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/framebuffer.h"
  35 #include "main/hash.h"
  36 #include "main/macros.h"
  37 #include "main/shaderobj.h"
  38 #include "program.h"
  39 #include "prog_cache.h"
  40 #include "prog_parameter.h"
  41 #include "prog_instruction.h"
  42 #include "util/bitscan.h"
  43 #include "util/ralloc.h"
  44 #include "util/u_atomic.h"
  45
  46
  47 /**
  48  * A pointer to this dummy program is put into the hash table when
  49  * glGenPrograms is called.
  50  */
  51 struct gl_program _mesa_DummyProgram;
  52
  53
  54 /**
  55  * Init context's vertex/fragment program state
  56  */
  57 void
  58 _mesa_init_program(struct gl_context *ctx)
  59 {
  60    /*
  61     * If this assertion fails, we need to increase the field
  62     * size for register indexes (see INST_INDEX_BITS).
  63     */
  64    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents / 4
  65           <= (1 << INST_INDEX_BITS));
  66    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents / 4
  67           <= (1 << INST_INDEX_BITS));
  68
  69    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps <= (1 << INST_INDEX_BITS));
  70    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxLocalParams <= (1 << INST_INDEX_BITS));
  71    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTemps <= (1 << INST_INDEX_BITS));
  72    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxLocalParams <= (1 << INST_INDEX_BITS));
  73
  74    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  75    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  76
  77    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  78    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  79
  80    /* If this fails, increase prog_instruction::TexSrcUnit size */
  81    STATIC_ASSERT(MAX_TEXTURE_UNITS <= (1 << 5));
  82
  83    /* If this fails, increase prog_instruction::TexSrcTarget size */
  84    STATIC_ASSERT(NUM_TEXTURE_TARGETS <= (1 << 4));
  85
  86    ctx->Program.ErrorPos = -1;
  87    ctx->Program.ErrorString = strdup("");
  88
  89    ctx->VertexProgram.Enabled = GL_FALSE;
  90    ctx->VertexProgram.PointSizeEnabled =
  91       (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE;
  92    ctx->VertexProgram.TwoSideEnabled = GL_FALSE;
  93    _mesa_reference_program(ctx, &ctx->VertexProgram.Current,
  94                            ctx->Shared->DefaultVertexProgram);
  95    assert(ctx->VertexProgram.Current);
  96    ctx->VertexProgram.Cache = _mesa_new_program_cache();
  97
  98    ctx->FragmentProgram.Enabled = GL_FALSE;
  99    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current,
 100                            ctx->Shared->DefaultFragmentProgram);
 101    assert(ctx->FragmentProgram.Current);
 102    ctx->FragmentProgram.Cache = _mesa_new_program_cache();
 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_program(ctx, &ctx->VertexProgram.Current, NULL);
 119    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 120    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current, NULL);
 121    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 122
 123    /* XXX probably move this stuff */
 124    if (ctx->ATIFragmentShader.Current) {
 125       ctx->ATIFragmentShader.Current->RefCount--;
 126       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 127          free(ctx->ATIFragmentShader.Current);
 128       }
 129    }
 130
 131    free((void *) ctx->Program.ErrorString);
 132 }
 133
 134
 135 /**
 136  * Update the default program objects in the given context to reference those
 137  * specified in the shared state and release those referencing the old
 138  * shared state.
 139  */
 140 void
 141 _mesa_update_default_objects_program(struct gl_context *ctx)
 142 {
 143    _mesa_reference_program(ctx, &ctx->VertexProgram.Current,
 144                            ctx->Shared->DefaultVertexProgram);
 145    assert(ctx->VertexProgram.Current);
 146
 147    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current,
 148                             ctx->Shared->DefaultFragmentProgram);
 149    assert(ctx->FragmentProgram.Current);
 150
 151    /* XXX probably move this stuff */
 152    if (ctx->ATIFragmentShader.Current) {
 153       ctx->ATIFragmentShader.Current->RefCount--;
 154       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 155          free(ctx->ATIFragmentShader.Current);
 156       }
 157    }
 158    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 159    assert(ctx->ATIFragmentShader.Current);
 160    ctx->ATIFragmentShader.Current->RefCount++;
 161 }
 162
 163
 164 /**
 165  * Set the vertex/fragment program error state (position and error string).
