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    ralloc_free(prog);
 275 }
 276
 277
 278 /**
 279  * Return the gl_program object for a given ID.
 280  * Basically just a wrapper for _mesa_HashLookup() to avoid a lot of
 281  * casts elsewhere.
 282  */
 283 struct gl_program *
 284 _mesa_lookup_program(struct gl_context *ctx, GLuint id)
 285 {
 286    if (id)
 287       return (struct gl_program *) _mesa_HashLookup(ctx->Shared->Programs, id);
 288    else
 289       return NULL;
 290 }
 291
 292
 293 /**
 294  * Reference counting for vertex/fragment programs
 295  * This is normally only called from the _mesa_reference_program() macro
 296  * when there's a real pointer change.
 297  */
 298 void
 299 _mesa_reference_program_(struct gl_context *ctx,
 300                          struct gl_program **ptr,
 301                          struct gl_program *prog)
 302 {
 303 #ifndef NDEBUG
 304    assert(ptr);
 305    if (*ptr && prog) {
 306       /* sanity check */
 307       if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB)
 308          assert(prog->Target == GL_VERTEX_PROGRAM_ARB);
 309       else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB)
 310          assert(prog->Target == GL_FRAGMENT_PROGRAM_ARB ||
 311                 prog->Target == GL_FRAGMENT_PROGRAM_NV);
 312       else if ((*ptr)->Target == GL_GEOMETRY_PROGRAM_NV)
 313          assert(prog->Target == GL_GEOMETRY_PROGRAM_NV);
 314    }
 315 #endif
 316
 317    if (*ptr) {
 318       struct gl_program *oldProg = *ptr;
 319
 320       assert(oldProg->RefCount > 0);
 321
 322       if (p_atomic_dec_zero(&oldProg->RefCount)) {
 323          assert(ctx);
 324          _mesa_reference_shader_program_data(ctx, &oldProg->sh.data, NULL);
 325          ctx->Driver.DeleteProgram(ctx, oldProg);
 326       }
 327
 328       *ptr = NULL;
 329    }
 330
 331    assert(!*ptr);
 332    if (prog) {
 333       p_atomic_inc(&prog->RefCount);
 334    }
 335
 336    *ptr = prog;
 337 }
 338
 339
 340 /**
 341  * Insert 'count' NOP instructions at 'start' in the given program.
 342  * Adjust branch targets accordingly.
 343  */
 344 GLboolean
 345 _mesa_insert_instructions(struct gl_program *prog, GLuint start, GLuint count)
 346 {
 347    const GLuint origLen = prog->arb.NumInstructions;
 348    const GLuint newLen = origLen + count;
 349    struct prog_instruction *newInst;
 350    GLuint i;
 351
 352    /* adjust branches */
 353    for (i = 0; i < prog->arb.NumInstructions; i++) {
 354       struct prog_instruction *inst = prog->arb.Instructions + i;
 355       if (inst->BranchTarget > 0) {
 356          if ((GLuint)inst->BranchTarget >= start) {
 357             inst->BranchTarget += count;
 358          }
 359       }
 360    }
 361
 362    /* Alloc storage for new instructions */
 363    newInst = rzalloc_array(prog, struct prog_instruction, newLen);
 364    if (!newInst) {
 365       return GL_FALSE;
 366    }
 367
 368    /* Copy 'start' instructions into new instruction buffer */
 369    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 370
 371    /* init the new instructions */
 372    _mesa_init_instructions(newInst + start, count);
 373
 374    /* Copy the remaining/tail instructions to new inst buffer */
 375    _mesa_copy_instructions(newInst + start + count,
 376                            prog->arb.Instructions + start,
 377                            origLen - start);
 378
 379    /* free old instructions */
 380    ralloc_free(prog->arb.Instructions);
 381
 382    /* install new instructions */
 383    prog->arb.Instructions = newInst;
 384    prog->arb.NumInstructions = newLen;
 385
 386    return GL_TRUE;
 387 }
 388
 389 /**
 390  * Delete 'count' instructions at 'start' in the given program.
 391  * Adjust branch targets accordingly.
