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 #include "util/ralloc.h"
  41
  42
  43 /**
  44  * A pointer to this dummy program is put into the hash table when
  45  * glGenPrograms is called.
  46  */
  47 struct gl_program _mesa_DummyProgram;
  48
  49
  50 /**
  51  * Init context's vertex/fragment program state
  52  */
  53 void
  54 _mesa_init_program(struct gl_context *ctx)
  55 {
  56    /*
  57     * If this assertion fails, we need to increase the field
  58     * size for register indexes (see INST_INDEX_BITS).
  59     */
  60    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents / 4
  61           <= (1 << INST_INDEX_BITS));
  62    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents / 4
  63           <= (1 << INST_INDEX_BITS));
  64
  65    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps <= (1 << INST_INDEX_BITS));
  66    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxLocalParams <= (1 << INST_INDEX_BITS));
  67    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTemps <= (1 << INST_INDEX_BITS));
  68    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxLocalParams <= (1 << INST_INDEX_BITS));
  69
  70    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  71    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  72
  73    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  74    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  75
  76    /* If this fails, increase prog_instruction::TexSrcUnit size */
  77    STATIC_ASSERT(MAX_TEXTURE_UNITS <= (1 << 5));
  78
  79    /* If this fails, increase prog_instruction::TexSrcTarget size */
  80    STATIC_ASSERT(NUM_TEXTURE_TARGETS <= (1 << 4));
  81
  82    ctx->Program.ErrorPos = -1;
  83    ctx->Program.ErrorString = strdup("");
  84
  85    ctx->VertexProgram.Enabled = GL_FALSE;
  86    ctx->VertexProgram.PointSizeEnabled =
  87       (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE;
  88    ctx->VertexProgram.TwoSideEnabled = GL_FALSE;
  89    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
  90                             ctx->Shared->DefaultVertexProgram);
  91    assert(ctx->VertexProgram.Current);
  92    ctx->VertexProgram.Cache = _mesa_new_program_cache();
  93
  94    ctx->FragmentProgram.Enabled = GL_FALSE;
  95    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
  96                             ctx->Shared->DefaultFragmentProgram);
  97    assert(ctx->FragmentProgram.Current);
  98    ctx->FragmentProgram.Cache = _mesa_new_program_cache();
  99
 100    /* XXX probably move this stuff */
 101    ctx->ATIFragmentShader.Enabled = GL_FALSE;
 102    ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
 103    assert(ctx->ATIFragmentShader.Current);
 104    ctx->ATIFragmentShader.Current->RefCount++;
 105 }
 106
 107
 108 /**
 109  * Free a context's vertex/fragment program state
 110  */
 111 void
 112 _mesa_free_program_data(struct gl_context *ctx)
 113 {
 114    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current, NULL);
 115    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 116    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current, NULL);
 117    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 118
 119    /* XXX probably move this stuff */
 120    if (ctx->ATIFragmentShader.Current) {
 121       ctx->ATIFragmentShader.Current->RefCount--;
 122       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 123          free(ctx->ATIFragmentShader.Current);
 124       }
 125    }
 126
 127    free((void *) ctx->Program.ErrorString);
 128 }
 129
 130
 131 /**
 132  * Update the default program objects in the given context to reference those
 133  * specified in the shared state and release those referencing the old
 134  * shared state.
 135  */
 136 void
 137 _mesa_update_default_objects_program(struct gl_context *ctx)
 138 {
 139    _mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
 140                             ctx->Shared->DefaultVertexProgram);
 141    assert(ctx->VertexProgram.Current);
 142
 143    _mesa_reference_fragprog(ctx, &ctx->FragmentProgram.Current,
 144                             ctx->Shared->DefaultFragmentProgram);
 145    assert(ctx->FragmentProgram.Current);
 146
 147    /* XXX probably move this stuff */
 148    if (ctx->ATIFragmentShader.Current) {
 149       ctx->ATIFragmentShader.Current->RefCount--;
 150       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 151          free(ctx->ATIFragmentShader.Current);
 152       }
 153    }
 154    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 155    assert(ctx->ATIFragmentShader.Current);
 156    ctx->ATIFragmentShader.Current->RefCount++;
 157 }
 158
 159
 160 /**
 161  * Set the vertex/fragment program error state (position and error string).
 162  * This is generally called from within the parsers.
 163  */
 164 void
 165 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 166 {
 167    ctx->Program.ErrorPos = pos;
 168    free((void *) ctx->Program.ErrorString);
 169    if (!string)
 170       string = "";
 171    ctx->Program.ErrorString = strdup(string);
 172 }
 173
 174
 175 /**
 176  * Initialize a new gl_program object.
