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 = 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
 104    /* XXX probably move this stuff */
 105    ctx->ATIFragmentShader.Enabled = GL_FALSE;
 106    ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
 107    assert(ctx->ATIFragmentShader.Current);
 108    ctx->ATIFragmentShader.Current->RefCount++;
 109 }
 110
 111
 112 /**
 113  * Free a context's vertex/fragment program state
 114  */
 115 void
 116 _mesa_free_program_data(struct gl_context *ctx)
 117 {
 118    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL);
 119    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 120    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL);
 121    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 122    _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current, NULL);
 123
 124    /* XXX probably move this stuff */
 125    if (ctx->ATIFragmentShader.Current) {
 126       ctx->ATIFragmentShader.Current->RefCount--;
 127       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 128          free(ctx->ATIFragmentShader.Current);
 129       }
 130    }
 131
 132    free((void *) ctx->Program.ErrorString);
 133 }
 134
 135
 136 /**
 137  * Update the default program objects in the given context to reference those
 138  * specified in the shared state and release those referencing the old
 139  * shared state.
 140  */
 141 void
 142 _mesa_update_default_objects_program(struct gl_context *ctx)
 143 {
 144    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
 145                             ctx->Shared->DefaultVertexProgram);
 146    assert(ctx->VertexProgram.Current);
 147
 148    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
 149                             ctx->Shared->DefaultFragmentProgram);
 150    assert(ctx->FragmentProgram.Current);
 151
 152    _mesa_reference_geomprog(ctx, &ctx->GeometryProgram.Current,
 153                       ctx->Shared->DefaultGeometryProgram);
 154
 155    /* XXX probably move this stuff */
 156    if (ctx->ATIFragmentShader.Current) {
 157       ctx->ATIFragmentShader.Current->RefCount--;
 158       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 159          free(ctx->ATIFragmentShader.Current);
 160       }
 161    }
 162    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 163    assert(ctx->ATIFragmentShader.Current);
 164    ctx->ATIFragmentShader.Current->RefCount++;
 165 }
 166
 167
 168 /**
 169  * Set the vertex/fragment program error state (position and error string).
 170  * This is generally called from within the parsers.
 171  */
 172 void
 173 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 174 {
 175    ctx->Program.ErrorPos = pos;
 176    free((void *) ctx->Program.ErrorString);
 177    if (!string)
 178       string = "";
 179    ctx->Program.ErrorString = strdup(string);
 180 }
 181
 182
 183 /**
 184  * Find the line number and column for 'pos' within 'string'.
 185  * Return a copy of the line which contains 'pos'.  Free the line with
 186  * free().
 187  * \param string  the program string
 188  * \param pos     the position within the string
 189  * \param line    returns the line number corresponding to 'pos'.
 190  * \param col     returns the column number corresponding to 'pos'.
 191  * \return copy of the line containing 'pos'.
 192  */
 193 const GLubyte *
 194 _mesa_find_line_column(const GLubyte *string, const GLubyte *pos,
 195                        GLint *line, GLint *col)
 196 {
 197    const GLubyte *lineStart = string;
 198    const GLubyte *p = string;
 199    GLubyte *s;
 200    int len;
 201
 202    *line = 1;
 203
 204    while (p != pos) {
 205       if (*p == (GLubyte) '\n') {
 206          (*line)++;
 207          lineStart = p + 1;
 208       }
 209       p++;
 210    }
 211
 212    *col = (pos - lineStart) + 1;
 213
 214    /* return copy of this line */
 215    while (*p != 0 && *p != '\n')
 216       p++;
 217    len = p - lineStart;
 218    s = malloc(len + 1);
 219    memcpy(s, lineStart, len);
 220    s[len] = 0;
 221
 222    return s;
 223 }
 224
 225
 226 /**
 227  * Initialize a new gl_program object.
 228  */
 229 static void
 230 init_program_struct(struct gl_program *prog, GLenum target, GLuint id)
 231 {
 232    GLuint i;
 233
 234    assert(prog);
 235
 236    memset(prog, 0, sizeof(*prog));
 237    prog->Id = id;
 238    prog->Target = target;
 239    prog->RefCount = 1;
 240    prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 241
 242    /* default mapping from samplers to texture units */
 243    for (i = 0; i < MAX_SAMPLERS; i++)
 244       prog->SamplerUnits[i] = i;
 245 }
 246
 247
 248 /**
 249  * Initialize a new fragment program object.
