linker: Improve handling of unread/unwritten shader inputs/outputs
[mesa.git] / src / glsl / linker.cpp
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 #include <cstdlib>
67 #include <cstdio>
68 #include <cstdarg>
69 #include <climits>
70
71 extern "C" {
72 #include <talloc.h>
73 }
74
75 #include "main/core.h"
76 #include "glsl_symbol_table.h"
77 #include "ir.h"
78 #include "program.h"
79 #include "program/hash_table.h"
80 #include "linker.h"
81 #include "ir_optimization.h"
82
83 extern "C" {
84 #include "main/shaderobj.h"
85 }
86
87 /**
88 * Visitor that determines whether or not a variable is ever written.
89 */
90 class find_assignment_visitor : public ir_hierarchical_visitor {
91 public:
92 find_assignment_visitor(const char *name)
93 : name(name), found(false)
94 {
95 /* empty */
96 }
97
98 virtual ir_visitor_status visit_enter(ir_assignment *ir)
99 {
100 ir_variable *const var = ir->lhs->variable_referenced();
101
102 if (strcmp(name, var->name) == 0) {
103 found = true;
104 return visit_stop;
105 }
106
107 return visit_continue_with_parent;
108 }
109
110 virtual ir_visitor_status visit_enter(ir_call *ir)
111 {
112 exec_list_iterator sig_iter = ir->get_callee()->parameters.iterator();
113 foreach_iter(exec_list_iterator, iter, *ir) {
114 ir_rvalue *param_rval = (ir_rvalue *)iter.get();
115 ir_variable *sig_param = (ir_variable *)sig_iter.get();
116
117 if (sig_param->mode == ir_var_out ||
118 sig_param->mode == ir_var_inout) {
119 ir_variable *var = param_rval->variable_referenced();
120 if (var && strcmp(name, var->name) == 0) {
121 found = true;
122 return visit_stop;
123 }
124 }
125 sig_iter.next();
126 }
127
128 return visit_continue_with_parent;
129 }
130
131 bool variable_found()
132 {
133 return found;
134 }
135
136 private:
137 const char *name; /**< Find writes to a variable with this name. */
138 bool found; /**< Was a write to the variable found? */
139 };
140
141
142 /**
143 * Visitor that determines whether or not a variable is ever read.
144 */
145 class find_deref_visitor : public ir_hierarchical_visitor {
146 public:
147 find_deref_visitor(const char *name)
148 : name(name), found(false)
149 {
150 /* empty */
151 }
152
153 virtual ir_visitor_status visit(ir_dereference_variable *ir)
154 {
155 if (strcmp(this->name, ir->var->name) == 0) {
156 this->found = true;
157 return visit_stop;
158 }
159
160 return visit_continue;
161 }
162
163 bool variable_found() const
164 {
165 return this->found;
166 }
167
168 private:
169 const char *name; /**< Find writes to a variable with this name. */
170 bool found; /**< Was a write to the variable found? */
171 };
172
173
174 void
175 linker_error_printf(gl_shader_program *prog, const char *fmt, ...)
176 {
177 va_list ap;
178
179 prog->InfoLog = talloc_strdup_append(prog->InfoLog, "error: ");
180 va_start(ap, fmt);
181 prog->InfoLog = talloc_vasprintf_append(prog->InfoLog, fmt, ap);
182 va_end(ap);
183 }
184
185
186 void
187 invalidate_variable_locations(gl_shader *sh, enum ir_variable_mode mode,
188 int generic_base)
189 {
190 foreach_list(node, sh->ir) {
191 ir_variable *const var = ((ir_instruction *) node)->as_variable();
192
193 if ((var == NULL) || (var->mode != (unsigned) mode))
194 continue;
195
196 /* Only assign locations for generic attributes / varyings / etc.
197 */
198 if ((var->location >= generic_base) && !var->explicit_location)
199 var->location = -1;
200 }
201 }
202
203
204 /**
205 * Determine the number of attribute slots required for a particular type
206 *
207 * This code is here because it implements the language rules of a specific
208 * GLSL version. Since it's a property of the language and not a property of
209 * types in general, it doesn't really belong in glsl_type.
210 */
211 unsigned
212 count_attribute_slots(const glsl_type *t)
213 {
214 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
215 *
216 * "A scalar input counts the same amount against this limit as a vec4,
217 * so applications may want to consider packing groups of four
218 * unrelated float inputs together into a vector to better utilize the
219 * capabilities of the underlying hardware. A matrix input will use up
220 * multiple locations. The number of locations used will equal the
221 * number of columns in the matrix."
222 *
223 * The spec does not explicitly say how arrays are counted. However, it
224 * should be safe to assume the total number of slots consumed by an array
225 * is the number of entries in the array multiplied by the number of slots
226 * consumed by a single element of the array.
227 */
228
229 if (t->is_array())
230 return t->array_size() * count_attribute_slots(t->element_type());
231
232 if (t->is_matrix())
233 return t->matrix_columns;
234
235 return 1;
236 }
237
238
239 /**
240 * Verify that a vertex shader executable meets all semantic requirements
241 *
242 * \param shader Vertex shader executable to be verified
243 */
244 bool
245 validate_vertex_shader_executable(struct gl_shader_program *prog,
246 struct gl_shader *shader)
247 {
248 if (shader == NULL)
249 return true;
250
251 find_assignment_visitor find("gl_Position");
252 find.run(shader->ir);
253 if (!find.variable_found()) {
254 linker_error_printf(prog,
255 "vertex shader does not write to `gl_Position'\n");
256 return false;
257 }
258
259 return true;
260 }
261
262
263 /**
264 * Verify that a fragment shader executable meets all semantic requirements
265 *
266 * \param shader Fragment shader executable to be verified
267 */
268 bool
269 validate_fragment_shader_executable(struct gl_shader_program *prog,
270 struct gl_shader *shader)
271 {
272 if (shader == NULL)
273 return true;
274
275 find_assignment_visitor frag_color("gl_FragColor");
276 find_assignment_visitor frag_data("gl_FragData");
277
278 frag_color.run(shader->ir);
279 frag_data.run(shader->ir);
280
281 if (frag_color.variable_found() && frag_data.variable_found()) {
282 linker_error_printf(prog, "fragment shader writes to both "
283 "`gl_FragColor' and `gl_FragData'\n");
284 return false;
285 }
286
287 return true;
288 }
289
290
291 /**
292 * Generate a string describing the mode of a variable
293 */
294 static const char *
295 mode_string(const ir_variable *var)
296 {
297 switch (var->mode) {
298 case ir_var_auto:
299 return (var->read_only) ? "global constant" : "global variable";
300
301 case ir_var_uniform: return "uniform";
302 case ir_var_in: return "shader input";
303 case ir_var_out: return "shader output";
304 case ir_var_inout: return "shader inout";
305
306 case ir_var_temporary:
307 default:
308 assert(!"Should not get here.");
309 return "invalid variable";
310 }
311 }
312
313
314 /**
315 * Perform validation of global variables used across multiple shaders
316 */
317 bool
318 cross_validate_globals(struct gl_shader_program *prog,
319 struct gl_shader **shader_list,
320 unsigned num_shaders,
321 bool uniforms_only)
322 {
323 /* Examine all of the uniforms in all of the shaders and cross validate
324 * them.
