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.Program[MESA_SHADER_VERTEX].MaxUniformComponents / 4
  60           <= (1 << INST_INDEX_BITS));
  61    ASSERT(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents / 4
  62           <= (1 << INST_INDEX_BITS));
  63
  64    ASSERT(ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps <= (1 << INST_INDEX_BITS));
  65    ASSERT(ctx->Const.Program[MESA_SHADER_VERTEX].MaxLocalParams <= (1 << INST_INDEX_BITS));
  66    ASSERT(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTemps <= (1 << INST_INDEX_BITS));
  67    ASSERT(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxLocalParams <= (1 << INST_INDEX_BITS));
  68
  69    ASSERT(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  70    ASSERT(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  71
  72    ASSERT(ctx->Const.Program[MESA_SHADER_VERTEX].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  73    ASSERT(ctx->Const.Program[MESA_SHADER_FRAGMENT].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 compute program object.
 283  */
 284 struct gl_program *
 285 _mesa_init_compute_program(struct gl_context *ctx,
 286                            struct gl_compute_program *prog, GLenum target,
 287                            GLuint id)
 288 {
 289    if (prog)
 290       return _mesa_init_program_struct( ctx, &prog->Base, target, id );
 291    else
 292       return NULL;
 293 }
 294
 295
 296 /**
 297  * Initialize a new geometry program object.
 298  */
 299 struct gl_program *
 300 _mesa_init_geometry_program( struct gl_context *ctx, struct gl_geometry_program *prog,
 301                              GLenum target, GLuint id)
 302 {
 303    if (prog)
 304       return _mesa_init_program_struct( ctx, &prog->Base, target, id );
 305    else
 306       return NULL;
 307 }
 308
 309
 310 /**
 311  * Allocate and initialize a new fragment/vertex program object but
 312  * don't put it into the program hash table.  Called via
 313  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 314  * device driver function to implement OO deriviation with additional
 315  * types not understood by this function.
 316  *
 317  * \param ctx  context
 318  * \param id   program id/number
 319  * \param target  program target/type
 320  * \return  pointer to new program object
 321  */
 322 struct gl_program *
 323 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id)
 324 {
 325    struct gl_program *prog;
 326    switch (target) {
 327    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 328       prog = _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 329                                        target, id );
 330       break;
 331    case GL_FRAGMENT_PROGRAM_NV:
 332    case GL_FRAGMENT_PROGRAM_ARB:
 333       prog =_mesa_init_fragment_program(ctx,
 334                                          CALLOC_STRUCT(gl_fragment_program),
 335                                          target, id );
 336       break;
 337    case MESA_GEOMETRY_PROGRAM:
 338       prog = _mesa_init_geometry_program(ctx,
 339                                          CALLOC_STRUCT(gl_geometry_program),
 340                                          target, id);
 341       break;
 342    case GL_COMPUTE_PROGRAM_NV:
 343       prog = _mesa_init_compute_program(ctx,
 344                                         CALLOC_STRUCT(gl_compute_program),
 345                                         target, id);
 346       break;
 347    default:
 348       _mesa_problem(ctx, "bad target in _mesa_new_program");
 349       prog = NULL;
 350    }
 351    return prog;
 352 }
 353
 354
 355 /**
 356  * Delete a program and remove it from the hash table, ignoring the
 357  * reference count.
 358  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 359  * by a device driver function.
 360  */
 361 void
 362 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 363 {
 364    (void) ctx;
 365    ASSERT(prog);
 366    ASSERT(prog->RefCount==0);
 367
 368    if (prog == &_mesa_DummyProgram)
 369       return;
 370
 371    free(prog->String);
 372    free(prog->LocalParams);
 373
 374    if (prog->Instructions) {
 375       _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
 376    }
 377    if (prog->Parameters) {
 378       _mesa_free_parameter_list(prog->Parameters);
 379    }
 380
 381    free(prog);
 382 }
 383
 384
 385 /**
 386  * Return the gl_program object for a given ID.
 387  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 388  * casts elsewhere.