 250  */
 251 struct gl_program *
 252 _mesa_init_fragment_program(struct gl_context *ctx,
 253                             struct gl_fragment_program *prog,
 254                             GLenum target, GLuint id)
 255 {
 256    if (prog) {
 257       init_program_struct(&prog->Base, target, id);
 258       return &prog->Base;
 259    }
 260    return NULL;
 261 }
 262
 263
 264 /**
 265  * Initialize a new vertex program object.
 266  */
 267 struct gl_program *
 268 _mesa_init_vertex_program(struct gl_context *ctx,
 269                           struct gl_vertex_program *prog,
 270                           GLenum target, GLuint id)
 271 {
 272    if (prog) {
 273       init_program_struct(&prog->Base, target, id);
 274       return &prog->Base;
 275    }
 276    return NULL;
 277 }
 278
 279
 280 /**
 281  * Initialize a new compute program object.
 282  */
 283 struct gl_program *
 284 _mesa_init_compute_program(struct gl_context *ctx,
 285                            struct gl_compute_program *prog,
 286                            GLenum target, GLuint id)
 287 {
 288    if (prog) {
 289       init_program_struct(&prog->Base, target, id);
 290       return &prog->Base;
 291    }
 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,
 301                             struct gl_geometry_program *prog,
 302                             GLenum target, GLuint id)
 303 {
 304    if (prog) {
 305       init_program_struct(&prog->Base, target, id);
 306       return &prog->Base;
 307    }
 308    return NULL;
 309 }
 310
 311
 312 /**
 313  * Allocate and initialize a new fragment/vertex program object but
 314  * don't put it into the program hash table.  Called via
 315  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 316  * device driver function to implement OO deriviation with additional
 317  * types not understood by this function.
 318  *
 319  * \param ctx  context
 320  * \param id   program id/number
 321  * \param target  program target/type
 322  * \return  pointer to new program object
 323  */
 324 struct gl_program *
 325 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id)
 326 {
 327    struct gl_program *prog;
 328    switch (target) {
 329    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 330       prog = _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
 331                                        target, id );
 332       break;
 333    case GL_FRAGMENT_PROGRAM_NV:
 334    case GL_FRAGMENT_PROGRAM_ARB:
 335       prog =_mesa_init_fragment_program(ctx,
 336                                          CALLOC_STRUCT(gl_fragment_program),
 337                                          target, id );
 338       break;
 339    case MESA_GEOMETRY_PROGRAM:
 340       prog = _mesa_init_geometry_program(ctx,
 341                                          CALLOC_STRUCT(gl_geometry_program),
 342                                          target, id);
 343       break;
 344    case GL_COMPUTE_PROGRAM_NV:
 345       prog = _mesa_init_compute_program(ctx,
 346                                         CALLOC_STRUCT(gl_compute_program),
 347                                         target, id);
 348       break;
 349    default:
 350       _mesa_problem(ctx, "bad target in _mesa_new_program");
 351       prog = NULL;
 352    }
 353    return prog;
 354 }
 355
 356
 357 /**
 358  * Delete a program and remove it from the hash table, ignoring the
 359  * reference count.
 360  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 361  * by a device driver function.
 362  */
 363 void
 364 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 365 {
 366    (void) ctx;
 367    assert(prog);
 368    assert(prog->RefCount==0);
 369
 370    if (prog == &_mesa_DummyProgram)
 371       return;
 372
 373    free(prog->String);
 374    free(prog->LocalParams);
 375
 376    if (prog->Instructions) {
 377       _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
 378    }
 379    if (prog->Parameters) {
 380       _mesa_free_parameter_list(prog->Parameters);
 381    }
 382
 383    free(prog);
 384 }
 385
 386
 387 /**
 388  * Return the gl_program object for a given ID.
 389  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 390  * casts elsewhere.
 391  */
 392 struct gl_program *
 393 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 394 {
 395    if (id)
 396       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 397    else
 398       return NULL;
 399 }
 400
 401
 402 /**
 403  * Reference counting for vertex/fragment programs
 404  * This is normally only called from the _mesa_reference_program() macro
 405  * when there's a real pointer change.