325 */
326 glsl_symbol_table variables;
327 for (unsigned i = 0; i < num_shaders; i++) {
328 if (shader_list[i] == NULL)
329 continue;
330
331 foreach_list(node, shader_list[i]->ir) {
332 ir_variable *const var = ((ir_instruction *) node)->as_variable();
333
334 if (var == NULL)
335 continue;
336
337 if (uniforms_only && (var->mode != ir_var_uniform))
338 continue;
339
340 /* Don't cross validate temporaries that are at global scope. These
341 * will eventually get pulled into the shaders 'main'.
342 */
343 if (var->mode == ir_var_temporary)
344 continue;
345
346 /* If a global with this name has already been seen, verify that the
347 * new instance has the same type. In addition, if the globals have
348 * initializers, the values of the initializers must be the same.
349 */
350 ir_variable *const existing = variables.get_variable(var->name);
351 if (existing != NULL) {
352 if (var->type != existing->type) {
353 /* Consider the types to be "the same" if both types are arrays
354 * of the same type and one of the arrays is implicitly sized.
355 * In addition, set the type of the linked variable to the
356 * explicitly sized array.
357 */
358 if (var->type->is_array()
359 && existing->type->is_array()
360 && (var->type->fields.array == existing->type->fields.array)
361 && ((var->type->length == 0)
362 || (existing->type->length == 0))) {
363 if (existing->type->length == 0)
364 existing->type = var->type;
365 } else {
366 linker_error_printf(prog, "%s `%s' declared as type "
367 "`%s' and type `%s'\n",
368 mode_string(var),
369 var->name, var->type->name,
370 existing->type->name);
371 return false;
372 }
373 }
374
375 if (var->explicit_location) {
376 if (existing->explicit_location
377 && (var->location != existing->location)) {
378 linker_error_printf(prog, "explicit locations for %s "
379 "`%s' have differing values\n",
380 mode_string(var), var->name);
381 return false;
382 }
383
384 existing->location = var->location;
385 existing->explicit_location = true;
386 }
387
388 /* FINISHME: Handle non-constant initializers.
389 */
390 if (var->constant_value != NULL) {
391 if (existing->constant_value != NULL) {
392 if (!var->constant_value->has_value(existing->constant_value)) {
393 linker_error_printf(prog, "initializers for %s "
394 "`%s' have differing values\n",
395 mode_string(var), var->name);
396 return false;
397 }
398 } else
399 /* If the first-seen instance of a particular uniform did not
400 * have an initializer but a later instance does, copy the
401 * initializer to the version stored in the symbol table.
402 */
403 /* FINISHME: This is wrong. The constant_value field should
404 * FINISHME: not be modified! Imagine a case where a shader
405 * FINISHME: without an initializer is linked in two different
406 * FINISHME: programs with shaders that have differing
407 * FINISHME: initializers. Linking with the first will
408 * FINISHME: modify the shader, and linking with the second
409 * FINISHME: will fail.
410 */
411 existing->constant_value =
412 var->constant_value->clone(talloc_parent(existing), NULL);
413 }
414 } else
415 variables.add_variable(var->name, var);
416 }
417 }
418
419 return true;
420 }
421
422
423 /**
424 * Perform validation of uniforms used across multiple shader stages
425 */
426 bool
427 cross_validate_uniforms(struct gl_shader_program *prog)
428 {
429 return cross_validate_globals(prog, prog->_LinkedShaders,
430 MESA_SHADER_TYPES, true);
431 }
432
433
434 /**
435 * Validate that outputs from one stage match inputs of another
436 */
437 bool
438 cross_validate_outputs_to_inputs(struct gl_shader_program *prog,
439 gl_shader *producer, gl_shader *consumer)
440 {
441 glsl_symbol_table parameters;
442 /* FINISHME: Figure these out dynamically. */
443 const char *const producer_stage = "vertex";
444 const char *const consumer_stage = "fragment";
445
446 /* Find all shader outputs in the "producer" stage.
447 */
448 foreach_list(node, producer->ir) {
449 ir_variable *const var = ((ir_instruction *) node)->as_variable();
450
451 /* FINISHME: For geometry shaders, this should also look for inout
452 * FINISHME: variables.
453 */
454 if ((var == NULL) || (var->mode != ir_var_out))
455 continue;
456
457 parameters.add_variable(var->name, var);
458 }
459
460
461 /* Find all shader inputs in the "consumer" stage. Any variables that have
462 * matching outputs already in the symbol table must have the same type and
463 * qualifiers.
464 */
465 foreach_list(node, consumer->ir) {
466 ir_variable *const input = ((ir_instruction *) node)->as_variable();
467
468 /* FINISHME: For geometry shaders, this should also look for inout
469 * FINISHME: variables.
470 */
471 if ((input == NULL) || (input->mode != ir_var_in))
472 continue;
473
474 ir_variable *const output = parameters.get_variable(input->name);
475 if (output != NULL) {
476 /* Check that the types match between stages.
477 */
478 if (input->type != output->type) {
479 linker_error_printf(prog,
480 "%s shader output `%s' declared as "
481 "type `%s', but %s shader input declared "
482 "as type `%s'\n",
483 producer_stage, output->name,
484 output->type->name,
485 consumer_stage, input->type->name);
486 return false;
487 }
488
489 /* Check that all of the qualifiers match between stages.