 166  * This is generally called from within the parsers.
 167  */
 168 void
 169 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 170 {
 171    ctx->Program.ErrorPos = pos;
 172    free((void *) ctx->Program.ErrorString);
 173    if (!string)
 174       string = "";
 175    ctx->Program.ErrorString = strdup(string);
 176 }
 177
 178
 179 /**
 180  * Initialize a new gl_program object.
 181  */
 182 struct gl_program *
 183 _mesa_init_gl_program(struct gl_program *prog, GLenum target, GLuint id,
 184                       bool is_arb_asm)
 185 {
 186    if (!prog)
 187       return NULL;
 188
 189    memset(prog, 0, sizeof(*prog));
 190    prog->Id = id;
 191    prog->Target = target;
 192    prog->RefCount = 1;
 193    prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 194    prog->info.stage = _mesa_program_enum_to_shader_stage(target);
 195    prog->is_arb_asm = is_arb_asm;
 196
 197    /* Uniforms that lack an initializer in the shader code have an initial
 198     * value of zero.  This includes sampler uniforms.
 199     *
 200     * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says:
 201     *
 202     *     "The link time initial value is either the value of the variable's
 203     *     initializer, if present, or 0 if no initializer is present. Sampler
 204     *     types cannot have initializers."
 205     *
 206     * So we only initialise ARB assembly style programs.
 207     */
 208    if (is_arb_asm) {
 209       /* default mapping from samplers to texture units */
 210       for (unsigned i = 0; i < MAX_SAMPLERS; i++)
 211          prog->SamplerUnits[i] = i;
 212    }
 213
 214    return prog;
 215 }
 216
 217
 218 /**
 219  * Allocate and initialize a new fragment/vertex program object but
 220  * don't put it into the program hash table.  Called via
 221  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 222  * device driver function to implement OO deriviation with additional
 223  * types not understood by this function.
 224  *
 225  * \param ctx  context
 226  * \param id   program id/number
 227  * \param target  program target/type
 228  * \return  pointer to new program object
 229  */
 230 struct gl_program *
 231 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id,
 232                   bool is_arb_asm)
 233 {
 234    switch (target) {
 235    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 236    case GL_GEOMETRY_PROGRAM_NV:
 237    case GL_TESS_CONTROL_PROGRAM_NV:
 238    case GL_TESS_EVALUATION_PROGRAM_NV:
 239    case GL_FRAGMENT_PROGRAM_ARB:
 240    case GL_COMPUTE_PROGRAM_NV: {
 241       struct gl_program *prog = rzalloc(NULL, struct gl_program);
 242       return _mesa_init_gl_program(prog, target, id, is_arb_asm);
 243    }
 244    default:
 245       _mesa_problem(ctx, "bad target in _mesa_new_program");
 246       return NULL;
 247    }
 248 }
 249
 250
 251 /**
 252  * Delete a program and remove it from the hash table, ignoring the
 253  * reference count.
 254  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 255  * by a device driver function.
 256  */
 257 void
 258 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 259 {
 260    (void) ctx;
 261    assert(prog);
 262    assert(prog->RefCount==0);
 263
 264    if (prog == &_mesa_DummyProgram)
 265       return;
 266
 267    if (prog->Parameters) {
 268       _mesa_free_parameter_list(prog->Parameters);
 269    }
 270
 271    if (prog->nir) {
 272       ralloc_free(prog->nir);
 273    }
 274
 275    if (prog->sh.BindlessSamplers) {
 276       ralloc_free(prog->sh.BindlessSamplers);
 277    }
 278
 279    if (prog->sh.BindlessImages) {
 280       ralloc_free(prog->sh.BindlessImages);
 281    }
 282
 283    if (prog->driver_cache_blob) {
 284       ralloc_free(prog->driver_cache_blob);
 285    }
 286
 287    ralloc_free(prog);
 288 }
 289
 290
 291 /**
 292  * Return the gl_program object for a given ID.
 293  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 294  * casts elsewhere.
 295  */
 296 struct gl_program *
 297 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 298 {
 299    if (id)
 300       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 301    else
 302       return NULL;
 303 }
 304
 305
 306 /**
 307  * Reference counting for vertex/fragment programs
 308  * This is normally only called from the _mesa_reference_program() macro
 309  * when there's a real pointer change.
