prog->NumInstructions; i++) {
 548       struct prog_instruction *inst = prog->Instructions + i;
 549       if (inst->BranchTarget > 0) {
 550          if ((GLuint)inst->BranchTarget >= start) {
 551             inst->BranchTarget += count;
 552          }
 553       }
 554    }
 555
 556    /* Alloc storage for new instructions */
 557    newInst = _mesa_alloc_instructions(newLen);
 558    if (!newInst) {
 559       return GL_FALSE;
 560    }
 561
 562    /* Copy 'start' instructions into new instruction buffer */
 563    _mesa_copy_instructions(newInst, prog->Instructions, start);
 564
 565    /* init the new instructions */
 566    _mesa_init_instructions(newInst + start, count);
 567
 568    /* Copy the remaining/tail instructions to new inst buffer */
 569    _mesa_copy_instructions(newInst + start + count,
 570                            prog->Instructions + start,
 571                            origLen - start);
 572
 573    /* free old instructions */
 574    _mesa_free_instructions(prog->Instructions, origLen);
 575
 576    /* install new instructions */
 577    prog->Instructions = newInst;
 578    prog->NumInstructions = newLen;
 579
 580    return GL_TRUE;
 581 }
 582
 583 /**
 584  * Delete 'count' instructions at 'start' in the given program.
 585  * Adjust branch targets accordingly.
 586  */
 587 GLboolean
 588 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
 589 {
 590    const GLuint origLen = prog->NumInstructions;
 591    const GLuint newLen = origLen - count;
 592    struct prog_instruction *newInst;
 593    GLuint i;
 594
 595    /* adjust branches */
 596    for (i = 0; i < prog->NumInstructions; i++) {
 597       struct prog_instruction *inst = prog->Instructions + i;
 598       if (inst->BranchTarget > 0) {
 599          if (inst->BranchTarget > (GLint) start) {
 600             inst->BranchTarget -= count;
 601          }
 602       }
 603    }
 604
 605    /* Alloc storage for new instructions */
 606    newInst = _mesa_alloc_instructions(newLen);
 607    if (!newInst) {
 608       return GL_FALSE;
 609    }
 610
 611    /* Copy 'start' instructions into new instruction buffer */
 612    _mesa_copy_instructions(newInst, prog->Instructions, start);
 613
 614    /* Copy the remaining/tail instructions to new inst buffer */
 615    _mesa_copy_instructions(newInst + start,
 616                            prog->Instructions + start + count,
 617                            newLen - start);
 618
 619    /* free old instructions */
 620    _mesa_free_instructions(prog->Instructions, origLen);
 621
 622    /* install new instructions */
 623    prog->Instructions = newInst;
 624    prog->NumInstructions = newLen;
 625
 626    return GL_TRUE;
 627 }
 628
 629
 630 /**
 631  * Search instructions for registers that match (oldFile, oldIndex),
 632  * replacing them with (newFile, newIndex).
 633  */
 634 static void
 635 replace_registers(struct prog_instruction *inst, GLuint numInst,
 636                   GLuint oldFile, GLuint oldIndex,
 637                   GLuint newFile, GLuint newIndex)
 638 {
 639    GLuint i, j;
 640    for (i = 0; i < numInst; i++) {
 641       /* src regs */
 642       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 643          if (inst[i].SrcReg[j].File == oldFile &&
 644              inst[i].SrcReg[j].Index == oldIndex) {
 645             inst[i].SrcReg[j].File = newFile;
 646             inst[i].SrcReg[j].Index = newIndex;
 647          }
 648       }
 649       /* dst reg */
 650       if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) {
 651          inst[i].DstReg.File = newFile;
 652          inst[i].DstReg.Index = newIndex;
 653       }
 654    }
 655 }
 656
 657
 658 /**
 659  * Search instructions for references to program parameters.  When found,
 660  * increment the parameter index by 'offset'.
 661  * Used when combining programs.
 662  */
 663 static void
 664 adjust_param_indexes(struct prog_instruction *inst, GLuint numInst,
 665                      GLuint offset)
 666 {
 667    GLuint i, j;
 668    for (i = 0; i < numInst; i++) {
 669       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 670          GLuint f = inst[i].SrcReg[j].File;
 671          if (f == PROGRAM_CONSTANT ||
 672              f == PROGRAM_UNIFORM ||
 673              f == PROGRAM_STATE_VAR) {
 674             inst[i].SrcReg[j].Index += offset;
 675          }
 676       }
 677    }
 678 }
 679
 680
 681 /**
 682  * Combine two programs into one.  Fix instructions so the outputs of
 683  * the first program go to the inputs of the second program.
