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 * Also sets prog->Vert.UsesClipDistance as a side effect.
251 * \param shader Vertex shader executable to be verified
254 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
255 struct gl_shader
*shader
)
260 find_assignment_visitor
find("gl_Position");
261 find
.run(shader
->ir
);
262 if (!find
.variable_found()) {
263 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
267 if (prog
->Version
>= 130) {
268 /* From section 7.1 (Vertex Shader Special Variables) of the
271 * "It is an error for a shader to statically write both
272 * gl_ClipVertex and gl_ClipDistance."
274 find_assignment_visitor
clip_vertex("gl_ClipVertex");
275 find_assignment_visitor
clip_distance("gl_ClipDistance");
277 clip_vertex
.run(shader
->ir
);
278 clip_distance
.run(shader
->ir
);
279 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
280 linker_error(prog
, "vertex shader writes to both `gl_ClipVertex' "
281 "and `gl_ClipDistance'\n");
284 prog
->Vert
.UsesClipDistance
= clip_distance
.variable_found();
292 * Verify that a fragment shader executable meets all semantic requirements
294 * \param shader Fragment shader executable to be verified
297 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
298 struct gl_shader
*shader
)
303 find_assignment_visitor
frag_color("gl_FragColor");
304 find_assignment_visitor
frag_data("gl_FragData");
306 frag_color
.run(shader
->ir
);
307 frag_data
.run(shader
->ir
);
309 if (frag_color
.variable_found() && frag_data
.variable_found()) {
310 linker_error(prog
, "fragment shader writes to both "
311 "`gl_FragColor' and `gl_FragData'\n");
320 * Generate a string describing the mode of a variable
323 mode_string(const ir_variable
*var
)
327 return (var
->read_only
) ? "global constant" : "global variable";
329 case ir_var_uniform
: return "uniform";
330 case ir_var_in
: return "shader input";
331 case ir_var_out
: return "shader output";
332 case ir_var_inout
: return "shader inout";
334 case ir_var_const_in
:
335 case ir_var_temporary
:
337 assert(!"Should not get here.");
338 return "invalid variable";
344 * Perform validation of global variables used across multiple shaders
347 cross_validate_globals(struct gl_shader_program
*prog
,
348 struct gl_shader
**shader_list
,
349 unsigned num_shaders
,
352 /* Examine all of the uniforms in all of the shaders and cross validate
355 glsl_symbol_table variables
;
356 for (unsigned i
= 0; i
< num_shaders
; i
++) {
357 if (shader_list
[i
] == NULL
)
360 foreach_list(node
, shader_list
[i
]->ir
) {
361 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
366 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
369 /* Don't cross validate temporaries that are at global scope. These
370 * will eventually get pulled into the shaders 'main'.
372 if (var
->mode
== ir_var_temporary
)
375 /* If a global with this name has already been seen, verify that the
376 * new instance has the same type. In addition, if the globals have
377 * initializers, the values of the initializers must be the same.
379 ir_variable
*const existing
= variables
.get_variable(var
->name
);
380 if (existing
!= NULL
) {
381 if (var
->type
!= existing
->type
) {
382 /* Consider the types to be "the same" if both types are arrays
383 * of the same type and one of the arrays is implicitly sized.
384 * In addition, set the type of the linked variable to the
385 * explicitly sized array.
387 if (var
->type
->is_array()
388 && existing
->type
->is_array()
389 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
390 && ((var
->type
->length
== 0)
391 || (existing
->type
->length
== 0))) {
392 if (var
->type
->length
!= 0) {
393 existing
->type
= var
->type
;
396 linker_error(prog
, "%s `%s' declared as type "
397 "`%s' and type `%s'\n",
399 var
->name
, var
->type
->name
,
400 existing
->type
->name
);
405 if (var
->explicit_location
) {
406 if (existing
->explicit_location
407 && (var
->location
!= existing
->location
)) {
408 linker_error(prog
, "explicit locations for %s "
409 "`%s' have differing values\n",
410 mode_string(var
), var
->name
);
414 existing
->location
= var
->location
;
415 existing
->explicit_location
= true;
418 /* Validate layout qualifiers for gl_FragDepth.
420 * From the AMD/ARB_conservative_depth specs:
421 * "If gl_FragDepth is redeclared in any fragment shader in
422 * a program, it must be redeclared in all fragment shaders in that
423 * program that have static assignments to gl_FragDepth. All
424 * redeclarations of gl_FragDepth in all fragment shaders in
425 * a single program must have the same set of qualifiers."
