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 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
185 ralloc_strcat(&prog
->InfoLog
, "error: ");
187 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
194 invalidate_variable_locations(gl_shader
*sh
, enum ir_variable_mode mode
,
197 foreach_list(node
, sh
->ir
) {
198 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
200 if ((var
== NULL
) || (var
->mode
!= (unsigned) mode
))
203 /* Only assign locations for generic attributes / varyings / etc.
205 if ((var
->location
>= generic_base
) && !var
->explicit_location
)
212 * Determine the number of attribute slots required for a particular type
214 * This code is here because it implements the language rules of a specific
215 * GLSL version. Since it's a property of the language and not a property of
216 * types in general, it doesn't really belong in glsl_type.
219 count_attribute_slots(const glsl_type
*t
)
221 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
223 * "A scalar input counts the same amount against this limit as a vec4,
224 * so applications may want to consider packing groups of four
225 * unrelated float inputs together into a vector to better utilize the
226 * capabilities of the underlying hardware. A matrix input will use up
227 * multiple locations. The number of locations used will equal the
228 * number of columns in the matrix."
230 * The spec does not explicitly say how arrays are counted. However, it
231 * should be safe to assume the total number of slots consumed by an array
232 * is the number of entries in the array multiplied by the number of slots
233 * consumed by a single element of the array.
237 return t
->array_size() * count_attribute_slots(t
->element_type());
240 return t
->matrix_columns
;
247 * Verify that a vertex shader executable meets all semantic requirements
249 * \param shader Vertex shader executable to be verified
252 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
253 struct gl_shader
*shader
)
258 find_assignment_visitor
find("gl_Position");
259 find
.run(shader
->ir
);
260 if (!find
.variable_found()) {
261 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
265 if (prog
->Version
>= 130) {
266 /* From section 7.1 (Vertex Shader Special Variables) of the
269 * "It is an error for a shader to statically write both
270 * gl_ClipVertex and gl_ClipDistance."
272 find_assignment_visitor
clip_vertex("gl_ClipVertex");
273 find_assignment_visitor
clip_distance("gl_ClipDistance");
275 clip_vertex
.run(shader
->ir
);
276 clip_distance
.run(shader
->ir
);
277 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
278 linker_error(prog
, "vertex shader writes to both `gl_ClipVertex' "
279 "and `gl_ClipDistance'\n");
289 * Verify that a fragment shader executable meets all semantic requirements
291 * \param shader Fragment shader executable to be verified
294 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
295 struct gl_shader
*shader
)
300 find_assignment_visitor
frag_color("gl_FragColor");
301 find_assignment_visitor
frag_data("gl_FragData");
303 frag_color
.run(shader
->ir
);
304 frag_data
.run(shader
->ir
);
306 if (frag_color
.variable_found() && frag_data
.variable_found()) {
307 linker_error(prog
, "fragment shader writes to both "
308 "`gl_FragColor' and `gl_FragData'\n");
317 * Generate a string describing the mode of a variable
320 mode_string(const ir_variable
*var
)
324 return (var
->read_only
) ? "global constant" : "global variable";
326 case ir_var_uniform
: return "uniform";
327 case ir_var_in
: return "shader input";
328 case ir_var_out
: return "shader output";
329 case ir_var_inout
: return "shader inout";
331 case ir_var_const_in
:
332 case ir_var_temporary
:
334 assert(!"Should not get here.");
335 return "invalid variable";
341 * Perform validation of global variables used across multiple shaders
344 cross_validate_globals(struct gl_shader_program
*prog
,
345 struct gl_shader
**shader_list
,
346 unsigned num_shaders
,
349 /* Examine all of the uniforms in all of the shaders and cross validate
352 glsl_symbol_table variables
;
353 for (unsigned i
= 0; i
< num_shaders
; i
++) {
354 if (shader_list
[i
] == NULL
)
357 foreach_list(node
, shader_list
[i
]->ir
) {
358 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
363 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
366 /* Don't cross validate temporaries that are at global scope. These
367 * will eventually get pulled into the shaders 'main'.
369 if (var
->mode
== ir_var_temporary
)
372 /* If a global with this name has already been seen, verify that the
373 * new instance has the same type. In addition, if the globals have
374 * initializers, the values of the initializers must be the same.
376 ir_variable
*const existing
= variables
.get_variable(var
->name
);
377 if (existing
!= NULL
) {
378 if (var
->type
!= existing
->type
) {
379 /* Consider the types to be "the same" if both types are arrays
380 * of the same type and one of the arrays is implicitly sized.
381 * In addition, set the type of the linked variable to the
382 * explicitly sized array.
384 if (var
->type
->is_array()
385 && existing
->type
->is_array()
386 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
387 && ((var
->type
->length
== 0)
388 || (existing
->type
->length
== 0))) {
389 if (var
->type
->length
!= 0) {
390 existing
->type
= var
->type
;
393 linker_error(prog
, "%s `%s' declared as type "
394 "`%s' and type `%s'\n",
396 var
->name
, var
->type
->name
,
397 existing
->type
->name
);
402 if (var
->explicit_location
) {
403 if (existing
->explicit_location
404 && (var
->location
!= existing
->location
)) {
405 linker_error(prog
, "explicit locations for %s "
406 "`%s' have differing values\n",
407 mode_string(var
), var
->name
);
411 existing
->location
= var
->location
;
412 existing
->explicit_location
= true;
415 /* Validate layout qualifiers for gl_FragDepth.