490 */
491 if (input->centroid != output->centroid) {
492 linker_error_printf(prog,
493 "%s shader output `%s' %s centroid qualifier, "
494 "but %s shader input %s centroid qualifier\n",
495 producer_stage,
496 output->name,
497 (output->centroid) ? "has" : "lacks",
498 consumer_stage,
499 (input->centroid) ? "has" : "lacks");
500 return false;
501 }
502
503 if (input->invariant != output->invariant) {
504 linker_error_printf(prog,
505 "%s shader output `%s' %s invariant qualifier, "
506 "but %s shader input %s invariant qualifier\n",
507 producer_stage,
508 output->name,
509 (output->invariant) ? "has" : "lacks",
510 consumer_stage,
511 (input->invariant) ? "has" : "lacks");
512 return false;
513 }
514
515 if (input->interpolation != output->interpolation) {
516 linker_error_printf(prog,
517 "%s shader output `%s' specifies %s "
518 "interpolation qualifier, "
519 "but %s shader input specifies %s "
520 "interpolation qualifier\n",
521 producer_stage,
522 output->name,
523 output->interpolation_string(),
524 consumer_stage,
525 input->interpolation_string());
526 return false;
527 }
528 }
529 }
530
531 return true;
532 }
533
534
535 /**
536 * Populates a shaders symbol table with all global declarations
537 */
538 static void
539 populate_symbol_table(gl_shader *sh)
540 {
541 sh->symbols = new(sh) glsl_symbol_table;
542
543 foreach_list(node, sh->ir) {
544 ir_instruction *const inst = (ir_instruction *) node;
545 ir_variable *var;
546 ir_function *func;
547
548 if ((func = inst->as_function()) != NULL) {
549 sh->symbols->add_function(func->name, func);
550 } else if ((var = inst->as_variable()) != NULL) {
551 sh->symbols->add_variable(var->name, var);
552 }
553 }
554 }
555
556
557 /**
558 * Remap variables referenced in an instruction tree
559 *
560 * This is used when instruction trees are cloned from one shader and placed in
561 * another. These trees will contain references to \c ir_variable nodes that
562 * do not exist in the target shader. This function finds these \c ir_variable
563 * references and replaces the references with matching variables in the target
564 * shader.
565 *
566 * If there is no matching variable in the target shader, a clone of the
567 * \c ir_variable is made and added to the target shader. The new variable is
568 * added to \b both the instruction stream and the symbol table.
569 *
570 * \param inst IR tree that is to be processed.
571 * \param symbols Symbol table containing global scope symbols in the
572 * linked shader.
573 * \param instructions Instruction stream where new variable declarations
574 * should be added.
575 */
576 void
577 remap_variables(ir_instruction *inst, struct gl_shader *target,
578 hash_table *temps)
579 {
580 class remap_visitor : public ir_hierarchical_visitor {
581 public:
582 remap_visitor(struct gl_shader *target,
583 hash_table *temps)
584 {
585 this->target = target;
586 this->symbols = target->symbols;
587 this->instructions = target->ir;
588 this->temps = temps;
589 }
590
591 virtual ir_visitor_status visit(ir_dereference_variable *ir)
592 {
593 if (ir->var->mode == ir_var_temporary) {
594 ir_variable *var = (ir_variable *) hash_table_find(temps, ir->var);
595
596 assert(var != NULL);
597 ir->var = var;
598 return visit_continue;
599 }
600
601 ir_variable *const existing =
602 this->symbols->get_variable(ir->var->name);
603 if (existing != NULL)
604 ir->var = existing;
605 else {
606 ir_variable *copy = ir->var->clone(this->target, NULL);
607
608 this->symbols->add_variable(copy->name, copy);
609 this->instructions->push_head(copy);
610 ir->var = copy;
611 }
612
613 return visit_continue;
614 }
615
616 private:
617 struct gl_shader *target;
618 glsl_symbol_table *symbols;
619 exec_list *instructions;
620 hash_table *temps;
621 };
622
623 remap_visitor v(target, temps);
624
625 inst->accept(&v);
626 }
627
628
629 /**
630 * Move non-declarations from one instruction stream to another
631 *
632 * The intended usage pattern of this function is to pass the pointer to the
633 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
634 * pointer) for \c last and \c false for \c make_copies on the first
635 * call. Successive calls pass the return value of the previous call for
636 * \c last and \c true for \c make_copies.
637 *
638 * \param instructions Source instruction stream
639 * \param last Instruction after which new instructions should be
640 * inserted in the target instruction stream
641 * \param make_copies Flag selecting whether instructions in \c instructions
642 * should be copied (via \c ir_instruction::clone) into the
643 * target list or moved.
644 *
645 * \return
646 * The new "last" instruction in the target instruction stream. This pointer
647 * is suitable for use as the \c last parameter of a later call to this
648 * function.
649 */
650 exec_node *
651 move_non_declarations(exec_list *instructions, exec_node *last,
652 bool make_copies, gl_shader *target)
653 {
654 hash_table *temps = NULL;
655
656 if (make_copies)
657 temps = hash_table_ctor(0, hash_table_pointer_hash,
658 hash_table_pointer_compare);
659
660 foreach_list_safe(node, instructions) {
661 ir_instruction *inst = (ir_instruction *) node;
662
663 if (inst->as_function())
664 continue;
665
666 ir_variable *var = inst->as_variable();
667 if ((var != NULL) && (var->mode != ir_var_temporary))
668 continue;
669
670 assert(inst->as_assignment()
671 || ((var != NULL) && (var->mode == ir_var_temporary)));
672
673 if (make_copies) {
674 inst = inst->clone(target, NULL);
675
676 if (var != NULL)
677 hash_table_insert(temps, inst, var);
678 else
679 remap_variables(inst, target, temps);
680 } else {
681 inst->remove();
682 }
683
684 last->insert_after(inst);
685 last = inst;
686 }
687
688 if (make_copies)
689 hash_table_dtor(temps);
690
691 return last;
692 }
693
694 /**
695 * Get the function signature for main from a shader
696 */
697 static ir_function_signature *
698 get_main_function_signature(gl_shader *sh)
699 {
700 ir_function *const f = sh->symbols->get_function("main");
701 if (f != NULL) {
702 exec_list void_parameters;
703
704 /* Look for the 'void main()' signature and ensure that it's defined.