 392  */
 393 GLboolean
 394 _mesa_delete_instructions(struct gl_program *prog, GLuint start, GLuint count,
 395                           void *mem_ctx)
 396 {
 397    const GLuint origLen = prog->arb.NumInstructions;
 398    const GLuint newLen = origLen - count;
 399    struct prog_instruction *newInst;
 400    GLuint i;
 401
 402    /* adjust branches */
 403    for (i = 0; i < prog->arb.NumInstructions; i++) {
 404       struct prog_instruction *inst = prog->arb.Instructions + i;
 405       if (inst->BranchTarget > 0) {
 406          if (inst->BranchTarget > (GLint) start) {
 407             inst->BranchTarget -= count;
 408          }
 409       }
 410    }
 411
 412    /* Alloc storage for new instructions */
 413    newInst = rzalloc_array(mem_ctx, struct prog_instruction, newLen);
 414    if (!newInst) {
 415       return GL_FALSE;
 416    }
 417
 418    /* Copy 'start' instructions into new instruction buffer */
 419    _mesa_copy_instructions(newInst, prog->arb.Instructions, start);
 420
 421    /* Copy the remaining/tail instructions to new inst buffer */
 422    _mesa_copy_instructions(newInst + start,
 423                            prog->arb.Instructions + start + count,
 424                            newLen - start);
 425
 426    /* free old instructions */
 427    ralloc_free(prog->arb.Instructions);
 428
 429    /* install new instructions */
 430    prog->arb.Instructions = newInst;
 431    prog->arb.NumInstructions = newLen;
 432
 433    return GL_TRUE;
 434 }
 435
 436
 437 /**
 438  * Populate the 'used' array with flags indicating which registers (TEMPs,
 439  * INPUTs, OUTPUTs, etc, are used by the given program.
 440  * \param file  type of register to scan for
 441  * \param used  returns true/false flags for in use / free
 442  * \param usedSize  size of the 'used' array
 443  */
 444 void
 445 _mesa_find_used_registers(const struct gl_program *prog,
 446                           gl_register_file file,
 447                           GLboolean used[], GLuint usedSize)
 448 {
 449    GLuint i, j;
 450
 451    memset(used, 0, usedSize);
 452
 453    for (i = 0; i < prog->arb.NumInstructions; i++) {
 454       const struct prog_instruction *inst = prog->arb.Instructions + i;
 455       const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
 456
 457       if (inst->DstReg.File == file) {
 458          assert(inst->DstReg.Index < usedSize);
 459          if(inst->DstReg.Index < usedSize)
 460             used[inst->DstReg.Index] = GL_TRUE;
 461       }
 462
 463       for (j = 0; j < n; j++) {
 464          if (inst->SrcReg[j].File == file) {
 465             assert(inst->SrcReg[j].Index < (GLint) usedSize);
 466             if (inst->SrcReg[j].Index < (GLint) usedSize)
 467                used[inst->SrcReg[j].Index] = GL_TRUE;
 468          }
 469       }
 470    }
 471 }
 472
 473
 474 /**
 475  * Scan the given 'used' register flag array for the first entry
 476  * that's >= firstReg.
 477  * \param used  vector of flags indicating registers in use (as returned
 478  *              by _mesa_find_used_registers())
 479  * \param usedSize  size of the 'used' array
 480  * \param firstReg  first register to start searching at
 481  * \return index of unused register, or -1 if none.
 482  */
 483 GLint
 484 _mesa_find_free_register(const GLboolean used[],
 485                          GLuint usedSize, GLuint firstReg)
 486 {
 487    GLuint i;
 488
 489    assert(firstReg < usedSize);
 490
 491    for (i = firstReg; i < usedSize; i++)
 492       if (!used[i])
 493          return i;
 494
 495    return -1;
 496 }
 497
 498
 499 /* Gets the minimum number of shader invocations per fragment.
 500  * This function is useful to determine if we need to do per
 501  * sample shading or per fragment shading.
 502  */
 503 GLint
 504 _mesa_get_min_invocations_per_fragment(struct gl_context *ctx,
 505                                        const struct gl_program *prog,
 506                                        bool ignore_sample_qualifier)
 507 {
 508    /* From ARB_sample_shading specification:
 509     * "Using gl_SampleID in a fragment shader causes the entire shader
 510     *  to be evaluated per-sample."
 511     *
 512     * "Using gl_SamplePosition in a fragment shader causes the entire
 513     *  shader to be evaluated per-sample."
 514     *
 515     * "If MULTISAMPLE or SAMPLE_SHADING_ARB is disabled, sample shading
 516     *  has no effect."
 517     */
 518    if (ctx->Multisample.Enabled) {
 519       /* The ARB_gpu_shader5 specification says:
 520        *
 521        * "Use of the "sample" qualifier on a fragment shader input
 522        *  forces per-sample shading"
 523        */
 524       if (prog->info.fs.uses_sample_qualifier && !ignore_sample_qualifier)
 525          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 526
 527       if (prog->info.system_values_read & (SYSTEM_BIT_SAMPLE_ID |
 528                                            SYSTEM_BIT_SAMPLE_POS))
 529          return MAX2(_mesa_geometric_samples(ctx->DrawBuffer), 1);
 530       else if (ctx->Multisample.SampleShading)
 531          return MAX2(ceil(ctx->Multisample.MinSampleShadingValue *
 532                           _mesa_geometric_samples(ctx->DrawBuffer)), 1);
 533       else
 534          return 1;
 535    }
 536    return 1;
 537 }