 177  */
 178 struct gl_program *
 179 _mesa_init_gl_program(void *_prog, GLenum target, GLuint id)
 180 {
 181    struct gl_program *prog = (struct gl_program*)_prog;
 182    GLuint i;
 183
 184    if (!prog)
 185       return NULL;
 186
 187    memset(prog, 0, sizeof(*prog));
 188    mtx_init(&prog->Mutex, mtx_plain);
 189    prog->Id = id;
 190    prog->Target = target;
 191    prog->RefCount = 1;
 192    prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 193
 194    /* default mapping from samplers to texture units */
 195    for (i = 0; i < MAX_SAMPLERS; i++)
 196       prog->SamplerUnits[i] = i;
 197
 198    return prog;
 199 }
 200
 201
 202 /**
 203  * Allocate and initialize a new fragment/vertex program object but
 204  * don't put it into the program hash table.  Called via
 205  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 206  * device driver function to implement OO deriviation with additional
 207  * types not understood by this function.
 208  *
 209  * \param ctx  context
 210  * \param id   program id/number
 211  * \param target  program target/type
 212  * \return  pointer to new program object
 213  */
 214 struct gl_program *
 215 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id)
 216 {
 217    struct gl_program *prog;
 218    switch (target) {
 219    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 220       prog = _mesa_init_gl_program(CALLOC_STRUCT(gl_vertex_program),
 221                                    target, id);
 222       break;
 223    case GL_FRAGMENT_PROGRAM_NV:
 224    case GL_FRAGMENT_PROGRAM_ARB:
 225       prog =_mesa_init_gl_program(CALLOC_STRUCT(gl_fragment_program),
 226                                   target, id);
 227       break;
 228    case GL_GEOMETRY_PROGRAM_NV:
 229       prog = _mesa_init_gl_program(CALLOC_STRUCT(gl_geometry_program),
 230                                    target, id);
 231       break;
 232    case GL_TESS_CONTROL_PROGRAM_NV:
 233       prog = _mesa_init_gl_program(CALLOC_STRUCT(gl_tess_ctrl_program),
 234                                    target, id);
 235       break;
 236    case GL_TESS_EVALUATION_PROGRAM_NV:
 237       prog = _mesa_init_gl_program(CALLOC_STRUCT(gl_tess_eval_program),
 238                                    target, id);
 239       break;
 240    case GL_COMPUTE_PROGRAM_NV:
 241       prog = _mesa_init_gl_program(CALLOC_STRUCT(gl_compute_program),
 242                                    target, id);
 243       break;
 244    default:
 245       _mesa_problem(ctx, "bad target in _mesa_new_program");
 246       prog = NULL;
 247    }
 248    return prog;
 249 }
 250
 251
 252 /**
 253  * Delete a program and remove it from the hash table, ignoring the
 254  * reference count.
 255  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 256  * by a device driver function.
 257  */
 258 void
 259 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 260 {
 261    (void) ctx;
 262    assert(prog);
 263    assert(prog->RefCount==0);
 264
 265    if (prog == &_mesa_DummyProgram)
 266       return;
 267
 268    free(prog->String);
 269    free(prog->LocalParams);
 270
 271    if (prog->Instructions) {
 272       _mesa_free_instructions(prog->Instructions, prog->NumInstructions);
 273    }
 274    if (prog->Parameters) {
 275       _mesa_free_parameter_list(prog->Parameters);
 276    }
 277
 278    if (prog->nir) {
 279       ralloc_free(prog->nir);
 280    }
 281
 282    mtx_destroy(&prog->Mutex);
 283    free(prog);
 284 }
 285
 286
 287 /**
 288  * Return the gl_program object for a given ID.
 289  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 290  * casts elsewhere.
 291  */
 292 struct gl_program *
 293 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 294 {
 295    if (id)
 296       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 297    else
 298       return NULL;
 299 }
 300
 301
 302 /**
 303  * Reference counting for vertex/fragment programs
 304  * This is normally only called from the _mesa_reference_program() macro
 305  * when there's a real pointer change.