705 * This keeps the linker from accidentally pick a shader that just
706 * contains a prototype for main.
707 *
708 * We don't have to check for multiple definitions of main (in multiple
709 * shaders) because that would have already been caught above.
710 */
711 ir_function_signature *sig = f->matching_signature(&void_parameters);
712 if ((sig != NULL) && sig->is_defined) {
713 return sig;
714 }
715 }
716
717 return NULL;
718 }
719
720
721 /**
722 * Combine a group of shaders for a single stage to generate a linked shader
723 *
724 * \note
725 * If this function is supplied a single shader, it is cloned, and the new
726 * shader is returned.
727 */
728 static struct gl_shader *
729 link_intrastage_shaders(struct gl_context *ctx,
730 struct gl_shader_program *prog,
731 struct gl_shader **shader_list,
732 unsigned num_shaders)
733 {
734 /* Check that global variables defined in multiple shaders are consistent.
735 */
736 if (!cross_validate_globals(prog, shader_list, num_shaders, false))
737 return NULL;
738
739 /* Check that there is only a single definition of each function signature
740 * across all shaders.
741 */
742 for (unsigned i = 0; i < (num_shaders - 1); i++) {
743 foreach_list(node, shader_list[i]->ir) {
744 ir_function *const f = ((ir_instruction *) node)->as_function();
745
746 if (f == NULL)
747 continue;
748
749 for (unsigned j = i + 1; j < num_shaders; j++) {
750 ir_function *const other =
751 shader_list[j]->symbols->get_function(f->name);
752
753 /* If the other shader has no function (and therefore no function
754 * signatures) with the same name, skip to the next shader.
755 */
756 if (other == NULL)
757 continue;
758
759 foreach_iter (exec_list_iterator, iter, *f) {
760 ir_function_signature *sig =
761 (ir_function_signature *) iter.get();
762
763 if (!sig->is_defined || sig->is_builtin)
764 continue;
765
766 ir_function_signature *other_sig =
767 other->exact_matching_signature(& sig->parameters);
768
769 if ((other_sig != NULL) && other_sig->is_defined
770 && !other_sig->is_builtin) {
771 linker_error_printf(prog,
772 "function `%s' is multiply defined",
773 f->name);
774 return NULL;
775 }
776 }
777 }
778 }
779 }
780
781 /* Find the shader that defines main, and make a clone of it.
782 *
783 * Starting with the clone, search for undefined references. If one is
784 * found, find the shader that defines it. Clone the reference and add
785 * it to the shader. Repeat until there are no undefined references or
786 * until a reference cannot be resolved.
787 */
788 gl_shader *main = NULL;
789 for (unsigned i = 0; i < num_shaders; i++) {
790 if (get_main_function_signature(shader_list[i]) != NULL) {
791 main = shader_list[i];
792 break;
793 }
794 }
795
796 if (main == NULL) {
797 linker_error_printf(prog, "%s shader lacks `main'\n",
798 (shader_list[0]->Type == GL_VERTEX_SHADER)
799 ? "vertex" : "fragment");
800 return NULL;
801 }
802
803 gl_shader *linked = ctx->Driver.NewShader(NULL, 0, main->Type);
804 linked->ir = new(linked) exec_list;
805 clone_ir_list(linked, linked->ir, main->ir);
806
807 populate_symbol_table(linked);
808
809 /* The a pointer to the main function in the final linked shader (i.e., the
810 * copy of the original shader that contained the main function).
811 */
812 ir_function_signature *const main_sig = get_main_function_signature(linked);
813
814 /* Move any instructions other than variable declarations or function
815 * declarations into main.
816 */
817 exec_node *insertion_point =
818 move_non_declarations(linked->ir, (exec_node *) &main_sig->body, false,
819 linked);
820
821 for (unsigned i = 0; i < num_shaders; i++) {
822 if (shader_list[i] == main)
823 continue;
824
825 insertion_point = move_non_declarations(shader_list[i]->ir,
826 insertion_point, true, linked);
827 }
828
829 /* Resolve initializers for global variables in the linked shader.
830 */
831 unsigned num_linking_shaders = num_shaders;
832 for (unsigned i = 0; i < num_shaders; i++)
833 num_linking_shaders += shader_list[i]->num_builtins_to_link;
834
835 gl_shader **linking_shaders =
836 (gl_shader **) calloc(num_linking_shaders, sizeof(gl_shader *));
837
838 memcpy(linking_shaders, shader_list,
839 sizeof(linking_shaders[0]) * num_shaders);
840
841 unsigned idx = num_shaders;
842 for (unsigned i = 0; i < num_shaders; i++) {
843 memcpy(&linking_shaders[idx], shader_list[i]->builtins_to_link,
844 sizeof(linking_shaders[0]) * shader_list[i]->num_builtins_to_link);
845 idx += shader_list[i]->num_builtins_to_link;
846 }
847
848 assert(idx == num_linking_shaders);
849
850 if (!link_function_calls(prog, linked, linking_shaders,
851 num_linking_shaders)) {
852 ctx->Driver.DeleteShader(ctx, linked);
853 linked = NULL;
854 }
855
856 free(linking_shaders);
857
858 return linked;
859 }
860
861
862 struct uniform_node {
863 exec_node link;
864 struct gl_uniform *u;
865 unsigned slots;
866 };
867
868 /**
869 * Update the sizes of linked shader uniform arrays to the maximum
870 * array index used.
871 *
872 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
873 *
874 * If one or more elements of an array are active,
875 * GetActiveUniform will return the name of the array in name,
876 * subject to the restrictions listed above. The type of the array
877 * is returned in type. The size parameter contains the highest
878 * array element index used, plus one. The compiler or linker
879 * determines the highest index used. There will be only one
880 * active uniform reported by the GL per uniform array.