427 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
428 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
429 bool layout_differs
= var
->depth_layout
!= existing
->depth_layout
;
430 if (layout_declared
&& layout_differs
) {
432 "All redeclarations of gl_FragDepth in all fragment shaders "
433 "in a single program must have the same set of qualifiers.");
435 if (var
->used
&& layout_differs
) {
437 "If gl_FragDepth is redeclared with a layout qualifier in"
438 "any fragment shader, it must be redeclared with the same"
439 "layout qualifier in all fragment shaders that have"
440 "assignments to gl_FragDepth");
444 /* FINISHME: Handle non-constant initializers.
446 if (var
->constant_value
!= NULL
) {
447 if (existing
->constant_value
!= NULL
) {
448 if (!var
->constant_value
->has_value(existing
->constant_value
)) {
449 linker_error(prog
, "initializers for %s "
450 "`%s' have differing values\n",
451 mode_string(var
), var
->name
);
455 /* If the first-seen instance of a particular uniform did not
456 * have an initializer but a later instance does, copy the
457 * initializer to the version stored in the symbol table.
459 /* FINISHME: This is wrong. The constant_value field should
460 * FINISHME: not be modified! Imagine a case where a shader
461 * FINISHME: without an initializer is linked in two different
462 * FINISHME: programs with shaders that have differing
463 * FINISHME: initializers. Linking with the first will
464 * FINISHME: modify the shader, and linking with the second
465 * FINISHME: will fail.
467 existing
->constant_value
=
468 var
->constant_value
->clone(ralloc_parent(existing
), NULL
);
471 if (existing
->invariant
!= var
->invariant
) {
472 linker_error(prog
, "declarations for %s `%s' have "
473 "mismatching invariant qualifiers\n",
474 mode_string(var
), var
->name
);
477 if (existing
->centroid
!= var
->centroid
) {
478 linker_error(prog
, "declarations for %s `%s' have "
479 "mismatching centroid qualifiers\n",
480 mode_string(var
), var
->name
);
484 variables
.add_variable(var
);
493 * Perform validation of uniforms used across multiple shader stages
496 cross_validate_uniforms(struct gl_shader_program
*prog
)
498 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
499 MESA_SHADER_TYPES
, true);
504 * Validate that outputs from one stage match inputs of another
507 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
508 gl_shader
*producer
, gl_shader
*consumer
)
510 glsl_symbol_table parameters
;
511 /* FINISHME: Figure these out dynamically. */
512 const char *const producer_stage
= "vertex";
513 const char *const consumer_stage
= "fragment";
515 /* Find all shader outputs in the "producer" stage.
517 foreach_list(node
, producer
->ir
) {
518 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
520 /* FINISHME: For geometry shaders, this should also look for inout
521 * FINISHME: variables.
523 if ((var
== NULL
) || (var
->mode
!= ir_var_out
))
526 parameters
.add_variable(var
);
530 /* Find all shader inputs in the "consumer" stage. Any variables that have
531 * matching outputs already in the symbol table must have the same type and
534 foreach_list(node
, consumer
->ir
) {
535 ir_variable
*const input
= ((ir_instruction
*) node
)->as_variable();
537 /* FINISHME: For geometry shaders, this should also look for inout
538 * FINISHME: variables.
540 if ((input
== NULL
) || (input
->mode
!= ir_var_in
))
543 ir_variable
*const output
= parameters
.get_variable(input
->name
);
544 if (output
!= NULL
) {
545 /* Check that the types match between stages.
547 if (input
->type
!= output
->type
) {
548 /* There is a bit of a special case for gl_TexCoord. This
549 * built-in is unsized by default. Applications that variable
550 * access it must redeclare it with a size. There is some
551 * language in the GLSL spec that implies the fragment shader
552 * and vertex shader do not have to agree on this size. Other
553 * driver behave this way, and one or two applications seem to
556 * Neither declaration needs to be modified here because the array
557 * sizes are fixed later when update_array_sizes is called.
559 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
561 * "Unlike user-defined varying variables, the built-in
562 * varying variables don't have a strict one-to-one
563 * correspondence between the vertex language and the
564 * fragment language."
566 if (!output
->type
->is_array()
567 || (strncmp("gl_", output
->name
, 3) != 0)) {
569 "%s shader output `%s' declared as type `%s', "
570 "but %s shader input declared as type `%s'\n",
571 producer_stage
, output
->name
,
573 consumer_stage
, input
->type
->name
);
578 /* Check that all of the qualifiers match between stages.
