2 * Copyright © 2010 Intel Corporation
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:
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
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.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
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
35 * - Undefined references in each shader are resolve to definitions in
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.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
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
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
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.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
67 #include "main/core.h"
68 #include "glsl_symbol_table.h"
71 #include "program/hash_table.h"
73 #include "ir_optimization.h"
76 #include "main/shaderobj.h"
80 * Visitor that determines whether or not a variable is ever written.
82 class find_assignment_visitor
: public ir_hierarchical_visitor
{
84 find_assignment_visitor(const char *name
)
85 : name(name
), found(false)
90 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
92 ir_variable
*const var
= ir
->lhs
->variable_referenced();
94 if (strcmp(name
, var
->name
) == 0) {
99 return visit_continue_with_parent
;
102 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
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();
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) {
120 return visit_continue_with_parent
;
123 bool variable_found()
129 const char *name
; /**< Find writes to a variable with this name. */
130 bool found
; /**< Was a write to the variable found? */
135 * Visitor that determines whether or not a variable is ever read.
137 class find_deref_visitor
: public ir_hierarchical_visitor
{
139 find_deref_visitor(const char *name
)
140 : name(name
), found(false)
145 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
147 if (strcmp(this->name
, ir
->var
->name
) == 0) {
152 return visit_continue
;
155 bool variable_found() const
161 const char *name
; /**< Find writes to a variable with this name. */
162 bool found
; /**< Was a write to the variable found? */
167 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
171 ralloc_strcat(&prog
->InfoLog
, "error: ");
173 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
176 prog
->LinkStatus
= false;
181 invalidate_variable_locations(gl_shader
*sh
, enum ir_variable_mode mode
,
184 foreach_list(node
, sh
->ir
) {
185 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
187 if ((var
== NULL
) || (var
->mode
!= (unsigned) mode
))
190 /* Only assign locations for generic attributes / varyings / etc.
192 if ((var
->location
>= generic_base
) && !var
->explicit_location
)
199 * Determine the number of attribute slots required for a particular type
201 * This code is here because it implements the language rules of a specific
202 * GLSL version. Since it's a property of the language and not a property of
203 * types in general, it doesn't really belong in glsl_type.
206 count_attribute_slots(const glsl_type
*t
)
208 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
210 * "A scalar input counts the same amount against this limit as a vec4,
211 * so applications may want to consider packing groups of four
212 * unrelated float inputs together into a vector to better utilize the
213 * capabilities of the underlying hardware. A matrix input will use up
214 * multiple locations. The number of locations used will equal the
215 * number of columns in the matrix."
217 * The spec does not explicitly say how arrays are counted. However, it
218 * should be safe to assume the total number of slots consumed by an array
219 * is the number of entries in the array multiplied by the number of slots
220 * consumed by a single element of the array.
224 return t
->array_size() * count_attribute_slots(t
->element_type());
227 return t
->matrix_columns
;
234 * Verify that a vertex shader executable meets all semantic requirements
236 * \param shader Vertex shader executable to be verified
239 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
240 struct gl_shader
*shader
)
245 find_assignment_visitor
find("gl_Position");
246 find
.run(shader
->ir
);
247 if (!find
.variable_found()) {
248 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
257 * Verify that a fragment shader executable meets all semantic requirements
259 * \param shader Fragment shader executable to be verified
262 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
263 struct gl_shader
*shader
)
268 find_assignment_visitor
frag_color("gl_FragColor");
269 find_assignment_visitor
frag_data("gl_FragData");
271 frag_color
.run(shader
->ir
);
272 frag_data
.run(shader
->ir
);
274 if (frag_color
.variable_found() && frag_data
.variable_found()) {
275 linker_error(prog
, "fragment shader writes to both "
276 "`gl_FragColor' and `gl_FragData'\n");
285 * Generate a string describing the mode of a variable
288 mode_string(const ir_variable
*var
)
292 return (var
->read_only
) ? "global constant" : "global variable";
294 case ir_var_uniform
: return "uniform";
295 case ir_var_in
: return "shader input";
296 case ir_var_out
: return "shader output";
297 case ir_var_inout
: return "shader inout";
299 case ir_var_const_in
:
300 case ir_var_temporary
:
302 assert(!"Should not get here.");
303 return "invalid variable";
309 * Perform validation of global variables used across multiple shaders
312 cross_validate_globals(struct gl_shader_program
*prog
,
313 struct gl_shader
**shader_list
,
314 unsigned num_shaders
,
317 /* Examine all of the uniforms in all of the shaders and cross validate
320 glsl_symbol_table variables
;
321 for (unsigned i
= 0; i
< num_shaders
; i
++) {
322 if (shader_list
[i
] == NULL
)
325 foreach_list(node
, shader_list
[i
]->ir
) {
326 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
331 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
334 /* Don't cross validate temporaries that are at global scope. These
335 * will eventually get pulled into the shaders 'main'.
337 if (var
->mode
== ir_var_temporary
)
340 /* If a global with this name has already been seen, verify that the
341 * new instance has the same type. In addition, if the globals have
342 * initializers, the values of the initializers must be the same.