 684  */
 685 struct gl_program *
 686 _mesa_combine_programs(struct gl_context *ctx,
 687                        const struct gl_program *progA,
 688                        const struct gl_program *progB)
 689 {
 690    struct prog_instruction *newInst;
 691    struct gl_program *newProg;
 692    const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
 693    const GLuint lenB = progB->NumInstructions;
 694    const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
 695    const GLuint newLength = lenA + lenB;
 696    GLboolean usedTemps[MAX_PROGRAM_TEMPS];
 697    GLuint firstTemp = 0;
 698    GLbitfield64 inputsB;
 699    GLuint i;
 700
 701    ASSERT(progA->Target == progB->Target);
 702
 703    newInst = _mesa_alloc_instructions(newLength);
 704    if (!newInst)
 705       return GL_FALSE;
 706
 707    _mesa_copy_instructions(newInst, progA->Instructions, lenA);
 708    _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
 709
 710    /* adjust branch / instruction addresses for B's instructions */
 711    for (i = 0; i < lenB; i++) {
 712       newInst[lenA + i].BranchTarget += lenA;
 713    }
 714
 715    newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
 716    newProg->Instructions = newInst;
 717    newProg->NumInstructions = newLength;
 718
 719    /* find used temp regs (we may need new temps below) */
 720    _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY,
 721                              usedTemps, MAX_PROGRAM_TEMPS);
 722
 723    if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
 724       const struct gl_fragment_program *fprogA, *fprogB;
 725       struct gl_fragment_program *newFprog;
 726       GLbitfield64 progB_inputsRead = progB->InputsRead;
 727       GLint progB_colorFile, progB_colorIndex;
 728
 729       fprogA = gl_fragment_program_const(progA);
 730       fprogB = gl_fragment_program_const(progB);
 731       newFprog = gl_fragment_program(newProg);
 732
 733       newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
 734       newFprog->UsesDFdy = fprogA->UsesDFdy || fprogB->UsesDFdy;
 735
 736       /* We'll do a search and replace for instances
 737        * of progB_colorFile/progB_colorIndex below...
 738        */
 739       progB_colorFile = PROGRAM_INPUT;
 740       progB_colorIndex = VARYING_SLOT_COL0;
 741
 742       /*
 743        * The fragment program may get color from a state var rather than
 744        * a fragment input (vertex output) if it's constant.
 745        * See the texenvprogram.c code.
 746        * So, search the program's parameter list now to see if the program
 747        * gets color from a state var instead of a conventional fragment
 748        * input register.
 749        */
 750       for (i = 0; i < progB->Parameters->NumParameters; i++) {
 751          struct gl_program_parameter *p = &progB->Parameters->Parameters[i];
 752          if (p->Type == PROGRAM_STATE_VAR &&
 753              p->StateIndexes[0] == STATE_INTERNAL &&
 754              p->StateIndexes[1] == STATE_CURRENT_ATTRIB &&
 755              (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) {
 756             progB_inputsRead |= VARYING_BIT_COL0;
 757             progB_colorFile = PROGRAM_STATE_VAR;
 758             progB_colorIndex = i;
 759             break;
 760          }
 761       }
 762
 763       /* Connect color outputs of fprogA to color inputs of fprogB, via a
 764        * new temporary register.
 765        */
 766       if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) &&
 767           (progB_inputsRead & VARYING_BIT_COL0)) {
 768          GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS,
 769                                                   firstTemp);
 770          if (tempReg < 0) {
 771             _mesa_problem(ctx, "No free temp regs found in "
 772                           "_mesa_combine_programs(), using 31");
 773             tempReg = 31;
 774          }
 775          firstTemp = tempReg + 1;
 776
 777          /* replace writes to result.color[0] with tempReg */
 778          replace_registers(newInst, lenA,
 779                            PROGRAM_OUTPUT, FRAG_RESULT_COLOR,
 780                            PROGRAM_TEMPORARY, tempReg);
 781          /* replace reads from the input color with tempReg */
 782          replace_registers(newInst + lenA, lenB,
 783                            progB_colorFile, progB_colorIndex, /* search for */
 784                            PROGRAM_TEMPORARY, tempReg  /* replace with */ );
 785       }
 786
 787       /* compute combined program's InputsRead */
 788       inputsB = progB_inputsRead;
 789       if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) {
 790          inputsB &= ~(1 << VARYING_SLOT_COL0);
 791       }
 792       newProg->InputsRead = progA->InputsRead | inputsB;
 793       newProg->OutputsWritten = progB->OutputsWritten;
 794       newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
 795    }
 796    else {
 797       /* vertex program */
 798       assert(0);      /* XXX todo */
 799    }
 800
 801    /*
 802     * Merge parameters (uniforms, constants, etc)
 803     */
 804    newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
 805                                                        progB->Parameters);
 806
 807    adjust_param_indexes(newInst + lenA, lenB, numParamsA);
 808
 809
 810    return newProg;
 811 }
 812
 813
 814 /**
 815  * Populate the 'used' array with flags indicating which registers (TEMPs,
 816  * INPUTs, OUTPUTs, etc, are used by the given program.