 389  */
 390 struct gl_program *
 391 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 392 {
 393    if (id)
 394       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 395    else
 396       return NULL;
 397 }
 398
 399
 400 /**
 401  * Reference counting for vertex/fragment programs
 402  * This is normally only called from the _mesa_reference_program() macro
 403  * when there's a real pointer change.
 404  */
 405 void
 406 _mesa_reference_program_(struct gl_context *ctx,
 407                          struct gl_program **ptr,
 408                          struct gl_program *prog)
 409 {
 410 #ifndef NDEBUG
 411    assert(ptr);
 412    if (*ptr && prog) {
 413       /* sanity check */
 414       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 415          ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB);
 416       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 417          ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 418                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 419       else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM)
 420          ASSERT(prog->Target == MESA_GEOMETRY_PROGRAM);
 421    }
 422 #endif
 423
 424    if (*ptr) {
 425       GLboolean deleteFlag;
 426
 427       /*_glthread_LOCK_MUTEX((*ptr)->Mutex);*/
 428 #if 0
 429       printf("Program %p ID=%u Target=%s  Refcount-- to %d\n",
 430              *ptr, (*ptr)->Id,
 431              ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 432               ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 433              (*ptr)->RefCount - 1);
 434 #endif
 435       ASSERT((*ptr)->RefCount > 0);
 436       (*ptr)->RefCount--;
 437
 438       deleteFlag = ((*ptr)->RefCount == 0);
 439       /*_glthread_UNLOCK_MUTEX((*ptr)->Mutex);*/
 440
 441       if (deleteFlag) {
 442          ASSERT(ctx);
 443          ctx->Driver.DeleteProgram(ctx, *ptr);
 444       }
 445
 446       *ptr = NULL;
 447    }
 448
 449    assert(!*ptr);
 450    if (prog) {
 451       /*_glthread_LOCK_MUTEX(prog->Mutex);*/
 452       prog->RefCount++;
 453 #if 0
 454       printf("Program %p ID=%u Target=%s  Refcount++ to %d\n",
 455              prog, prog->Id,
 456              (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 457               (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 458              prog->RefCount);
 459 #endif
 460       /*_glthread_UNLOCK_MUTEX(prog->Mutex);*/
 461    }
 462
 463    *ptr = prog;
 464 }
 465
 466
 467 /**
 468  * Return a copy of a program.
 469  * XXX Problem here if the program object is actually OO-derivation
 470  * made by a device driver.
 471  */
 472 struct gl_program *
 473 _mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog)
 474 {
 475    struct gl_program *clone;
 476
 477    clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id);
 478    if (!clone)
 479       return NULL;
 480
 481    assert(clone->Target == prog->Target);
 482    assert(clone->RefCount == 1);
 483
 484    clone->String = (GLubyte *) _mesa_strdup((char *) prog->String);
 485    clone->Format = prog->Format;
 486    clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions);
 487    if (!clone->Instructions) {
 488       _mesa_reference_program(ctx, &clone, NULL);
 489       return NULL;
 490    }
 491    _mesa_copy_instructions(clone->Instructions, prog->Instructions,
 492                            prog->NumInstructions);
 493    clone->InputsRead = prog->InputsRead;
 494    clone->OutputsWritten = prog->OutputsWritten;
 495    clone->SamplersUsed = prog->SamplersUsed;
 496    clone->ShadowSamplers = prog->ShadowSamplers;
 497    memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed));
 498
 499    if (prog->Parameters)
 500       clone->Parameters = _mesa_clone_parameter_list(prog->Parameters);
 501    if (prog->LocalParams) {
 502       clone->LocalParams = malloc(MAX_PROGRAM_LOCAL_PARAMS *
 503                                   sizeof(float[4]));
 504       if (!clone->LocalParams) {
 505          _mesa_reference_program(ctx, &clone, NULL);
 506          return NULL;
 507       }
 508       memcpy(clone->LocalParams, prog->LocalParams,