881
882 */
883 static void
884 update_array_sizes(struct gl_shader_program *prog)
885 {
886 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
887 if (prog->_LinkedShaders[i] == NULL)
888 continue;
889
890 foreach_list(node, prog->_LinkedShaders[i]->ir) {
891 ir_variable *const var = ((ir_instruction *) node)->as_variable();
892
893 if ((var == NULL) || (var->mode != ir_var_uniform &&
894 var->mode != ir_var_in &&
895 var->mode != ir_var_out) ||
896 !var->type->is_array())
897 continue;
898
899 unsigned int size = var->max_array_access;
900 for (unsigned j = 0; j < MESA_SHADER_TYPES; j++) {
901 if (prog->_LinkedShaders[j] == NULL)
902 continue;
903
904 foreach_list(node2, prog->_LinkedShaders[j]->ir) {
905 ir_variable *other_var = ((ir_instruction *) node2)->as_variable();
906 if (!other_var)
907 continue;
908
909 if (strcmp(var->name, other_var->name) == 0 &&
910 other_var->max_array_access > size) {
911 size = other_var->max_array_access;
912 }
913 }
914 }
915
916 if (size + 1 != var->type->fields.array->length) {
917 var->type = glsl_type::get_array_instance(var->type->fields.array,
918 size + 1);
919 /* FINISHME: We should update the types of array
920 * dereferences of this variable now.
921 */
922 }
923 }
924 }
925 }
926
927 static void
928 add_uniform(void *mem_ctx, exec_list *uniforms, struct hash_table *ht,
929 const char *name, const glsl_type *type, GLenum shader_type,
930 unsigned *next_shader_pos, unsigned *total_uniforms)
931 {
932 if (type->is_record()) {
933 for (unsigned int i = 0; i < type->length; i++) {
934 const glsl_type *field_type = type->fields.structure[i].type;
935 char *field_name = talloc_asprintf(mem_ctx, "%s.%s", name,
936 type->fields.structure[i].name);
937
938 add_uniform(mem_ctx, uniforms, ht, field_name, field_type,
939 shader_type, next_shader_pos, total_uniforms);
940 }
941 } else {
942 uniform_node *n = (uniform_node *) hash_table_find(ht, name);
943 unsigned int vec4_slots;
944 const glsl_type *array_elem_type = NULL;
945
946 if (type->is_array()) {
947 array_elem_type = type->fields.array;
948 /* Array of structures. */
949 if (array_elem_type->is_record()) {
950 for (unsigned int i = 0; i < type->length; i++) {
951 char *elem_name = talloc_asprintf(mem_ctx, "%s[%d]", name, i);
952 add_uniform(mem_ctx, uniforms, ht, elem_name, array_elem_type,
953 shader_type, next_shader_pos, total_uniforms);
954 }
955 return;
956 }
957 }
958
959 /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
960 * vectors to vec4 slots.
961 */
962 if (type->is_array()) {
963 if (array_elem_type->is_sampler())
964 vec4_slots = type->length;
965 else
966 vec4_slots = type->length * array_elem_type->matrix_columns;
967 } else if (type->is_sampler()) {
968 vec4_slots = 1;
969 } else {
970 vec4_slots = type->matrix_columns;
971 }
972
973 if (n == NULL) {
974 n = (uniform_node *) calloc(1, sizeof(struct uniform_node));
975 n->u = (gl_uniform *) calloc(1, sizeof(struct gl_uniform));
976 n->slots = vec4_slots;
977
978 n->u->Name = strdup(name);
979 n->u->Type = type;
980 n->u->VertPos = -1;
981 n->u->FragPos = -1;
982 n->u->GeomPos = -1;
983 (*total_uniforms)++;
984
985 hash_table_insert(ht, n, name);
986 uniforms->push_tail(& n->link);
987 }
988
989 switch (shader_type) {
990 case GL_VERTEX_SHADER:
991 n->u->VertPos = *next_shader_pos;
992 break;
993 case GL_FRAGMENT_SHADER:
994 n->u->FragPos = *next_shader_pos;
995 break;
996 case GL_GEOMETRY_SHADER:
997 n->u->GeomPos = *next_shader_pos;
998 break;
999 }
1000
1001 (*next_shader_pos) += vec4_slots;
1002 }
1003 }
1004
1005 void
1006 assign_uniform_locations(struct gl_shader_program *prog)
1007 {
1008 /* */
1009 exec_list uniforms;
1010 unsigned total_uniforms = 0;
1011 hash_table *ht = hash_table_ctor(32, hash_table_string_hash,
1012 hash_table_string_compare);
1013 void *mem_ctx = talloc_new(NULL);
1014
1015 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
1016 if (prog->_LinkedShaders[i] == NULL)
1017 continue;
1018
1019 unsigned next_position = 0;
1020
1021 foreach_list(node, prog->_LinkedShaders[i]->ir) {
1022 ir_variable *const var = ((ir_instruction *) node)->as_variable();
1023
1024 if ((var == NULL) || (var->mode != ir_var_uniform))
1025 continue;
1026
1027 if (strncmp(var->name, "gl_", 3) == 0) {
1028 /* At the moment, we don't allocate uniform locations for
1029 * builtin uniforms. It's permitted by spec, and we'll
1030 * likely switch to doing that at some point, but not yet.
1031 */
1032 continue;
1033 }
1034
1035 var->location = next_position;
1036 add_uniform(mem_ctx, &uniforms, ht, var->name, var->type,
1037 prog->_LinkedShaders[i]->Type,
1038 &next_position, &total_uniforms);
1039 }
1040 }
1041
1042 talloc_free(mem_ctx);
1043
1044 gl_uniform_list *ul = (gl_uniform_list *)
1045 calloc(1, sizeof(gl_uniform_list));
1046
1047 ul->Size = total_uniforms;
1048 ul->NumUniforms = total_uniforms;
1049 ul->Uniforms = (gl_uniform *) calloc(total_uniforms, sizeof(gl_uniform));
1050
1051 unsigned idx = 0;
1052 uniform_node *next;
1053 for (uniform_node *node = (uniform_node *) uniforms.head
1054 ; node->link.next != NULL
1055 ; node = next) {
1056 next = (uniform_node *) node->link.next;
1057
1058 node->link.remove();
1059 memcpy(&ul->Uniforms[idx], node->u, sizeof(gl_uniform));
1060 idx++;
1061
1062 free(node->u);
1063 free(node);
1064 }
1065
1066 hash_table_dtor(ht);
1067
1068 prog->Uniforms = ul;
1069 }
1070
1071
1072 /**
1073 * Find a contiguous set of available bits in a bitmask
1074 *
1075 * \param used_mask Bits representing used (1) and unused (0) locations
1076 * \param needed_count Number of contiguous bits needed.