344 ir_variable
*const existing
= variables
.get_variable(var
->name
);
345 if (existing
!= NULL
) {
346 if (var
->type
!= existing
->type
) {
347 /* Consider the types to be "the same" if both types are arrays
348 * of the same type and one of the arrays is implicitly sized.
349 * In addition, set the type of the linked variable to the
350 * explicitly sized array.
352 if (var
->type
->is_array()
353 && existing
->type
->is_array()
354 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
355 && ((var
->type
->length
== 0)
356 || (existing
->type
->length
== 0))) {
357 if (var
->type
->length
!= 0) {
358 existing
->type
= var
->type
;
361 linker_error(prog
, "%s `%s' declared as type "
362 "`%s' and type `%s'\n",
364 var
->name
, var
->type
->name
,
365 existing
->type
->name
);
370 if (var
->explicit_location
) {
371 if (existing
->explicit_location
372 && (var
->location
!= existing
->location
)) {
373 linker_error(prog
, "explicit locations for %s "
374 "`%s' have differing values\n",
375 mode_string(var
), var
->name
);
379 existing
->location
= var
->location
;
380 existing
->explicit_location
= true;
383 /* Validate layout qualifiers for gl_FragDepth.
385 * From the AMD_conservative_depth spec:
386 * "If gl_FragDepth is redeclared in any fragment shader in
387 * a program, it must be redeclared in all fragment shaders in that
388 * program that have static assignments to gl_FragDepth. All
389 * redeclarations of gl_FragDepth in all fragment shaders in
390 * a single program must have the same set of qualifiers."
392 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
393 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
394 bool layout_differs
= var
->depth_layout
!= existing
->depth_layout
;
395 if (layout_declared
&& layout_differs
) {
397 "All redeclarations of gl_FragDepth in all fragment shaders "
398 "in a single program must have the same set of qualifiers.");
400 if (var
->used
&& layout_differs
) {
402 "If gl_FragDepth is redeclared with a layout qualifier in"
403 "any fragment shader, it must be redeclared with the same"
404 "layout qualifier in all fragment shaders that have"
405 "assignments to gl_FragDepth");
409 /* FINISHME: Handle non-constant initializers.
411 if (var
->constant_value
!= NULL
) {
412 if (existing
->constant_value
!= NULL
) {
413 if (!var
->constant_value
->has_value(existing
->constant_value
)) {
414 linker_error(prog
, "initializers for %s "
415 "`%s' have differing values\n",
416 mode_string(var
), var
->name
);
420 /* If the first-seen instance of a particular uniform did not
421 * have an initializer but a later instance does, copy the
422 * initializer to the version stored in the symbol table.
424 /* FINISHME: This is wrong. The constant_value field should
425 * FINISHME: not be modified! Imagine a case where a shader
426 * FINISHME: without an initializer is linked in two different
427 * FINISHME: programs with shaders that have differing
428 * FINISHME: initializers. Linking with the first will
429 * FINISHME: modify the shader, and linking with the second
430 * FINISHME: will fail.
432 existing
->constant_value
=
433 var
->constant_value
->clone(ralloc_parent(existing
), NULL
);
436 if (existing
->invariant
!= var
->invariant
) {
437 linker_error(prog
, "declarations for %s `%s' have "
438 "mismatching invariant qualifiers\n",
439 mode_string(var
), var
->name
);
442 if (existing
->centroid
!= var
->centroid
) {
443 linker_error(prog
, "declarations for %s `%s' have "
444 "mismatching centroid qualifiers\n",
445 mode_string(var
), var
->name
);
449 variables
.add_variable(var
);
458 * Perform validation of uniforms used across multiple shader stages
461 cross_validate_uniforms(struct gl_shader_program
*prog
)
463 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
464 MESA_SHADER_TYPES
, true);
469 * Validate that outputs from one stage match inputs of another
472 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
473 gl_shader
*producer
, gl_shader
*consumer
)
475 glsl_symbol_table parameters
;
476 /* FINISHME: Figure these out dynamically. */
477 const char *const producer_stage
= "vertex";
478 const char *const consumer_stage
= "fragment";
480 /* Find all shader outputs in the "producer" stage.
482 foreach_list(node
, producer
->ir
) {
483 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
485 /* FINISHME: For geometry shaders, this should also look for inout
486 * FINISHME: variables.
488 if ((var
== NULL
) || (var
->mode
!= ir_var_out
))
491 parameters
.add_variable(var
);
495 /* Find all shader inputs in the "consumer" stage. Any variables that have
496 * matching outputs already in the symbol table must have the same type and
499 foreach_list(node
, consumer
->ir
) {
500 ir_variable
*const input
= ((ir_instruction
*) node
)->as_variable();
502 /* FINISHME: For geometry shaders, this should also look for inout
503 * FINISHME: variables.
505 if ((input
== NULL
) || (input
->mode
!= ir_var_in
))
508 ir_variable
*const output
= parameters
.get_variable(input
->name
);
509 if (output
!= NULL
) {
510 /* Check that the types match between stages.