417 * From the AMD/ARB_conservative_depth specs:
418 * "If gl_FragDepth is redeclared in any fragment shader in
419 * a program, it must be redeclared in all fragment shaders in that
420 * program that have static assignments to gl_FragDepth. All
421 * redeclarations of gl_FragDepth in all fragment shaders in
422 * a single program must have the same set of qualifiers."
424 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
425 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
426 bool layout_differs
= var
->depth_layout
!= existing
->depth_layout
;
427 if (layout_declared
&& layout_differs
) {
429 "All redeclarations of gl_FragDepth in all fragment shaders "
430 "in a single program must have the same set of qualifiers.");
432 if (var
->used
&& layout_differs
) {
434 "If gl_FragDepth is redeclared with a layout qualifier in"
435 "any fragment shader, it must be redeclared with the same"
436 "layout qualifier in all fragment shaders that have"
437 "assignments to gl_FragDepth");
441 /* FINISHME: Handle non-constant initializers.
443 if (var
->constant_value
!= NULL
) {
444 if (existing
->constant_value
!= NULL
) {
445 if (!var
->constant_value
->has_value(existing
->constant_value
)) {
446 linker_error(prog
, "initializers for %s "
447 "`%s' have differing values\n",
448 mode_string(var
), var
->name
);
452 /* If the first-seen instance of a particular uniform did not
453 * have an initializer but a later instance does, copy the
454 * initializer to the version stored in the symbol table.
456 /* FINISHME: This is wrong. The constant_value field should
457 * FINISHME: not be modified! Imagine a case where a shader
458 * FINISHME: without an initializer is linked in two different
459 * FINISHME: programs with shaders that have differing
460 * FINISHME: initializers. Linking with the first will
461 * FINISHME: modify the shader, and linking with the second
462 * FINISHME: will fail.
464 existing
->constant_value
=
465 var
->constant_value
->clone(ralloc_parent(existing
), NULL
);
468 if (existing
->invariant
!= var
->invariant
) {
469 linker_error(prog
, "declarations for %s `%s' have "
470 "mismatching invariant qualifiers\n",
471 mode_string(var
), var
->name
);
474 if (existing
->centroid
!= var
->centroid
) {
475 linker_error(prog
, "declarations for %s `%s' have "
476 "mismatching centroid qualifiers\n",
477 mode_string(var
), var
->name
);
481 variables
.add_variable(var
);
490 * Perform validation of uniforms used across multiple shader stages
493 cross_validate_uniforms(struct gl_shader_program
*prog
)
495 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
496 MESA_SHADER_TYPES
, true);
501 * Validate that outputs from one stage match inputs of another
504 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
505 gl_shader
*producer
, gl_shader
*consumer
)
507 glsl_symbol_table parameters
;
508 /* FINISHME: Figure these out dynamically. */
509 const char *const producer_stage
= "vertex";
510 const char *const consumer_stage
= "fragment";
512 /* Find all shader outputs in the "producer" stage.
514 foreach_list(node
, producer
->ir
) {
515 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
517 /* FINISHME: For geometry shaders, this should also look for inout
518 * FINISHME: variables.
520 if ((var
== NULL
) || (var
->mode
!= ir_var_out
))
523 parameters
.add_variable(var
);
527 /* Find all shader inputs in the "consumer" stage. Any variables that have
528 * matching outputs already in the symbol table must have the same type and
531 foreach_list(node
, consumer
->ir
) {
532 ir_variable
*const input
= ((ir_instruction
*) node
)->as_variable();
534 /* FINISHME: For geometry shaders, this should also look for inout
535 * FINISHME: variables.
537 if ((input
== NULL
) || (input
->mode
!= ir_var_in
))
540 ir_variable
*const output
= parameters
.get_variable(input
->name
);
541 if (output
!= NULL
) {
542 /* Check that the types match between stages.
544 if (input
->type
!= output
->type
) {
545 /* There is a bit of a special case for gl_TexCoord. This
546 * built-in is unsized by default. Applications that variable
547 * access it must redeclare it with a size. There is some
548 * language in the GLSL spec that implies the fragment shader
549 * and vertex shader do not have to agree on this size. Other
550 * driver behave this way, and one or two applications seem to
553 * Neither declaration needs to be modified here because the array
554 * sizes are fixed later when update_array_sizes is called.
556 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
558 * "Unlike user-defined varying variables, the built-in
559 * varying variables don't have a strict one-to-one
560 * correspondence between the vertex language and the
561 * fragment language."