 817  * \param file  type of register to scan for
 818  * \param used  returns true/false flags for in use / free
 819  * \param usedSize  size of the 'used' array
 820  */
 821 void
 822 _mesa_find_used_registers(const struct gl_program *prog,
 823                           gl_register_file file,
 824                           GLboolean used[], GLuint usedSize)
 825 {
 826    GLuint i, j;
 827
 828    memset(used, 0, usedSize);
 829
 830    for (i = 0; i < prog->NumInstructions; i++) {
 831       const struct prog_instruction *inst = prog->Instructions + i;
 832       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 833
 834       if (inst->DstReg.File == file) {
 835          ASSERT(inst->DstReg.Index < usedSize);
 836          if(inst->DstReg.Index < usedSize)
 837             used[inst->DstReg.Index] = GL_TRUE;
 838       }
 839
 840       for (j = 0; j < n; j++) {
 841          if (inst->SrcReg[j].File == file) {
 842             ASSERT(inst->SrcReg[j].Index < (GLint) usedSize);
 843             if (inst->SrcReg[j].Index < (GLint) usedSize)
 844                used[inst->SrcReg[j].Index] = GL_TRUE;
 845          }
 846       }
 847    }
 848 }
 849
 850
 851 /**
 852  * Scan the given 'used' register flag array for the first entry
 853  * that's >= firstReg.
 854  * \param used  vector of flags indicating registers in use (as returned
 855  *              by _mesa_find_used_registers())
 856  * \param usedSize  size of the 'used' array
 857  * \param firstReg  first register to start searching at
 858  * \return index of unused register, or -1 if none.
 859  */
 860 GLint
 861 _mesa_find_free_register(const GLboolean used[],
 862                          GLuint usedSize, GLuint firstReg)
 863 {
 864    GLuint i;
 865
 866    assert(firstReg < usedSize);
 867
 868    for (i = firstReg; i < usedSize; i++)
 869       if (!used[i])
 870          return i;
 871
 872    return -1;
 873 }
 874
 875
 876
 877 /**
 878  * Check if the given register index is valid (doesn't exceed implementation-
 879  * dependent limits).
 880  * \return GL_TRUE if OK, GL_FALSE if bad index
 881  */
 882 GLboolean
 883 _mesa_valid_register_index(const struct gl_context *ctx,
 884                            gl_shader_type shaderType,
 885                            gl_register_file file, GLint index)
 886 {
 887    const struct gl_program_constants *c;
 888
 889    switch (shaderType) {
 890    case MESA_SHADER_VERTEX:
 891       c = &ctx->Const.VertexProgram;
 892       break;
 893    case MESA_SHADER_FRAGMENT:
 894       c = &ctx->Const.FragmentProgram;
 895       break;
 896    case MESA_SHADER_GEOMETRY:
 897       c = &ctx->Const.GeometryProgram;
 898       break;
 899    default:
 900       _mesa_problem(ctx,
 901                     "unexpected shader type in _mesa_valid_register_index()");
 902       return GL_FALSE;
 903    }
 904
 905    switch (file) {
 906    case PROGRAM_UNDEFINED:
 907       return GL_TRUE;  /* XXX or maybe false? */
 908
 909    case PROGRAM_TEMPORARY:
 910       return index >= 0 && index < (GLint) c->MaxTemps;
 911
 912    case PROGRAM_UNIFORM:
 913    case PROGRAM_STATE_VAR:
 914       /* aka constant buffer */
 915       return index >= 0 && index < (GLint) c->MaxUniformComponents / 4;
 916
 917    case PROGRAM_CONSTANT:
 918       /* constant buffer w/ possible relative negative addressing */
 919       return (index > (int) c->MaxUniformComponents / -4 &&
 920               index < (int) c->MaxUniformComponents / 4);
 921
 922    case PROGRAM_INPUT:
 923       if (index < 0)
 924          return GL_FALSE;
 925
 926       switch (shaderType) {
 927       case MESA_SHADER_VERTEX:
 928          return index < VERT_ATTRIB_GENERIC0 + (GLint) c->MaxAttribs;
 929       case MESA_SHADER_FRAGMENT:
 930          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 931       case MESA_SHADER_GEOMETRY:
 932          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 933       default:
 934          return GL_FALSE;
 935       }
 936