580 if (input
->centroid
!= output
->centroid
) {
582 "%s shader output `%s' %s centroid qualifier, "
583 "but %s shader input %s centroid qualifier\n",
586 (output
->centroid
) ? "has" : "lacks",
588 (input
->centroid
) ? "has" : "lacks");
592 if (input
->invariant
!= output
->invariant
) {
594 "%s shader output `%s' %s invariant qualifier, "
595 "but %s shader input %s invariant qualifier\n",
598 (output
->invariant
) ? "has" : "lacks",
600 (input
->invariant
) ? "has" : "lacks");
604 if (input
->interpolation
!= output
->interpolation
) {
606 "%s shader output `%s' specifies %s "
607 "interpolation qualifier, "
608 "but %s shader input specifies %s "
609 "interpolation qualifier\n",
612 output
->interpolation_string(),
614 input
->interpolation_string());
625 * Populates a shaders symbol table with all global declarations
628 populate_symbol_table(gl_shader
*sh
)
630 sh
->symbols
= new(sh
) glsl_symbol_table
;
632 foreach_list(node
, sh
->ir
) {
633 ir_instruction
*const inst
= (ir_instruction
*) node
;
637 if ((func
= inst
->as_function()) != NULL
) {
638 sh
->symbols
->add_function(func
);
639 } else if ((var
= inst
->as_variable()) != NULL
) {
640 sh
->symbols
->add_variable(var
);
647 * Remap variables referenced in an instruction tree
649 * This is used when instruction trees are cloned from one shader and placed in
650 * another. These trees will contain references to \c ir_variable nodes that
651 * do not exist in the target shader. This function finds these \c ir_variable
652 * references and replaces the references with matching variables in the target
655 * If there is no matching variable in the target shader, a clone of the
656 * \c ir_variable is made and added to the target shader. The new variable is
657 * added to \b both the instruction stream and the symbol table.
659 * \param inst IR tree that is to be processed.
660 * \param symbols Symbol table containing global scope symbols in the
662 * \param instructions Instruction stream where new variable declarations
666 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
669 class remap_visitor
: public ir_hierarchical_visitor
{
671 remap_visitor(struct gl_shader
*target
,
674 this->target
= target
;
675 this->symbols
= target
->symbols
;
676 this->instructions
= target
->ir
;
680 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
682 if (ir
->var
->mode
== ir_var_temporary
) {
683 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
687 return visit_continue
;
690 ir_variable
*const existing
=
691 this->symbols
->get_variable(ir
->var
->name
);
692 if (existing
!= NULL
)
695 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
697 this->symbols
->add_variable(copy
);
698 this->instructions
->push_head(copy
);
702 return visit_continue
;
706 struct gl_shader
*target
;
707 glsl_symbol_table
*symbols
;
708 exec_list
*instructions
;
712 remap_visitor
v(target
, temps
);
719 * Move non-declarations from one instruction stream to another
721 * The intended usage pattern of this function is to pass the pointer to the
722 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
723 * pointer) for \c last and \c false for \c make_copies on the first
724 * call. Successive calls pass the return value of the previous call for
725 * \c last and \c true for \c make_copies.
727 * \param instructions Source instruction stream
728 * \param last Instruction after which new instructions should be
729 * inserted in the target instruction stream
730 * \param make_copies Flag selecting whether instructions in \c instructions
731 * should be copied (via \c ir_instruction::clone) into the
732 * target list or moved.
735 * The new "last" instruction in the target instruction stream. This pointer
736 * is suitable for use as the \c last parameter of a later call to this
740 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
741 bool make_copies
, gl_shader
*target
)
743 hash_table
*temps
= NULL
;
746 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
747 hash_table_pointer_compare
);
749 foreach_list_safe(node
, instructions
) {
750 ir_instruction
*inst
= (ir_instruction
*) node
;
752 if (inst
->as_function())
755 ir_variable
*var
= inst
->as_variable();
756 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
759 assert(inst
->as_assignment()
760 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
763 inst
= inst
->clone(target
, NULL
);
766 hash_table_insert(temps
, inst
, var
);
768 remap_variables(inst
, target
, temps
);
773 last
->insert_after(inst
);
778 hash_table_dtor(temps
);
784 * Get the function signature for main from a shader
786 static ir_function_signature
*
787 get_main_function_signature(gl_shader
*sh
)
789 ir_function
*const f
= sh
->symbols
->get_function("main");
791 exec_list void_parameters
;
793 /* Look for the 'void main()' signature and ensure that it's defined.
794 * This keeps the linker from accidentally pick a shader that just
795 * contains a prototype for main.