563 if (!output
->type
->is_array()
564 || (strncmp("gl_", output
->name
, 3) != 0)) {
566 "%s shader output `%s' declared as type `%s', "
567 "but %s shader input declared as type `%s'\n",
568 producer_stage
, output
->name
,
570 consumer_stage
, input
->type
->name
);
575 /* Check that all of the qualifiers match between stages.
577 if (input
->centroid
!= output
->centroid
) {
579 "%s shader output `%s' %s centroid qualifier, "
580 "but %s shader input %s centroid qualifier\n",
583 (output
->centroid
) ? "has" : "lacks",
585 (input
->centroid
) ? "has" : "lacks");
589 if (input
->invariant
!= output
->invariant
) {
591 "%s shader output `%s' %s invariant qualifier, "
592 "but %s shader input %s invariant qualifier\n",
595 (output
->invariant
) ? "has" : "lacks",
597 (input
->invariant
) ? "has" : "lacks");
601 if (input
->interpolation
!= output
->interpolation
) {
603 "%s shader output `%s' specifies %s "
604 "interpolation qualifier, "
605 "but %s shader input specifies %s "
606 "interpolation qualifier\n",
609 output
->interpolation_string(),
611 input
->interpolation_string());
622 * Populates a shaders symbol table with all global declarations
625 populate_symbol_table(gl_shader
*sh
)
627 sh
->symbols
= new(sh
) glsl_symbol_table
;
629 foreach_list(node
, sh
->ir
) {
630 ir_instruction
*const inst
= (ir_instruction
*) node
;
634 if ((func
= inst
->as_function()) != NULL
) {
635 sh
->symbols
->add_function(func
);
636 } else if ((var
= inst
->as_variable()) != NULL
) {
637 sh
->symbols
->add_variable(var
);
644 * Remap variables referenced in an instruction tree
646 * This is used when instruction trees are cloned from one shader and placed in
647 * another. These trees will contain references to \c ir_variable nodes that
648 * do not exist in the target shader. This function finds these \c ir_variable
649 * references and replaces the references with matching variables in the target
652 * If there is no matching variable in the target shader, a clone of the
653 * \c ir_variable is made and added to the target shader. The new variable is
654 * added to \b both the instruction stream and the symbol table.
656 * \param inst IR tree that is to be processed.
657 * \param symbols Symbol table containing global scope symbols in the
659 * \param instructions Instruction stream where new variable declarations
663 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
666 class remap_visitor
: public ir_hierarchical_visitor
{
668 remap_visitor(struct gl_shader
*target
,
671 this->target
= target
;
672 this->symbols
= target
->symbols
;
673 this->instructions
= target
->ir
;
677 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
679 if (ir
->var
->mode
== ir_var_temporary
) {
680 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
684 return visit_continue
;
687 ir_variable
*const existing
=
688 this->symbols
->get_variable(ir
->var
->name
);
689 if (existing
!= NULL
)
692 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
694 this->symbols
->add_variable(copy
);
695 this->instructions
->push_head(copy
);
699 return visit_continue
;
703 struct gl_shader
*target
;
704 glsl_symbol_table
*symbols
;
705 exec_list
*instructions
;
709 remap_visitor
v(target
, temps
);
716 * Move non-declarations from one instruction stream to another
718 * The intended usage pattern of this function is to pass the pointer to the
719 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
720 * pointer) for \c last and \c false for \c make_copies on the first
721 * call. Successive calls pass the return value of the previous call for
722 * \c last and \c true for \c make_copies.
724 * \param instructions Source instruction stream
725 * \param last Instruction after which new instructions should be
726 * inserted in the target instruction stream
727 * \param make_copies Flag selecting whether instructions in \c instructions
728 * should be copied (via \c ir_instruction::clone) into the
729 * target list or moved.
732 * The new "last" instruction in the target instruction stream. This pointer
733 * is suitable for use as the \c last parameter of a later call to this
737 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
738 bool make_copies
, gl_shader
*target
)
740 hash_table
*temps
= NULL
;
743 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
744 hash_table_pointer_compare
);
746 foreach_list_safe(node
, instructions
) {
747 ir_instruction
*inst
= (ir_instruction
*) node
;
749 if (inst
->as_function())
752 ir_variable
*var
= inst
->as_variable();
753 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
756 assert(inst
->as_assignment()
757 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
760 inst
= inst
->clone(target
, NULL
);
763 hash_table_insert(temps
, inst
, var
);
765 remap_variables(inst
, target
, temps
);
770 last
->insert_after(inst
);
775 hash_table_dtor(temps
);
781 * Get the function signature for main from a shader
783 static ir_function_signature
*
784 get_main_function_signature(gl_shader
*sh
)
786 ir_function
*const f
= sh
->symbols
->get_function("main");
788 exec_list void_parameters
;
790 /* Look for the 'void main()' signature and ensure that it's defined.
791 * This keeps the linker from accidentally pick a shader that just
792 * contains a prototype for main.
794 * We don't have to check for multiple definitions of main (in multiple
795 * shaders) because that would have already been caught above.
