linker.cpp revision ebfad9f6a125738b9bfc5d5f7d09a8b57856674a
1/* 2 * Copyright © 2010 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 21 * DEALINGS IN THE SOFTWARE. 22 */ 23 24/** 25 * \file linker.cpp 26 * GLSL linker implementation 27 * 28 * Given a set of shaders that are to be linked to generate a final program, 29 * there are three distinct stages. 30 * 31 * In the first stage shaders are partitioned into groups based on the shader 32 * type. All shaders of a particular type (e.g., vertex shaders) are linked 33 * together. 34 * 35 * - Undefined references in each shader are resolve to definitions in 36 * another shader. 37 * - Types and qualifiers of uniforms, outputs, and global variables defined 38 * in multiple shaders with the same name are verified to be the same. 39 * - Initializers for uniforms and global variables defined 40 * in multiple shaders with the same name are verified to be the same. 41 * 42 * The result, in the terminology of the GLSL spec, is a set of shader 43 * executables for each processing unit. 44 * 45 * After the first stage is complete, a series of semantic checks are performed 46 * on each of the shader executables. 47 * 48 * - Each shader executable must define a \c main function. 49 * - Each vertex shader executable must write to \c gl_Position. 50 * - Each fragment shader executable must write to either \c gl_FragData or 51 * \c gl_FragColor. 52 * 53 * In the final stage individual shader executables are linked to create a 54 * complete exectuable. 55 * 56 * - Types of uniforms defined in multiple shader stages with the same name 57 * are verified to be the same. 58 * - Initializers for uniforms defined in multiple shader stages with the 59 * same name are verified to be the same. 60 * - Types and qualifiers of outputs defined in one stage are verified to 61 * be the same as the types and qualifiers of inputs defined with the same 62 * name in a later stage. 63 * 64 * \author Ian Romanick <ian.d.romanick@intel.com> 65 */ 66 67#include "main/core.h" 68#include "glsl_symbol_table.h" 69#include "ir.h" 70#include "program.h" 71#include "program/hash_table.h" 72#include "linker.h" 73#include "ir_optimization.h" 74 75extern "C" { 76#include "main/shaderobj.h" 77} 78 79/** 80 * Visitor that determines whether or not a variable is ever written. 81 */ 82class find_assignment_visitor : public ir_hierarchical_visitor { 83public: 84 find_assignment_visitor(const char *name) 85 : name(name), found(false) 86 { 87 /* empty */ 88 } 89 90 virtual ir_visitor_status visit_enter(ir_assignment *ir) 91 { 92 ir_variable *const var = ir->lhs->variable_referenced(); 93 94 if (strcmp(name, var->name) == 0) { 95 found = true; 96 return visit_stop; 97 } 98 99 return visit_continue_with_parent; 100 } 101 102 virtual ir_visitor_status visit_enter(ir_call *ir) 103 { 104 exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator(); 105 foreach_iter(exec_list_iterator, iter, *ir) { 106 ir_rvalue *param_rval = (ir_rvalue *)iter.get(); 107 ir_variable *sig_param = (ir_variable *)sig_iter.get(); 108 109 if (sig_param->mode == ir_var_out || 110 sig_param->mode == ir_var_inout) { 111 ir_variable *var = param_rval->variable_referenced(); 112 if (var && strcmp(name, var->name) == 0) { 113 found = true; 114 return visit_stop; 115 } 116 } 117 sig_iter.next(); 118 } 119 120 return visit_continue_with_parent; 121 } 122 123 bool variable_found() 124 { 125 return found; 126 } 127 128private: 129 const char *name; /**< Find writes to a variable with this name. */ 130 bool found; /**< Was a write to the variable found? */ 131}; 132 133 134/** 135 * Visitor that determines whether or not a variable is ever read. 136 */ 137class find_deref_visitor : public ir_hierarchical_visitor { 138public: 139 find_deref_visitor(const char *name) 140 : name(name), found(false) 141 { 142 /* empty */ 143 } 144 145 virtual ir_visitor_status visit(ir_dereference_variable *ir) 146 { 147 if (strcmp(this->name, ir->var->name) == 0) { 148 this->found = true; 149 return visit_stop; 150 } 151 152 return visit_continue; 153 } 154 155 bool variable_found() const 156 { 157 return this->found; 158 } 159 160private: 161 const char *name; /**< Find writes to a variable with this name. */ 162 bool found; /**< Was a write to the variable found? */ 163}; 164 165 166void 167linker_error(gl_shader_program *prog, const char *fmt, ...) 168{ 169 va_list ap; 170 171 ralloc_strcat(&prog->InfoLog, "error: "); 172 va_start(ap, fmt); 173 ralloc_vasprintf_append(&prog->InfoLog, fmt, ap); 174 va_end(ap); 175 176 prog->LinkStatus = false; 177} 178 179 180void 181linker_warning(gl_shader_program *prog, const char *fmt, ...) 182{ 183 va_list ap; 184 185 ralloc_strcat(&prog->InfoLog, "error: "); 186 va_start(ap, fmt); 187 ralloc_vasprintf_append(&prog->InfoLog, fmt, ap); 188 va_end(ap); 189 190} 191 192 193void 194link_invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode, 195 int generic_base) 196{ 197 foreach_list(node, sh->ir) { 198 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 199 200 if ((var == NULL) || (var->mode != (unsigned) mode)) 201 continue; 202 203 /* Only assign locations for generic attributes / varyings / etc. 204 */ 205 if ((var->location >= generic_base) && !var->explicit_location) 206 var->location = -1; 207 } 208} 209 210 211/** 212 * Determine the number of attribute slots required for a particular type 213 * 214 * This code is here because it implements the language rules of a specific 215 * GLSL version. Since it's a property of the language and not a property of 216 * types in general, it doesn't really belong in glsl_type. 217 */ 218unsigned 219count_attribute_slots(const glsl_type *t) 220{ 221 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec: 222 * 223 * "A scalar input counts the same amount against this limit as a vec4, 224 * so applications may want to consider packing groups of four 225 * unrelated float inputs together into a vector to better utilize the 226 * capabilities of the underlying hardware. A matrix input will use up 227 * multiple locations. The number of locations used will equal the 228 * number of columns in the matrix." 229 * 230 * The spec does not explicitly say how arrays are counted. However, it 231 * should be safe to assume the total number of slots consumed by an array 232 * is the number of entries in the array multiplied by the number of slots 233 * consumed by a single element of the array. 234 */ 235 236 if (t->is_array()) 237 return t->array_size() * count_attribute_slots(t->element_type()); 238 239 if (t->is_matrix()) 240 return t->matrix_columns; 241 242 return 1; 243} 244 245 246/** 247 * Verify that a vertex shader executable meets all semantic requirements. 248 * 249 * Also sets prog->Vert.UsesClipDistance as a side effect. 250 * 251 * \param shader Vertex shader executable to be verified 252 */ 253bool 254validate_vertex_shader_executable(struct gl_shader_program *prog, 255 struct gl_shader *shader) 256{ 257 if (shader == NULL) 258 return true; 259 260 find_assignment_visitor find("gl_Position"); 261 find.run(shader->ir); 262 if (!find.variable_found()) { 263 linker_error(prog, "vertex shader does not write to `gl_Position'\n"); 264 return false; 265 } 266 267 if (prog->Version >= 130) { 268 /* From section 7.1 (Vertex Shader Special Variables) of the 269 * GLSL 1.30 spec: 270 * 271 * "It is an error for a shader to statically write both 272 * gl_ClipVertex and gl_ClipDistance." 273 */ 274 find_assignment_visitor clip_vertex("gl_ClipVertex"); 275 find_assignment_visitor clip_distance("gl_ClipDistance"); 276 277 clip_vertex.run(shader->ir); 278 clip_distance.run(shader->ir); 279 if (clip_vertex.variable_found() && clip_distance.variable_found()) { 280 linker_error(prog, "vertex shader writes to both `gl_ClipVertex' " 281 "and `gl_ClipDistance'\n"); 282 return false; 283 } 284 prog->Vert.UsesClipDistance = clip_distance.variable_found(); 285 } 286 287 return true; 288} 289 290 291/** 292 * Verify that a fragment shader executable meets all semantic requirements 293 * 294 * \param shader Fragment shader executable to be verified 295 */ 296bool 297validate_fragment_shader_executable(struct gl_shader_program *prog, 298 struct gl_shader *shader) 299{ 300 if (shader == NULL) 301 return true; 302 303 find_assignment_visitor frag_color("gl_FragColor"); 304 find_assignment_visitor frag_data("gl_FragData"); 305 306 frag_color.run(shader->ir); 307 frag_data.run(shader->ir); 308 309 if (frag_color.variable_found() && frag_data.variable_found()) { 310 linker_error(prog, "fragment shader writes to both " 311 "`gl_FragColor' and `gl_FragData'\n"); 312 return false; 313 } 314 315 return true; 316} 317 318 319/** 320 * Generate a string describing the mode of a variable 321 */ 322static const char * 323mode_string(const ir_variable *var) 324{ 325 switch (var->mode) { 326 case ir_var_auto: 327 return (var->read_only) ? "global constant" : "global variable"; 328 329 case ir_var_uniform: return "uniform"; 330 case ir_var_in: return "shader input"; 331 case ir_var_out: return "shader output"; 332 case ir_var_inout: return "shader inout"; 333 334 case ir_var_const_in: 335 case ir_var_temporary: 336 default: 337 assert(!"Should not get here."); 338 return "invalid variable"; 339 } 340} 341 342 343/** 344 * Perform validation of global variables used across multiple shaders 345 */ 346bool 347cross_validate_globals(struct gl_shader_program *prog, 348 struct gl_shader **shader_list, 349 unsigned num_shaders, 350 bool uniforms_only) 351{ 352 /* Examine all of the uniforms in all of the shaders and cross validate 353 * them. 354 */ 355 glsl_symbol_table variables; 356 for (unsigned i = 0; i < num_shaders; i++) { 357 if (shader_list[i] == NULL) 358 continue; 359 360 foreach_list(node, shader_list[i]->ir) { 361 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 362 363 if (var == NULL) 364 continue; 365 366 if (uniforms_only && (var->mode != ir_var_uniform)) 367 continue; 368 369 /* Don't cross validate temporaries that are at global scope. These 370 * will eventually get pulled into the shaders 'main'. 