512 if (input
->type
!= output
->type
) {
513 /* There is a bit of a special case for gl_TexCoord. This
514 * built-in is unsized by default. Applications that variable
515 * access it must redeclare it with a size. There is some
516 * language in the GLSL spec that implies the fragment shader
517 * and vertex shader do not have to agree on this size. Other
518 * driver behave this way, and one or two applications seem to
521 * Neither declaration needs to be modified here because the array
522 * sizes are fixed later when update_array_sizes is called.
524 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
526 * "Unlike user-defined varying variables, the built-in
527 * varying variables don't have a strict one-to-one
528 * correspondence between the vertex language and the
529 * fragment language."
531 if (!output
->type
->is_array()
532 || (strncmp("gl_", output
->name
, 3) != 0)) {
534 "%s shader output `%s' declared as type `%s', "
535 "but %s shader input declared as type `%s'\n",
536 producer_stage
, output
->name
,
538 consumer_stage
, input
->type
->name
);
543 /* Check that all of the qualifiers match between stages.
545 if (input
->centroid
!= output
->centroid
) {
547 "%s shader output `%s' %s centroid qualifier, "
548 "but %s shader input %s centroid qualifier\n",
551 (output
->centroid
) ? "has" : "lacks",
553 (input
->centroid
) ? "has" : "lacks");
557 if (input
->invariant
!= output
->invariant
) {
559 "%s shader output `%s' %s invariant qualifier, "
560 "but %s shader input %s invariant qualifier\n",
563 (output
->invariant
) ? "has" : "lacks",
565 (input
->invariant
) ? "has" : "lacks");
569 if (input
->interpolation
!= output
->interpolation
) {
571 "%s shader output `%s' specifies %s "
572 "interpolation qualifier, "
573 "but %s shader input specifies %s "
574 "interpolation qualifier\n",
577 output
->interpolation_string(),
579 input
->interpolation_string());
590 * Populates a shaders symbol table with all global declarations
593 populate_symbol_table(gl_shader
*sh
)
595 sh
->symbols
= new(sh
) glsl_symbol_table
;
597 foreach_list(node
, sh
->ir
) {
598 ir_instruction
*const inst
= (ir_instruction
*) node
;
602 if ((func
= inst
->as_function()) != NULL
) {
603 sh
->symbols
->add_function(func
);
604 } else if ((var
= inst
->as_variable()) != NULL
) {
605 sh
->symbols
->add_variable(var
);
612 * Remap variables referenced in an instruction tree
614 * This is used when instruction trees are cloned from one shader and placed in
615 * another. These trees will contain references to \c ir_variable nodes that
616 * do not exist in the target shader. This function finds these \c ir_variable
617 * references and replaces the references with matching variables in the target
620 * If there is no matching variable in the target shader, a clone of the
621 * \c ir_variable is made and added to the target shader. The new variable is
622 * added to \b both the instruction stream and the symbol table.
624 * \param inst IR tree that is to be processed.
625 * \param symbols Symbol table containing global scope symbols in the
627 * \param instructions Instruction stream where new variable declarations
631 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
634 class remap_visitor
: public ir_hierarchical_visitor
{
636 remap_visitor(struct gl_shader
*target
,
639 this->target
= target
;
640 this->symbols
= target
->symbols
;
641 this->instructions
= target
->ir
;
645 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
647 if (ir
->var
->mode
== ir_var_temporary
) {
648 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
652 return visit_continue
;
655 ir_variable
*const existing
=
656 this->symbols
->get_variable(ir
->var
->name
);
657 if (existing
!= NULL
)
660 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
662 this->symbols
->add_variable(copy
);
663 this->instructions
->push_head(copy
);
667 return visit_continue
;
671 struct gl_shader
*target
;
672 glsl_symbol_table
*symbols
;
673 exec_list
*instructions
;
677 remap_visitor
v(target
, temps
);
684 * Move non-declarations from one instruction stream to another
686 * The intended usage pattern of this function is to pass the pointer to the
687 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
688 * pointer) for \c last and \c false for \c make_copies on the first
689 * call. Successive calls pass the return value of the previous call for
690 * \c last and \c true for \c make_copies.
692 * \param instructions Source instruction stream
693 * \param last Instruction after which new instructions should be
694 * inserted in the target instruction stream
695 * \param make_copies Flag selecting whether instructions in \c instructions
696 * should be copied (via \c ir_instruction::clone) into the
697 * target list or moved.
700 * The new "last" instruction in the target instruction stream. This pointer
701 * is suitable for use as the \c last parameter of a later call to this
705 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
706 bool make_copies
, gl_shader
*target
)
708 hash_table
*temps
= NULL
;
711 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
712 hash_table_pointer_compare
);
714 foreach_list_safe(node
, instructions
) {
715 ir_instruction
*inst
= (ir_instruction
*) node
;
717 if (inst
->as_function())
720 ir_variable
*var
= inst
->as_variable();
721 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
724 assert(inst
->as_assignment()
725 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
728 inst
= inst
->clone(target
, NULL
);
731 hash_table_insert(temps
, inst
, var
);
733 remap_variables(inst
, target
, temps
);
738 last
->insert_after(inst
);
743 hash_table_dtor(temps
);
749 * Get the function signature for main from a shader
751 static ir_function_signature
*
752 get_main_function_signature(gl_shader
*sh
)
754 ir_function
*const f
= sh
->symbols
->get_function("main");
756 exec_list void_parameters
;
758 /* Look for the 'void main()' signature and ensure that it's defined.