797 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
798 if ((sig
!= NULL
) && sig
->is_defined
) {
808 * Combine a group of shaders for a single stage to generate a linked shader
811 * If this function is supplied a single shader, it is cloned, and the new
812 * shader is returned.
814 static struct gl_shader
*
815 link_intrastage_shaders(void *mem_ctx
,
816 struct gl_context
*ctx
,
817 struct gl_shader_program
*prog
,
818 struct gl_shader
**shader_list
,
819 unsigned num_shaders
)
821 /* Check that global variables defined in multiple shaders are consistent.
823 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
826 /* Check that there is only a single definition of each function signature
827 * across all shaders.
829 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
830 foreach_list(node
, shader_list
[i
]->ir
) {
831 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
836 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
837 ir_function
*const other
=
838 shader_list
[j
]->symbols
->get_function(f
->name
);
840 /* If the other shader has no function (and therefore no function
841 * signatures) with the same name, skip to the next shader.
846 foreach_iter (exec_list_iterator
, iter
, *f
) {
847 ir_function_signature
*sig
=
848 (ir_function_signature
*) iter
.get();
850 if (!sig
->is_defined
|| sig
->is_builtin
)
853 ir_function_signature
*other_sig
=
854 other
->exact_matching_signature(& sig
->parameters
);
856 if ((other_sig
!= NULL
) && other_sig
->is_defined
857 && !other_sig
->is_builtin
) {
858 linker_error(prog
, "function `%s' is multiply defined",
867 /* Find the shader that defines main, and make a clone of it.
869 * Starting with the clone, search for undefined references. If one is
870 * found, find the shader that defines it. Clone the reference and add
871 * it to the shader. Repeat until there are no undefined references or
872 * until a reference cannot be resolved.
874 gl_shader
*main
= NULL
;
875 for (unsigned i
= 0; i
< num_shaders
; i
++) {
876 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
877 main
= shader_list
[i
];
883 linker_error(prog
, "%s shader lacks `main'\n",
884 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
885 ? "vertex" : "fragment");
889 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
890 linked
->ir
= new(linked
) exec_list
;
891 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
893 populate_symbol_table(linked
);
895 /* The a pointer to the main function in the final linked shader (i.e., the
896 * copy of the original shader that contained the main function).
898 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
900 /* Move any instructions other than variable declarations or function
901 * declarations into main.
903 exec_node
*insertion_point
=
904 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
907 for (unsigned i
= 0; i
< num_shaders
; i
++) {
908 if (shader_list
[i
] == main
)
911 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
912 insertion_point
, true, linked
);
915 /* Resolve initializers for global variables in the linked shader.
917 unsigned num_linking_shaders
= num_shaders
;
918 for (unsigned i
= 0; i
< num_shaders
; i
++)
919 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
921 gl_shader
**linking_shaders
=
922 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
924 memcpy(linking_shaders
, shader_list
,
925 sizeof(linking_shaders
[0]) * num_shaders
);
927 unsigned idx
= num_shaders
;
928 for (unsigned i
= 0; i
< num_shaders
; i
++) {
929 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
930 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
931 idx
+= shader_list
[i
]->num_builtins_to_link
;
934 assert(idx
== num_linking_shaders
);
936 if (!link_function_calls(prog
, linked
, linking_shaders
,
937 num_linking_shaders
)) {
938 ctx
->Driver
.DeleteShader(ctx
, linked
);
942 free(linking_shaders
);
945 /* At this point linked should contain all of the linked IR, so
946 * validate it to make sure nothing went wrong.
949 validate_ir_tree(linked
->ir
);
952 /* Make a pass over all variable declarations to ensure that arrays with
953 * unspecified sizes have a size specified. The size is inferred from the
954 * max_array_access field.
956 if (linked
!= NULL
) {
957 class array_sizing_visitor
: public ir_hierarchical_visitor
{
959 virtual ir_visitor_status
visit(ir_variable
*var
)
961 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
962 const glsl_type
*type
=
963 glsl_type::get_array_instance(var
->type
->fields
.array
,
964 var
->max_array_access
+ 1);
966 assert(type
!= NULL
);
970 return visit_continue
;
981 struct uniform_node
{
983 struct gl_uniform
*u
;
988 * Update the sizes of linked shader uniform arrays to the maximum
991 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
993 * If one or more elements of an array are active,
994 * GetActiveUniform will return the name of the array in name,
995 * subject to the restrictions listed above. The type of the array
996 * is returned in type. The size parameter contains the highest
997 * array element index used, plus one. The compiler or linker
998 * determines the highest index used. There will be only one
999 * active uniform reported by the GL per uniform array.
1003 update_array_sizes(struct gl_shader_program
*prog
)
1005 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1006 if (prog
->_LinkedShaders
[i
] == NULL
)
1009 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1010 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1012 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1013 var
->mode
!= ir_var_in
&&
1014 var
->mode
!= ir_var_out
) ||
1015 !var
->type
->is_array())
1018 unsigned int size
= var
->max_array_access
;
1019 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1020 if (prog
->_LinkedShaders
[j
] == NULL
)
1023 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1024 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1028 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1029 other_var
->max_array_access
> size
) {
1030 size
= other_var
->max_array_access
;
1035 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1036 /* If this is a built-in uniform (i.e., it's backed by some
1037 * fixed-function state), adjust the number of state slots to
1038 * match the new array size. The number of slots per array entry
1039 * is not known. It seems safe to assume that the total number of
1040 * slots is an integer multiple of the number of array elements.