371 */ 372 if (var->mode == ir_var_temporary) 373 continue; 374 375 /* If a global with this name has already been seen, verify that the 376 * new instance has the same type. In addition, if the globals have 377 * initializers, the values of the initializers must be the same. 378 */ 379 ir_variable *const existing = variables.get_variable(var->name); 380 if (existing != NULL) { 381 if (var->type != existing->type) { 382 /* Consider the types to be "the same" if both types are arrays 383 * of the same type and one of the arrays is implicitly sized. 384 * In addition, set the type of the linked variable to the 385 * explicitly sized array. 386 */ 387 if (var->type->is_array() 388 && existing->type->is_array() 389 && (var->type->fields.array == existing->type->fields.array) 390 && ((var->type->length == 0) 391 || (existing->type->length == 0))) { 392 if (var->type->length != 0) { 393 existing->type = var->type; 394 } 395 } else { 396 linker_error(prog, "%s `%s' declared as type " 397 "`%s' and type `%s'\n", 398 mode_string(var), 399 var->name, var->type->name, 400 existing->type->name); 401 return false; 402 } 403 } 404 405 if (var->explicit_location) { 406 if (existing->explicit_location 407 && (var->location != existing->location)) { 408 linker_error(prog, "explicit locations for %s " 409 "`%s' have differing values\n", 410 mode_string(var), var->name); 411 return false; 412 } 413 414 existing->location = var->location; 415 existing->explicit_location = true; 416 } 417 418 /* Validate layout qualifiers for gl_FragDepth. 419 * 420 * From the AMD/ARB_conservative_depth specs: 421 * 422 * "If gl_FragDepth is redeclared in any fragment shader in a 423 * program, it must be redeclared in all fragment shaders in 424 * that program that have static assignments to 425 * gl_FragDepth. All redeclarations of gl_FragDepth in all 426 * fragment shaders in a single program must have the same set 427 * of qualifiers." 428 */ 429 if (strcmp(var->name, "gl_FragDepth") == 0) { 430 bool layout_declared = var->depth_layout != ir_depth_layout_none; 431 bool layout_differs = 432 var->depth_layout != existing->depth_layout; 433 434 if (layout_declared && layout_differs) { 435 linker_error(prog, 436 "All redeclarations of gl_FragDepth in all " 437 "fragment shaders in a single program must have " 438 "the same set of qualifiers."); 439 } 440 441 if (var->used && layout_differs) { 442 linker_error(prog, 443 "If gl_FragDepth is redeclared with a layout " 444 "qualifier in any fragment shader, it must be " 445 "redeclared with the same layout qualifier in " 446 "all fragment shaders that have assignments to " 447 "gl_FragDepth"); 448 } 449 } 450 451 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says: 452 * 453 * "If a shared global has multiple initializers, the 454 * initializers must all be constant expressions, and they 455 * must all have the same value. Otherwise, a link error will 456 * result. (A shared global having only one initializer does 457 * not require that initializer to be a constant expression.)" 458 * 459 * Previous to 4.20 the GLSL spec simply said that initializers 460 * must have the same value. In this case of non-constant 461 * initializers, this was impossible to determine. As a result, 462 * no vendor actually implemented that behavior. The 4.20 463 * behavior matches the implemented behavior of at least one other 464 * vendor, so we'll implement that for all GLSL versions. 465 */ 466 if (var->constant_initializer != NULL) { 467 if (existing->constant_initializer != NULL) { 468 if (!var->constant_initializer->has_value(existing->constant_initializer)) { 469 linker_error(prog, "initializers for %s " 470 "`%s' have differing values\n", 471 mode_string(var), var->name); 472 return false; 473 } 474 } else { 475 /* If the first-seen instance of a particular uniform did not 476 * have an initializer but a later instance does, copy the 477 * initializer to the version stored in the symbol table. 478 */ 479 /* FINISHME: This is wrong. The constant_value field should 480 * FINISHME: not be modified! Imagine a case where a shader 481 * FINISHME: without an initializer is linked in two different 482 * FINISHME: programs with shaders that have differing 483 * FINISHME: initializers. Linking with the first will 484 * FINISHME: modify the shader, and linking with the second 485 * FINISHME: will fail. 486 */ 487 existing->constant_initializer = 488 var->constant_initializer->clone(ralloc_parent(existing), 489 NULL); 490 } 491 } 492 493 if (var->has_initializer) { 494 if (existing->has_initializer 495 && (var->constant_initializer == NULL 496 || existing->constant_initializer == NULL)) { 497 linker_error(prog, 498 "shared global variable `%s' has multiple " 499 "non-constant initializers.\n", 500 var->name); 501 return false; 502 } 503 504 /* Some instance had an initializer, so keep track of that. In 505 * this location, all sorts of initializers (constant or 506 * otherwise) will propagate the existence to the variable 507 * stored in the symbol table. 508 */ 509 existing->has_initializer = true; 510 } 511 512 if (existing->invariant != var->invariant) { 513 linker_error(prog, "declarations for %s `%s' have " 514 "mismatching invariant qualifiers\n", 515 mode_string(var), var->name); 516 return false; 517 } 518 if (existing->centroid != var->centroid) { 519 linker_error(prog, "declarations for %s `%s' have " 520 "mismatching centroid qualifiers\n", 521 mode_string(var), var->name); 522 return false; 523 } 524 } else 525 variables.add_variable(var); 526 } 527 } 528 529 return true; 530} 531 532 533/** 534 * Perform validation of uniforms used across multiple shader stages 535 */ 536bool 537cross_validate_uniforms(struct gl_shader_program *prog) 538{ 539 return cross_validate_globals(prog, prog->_LinkedShaders, 540 MESA_SHADER_TYPES, true); 541} 542 543 544/** 545 * Validate that outputs from one stage match inputs of another 546 */ 547bool 548cross_validate_outputs_to_inputs(struct gl_shader_program *prog, 549 gl_shader *producer, gl_shader *consumer) 550{ 551 glsl_symbol_table parameters; 552 /* FINISHME: Figure these out dynamically. */ 553 const char *const producer_stage = "vertex"; 554 const char *const consumer_stage = "fragment"; 555 556 /* Find all shader outputs in the "producer" stage. 557 */ 558 foreach_list(node, producer->ir) { 559 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 560 561 /* FINISHME: For geometry shaders, this should also look for inout 562 * FINISHME: variables. 563 */ 564 if ((var == NULL) || (var->mode != ir_var_out)) 565 continue; 566 567 parameters.add_variable(var); 568 } 569 570 571 /* Find all shader inputs in the "consumer" stage. Any variables that have 572 * matching outputs already in the symbol table must have the same type and 573 * qualifiers. 574 */ 575 foreach_list(node, consumer->ir) { 576 ir_variable *const input = ((ir_instruction *) node)->as_variable(); 577 578 /* FINISHME: For geometry shaders, this should also look for inout 579 * FINISHME: variables. 580 */ 581 if ((input == NULL) || (input->mode != ir_var_in)) 582 continue; 583 584 ir_variable *const output = parameters.get_variable(input->name); 585 if (output != NULL) { 586 /* Check that the types match between stages. 587 */ 588 if (input->type != output->type) { 589 /* There is a bit of a special case for gl_TexCoord. This 590 * built-in is unsized by default. Applications that variable 591 * access it must redeclare it with a size. There is some 592 * language in the GLSL spec that implies the fragment shader 593 * and vertex shader do not have to agree on this size. Other 594 * driver behave this way, and one or two applications seem to 595 * rely on it. 596 * 597 * Neither declaration needs to be modified here because the array 598 * sizes are fixed later when update_array_sizes is called. 599 * 600 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec: 601 * 602 * "Unlike user-defined varying variables, the built-in 603 * varying variables don't have a strict one-to-one 604 * correspondence between the vertex language and the 605 * fragment language." 