 310  */
 311 void
 312 _mesa_reference_program_(struct gl_context *ctx,
 313                          struct gl_program **ptr,
 314                          struct gl_program *prog)
 315 {
 316 #ifndef NDEBUG
 317    assert(ptr);
 318    if (*ptr && prog) {
 319       /* sanity check */
 320       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 321          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 322       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 323          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 324                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 325       else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
 326          assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
 327    }
 328 #endif
 329
 330    if (*ptr) {
 331       struct gl_program *oldProg = *ptr;
 332
 333       assert(oldProg->RefCount > 0);
 334
 335       if (p_atomic_dec_zero(&oldProg->RefCount)) {
 336          assert(ctx);
 337          _mesa_reference_shader_program_data(ctx, &oldProg->sh.data, NULL);
 338          ctx->Driver.DeleteProgram(ctx, oldProg);
 339       }
 340
 341       *ptr = NULL;
 342    }
 343
 344    assert(!*ptr);
 345    if (prog) {
 346       p_atomic_inc(&prog->RefCount);
 347    }
 348
 349    *ptr = prog;
 350 }
 351
 352
 353 /**
 354  * Insert 'count' NOP instructions at 'start' in the given program.
 355  * Adjust branch targets accordingly.
 356  */
 357 GLboolean
 358 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 359 {
 360    const GLuint origLen = prog->arb.NumInstructions;
 361    const GLuint newLen = origLen + count;
 362    struct prog_instruction *newInst;
 363    GLuint i;
 364
 365    /* adjust branches */
 366    for (i = 0; i < prog->arb.NumInstructions; i++) {
 367       struct prog_instruction *inst = prog->arb.Instructions + i;
 368       if (inst->BranchTarget > 0) {
 369          if ((GLuint)inst->BranchTarget >= start) {
 370             inst->BranchTarget += count;
 371          }
 372       }
 373    }
 374
 375    /* Alloc storage for new instructions */
 376    newInst = rzalloc_array(prog, struct prog_instruction, newLen);
 377    if (!newInst) {
 378       return GL_FALSE;
 379    }
 380
 381    /* Copy 'start' instructions into new instruction buffer */
 382    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 383
 384    /* init the new instructions */
 385    _mesa_init_instructions(newInst + start, count);
 386
 387    /* Copy the remaining/tail instructions to new inst buffer */
 388    _mesa_copy_instructions(newInst + start + count,
 389                            prog->arb.Instructions + start,
 390                            origLen - start);
 391
 392    /* free old instructions */
 393    ralloc_free(prog->arb.Instructions);
 394
 395    /* install new instructions */
 396    prog->arb.Instructions = newInst;
 397    prog->arb.NumInstructions = newLen;
 398
 399    return GL_TRUE;
 400 }
 401
 402 /**
 403  * Delete 'count' instructions at 'start' in the given program.
 404  * Adjust branch targets accordingly.
 405  */
 406 GLboolean
 407 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count,
 408                           void *mem_ctx)
 409 {
 410    const GLuint origLen = prog->arb.NumInstructions;
 411    const GLuint newLen = origLen - count;
 412    struct prog_instruction *newInst;
 413    GLuint i;
 414
 415    /* adjust branches */
 416    for (i = 0; i < prog->arb.NumInstructions; i++) {
 417       struct prog_instruction *inst = prog->arb.Instructions + i;
 418       if (inst->BranchTarget > 0) {
 419          if (inst->BranchTarget > (GLint) start) {
 420             inst->BranchTarget -= count;
 421          }
 422       }
 423    }
 424
 425    /* Alloc storage for new instructions */
 426    newInst = rzalloc_array(mem_ctx, struct prog_instruction, newLen);
 427    if (!newInst) {
 428       return GL_FALSE;
 429    }
 430
 431    /* Copy 'start' instructions into new instruction buffer */
 432    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 433
 434    /* Copy the remaining/tail instructions to new inst buffer */
 435    _mesa_copy_instructions(newInst + start,
 436                            prog->arb.Instructions + start + count,
 437                            newLen - start);
 438
 439    /* free old instructions */
 440    ralloc_free(prog->arb.Instructions);
 441
 442    /* install new instructions */
 443    prog->arb.Instructions = newInst;
 444    prog->arb.NumInstructions = newLen;
 445
 446    return GL_TRUE;
 447 }
 448
 449
 450 /**
 451  * Populate the 'used' array with flags indicating which registers (TEMPs,
 452  * INPUTs, OUTPUTs, etc, are used by the given program.