 509              MAX_PROGRAM_LOCAL_PARAMS * sizeof(float[4]));
 510    }
 511    clone->IndirectRegisterFiles = prog->IndirectRegisterFiles;
 512    clone->NumInstructions = prog->NumInstructions;
 513    clone->NumTemporaries = prog->NumTemporaries;
 514    clone->NumParameters = prog->NumParameters;
 515    clone->NumAttributes = prog->NumAttributes;
 516    clone->NumAddressRegs = prog->NumAddressRegs;
 517    clone->NumNativeInstructions = prog->NumNativeInstructions;
 518    clone->NumNativeTemporaries = prog->NumNativeTemporaries;
 519    clone->NumNativeParameters = prog->NumNativeParameters;
 520    clone->NumNativeAttributes = prog->NumNativeAttributes;
 521    clone->NumNativeAddressRegs = prog->NumNativeAddressRegs;
 522    clone->NumAluInstructions = prog->NumAluInstructions;
 523    clone->NumTexInstructions = prog->NumTexInstructions;
 524    clone->NumTexIndirections = prog->NumTexIndirections;
 525    clone->NumNativeAluInstructions = prog->NumNativeAluInstructions;
 526    clone->NumNativeTexInstructions = prog->NumNativeTexInstructions;
 527    clone->NumNativeTexIndirections = prog->NumNativeTexIndirections;
 528
 529    switch (prog->Target) {
 530    case GL_VERTEX_PROGRAM_ARB:
 531       {
 532          const struct gl_vertex_program *vp = gl_vertex_program_const(prog);
 533          struct gl_vertex_program *vpc = gl_vertex_program(clone);
 534          vpc->IsPositionInvariant = vp->IsPositionInvariant;
 535       }
 536       break;
 537    case GL_FRAGMENT_PROGRAM_ARB:
 538       {
 539          const struct gl_fragment_program *fp = gl_fragment_program_const(prog);
 540          struct gl_fragment_program *fpc = gl_fragment_program(clone);
 541          fpc->UsesKill = fp->UsesKill;
 542          fpc->UsesDFdy = fp->UsesDFdy;
 543          fpc->OriginUpperLeft = fp->OriginUpperLeft;
 544          fpc->PixelCenterInteger = fp->PixelCenterInteger;
 545       }
 546       break;
 547    case MESA_GEOMETRY_PROGRAM:
 548       {
 549          const struct gl_geometry_program *gp = gl_geometry_program_const(prog);
 550          struct gl_geometry_program *gpc = gl_geometry_program(clone);
 551          gpc->VerticesOut = gp->VerticesOut;
 552          gpc->InputType = gp->InputType;
 553          gpc->OutputType = gp->OutputType;
 554       }
 555       break;
 556    default:
 557       _mesa_problem(NULL, "Unexpected target in _mesa_clone_program");
 558    }
 559
 560    return clone;
 561 }
 562
 563
 564 /**
 565  * Insert 'count' NOP instructions at 'start' in the given program.
 566  * Adjust branch targets accordingly.
 567  */
 568 GLboolean
 569 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 570 {
 571    const GLuint origLen = prog->NumInstructions;
 572    const GLuint newLen = origLen + count;
 573    struct prog_instruction *newInst;
 574    GLuint i;
 575
 576    /* adjust branches */
 577    for (i = 0; i < prog->NumInstructions; i++) {
 578       struct prog_instruction *inst = prog->Instructions + i;
 579       if (inst->BranchTarget > 0) {
 580          if ((GLuint)inst->BranchTarget >= start) {
 581             inst->BranchTarget += count;
 582          }
 583       }
 584    }
 585
 586    /* Alloc storage for new instructions */
 587    newInst = _mesa_alloc_instructions(newLen);
 588    if (!newInst) {
 589       return GL_FALSE;
 590    }
 591
 592    /* Copy 'start' instructions into new instruction buffer */
 593    _mesa_copy_instructions(newInst, prog->Instructions, start);
 594
 595    /* init the new instructions */
 596    _mesa_init_instructions(newInst + start, count);
 597
 598    /* Copy the remaining/tail instructions to new inst buffer */
 599    _mesa_copy_instructions(newInst + start + count,
 600                            prog->Instructions + start,
 601                            origLen - start);
 602
 603    /* free old instructions */
 604    _mesa_free_instructions(prog->Instructions, origLen);
 605
 606    /* install new instructions */
 607    prog->Instructions = newInst;
 608    prog->NumInstructions = newLen;
 609
 610    return GL_TRUE;
 611 }
 612
 613 /**
 614  * Delete 'count' instructions at 'start' in the given program.