797 * We don't have to check for multiple definitions of main (in multiple
798 * shaders) because that would have already been caught above.
800 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
801 if ((sig
!= NULL
) && sig
->is_defined
) {
811 * Combine a group of shaders for a single stage to generate a linked shader
814 * If this function is supplied a single shader, it is cloned, and the new
815 * shader is returned.
817 static struct gl_shader
*
818 link_intrastage_shaders(void *mem_ctx
,
819 struct gl_context
*ctx
,
820 struct gl_shader_program
*prog
,
821 struct gl_shader
**shader_list
,
822 unsigned num_shaders
)
824 /* Check that global variables defined in multiple shaders are consistent.
826 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
829 /* Check that there is only a single definition of each function signature
830 * across all shaders.
832 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
833 foreach_list(node
, shader_list
[i
]->ir
) {
834 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
839 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
840 ir_function
*const other
=
841 shader_list
[j
]->symbols
->get_function(f
->name
);
843 /* If the other shader has no function (and therefore no function
844 * signatures) with the same name, skip to the next shader.
849 foreach_iter (exec_list_iterator
, iter
, *f
) {
850 ir_function_signature
*sig
=
851 (ir_function_signature
*) iter
.get();
853 if (!sig
->is_defined
|| sig
->is_builtin
)
856 ir_function_signature
*other_sig
=
857 other
->exact_matching_signature(& sig
->parameters
);
859 if ((other_sig
!= NULL
) && other_sig
->is_defined
860 && !other_sig
->is_builtin
) {
861 linker_error(prog
, "function `%s' is multiply defined",
870 /* Find the shader that defines main, and make a clone of it.
872 * Starting with the clone, search for undefined references. If one is
873 * found, find the shader that defines it. Clone the reference and add
874 * it to the shader. Repeat until there are no undefined references or
875 * until a reference cannot be resolved.
877 gl_shader
*main
= NULL
;
878 for (unsigned i
= 0; i
< num_shaders
; i
++) {
879 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
880 main
= shader_list
[i
];
886 linker_error(prog
, "%s shader lacks `main'\n",
887 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
888 ? "vertex" : "fragment");
892 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
893 linked
->ir
= new(linked
) exec_list
;
894 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
896 populate_symbol_table(linked
);
898 /* The a pointer to the main function in the final linked shader (i.e., the
899 * copy of the original shader that contained the main function).
901 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
903 /* Move any instructions other than variable declarations or function
904 * declarations into main.
906 exec_node
*insertion_point
=
907 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
910 for (unsigned i
= 0; i
< num_shaders
; i
++) {
911 if (shader_list
[i
] == main
)
914 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
915 insertion_point
, true, linked
);
918 /* Resolve initializers for global variables in the linked shader.
920 unsigned num_linking_shaders
= num_shaders
;
921 for (unsigned i
= 0; i
< num_shaders
; i
++)
922 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
924 gl_shader
**linking_shaders
=
925 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
927 memcpy(linking_shaders
, shader_list
,
928 sizeof(linking_shaders
[0]) * num_shaders
);
930 unsigned idx
= num_shaders
;
931 for (unsigned i
= 0; i
< num_shaders
; i
++) {
932 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
933 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
934 idx
+= shader_list
[i
]->num_builtins_to_link
;
937 assert(idx
== num_linking_shaders
);
939 if (!link_function_calls(prog
, linked
, linking_shaders
,
940 num_linking_shaders
)) {
941 ctx
->Driver
.DeleteShader(ctx
, linked
);
945 free(linking_shaders
);
948 /* At this point linked should contain all of the linked IR, so
949 * validate it to make sure nothing went wrong.
952 validate_ir_tree(linked
->ir
);
955 /* Make a pass over all variable declarations to ensure that arrays with
956 * unspecified sizes have a size specified. The size is inferred from the
957 * max_array_access field.
959 if (linked
!= NULL
) {
960 class array_sizing_visitor
: public ir_hierarchical_visitor
{
962 virtual ir_visitor_status
visit(ir_variable
*var
)
964 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
965 const glsl_type
*type
=
966 glsl_type::get_array_instance(var
->type
->fields
.array
,
967 var
->max_array_access
+ 1);
969 assert(type
!= NULL
);
973 return visit_continue
;
984 struct uniform_node
{
986 struct gl_uniform
*u
;
991 * Update the sizes of linked shader uniform arrays to the maximum
994 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
996 * If one or more elements of an array are active,
997 * GetActiveUniform will return the name of the array in name,
998 * subject to the restrictions listed above. The type of the array
999 * is returned in type. The size parameter contains the highest
1000 * array element index used, plus one. The compiler or linker
1001 * determines the highest index used. There will be only one
1002 * active uniform reported by the GL per uniform array.