759 * This keeps the linker from accidentally pick a shader that just
760 * contains a prototype for main.
762 * We don't have to check for multiple definitions of main (in multiple
763 * shaders) because that would have already been caught above.
765 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
766 if ((sig
!= NULL
) && sig
->is_defined
) {
776 * Combine a group of shaders for a single stage to generate a linked shader
779 * If this function is supplied a single shader, it is cloned, and the new
780 * shader is returned.
782 static struct gl_shader
*
783 link_intrastage_shaders(void *mem_ctx
,
784 struct gl_context
*ctx
,
785 struct gl_shader_program
*prog
,
786 struct gl_shader
**shader_list
,
787 unsigned num_shaders
)
789 /* Check that global variables defined in multiple shaders are consistent.
791 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
794 /* Check that there is only a single definition of each function signature
795 * across all shaders.
797 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
798 foreach_list(node
, shader_list
[i
]->ir
) {
799 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
804 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
805 ir_function
*const other
=
806 shader_list
[j
]->symbols
->get_function(f
->name
);
808 /* If the other shader has no function (and therefore no function
809 * signatures) with the same name, skip to the next shader.
814 foreach_iter (exec_list_iterator
, iter
, *f
) {
815 ir_function_signature
*sig
=
816 (ir_function_signature
*) iter
.get();
818 if (!sig
->is_defined
|| sig
->is_builtin
)
821 ir_function_signature
*other_sig
=
822 other
->exact_matching_signature(& sig
->parameters
);
824 if ((other_sig
!= NULL
) && other_sig
->is_defined
825 && !other_sig
->is_builtin
) {
826 linker_error(prog
, "function `%s' is multiply defined",
835 /* Find the shader that defines main, and make a clone of it.
837 * Starting with the clone, search for undefined references. If one is
838 * found, find the shader that defines it. Clone the reference and add
839 * it to the shader. Repeat until there are no undefined references or
840 * until a reference cannot be resolved.
842 gl_shader
*main
= NULL
;
843 for (unsigned i
= 0; i
< num_shaders
; i
++) {
844 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
845 main
= shader_list
[i
];
851 linker_error(prog
, "%s shader lacks `main'\n",
852 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
853 ? "vertex" : "fragment");
857 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
858 linked
->ir
= new(linked
) exec_list
;
859 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
861 populate_symbol_table(linked
);
863 /* The a pointer to the main function in the final linked shader (i.e., the
864 * copy of the original shader that contained the main function).
866 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
868 /* Move any instructions other than variable declarations or function
869 * declarations into main.
871 exec_node
*insertion_point
=
872 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
875 for (unsigned i
= 0; i
< num_shaders
; i
++) {
876 if (shader_list
[i
] == main
)
879 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
880 insertion_point
, true, linked
);
883 /* Resolve initializers for global variables in the linked shader.
885 unsigned num_linking_shaders
= num_shaders
;
886 for (unsigned i
= 0; i
< num_shaders
; i
++)
887 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
889 gl_shader
**linking_shaders
=
890 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
892 memcpy(linking_shaders
, shader_list
,
893 sizeof(linking_shaders
[0]) * num_shaders
);
895 unsigned idx
= num_shaders
;
896 for (unsigned i
= 0; i
< num_shaders
; i
++) {
897 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
898 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
899 idx
+= shader_list
[i
]->num_builtins_to_link
;
902 assert(idx
== num_linking_shaders
);
904 if (!link_function_calls(prog
, linked
, linking_shaders
,
905 num_linking_shaders
)) {
906 ctx
->Driver
.DeleteShader(ctx
, linked
);
910 free(linking_shaders
);
912 /* Make a pass over all variable declarations to ensure that arrays with
913 * unspecified sizes have a size specified. The size is inferred from the
914 * max_array_access field.
916 if (linked
!= NULL
) {
917 class array_sizing_visitor
: public ir_hierarchical_visitor
{
919 virtual ir_visitor_status
visit(ir_variable
*var
)
921 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
922 const glsl_type
*type
=
923 glsl_type::get_array_instance(var
->type
->fields
.array
,
924 var
->max_array_access
+ 1);
926 assert(type
!= NULL
);
930 return visit_continue
;
941 struct uniform_node
{
943 struct gl_uniform
*u
;
948 * Update the sizes of linked shader uniform arrays to the maximum
951 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
953 * If one or more elements of an array are active,
954 * GetActiveUniform will return the name of the array in name,
955 * subject to the restrictions listed above. The type of the array
956 * is returned in type. The size parameter contains the highest
957 * array element index used, plus one. The compiler or linker
958 * determines the highest index used. There will be only one
959 * active uniform reported by the GL per uniform array.