606 */ 607 if (!output->type->is_array() 608 || (strncmp("gl_", output->name, 3) != 0)) { 609 linker_error(prog, 610 "%s shader output `%s' declared as type `%s', " 611 "but %s shader input declared as type `%s'\n", 612 producer_stage, output->name, 613 output->type->name, 614 consumer_stage, input->type->name); 615 return false; 616 } 617 } 618 619 /* Check that all of the qualifiers match between stages. 620 */ 621 if (input->centroid != output->centroid) { 622 linker_error(prog, 623 "%s shader output `%s' %s centroid qualifier, " 624 "but %s shader input %s centroid qualifier\n", 625 producer_stage, 626 output->name, 627 (output->centroid) ? "has" : "lacks", 628 consumer_stage, 629 (input->centroid) ? "has" : "lacks"); 630 return false; 631 } 632 633 if (input->invariant != output->invariant) { 634 linker_error(prog, 635 "%s shader output `%s' %s invariant qualifier, " 636 "but %s shader input %s invariant qualifier\n", 637 producer_stage, 638 output->name, 639 (output->invariant) ? "has" : "lacks", 640 consumer_stage, 641 (input->invariant) ? "has" : "lacks"); 642 return false; 643 } 644 645 if (input->interpolation != output->interpolation) { 646 linker_error(prog, 647 "%s shader output `%s' specifies %s " 648 "interpolation qualifier, " 649 "but %s shader input specifies %s " 650 "interpolation qualifier\n", 651 producer_stage, 652 output->name, 653 output->interpolation_string(), 654 consumer_stage, 655 input->interpolation_string()); 656 return false; 657 } 658 } 659 } 660 661 return true; 662} 663 664 665/** 666 * Populates a shaders symbol table with all global declarations 667 */ 668static void 669populate_symbol_table(gl_shader *sh) 670{ 671 sh->symbols = new(sh) glsl_symbol_table; 672 673 foreach_list(node, sh->ir) { 674 ir_instruction *const inst = (ir_instruction *) node; 675 ir_variable *var; 676 ir_function *func; 677 678 if ((func = inst->as_function()) != NULL) { 679 sh->symbols->add_function(func); 680 } else if ((var = inst->as_variable()) != NULL) { 681 sh->symbols->add_variable(var); 682 } 683 } 684} 685 686 687/** 688 * Remap variables referenced in an instruction tree 689 * 690 * This is used when instruction trees are cloned from one shader and placed in 691 * another. These trees will contain references to \c ir_variable nodes that 692 * do not exist in the target shader. This function finds these \c ir_variable 693 * references and replaces the references with matching variables in the target 694 * shader. 695 * 696 * If there is no matching variable in the target shader, a clone of the 697 * \c ir_variable is made and added to the target shader. The new variable is 698 * added to \b both the instruction stream and the symbol table. 699 * 700 * \param inst IR tree that is to be processed. 701 * \param symbols Symbol table containing global scope symbols in the 702 * linked shader. 703 * \param instructions Instruction stream where new variable declarations 704 * should be added. 705 */ 706void 707remap_variables(ir_instruction *inst, struct gl_shader *target, 708 hash_table *temps) 709{ 710 class remap_visitor : public ir_hierarchical_visitor { 711 public: 712 remap_visitor(struct gl_shader *target, 713 hash_table *temps) 714 { 715 this->target = target; 716 this->symbols = target->symbols; 717 this->instructions = target->ir; 718 this->temps = temps; 719 } 720 721 virtual ir_visitor_status visit(ir_dereference_variable *ir) 722 { 723 if (ir->var->mode == ir_var_temporary) { 724 ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var); 725 726 assert(var != NULL); 727 ir->var = var; 728 return visit_continue; 729 } 730 731 ir_variable *const existing = 732 this->symbols->get_variable(ir->var->name); 733 if (existing != NULL) 734 ir->var = existing; 735 else { 736 ir_variable *copy = ir->var->clone(this->target, NULL); 737 738 this->symbols->add_variable(copy); 739 this->instructions->push_head(copy); 740 ir->var = copy; 741 } 742 743 return visit_continue; 744 } 745 746 private: 747 struct gl_shader *target; 748 glsl_symbol_table *symbols; 749 exec_list *instructions; 750 hash_table *temps; 751 }; 752 753 remap_visitor v(target, temps); 754 755 inst->accept(&v); 756} 757 758 759/** 760 * Move non-declarations from one instruction stream to another 761 * 762 * The intended usage pattern of this function is to pass the pointer to the 763 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node 764 * pointer) for \c last and \c false for \c make_copies on the first 765 * call. Successive calls pass the return value of the previous call for 766 * \c last and \c true for \c make_copies. 767 * 768 * \param instructions Source instruction stream 769 * \param last Instruction after which new instructions should be 770 * inserted in the target instruction stream 771 * \param make_copies Flag selecting whether instructions in \c instructions 772 * should be copied (via \c ir_instruction::clone) into the 773 * target list or moved. 774 * 775 * \return 776 * The new "last" instruction in the target instruction stream. This pointer 777 * is suitable for use as the \c last parameter of a later call to this 778 * function. 779 */ 780exec_node * 781move_non_declarations(exec_list *instructions, exec_node *last, 782 bool make_copies, gl_shader *target) 783{ 784 hash_table *temps = NULL; 785 786 if (make_copies) 787 temps = hash_table_ctor(0, hash_table_pointer_hash, 788 hash_table_pointer_compare); 789 790 foreach_list_safe(node, instructions) { 791 ir_instruction *inst = (ir_instruction *) node; 792 793 if (inst->as_function()) 794 continue; 795 796 ir_variable *var = inst->as_variable(); 797 if ((var != NULL) && (var->mode != ir_var_temporary)) 798 continue; 799 800 assert(inst->as_assignment() 801 || ((var != NULL) && (var->mode == ir_var_temporary))); 802 803 if (make_copies) { 804 inst = inst->clone(target, NULL); 805 806 if (var != NULL) 807 hash_table_insert(temps, inst, var); 808 else 809 remap_variables(inst, target, temps); 810 } else { 811 inst->remove(); 812 } 813 814 last->insert_after(inst); 815 last = inst; 816 } 817 818 if (make_copies) 819 hash_table_dtor(temps); 820 821 return last; 822} 823 824/** 825 * Get the function signature for main from a shader 826 */ 827static ir_function_signature * 828get_main_function_signature(gl_shader *sh) 829{ 830 ir_function *const f = sh->symbols->get_function("main"); 831 if (f != NULL) { 832 exec_list void_parameters; 833 834 /* Look for the 'void main()' signature and ensure that it's defined. 835 * This keeps the linker from accidentally pick a shader that just 836 * contains a prototype for main. 837 * 838 * We don't have to check for multiple definitions of main (in multiple 839 * shaders) because that would have already been caught above. 840 */ 841 ir_function_signature *sig = f->matching_signature(&void_parameters); 842 if ((sig != NULL) && sig->is_defined) { 843 return sig; 844 } 845 } 846 847 return NULL; 848} 849 850 851/** 852 * Combine a group of shaders for a single stage to generate a linked shader 853 * 854 * \note 855 * If this function is supplied a single shader, it is cloned, and the new 856 * shader is returned. 857 */ 858static struct gl_shader * 859link_intrastage_shaders(void *mem_ctx, 860 struct gl_context *ctx, 861 struct gl_shader_program *prog, 862 struct gl_shader **shader_list, 863 unsigned num_shaders) 864{ 865 /* Check that global variables defined in multiple shaders are consistent. 866 */ 867 if (!cross_validate_globals(prog, shader_list, num_shaders, false)) 868 return NULL; 869 870 /* Check that there is only a single definition of each function signature 871 * across all shaders. 872 */ 873 for (unsigned i = 0; i < (num_shaders - 1); i++) { 874 foreach_list(node, shader_list[i]->ir) { 875 ir_function *const f = ((ir_instruction *) node)->as_function(); 876 877 if (f == NULL) 878 continue; 879 880 for (unsigned j = i + 1; j < num_shaders; j++) { 881 ir_function *const other = 882 shader_list[j]->symbols->get_function(f->name); 883 884 /* If the other shader has no function (and therefore no function 885 * signatures) with the same name, skip to the next shader. 886 */ 887 if (other == NULL) 888 continue; 889 890 foreach_iter (exec_list_iterator, iter, *f) { 891 ir_function_signature *sig = 892 (ir_function_signature *) iter.get(); 893 894 if (!sig->is_defined || sig->is_builtin) 895 continue; 896 897 ir_function_signature *other_sig = 898 other->exact_matching_signature(& sig->parameters); 899 900 if ((other_sig != NULL) && other_sig->is_defined 901 && !other_sig->is_builtin) { 902 linker_error(prog, "function `%s' is multiply defined", 903 f->name); 904 return NULL; 905 } 906 } 907 } 908 } 909 } 910 911 /* Find the shader that defines main, and make a clone of it. 912 * 913 * Starting with the clone, search for undefined references. If one is 914 * found, find the shader that defines it. Clone the reference and add 915 * it to the shader. Repeat until there are no undefined references or 916 * until a reference cannot be resolved. 