1077 *
1078 * \return
1079 * Base location of the available bits on success or -1 on failure.
1080 */
1081 int
1082 find_available_slots(unsigned used_mask, unsigned needed_count)
1083 {
1084 unsigned needed_mask = (1 << needed_count) - 1;
1085 const int max_bit_to_test = (8 * sizeof(used_mask)) - needed_count;
1086
1087 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1088 * cannot optimize possibly infinite loops" for the loop below.
1089 */
1090 if ((needed_count == 0) || (max_bit_to_test < 0) || (max_bit_to_test > 32))
1091 return -1;
1092
1093 for (int i = 0; i <= max_bit_to_test; i++) {
1094 if ((needed_mask & ~used_mask) == needed_mask)
1095 return i;
1096
1097 needed_mask <<= 1;
1098 }
1099
1100 return -1;
1101 }
1102
1103
1104 bool
1105 assign_attribute_locations(gl_shader_program *prog, unsigned max_attribute_index)
1106 {
1107 /* Mark invalid attribute locations as being used.
1108 */
1109 unsigned used_locations = (max_attribute_index >= 32)
1110 ? ~0 : ~((1 << max_attribute_index) - 1);
1111
1112 gl_shader *const sh = prog->_LinkedShaders[0];
1113 assert(sh->Type == GL_VERTEX_SHADER);
1114
1115 /* Operate in a total of four passes.
1116 *
1117 * 1. Invalidate the location assignments for all vertex shader inputs.
1118 *
1119 * 2. Assign locations for inputs that have user-defined (via
1120 * glBindVertexAttribLocation) locatoins.
1121 *
1122 * 3. Sort the attributes without assigned locations by number of slots
1123 * required in decreasing order. Fragmentation caused by attribute
1124 * locations assigned by the application may prevent large attributes
1125 * from having enough contiguous space.
1126 *
1127 * 4. Assign locations to any inputs without assigned locations.
1128 */
1129
1130 invalidate_variable_locations(sh, ir_var_in, VERT_ATTRIB_GENERIC0);
1131
1132 if (prog->Attributes != NULL) {
1133 for (unsigned i = 0; i < prog->Attributes->NumParameters; i++) {
1134 ir_variable *const var =
1135 sh->symbols->get_variable(prog->Attributes->Parameters[i].Name);
1136
1137 /* Note: attributes that occupy multiple slots, such as arrays or
1138 * matrices, may appear in the attrib array multiple times.
1139 */
1140 if ((var == NULL) || (var->location != -1))
1141 continue;
1142
1143 /* From page 61 of the OpenGL 4.0 spec:
1144 *
1145 * "LinkProgram will fail if the attribute bindings assigned by
1146 * BindAttribLocation do not leave not enough space to assign a
1147 * location for an active matrix attribute or an active attribute
1148 * array, both of which require multiple contiguous generic
1149 * attributes."
1150 *
1151 * Previous versions of the spec contain similar language but omit the
1152 * bit about attribute arrays.
1153 *
1154 * Page 61 of the OpenGL 4.0 spec also says:
1155 *
1156 * "It is possible for an application to bind more than one
1157 * attribute name to the same location. This is referred to as
1158 * aliasing. This will only work if only one of the aliased
1159 * attributes is active in the executable program, or if no path
1160 * through the shader consumes more than one attribute of a set
1161 * of attributes aliased to the same location. A link error can
1162 * occur if the linker determines that every path through the
1163 * shader consumes multiple aliased attributes, but
1164 * implementations are not required to generate an error in this
1165 * case."
1166 *
1167 * These two paragraphs are either somewhat contradictory, or I don't
1168 * fully understand one or both of them.
1169 */
1170 /* FINISHME: The code as currently written does not support attribute
1171 * FINISHME: location aliasing (see comment above).
1172 */
1173 const int attr = prog->Attributes->Parameters[i].StateIndexes[0];
1174 const unsigned slots = count_attribute_slots(var->type);
1175
1176 /* Mask representing the contiguous slots that will be used by this
1177 * attribute.
1178 */
1179 const unsigned use_mask = (1 << slots) - 1;
1180
1181 /* Generate a link error if the set of bits requested for this
1182 * attribute overlaps any previously allocated bits.
1183 */
1184 if ((~(use_mask << attr) & used_locations) != used_locations) {
1185 linker_error_printf(prog,
1186 "insufficient contiguous attribute locations "
1187 "available for vertex shader input `%s'",
1188 var->name);
1189 return false;
1190 }
1191
1192 var->location = VERT_ATTRIB_GENERIC0 + attr;
1193 used_locations |= (use_mask << attr);
1194 }
1195 }
1196
1197 /* Temporary storage for the set of attributes that need locations assigned.
1198 */
1199 struct temp_attr {
1200 unsigned slots;
1201 ir_variable *var;
1202
1203 /* Used below in the call to qsort. */
1204 static int compare(const void *a, const void *b)
1205 {
1206 const temp_attr *const l = (const temp_attr *) a;
1207 const temp_attr *const r = (const temp_attr *) b;
1208
1209 /* Reversed because we want a descending order sort below. */
1210 return r->slots - l->slots;
1211 }
1212 } to_assign[16];
1213
1214 unsigned num_attr = 0;
1215
1216 foreach_list(node, sh->ir) {
1217 ir_variable *const var = ((ir_instruction *) node)->as_variable();
1218
1219 if ((var == NULL) || (var->mode != ir_var_in))
1220 continue;
1221
1222 if (var->explicit_location) {
1223 const unsigned slots = count_attribute_slots(var->type);
1224 const unsigned use_mask = (1 << slots) - 1;
1225 const int attr = var->location - VERT_ATTRIB_GENERIC0;
1226
1227 if ((var->location >= (int)(max_attribute_index + VERT_ATTRIB_GENERIC0))
1228 || (var->location < 0)) {
1229 linker_error_printf(prog,
1230 "invalid explicit location %d specified for "
1231 "`%s'\n",
1232 (var->location < 0) ? var->location : attr,
1233 var->name);
1234 return false;
1235 } else if (var->location >= VERT_ATTRIB_GENERIC0) {
1236 used_locations |= (use_mask << attr);
1237 }
1238 }
1239
1240 /* The location was explicitly assigned, nothing to do here.