 615  * Adjust branch targets accordingly.
 616  */
 617 GLboolean
 618 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
 619 {
 620    const GLuint origLen = prog->NumInstructions;
 621    const GLuint newLen = origLen - count;
 622    struct prog_instruction *newInst;
 623    GLuint i;
 624
 625    /* adjust branches */
 626    for (i = 0; i < prog->NumInstructions; i++) {
 627       struct prog_instruction *inst = prog->Instructions + i;
 628       if (inst->BranchTarget > 0) {
 629          if (inst->BranchTarget > (GLint) start) {
 630             inst->BranchTarget -= count;
 631          }
 632       }
 633    }
 634
 635    /* Alloc storage for new instructions */
 636    newInst = _mesa_alloc_instructions(newLen);
 637    if (!newInst) {
 638       return GL_FALSE;
 639    }
 640
 641    /* Copy 'start' instructions into new instruction buffer */
 642    _mesa_copy_instructions(newInst, prog->Instructions, start);
 643
 644    /* Copy the remaining/tail instructions to new inst buffer */
 645    _mesa_copy_instructions(newInst + start,
 646                            prog->Instructions + start + count,
 647                            newLen - start);
 648
 649    /* free old instructions */
 650    _mesa_free_instructions(prog->Instructions, origLen);
 651
 652    /* install new instructions */
 653    prog->Instructions = newInst;
 654    prog->NumInstructions = newLen;
 655
 656    return GL_TRUE;
 657 }
 658
 659
 660 /**
 661  * Search instructions for registers that match (oldFile, oldIndex),
 662  * replacing them with (newFile, newIndex).
 663  */
 664 static void
 665 replace_registers(struct prog_instruction *inst, GLuint numInst,
 666                   GLuint oldFile, GLuint oldIndex,
 667                   GLuint newFile, GLuint newIndex)
 668 {
 669    GLuint i, j;
 670    for (i = 0; i < numInst; i++) {
 671       /* src regs */
 672       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 673          if (inst[i].SrcReg[j].File == oldFile &&
 674              inst[i].SrcReg[j].Index == oldIndex) {
 675             inst[i].SrcReg[j].File = newFile;
 676             inst[i].SrcReg[j].Index = newIndex;
 677          }
 678       }
 679       /* dst reg */
 680       if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) {
 681          inst[i].DstReg.File = newFile;
 682          inst[i].DstReg.Index = newIndex;
 683       }
 684    }
 685 }
 686
 687
 688 /**
 689  * Search instructions for references to program parameters.  When found,
 690  * increment the parameter index by 'offset'.
 691  * Used when combining programs.
 692  */
 693 static void
 694 adjust_param_indexes(struct prog_instruction *inst, GLuint numInst,
 695                      GLuint offset)
 696 {
 697    GLuint i, j;
 698    for (i = 0; i < numInst; i++) {
 699       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 700          GLuint f = inst[i].SrcReg[j].File;
 701          if (f == PROGRAM_CONSTANT ||
 702              f == PROGRAM_UNIFORM ||
 703              f == PROGRAM_STATE_VAR) {
 704             inst[i].SrcReg[j].Index += offset;
 705          }
 706       }
 707    }
 708 }
 709
 710
 711 /**
 712  * Combine two programs into one.  Fix instructions so the outputs of
 713  * the first program go to the inputs of the second program.
