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/ralloc.h"
  43 #include "util/u_atomic.h"
  44
  45
  46 /**
  47  * A pointer to this dummy program is put into the hash table when
  48  * glGenPrograms is called.
  49  */
  50 struct gl_program _mesa_DummyProgram;
  51
  52
  53 /**
  54  * Init context's vertex/fragment program state
  55  */
  56 void
  57 _mesa_init_program(struct gl_context *ctx)
  58 {
  59    /*
  60     * If this assertion fails, we need to increase the field
  61     * size for register indexes (see INST_INDEX_BITS).
  62     */
  63    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents / 4
  64           <= (1 << INST_INDEX_BITS));
  65    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents / 4
  66           <= (1 << INST_INDEX_BITS));
  67
  68    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxTemps <= (1 << INST_INDEX_BITS));
  69    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxLocalParams <= (1 << INST_INDEX_BITS));
  70    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTemps <= (1 << INST_INDEX_BITS));
  71    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxLocalParams <= (1 << INST_INDEX_BITS));
  72
  73    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  74    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxUniformComponents <= 4 * MAX_UNIFORMS);
  75
  76    assert(ctx->Const.Program[MESA_SHADER_VERTEX].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  77    assert(ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxAddressOffset <= (1 << INST_INDEX_BITS));
  78
  79    /* If this fails, increase prog_instruction::TexSrcUnit size */
  80    STATIC_ASSERT(MAX_TEXTURE_UNITS <= (1 << 5));
  81
  82    /* If this fails, increase prog_instruction::TexSrcTarget size */
  83    STATIC_ASSERT(NUM_TEXTURE_TARGETS <= (1 << 4));
  84
  85    ctx->Program.ErrorPos = -1;
  86    ctx->Program.ErrorString = strdup("");
  87
  88    ctx->VertexProgram.Enabled = GL_FALSE;
  89    ctx->VertexProgram.PointSizeEnabled =
  90       (ctx->API == API_OPENGLES2) ? GL_TRUE : GL_FALSE;
  91    ctx->VertexProgram.TwoSideEnabled = GL_FALSE;
  92    _mesa_reference_program(ctx, &ctx->VertexProgram.Current,
  93                            ctx->Shared->DefaultVertexProgram);
  94    assert(ctx->VertexProgram.Current);
  95    ctx->VertexProgram.Cache = _mesa_new_program_cache();
  96
  97    ctx->FragmentProgram.Enabled = GL_FALSE;
  98    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current,
  99                            ctx->Shared->DefaultFragmentProgram);
 100    assert(ctx->FragmentProgram.Current);
 101    ctx->FragmentProgram.Cache = _mesa_new_program_cache();
 102
 103    /* XXX probably move this stuff */
 104    ctx->ATIFragmentShader.Enabled = GL_FALSE;
 105    ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
 106    assert(ctx->ATIFragmentShader.Current);
 107    ctx->ATIFragmentShader.Current->RefCount++;
 108 }
 109
 110
 111 /**
 112  * Free a context's vertex/fragment program state
 113  */
 114 void
 115 _mesa_free_program_data(struct gl_context *ctx)
 116 {
 117    _mesa_reference_program(ctx, &ctx->VertexProgram.Current, NULL);
 118    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 119    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current, NULL);
 120    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 121
 122    /* XXX probably move this stuff */
 123    if (ctx->ATIFragmentShader.Current) {
 124       ctx->ATIFragmentShader.Current->RefCount--;
 125       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 126          free(ctx->ATIFragmentShader.Current);
 127       }
 128    }
 129
 130    free((void *) ctx->Program.ErrorString);
 131 }
 132
 133
 134 /**
 135  * Update the default program objects in the given context to reference those
 136  * specified in the shared state and release those referencing the old
 137  * shared state.
 138  */
 139 void
 140 _mesa_update_default_objects_program(struct gl_context *ctx)
 141 {
 142    _mesa_reference_program(ctx, &ctx->VertexProgram.Current,
 143                            ctx->Shared->DefaultVertexProgram);
 144    assert(ctx->VertexProgram.Current);
 145
 146    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current,
 147                             ctx->Shared->DefaultFragmentProgram);
 148    assert(ctx->FragmentProgram.Current);
 149
 150    /* XXX probably move this stuff */
 151    if (ctx->ATIFragmentShader.Current) {
 152       ctx->ATIFragmentShader.Current->RefCount--;
 153       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 154          free(ctx->ATIFragmentShader.Current);
 155       }
 156    }
 157    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 158    assert(ctx->ATIFragmentShader.Current);
 159    ctx->ATIFragmentShader.Current->RefCount++;
 160 }
 161
 162
 163 /**
 164  * Set the vertex/fragment program error state (position and error string).