917 */ 918 gl_shader *main = NULL; 919 for (unsigned i = 0; i < num_shaders; i++) { 920 if (get_main_function_signature(shader_list[i]) != NULL) { 921 main = shader_list[i]; 922 break; 923 } 924 } 925 926 if (main == NULL) { 927 linker_error(prog, "%s shader lacks `main'\n", 928 (shader_list[0]->Type == GL_VERTEX_SHADER) 929 ? "vertex" : "fragment"); 930 return NULL; 931 } 932 933 gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type); 934 linked->ir = new(linked) exec_list; 935 clone_ir_list(mem_ctx, linked->ir, main->ir); 936 937 populate_symbol_table(linked); 938 939 /* The a pointer to the main function in the final linked shader (i.e., the 940 * copy of the original shader that contained the main function). 941 */ 942 ir_function_signature *const main_sig = get_main_function_signature(linked); 943 944 /* Move any instructions other than variable declarations or function 945 * declarations into main. 946 */ 947 exec_node *insertion_point = 948 move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false, 949 linked); 950 951 for (unsigned i = 0; i < num_shaders; i++) { 952 if (shader_list[i] == main) 953 continue; 954 955 insertion_point = move_non_declarations(shader_list[i]->ir, 956 insertion_point, true, linked); 957 } 958 959 /* Resolve initializers for global variables in the linked shader. 960 */ 961 unsigned num_linking_shaders = num_shaders; 962 for (unsigned i = 0; i < num_shaders; i++) 963 num_linking_shaders += shader_list[i]->num_builtins_to_link; 964 965 gl_shader **linking_shaders = 966 (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *)); 967 968 memcpy(linking_shaders, shader_list, 969 sizeof(linking_shaders[0]) * num_shaders); 970 971 unsigned idx = num_shaders; 972 for (unsigned i = 0; i < num_shaders; i++) { 973 memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link, 974 sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link); 975 idx += shader_list[i]->num_builtins_to_link; 976 } 977 978 assert(idx == num_linking_shaders); 979 980 if (!link_function_calls(prog, linked, linking_shaders, 981 num_linking_shaders)) { 982 ctx->Driver.DeleteShader(ctx, linked); 983 linked = NULL; 984 } 985 986 free(linking_shaders); 987 988#ifdef DEBUG 989 /* At this point linked should contain all of the linked IR, so 990 * validate it to make sure nothing went wrong. 991 */ 992 if (linked) 993 validate_ir_tree(linked->ir); 994#endif 995 996 /* Make a pass over all variable declarations to ensure that arrays with 997 * unspecified sizes have a size specified. The size is inferred from the 998 * max_array_access field. 999 */ 1000 if (linked != NULL) { 1001 class array_sizing_visitor : public ir_hierarchical_visitor { 1002 public: 1003 virtual ir_visitor_status visit(ir_variable *var) 1004 { 1005 if (var->type->is_array() && (var->type->length == 0)) { 1006 const glsl_type *type = 1007 glsl_type::get_array_instance(var->type->fields.array, 1008 var->max_array_access + 1); 1009 1010 assert(type != NULL); 1011 var->type = type; 1012 } 1013 1014 return visit_continue; 1015 } 1016 } v; 1017 1018 v.run(linked->ir); 1019 } 1020 1021 return linked; 1022} 1023 1024/** 1025 * Update the sizes of linked shader uniform arrays to the maximum 1026 * array index used. 1027 * 1028 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec: 1029 * 1030 * If one or more elements of an array are active, 1031 * GetActiveUniform will return the name of the array in name, 1032 * subject to the restrictions listed above. The type of the array 1033 * is returned in type. The size parameter contains the highest 1034 * array element index used, plus one. The compiler or linker 1035 * determines the highest index used. There will be only one 1036 * active uniform reported by the GL per uniform array. 1037 1038 */ 1039static void 1040update_array_sizes(struct gl_shader_program *prog) 1041{ 1042 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { 1043 if (prog->_LinkedShaders[i] == NULL) 1044 continue; 1045 1046 foreach_list(node, prog->_LinkedShaders[i]->ir) { 1047 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 1048 1049 if ((var == NULL) || (var->mode != ir_var_uniform && 1050 var->mode != ir_var_in && 1051 var->mode != ir_var_out) || 1052 !var->type->is_array()) 1053 continue; 1054 1055 unsigned int size = var->max_array_access; 1056 for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) { 1057 if (prog->_LinkedShaders[j] == NULL) 1058 continue; 1059 1060 foreach_list(node2, prog->_LinkedShaders[j]->ir) { 1061 ir_variable *other_var = ((ir_instruction *) node2)->as_variable(); 1062 if (!other_var) 1063 continue; 1064 1065 if (strcmp(var->name, other_var->name) == 0 && 1066 other_var->max_array_access > size) { 1067 size = other_var->max_array_access; 1068 } 1069 } 1070 } 1071 1072 if (size + 1 != var->type->fields.array->length) { 1073 /* If this is a built-in uniform (i.e., it's backed by some 1074 * fixed-function state), adjust the number of state slots to 1075 * match the new array size. The number of slots per array entry 1076 * is not known. It seems safe to assume that the total number of 1077 * slots is an integer multiple of the number of array elements. 1078 * Determine the number of slots per array element by dividing by 1079 * the old (total) size. 1080 */ 1081 if (var->num_state_slots > 0) { 1082 var->num_state_slots = (size + 1) 1083 * (var->num_state_slots / var->type->length); 1084 } 1085 1086 var->type = glsl_type::get_array_instance(var->type->fields.array, 1087 size + 1); 1088 /* FINISHME: We should update the types of array 1089 * dereferences of this variable now. 1090 */ 1091 } 1092 } 1093 } 1094} 1095 1096/** 1097 * Find a contiguous set of available bits in a bitmask. 1098 * 1099 * \param used_mask Bits representing used (1) and unused (0) locations 1100 * \param needed_count Number of contiguous bits needed. 1101 * 1102 * \return 1103 * Base location of the available bits on success or -1 on failure. 1104 */ 1105int 1106find_available_slots(unsigned used_mask, unsigned needed_count) 1107{ 1108 unsigned needed_mask = (1 << needed_count) - 1; 1109 const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count; 1110 1111 /* The comparison to 32 is redundant, but without it GCC emits "warning: 1112 * cannot optimize possibly infinite loops" for the loop below. 1113 */ 1114 if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32)) 1115 return -1; 1116 1117 for (int i = 0; i <= max_bit_to_test; i++) { 1118 if ((needed_mask & ~used_mask) == needed_mask) 1119 return i; 1120 1121 needed_mask <<= 1; 1122 } 1123 1124 return -1; 1125} 1126 1127 1128/** 1129 * Assign locations for either VS inputs for FS outputs 1130 * 1131 * \param prog Shader program whose variables need locations assigned 1132 * \param target_index Selector for the program target to receive location 1133 * assignmnets. Must be either \c MESA_SHADER_VERTEX or 1134 * \c MESA_SHADER_FRAGMENT. 1135 * \param max_index Maximum number of generic locations. This corresponds 1136 * to either the maximum number of draw buffers or the 1137 * maximum number of generic attributes. 1138 * 1139 * \return 1140 * If locations are successfully assigned, true is returned. Otherwise an 1141 * error is emitted to the shader link log and false is returned. 1142 */ 1143bool 1144assign_attribute_or_color_locations(gl_shader_program *prog, 1145 unsigned target_index, 1146 unsigned max_index) 1147{ 1148 /* Mark invalid locations as being used. 1149 */ 1150 unsigned used_locations = (max_index >= 32) 1151 ? ~0 : ~((1 << max_index) - 1); 1152 1153 assert((target_index == MESA_SHADER_VERTEX) 1154 || (target_index == MESA_SHADER_FRAGMENT)); 1155 1156 gl_shader *const sh = prog->_LinkedShaders[target_index]; 1157 if (sh == NULL) 1158 return true; 1159 1160 /* Operate in a total of four passes. 1161 * 1162 * 1. Invalidate the location assignments for all vertex shader inputs. 1163 * 1164 * 2. Assign locations for inputs that have user-defined (via 1165 * glBindVertexAttribLocation) locations and outputs that have 1166 * user-defined locations (via glBindFragDataLocation). 1167 * 1168 * 3. Sort the attributes without assigned locations by number of slots 1169 * required in decreasing order. Fragmentation caused by attribute 1170 * locations assigned by the application may prevent large attributes 1171 * from having enough contiguous space. 1172 * 1173 * 4. Assign locations to any inputs without assigned locations. 1174 */ 1175 1176 const int generic_base = (target_index == MESA_SHADER_VERTEX) 1177 ? (int) VERT_ATTRIB_GENERIC0 : (int) FRAG_RESULT_DATA0; 1178 1179 const enum ir_variable_mode direction = 1180 (target_index == MESA_SHADER_VERTEX) ? ir_var_in : ir_var_out; 1181 1182 1183 link_invalidate_variable_locations(sh, direction, generic_base); 1184 1185 /* Temporary storage for the set of attributes that need locations assigned. 