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    ctx->VertexProgram._VPMode = VP_MODE_FF;
 104
 105    /* XXX probably move this stuff */
 106    ctx->ATIFragmentShader.Enabled = GL_FALSE;
 107    ctx->ATIFragmentShader.Current = ctx->Shared->DefaultFragmentShader;
 108    assert(ctx->ATIFragmentShader.Current);
 109    ctx->ATIFragmentShader.Current->RefCount++;
 110 }
 111
 112
 113 /**
 114  * Free a context's vertex/fragment program state
 115  */
 116 void
 117 _mesa_free_program_data(struct gl_context *ctx)
 118 {
 119    _mesa_reference_program(ctx, &ctx->VertexProgram.Current, NULL);
 120    _mesa_delete_program_cache(ctx, ctx->VertexProgram.Cache);
 121    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current, NULL);
 122    _mesa_delete_shader_cache(ctx, ctx->FragmentProgram.Cache);
 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_program(ctx, &ctx->VertexProgram.Current,
 145                            ctx->Shared->DefaultVertexProgram);
 146    assert(ctx->VertexProgram.Current);
 147
 148    _mesa_reference_program(ctx, &ctx->FragmentProgram.Current,
 149                             ctx->Shared->DefaultFragmentProgram);
 150    assert(ctx->FragmentProgram.Current);
 151
 152    /* XXX probably move this stuff */
 153    if (ctx->ATIFragmentShader.Current) {
 154       ctx->ATIFragmentShader.Current->RefCount--;
 155       if (ctx->ATIFragmentShader.Current->RefCount <= 0) {
 156          free(ctx->ATIFragmentShader.Current);
 157       }
 158    }
 159    ctx->ATIFragmentShader.Current = (struct ati_fragment_shader *) ctx->Shared->DefaultFragmentShader;
 160    assert(ctx->ATIFragmentShader.Current);
 161    ctx->ATIFragmentShader.Current->RefCount++;
 162 }
 163
 164
 165 /**
 166  * Set the vertex/fragment program error state (position and error string).
 167  * This is generally called from within the parsers.
 168  */
 169 void
 170 _mesa_set_program_error(struct gl_context *ctx, GLint pos, const char *string)
 171 {
 172    ctx->Program.ErrorPos = pos;
 173    free((void *) ctx->Program.ErrorString);
 174    if (!string)
 175       string = "";
 176    ctx->Program.ErrorString = strdup(string);
 177 }
 178
 179
 180 /**
 181  * Initialize a new gl_program object.
 182  */
 183 struct gl_program *
 184 _mesa_init_gl_program(struct gl_program *prog, GLenum target, GLuint id,
 185                       bool is_arb_asm)
 186 {
 187    if (!prog)
 188       return NULL;
 189
 190    memset(prog, 0, sizeof(*prog));
 191    prog->Id = id;
 192    prog->Target = target;
 193    prog->RefCount = 1;
 194    prog->Format = GL_PROGRAM_FORMAT_ASCII_ARB;
 195    prog->info.stage = _mesa_program_enum_to_shader_stage(target);
 196    prog->is_arb_asm = is_arb_asm;
 197
 198    /* Uniforms that lack an initializer in the shader code have an initial
 199     * value of zero.  This includes sampler uniforms.
 200     *
 201     * Page 24 (page 30 of the PDF) of the GLSL 1.20 spec says:
 202     *
 203     *     "The link time initial value is either the value of the variable's
 204     *     initializer, if present, or 0 if no initializer is present. Sampler
 205     *     types cannot have initializers."
 206     *
 207     * So we only initialise ARB assembly style programs.
 208     */
 209    if (is_arb_asm) {
 210       /* default mapping from samplers to texture units */
 211       for (unsigned i = 0; i < MAX_SAMPLERS; i++)
 212          prog->SamplerUnits[i] = i;
 213    }
 214
 215    return prog;
 216 }
 217
 218
 219 /**
 220  * Allocate and initialize a new fragment/vertex program object but
 221  * don't put it into the program hash table.  Called via
 222  * ctx->Driver.NewProgram.  May be overridden (ie. replaced) by a
 223  * device driver function to implement OO deriviation with additional
 224  * types not understood by this function.