963 update_array_sizes(struct gl_shader_program
*prog
)
965 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
966 if (prog
->_LinkedShaders
[i
] == NULL
)
969 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
970 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
972 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
973 var
->mode
!= ir_var_in
&&
974 var
->mode
!= ir_var_out
) ||
975 !var
->type
->is_array())
978 unsigned int size
= var
->max_array_access
;
979 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
980 if (prog
->_LinkedShaders
[j
] == NULL
)
983 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
984 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
988 if (strcmp(var
->name
, other_var
->name
) == 0 &&
989 other_var
->max_array_access
> size
) {
990 size
= other_var
->max_array_access
;
995 if (size
+ 1 != var
->type
->fields
.array
->length
) {
996 /* If this is a built-in uniform (i.e., it's backed by some
997 * fixed-function state), adjust the number of state slots to
998 * match the new array size. The number of slots per array entry
999 * is not known. It seems safe to assume that the total number of
1000 * slots is an integer multiple of the number of array elements.
1001 * Determine the number of slots per array element by dividing by
1002 * the old (total) size.
1004 if (var
->num_state_slots
> 0) {
1005 var
->num_state_slots
= (size
+ 1)
1006 * (var
->num_state_slots
/ var
->type
->length
);
1009 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1011 /* FINISHME: We should update the types of array
1012 * dereferences of this variable now.
1020 add_uniform(void *mem_ctx
, exec_list
*uniforms
, struct hash_table
*ht
,
1021 const char *name
, const glsl_type
*type
, GLenum shader_type
,
1022 unsigned *next_shader_pos
, unsigned *total_uniforms
)
1024 if (type
->is_record()) {
1025 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1026 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
1027 char *field_name
= ralloc_asprintf(mem_ctx
, "%s.%s", name
,
1028 type
->fields
.structure
[i
].name
);
1030 add_uniform(mem_ctx
, uniforms
, ht
, field_name
, field_type
,
1031 shader_type
, next_shader_pos
, total_uniforms
);
1034 uniform_node
*n
= (uniform_node
*) hash_table_find(ht
, name
);
1035 unsigned int vec4_slots
;
1036 const glsl_type
*array_elem_type
= NULL
;
1038 if (type
->is_array()) {
1039 array_elem_type
= type
->fields
.array
;
1040 /* Array of structures. */
1041 if (array_elem_type
->is_record()) {
1042 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1043 char *elem_name
= ralloc_asprintf(mem_ctx
, "%s[%d]", name
, i
);
1044 add_uniform(mem_ctx
, uniforms
, ht
, elem_name
, array_elem_type
,
1045 shader_type
, next_shader_pos
, total_uniforms
);
1051 /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
1052 * vectors to vec4 slots.
1054 if (type
->is_array()) {
1055 if (array_elem_type
->is_sampler())
1056 vec4_slots
= type
->length
;
1058 vec4_slots
= type
->length
* array_elem_type
->matrix_columns
;
1059 } else if (type
->is_sampler()) {
1062 vec4_slots
= type
->matrix_columns
;
1066 n
= (uniform_node
*) calloc(1, sizeof(struct uniform_node
));
1067 n
->u
= (gl_uniform
*) calloc(1, sizeof(struct gl_uniform
));
1068 n
->slots
= vec4_slots
;
1070 n
->u
->Name
= strdup(name
);
1075 (*total_uniforms
)++;
1077 hash_table_insert(ht
, n
, name
);
1078 uniforms
->push_tail(& n
->link
);
1081 switch (shader_type
) {
1082 case GL_VERTEX_SHADER
:
1083 n
->u
->VertPos
= *next_shader_pos
;
1085 case GL_FRAGMENT_SHADER
:
1086 n
->u
->FragPos
= *next_shader_pos
;
1088 case GL_GEOMETRY_SHADER
:
1089 n
->u
->GeomPos
= *next_shader_pos
;
1093 (*next_shader_pos
) += vec4_slots
;
1098 assign_uniform_locations(struct gl_shader_program
*prog
)
1102 unsigned total_uniforms
= 0;
1103 hash_table
*ht
= hash_table_ctor(32, hash_table_string_hash
,
1104 hash_table_string_compare
);
1105 void *mem_ctx
= ralloc_context(NULL
);
1107 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1108 if (prog
->_LinkedShaders
[i
] == NULL
)
1111 unsigned next_position
= 0;
1113 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1114 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1116 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
))
1119 if (strncmp(var
->name
, "gl_", 3) == 0) {
1120 /* At the moment, we don't allocate uniform locations for
1121 * builtin uniforms. It's permitted by spec, and we'll
1122 * likely switch to doing that at some point, but not yet.
1127 var
->location
= next_position
;
1128 add_uniform(mem_ctx
, &uniforms
, ht
, var
->name
, var
->type
,
1129 prog
->_LinkedShaders
[i
]->Type
,
1130 &next_position
, &total_uniforms
);
1134 ralloc_free(mem_ctx
);
1136 gl_uniform_list
*ul
= (gl_uniform_list
*)
1137 calloc(1, sizeof(gl_uniform_list
));
1139 ul
->Size
= total_uniforms
;
1140 ul
->NumUniforms
= total_uniforms
;
1141 ul
->Uniforms
= (gl_uniform
*) calloc(total_uniforms
, sizeof(gl_uniform
));
1145 for (uniform_node
*node
= (uniform_node
*) uniforms
.head
1146 ; node
->link
.next
!= NULL
1148 next
= (uniform_node
*) node
->link
.next
;
1150 node
->link
.remove();
1151 memcpy(&ul
->Uniforms
[idx
], node
->u
, sizeof(gl_uniform
));
1158 hash_table_dtor(ht
);
1160 prog
->Uniforms
= ul
;
1165 * Find a contiguous set of available bits in a bitmask.