1186 */ 1187 struct temp_attr { 1188 unsigned slots; 1189 ir_variable *var; 1190 1191 /* Used below in the call to qsort. */ 1192 static int compare(const void *a, const void *b) 1193 { 1194 const temp_attr *const l = (const temp_attr *) a; 1195 const temp_attr *const r = (const temp_attr *) b; 1196 1197 /* Reversed because we want a descending order sort below. */ 1198 return r->slots - l->slots; 1199 } 1200 } to_assign[16]; 1201 1202 unsigned num_attr = 0; 1203 1204 foreach_list(node, sh->ir) { 1205 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 1206 1207 if ((var == NULL) || (var->mode != (unsigned) direction)) 1208 continue; 1209 1210 if (var->explicit_location) { 1211 if ((var->location >= (int)(max_index + generic_base)) 1212 || (var->location < 0)) { 1213 linker_error(prog, 1214 "invalid explicit location %d specified for `%s'\n", 1215 (var->location < 0) 1216 ? var->location : var->location - generic_base, 1217 var->name); 1218 return false; 1219 } 1220 } else if (target_index == MESA_SHADER_VERTEX) { 1221 unsigned binding; 1222 1223 if (prog->AttributeBindings->get(binding, var->name)) { 1224 assert(binding >= VERT_ATTRIB_GENERIC0); 1225 var->location = binding; 1226 } 1227 } else if (target_index == MESA_SHADER_FRAGMENT) { 1228 unsigned binding; 1229 1230 if (prog->FragDataBindings->get(binding, var->name)) { 1231 assert(binding >= FRAG_RESULT_DATA0); 1232 var->location = binding; 1233 } 1234 } 1235 1236 /* If the variable is not a built-in and has a location statically 1237 * assigned in the shader (presumably via a layout qualifier), make sure 1238 * that it doesn't collide with other assigned locations. Otherwise, 1239 * add it to the list of variables that need linker-assigned locations. 1240 */ 1241 const unsigned slots = count_attribute_slots(var->type); 1242 if (var->location != -1) { 1243 if (var->location >= generic_base) { 1244 /* From page 61 of the OpenGL 4.0 spec: 1245 * 1246 * "LinkProgram will fail if the attribute bindings assigned 1247 * by BindAttribLocation do not leave not enough space to 1248 * assign a location for an active matrix attribute or an 1249 * active attribute array, both of which require multiple 1250 * contiguous generic attributes." 1251 * 1252 * Previous versions of the spec contain similar language but omit 1253 * the bit about attribute arrays. 1254 * 1255 * Page 61 of the OpenGL 4.0 spec also says: 1256 * 1257 * "It is possible for an application to bind more than one 1258 * attribute name to the same location. This is referred to as 1259 * aliasing. This will only work if only one of the aliased 1260 * attributes is active in the executable program, or if no 1261 * path through the shader consumes more than one attribute of 1262 * a set of attributes aliased to the same location. A link 1263 * error can occur if the linker determines that every path 1264 * through the shader consumes multiple aliased attributes, 1265 * but implementations are not required to generate an error 1266 * in this case." 1267 * 1268 * These two paragraphs are either somewhat contradictory, or I 1269 * don't fully understand one or both of them. 1270 */ 1271 /* FINISHME: The code as currently written does not support 1272 * FINISHME: attribute location aliasing (see comment above). 1273 */ 1274 /* Mask representing the contiguous slots that will be used by 1275 * this attribute. 1276 */ 1277 const unsigned attr = var->location - generic_base; 1278 const unsigned use_mask = (1 << slots) - 1; 1279 1280 /* Generate a link error if the set of bits requested for this 1281 * attribute overlaps any previously allocated bits. 1282 */ 1283 if ((~(use_mask << attr) & used_locations) != used_locations) { 1284 linker_error(prog, 1285 "insufficient contiguous attribute locations " 1286 "available for vertex shader input `%s'", 1287 var->name); 1288 return false; 1289 } 1290 1291 used_locations |= (use_mask << attr); 1292 } 1293 1294 continue; 1295 } 1296 1297 to_assign[num_attr].slots = slots; 1298 to_assign[num_attr].var = var; 1299 num_attr++; 1300 } 1301 1302 /* If all of the attributes were assigned locations by the application (or 1303 * are built-in attributes with fixed locations), return early. This should 1304 * be the common case. 1305 */ 1306 if (num_attr == 0) 1307 return true; 1308 1309 qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare); 1310 1311 if (target_index == MESA_SHADER_VERTEX) { 1312 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can 1313 * only be explicitly assigned by via glBindAttribLocation. Mark it as 1314 * reserved to prevent it from being automatically allocated below. 1315 */ 1316 find_deref_visitor find("gl_Vertex"); 1317 find.run(sh->ir); 1318 if (find.variable_found()) 1319 used_locations |= (1 << 0); 1320 } 1321 1322 for (unsigned i = 0; i < num_attr; i++) { 1323 /* Mask representing the contiguous slots that will be used by this 1324 * attribute. 1325 */ 1326 const unsigned use_mask = (1 << to_assign[i].slots) - 1; 1327 1328 int location = find_available_slots(used_locations, to_assign[i].slots); 1329 1330 if (location < 0) { 1331 const char *const string = (target_index == MESA_SHADER_VERTEX) 1332 ? "vertex shader input" : "fragment shader output"; 1333 1334 linker_error(prog, 1335 "insufficient contiguous attribute locations " 1336 "available for %s `%s'", 1337 string, to_assign[i].var->name); 1338 return false; 1339 } 1340 1341 to_assign[i].var->location = generic_base + location; 1342 used_locations |= (use_mask << location); 1343 } 1344 1345 return true; 1346} 1347 1348 1349/** 1350 * Demote shader inputs and outputs that are not used in other stages 1351 */ 1352void 1353demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode) 1354{ 1355 foreach_list(node, sh->ir) { 1356 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 1357 1358 if ((var == NULL) || (var->mode != int(mode))) 1359 continue; 1360 1361 /* A shader 'in' or 'out' variable is only really an input or output if 1362 * its value is used by other shader stages. This will cause the variable 1363 * to have a location assigned. 1364 */ 1365 if (var->location == -1) { 1366 var->mode = ir_var_auto; 1367 } 1368 } 1369} 1370 1371 1372/** 1373 * Data structure tracking information about a transform feedback declaration 1374 * during linking. 1375 */ 1376class tfeedback_decl 1377{ 1378public: 1379 bool init(struct gl_shader_program *prog, const void *mem_ctx, 1380 const char *input); 1381 static bool is_same(const tfeedback_decl &x, const tfeedback_decl &y); 1382 bool assign_location(struct gl_context *ctx, struct gl_shader_program *prog, 1383 ir_variable *output_var); 1384 bool store(struct gl_shader_program *prog, 1385 struct gl_transform_feedback_info *info, unsigned buffer) const; 1386 1387 1388 /** 1389 * True if assign_location() has been called for this object. 1390 */ 1391 bool is_assigned() const 1392 { 1393 return this->location != -1; 1394 } 1395 1396 /** 1397 * Determine whether this object refers to the variable var. 1398 */ 1399 bool matches_var(ir_variable *var) const 1400 { 1401 return strcmp(var->name, this->var_name) == 0; 1402 } 1403 1404 /** 1405 * The total number of varying components taken up by this variable. Only 1406 * valid if is_assigned() is true. 1407 */ 1408 unsigned num_components() const 1409 { 1410 return this->vector_elements * this->matrix_columns; 1411 } 1412 1413private: 1414 /** 1415 * The name that was supplied to glTransformFeedbackVaryings. Used for 1416 * error reporting. 1417 */ 1418 const char *orig_name; 1419 1420 /** 1421 * The name of the variable, parsed from orig_name. 1422 */ 1423 char *var_name; 1424 1425 /** 1426 * True if the declaration in orig_name represents an array. 1427 */ 1428 bool is_array; 1429 1430 /** 1431 * If is_array is true, the array index that was specified in orig_name. 1432 */ 1433 unsigned array_index; 1434 1435 /** 1436 * The vertex shader output location that the linker assigned for this 1437 * variable. -1 if a location hasn't been assigned yet. 1438 */ 1439 int location; 1440 1441 /** 1442 * If location != -1, the number of vector elements in this variable, or 1 1443 * if this variable is a scalar. 1444 */ 1445 unsigned vector_elements; 1446 1447 /** 1448 * If location != -1, the number of matrix columns in this variable, or 1 1449 * if this variable is not a matrix. 1450 */ 1451 unsigned matrix_columns; 1452}; 1453 1454 1455/** 1456 * Initialize this object based on a string that was passed to 1457 * glTransformFeedbackVaryings. If there is a parse error, the error is 1458 * reported using linker_error(), and false is returned. 1459 */ 1460bool 1461tfeedback_decl::init(struct gl_shader_program *prog, const void *mem_ctx, 1462 const char *input) 1463{ 1464 /* We don't have to be pedantic about what is a valid GLSL variable name, 1465 * because any variable with an invalid name can't exist in the IR anyway. 