 714  */
 715 struct gl_program *
 716 _mesa_combine_programs(struct gl_context *ctx,
 717                        const struct gl_program *progA,
 718                        const struct gl_program *progB)
 719 {
 720    struct prog_instruction *newInst;
 721    struct gl_program *newProg;
 722    const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
 723    const GLuint lenB = progB->NumInstructions;
 724    const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
 725    const GLuint newLength = lenA + lenB;
 726    GLboolean usedTemps[MAX_PROGRAM_TEMPS];
 727    GLuint firstTemp = 0;
 728    GLbitfield64 inputsB;
 729    GLuint i;
 730
 731    ASSERT(progA->Target == progB->Target);
 732
 733    newInst = _mesa_alloc_instructions(newLength);
 734    if (!newInst)
 735       return GL_FALSE;
 736
 737    _mesa_copy_instructions(newInst, progA->Instructions, lenA);
 738    _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
 739
 740    /* adjust branch / instruction addresses for B's instructions */
 741    for (i = 0; i < lenB; i++) {
 742       newInst[lenA + i].BranchTarget += lenA;
 743    }
 744
 745    newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
 746    newProg->Instructions = newInst;
 747    newProg->NumInstructions = newLength;
 748
 749    /* find used temp regs (we may need new temps below) */
 750    _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY,
 751                              usedTemps, MAX_PROGRAM_TEMPS);
 752
 753    if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
 754       const struct gl_fragment_program *fprogA, *fprogB;
 755       struct gl_fragment_program *newFprog;
 756       GLbitfield64 progB_inputsRead = progB->InputsRead;
 757       GLint progB_colorFile, progB_colorIndex;
 758
 759       fprogA = gl_fragment_program_const(progA);
 760       fprogB = gl_fragment_program_const(progB);
 761       newFprog = gl_fragment_program(newProg);
 762
 763       newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
 764       newFprog->UsesDFdy = fprogA->UsesDFdy || fprogB->UsesDFdy;
 765
 766       /* We'll do a search and replace for instances
 767        * of progB_colorFile/progB_colorIndex below...
 768        */
 769       progB_colorFile = PROGRAM_INPUT;
 770       progB_colorIndex = VARYING_SLOT_COL0;
 771
 772       /*
 773        * The fragment program may get color from a state var rather than
 774        * a fragment input (vertex output) if it's constant.
 775        * See the texenvprogram.c code.
 776        * So, search the program's parameter list now to see if the program
 777        * gets color from a state var instead of a conventional fragment
 778        * input register.
 779        */
 780       for (i = 0; i < progB->Parameters->NumParameters; i++) {
 781          struct gl_program_parameter *p = &progB->Parameters->Parameters[i];
 782          if (p->Type == PROGRAM_STATE_VAR &&
 783              p->StateIndexes[0] == STATE_INTERNAL &&
 784              p->StateIndexes[1] == STATE_CURRENT_ATTRIB &&
 785              (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) {
 786             progB_inputsRead |= VARYING_BIT_COL0;
 787             progB_colorFile = PROGRAM_STATE_VAR;
 788             progB_colorIndex = i;
 789             break;
 790          }
 791       }
 792
 793       /* Connect color outputs of fprogA to color inputs of fprogB, via a
 794        * new temporary register.
 795        */
 796       if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) &&
 797           (progB_inputsRead & VARYING_BIT_COL0)) {
 798          GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS,
 799                                                   firstTemp);
 800          if (tempReg < 0) {
 801             _mesa_problem(ctx, "No free temp regs found in "
 802                           "_mesa_combine_programs(), using 31");
 803             tempReg = 31;
 804          }
 805          firstTemp = tempReg + 1;
 806
 807          /* replace writes to result.color[0] with tempReg */
 808          replace_registers(newInst, lenA,
 809                            PROGRAM_OUTPUT, FRAG_RESULT_COLOR,
 810                            PROGRAM_TEMPORARY, tempReg);
 811          /* replace reads from the input color with tempReg */
 812          replace_registers(newInst + lenA, lenB,
 813                            progB_colorFile, progB_colorIndex, /* search for */
 814                            PROGRAM_TEMPORARY, tempReg  /* replace with */ );
 815       }
 816
 817       /* compute combined program's InputsRead */
 818       inputsB = progB_inputsRead;
 819       if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) {
 820          inputsB &= ~(1 << VARYING_SLOT_COL0);
 821       }
 822       newProg->InputsRead = progA->InputsRead | inputsB;
 823       newProg->OutputsWritten = progB->OutputsWritten;
 824       newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
 825    }
 826    else {
 827       /* vertex program */
 828       assert(0);      /* XXX todo */
 829    }
 830
 831    /*
 832     * Merge parameters (uniforms, constants, etc)
 833     */
 834    newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
 835                                                        progB->Parameters);
 836
 837    adjust_param_indexes(newInst + lenA, lenB, numParamsA);
 838
 839
 840    return newProg;
 841 }
 842
 843
 844 /**
 845  * Populate the 'used' array with flags indicating which registers (TEMPs,
 846  * INPUTs, OUTPUTs, etc, are used by the given program.