 225  *
 226  * \param ctx  context
 227  * \param id   program id/number
 228  * \param target  program target/type
 229  * \return  pointer to new program object
 230  */
 231 struct gl_program *
 232 _mesa_new_program(struct gl_context *ctx, GLenum target, GLuint id,
 233                   bool is_arb_asm)
 234 {
 235    switch (target) {
 236    case GL_VERTEX_PROGRAM_ARB: /* == GL_VERTEX_PROGRAM_NV */
 237    case GL_GEOMETRY_PROGRAM_NV:
 238    case GL_TESS_CONTROL_PROGRAM_NV:
 239    case GL_TESS_EVALUATION_PROGRAM_NV:
 240    case GL_FRAGMENT_PROGRAM_ARB:
 241    case GL_COMPUTE_PROGRAM_NV: {
 242       struct gl_program *prog = rzalloc(NULL, struct gl_program);
 243       return _mesa_init_gl_program(prog, target, id, is_arb_asm);
 244    }
 245    default:
 246       _mesa_problem(ctx, "bad target in _mesa_new_program");
 247       return NULL;
 248    }
 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    if (prog->Parameters) {
 269       _mesa_free_parameter_list(prog->Parameters);
 270    }
 271
 272    if (prog->nir) {
 273       ralloc_free(prog->nir);
 274    }
 275
 276    if (prog->sh.BindlessSamplers) {
 277       ralloc_free(prog->sh.BindlessSamplers);
 278    }
 279
 280    if (prog->sh.BindlessImages) {
 281       ralloc_free(prog->sh.BindlessImages);
 282    }
 283
 284    if (prog->driver_cache_blob) {
 285       ralloc_free(prog->driver_cache_blob);
 286    }
 287
 288    ralloc_free(prog);
 289 }
 290
 291
 292 /**
 293  * Return the gl_program object for a given ID.
 294  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 295  * casts elsewhere.
 296  */
 297 struct gl_program *
 298 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 299 {
 300    if (id)
 301       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 302    else
 303       return NULL;
 304 }
 305
 306
 307 /**
 308  * Reference counting for vertex/fragment programs
 309  * This is normally only called from the _mesa_reference_program() macro
 310  * when there's a real pointer change.
 311  */
 312 void
 313 _mesa_reference_program_(struct gl_context *ctx,
 314                          struct gl_program **ptr,
 315                          struct gl_program *prog)
 316 {
 317 #ifndef NDEBUG
 318    assert(ptr);
 319    if (*ptr && prog) {
 320       /* sanity check */
 321       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 322          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 323       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 324          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 325                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 326       else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
 327          assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
 328    }
 329 #endif
 330
 331    if (*ptr) {
 332       struct gl_program *oldProg = *ptr;
 333
 334       assert(oldProg->RefCount > 0);
 335
 336       if (p_atomic_dec_zero(&oldProg->RefCount)) {
 337          assert(ctx);
 338          _mesa_reference_shader_program_data(ctx, &oldProg->sh.data, NULL);
 339          ctx->Driver.DeleteProgram(ctx, oldProg);
 340       }
 341
 342       *ptr = NULL;
 343    }
 344
 345    assert(!*ptr);
 346    if (prog) {
 347       p_atomic_inc(&prog->RefCount);
 348    }
 349
 350    *ptr = prog;
 351 }
 352
 353
 354 /**
 355  * Insert 'count' NOP instructions at 'start' in the given program.
 356  * Adjust branch targets accordingly.
 357  */
 358 GLboolean
 359 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 360 {
 361    const GLuint origLen = prog->arb.NumInstructions;
 362    const GLuint newLen = origLen + count;
 363    struct prog_instruction *newInst;
 364    GLuint i;
 365
 366    /* adjust branches */
 367    for (i = 0; i < prog->arb.NumInstructions; i++) {
 368       struct prog_instruction *inst = prog->arb.Instructions + i;
 369       if (inst->BranchTarget > 0) {
 370          if ((GLuint)inst->BranchTarget >= start) {
 371             inst->BranchTarget += count;
 372          }
 373       }
 374    }
 375
 376    /* Alloc storage for new instructions */
 377    newInst = rzalloc_array(prog, struct prog_instruction, newLen);
 378    if (!newInst) {
 379       return GL_FALSE;
 380    }
 381
 382    /* Copy 'start' instructions into new instruction buffer */
 383    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 384
 385    /* init the new instructions */
 386    _mesa_init_instructions(newInst + start, count);
 387
 388    /* Copy the remaining/tail instructions to new inst buffer */
 389    _mesa_copy_instructions(newInst + start + count,
 390                            prog->arb.Instructions + start,
 391                            origLen - start);
 392
 393    /* free old instructions */
 394    ralloc_free(prog->arb.Instructions);
 395
 396    /* install new instructions */
 397    prog->arb.Instructions = newInst;
 398    prog->arb.NumInstructions = newLen;
 399
 400    return GL_TRUE;
 401 }
 402
 403 /**
 404  * Delete 'count' instructions at 'start' in the given program.
 405  * Adjust branch targets accordingly.