 165  * This is generally called from within the parsers.
 166  */
 167 void
 168 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 169 {
 170    ctx->Program.ErrorPos = pos;
 171    free((void *) ctx->Program.ErrorString);
 172    if (!string)
 173       string = "";
 174    ctx->Program.ErrorString = strdup(string);
 175 }
 176
 177
 178 /**
 179  * Initialize a new gl_program object.
 180  */
 181 struct gl_program *
 182 _mesa_init_gl_program(struct gl_program *prog, GLenum target, GLuint id,
 183                       bool is_arb_asm)
 184 {
 185    if (!prog)
 186       return NULL;
 187
 188    memset(prog, 0, sizeof(*prog));
 189    prog->Id = id;
 190    prog->Target = target;
 191    prog->RefCount = 1;
 192    prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 193    prog->info.stage = _mesa_program_enum_to_shader_stage(target);
 194    prog->is_arb_asm = is_arb_asm;
 195
 196    /* Uniforms that lack an initializer in the shader code have an initial
 197     * value of zero.  This includes sampler uniforms.
 198     *
 199     * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says:
 200     *
 201     *     "The link time initial value is either the value of the variable's
 202     *     initializer, if present, or 0 if no initializer is present. Sampler
 203     *     types cannot have initializers."
 204     *
 205     * So we only initialise ARB assembly style programs.
 206     */
 207    if (is_arb_asm) {
 208       /* default mapping from samplers to texture units */
 209       for (unsigned i = 0; i < MAX_SAMPLERS; i++)
 210          prog->SamplerUnits[i] = i;
 211    }
 212
 213    return prog;
 214 }
 215
 216
 217 /**
 218  * Allocate and initialize a new fragment/vertex program object but
 219  * don't put it into the program hash table.  Called via
 220  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 221  * device driver function to implement OO deriviation with additional
 222  * types not understood by this function.
 223  *
 224  * \param ctx  context
 225  * \param id   program id/number
 226  * \param target  program target/type
 227  * \return  pointer to new program object
 228  */
 229 struct gl_program *
 230 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id,
 231                   bool is_arb_asm)
 232 {
 233    switch (target) {
 234    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 235    case GL_GEOMETRY_PROGRAM_NV:
 236    case GL_TESS_CONTROL_PROGRAM_NV:
 237    case GL_TESS_EVALUATION_PROGRAM_NV:
 238    case GL_FRAGMENT_PROGRAM_ARB:
 239    case GL_COMPUTE_PROGRAM_NV: {
 240       struct gl_program *prog = rzalloc(NULL, struct gl_program);
 241       return _mesa_init_gl_program(prog, target, id, is_arb_asm);
 242    }
 243    default:
 244       _mesa_problem(ctx, "bad target in _mesa_new_program");
 245       return NULL;
 246    }
 247 }
 248
 249
 250 /**
 251  * Delete a program and remove it from the hash table, ignoring the
 252  * reference count.
 253  * Called via ctx->Driver.DeleteProgram.  May be wrapped (OO deriviation)
 254  * by a device driver function.
 255  */
 256 void
 257 _mesa_delete_program(struct gl_context *ctx, struct gl_program *prog)
 258 {
 259    (void) ctx;
 260    assert(prog);
 261    assert(prog->RefCount==0);
 262
 263    if (prog == &_mesa_DummyProgram)
 264       return;
 265
 266    if (prog->Parameters) {
 267       _mesa_free_parameter_list(prog->Parameters);
 268    }
 269
 270    if (prog->nir) {
 271       ralloc_free(prog->nir);
 272    }
 273
 274    if (prog->sh.BindlessSamplers) {
 275       ralloc_free(prog->sh.BindlessSamplers);
 276    }
 277
 278    if (prog->sh.BindlessImages) {
 279       ralloc_free(prog->sh.BindlessImages);
 280    }
 281
 282    if (prog->driver_cache_blob) {
 283       ralloc_free(prog->driver_cache_blob);
 284    }
 285
 286    ralloc_free(prog);
 287 }
 288
 289
 290 /**
 291  * Return the gl_program object for a given ID.