 847  * \param file  type of register to scan for
 848  * \param used  returns true/false flags for in use / free
 849  * \param usedSize  size of the 'used' array
 850  */
 851 void
 852 _mesa_find_used_registers(const struct gl_program *prog,
 853                           gl_register_file file,
 854                           GLboolean used[], GLuint usedSize)
 855 {
 856    GLuint i, j;
 857
 858    memset(used, 0, usedSize);
 859
 860    for (i = 0; i < prog->NumInstructions; i++) {
 861       const struct prog_instruction *inst = prog->Instructions + i;
 862       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 863
 864       if (inst->DstReg.File == file) {
 865          ASSERT(inst->DstReg.Index < usedSize);
 866          if(inst->DstReg.Index < usedSize)
 867             used[inst->DstReg.Index] = GL_TRUE;
 868       }
 869
 870       for (j = 0; j < n; j++) {
 871          if (inst->SrcReg[j].File == file) {
 872             ASSERT(inst->SrcReg[j].Index < (GLint) usedSize);
 873             if (inst->SrcReg[j].Index < (GLint) usedSize)
 874                used[inst->SrcReg[j].Index] = GL_TRUE;
 875          }
 876       }
 877    }
 878 }
 879
 880
 881 /**
 882  * Scan the given 'used' register flag array for the first entry
 883  * that's >= firstReg.
 884  * \param used  vector of flags indicating registers in use (as returned
 885  *              by _mesa_find_used_registers())
 886  * \param usedSize  size of the 'used' array
 887  * \param firstReg  first register to start searching at
 888  * \return index of unused register, or -1 if none.
 889  */
 890 GLint
 891 _mesa_find_free_register(const GLboolean used[],
 892                          GLuint usedSize, GLuint firstReg)
 893 {
 894    GLuint i;
 895
 896    assert(firstReg < usedSize);
 897
 898    for (i = firstReg; i < usedSize; i++)
 899       if (!used[i])
 900          return i;
 901
 902    return -1;
 903 }
 904
 905
 906
 907 /**
 908  * Check if the given register index is valid (doesn't exceed implementation-
 909  * dependent limits).
 910  * \return GL_TRUE if OK, GL_FALSE if bad index
 911  */
 912 GLboolean
 913 _mesa_valid_register_index(const struct gl_context *ctx,
 914                            gl_shader_stage shaderType,
 915                            gl_register_file file, GLint index)
 916 {
 917    const struct gl_program_constants *c;
 918
 919    assert(0 <= shaderType && shaderType < MESA_SHADER_STAGES);
 920    c = &ctx->Const.Program[shaderType];
 921
 922    switch (file) {
 923    case PROGRAM_UNDEFINED:
 924       return GL_TRUE;  /* XXX or maybe false? */
 925
 926    case PROGRAM_TEMPORARY:
 927       return index >= 0 && index < (GLint) c->MaxTemps;
 928
 929    case PROGRAM_UNIFORM:
 930    case PROGRAM_STATE_VAR:
 931       /* aka constant buffer */
 932       return index >= 0 && index < (GLint) c->MaxUniformComponents / 4;
 933
 934    case PROGRAM_CONSTANT:
 935       /* constant buffer w/ possible relative negative addressing */
 936       return (index > (int) c->MaxUniformComponents / -4 &&
 937               index < (int) c->MaxUniformComponents / 4);
 938
 939    case PROGRAM_INPUT:
 940       if (index < 0)
 941          return GL_FALSE;
 942
 943       switch (shaderType) {
 944       case MESA_SHADER_VERTEX:
 945          return index < VERT_ATTRIB_GENERIC0 + (GLint) c->MaxAttribs;
 946       case MESA_SHADER_FRAGMENT:
 947          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 948       case MESA_SHADER_GEOMETRY:
 949          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 950       default:
 951          return GL_FALSE;
 952       }
 953
 954    case PROGRAM_OUTPUT:
 955       if (index < 0)
 956          return GL_FALSE;
 957
 958       switch (shaderType) {
 959       case MESA_SHADER_VERTEX:
 960          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 961       case MESA_SHADER_FRAGMENT:
 962          return index < FRAG_RESULT_DATA0 + (GLint) ctx->Const.MaxDrawBuffers;
 963       case MESA_SHADER_GEOMETRY:
 964          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 965       default:
 966          return GL_FALSE;
 967       }
 968
 969    case PROGRAM_ADDRESS:
 970       return index >= 0 && index < (GLint) c->MaxAddressRegs;
 971
 972    default:
 973       _mesa_problem(ctx,
 974                     "unexpected register file in _mesa_valid_register_index()");
 975       return GL_FALSE;
 976    }
 977 }
 978
 979
 980
 981 /**
 982  * "Post-process" a GPU program.  This is intended to be used for debugging.