 406  */
 407 void
 408 _mesa_reference_program_(struct gl_context *ctx,
 409                          struct gl_program **ptr,
 410                          struct gl_program *prog)
 411 {
 412 #ifndef NDEBUG
 413    assert(ptr);
 414    if (*ptr && prog) {
 415       /* sanity check */
 416       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 417          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 418       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 419          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 420                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 421       else if ((*ptr)->Target == MESA_GEOMETRY_PROGRAM)
 422          assert(prog->Target == MESA_GEOMETRY_PROGRAM);
 423    }
 424 #endif
 425
 426    if (*ptr) {
 427       GLboolean deleteFlag;
 428
 429       /*mtx_lock(&(*ptr)->Mutex);*/
 430 #if 0
 431       printf("Program %p ID=%u Target=%s  Refcount-- to %d\n",
 432              *ptr, (*ptr)->Id,
 433              ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 434               ((*ptr)->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 435              (*ptr)->RefCount - 1);
 436 #endif
 437       assert((*ptr)->RefCount > 0);
 438       (*ptr)->RefCount--;
 439
 440       deleteFlag = ((*ptr)->RefCount == 0);
 441       /*mtx_lock(&(*ptr)->Mutex);*/
 442
 443       if (deleteFlag) {
 444          assert(ctx);
 445          ctx->Driver.DeleteProgram(ctx, *ptr);
 446       }
 447
 448       *ptr = NULL;
 449    }
 450
 451    assert(!*ptr);
 452    if (prog) {
 453       /*mtx_lock(&prog->Mutex);*/
 454       prog->RefCount++;
 455 #if 0
 456       printf("Program %p ID=%u Target=%s  Refcount++ to %d\n",
 457              prog, prog->Id,
 458              (prog->Target == GL_VERTEX_PROGRAM_ARB ? "VP" :
 459               (prog->Target == MESA_GEOMETRY_PROGRAM ? "GP" : "FP")),
 460              prog->RefCount);
 461 #endif
 462       /*mtx_unlock(&prog->Mutex);*/
 463    }
 464
 465    *ptr = prog;
 466 }
 467
 468
 469 /**
 470  * Return a copy of a program.
 471  * XXX Problem here if the program object is actually OO-derivation
 472  * made by a device driver.
 473  */
 474 struct gl_program *
 475 _mesa_clone_program(struct gl_context *ctx, const struct gl_program *prog)
 476 {
 477    struct gl_program *clone;
 478
 479    clone = ctx->Driver.NewProgram(ctx, prog->Target, prog->Id);
 480    if (!clone)
 481       return NULL;
 482
 483    assert(clone->Target == prog->Target);
 484    assert(clone->RefCount == 1);
 485
 486    clone->String = (GLubyte *) strdup((char *) prog->String);
 487    clone->Format = prog->Format;
 488    clone->Instructions = _mesa_alloc_instructions(prog->NumInstructions);
 489    if (!clone->Instructions) {
 490       _mesa_reference_program(ctx, &clone, NULL);
 491       return NULL;
 492    }
 493    _mesa_copy_instructions(clone->Instructions, prog->Instructions,
 494                            prog->NumInstructions);
 495    clone->InputsRead = prog->InputsRead;
 496    clone->OutputsWritten = prog->OutputsWritten;
 497    clone->SamplersUsed = prog->SamplersUsed;
 498    clone->ShadowSamplers = prog->ShadowSamplers;
 499    memcpy(clone->TexturesUsed, prog->TexturesUsed, sizeof(prog->TexturesUsed));
 500
 501    if (prog->Parameters)
 502       clone->Parameters = _mesa_clone_parameter_list(prog->Parameters);
 503    if (prog->LocalParams) {
 504       clone->LocalParams = malloc(MAX_PROGRAM_LOCAL_PARAMS *
 505                                   sizeof(float[4]));
 506       if (!clone->LocalParams) {
 507          _mesa_reference_program(ctx, &clone, NULL);
 508          return NULL;
 509       }
 510       memcpy(clone->LocalParams, prog->LocalParams,
 511              MAX_PROGRAM_LOCAL_PARAMS * sizeof(float[4]));
 512    }
 513    clone->IndirectRegisterFiles = prog->IndirectRegisterFiles;
 514    clone->NumInstructions = prog->NumInstructions;
 515    clone->NumTemporaries = prog->NumTemporaries;
 516    clone->NumParameters = prog->NumParameters;