 292  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 293  * casts elsewhere.
 294  */
 295 struct gl_program *
 296 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 297 {
 298    if (id)
 299       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 300    else
 301       return NULL;
 302 }
 303
 304
 305 /**
 306  * Reference counting for vertex/fragment programs
 307  * This is normally only called from the _mesa_reference_program() macro
 308  * when there's a real pointer change.
 309  */
 310 void
 311 _mesa_reference_program_(struct gl_context *ctx,
 312                          struct gl_program **ptr,
 313                          struct gl_program *prog)
 314 {
 315 #ifndef NDEBUG
 316    assert(ptr);
 317    if (*ptr && prog) {
 318       /* sanity check */
 319       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 320          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 321       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 322          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 323                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 324       else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
 325          assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
 326    }
 327 #endif
 328
 329    if (*ptr) {
 330       struct gl_program *oldProg = *ptr;
 331
 332       assert(oldProg->RefCount > 0);
 333
 334       if (p_atomic_dec_zero(&oldProg->RefCount)) {
 335          assert(ctx);
 336          _mesa_reference_shader_program_data(ctx, &oldProg->sh.data, NULL);
 337          ctx->Driver.DeleteProgram(ctx, oldProg);
 338       }
 339
 340       *ptr = NULL;
 341    }
 342
 343    assert(!*ptr);
 344    if (prog) {
 345       p_atomic_inc(&prog->RefCount);
 346    }
 347
 348    *ptr = prog;
 349 }
 350
 351
 352 /**
 353  * Insert 'count' NOP instructions at 'start' in the given program.
 354  * Adjust branch targets accordingly.
 355  */
 356 GLboolean
 357 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 358 {
 359    const GLuint origLen = prog->arb.NumInstructions;
 360    const GLuint newLen = origLen + count;
 361    struct prog_instruction *newInst;
 362    GLuint i;
 363
 364    /* adjust branches */
 365    for (i = 0; i < prog->arb.NumInstructions; i++) {
 366       struct prog_instruction *inst = prog->arb.Instructions + i;
 367       if (inst->BranchTarget > 0) {
 368          if ((GLuint)inst->BranchTarget >= start) {
 369             inst->BranchTarget += count;
 370          }
 371       }
 372    }
 373
 374    /* Alloc storage for new instructions */
 375    newInst = rzalloc_array(prog, struct prog_instruction, newLen);
 376    if (!newInst) {
 377       return GL_FALSE;
 378    }
 379
 380    /* Copy 'start' instructions into new instruction buffer */
 381    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 382
 383    /* init the new instructions */
 384    _mesa_init_instructions(newInst + start, count);
 385
 386    /* Copy the remaining/tail instructions to new inst buffer */
 387    _mesa_copy_instructions(newInst + start + count,
 388                            prog->arb.Instructions + start,
 389                            origLen - start);
 390
 391    /* free old instructions */
 392    ralloc_free(prog->arb.Instructions);
 393
 394    /* install new instructions */
 395    prog->arb.Instructions = newInst;
 396    prog->arb.NumInstructions = newLen;
 397
 398    return GL_TRUE;
 399 }
 400
 401 /**
 402  * Delete 'count' instructions at 'start' in the given program.
 403  * Adjust branch targets accordingly.