 983  * Example actions include no-op'ing instructions or changing instruction
 984  * behaviour.
 985  */
 986 void
 987 _mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog)
 988 {
 989    static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 };
 990    GLuint i;
 991    GLuint whiteSwizzle;
 992    GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters,
 993                                                  (gl_constant_value *) white,
 994                                                  4, &whiteSwizzle);
 995
 996    (void) whiteIndex;
 997
 998    for (i = 0; i < prog->NumInstructions; i++) {
 999       struct prog_instruction *inst = prog->Instructions + i;
1000       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
1001
1002       (void) n;
1003
1004       if (_mesa_is_tex_instruction(inst->Opcode)) {
1005 #if 0
1006          /* replace TEX/TXP/TXB with MOV */
1007          inst->Opcode = OPCODE_MOV;
1008          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1009          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1010          inst->SrcReg[0].Negate = NEGATE_NONE;
1011 #endif
1012
1013 #if 0
1014          /* disable shadow texture mode */
1015          inst->TexShadow = 0;
1016 #endif
1017       }
1018
1019       if (inst->Opcode == OPCODE_TXP) {
1020 #if 0
1021          inst->Opcode = OPCODE_MOV;
1022          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1023          inst->SrcReg[0].File = PROGRAM_CONSTANT;
1024          inst->SrcReg[0].Index = whiteIndex;
1025          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1026          inst->SrcReg[0].Negate = NEGATE_NONE;
1027 #endif
1028 #if 0
1029          inst->TexShadow = 0;
1030 #endif
1031 #if 0
1032          inst->Opcode = OPCODE_TEX;
1033          inst->TexShadow = 0;
1034 #endif
1035       }
1036
1037    }
1038 }
1039
1040 /* Gets the minimum number of shader invocations per fragment.
1041  * This function is useful to determine if we need to do per
1042  * sample shading or per fragment shading.
1043  */
1044 GLint
1045 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
1046                                        const struct gl_fragment_program *prog,
1047                                        bool ignore_sample_qualifier)
1048 {
1049    /* From ARB_sample_shading specification:
1050     * "Using gl_SampleID in a fragment shader causes the entire shader
1051     *  to be evaluated per-sample."
1052     *
1053     * "Using gl_SamplePosition in a fragment shader causes the entire
1054     *  shader to be evaluated per-sample."
1055     *
1056     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
1057     *  has no effect."
1058     */
1059    if (ctx->Multisample.Enabled) {
1060       /* The ARB_gpu_shader5 specification says:
1061        *
1062        * "Use of the "sample" qualifier on a fragment shader input
1063        *  forces per-sample shading"
1064        */
1065       if (prog->IsSample && !ignore_sample_qualifier)
1066          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1067
1068       if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
1069                                          SYSTEM_BIT_SAMPLE_POS))
1070          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1071       else if (ctx->Multisample.SampleShading)
1072          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
1073                           ctx->DrawBuffer->Visual.samples), 1);
1074       else
1075          return 1;
1076    }
1077    return 1;
1078 }