1167 * \param used_mask Bits representing used (1) and unused (0) locations
1168 * \param needed_count Number of contiguous bits needed.
1171 * Base location of the available bits on success or -1 on failure.
1174 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1176 unsigned needed_mask
= (1 << needed_count
) - 1;
1177 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1179 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1180 * cannot optimize possibly infinite loops" for the loop below.
1182 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1185 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1186 if ((needed_mask
& ~used_mask
) == needed_mask
)
1197 * Assign locations for either VS inputs for FS outputs
1199 * \param prog Shader program whose variables need locations assigned
1200 * \param target_index Selector for the program target to receive location
1201 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1202 * \c MESA_SHADER_FRAGMENT.
1203 * \param max_index Maximum number of generic locations. This corresponds
1204 * to either the maximum number of draw buffers or the
1205 * maximum number of generic attributes.
1208 * If locations are successfully assigned, true is returned. Otherwise an
1209 * error is emitted to the shader link log and false is returned.
1212 * Locations set via \c glBindFragDataLocation are not currently supported.
1213 * Only locations assigned automatically by the linker, explicitly set by a
1214 * layout qualifier, or explicitly set by a built-in variable (e.g., \c
1215 * gl_FragColor) are supported for fragment shaders.
1218 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1219 unsigned target_index
,
1222 /* Mark invalid locations as being used.
1224 unsigned used_locations
= (max_index
>= 32)
1225 ? ~0 : ~((1 << max_index
) - 1);
1227 assert((target_index
== MESA_SHADER_VERTEX
)
1228 || (target_index
== MESA_SHADER_FRAGMENT
));
1230 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1234 /* Operate in a total of four passes.
1236 * 1. Invalidate the location assignments for all vertex shader inputs.
1238 * 2. Assign locations for inputs that have user-defined (via
1239 * glBindVertexAttribLocation) locations.
1241 * 3. Sort the attributes without assigned locations by number of slots
1242 * required in decreasing order. Fragmentation caused by attribute
1243 * locations assigned by the application may prevent large attributes
1244 * from having enough contiguous space.
1246 * 4. Assign locations to any inputs without assigned locations.
1249 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1250 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1252 const enum ir_variable_mode direction
=
1253 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1256 invalidate_variable_locations(sh
, direction
, generic_base
);
1258 if ((target_index
== MESA_SHADER_VERTEX
) && (prog
->Attributes
!= NULL
)) {
1259 for (unsigned i
= 0; i
< prog
->Attributes
->NumParameters
; i
++) {
1260 ir_variable
*const var
=
1261 sh
->symbols
->get_variable(prog
->Attributes
->Parameters
[i
].Name
);
1263 /* Note: attributes that occupy multiple slots, such as arrays or
1264 * matrices, may appear in the attrib array multiple times.
1266 if ((var
== NULL
) || (var
->location
!= -1))
1269 /* From page 61 of the OpenGL 4.0 spec:
1271 * "LinkProgram will fail if the attribute bindings assigned by
1272 * BindAttribLocation do not leave not enough space to assign a
1273 * location for an active matrix attribute or an active attribute
1274 * array, both of which require multiple contiguous generic
1277 * Previous versions of the spec contain similar language but omit the
1278 * bit about attribute arrays.
1280 * Page 61 of the OpenGL 4.0 spec also says:
1282 * "It is possible for an application to bind more than one
1283 * attribute name to the same location. This is referred to as
1284 * aliasing. This will only work if only one of the aliased
1285 * attributes is active in the executable program, or if no path
1286 * through the shader consumes more than one attribute of a set
1287 * of attributes aliased to the same location. A link error can
1288 * occur if the linker determines that every path through the
1289 * shader consumes multiple aliased attributes, but
1290 * implementations are not required to generate an error in this
1293 * These two paragraphs are either somewhat contradictory, or I don't
1294 * fully understand one or both of them.
1296 /* FINISHME: The code as currently written does not support attribute
1297 * FINISHME: location aliasing (see comment above).
1299 const int attr
= prog
->Attributes
->Parameters
[i
].StateIndexes
[0];
1300 const unsigned slots
= count_attribute_slots(var
->type
);
1302 /* Mask representing the contiguous slots that will be used by this
1305 const unsigned use_mask
= (1 << slots
) - 1;
1307 /* Generate a link error if the set of bits requested for this
1308 * attribute overlaps any previously allocated bits.
1310 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1312 "insufficient contiguous attribute locations "
1313 "available for vertex shader input `%s'",
1318 var
->location
= VERT_ATTRIB_GENERIC0
+ attr
;
1319 used_locations
|= (use_mask
<< attr
);
1323 /* Temporary storage for the set of attributes that need locations assigned.