 517    clone->NumAttributes = prog->NumAttributes;
 518    clone->NumAddressRegs = prog->NumAddressRegs;
 519    clone->NumNativeInstructions = prog->NumNativeInstructions;
 520    clone->NumNativeTemporaries = prog->NumNativeTemporaries;
 521    clone->NumNativeParameters = prog->NumNativeParameters;
 522    clone->NumNativeAttributes = prog->NumNativeAttributes;
 523    clone->NumNativeAddressRegs = prog->NumNativeAddressRegs;
 524    clone->NumAluInstructions = prog->NumAluInstructions;
 525    clone->NumTexInstructions = prog->NumTexInstructions;
 526    clone->NumTexIndirections = prog->NumTexIndirections;
 527    clone->NumNativeAluInstructions = prog->NumNativeAluInstructions;
 528    clone->NumNativeTexInstructions = prog->NumNativeTexInstructions;
 529    clone->NumNativeTexIndirections = prog->NumNativeTexIndirections;
 530
 531    switch (prog->Target) {
 532    case GL_VERTEX_PROGRAM_ARB:
 533       {
 534          const struct gl_vertex_program *vp = gl_vertex_program_const(prog);
 535          struct gl_vertex_program *vpc = gl_vertex_program(clone);
 536          vpc->IsPositionInvariant = vp->IsPositionInvariant;
 537       }
 538       break;
 539    case GL_FRAGMENT_PROGRAM_ARB:
 540       {
 541          const struct gl_fragment_program *fp = gl_fragment_program_const(prog);
 542          struct gl_fragment_program *fpc = gl_fragment_program(clone);
 543          fpc->UsesKill = fp->UsesKill;
 544          fpc->UsesDFdy = fp->UsesDFdy;
 545          fpc->OriginUpperLeft = fp->OriginUpperLeft;
 546          fpc->PixelCenterInteger = fp->PixelCenterInteger;
 547       }
 548       break;
 549    case MESA_GEOMETRY_PROGRAM:
 550       {
 551          const struct gl_geometry_program *gp = gl_geometry_program_const(prog);
 552          struct gl_geometry_program *gpc = gl_geometry_program(clone);
 553          gpc->VerticesOut = gp->VerticesOut;
 554          gpc->InputType = gp->InputType;
 555          gpc->Invocations = gp->Invocations;
 556          gpc->OutputType = gp->OutputType;
 557          gpc->UsesEndPrimitive = gp->UsesEndPrimitive;
 558          gpc->UsesStreams = gp->UsesStreams;
 559       }
 560       break;
 561    default:
 562       _mesa_problem(NULL, "Unexpected target in _mesa_clone_program");
 563    }
 564
 565    return clone;
 566 }
 567
 568
 569 /**
 570  * Insert 'count' NOP instructions at 'start' in the given program.
 571  * Adjust branch targets accordingly.
 572  */
 573 GLboolean
 574 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 575 {
 576    const GLuint origLen = prog->NumInstructions;
 577    const GLuint newLen = origLen + count;
 578    struct prog_instruction *newInst;
 579    GLuint i;
 580
 581    /* adjust branches */
 582    for (i = 0; i < prog->NumInstructions; i++) {
 583       struct prog_instruction *inst = prog->Instructions + i;
 584       if (inst->BranchTarget > 0) {
 585          if ((GLuint)inst->BranchTarget >= start) {
 586             inst->BranchTarget += count;
 587          }
 588       }
 589    }
 590
 591    /* Alloc storage for new instructions */
 592    newInst = _mesa_alloc_instructions(newLen);
 593    if (!newInst) {
 594       return GL_FALSE;
 595    }
 596
 597    /* Copy 'start' instructions into new instruction buffer */
 598    _mesa_copy_instructions(newInst, prog->Instructions, start);
 599
 600    /* init the new instructions */
 601    _mesa_init_instructions(newInst + start, count);
 602
 603    /* Copy the remaining/tail instructions to new inst buffer */
 604    _mesa_copy_instructions(newInst + start + count,
 605                            prog->Instructions + start,
 606                            origLen - start);
 607
 608    /* free old instructions */
 609    _mesa_free_instructions(prog->Instructions, origLen);
 610
 611    /* install new instructions */
 612    prog->Instructions = newInst;
 613    prog->NumInstructions = newLen;
 614
 615    return GL_TRUE;
 616 }
 617
 618 /**
 619  * Delete 'count' instructions at 'start' in the given program.