 404  */
 405 GLboolean
 406 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count,
 407                           void *mem_ctx)
 408 {
 409    const GLuint origLen = prog->arb.NumInstructions;
 410    const GLuint newLen = origLen - count;
 411    struct prog_instruction *newInst;
 412    GLuint i;
 413
 414    /* adjust branches */
 415    for (i = 0; i < prog->arb.NumInstructions; i++) {
 416       struct prog_instruction *inst = prog->arb.Instructions + i;
 417       if (inst->BranchTarget > 0) {
 418          if (inst->BranchTarget > (GLint) start) {
 419             inst->BranchTarget -= count;
 420          }
 421       }
 422    }
 423
 424    /* Alloc storage for new instructions */
 425    newInst = rzalloc_array(mem_ctx, struct prog_instruction, newLen);
 426    if (!newInst) {
 427       return GL_FALSE;
 428    }
 429
 430    /* Copy 'start' instructions into new instruction buffer */
 431    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 432
 433    /* Copy the remaining/tail instructions to new inst buffer */
 434    _mesa_copy_instructions(newInst + start,
 435                            prog->arb.Instructions + start + count,
 436                            newLen - start);
 437
 438    /* free old instructions */
 439    ralloc_free(prog->arb.Instructions);
 440
 441    /* install new instructions */
 442    prog->arb.Instructions = newInst;
 443    prog->arb.NumInstructions = newLen;
 444
 445    return GL_TRUE;
 446 }
 447
 448
 449 /**
 450  * Populate the 'used' array with flags indicating which registers (TEMPs,
 451  * INPUTs, OUTPUTs, etc, are used by the given program.
 452  * \param file  type of register to scan for
 453  * \param used  returns true/false flags for in use / free
 454  * \param usedSize  size of the 'used' array
 455  */
 456 void
 457 _mesa_find_used_registers(const struct gl_program *prog,
 458                           gl_register_file file,
 459                           GLboolean used[], GLuint usedSize)
 460 {
 461    GLuint i, j;
 462
 463    memset(used, 0, usedSize);
 464
 465    for (i = 0; i < prog->arb.NumInstructions; i++) {
 466       const struct prog_instruction *inst = prog->arb.Instructions + i;
 467       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 468
 469       if (inst->DstReg.File == file) {
 470          assert(inst->DstReg.Index < usedSize);
 471          if(inst->DstReg.Index < usedSize)
 472             used[inst->DstReg.Index] = GL_TRUE;
 473       }
 474
 475       for (j = 0; j < n; j++) {
 476          if (inst->SrcReg[j].File == file) {
 477             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 478             if (inst->SrcReg[j].Index < (GLint) usedSize)
 479                used[inst->SrcReg[j].Index] = GL_TRUE;
 480          }
 481       }
 482    }
 483 }
 484
 485
 486 /**
 487  * Scan the given 'used' register flag array for the first entry
 488  * that's >= firstReg.
 489  * \param used  vector of flags indicating registers in use (as returned
 490  *              by _mesa_find_used_registers())
 491  * \param usedSize  size of the 'used' array
 492  * \param firstReg  first register to start searching at
 493  * \return index of unused register, or -1 if none.
 494  */
 495 GLint
 496 _mesa_find_free_register(const GLboolean used[],
 497                          GLuint usedSize, GLuint firstReg)
 498 {
 499    GLuint i;
 500
 501    assert(firstReg < usedSize);
 502
 503    for (i = firstReg; i < usedSize; i++)
 504       if (!used[i])
 505          return i;
 506
 507    return -1;
 508 }
 509
 510
 511 /* Gets the minimum number of shader invocations per fragment.
 512  * This function is useful to determine if we need to do per
 513  * sample shading or per fragment shading.
 514  */
 515 GLint
 516 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
 517                                        const struct gl_program *prog,
 518                                        bool ignore_sample_qualifier)
 519 {
 520    /* From ARB_sample_shading specification:
 521     * "Using gl_SampleID in a fragment shader causes the entire shader
 522     *  to be evaluated per-sample."
 523     *
 524     * "Using gl_SamplePosition in a fragment shader causes the entire
 525     *  shader to be evaluated per-sample."
 526     *
 527     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
 528     *  has no effect."
 529     */
 530    if (ctx->Multisample.Enabled) {
 531       /* The ARB_gpu_shader5 specification says:
 532        *
 533        * "Use of the "sample" qualifier on a fragment shader input
 534        *  forces per-sample shading"
 535        */
 536       if (prog->info.fs.uses_sample_qualifier && !ignore_sample_qualifier)
 537          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 538
 539       if (prog->info.system_values_read & (SYSTEM_BIT_SAMPLE_ID |
 540                                            SYSTEM_BIT_SAMPLE_POS))
 541          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 542       else if (ctx->Multisample.SampleShading)
 543          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
 544                           _mesa_geometric_samples(ctx->DrawBuffer)), 1);
 545       else
 546          return 1;
 547    }
 548    return 1;
 549 }