1041 * Determine the number of slots per array element by dividing by
1042 * the old (total) size.
1044 if (var
->num_state_slots
> 0) {
1045 var
->num_state_slots
= (size
+ 1)
1046 * (var
->num_state_slots
/ var
->type
->length
);
1049 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1051 /* FINISHME: We should update the types of array
1052 * dereferences of this variable now.
1060 add_uniform(void *mem_ctx
, exec_list
*uniforms
, struct hash_table
*ht
,
1061 const char *name
, const glsl_type
*type
, GLenum shader_type
,
1062 unsigned *next_shader_pos
, unsigned *total_uniforms
)
1064 if (type
->is_record()) {
1065 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1066 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
1067 char *field_name
= ralloc_asprintf(mem_ctx
, "%s.%s", name
,
1068 type
->fields
.structure
[i
].name
);
1070 add_uniform(mem_ctx
, uniforms
, ht
, field_name
, field_type
,
1071 shader_type
, next_shader_pos
, total_uniforms
);
1074 uniform_node
*n
= (uniform_node
*) hash_table_find(ht
, name
);
1075 unsigned int vec4_slots
;
1076 const glsl_type
*array_elem_type
= NULL
;
1078 if (type
->is_array()) {
1079 array_elem_type
= type
->fields
.array
;
1080 /* Array of structures. */
1081 if (array_elem_type
->is_record()) {
1082 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1083 char *elem_name
= ralloc_asprintf(mem_ctx
, "%s[%d]", name
, i
);
1084 add_uniform(mem_ctx
, uniforms
, ht
, elem_name
, array_elem_type
,
1085 shader_type
, next_shader_pos
, total_uniforms
);
1091 /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
1092 * vectors to vec4 slots.
1094 if (type
->is_array()) {
1095 if (array_elem_type
->is_sampler())
1096 vec4_slots
= type
->length
;
1098 vec4_slots
= type
->length
* array_elem_type
->matrix_columns
;
1099 } else if (type
->is_sampler()) {
1102 vec4_slots
= type
->matrix_columns
;
1106 n
= (uniform_node
*) calloc(1, sizeof(struct uniform_node
));
1107 n
->u
= (gl_uniform
*) calloc(1, sizeof(struct gl_uniform
));
1108 n
->slots
= vec4_slots
;
1110 n
->u
->Name
= strdup(name
);
1115 (*total_uniforms
)++;
1117 hash_table_insert(ht
, n
, name
);
1118 uniforms
->push_tail(& n
->link
);
1121 switch (shader_type
) {
1122 case GL_VERTEX_SHADER
:
1123 n
->u
->VertPos
= *next_shader_pos
;
1125 case GL_FRAGMENT_SHADER
:
1126 n
->u
->FragPos
= *next_shader_pos
;
1128 case GL_GEOMETRY_SHADER
:
1129 n
->u
->GeomPos
= *next_shader_pos
;
1133 (*next_shader_pos
) += vec4_slots
;
1138 assign_uniform_locations(struct gl_shader_program
*prog
)
1142 unsigned total_uniforms
= 0;
1143 hash_table
*ht
= hash_table_ctor(32, hash_table_string_hash
,
1144 hash_table_string_compare
);
1145 void *mem_ctx
= ralloc_context(NULL
);
1147 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1148 if (prog
->_LinkedShaders
[i
] == NULL
)
1151 unsigned next_position
= 0;
1153 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1154 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1156 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
))
1159 if (strncmp(var
->name
, "gl_", 3) == 0) {
1160 /* At the moment, we don't allocate uniform locations for
1161 * builtin uniforms. It's permitted by spec, and we'll
1162 * likely switch to doing that at some point, but not yet.
1167 var
->location
= next_position
;
1168 add_uniform(mem_ctx
, &uniforms
, ht
, var
->name
, var
->type
,
1169 prog
->_LinkedShaders
[i
]->Type
,
1170 &next_position
, &total_uniforms
);
1174 ralloc_free(mem_ctx
);
1176 gl_uniform_list
*ul
= (gl_uniform_list
*)
1177 calloc(1, sizeof(gl_uniform_list
));
1179 ul
->Size
= total_uniforms
;
1180 ul
->NumUniforms
= total_uniforms
;
1181 ul
->Uniforms
= (gl_uniform
*) calloc(total_uniforms
, sizeof(gl_uniform
));
1185 for (uniform_node
*node
= (uniform_node
*) uniforms
.head
1186 ; node
->link
.next
!= NULL
1188 next
= (uniform_node
*) node
->link
.next
;
1190 node
->link
.remove();
1191 memcpy(&ul
->Uniforms
[idx
], node
->u
, sizeof(gl_uniform
));
1198 hash_table_dtor(ht
);
1200 prog
->Uniforms
= ul
;
1205 * Find a contiguous set of available bits in a bitmask.