1329 /* Used below in the call to qsort. */
1330 static int compare(const void *a
, const void *b
)
1332 const temp_attr
*const l
= (const temp_attr
*) a
;
1333 const temp_attr
*const r
= (const temp_attr
*) b
;
1335 /* Reversed because we want a descending order sort below. */
1336 return r
->slots
- l
->slots
;
1340 unsigned num_attr
= 0;
1342 foreach_list(node
, sh
->ir
) {
1343 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1345 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1348 if (var
->explicit_location
) {
1349 const unsigned slots
= count_attribute_slots(var
->type
);
1350 const unsigned use_mask
= (1 << slots
) - 1;
1351 const int attr
= var
->location
- generic_base
;
1353 if ((var
->location
>= (int)(max_index
+ generic_base
))
1354 || (var
->location
< 0)) {
1356 "invalid explicit location %d specified for `%s'\n",
1357 (var
->location
< 0) ? var
->location
: attr
,
1360 } else if (var
->location
>= generic_base
) {
1361 used_locations
|= (use_mask
<< attr
);
1365 /* The location was explicitly assigned, nothing to do here.
1367 if (var
->location
!= -1)
1370 to_assign
[num_attr
].slots
= count_attribute_slots(var
->type
);
1371 to_assign
[num_attr
].var
= var
;
1375 /* If all of the attributes were assigned locations by the application (or
1376 * are built-in attributes with fixed locations), return early. This should
1377 * be the common case.
1382 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1384 if (target_index
== MESA_SHADER_VERTEX
) {
1385 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1386 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1387 * reserved to prevent it from being automatically allocated below.
1389 find_deref_visitor
find("gl_Vertex");
1391 if (find
.variable_found())
1392 used_locations
|= (1 << 0);
1395 for (unsigned i
= 0; i
< num_attr
; i
++) {
1396 /* Mask representing the contiguous slots that will be used by this
1399 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1401 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1404 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1405 ? "vertex shader input" : "fragment shader output";
1408 "insufficient contiguous attribute locations "
1409 "available for %s `%s'",
1410 string
, to_assign
[i
].var
->name
);
1414 to_assign
[i
].var
->location
= generic_base
+ location
;
1415 used_locations
|= (use_mask
<< location
);
1423 * Demote shader inputs and outputs that are not used in other stages
1426 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1428 foreach_list(node
, sh
->ir
) {
1429 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1431 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1434 /* A shader 'in' or 'out' variable is only really an input or output if
1435 * its value is used by other shader stages. This will cause the variable
1436 * to have a location assigned.
1438 if (var
->location
== -1) {
1439 var
->mode
= ir_var_auto
;
1446 assign_varying_locations(struct gl_context
*ctx
,
1447 struct gl_shader_program
*prog
,
1448 gl_shader
*producer
, gl_shader
*consumer
)
1450 /* FINISHME: Set dynamically when geometry shader support is added. */
1451 unsigned output_index
= VERT_RESULT_VAR0
;
1452 unsigned input_index
= FRAG_ATTRIB_VAR0
;
1454 /* Operate in a total of three passes.
1456 * 1. Assign locations for any matching inputs and outputs.
1458 * 2. Mark output variables in the producer that do not have locations as
1459 * not being outputs. This lets the optimizer eliminate them.
1461 * 3. Mark input variables in the consumer that do not have locations as
1462 * not being inputs. This lets the optimizer eliminate them.
1465 invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
1466 invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
1468 foreach_list(node
, producer
->ir
) {
1469 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
1471 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
)
1472 || (output_var
->location
!= -1))
1475 ir_variable
*const input_var
=
1476 consumer
->symbols
->get_variable(output_var
->name
);
1478 if ((input_var
== NULL
) || (input_var
->mode
!= ir_var_in
))
1481 assert(input_var
->location
== -1);
1483 output_var
->location
= output_index
;
1484 input_var
->location
= input_index
;
1486 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1487 assert(!output_var
->type
->is_record());
1489 if (output_var
->type
->is_array()) {
1490 const unsigned slots
= output_var
->type
->length
1491 * output_var
->type
->fields
.array
->matrix_columns
;
1493 output_index
+= slots
;
1494 input_index
+= slots
;
1496 const unsigned slots
= output_var
->type
->matrix_columns
;
1498 output_index
+= slots
;
1499 input_index
+= slots
;
1503 unsigned varying_vectors
= 0;
1505 foreach_list(node
, consumer
->ir
) {
1506 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1508 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
1511 if (var
->location
== -1) {
1512 if (prog
->Version
<= 120) {
1513 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1515 * Only those varying variables used (i.e. read) in
1516 * the fragment shader executable must be written to
1517 * by the vertex shader executable; declaring
1518 * superfluous varying variables in a vertex shader is
1521 * We interpret this text as meaning that the VS must
1522 * write the variable for the FS to read it. See
1523 * "glsl1-varying read but not written" in piglit.
1526 linker_error(prog
, "fragment shader varying %s not written "
1527 "by vertex shader\n.", var
->name
);
1530 /* An 'in' variable is only really a shader input if its
1531 * value is written by the previous stage.