1006 update_array_sizes(struct gl_shader_program
*prog
)
1008 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1009 if (prog
->_LinkedShaders
[i
] == NULL
)
1012 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1013 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1015 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1016 var
->mode
!= ir_var_in
&&
1017 var
->mode
!= ir_var_out
) ||
1018 !var
->type
->is_array())
1021 unsigned int size
= var
->max_array_access
;
1022 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1023 if (prog
->_LinkedShaders
[j
] == NULL
)
1026 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1027 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1031 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1032 other_var
->max_array_access
> size
) {
1033 size
= other_var
->max_array_access
;
1038 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1039 /* If this is a built-in uniform (i.e., it's backed by some
1040 * fixed-function state), adjust the number of state slots to
1041 * match the new array size. The number of slots per array entry
1042 * is not known. It seems safe to assume that the total number of
1043 * slots is an integer multiple of the number of array elements.
1044 * Determine the number of slots per array element by dividing by
1045 * the old (total) size.
1047 if (var
->num_state_slots
> 0) {
1048 var
->num_state_slots
= (size
+ 1)
1049 * (var
->num_state_slots
/ var
->type
->length
);
1052 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1054 /* FINISHME: We should update the types of array
1055 * dereferences of this variable now.
1063 add_uniform(void *mem_ctx
, exec_list
*uniforms
, struct hash_table
*ht
,
1064 const char *name
, const glsl_type
*type
, GLenum shader_type
,
1065 unsigned *next_shader_pos
, unsigned *total_uniforms
)
1067 if (type
->is_record()) {
1068 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1069 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
1070 char *field_name
= ralloc_asprintf(mem_ctx
, "%s.%s", name
,
1071 type
->fields
.structure
[i
].name
);
1073 add_uniform(mem_ctx
, uniforms
, ht
, field_name
, field_type
,
1074 shader_type
, next_shader_pos
, total_uniforms
);
1077 uniform_node
*n
= (uniform_node
*) hash_table_find(ht
, name
);
1078 unsigned int vec4_slots
;
1079 const glsl_type
*array_elem_type
= NULL
;
1081 if (type
->is_array()) {
1082 array_elem_type
= type
->fields
.array
;
1083 /* Array of structures. */
1084 if (array_elem_type
->is_record()) {
1085 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1086 char *elem_name
= ralloc_asprintf(mem_ctx
, "%s[%d]", name
, i
);
1087 add_uniform(mem_ctx
, uniforms
, ht
, elem_name
, array_elem_type
,
1088 shader_type
, next_shader_pos
, total_uniforms
);
1094 /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
1095 * vectors to vec4 slots.
1097 if (type
->is_array()) {
1098 if (array_elem_type
->is_sampler())
1099 vec4_slots
= type
->length
;
1101 vec4_slots
= type
->length
* array_elem_type
->matrix_columns
;
1102 } else if (type
->is_sampler()) {
1105 vec4_slots
= type
->matrix_columns
;
1109 n
= (uniform_node
*) calloc(1, sizeof(struct uniform_node
));
1110 n
->u
= (gl_uniform
*) calloc(1, sizeof(struct gl_uniform
));
1111 n
->slots
= vec4_slots
;
1113 n
->u
->Name
= strdup(name
);
1118 (*total_uniforms
)++;
1120 hash_table_insert(ht
, n
, name
);
1121 uniforms
->push_tail(& n
->link
);
1124 switch (shader_type
) {
1125 case GL_VERTEX_SHADER
:
1126 n
->u
->VertPos
= *next_shader_pos
;
1128 case GL_FRAGMENT_SHADER
:
1129 n
->u
->FragPos
= *next_shader_pos
;
1131 case GL_GEOMETRY_SHADER
:
1132 n
->u
->GeomPos
= *next_shader_pos
;
1136 (*next_shader_pos
) += vec4_slots
;
1141 assign_uniform_locations(struct gl_shader_program
*prog
)
1145 unsigned total_uniforms
= 0;
1146 hash_table
*ht
= hash_table_ctor(32, hash_table_string_hash
,
1147 hash_table_string_compare
);
1148 void *mem_ctx
= ralloc_context(NULL
);
1150 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1151 if (prog
->_LinkedShaders
[i
] == NULL
)
1154 unsigned next_position
= 0;
1156 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1157 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1159 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
))
1162 if (strncmp(var
->name
, "gl_", 3) == 0) {
1163 /* At the moment, we don't allocate uniform locations for
1164 * builtin uniforms. It's permitted by spec, and we'll
1165 * likely switch to doing that at some point, but not yet.