1466 */ 1467 1468 this->location = -1; 1469 this->orig_name = input; 1470 1471 const char *bracket = strrchr(input, '['); 1472 1473 if (bracket) { 1474 this->var_name = ralloc_strndup(mem_ctx, input, bracket - input); 1475 if (sscanf(bracket, "[%u]", &this->array_index) == 1) { 1476 this->is_array = true; 1477 return true; 1478 } 1479 } else { 1480 this->var_name = ralloc_strdup(mem_ctx, input); 1481 this->is_array = false; 1482 return true; 1483 } 1484 1485 linker_error(prog, "Cannot parse transform feedback varying %s", input); 1486 return false; 1487} 1488 1489 1490/** 1491 * Determine whether two tfeedback_decl objects refer to the same variable and 1492 * array index (if applicable). 1493 */ 1494bool 1495tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y) 1496{ 1497 if (strcmp(x.var_name, y.var_name) != 0) 1498 return false; 1499 if (x.is_array != y.is_array) 1500 return false; 1501 if (x.is_array && x.array_index != y.array_index) 1502 return false; 1503 return true; 1504} 1505 1506 1507/** 1508 * Assign a location for this tfeedback_decl object based on the location 1509 * assignment in output_var. 1510 * 1511 * If an error occurs, the error is reported through linker_error() and false 1512 * is returned. 1513 */ 1514bool 1515tfeedback_decl::assign_location(struct gl_context *ctx, 1516 struct gl_shader_program *prog, 1517 ir_variable *output_var) 1518{ 1519 if (output_var->type->is_array()) { 1520 /* Array variable */ 1521 if (!this->is_array) { 1522 linker_error(prog, "Transform feedback varying %s found, " 1523 "but it's not an array ([] not expected).", 1524 this->orig_name); 1525 return false; 1526 } 1527 /* Check array bounds. */ 1528 if (this->array_index >= 1529 (unsigned) output_var->type->array_size()) { 1530 linker_error(prog, "Transform feedback varying %s has index " 1531 "%i, but the array size is %i.", 1532 this->orig_name, this->array_index, 1533 output_var->type->array_size()); 1534 return false; 1535 } 1536 const unsigned matrix_cols = 1537 output_var->type->fields.array->matrix_columns; 1538 this->location = output_var->location + this->array_index * matrix_cols; 1539 this->vector_elements = output_var->type->fields.array->vector_elements; 1540 this->matrix_columns = matrix_cols; 1541 } else { 1542 /* Regular variable (scalar, vector, or matrix) */ 1543 if (this->is_array) { 1544 linker_error(prog, "Transform feedback varying %s found, " 1545 "but it's an array ([] expected).", 1546 this->orig_name); 1547 return false; 1548 } 1549 this->location = output_var->location; 1550 this->vector_elements = output_var->type->vector_elements; 1551 this->matrix_columns = output_var->type->matrix_columns; 1552 } 1553 /* From GL_EXT_transform_feedback: 1554 * A program will fail to link if: 1555 * 1556 * * the total number of components to capture in any varying 1557 * variable in <varyings> is greater than the constant 1558 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the 1559 * buffer mode is SEPARATE_ATTRIBS_EXT; 1560 */ 1561 if (prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS && 1562 this->num_components() > 1563 ctx->Const.MaxTransformFeedbackSeparateComponents) { 1564 linker_error(prog, "Transform feedback varying %s exceeds " 1565 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.", 1566 this->orig_name); 1567 return false; 1568 } 1569 1570 return true; 1571} 1572 1573 1574/** 1575 * Update gl_transform_feedback_info to reflect this tfeedback_decl. 1576 * 1577 * If an error occurs, the error is reported through linker_error() and false 1578 * is returned. 1579 */ 1580bool 1581tfeedback_decl::store(struct gl_shader_program *prog, 1582 struct gl_transform_feedback_info *info, 1583 unsigned buffer) const 1584{ 1585 if (!this->is_assigned()) { 1586 /* From GL_EXT_transform_feedback: 1587 * A program will fail to link if: 1588 * 1589 * * any variable name specified in the <varyings> array is not 1590 * declared as an output in the geometry shader (if present) or 1591 * the vertex shader (if no geometry shader is present); 1592 */ 1593 linker_error(prog, "Transform feedback varying %s undeclared.", 1594 this->orig_name); 1595 return false; 1596 } 1597 for (unsigned v = 0; v < this->matrix_columns; ++v) { 1598 info->Outputs[info->NumOutputs].OutputRegister = this->location + v; 1599 info->Outputs[info->NumOutputs].NumComponents = this->vector_elements; 1600 info->Outputs[info->NumOutputs].OutputBuffer = buffer; 1601 info->Outputs[info->NumOutputs].DstOffset = info->BufferStride[buffer]; 1602 ++info->NumOutputs; 1603 info->BufferStride[buffer] += this->vector_elements; 1604 } 1605 return true; 1606} 1607 1608 1609/** 1610 * Parse all the transform feedback declarations that were passed to 1611 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects. 1612 * 1613 * If an error occurs, the error is reported through linker_error() and false 1614 * is returned. 1615 */ 1616static bool 1617parse_tfeedback_decls(struct gl_shader_program *prog, const void *mem_ctx, 1618 unsigned num_names, char **varying_names, 1619 tfeedback_decl *decls) 1620{ 1621 for (unsigned i = 0; i < num_names; ++i) { 1622 if (!decls[i].init(prog, mem_ctx, varying_names[i])) 1623 return false; 1624 /* From GL_EXT_transform_feedback: 1625 * A program will fail to link if: 1626 * 1627 * * any two entries in the <varyings> array specify the same varying 1628 * variable; 1629 * 1630 * We interpret this to mean "any two entries in the <varyings> array 1631 * specify the same varying variable and array index", since transform 1632 * feedback of arrays would be useless otherwise. 1633 */ 1634 for (unsigned j = 0; j < i; ++j) { 1635 if (tfeedback_decl::is_same(decls[i], decls[j])) { 1636 linker_error(prog, "Transform feedback varying %s specified " 1637 "more than once.", varying_names[i]); 1638 return false; 1639 } 1640 } 1641 } 1642 return true; 1643} 1644 1645 1646/** 1647 * Assign a location for a variable that is produced in one pipeline stage 1648 * (the "producer") and consumed in the next stage (the "consumer"). 1649 * 1650 * \param input_var is the input variable declaration in the consumer. 1651 * 1652 * \param output_var is the output variable declaration in the producer. 1653 * 1654 * \param input_index is the counter that keeps track of assigned input 1655 * locations in the consumer. 1656 * 1657 * \param output_index is the counter that keeps track of assigned output 1658 * locations in the producer. 1659 * 1660 * It is permissible for \c input_var to be NULL (this happens if a variable 1661 * is output by the producer and consumed by transform feedback, but not 1662 * consumed by the consumer). 1663 * 1664 * If the variable has already been assigned a location, this function has no 1665 * effect. 1666 */ 1667void 1668assign_varying_location(ir_variable *input_var, ir_variable *output_var, 1669 unsigned *input_index, unsigned *output_index) 1670{ 1671 if (output_var->location != -1) { 1672 /* Location already assigned. */ 1673 return; 1674 } 1675 1676 if (input_var) { 1677 assert(input_var->location == -1); 1678 input_var->location = *input_index; 1679 } 1680 1681 output_var->location = *output_index; 1682 1683 /* FINISHME: Support for "varying" records in GLSL 1.50. */ 1684 assert(!output_var->type->is_record()); 1685 1686 if (output_var->type->is_array()) { 1687 const unsigned slots = output_var->type->length 1688 * output_var->type->fields.array->matrix_columns; 1689 1690 *output_index += slots; 1691 *input_index += slots; 1692 } else { 1693 const unsigned slots = output_var->type->matrix_columns; 1694 1695 *output_index += slots; 1696 *input_index += slots; 1697 } 1698} 1699 1700 1701/** 1702 * Assign locations for all variables that are produced in one pipeline stage 1703 * (the "producer") and consumed in the next stage (the "consumer"). 1704 * 1705 * Variables produced by the producer may also be consumed by transform 1706 * feedback. 1707 * 1708 * \param num_tfeedback_decls is the number of declarations indicating 1709 * variables that may be consumed by transform feedback. 1710 * 1711 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects 1712 * representing the result of parsing the strings passed to 1713 * glTransformFeedbackVaryings(). assign_location() will be called for 1714 * each of these objects that matches one of the outputs of the 1715 * producer. 1716 * 1717 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to 1718 * be NULL. In this case, varying locations are assigned solely based on the 1719 * requirements of transform feedback. 1720 */ 1721bool 1722assign_varying_locations(struct gl_context *ctx, 1723 struct gl_shader_program *prog, 1724 gl_shader *producer, gl_shader *consumer, 1725 unsigned num_tfeedback_decls, 1726 tfeedback_decl *tfeedback_decls) 1727{ 1728 /* FINISHME: Set dynamically when geometry shader support is added. */ 1729 unsigned output_index = VERT_RESULT_VAR0; 1730 unsigned input_index = FRAG_ATTRIB_VAR0; 1731 1732 /* Operate in a total of three passes. 1733 * 1734 * 1. Assign locations for any matching inputs and outputs. 