1533 var
->mode
= ir_var_auto
;
1535 /* The packing rules are used for vertex shader inputs are also used
1536 * for fragment shader inputs.
1538 varying_vectors
+= count_attribute_slots(var
->type
);
1542 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1543 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
1544 linker_error(prog
, "shader uses too many varying vectors "
1546 varying_vectors
, ctx
->Const
.MaxVarying
);
1550 const unsigned float_components
= varying_vectors
* 4;
1551 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
1552 linker_error(prog
, "shader uses too many varying components "
1554 float_components
, ctx
->Const
.MaxVarying
* 4);
1564 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1566 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1568 prog
->LinkStatus
= false;
1569 prog
->Validated
= false;
1570 prog
->_Used
= false;
1572 if (prog
->InfoLog
!= NULL
)
1573 ralloc_free(prog
->InfoLog
);
1575 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1577 /* Separate the shaders into groups based on their type.
1579 struct gl_shader
**vert_shader_list
;
1580 unsigned num_vert_shaders
= 0;
1581 struct gl_shader
**frag_shader_list
;
1582 unsigned num_frag_shaders
= 0;
1584 vert_shader_list
= (struct gl_shader
**)
1585 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
1586 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
1588 unsigned min_version
= UINT_MAX
;
1589 unsigned max_version
= 0;
1590 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1591 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1592 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1594 switch (prog
->Shaders
[i
]->Type
) {
1595 case GL_VERTEX_SHADER
:
1596 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1599 case GL_FRAGMENT_SHADER
:
1600 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1603 case GL_GEOMETRY_SHADER
:
1604 /* FINISHME: Support geometry shaders. */
1605 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
1610 /* Previous to GLSL version 1.30, different compilation units could mix and
1611 * match shading language versions. With GLSL 1.30 and later, the versions
1612 * of all shaders must match.
1614 assert(min_version
>= 100);
1615 assert(max_version
<= 130);
1616 if ((max_version
>= 130 || min_version
== 100)
1617 && min_version
!= max_version
) {
1618 linker_error(prog
, "all shaders must use same shading "
1619 "language version\n");
1623 prog
->Version
= max_version
;
1625 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1626 if (prog
->_LinkedShaders
[i
] != NULL
)
1627 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1629 prog
->_LinkedShaders
[i
] = NULL
;
1632 /* Link all shaders for a particular stage and validate the result.
1634 if (num_vert_shaders
> 0) {
1635 gl_shader
*const sh
=
1636 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1642 if (!validate_vertex_shader_executable(prog
, sh
))
1645 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1649 if (num_frag_shaders
> 0) {
1650 gl_shader
*const sh
=
1651 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1657 if (!validate_fragment_shader_executable(prog
, sh
))
1660 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1664 /* Here begins the inter-stage linking phase. Some initial validation is
1665 * performed, then locations are assigned for uniforms, attributes, and
1668 if (cross_validate_uniforms(prog
)) {
1671 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1672 if (prog
->_LinkedShaders
[prev
] != NULL
)
1676 /* Validate the inputs of each stage with the output of the preceding
1679 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1680 if (prog
->_LinkedShaders
[i
] == NULL
)
1683 if (!cross_validate_outputs_to_inputs(prog
,
1684 prog
->_LinkedShaders
[prev
],
1685 prog
->_LinkedShaders
[i
]))
1691 prog
->LinkStatus
= true;
1694 /* Do common optimization before assigning storage for attributes,
1695 * uniforms, and varyings. Later optimization could possibly make
1696 * some of that unused.
1698 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1699 if (prog
->_LinkedShaders
[i
] == NULL
)
1702 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
1703 if (!prog
->LinkStatus
)
1706 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, 32))
1710 update_array_sizes(prog
);
1712 assign_uniform_locations(prog
);
1714 /* FINISHME: The value of the max_attribute_index parameter is
1715 * FINISHME: implementation dependent based on the value of
1716 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1717 * FINISHME: at least 16, so hardcode 16 for now.
1719 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
1723 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, ctx
->Const
.MaxDrawBuffers
)) {
1728 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1729 if (prog
->_LinkedShaders
[prev
] != NULL
)
1733 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1734 if (prog
->_LinkedShaders
[i
] == NULL
)
1737 if (!assign_varying_locations(ctx
, prog
,
1738 prog
->_LinkedShaders
[prev
],
1739 prog
->_LinkedShaders
[i
])) {
1746 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1747 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1751 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
1752 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
1754 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1755 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
1756 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
1759 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1760 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1762 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1765 /* OpenGL ES requires that a vertex shader and a fragment shader both be
1766 * present in a linked program. By checking for use of shading language
1767 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
1769 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1770 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
1771 linker_error(prog
, "program lacks a vertex shader\n");
1772 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1773 linker_error(prog
, "program lacks a fragment shader\n");
1777 /* FINISHME: Assign fragment shader output locations. */
1780 free(vert_shader_list
);
1782 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1783 if (prog
->_LinkedShaders
[i
] == NULL
)
1786 /* Retain any live IR, but trash the rest. */
1787 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
1790 ralloc_free(mem_ctx
);