1170 var
->location
= next_position
;
1171 add_uniform(mem_ctx
, &uniforms
, ht
, var
->name
, var
->type
,
1172 prog
->_LinkedShaders
[i
]->Type
,
1173 &next_position
, &total_uniforms
);
1177 ralloc_free(mem_ctx
);
1179 gl_uniform_list
*ul
= (gl_uniform_list
*)
1180 calloc(1, sizeof(gl_uniform_list
));
1182 ul
->Size
= total_uniforms
;
1183 ul
->NumUniforms
= total_uniforms
;
1184 ul
->Uniforms
= (gl_uniform
*) calloc(total_uniforms
, sizeof(gl_uniform
));
1188 for (uniform_node
*node
= (uniform_node
*) uniforms
.head
1189 ; node
->link
.next
!= NULL
1191 next
= (uniform_node
*) node
->link
.next
;
1193 node
->link
.remove();
1194 memcpy(&ul
->Uniforms
[idx
], node
->u
, sizeof(gl_uniform
));
1201 hash_table_dtor(ht
);
1203 prog
->Uniforms
= ul
;
1208 * Find a contiguous set of available bits in a bitmask.
1210 * \param used_mask Bits representing used (1) and unused (0) locations
1211 * \param needed_count Number of contiguous bits needed.
1214 * Base location of the available bits on success or -1 on failure.
1217 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1219 unsigned needed_mask
= (1 << needed_count
) - 1;
1220 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1222 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1223 * cannot optimize possibly infinite loops" for the loop below.
1225 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1228 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1229 if ((needed_mask
& ~used_mask
) == needed_mask
)
1240 * Assign locations for either VS inputs for FS outputs
1242 * \param prog Shader program whose variables need locations assigned
1243 * \param target_index Selector for the program target to receive location
1244 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1245 * \c MESA_SHADER_FRAGMENT.
1246 * \param max_index Maximum number of generic locations. This corresponds
1247 * to either the maximum number of draw buffers or the
1248 * maximum number of generic attributes.
1251 * If locations are successfully assigned, true is returned. Otherwise an
1252 * error is emitted to the shader link log and false is returned.
1255 * Locations set via \c glBindFragDataLocation are not currently supported.
1256 * Only locations assigned automatically by the linker, explicitly set by a
1257 * layout qualifier, or explicitly set by a built-in variable (e.g., \c
1258 * gl_FragColor) are supported for fragment shaders.
1261 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1262 unsigned target_index
,
1265 /* Mark invalid locations as being used.
1267 unsigned used_locations
= (max_index
>= 32)
1268 ? ~0 : ~((1 << max_index
) - 1);
1270 assert((target_index
== MESA_SHADER_VERTEX
)
1271 || (target_index
== MESA_SHADER_FRAGMENT
));
1273 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1277 /* Operate in a total of four passes.
1279 * 1. Invalidate the location assignments for all vertex shader inputs.
1281 * 2. Assign locations for inputs that have user-defined (via
1282 * glBindVertexAttribLocation) locations.
1284 * 3. Sort the attributes without assigned locations by number of slots
1285 * required in decreasing order. Fragmentation caused by attribute
1286 * locations assigned by the application may prevent large attributes
1287 * from having enough contiguous space.
1289 * 4. Assign locations to any inputs without assigned locations.
1292 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1293 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1295 const enum ir_variable_mode direction
=
1296 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1299 invalidate_variable_locations(sh
, direction
, generic_base
);
1301 /* Temporary storage for the set of attributes that need locations assigned.
1307 /* Used below in the call to qsort. */
1308 static int compare(const void *a
, const void *b
)
1310 const temp_attr
*const l
= (const temp_attr
*) a
;
1311 const temp_attr
*const r
= (const temp_attr
*) b
;
1313 /* Reversed because we want a descending order sort below. */
1314 return r
->slots
- l
->slots
;
1318 unsigned num_attr
= 0;
1320 foreach_list(node
, sh
->ir
) {
1321 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1323 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1326 if (var
->explicit_location
) {
1327 if ((var
->location
>= (int)(max_index
+ generic_base
))
1328 || (var
->location
< 0)) {
1330 "invalid explicit location %d specified for `%s'\n",
1332 ? var
->location
: var
->location
- generic_base
,
1336 } else if (target_index
== MESA_SHADER_VERTEX
) {
1339 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1340 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1341 var
->location
= binding
;
1345 /* If the variable is not a built-in and has a location statically
1346 * assigned in the shader (presumably via a layout qualifier), make sure
1347 * that it doesn't collide with other assigned locations. Otherwise,
1348 * add it to the list of variables that need linker-assigned locations.