 306  */
 307 void
 308 _mesa_reference_program_(struct gl_context *ctx,
 309                          struct gl_program **ptr,
 310                          struct gl_program *prog)
 311 {
 312 #ifndef NDEBUG
 313    assert(ptr);
 314    if (*ptr && prog) {
 315       /* sanity check */
 316       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 317          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 318       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 319          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 320                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 321       else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
 322          assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
 323    }
 324 #endif
 325
 326    if (*ptr) {
 327       GLboolean deleteFlag;
 328       struct gl_program *oldProg = *ptr;
 329
 330       mtx_lock(&oldProg->Mutex);
 331       assert(oldProg->RefCount > 0);
 332       oldProg->RefCount--;
 333
 334       deleteFlag = (oldProg->RefCount == 0);
 335       mtx_unlock(&oldProg->Mutex);
 336
 337       if (deleteFlag) {
 338          assert(ctx);
 339          ctx->Driver.DeleteProgram(ctx, oldProg);
 340       }
 341
 342       *ptr = NULL;
 343    }
 344
 345    assert(!*ptr);
 346    if (prog) {
 347       mtx_lock(&prog->Mutex);
 348       prog->RefCount++;
 349       mtx_unlock(&prog->Mutex);
 350    }
 351
 352    *ptr = prog;
 353 }
 354
 355
 356 /**
 357  * Insert 'count' NOP instructions at 'start' in the given program.
 358  * Adjust branch targets accordingly.
 359  */
 360 GLboolean
 361 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 362 {
 363    const GLuint origLen = prog->NumInstructions;
 364    const GLuint newLen = origLen + count;
 365    struct prog_instruction *newInst;
 366    GLuint i;
 367
 368    /* adjust branches */
 369    for (i = 0; i < prog->NumInstructions; i++) {
 370       struct prog_instruction *inst = prog->Instructions + i;
 371       if (inst->BranchTarget > 0) {
 372          if ((GLuint)inst->BranchTarget >= start) {
 373             inst->BranchTarget += count;
 374          }
 375       }
 376    }
 377
 378    /* Alloc storage for new instructions */
 379    newInst = _mesa_alloc_instructions(newLen);
 380    if (!newInst) {
 381       return GL_FALSE;
 382    }
 383
 384    /* Copy 'start' instructions into new instruction buffer */
 385    _mesa_copy_instructions(newInst, prog->Instructions, start);
 386
 387    /* init the new instructions */
 388    _mesa_init_instructions(newInst + start, count);
 389
 390    /* Copy the remaining/tail instructions to new inst buffer */
 391    _mesa_copy_instructions(newInst + start + count,
 392                            prog->Instructions + start,
 393                            origLen - start);
 394
 395    /* free old instructions */
 396    _mesa_free_instructions(prog->Instructions, origLen);
 397
 398    /* install new instructions */
 399    prog->Instructions = newInst;
 400    prog->NumInstructions = newLen;
 401
 402    return GL_TRUE;
 403 }
 404
 405 /**
 406  * Delete 'count' instructions at 'start' in the given program.
 407  * Adjust branch targets accordingly.
 408  */
 409 GLboolean
 410 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count)
 411 {
 412    const GLuint origLen = prog->NumInstructions;
 413    const GLuint newLen = origLen - count;
 414    struct prog_instruction *newInst;
 415    GLuint i;
 416
 417    /* adjust branches */
 418    for (i = 0; i < prog->NumInstructions; i++) {
 419       struct prog_instruction *inst = prog->Instructions + i;
 420       if (inst->BranchTarget > 0) {
 421          if (inst->BranchTarget > (GLint) start) {
 422             inst->BranchTarget -= count;
 423          }
 424       }
 425    }
 426
 427    /* Alloc storage for new instructions */
 428    newInst = _mesa_alloc_instructions(newLen);
 429    if (!newInst) {
 430       return GL_FALSE;
 431    }
 432
 433    /* Copy 'start' instructions into new instruction buffer */
 434    _mesa_copy_instructions(newInst, prog->Instructions, start);
 435
 436    /* Copy the remaining/tail instructions to new inst buffer */
 437    _mesa_copy_instructions(newInst + start,
 438                            prog->Instructions + start + count,
 439                            newLen - start);
 440
 441    /* free old instructions */
 442    _mesa_free_instructions(prog->Instructions, origLen);
 443
 444    /* install new instructions */
 445    prog->Instructions = newInst;
 446    prog->NumInstructions = newLen;
 447
 448    return GL_TRUE;
 449 }
 450
 451
 452 /**
 453  * Search instructions for registers that match (oldFile, oldIndex),
 454  * replacing them with (newFile, newIndex).
 455  */
 456 static void
 457 replace_registers(struct prog_instruction *inst, GLuint numInst,
 458                   GLuint oldFile, GLuint oldIndex,
 459                   GLuint newFile, GLuint newIndex)
 460 {
 461    GLuint i, j;
 462    for (i = 0; i < numInst; i++) {
 463       /* src regs */
 464       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 465          if (inst[i].SrcReg[j].File == oldFile &&
 466              inst[i].SrcReg[j].Index == oldIndex) {
 467             inst[i].SrcReg[j].File = newFile;
 468             inst[i].SrcReg[j].Index = newIndex;
 469          }
 470       }
 471       /* dst reg */
 472       if (inst[i].DstReg.File == oldFile && inst[i].DstReg.Index == oldIndex) {
 473          inst[i].DstReg.File = newFile;
 474          inst[i].DstReg.Index = newIndex;
 475       }
 476    }
 477 }
 478
 479
 480 /**
 481  * Search instructions for references to program parameters.  When found,
 482  * increment the parameter index by 'offset'.