 937    case PROGRAM_OUTPUT:
 938       if (index < 0)
 939          return GL_FALSE;
 940
 941       switch (shaderType) {
 942       case MESA_SHADER_VERTEX:
 943          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 944       case MESA_SHADER_FRAGMENT:
 945          return index < FRAG_RESULT_DATA0 + (GLint) ctx->Const.MaxDrawBuffers;
 946       case MESA_SHADER_GEOMETRY:
 947          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 948       default:
 949          return GL_FALSE;
 950       }
 951
 952    case PROGRAM_ADDRESS:
 953       return index >= 0 && index < (GLint) c->MaxAddressRegs;
 954
 955    default:
 956       _mesa_problem(ctx,
 957                     "unexpected register file in _mesa_valid_register_index()");
 958       return GL_FALSE;
 959    }
 960 }
 961
 962
 963
 964 /**
 965  * "Post-process" a GPU program.  This is intended to be used for debugging.
 966  * Example actions include no-op'ing instructions or changing instruction
 967  * behaviour.
 968  */
 969 void
 970 _mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog)
 971 {
 972    static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 };
 973    GLuint i;
 974    GLuint whiteSwizzle;
 975    GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters,
 976                                                  (gl_constant_value *) white,
 977                                                  4, &whiteSwizzle);
 978
 979    (void) whiteIndex;
 980
 981    for (i = 0; i < prog->NumInstructions; i++) {
 982       struct prog_instruction *inst = prog->Instructions + i;
 983       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 984
 985       (void) n;
 986
 987       if (_mesa_is_tex_instruction(inst->Opcode)) {
 988 #if 0
 989          /* replace TEX/TXP/TXB with MOV */
 990          inst->Opcode = OPCODE_MOV;
 991          inst->DstReg.WriteMask = WRITEMASK_XYZW;
 992          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
 993          inst->SrcReg[0].Negate = NEGATE_NONE;
 994 #endif
 995
 996 #if 0
 997          /* disable shadow texture mode */
 998          inst->TexShadow = 0;
 999 #endif
1000       }
1001
1002       if (inst->Opcode == OPCODE_TXP) {
1003 #if 0
1004          inst->Opcode = OPCODE_MOV;
1005          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1006          inst->SrcReg[0].File = PROGRAM_CONSTANT;
1007          inst->SrcReg[0].Index = whiteIndex;
1008          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1009          inst->SrcReg[0].Negate = NEGATE_NONE;
1010 #endif
1011 #if 0
1012          inst->TexShadow = 0;
1013 #endif
1014 #if 0
1015          inst->Opcode = OPCODE_TEX;
1016          inst->TexShadow = 0;
1017 #endif
1018       }
1019
1020    }
1021 }
1022
1023 /* Gets the minimum number of shader invocations per fragment.
1024  * This function is useful to determine if we need to do per
1025  * sample shading or per fragment shading.
1026  */
1027 GLint
1028 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
1029                                        const struct gl_fragment_program *prog)
1030 {
1031    /* From ARB_sample_shading specification:
1032     * "Using gl_SampleID in a fragment shader causes the entire shader
1033     *  to be evaluated per-sample."
1034     *
1035     * "Using gl_SamplePosition in a fragment shader causes the entire
1036     *  shader to be evaluated per-sample."
1037     *
1038     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
1039     *  has no effect."
1040     */
1041    if (ctx->Multisample.Enabled) {
1042       if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
1043                                          SYSTEM_BIT_SAMPLE_POS))
1044          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1045       else if (ctx->Multisample.SampleShading)
1046          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
1047                           ctx->DrawBuffer->Visual.samples), 1);
1048       else
1049          return 1;
1050    }
1051    return 1;
1052 }