1350 const unsigned slots
= count_attribute_slots(var
->type
);
1351 if (var
->location
!= -1) {
1352 if (var
->location
>= generic_base
) {
1353 /* From page 61 of the OpenGL 4.0 spec:
1355 * "LinkProgram will fail if the attribute bindings assigned
1356 * by BindAttribLocation do not leave not enough space to
1357 * assign a location for an active matrix attribute or an
1358 * active attribute array, both of which require multiple
1359 * contiguous generic attributes."
1361 * Previous versions of the spec contain similar language but omit
1362 * the bit about attribute arrays.
1364 * Page 61 of the OpenGL 4.0 spec also says:
1366 * "It is possible for an application to bind more than one
1367 * attribute name to the same location. This is referred to as
1368 * aliasing. This will only work if only one of the aliased
1369 * attributes is active in the executable program, or if no
1370 * path through the shader consumes more than one attribute of
1371 * a set of attributes aliased to the same location. A link
1372 * error can occur if the linker determines that every path
1373 * through the shader consumes multiple aliased attributes,
1374 * but implementations are not required to generate an error
1377 * These two paragraphs are either somewhat contradictory, or I
1378 * don't fully understand one or both of them.
1380 /* FINISHME: The code as currently written does not support
1381 * FINISHME: attribute location aliasing (see comment above).
1383 /* Mask representing the contiguous slots that will be used by
1386 const unsigned attr
= var
->location
- generic_base
;
1387 const unsigned use_mask
= (1 << slots
) - 1;
1389 /* Generate a link error if the set of bits requested for this
1390 * attribute overlaps any previously allocated bits.
1392 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1394 "insufficient contiguous attribute locations "
1395 "available for vertex shader input `%s'",
1400 used_locations
|= (use_mask
<< attr
);
1406 to_assign
[num_attr
].slots
= slots
;
1407 to_assign
[num_attr
].var
= var
;
1411 /* If all of the attributes were assigned locations by the application (or
1412 * are built-in attributes with fixed locations), return early. This should
1413 * be the common case.
1418 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1420 if (target_index
== MESA_SHADER_VERTEX
) {
1421 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1422 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1423 * reserved to prevent it from being automatically allocated below.
1425 find_deref_visitor
find("gl_Vertex");
1427 if (find
.variable_found())
1428 used_locations
|= (1 << 0);
1431 for (unsigned i
= 0; i
< num_attr
; i
++) {
1432 /* Mask representing the contiguous slots that will be used by this
1435 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1437 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1440 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1441 ? "vertex shader input" : "fragment shader output";
1444 "insufficient contiguous attribute locations "
1445 "available for %s `%s'",
1446 string
, to_assign
[i
].var
->name
);
1450 to_assign
[i
].var
->location
= generic_base
+ location
;
1451 used_locations
|= (use_mask
<< location
);
1459 * Demote shader inputs and outputs that are not used in other stages
1462 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1464 foreach_list(node
, sh
->ir
) {
1465 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1467 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1470 /* A shader 'in' or 'out' variable is only really an input or output if
1471 * its value is used by other shader stages. This will cause the variable
1472 * to have a location assigned.
1474 if (var
->location
== -1) {
1475 var
->mode
= ir_var_auto
;
1482 assign_varying_locations(struct gl_context
*ctx
,
1483 struct gl_shader_program
*prog
,
1484 gl_shader
*producer
, gl_shader
*consumer
)
1486 /* FINISHME: Set dynamically when geometry shader support is added. */
1487 unsigned output_index
= VERT_RESULT_VAR0
;
1488 unsigned input_index
= FRAG_ATTRIB_VAR0
;
1490 /* Operate in a total of three passes.
1492 * 1. Assign locations for any matching inputs and outputs.
1494 * 2. Mark output variables in the producer that do not have locations as
1495 * not being outputs. This lets the optimizer eliminate them.
1497 * 3. Mark input variables in the consumer that do not have locations as
1498 * not being inputs. This lets the optimizer eliminate them.