 453  * \param file  type of register to scan for
 454  * \param used  returns true/false flags for in use / free
 455  * \param usedSize  size of the 'used' array
 456  */
 457 void
 458 _mesa_find_used_registers(const struct gl_program *prog,
 459                           gl_register_file file,
 460                           GLboolean used[], GLuint usedSize)
 461 {
 462    GLuint i, j;
 463
 464    memset(used, 0, usedSize);
 465
 466    for (i = 0; i < prog->arb.NumInstructions; i++) {
 467       const struct prog_instruction *inst = prog->arb.Instructions + i;
 468       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 469
 470       if (inst->DstReg.File == file) {
 471          assert(inst->DstReg.Index < usedSize);
 472          if(inst->DstReg.Index < usedSize)
 473             used[inst->DstReg.Index] = GL_TRUE;
 474       }
 475
 476       for (j = 0; j < n; j++) {
 477          if (inst->SrcReg[j].File == file) {
 478             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 479             if (inst->SrcReg[j].Index < (GLint) usedSize)
 480                used[inst->SrcReg[j].Index] = GL_TRUE;
 481          }
 482       }
 483    }
 484 }
 485
 486
 487 /**
 488  * Scan the given 'used' register flag array for the first entry
 489  * that's >= firstReg.
 490  * \param used  vector of flags indicating registers in use (as returned
 491  *              by _mesa_find_used_registers())
 492  * \param usedSize  size of the 'used' array
 493  * \param firstReg  first register to start searching at
 494  * \return index of unused register, or -1 if none.
 495  */
 496 GLint
 497 _mesa_find_free_register(const GLboolean used[],
 498                          GLuint usedSize, GLuint firstReg)
 499 {
 500    GLuint i;
 501
 502    assert(firstReg < usedSize);
 503
 504    for (i = firstReg; i < usedSize; i++)
 505       if (!used[i])
 506          return i;
 507
 508    return -1;
 509 }
 510
 511
 512 /* Gets the minimum number of shader invocations per fragment.
 513  * This function is useful to determine if we need to do per
 514  * sample shading or per fragment shading.
 515  */
 516 GLint
 517 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
 518                                        const struct gl_program *prog,
 519                                        bool ignore_sample_qualifier)
 520 {
 521    /* From ARB_sample_shading specification:
 522     * "Using gl_SampleID in a fragment shader causes the entire shader
 523     *  to be evaluated per-sample."
 524     *
 525     * "Using gl_SamplePosition in a fragment shader causes the entire
 526     *  shader to be evaluated per-sample."
 527     *
 528     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
 529     *  has no effect."
 530     */
 531    if (ctx->Multisample.Enabled) {
 532       /* The ARB_gpu_shader5 specification says:
 533        *
 534        * "Use of the "sample" qualifier on a fragment shader input
 535        *  forces per-sample shading"
 536        */
 537       if (prog->info.fs.uses_sample_qualifier && !ignore_sample_qualifier)
 538          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 539
 540       if (prog->info.system_values_read & (SYSTEM_BIT_SAMPLE_ID |
 541                                            SYSTEM_BIT_SAMPLE_POS))
 542          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 543       else if (ctx->Multisample.SampleShading)
 544          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
 545                           _mesa_geometric_samples(ctx->DrawBuffer)), 1);
 546       else
 547          return 1;
 548    }
 549    return 1;
 550 }
 551
 552
 553 GLbitfield
 554 gl_external_samplers(const struct gl_program *prog)
 555 {
 556    GLbitfield external_samplers = 0;
 557    GLbitfield mask = prog->SamplersUsed;
 558
 559    while (mask) {
 560       int idx = u_bit_scan(&mask);
 561       if (prog->sh.SamplerTargets[idx] == TEXTURE_EXTERNAL_INDEX)
 562          external_samplers |= (1 << idx);
 563    }
 564
 565    return external_samplers;
 566 }