1207 * \param used_mask Bits representing used (1) and unused (0) locations
1208 * \param needed_count Number of contiguous bits needed.
1211 * Base location of the available bits on success or -1 on failure.
1214 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1216 unsigned needed_mask
= (1 << needed_count
) - 1;
1217 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1219 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1220 * cannot optimize possibly infinite loops" for the loop below.
1222 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1225 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1226 if ((needed_mask
& ~used_mask
) == needed_mask
)
1237 * Assign locations for either VS inputs for FS outputs
1239 * \param prog Shader program whose variables need locations assigned
1240 * \param target_index Selector for the program target to receive location
1241 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1242 * \c MESA_SHADER_FRAGMENT.
1243 * \param max_index Maximum number of generic locations. This corresponds
1244 * to either the maximum number of draw buffers or the
1245 * maximum number of generic attributes.
1248 * If locations are successfully assigned, true is returned. Otherwise an
1249 * error is emitted to the shader link log and false is returned.
1252 * Locations set via \c glBindFragDataLocation are not currently supported.
1253 * Only locations assigned automatically by the linker, explicitly set by a
1254 * layout qualifier, or explicitly set by a built-in variable (e.g., \c
1255 * gl_FragColor) are supported for fragment shaders.
1258 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1259 unsigned target_index
,
1262 /* Mark invalid locations as being used.
1264 unsigned used_locations
= (max_index
>= 32)
1265 ? ~0 : ~((1 << max_index
) - 1);
1267 assert((target_index
== MESA_SHADER_VERTEX
)
1268 || (target_index
== MESA_SHADER_FRAGMENT
));
1270 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1274 /* Operate in a total of four passes.
1276 * 1. Invalidate the location assignments for all vertex shader inputs.
1278 * 2. Assign locations for inputs that have user-defined (via
1279 * glBindVertexAttribLocation) locations.
1281 * 3. Sort the attributes without assigned locations by number of slots
1282 * required in decreasing order. Fragmentation caused by attribute
1283 * locations assigned by the application may prevent large attributes
1284 * from having enough contiguous space.
1286 * 4. Assign locations to any inputs without assigned locations.
1289 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1290 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1292 const enum ir_variable_mode direction
=
1293 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1296 invalidate_variable_locations(sh
, direction
, generic_base
);
1298 if ((target_index
== MESA_SHADER_VERTEX
) && (prog
->Attributes
!= NULL
)) {
1299 for (unsigned i
= 0; i
< prog
->Attributes
->NumParameters
; i
++) {
1300 ir_variable
*const var
=
1301 sh
->symbols
->get_variable(prog
->Attributes
->Parameters
[i
].Name
);
1303 /* Note: attributes that occupy multiple slots, such as arrays or
1304 * matrices, may appear in the attrib array multiple times.
1306 if ((var
== NULL
) || (var
->location
!= -1))
1309 /* From page 61 of the OpenGL 4.0 spec:
1311 * "LinkProgram will fail if the attribute bindings assigned by
1312 * BindAttribLocation do not leave not enough space to assign a
1313 * location for an active matrix attribute or an active attribute
1314 * array, both of which require multiple contiguous generic
1317 * Previous versions of the spec contain similar language but omit the
1318 * bit about attribute arrays.
1320 * Page 61 of the OpenGL 4.0 spec also says:
1322 * "It is possible for an application to bind more than one
1323 * attribute name to the same location. This is referred to as
1324 * aliasing. This will only work if only one of the aliased
1325 * attributes is active in the executable program, or if no path
1326 * through the shader consumes more than one attribute of a set
1327 * of attributes aliased to the same location. A link error can
1328 * occur if the linker determines that every path through the
1329 * shader consumes multiple aliased attributes, but
1330 * implementations are not required to generate an error in this
1333 * These two paragraphs are either somewhat contradictory, or I don't
1334 * fully understand one or both of them.
1336 /* FINISHME: The code as currently written does not support attribute
1337 * FINISHME: location aliasing (see comment above).
1339 const int attr
= prog
->Attributes
->Parameters
[i
].StateIndexes
[0];
1340 const unsigned slots
= count_attribute_slots(var
->type
);
1342 /* Mask representing the contiguous slots that will be used by this
1345 const unsigned use_mask
= (1 << slots
) - 1;
1347 /* Generate a link error if the set of bits requested for this
1348 * attribute overlaps any previously allocated bits.
1350 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1352 "insufficient contiguous attribute locations "
1353 "available for vertex shader input `%s'",
1358 var
->location
= VERT_ATTRIB_GENERIC0
+ attr
;
1359 used_locations
|= (use_mask
<< attr
);
1363 /* Temporary storage for the set of attributes that need locations assigned.