 620  * Adjust branch targets accordingly.
 621  */
 622 GLboolean
 623 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
 624 {
 625    const GLuint origLen = prog->NumInstructions;
 626    const GLuint newLen = origLen - count;
 627    struct prog_instruction *newInst;
 628    GLuint i;
 629
 630    /* adjust branches */
 631    for (i = 0; i < prog->NumInstructions; i++) {
 632       struct prog_instruction *inst = prog->Instructions + i;
 633       if (inst->BranchTarget > 0) {
 634          if (inst->BranchTarget > (GLint) start) {
 635             inst->BranchTarget -= count;
 636          }
 637       }
 638    }
 639
 640    /* Alloc storage for new instructions */
 641    newInst = _mesa_alloc_instructions(newLen);
 642    if (!newInst) {
 643       return GL_FALSE;
 644    }
 645
 646    /* Copy 'start' instructions into new instruction buffer */
 647    _mesa_copy_instructions(newInst, prog->Instructions, start);
 648
 649    /* Copy the remaining/tail instructions to new inst buffer */
 650    _mesa_copy_instructions(newInst + start,
 651                            prog->Instructions + start + count,
 652                            newLen - start);
 653
 654    /* free old instructions */
 655    _mesa_free_instructions(prog->Instructions, origLen);
 656
 657    /* install new instructions */
 658    prog->Instructions = newInst;
 659    prog->NumInstructions = newLen;
 660
 661    return GL_TRUE;
 662 }
 663
 664
 665 /**
 666  * Search instructions for registers that match (oldFile, oldIndex),
 667  * replacing them with (newFile, newIndex).
 668  */
 669 static void
 670 replace_registers(struct prog_instruction *inst, GLuint numInst,
 671                   GLuint oldFile, GLuint oldIndex,
 672                   GLuint newFile, GLuint newIndex)
 673 {
 674    GLuint i, j;
 675    for (i = 0; i < numInst; i++) {
 676       /* src regs */
 677       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 678          if (inst[i].SrcReg[j].File == oldFile &&
 679              inst[i].SrcReg[j].Index == oldIndex) {
 680             inst[i].SrcReg[j].File = newFile;
 681             inst[i].SrcReg[j].Index = newIndex;
 682          }
 683       }
 684       /* dst reg */
 685       if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) {
 686          inst[i].DstReg.File = newFile;
 687          inst[i].DstReg.Index = newIndex;
 688       }
 689    }
 690 }
 691
 692
 693 /**
 694  * Search instructions for references to program parameters.  When found,
 695  * increment the parameter index by 'offset'.
 696  * Used when combining programs.
 697  */
 698 static void
 699 adjust_param_indexes(struct prog_instruction *inst, GLuint numInst,
 700                      GLuint offset)
 701 {
 702    GLuint i, j;
 703    for (i = 0; i < numInst; i++) {
 704       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 705          GLuint f = inst[i].SrcReg[j].File;
 706          if (f == PROGRAM_CONSTANT ||
 707              f == PROGRAM_UNIFORM ||
 708              f == PROGRAM_STATE_VAR) {
 709             inst[i].SrcReg[j].Index += offset;
 710          }
 711       }
 712    }
 713 }
 714
 715
 716 /**
 717  * Combine two programs into one.  Fix instructions so the outputs of
 718  * the first program go to the inputs of the second program.