1735 * 1736 * 2. Mark output variables in the producer that do not have locations as 1737 * not being outputs. This lets the optimizer eliminate them. 1738 * 1739 * 3. Mark input variables in the consumer that do not have locations as 1740 * not being inputs. This lets the optimizer eliminate them. 1741 */ 1742 1743 link_invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0); 1744 if (consumer) 1745 link_invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0); 1746 1747 foreach_list(node, producer->ir) { 1748 ir_variable *const output_var = ((ir_instruction *) node)->as_variable(); 1749 1750 if ((output_var == NULL) || (output_var->mode != ir_var_out)) 1751 continue; 1752 1753 ir_variable *input_var = 1754 consumer ? consumer->symbols->get_variable(output_var->name) : NULL; 1755 1756 if (input_var && input_var->mode != ir_var_in) 1757 input_var = NULL; 1758 1759 if (input_var) { 1760 assign_varying_location(input_var, output_var, &input_index, 1761 &output_index); 1762 } 1763 1764 for (unsigned i = 0; i < num_tfeedback_decls; ++i) { 1765 if (!tfeedback_decls[i].is_assigned() && 1766 tfeedback_decls[i].matches_var(output_var)) { 1767 if (output_var->location == -1) { 1768 assign_varying_location(input_var, output_var, &input_index, 1769 &output_index); 1770 } 1771 if (!tfeedback_decls[i].assign_location(ctx, prog, output_var)) 1772 return false; 1773 } 1774 } 1775 } 1776 1777 unsigned varying_vectors = 0; 1778 1779 if (consumer) { 1780 foreach_list(node, consumer->ir) { 1781 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 1782 1783 if ((var == NULL) || (var->mode != ir_var_in)) 1784 continue; 1785 1786 if (var->location == -1) { 1787 if (prog->Version <= 120) { 1788 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec: 1789 * 1790 * Only those varying variables used (i.e. read) in 1791 * the fragment shader executable must be written to 1792 * by the vertex shader executable; declaring 1793 * superfluous varying variables in a vertex shader is 1794 * permissible. 1795 * 1796 * We interpret this text as meaning that the VS must 1797 * write the variable for the FS to read it. See 1798 * "glsl1-varying read but not written" in piglit. 1799 */ 1800 1801 linker_error(prog, "fragment shader varying %s not written " 1802 "by vertex shader\n.", var->name); 1803 } 1804 1805 /* An 'in' variable is only really a shader input if its 1806 * value is written by the previous stage. 1807 */ 1808 var->mode = ir_var_auto; 1809 } else { 1810 /* The packing rules are used for vertex shader inputs are also 1811 * used for fragment shader inputs. 1812 */ 1813 varying_vectors += count_attribute_slots(var->type); 1814 } 1815 } 1816 } 1817 1818 if (ctx->API == API_OPENGLES2 || prog->Version == 100) { 1819 if (varying_vectors > ctx->Const.MaxVarying) { 1820 if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { 1821 linker_warning(prog, "shader uses too many varying vectors " 1822 "(%u > %u), but the driver will try to optimize " 1823 "them out; this is non-portable out-of-spec " 1824 "behavior\n", 1825 varying_vectors, ctx->Const.MaxVarying); 1826 } else { 1827 linker_error(prog, "shader uses too many varying vectors " 1828 "(%u > %u)\n", 1829 varying_vectors, ctx->Const.MaxVarying); 1830 return false; 1831 } 1832 } 1833 } else { 1834 const unsigned float_components = varying_vectors * 4; 1835 if (float_components > ctx->Const.MaxVarying * 4) { 1836 if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) { 1837 linker_warning(prog, "shader uses too many varying components " 1838 "(%u > %u), but the driver will try to optimize " 1839 "them out; this is non-portable out-of-spec " 1840 "behavior\n", 1841 float_components, ctx->Const.MaxVarying * 4); 1842 } else { 1843 linker_error(prog, "shader uses too many varying components " 1844 "(%u > %u)\n", 1845 float_components, ctx->Const.MaxVarying * 4); 1846 return false; 1847 } 1848 } 1849 } 1850 1851 return true; 1852} 1853 1854 1855/** 1856 * Store transform feedback location assignments into 1857 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls. 1858 * 1859 * If an error occurs, the error is reported through linker_error() and false 1860 * is returned. 1861 */ 1862static bool 1863store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog, 1864 unsigned num_tfeedback_decls, 1865 tfeedback_decl *tfeedback_decls) 1866{ 1867 unsigned total_tfeedback_components = 0; 1868 bool separate_attribs_mode = 1869 prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS; 1870 memset(&prog->LinkedTransformFeedback, 0, 1871 sizeof(prog->LinkedTransformFeedback)); 1872 prog->LinkedTransformFeedback.NumBuffers = 1873 separate_attribs_mode ? num_tfeedback_decls : 1; 1874 for (unsigned i = 0; i < num_tfeedback_decls; ++i) { 1875 unsigned buffer = separate_attribs_mode ? i : 0; 1876 if (!tfeedback_decls[i].store(prog, &prog->LinkedTransformFeedback, 1877 buffer)) 1878 return false; 1879 total_tfeedback_components += tfeedback_decls[i].num_components(); 1880 } 1881 1882 /* From GL_EXT_transform_feedback: 1883 * A program will fail to link if: 1884 * 1885 * * the total number of components to capture is greater than 1886 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT 1887 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT. 1888 */ 1889 if (prog->TransformFeedback.BufferMode == GL_INTERLEAVED_ATTRIBS && 1890 total_tfeedback_components > 1891 ctx->Const.MaxTransformFeedbackInterleavedComponents) { 1892 linker_error(prog, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS " 1893 "limit has been exceeded."); 1894 return false; 1895 } 1896 1897 return true; 1898} 1899 1900/** 1901 * Store the gl_FragDepth layout in the gl_shader_program struct. 1902 */ 1903static void 1904store_fragdepth_layout(struct gl_shader_program *prog) 1905{ 1906 if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] == NULL) { 1907 return; 1908 } 1909 1910 struct exec_list *ir = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]->ir; 1911 1912 /* We don't look up the gl_FragDepth symbol directly because if 1913 * gl_FragDepth is not used in the shader, it's removed from the IR. 1914 * However, the symbol won't be removed from the symbol table. 1915 * 1916 * We're only interested in the cases where the variable is NOT removed 1917 * from the IR. 1918 */ 1919 foreach_list(node, ir) { 1920 ir_variable *const var = ((ir_instruction *) node)->as_variable(); 1921 1922 if (var == NULL || var->mode != ir_var_out) { 1923 continue; 1924 } 1925 1926 if (strcmp(var->name, "gl_FragDepth") == 0) { 1927 switch (var->depth_layout) { 1928 case ir_depth_layout_none: 1929 prog->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE; 1930 return; 1931 case ir_depth_layout_any: 1932 prog->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY; 1933 return; 1934 case ir_depth_layout_greater: 1935 prog->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER; 1936 return; 1937 case ir_depth_layout_less: 1938 prog->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS; 1939 return; 1940 case ir_depth_layout_unchanged: 1941 prog->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED; 1942 return; 1943 default: 1944 assert(0); 1945 return; 1946 } 1947 } 1948 } 1949} 1950 1951/** 1952 * Validate the resources used by a program versus the implementation limits 1953 */ 1954static bool 1955check_resources(struct gl_context *ctx, struct gl_shader_program *prog) 1956{ 1957 static const char *const shader_names[MESA_SHADER_TYPES] = { 1958 "vertex", "fragment", "geometry" 1959 }; 1960 1961 const unsigned max_samplers[MESA_SHADER_TYPES] = { 1962 ctx->Const.MaxVertexTextureImageUnits, 1963 ctx->Const.MaxTextureImageUnits, 1964 ctx->Const.MaxGeometryTextureImageUnits 1965 }; 1966 1967 const unsigned max_uniform_components[MESA_SHADER_TYPES] = { 1968 ctx->Const.VertexProgram.MaxUniformComponents, 1969 ctx->Const.FragmentProgram.MaxUniformComponents, 1970 0 /* FINISHME: Geometry shaders. */ 1971 }; 1972 1973 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { 1974 struct gl_shader *sh = prog->_LinkedShaders[i]; 1975 1976 if (sh == NULL) 1977 continue; 1978 1979 if (sh->num_samplers > max_samplers[i]) { 1980 linker_error(prog, "Too many %s shader texture samplers", 1981 shader_names[i]); 1982 } 1983 1984 if (sh->num_uniform_components > max_uniform_components[i]) { 1985 if (ctx->Const.GLSLSkipStrictMaxUniformLimitCheck) { 1986 linker_warning(prog, "Too many %s shader uniform components, " 1987 "but the driver will try to optimize them out; " 1988 "this is non-portable out-of-spec behavior\n", 1989 shader_names[i]); 1990 } else { 1991 linker_error(prog, "Too many %s shader uniform components", 1992 shader_names[i]); 1993 } 1994 } 1995 } 1996 1997 return prog->LinkStatus; 1998} 1999 2000void 2001link_shaders(struct gl_context *ctx, struct gl_shader_program *prog) 2002{ 2003 tfeedback_decl *tfeedback_decls = NULL; 2004 unsigned num_tfeedback_decls = prog->TransformFeedback.NumVarying; 2005 2006 void *mem_ctx = ralloc_context(NULL); // temporary linker context 2007 2008 prog->LinkStatus = false; 2009 prog->Validated = false; 2010 prog->_Used = false; 2011 2012 if (prog->InfoLog != NULL) 2013 ralloc_free(prog->InfoLog); 2014 2015 prog->InfoLog = ralloc_strdup(NULL, ""); 2016 2017 /* Separate the shaders into groups based on their type. 2018 */ 2019 struct gl_shader **vert_shader_list; 2020 unsigned num_vert_shaders = 0; 2021 struct gl_shader **frag_shader_list; 2022 unsigned num_frag_shaders = 0; 2023 2024 vert_shader_list = (struct gl_shader **) 2025 calloc(2 * prog->NumShaders, sizeof(struct gl_shader *)); 2026 frag_shader_list = &vert_shader_list[prog->NumShaders]; 2027 2028 unsigned min_version = UINT_MAX; 2029 unsigned max_version = 0; 2030 for (unsigned i = 0; i < prog->NumShaders; i++) { 2031 min_version = MIN2(min_version, prog->Shaders[i]->Version); 2032 max_version = MAX2(max_version, prog->Shaders[i]->Version); 2033 2034 switch (prog->Shaders[i]->Type) { 2035 case GL_VERTEX_SHADER: 2036 vert_shader_list[num_vert_shaders] = prog->Shaders[i]; 2037 num_vert_shaders++; 2038 break; 2039 case GL_FRAGMENT_SHADER: 2040 frag_shader_list[num_frag_shaders] = prog->Shaders[i]; 2041 num_frag_shaders++; 2042 break; 2043 case GL_GEOMETRY_SHADER: 2044 /* FINISHME: Support geometry shaders. */ 2045 assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER); 2046 break; 2047 } 2048 } 2049 2050 /* Previous to GLSL version 1.30, different compilation units could mix and 2051 * match shading language versions. With GLSL 1.30 and later, the versions 2052 * of all shaders must match. 