 483  * Used when combining programs.
 484  */
 485 static void
 486 adjust_param_indexes(struct prog_instruction *inst, GLuint numInst,
 487                      GLuint offset)
 488 {
 489    GLuint i, j;
 490    for (i = 0; i < numInst; i++) {
 491       for (j = 0; j < _mesa_num_inst_src_regs(inst[i].Opcode); j++) {
 492          GLuint f = inst[i].SrcReg[j].File;
 493          if (f == PROGRAM_CONSTANT ||
 494              f == PROGRAM_UNIFORM ||
 495              f == PROGRAM_STATE_VAR) {
 496             inst[i].SrcReg[j].Index += offset;
 497          }
 498       }
 499    }
 500 }
 501
 502
 503 /**
 504  * Populate the 'used' array with flags indicating which registers (TEMPs,
 505  * INPUTs, OUTPUTs, etc, are used by the given program.
 506  * \param file  type of register to scan for
 507  * \param used  returns true/false flags for in use / free
 508  * \param usedSize  size of the 'used' array
 509  */
 510 void
 511 _mesa_find_used_registers(const struct gl_program *prog,
 512                           gl_register_file file,
 513                           GLboolean used[], GLuint usedSize)
 514 {
 515    GLuint i, j;
 516
 517    memset(used, 0, usedSize);
 518
 519    for (i = 0; i < prog->NumInstructions; i++) {
 520       const struct prog_instruction *inst = prog->Instructions + i;
 521       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 522
 523       if (inst->DstReg.File == file) {
 524          assert(inst->DstReg.Index < usedSize);
 525          if(inst->DstReg.Index < usedSize)
 526             used[inst->DstReg.Index] = GL_TRUE;
 527       }
 528
 529       for (j = 0; j < n; j++) {
 530          if (inst->SrcReg[j].File == file) {
 531             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 532             if (inst->SrcReg[j].Index < (GLint) usedSize)
 533                used[inst->SrcReg[j].Index] = GL_TRUE;
 534          }
 535       }
 536    }
 537 }
 538
 539
 540 /**
 541  * Scan the given 'used' register flag array for the first entry
 542  * that's >= firstReg.
 543  * \param used  vector of flags indicating registers in use (as returned
 544  *              by _mesa_find_used_registers())
 545  * \param usedSize  size of the 'used' array
 546  * \param firstReg  first register to start searching at
 547  * \return index of unused register, or -1 if none.
 548  */
 549 GLint
 550 _mesa_find_free_register(const GLboolean used[],
 551                          GLuint usedSize, GLuint firstReg)
 552 {
 553    GLuint i;
 554
 555    assert(firstReg < usedSize);
 556
 557    for (i = firstReg; i < usedSize; i++)
 558       if (!used[i])
 559          return i;
 560
 561    return -1;
 562 }
 563
 564
 565 /* Gets the minimum number of shader invocations per fragment.
 566  * This function is useful to determine if we need to do per
 567  * sample shading or per fragment shading.
 568  */
 569 GLint
 570 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
 571                                        const struct gl_fragment_program *prog,
 572                                        bool ignore_sample_qualifier)
 573 {
 574    /* From ARB_sample_shading specification:
 575     * "Using gl_SampleID in a fragment shader causes the entire shader
 576     *  to be evaluated per-sample."
 577     *
 578     * "Using gl_SamplePosition in a fragment shader causes the entire
 579     *  shader to be evaluated per-sample."
 580     *
 581     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
 582     *  has no effect."
 583     */
 584    if (ctx->Multisample.Enabled) {
 585       /* The ARB_gpu_shader5 specification says:
 586        *
 587        * "Use of the "sample" qualifier on a fragment shader input
 588        *  forces per-sample shading"
 589        */
 590       if (prog->IsSample && !ignore_sample_qualifier)
 591          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
 592
 593       if (prog->Base.SystemValuesRead & (SYSTEM_BIT_SAMPLE_ID |
 594                                          SYSTEM_BIT_SAMPLE_POS))
 595          return MAX2(ctx->DrawBuffer->Visual.samples, 1);
 596       else if (ctx->Multisample.SampleShading)
 597          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
 598                           ctx->DrawBuffer->Visual.samples), 1);
 599       else
 600          return 1;
 601    }
 602    return 1;
 603 }