 719  */
 720 struct gl_program *
 721 _mesa_combine_programs(struct gl_context *ctx,
 722                        const struct gl_program *progA,
 723                        const struct gl_program *progB)
 724 {
 725    struct prog_instruction *newInst;
 726    struct gl_program *newProg;
 727    const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
 728    const GLuint lenB = progB->NumInstructions;
 729    const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
 730    const GLuint newLength = lenA + lenB;
 731    GLboolean usedTemps[MAX_PROGRAM_TEMPS];
 732    GLuint firstTemp = 0;
 733    GLbitfield64 inputsB;
 734    GLuint i;
 735
 736    assert(progA->Target == progB->Target);
 737
 738    newInst = _mesa_alloc_instructions(newLength);
 739    if (!newInst)
 740       return GL_FALSE;
 741
 742    _mesa_copy_instructions(newInst, progA->Instructions, lenA);
 743    _mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
 744
 745    /* adjust branch / instruction addresses for B's instructions */
 746    for (i = 0; i < lenB; i++) {
 747       newInst[lenA + i].BranchTarget += lenA;
 748    }
 749
 750    newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
 751    newProg->Instructions = newInst;
 752    newProg->NumInstructions = newLength;
 753
 754    /* find used temp regs (we may need new temps below) */
 755    _mesa_find_used_registers(newProg, PROGRAM_TEMPORARY,
 756                              usedTemps, MAX_PROGRAM_TEMPS);
 757
 758    if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
 759       const struct gl_fragment_program *fprogA, *fprogB;
 760       struct gl_fragment_program *newFprog;
 761       GLbitfield64 progB_inputsRead = progB->InputsRead;
 762       GLint progB_colorFile, progB_colorIndex;
 763
 764       fprogA = gl_fragment_program_const(progA);
 765       fprogB = gl_fragment_program_const(progB);
 766       newFprog = gl_fragment_program(newProg);
 767
 768       newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
 769       newFprog->UsesDFdy = fprogA->UsesDFdy || fprogB->UsesDFdy;
 770
 771       /* We'll do a search and replace for instances
 772        * of progB_colorFile/progB_colorIndex below...
 773        */
 774       progB_colorFile = PROGRAM_INPUT;
 775       progB_colorIndex = VARYING_SLOT_COL0;
 776
 777       /*
 778        * The fragment program may get color from a state var rather than
 779        * a fragment input (vertex output) if it's constant.
 780        * See the texenvprogram.c code.
 781        * So, search the program's parameter list now to see if the program
 782        * gets color from a state var instead of a conventional fragment
 783        * input register.
 784        */
 785       for (i = 0; i < progB->Parameters->NumParameters; i++) {
 786          struct gl_program_parameter *p = &progB->Parameters->Parameters[i];
 787          if (p->Type == PROGRAM_STATE_VAR &&
 788              p->StateIndexes[0] == STATE_INTERNAL &&
 789              p->StateIndexes[1] == STATE_CURRENT_ATTRIB &&
 790              (int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) {
 791             progB_inputsRead |= VARYING_BIT_COL0;
 792             progB_colorFile = PROGRAM_STATE_VAR;
 793             progB_colorIndex = i;
 794             break;
 795          }
 796       }
 797
 798       /* Connect color outputs of fprogA to color inputs of fprogB, via a
 799        * new temporary register.
 800        */
 801       if ((progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) &&
 802           (progB_inputsRead & VARYING_BIT_COL0)) {
 803          GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS,
 804                                                   firstTemp);
 805          if (tempReg < 0) {
 806             _mesa_problem(ctx, "No free temp regs found in "
 807                           "_mesa_combine_programs(), using 31");
 808             tempReg = 31;
 809          }
 810          firstTemp = tempReg + 1;
 811
 812          /* replace writes to result.color[0] with tempReg */
 813          replace_registers(newInst, lenA,
 814                            PROGRAM_OUTPUT, FRAG_RESULT_COLOR,
 815                            PROGRAM_TEMPORARY, tempReg);
 816          /* replace reads from the input color with tempReg */
 817          replace_registers(newInst + lenA, lenB,
 818                            progB_colorFile, progB_colorIndex, /* search for */
 819                            PROGRAM_TEMPORARY, tempReg  /* replace with */ );
 820       }
 821
 822       /* compute combined program's InputsRead */
 823       inputsB = progB_inputsRead;
 824       if (progA->OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) {
 825          inputsB &= ~(1 << VARYING_SLOT_COL0);
 826       }
 827       newProg->InputsRead = progA->InputsRead | inputsB;
 828       newProg->OutputsWritten = progB->OutputsWritten;
 829       newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
 830    }
 831    else {
 832       /* vertex program */
 833       assert(0);      /* XXX todo */
 834    }
 835
 836    /*
 837     * Merge parameters (uniforms, constants, etc)
 838     */
 839    newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
 840                                                        progB->Parameters);
 841
 842    adjust_param_indexes(newInst + lenA, lenB, numParamsA);
 843
 844
 845    return newProg;
 846 }
 847
 848
 849 /**
 850  * Populate the 'used' array with flags indicating which registers (TEMPs,
 851  * INPUTs, OUTPUTs, etc, are used by the given program.