1241 */
1242 if (var->location != -1)
1243 continue;
1244
1245 to_assign[num_attr].slots = count_attribute_slots(var->type);
1246 to_assign[num_attr].var = var;
1247 num_attr++;
1248 }
1249
1250 /* If all of the attributes were assigned locations by the application (or
1251 * are built-in attributes with fixed locations), return early. This should
1252 * be the common case.
1253 */
1254 if (num_attr == 0)
1255 return true;
1256
1257 qsort(to_assign, num_attr, sizeof(to_assign[0]), temp_attr::compare);
1258
1259 /* VERT_ATTRIB_GENERIC0 is a psdueo-alias for VERT_ATTRIB_POS. It can only
1260 * be explicitly assigned by via glBindAttribLocation. Mark it as reserved
1261 * to prevent it from being automatically allocated below.
1262 */
1263 find_deref_visitor find("gl_Vertex");
1264 find.run(sh->ir);
1265 if (find.variable_found())
1266 used_locations |= (1 << 0);
1267
1268 for (unsigned i = 0; i < num_attr; i++) {
1269 /* Mask representing the contiguous slots that will be used by this
1270 * attribute.
1271 */
1272 const unsigned use_mask = (1 << to_assign[i].slots) - 1;
1273
1274 int location = find_available_slots(used_locations, to_assign[i].slots);
1275
1276 if (location < 0) {
1277 linker_error_printf(prog,
1278 "insufficient contiguous attribute locations "
1279 "available for vertex shader input `%s'",
1280 to_assign[i].var->name);
1281 return false;
1282 }
1283
1284 to_assign[i].var->location = VERT_ATTRIB_GENERIC0 + location;
1285 used_locations |= (use_mask << location);
1286 }
1287
1288 return true;
1289 }
1290
1291
1292 /**
1293 * Demote shader inputs and outputs that are not used in other stages
1294 */
1295 void
1296 demote_shader_inputs_and_outputs(gl_shader *sh, enum ir_variable_mode mode)
1297 {
1298 foreach_list(node, sh->ir) {
1299 ir_variable *const var = ((ir_instruction *) node)->as_variable();
1300
1301 if ((var == NULL) || (var->mode != int(mode)))
1302 continue;
1303
1304 /* A shader 'in' or 'out' variable is only really an input or output if
1305 * its value is used by other shader stages. This will cause the variable
1306 * to have a location assigned.
1307 */
1308 if (var->location == -1) {
1309 var->mode = ir_var_auto;
1310 }
1311 }
1312 }
1313
1314
1315 void
1316 assign_varying_locations(struct gl_shader_program *prog,
1317 gl_shader *producer, gl_shader *consumer)
1318 {
1319 /* FINISHME: Set dynamically when geometry shader support is added. */
1320 unsigned output_index = VERT_RESULT_VAR0;
1321 unsigned input_index = FRAG_ATTRIB_VAR0;
1322
1323 /* Operate in a total of three passes.
1324 *
1325 * 1. Assign locations for any matching inputs and outputs.
1326 *
1327 * 2. Mark output variables in the producer that do not have locations as
1328 * not being outputs. This lets the optimizer eliminate them.
1329 *
1330 * 3. Mark input variables in the consumer that do not have locations as
1331 * not being inputs. This lets the optimizer eliminate them.
1332 */
1333
1334 invalidate_variable_locations(producer, ir_var_out, VERT_RESULT_VAR0);
1335 invalidate_variable_locations(consumer, ir_var_in, FRAG_ATTRIB_VAR0);
1336
1337 foreach_list(node, producer->ir) {
1338 ir_variable *const output_var = ((ir_instruction *) node)->as_variable();
1339
1340 if ((output_var == NULL) || (output_var->mode != ir_var_out)
1341 || (output_var->location != -1))
1342 continue;
1343
1344 ir_variable *const input_var =
1345 consumer->symbols->get_variable(output_var->name);
1346
1347 if ((input_var == NULL) || (input_var->mode != ir_var_in))
1348 continue;
1349
1350 assert(input_var->location == -1);
1351
1352 output_var->location = output_index;
1353 input_var->location = input_index;
1354
1355 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1356 assert(!output_var->type->is_record());
1357
1358 if (output_var->type->is_array()) {
1359 const unsigned slots = output_var->type->length
1360 * output_var->type->fields.array->matrix_columns;
1361
1362 output_index += slots;
1363 input_index += slots;
1364 } else {
1365 const unsigned slots = output_var->type->matrix_columns;
1366
1367 output_index += slots;
1368 input_index += slots;
1369 }
1370 }
1371
1372 foreach_list(node, consumer->ir) {
1373 ir_variable *const var = ((ir_instruction *) node)->as_variable();
1374
1375 if ((var == NULL) || (var->mode != ir_var_in))
1376 continue;
1377
1378 if (var->location == -1) {
1379 if (prog->Version <= 120) {
1380 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1381 *
1382 * Only those varying variables used (i.e. read) in
1383 * the fragment shader executable must be written to
1384 * by the vertex shader executable; declaring
1385 * superfluous varying variables in a vertex shader is
1386 * permissible.
1387 *
1388 * We interpret this text as meaning that the VS must
1389 * write the variable for the FS to read it. See
1390 * "glsl1-varying read but not written" in piglit.
1391 */
1392
1393 linker_error_printf(prog, "fragment shader varying %s not written "
1394 "by vertex shader\n.", var->name);
1395 prog->LinkStatus = false;
1396 }
1397
1398 /* An 'in' variable is only really a shader input if its
1399 * value is written by the previous stage.
1400 */
1401 var->mode = ir_var_auto;
1402 }
1403 }
1404 }
1405
1406
1407 void
1408 link_shaders(struct gl_context *ctx, struct gl_shader_program *prog)
1409 {
1410 prog->LinkStatus = false;
1411 prog->Validated = false;
1412 prog->_Used = false;
1413
1414 if (prog->InfoLog != NULL)
1415 talloc_free(prog->InfoLog);
1416
1417 prog->InfoLog = talloc_strdup(NULL, "");
1418
1419 /* Separate the shaders into groups based on their type.