1369 /* Used below in the call to qsort. */
1370 static int compare(const void *a
, const void *b
)
1372 const temp_attr
*const l
= (const temp_attr
*) a
;
1373 const temp_attr
*const r
= (const temp_attr
*) b
;
1375 /* Reversed because we want a descending order sort below. */
1376 return r
->slots
- l
->slots
;
1380 unsigned num_attr
= 0;
1382 foreach_list(node
, sh
->ir
) {
1383 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1385 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1388 if (var
->explicit_location
) {
1389 const unsigned slots
= count_attribute_slots(var
->type
);
1390 const unsigned use_mask
= (1 << slots
) - 1;
1391 const int attr
= var
->location
- generic_base
;
1393 if ((var
->location
>= (int)(max_index
+ generic_base
))
1394 || (var
->location
< 0)) {
1396 "invalid explicit location %d specified for `%s'\n",
1397 (var
->location
< 0) ? var
->location
: attr
,
1400 } else if (var
->location
>= generic_base
) {
1401 used_locations
|= (use_mask
<< attr
);
1405 /* The location was explicitly assigned, nothing to do here.
1407 if (var
->location
!= -1)
1410 to_assign
[num_attr
].slots
= count_attribute_slots(var
->type
);
1411 to_assign
[num_attr
].var
= var
;
1415 /* If all of the attributes were assigned locations by the application (or
1416 * are built-in attributes with fixed locations), return early. This should
1417 * be the common case.
1422 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1424 if (target_index
== MESA_SHADER_VERTEX
) {
1425 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1426 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1427 * reserved to prevent it from being automatically allocated below.
1429 find_deref_visitor
find("gl_Vertex");
1431 if (find
.variable_found())
1432 used_locations
|= (1 << 0);
1435 for (unsigned i
= 0; i
< num_attr
; i
++) {
1436 /* Mask representing the contiguous slots that will be used by this
1439 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1441 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1444 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1445 ? "vertex shader input" : "fragment shader output";
1448 "insufficient contiguous attribute locations "
1449 "available for %s `%s'",
1450 string
, to_assign
[i
].var
->name
);
1454 to_assign
[i
].var
->location
= generic_base
+ location
;
1455 used_locations
|= (use_mask
<< location
);
1463 * Demote shader inputs and outputs that are not used in other stages
1466 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1468 foreach_list(node
, sh
->ir
) {
1469 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1471 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1474 /* A shader 'in' or 'out' variable is only really an input or output if
1475 * its value is used by other shader stages. This will cause the variable
1476 * to have a location assigned.
1478 if (var
->location
== -1) {
1479 var
->mode
= ir_var_auto
;
1486 assign_varying_locations(struct gl_context
*ctx
,
1487 struct gl_shader_program
*prog
,
1488 gl_shader
*producer
, gl_shader
*consumer
)
1490 /* FINISHME: Set dynamically when geometry shader support is added. */
1491 unsigned output_index
= VERT_RESULT_VAR0
;
1492 unsigned input_index
= FRAG_ATTRIB_VAR0
;
1494 /* Operate in a total of three passes.
1496 * 1. Assign locations for any matching inputs and outputs.
1498 * 2. Mark output variables in the producer that do not have locations as
1499 * not being outputs. This lets the optimizer eliminate them.
1501 * 3. Mark input variables in the consumer that do not have locations as
1502 * not being inputs. This lets the optimizer eliminate them.
1505 invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
1506 invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
1508 foreach_list(node
, producer
->ir
) {
1509 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
1511 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
)
1512 || (output_var
->location
!= -1))
1515 ir_variable
*const input_var
=
1516 consumer
->symbols
->get_variable(output_var
->name
);
1518 if ((input_var
== NULL
) || (input_var
->mode
!= ir_var_in
))
1521 assert(input_var
->location
== -1);
1523 output_var
->location
= output_index
;
1524 input_var
->location
= input_index
;
1526 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1527 assert(!output_var
->type
->is_record());
1529 if (output_var
->type
->is_array()) {
1530 const unsigned slots
= output_var
->type
->length
1531 * output_var
->type
->fields
.array
->matrix_columns
;
1533 output_index
+= slots
;
1534 input_index
+= slots
;
1536 const unsigned slots
= output_var
->type
->matrix_columns
;
1538 output_index
+= slots
;
1539 input_index
+= slots
;
1543 unsigned varying_vectors
= 0;
1545 foreach_list(node
, consumer
->ir
) {
1546 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1548 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
1551 if (var
->location
== -1) {
1552 if (prog
->Version
<= 120) {
1553 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1555 * Only those varying variables used (i.e. read) in
1556 * the fragment shader executable must be written to
1557 * by the vertex shader executable; declaring
1558 * superfluous varying variables in a vertex shader is
1561 * We interpret this text as meaning that the VS must
1562 * write the variable for the FS to read it. See
1563 * "glsl1-varying read but not written" in piglit.
1566 linker_error(prog
, "fragment shader varying %s not written "
1567 "by vertex shader\n.", var
->name
);
1570 /* An 'in' variable is only really a shader input if its
1571 * value is written by the previous stage.
1573 var
->mode
= ir_var_auto
;
1575 /* The packing rules are used for vertex shader inputs are also used
1576 * for fragment shader inputs.