 852  * \param file  type of register to scan for
 853  * \param used  returns true/false flags for in use / free
 854  * \param usedSize  size of the 'used' array
 855  */
 856 void
 857 _mesa_find_used_registers(const struct gl_program *prog,
 858                           gl_register_file file,
 859                           GLboolean used[], GLuint usedSize)
 860 {
 861    GLuint i, j;
 862
 863    memset(used, 0, usedSize);
 864
 865    for (i = 0; i < prog->NumInstructions; i++) {
 866       const struct prog_instruction *inst = prog->Instructions + i;
 867       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 868
 869       if (inst->DstReg.File == file) {
 870          assert(inst->DstReg.Index < usedSize);
 871          if(inst->DstReg.Index < usedSize)
 872             used[inst->DstReg.Index] = GL_TRUE;
 873       }
 874
 875       for (j = 0; j < n; j++) {
 876          if (inst->SrcReg[j].File == file) {
 877             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 878             if (inst->SrcReg[j].Index < (GLint) usedSize)
 879                used[inst->SrcReg[j].Index] = GL_TRUE;
 880          }
 881       }
 882    }
 883 }
 884
 885
 886 /**
 887  * Scan the given 'used' register flag array for the first entry
 888  * that's >= firstReg.
 889  * \param used  vector of flags indicating registers in use (as returned
 890  *              by _mesa_find_used_registers())
 891  * \param usedSize  size of the 'used' array
 892  * \param firstReg  first register to start searching at
 893  * \return index of unused register, or -1 if none.
 894  */
 895 GLint
 896 _mesa_find_free_register(const GLboolean used[],
 897                          GLuint usedSize, GLuint firstReg)
 898 {
 899    GLuint i;
 900
 901    assert(firstReg < usedSize);
 902
 903    for (i = firstReg; i < usedSize; i++)
 904       if (!used[i])
 905          return i;
 906
 907    return -1;
 908 }
 909
 910
 911
 912 /**
 913  * Check if the given register index is valid (doesn't exceed implementation-
 914  * dependent limits).
 915  * \return GL_TRUE if OK, GL_FALSE if bad index
 916  */
 917 GLboolean
 918 _mesa_valid_register_index(const struct gl_context *ctx,
 919                            gl_shader_stage shaderType,
 920                            gl_register_file file, GLint index)
 921 {
 922    const struct gl_program_constants *c;
 923
 924    assert(0 <= shaderType && shaderType < MESA_SHADER_STAGES);
 925    c = &ctx->Const.Program[shaderType];
 926
 927    switch (file) {
 928    case PROGRAM_UNDEFINED:
 929       return GL_TRUE;  /* XXX or maybe false? */
 930
 931    case PROGRAM_TEMPORARY:
 932       return index >= 0 && index < (GLint) c->MaxTemps;
 933
 934    case PROGRAM_UNIFORM:
 935    case PROGRAM_STATE_VAR:
 936       /* aka constant buffer */
 937       return index >= 0 && index < (GLint) c->MaxUniformComponents / 4;
 938
 939    case PROGRAM_CONSTANT:
 940       /* constant buffer w/ possible relative negative addressing */
 941       return (index > (int) c->MaxUniformComponents / -4 &&
 942               index < (int) c->MaxUniformComponents / 4);
 943
 944    case PROGRAM_INPUT:
 945       if (index < 0)
 946          return GL_FALSE;
 947
 948       switch (shaderType) {
 949       case MESA_SHADER_VERTEX:
 950          return index < VERT_ATTRIB_GENERIC0 + (GLint) c->MaxAttribs;
 951       case MESA_SHADER_FRAGMENT:
 952          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 953       case MESA_SHADER_GEOMETRY:
 954          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 955       default:
 956          return GL_FALSE;
 957       }
 958
 959    case PROGRAM_OUTPUT:
 960       if (index < 0)
 961          return GL_FALSE;
 962
 963       switch (shaderType) {
 964       case MESA_SHADER_VERTEX:
 965          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 966       case MESA_SHADER_FRAGMENT:
 967          return index < FRAG_RESULT_DATA0 + (GLint) ctx->Const.MaxDrawBuffers;
 968       case MESA_SHADER_GEOMETRY:
 969          return index < VARYING_SLOT_VAR0 + (GLint) ctx->Const.MaxVarying;
 970       default:
 971          return GL_FALSE;
 972       }
 973
 974    case PROGRAM_ADDRESS:
 975       return index >= 0 && index < (GLint) c->MaxAddressRegs;
 976
 977    default:
 978       _mesa_problem(ctx,
 979                     "unexpected register file in _mesa_valid_register_index()");
 980       return GL_FALSE;
 981    }
 982 }
 983
 984
 985
 986 /**
 987  * "Post-process" a GPU program.  This is intended to be used for debugging.