1420 */
1421 struct gl_shader **vert_shader_list;
1422 unsigned num_vert_shaders = 0;
1423 struct gl_shader **frag_shader_list;
1424 unsigned num_frag_shaders = 0;
1425
1426 vert_shader_list = (struct gl_shader **)
1427 calloc(2 * prog->NumShaders, sizeof(struct gl_shader *));
1428 frag_shader_list = &vert_shader_list[prog->NumShaders];
1429
1430 unsigned min_version = UINT_MAX;
1431 unsigned max_version = 0;
1432 for (unsigned i = 0; i < prog->NumShaders; i++) {
1433 min_version = MIN2(min_version, prog->Shaders[i]->Version);
1434 max_version = MAX2(max_version, prog->Shaders[i]->Version);
1435
1436 switch (prog->Shaders[i]->Type) {
1437 case GL_VERTEX_SHADER:
1438 vert_shader_list[num_vert_shaders] = prog->Shaders[i];
1439 num_vert_shaders++;
1440 break;
1441 case GL_FRAGMENT_SHADER:
1442 frag_shader_list[num_frag_shaders] = prog->Shaders[i];
1443 num_frag_shaders++;
1444 break;
1445 case GL_GEOMETRY_SHADER:
1446 /* FINISHME: Support geometry shaders. */
1447 assert(prog->Shaders[i]->Type != GL_GEOMETRY_SHADER);
1448 break;
1449 }
1450 }
1451
1452 /* Previous to GLSL version 1.30, different compilation units could mix and
1453 * match shading language versions. With GLSL 1.30 and later, the versions
1454 * of all shaders must match.
1455 */
1456 assert(min_version >= 100);
1457 assert(max_version <= 130);
1458 if ((max_version >= 130 || min_version == 100)
1459 && min_version != max_version) {
1460 linker_error_printf(prog, "all shaders must use same shading "
1461 "language version\n");
1462 goto done;
1463 }
1464
1465 prog->Version = max_version;
1466
1467 for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) {
1468 if (prog->_LinkedShaders[i] != NULL)
1469 ctx->Driver.DeleteShader(ctx, prog->_LinkedShaders[i]);
1470
1471 prog->_LinkedShaders[i] = NULL;
1472 }
1473
1474 /* Link all shaders for a particular stage and validate the result.
1475 */
1476 if (num_vert_shaders > 0) {
1477 gl_shader *const sh =
1478 link_intrastage_shaders(ctx, prog, vert_shader_list, num_vert_shaders);
1479
1480 if (sh == NULL)
1481 goto done;
1482
1483 if (!validate_vertex_shader_executable(prog, sh))
1484 goto done;
1485
1486 _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_VERTEX],
1487 sh);
1488 }
1489
1490 if (num_frag_shaders > 0) {
1491 gl_shader *const sh =
1492 link_intrastage_shaders(ctx, prog, frag_shader_list, num_frag_shaders);
1493
1494 if (sh == NULL)
1495 goto done;
1496
1497 if (!validate_fragment_shader_executable(prog, sh))
1498 goto done;
1499
1500 _mesa_reference_shader(ctx, &prog->_LinkedShaders[MESA_SHADER_FRAGMENT],
1501 sh);
1502 }
1503
1504 /* Here begins the inter-stage linking phase. Some initial validation is
1505 * performed, then locations are assigned for uniforms, attributes, and
1506 * varyings.
1507 */
1508 if (cross_validate_uniforms(prog)) {
1509 unsigned prev;
1510
1511 for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {
1512 if (prog->_LinkedShaders[prev] != NULL)
1513 break;
1514 }
1515
1516 /* Validate the inputs of each stage with the output of the preceeding
1517 * stage.
1518 */
1519 for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {
1520 if (prog->_LinkedShaders[i] == NULL)
1521 continue;
1522
1523 if (!cross_validate_outputs_to_inputs(prog,
1524 prog->_LinkedShaders[prev],
1525 prog->_LinkedShaders[i]))
1526 goto done;
1527
1528 prev = i;
1529 }
1530
1531 prog->LinkStatus = true;
1532 }
1533
1534 /* Do common optimization before assigning storage for attributes,
1535 * uniforms, and varyings. Later optimization could possibly make
1536 * some of that unused.
1537 */
1538 for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
1539 if (prog->_LinkedShaders[i] == NULL)
1540 continue;
1541
1542 while (do_common_optimization(prog->_LinkedShaders[i]->ir, true, 32))
1543 ;
1544 }
1545
1546 update_array_sizes(prog);
1547
1548 assign_uniform_locations(prog);
1549
1550 if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {
1551 /* FINISHME: The value of the max_attribute_index parameter is
1552 * FINISHME: implementation dependent based on the value of
1553 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1554 * FINISHME: at least 16, so hardcode 16 for now.
1555 */
1556 if (!assign_attribute_locations(prog, 16)) {
1557 prog->LinkStatus = false;
1558 goto done;
1559 }
1560 }
1561
1562 unsigned prev;
1563 for (prev = 0; prev < MESA_SHADER_TYPES; prev++) {
1564 if (prog->_LinkedShaders[prev] != NULL)
1565 break;
1566 }
1567
1568 for (unsigned i = prev + 1; i < MESA_SHADER_TYPES; i++) {
1569 if (prog->_LinkedShaders[i] == NULL)
1570 continue;
1571
1572 assign_varying_locations(prog,
1573 prog->_LinkedShaders[prev],
1574 prog->_LinkedShaders[i]);
1575 prev = i;
1576 }
1577
1578 if (prog->_LinkedShaders[MESA_SHADER_VERTEX] != NULL) {
1579 demote_shader_inputs_and_outputs(prog->_LinkedShaders[MESA_SHADER_VERTEX],
1580 ir_var_out);
1581 }
1582
1583 if (prog->_LinkedShaders[MESA_SHADER_GEOMETRY] != NULL) {
1584 gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
1585
1586 demote_shader_inputs_and_outputs(sh, ir_var_in);
1587 demote_shader_inputs_and_outputs(sh, ir_var_inout);
1588 demote_shader_inputs_and_outputs(sh, ir_var_out);
1589 }
1590
1591 if (prog->_LinkedShaders[MESA_SHADER_FRAGMENT] != NULL) {
1592 gl_shader *const sh = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
1593
1594 demote_shader_inputs_and_outputs(sh, ir_var_in);
1595 }
1596
1597 /* FINISHME: Assign fragment shader output locations. */
1598
1599 done:
1600 free(vert_shader_list);
1601 }