 406  */
 407 GLboolean
 408 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count,
 409                           void *mem_ctx)
 410 {
 411    const GLuint origLen = prog->arb.NumInstructions;
 412    const GLuint newLen = origLen - count;
 413    struct prog_instruction *newInst;
 414    GLuint i;
 415
 416    /* adjust branches */
 417    for (i = 0; i < prog->arb.NumInstructions; i++) {
 418       struct prog_instruction *inst = prog->arb.Instructions + i;
 419       if (inst->BranchTarget > 0) {
 420          if (inst->BranchTarget > (GLint) start) {
 421             inst->BranchTarget -= count;
 422          }
 423       }
 424    }
 425
 426    /* Alloc storage for new instructions */
 427    newInst = rzalloc_array(mem_ctx, struct prog_instruction, newLen);
 428    if (!newInst) {
 429       return GL_FALSE;
 430    }
 431
 432    /* Copy 'start' instructions into new instruction buffer */
 433    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 434
 435    /* Copy the remaining/tail instructions to new inst buffer */
 436    _mesa_copy_instructions(newInst + start,
 437                            prog->arb.Instructions + start + count,
 438                            newLen - start);
 439
 440    /* free old instructions */
 441    ralloc_free(prog->arb.Instructions);
 442
 443    /* install new instructions */
 444    prog->arb.Instructions = newInst;
 445    prog->arb.NumInstructions = newLen;
 446
 447    return GL_TRUE;
 448 }
 449
 450
 451 /**
 452  * Populate the 'used' array with flags indicating which registers (TEMPs,
 453  * INPUTs, OUTPUTs, etc, are used by the given program.
 454  * \param file  type of register to scan for
 455  * \param used  returns true/false flags for in use / free
 456  * \param usedSize  size of the 'used' array
 457  */
 458 void
 459 _mesa_find_used_registers(const struct gl_program *prog,
 460                           gl_register_file file,
 461                           GLboolean used[], GLuint usedSize)
 462 {
 463    GLuint i, j;
 464
 465    memset(used, 0, usedSize);
 466
 467    for (i = 0; i < prog->arb.NumInstructions; i++) {
 468       const struct prog_instruction *inst = prog->arb.Instructions + i;
 469       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 470
 471       if (inst->DstReg.File == file) {
 472          assert(inst->DstReg.Index < usedSize);
 473          if(inst->DstReg.Index < usedSize)
 474             used[inst->DstReg.Index] = GL_TRUE;
 475       }
 476
 477       for (j = 0; j < n; j++) {
 478          if (inst->SrcReg[j].File == file) {
 479             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 480             if (inst->SrcReg[j].Index < (GLint) usedSize)
 481                used[inst->SrcReg[j].Index] = GL_TRUE;
 482          }
 483       }
 484    }
 485 }
 486
 487
 488 /**
 489  * Scan the given 'used' register flag array for the first entry
 490  * that's >= firstReg.
 491  * \param used  vector of flags indicating registers in use (as returned
 492  *              by _mesa_find_used_registers())
 493  * \param usedSize  size of the 'used' array
 494  * \param firstReg  first register to start searching at
 495  * \return index of unused register, or -1 if none.
 496  */
 497 GLint
 498 _mesa_find_free_register(const GLboolean used[],
 499                          GLuint usedSize, GLuint firstReg)
 500 {
 501    GLuint i;
 502
 503    assert(firstReg < usedSize);
 504
 505    for (i = firstReg; i < usedSize; i++)
 506       if (!used[i])
 507          return i;
 508
 509    return -1;
 510 }
 511
 512
 513 /* Gets the minimum number of shader invocations per fragment.
 514  * This function is useful to determine if we need to do per
 515  * sample shading or per fragment shading.
 516  */
 517 GLint
 518 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
 519                                        const struct gl_program *prog)
 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 ||
 538           (prog->info.system_values_read & (SYSTEM_BIT_SAMPLE_ID |
 539                                             SYSTEM_BIT_SAMPLE_POS)))
 540          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 541       else if (ctx->Multisample.SampleShading)
 542          return MAX2(ceilf(ctx->Multisample.MinSampleShadingValue *
 543                           _mesa_geometric_samples(ctx->DrawBuffer)), 1);
 544       else
 545          return 1;
 546    }
 547    return 1;
 548 }
 549
 550
 551 GLbitfield
 552 gl_external_samplers(const struct gl_program *prog)
 553 {
 554    GLbitfield external_samplers = 0;
 555    GLbitfield mask = prog->SamplersUsed;
 556
 557    while (mask) {
 558       int idx = u_bit_scan(&mask);
 559       if (prog->sh.SamplerTargets[idx] == TEXTURE_EXTERNAL_INDEX)
 560          external_samplers |= (1 << idx);
 561    }
 562
 563    return external_samplers;
 564 }