 988  * Example actions include no-op'ing instructions or changing instruction
 989  * behaviour.
 990  */
 991 void
 992 _mesa_postprocess_program(struct gl_context *ctx, struct gl_program *prog)
 993 {
 994    static const GLfloat white[4] = { 0.5, 0.5, 0.5, 0.5 };
 995    GLuint i;
 996    GLuint whiteSwizzle;
 997    GLint whiteIndex = _mesa_add_unnamed_constant(prog->Parameters,
 998                                                  (gl_constant_value *) white,
 999                                                  4, &whiteSwizzle);
1000
1001    (void) whiteIndex;
1002
1003    for (i = 0; i < prog->NumInstructions; i++) {
1004       struct prog_instruction *inst = prog->Instructions + i;
1005       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
1006
1007       (void) n;
1008
1009       if (_mesa_is_tex_instruction(inst->Opcode)) {
1010 #if 0
1011          /* replace TEX/TXP/TXB with MOV */
1012          inst->Opcode = OPCODE_MOV;
1013          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1014          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1015          inst->SrcReg[0].Negate = NEGATE_NONE;
1016 #endif
1017
1018 #if 0
1019          /* disable shadow texture mode */
1020          inst->TexShadow = 0;
1021 #endif
1022       }
1023
1024       if (inst->Opcode == OPCODE_TXP) {
1025 #if 0
1026          inst->Opcode = OPCODE_MOV;
1027          inst->DstReg.WriteMask = WRITEMASK_XYZW;
1028          inst->SrcReg[0].File = PROGRAM_CONSTANT;
1029          inst->SrcReg[0].Index = whiteIndex;
1030          inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
1031          inst->SrcReg[0].Negate = NEGATE_NONE;
1032 #endif
1033 #if 0
1034          inst->TexShadow = 0;
1035 #endif
1036 #if 0
1037          inst->Opcode = OPCODE_TEX;
1038          inst->TexShadow = 0;
1039 #endif
1040       }
1041
1042    }
1043 }
1044
1045 /* Gets the minimum number of shader invocations per fragment.
1046  * This function is useful to determine if we need to do per
1047  * sample shading or per fragment shading.
1048  */
1049 GLint
1050 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
1051                                        const struct gl_fragment_program *prog,
1052                                        bool ignore_sample_qualifier)
1053 {
1054    /* From ARB_sample_shading specification:
1055     * "Using gl_SampleID in a fragment shader causes the entire shader
1056     *  to be evaluated per-sample."
1057     *
1058     * "Using gl_SamplePosition in a fragment shader causes the entire
1059     *  shader to be evaluated per-sample."
1060     *
1061     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
1062     *  has no effect."
1063     */
1064    if (ctx->Multisample.Enabled) {
1065       /* The ARB_gpu_shader5 specification says:
1066        *
1067        * "Use of the "sample" qualifier on a fragment shader input
1068        *  forces per-sample shading"
1069        */
1070       if (prog->IsSample && !ignore_sample_qualifier)
1071          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1072
1073       if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
1074                                          SYSTEM_BIT_SAMPLE_POS))
1075          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
1076       else if (ctx->Multisample.SampleShading)
1077          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
1078                           ctx->DrawBuffer->Visual.samples), 1);
1079       else
1080          return 1;
1081    }
1082    return 1;
1083 }