1578 varying_vectors
+= count_attribute_slots(var
->type
);
1582 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1583 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
1584 linker_error(prog
, "shader uses too many varying vectors "
1586 varying_vectors
, ctx
->Const
.MaxVarying
);
1590 const unsigned float_components
= varying_vectors
* 4;
1591 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
1592 linker_error(prog
, "shader uses too many varying components "
1594 float_components
, ctx
->Const
.MaxVarying
* 4);
1604 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1606 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1608 prog
->LinkStatus
= false;
1609 prog
->Validated
= false;
1610 prog
->_Used
= false;
1612 if (prog
->InfoLog
!= NULL
)
1613 ralloc_free(prog
->InfoLog
);
1615 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1617 /* Separate the shaders into groups based on their type.
1619 struct gl_shader
**vert_shader_list
;
1620 unsigned num_vert_shaders
= 0;
1621 struct gl_shader
**frag_shader_list
;
1622 unsigned num_frag_shaders
= 0;
1624 vert_shader_list
= (struct gl_shader
**)
1625 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
1626 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
1628 unsigned min_version
= UINT_MAX
;
1629 unsigned max_version
= 0;
1630 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1631 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1632 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1634 switch (prog
->Shaders
[i
]->Type
) {
1635 case GL_VERTEX_SHADER
:
1636 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1639 case GL_FRAGMENT_SHADER
:
1640 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1643 case GL_GEOMETRY_SHADER
:
1644 /* FINISHME: Support geometry shaders. */
1645 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
1650 /* Previous to GLSL version 1.30, different compilation units could mix and
1651 * match shading language versions. With GLSL 1.30 and later, the versions
1652 * of all shaders must match.
1654 assert(min_version
>= 100);
1655 assert(max_version
<= 130);
1656 if ((max_version
>= 130 || min_version
== 100)
1657 && min_version
!= max_version
) {
1658 linker_error(prog
, "all shaders must use same shading "
1659 "language version\n");
1663 prog
->Version
= max_version
;
1665 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1666 if (prog
->_LinkedShaders
[i
] != NULL
)
1667 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1669 prog
->_LinkedShaders
[i
] = NULL
;
1672 /* Link all shaders for a particular stage and validate the result.
1674 if (num_vert_shaders
> 0) {
1675 gl_shader
*const sh
=
1676 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1682 if (!validate_vertex_shader_executable(prog
, sh
))
1685 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1689 if (num_frag_shaders
> 0) {
1690 gl_shader
*const sh
=
1691 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1697 if (!validate_fragment_shader_executable(prog
, sh
))
1700 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1704 /* Here begins the inter-stage linking phase. Some initial validation is
1705 * performed, then locations are assigned for uniforms, attributes, and
1708 if (cross_validate_uniforms(prog
)) {
1711 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1712 if (prog
->_LinkedShaders
[prev
] != NULL
)
1716 /* Validate the inputs of each stage with the output of the preceding
1719 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1720 if (prog
->_LinkedShaders
[i
] == NULL
)
1723 if (!cross_validate_outputs_to_inputs(prog
,
1724 prog
->_LinkedShaders
[prev
],
1725 prog
->_LinkedShaders
[i
]))
1731 prog
->LinkStatus
= true;
1734 /* Do common optimization before assigning storage for attributes,
1735 * uniforms, and varyings. Later optimization could possibly make
1736 * some of that unused.
1738 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1739 if (prog
->_LinkedShaders
[i
] == NULL
)
1742 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
1743 if (!prog
->LinkStatus
)
1746 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, 32))
1750 update_array_sizes(prog
);
1752 assign_uniform_locations(prog
);
1754 /* FINISHME: The value of the max_attribute_index parameter is
1755 * FINISHME: implementation dependent based on the value of
1756 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1757 * FINISHME: at least 16, so hardcode 16 for now.
1759 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
1763 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, ctx
->Const
.MaxDrawBuffers
)) {
1768 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1769 if (prog
->_LinkedShaders
[prev
] != NULL
)
1773 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1774 if (prog
->_LinkedShaders
[i
] == NULL
)
1777 if (!assign_varying_locations(ctx
, prog
,
1778 prog
->_LinkedShaders
[prev
],
1779 prog
->_LinkedShaders
[i
])) {
1786 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1787 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1791 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
1792 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
1794 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1795 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
1796 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
1799 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1800 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1802 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1805 /* OpenGL ES requires that a vertex shader and a fragment shader both be
1806 * present in a linked program. By checking for use of shading language
1807 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
1809 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1810 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
1811 linker_error(prog
, "program lacks a vertex shader\n");
1812 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1813 linker_error(prog
, "program lacks a fragment shader\n");
1817 /* FINISHME: Assign fragment shader output locations. */
1820 free(vert_shader_list
);
1822 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1823 if (prog
->_LinkedShaders
[i
] == NULL
)
1826 /* Retain any live IR, but trash the rest. */
1827 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
1830 ralloc_free(mem_ctx
);