2053 */ 2054 assert(min_version >= 100); 2055 assert(max_version <= 130); 2056 if ((max_version >= 130 || min_version == 100) 2057 && min_version != max_version) { 2058 linker_error(prog, "all shaders must use same shading " 2059 "language version\n"); 2060 goto done; 2061 } 2062 2063 prog->Version = max_version; 2064 2065 for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) { 2066 if (prog->_LinkedShaders[i] != NULL) 2067 ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]); 2068 2069 prog->_LinkedShaders[i] = NULL; 2070 } 2071 2072 /* Link all shaders for a particular stage and validate the result. 2073 */ 2074 if (num_vert_shaders > 0) { 2075 gl_shader *const sh = 2076 link_intrastage_shaders(mem_ctx, ctx, prog, vert_shader_list, 2077 num_vert_shaders); 2078 2079 if (sh == NULL) 2080 goto done; 2081 2082 if (!validate_vertex_shader_executable(prog, sh)) 2083 goto done; 2084 2085 _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX], 2086 sh); 2087 } 2088 2089 if (num_frag_shaders > 0) { 2090 gl_shader *const sh = 2091 link_intrastage_shaders(mem_ctx, ctx, prog, frag_shader_list, 2092 num_frag_shaders); 2093 2094 if (sh == NULL) 2095 goto done; 2096 2097 if (!validate_fragment_shader_executable(prog, sh)) 2098 goto done; 2099 2100 _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT], 2101 sh); 2102 } 2103 2104 /* Here begins the inter-stage linking phase. Some initial validation is 2105 * performed, then locations are assigned for uniforms, attributes, and 2106 * varyings. 2107 */ 2108 if (cross_validate_uniforms(prog)) { 2109 unsigned prev; 2110 2111 for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { 2112 if (prog->_LinkedShaders[prev] != NULL) 2113 break; 2114 } 2115 2116 /* Validate the inputs of each stage with the output of the preceding 2117 * stage. 2118 */ 2119 for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { 2120 if (prog->_LinkedShaders[i] == NULL) 2121 continue; 2122 2123 if (!cross_validate_outputs_to_inputs(prog, 2124 prog->_LinkedShaders[prev], 2125 prog->_LinkedShaders[i])) 2126 goto done; 2127 2128 prev = i; 2129 } 2130 2131 prog->LinkStatus = true; 2132 } 2133 2134 /* Do common optimization before assigning storage for attributes, 2135 * uniforms, and varyings. Later optimization could possibly make 2136 * some of that unused. 2137 */ 2138 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { 2139 if (prog->_LinkedShaders[i] == NULL) 2140 continue; 2141 2142 detect_recursion_linked(prog, prog->_LinkedShaders[i]->ir); 2143 if (!prog->LinkStatus) 2144 goto done; 2145 2146 if (ctx->ShaderCompilerOptions[i].LowerClipDistance) 2147 lower_clip_distance(prog->_LinkedShaders[i]->ir); 2148 2149 while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, false, 32)) 2150 ; 2151 } 2152 2153 /* FINISHME: The value of the max_attribute_index parameter is 2154 * FINISHME: implementation dependent based on the value of 2155 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be 2156 * FINISHME: at least 16, so hardcode 16 for now. 2157 */ 2158 if (!assign_attribute_or_color_locations(prog, MESA_SHADER_VERTEX, 16)) { 2159 goto done; 2160 } 2161 2162 if (!assign_attribute_or_color_locations(prog, MESA_SHADER_FRAGMENT, ctx->Const.MaxDrawBuffers)) { 2163 goto done; 2164 } 2165 2166 unsigned prev; 2167 for (prev = 0; prev < MESA_SHADER_TYPES; prev++) { 2168 if (prog->_LinkedShaders[prev] != NULL) 2169 break; 2170 } 2171 2172 if (num_tfeedback_decls != 0) { 2173 /* From GL_EXT_transform_feedback: 2174 * A program will fail to link if: 2175 * 2176 * * the <count> specified by TransformFeedbackVaryingsEXT is 2177 * non-zero, but the program object has no vertex or geometry 2178 * shader; 2179 */ 2180 if (prev >= MESA_SHADER_FRAGMENT) { 2181 linker_error(prog, "Transform feedback varyings specified, but " 2182 "no vertex or geometry shader is present."); 2183 goto done; 2184 } 2185 2186 tfeedback_decls = ralloc_array(mem_ctx, tfeedback_decl, 2187 prog->TransformFeedback.NumVarying); 2188 if (!parse_tfeedback_decls(prog, mem_ctx, num_tfeedback_decls, 2189 prog->TransformFeedback.VaryingNames, 2190 tfeedback_decls)) 2191 goto done; 2192 } 2193 2194 for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) { 2195 if (prog->_LinkedShaders[i] == NULL) 2196 continue; 2197 2198 if (!assign_varying_locations( 2199 ctx, prog, prog->_LinkedShaders[prev], prog->_LinkedShaders[i], 2200 i == MESA_SHADER_FRAGMENT ? num_tfeedback_decls : 0, 2201 tfeedback_decls)) 2202 goto done; 2203 2204 prev = i; 2205 } 2206 2207 if (prev != MESA_SHADER_FRAGMENT && num_tfeedback_decls != 0) { 2208 /* There was no fragment shader, but we still have to assign varying 2209 * locations for use by transform feedback. 2210 */ 2211 if (!assign_varying_locations( 2212 ctx, prog, prog->_LinkedShaders[prev], NULL, num_tfeedback_decls, 2213 tfeedback_decls)) 2214 goto done; 2215 } 2216 2217 if (!store_tfeedback_info(ctx, prog, num_tfeedback_decls, tfeedback_decls)) 2218 goto done; 2219 2220 if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) { 2221 demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX], 2222 ir_var_out); 2223 2224 /* Eliminate code that is now dead due to unused vertex outputs being 2225 * demoted. 2226 */ 2227 while (do_dead_code(prog->_LinkedShaders[MESA_SHADER_VERTEX]->ir, false)) 2228 ; 2229 } 2230 2231 if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) { 2232 gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY]; 2233 2234 demote_shader_inputs_and_outputs(sh, ir_var_in); 2235 demote_shader_inputs_and_outputs(sh, ir_var_inout); 2236 demote_shader_inputs_and_outputs(sh, ir_var_out); 2237 2238 /* Eliminate code that is now dead due to unused geometry outputs being 2239 * demoted. 2240 */ 2241 while (do_dead_code(prog->_LinkedShaders[MESA_SHADER_GEOMETRY]->ir, false)) 2242 ; 2243 } 2244 2245 if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) { 2246 gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT]; 2247 2248 demote_shader_inputs_and_outputs(sh, ir_var_in); 2249 2250 /* Eliminate code that is now dead due to unused fragment inputs being 2251 * demoted. This shouldn't actually do anything other than remove 2252 * declarations of the (now unused) global variables. 2253 */ 2254 while (do_dead_code(prog->_LinkedShaders[MESA_SHADER_FRAGMENT]->ir, false)) 2255 ; 2256 } 2257 2258 update_array_sizes(prog); 2259 link_assign_uniform_locations(prog); 2260 store_fragdepth_layout(prog); 2261 2262 if (!check_resources(ctx, prog)) 2263 goto done; 2264 2265 /* OpenGL ES requires that a vertex shader and a fragment shader both be 2266 * present in a linked program. By checking for use of shading language 2267 * version 1.00, we also catch the GL_ARB_ES2_compatibility case. 2268 */ 2269 if (!prog->InternalSeparateShader && 2270 (ctx->API == API_OPENGLES2 || prog->Version == 100)) { 2271 if (prog->_LinkedShaders[MESA_SHADER_VERTEX] == NULL) { 2272 linker_error(prog, "program lacks a vertex shader\n"); 2273 } else if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] == NULL) { 2274 linker_error(prog, "program lacks a fragment shader\n"); 2275 } 2276 } 2277 2278 /* FINISHME: Assign fragment shader output locations. */ 2279 2280done: 2281 free(vert_shader_list); 2282 2283 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) { 2284 if (prog->_LinkedShaders[i] == NULL) 2285 continue; 2286 2287 /* Retain any live IR, but trash the rest. */ 2288 reparent_ir(prog->_LinkedShaders[i]->ir, prog->_LinkedShaders[i]->ir); 2289 2290 /* The symbol table in the linked shaders may contain references to 2291 * variables that were removed (e.g., unused uniforms). Since it may 2292 * contain junk, there is no possible valid use. Delete it and set the 2293 * pointer to NULL. 2294 */ 2295 delete prog->_LinkedShaders[i]->symbols; 2296 prog->_LinkedShaders[i]->symbols = NULL; 2297 } 2298 2299 ralloc_free(mem_ctx); 2300} 2301