1501 invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
1502 invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
1504 foreach_list(node
, producer
->ir
) {
1505 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
1507 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
)
1508 || (output_var
->location
!= -1))
1511 ir_variable
*const input_var
=
1512 consumer
->symbols
->get_variable(output_var
->name
);
1514 if ((input_var
== NULL
) || (input_var
->mode
!= ir_var_in
))
1517 assert(input_var
->location
== -1);
1519 output_var
->location
= output_index
;
1520 input_var
->location
= input_index
;
1522 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1523 assert(!output_var
->type
->is_record());
1525 if (output_var
->type
->is_array()) {
1526 const unsigned slots
= output_var
->type
->length
1527 * output_var
->type
->fields
.array
->matrix_columns
;
1529 output_index
+= slots
;
1530 input_index
+= slots
;
1532 const unsigned slots
= output_var
->type
->matrix_columns
;
1534 output_index
+= slots
;
1535 input_index
+= slots
;
1539 unsigned varying_vectors
= 0;
1541 foreach_list(node
, consumer
->ir
) {
1542 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1544 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
1547 if (var
->location
== -1) {
1548 if (prog
->Version
<= 120) {
1549 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1551 * Only those varying variables used (i.e. read) in
1552 * the fragment shader executable must be written to
1553 * by the vertex shader executable; declaring
1554 * superfluous varying variables in a vertex shader is
1557 * We interpret this text as meaning that the VS must
1558 * write the variable for the FS to read it. See
1559 * "glsl1-varying read but not written" in piglit.
1562 linker_error(prog
, "fragment shader varying %s not written "
1563 "by vertex shader\n.", var
->name
);
1566 /* An 'in' variable is only really a shader input if its
1567 * value is written by the previous stage.
1569 var
->mode
= ir_var_auto
;
1571 /* The packing rules are used for vertex shader inputs are also used
1572 * for fragment shader inputs.
1574 varying_vectors
+= count_attribute_slots(var
->type
);
1578 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1579 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
1580 linker_error(prog
, "shader uses too many varying vectors "
1582 varying_vectors
, ctx
->Const
.MaxVarying
);
1586 const unsigned float_components
= varying_vectors
* 4;
1587 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
1588 linker_error(prog
, "shader uses too many varying components "
1590 float_components
, ctx
->Const
.MaxVarying
* 4);
1600 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1602 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1604 prog
->LinkStatus
= false;
1605 prog
->Validated
= false;
1606 prog
->_Used
= false;
1608 if (prog
->InfoLog
!= NULL
)
1609 ralloc_free(prog
->InfoLog
);
1611 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1613 /* Separate the shaders into groups based on their type.
1615 struct gl_shader
**vert_shader_list
;
1616 unsigned num_vert_shaders
= 0;
1617 struct gl_shader
**frag_shader_list
;
1618 unsigned num_frag_shaders
= 0;
1620 vert_shader_list
= (struct gl_shader
**)
1621 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
1622 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
1624 unsigned min_version
= UINT_MAX
;
1625 unsigned max_version
= 0;
1626 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1627 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1628 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1630 switch (prog
->Shaders
[i
]->Type
) {
1631 case GL_VERTEX_SHADER
:
1632 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1635 case GL_FRAGMENT_SHADER
:
1636 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1639 case GL_GEOMETRY_SHADER
:
1640 /* FINISHME: Support geometry shaders. */
1641 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
1646 /* Previous to GLSL version 1.30, different compilation units could mix and
1647 * match shading language versions. With GLSL 1.30 and later, the versions
1648 * of all shaders must match.
1650 assert(min_version
>= 100);
1651 assert(max_version
<= 130);
1652 if ((max_version
>= 130 || min_version
== 100)
1653 && min_version
!= max_version
) {
1654 linker_error(prog
, "all shaders must use same shading "
1655 "language version\n");
1659 prog
->Version
= max_version
;
1661 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1662 if (prog
->_LinkedShaders
[i
] != NULL
)
1663 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1665 prog
->_LinkedShaders
[i
] = NULL
;
1668 /* Link all shaders for a particular stage and validate the result.
1670 if (num_vert_shaders
> 0) {
1671 gl_shader
*const sh
=
1672 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1678 if (!validate_vertex_shader_executable(prog
, sh
))
1681 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1685 if (num_frag_shaders
> 0) {
1686 gl_shader
*const sh
=
1687 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1693 if (!validate_fragment_shader_executable(prog
, sh
))
1696 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1700 /* Here begins the inter-stage linking phase. Some initial validation is
1701 * performed, then locations are assigned for uniforms, attributes, and
1704 if (cross_validate_uniforms(prog
)) {
1707 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1708 if (prog
->_LinkedShaders
[prev
] != NULL
)
1712 /* Validate the inputs of each stage with the output of the preceding
1715 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1716 if (prog
->_LinkedShaders
[i
] == NULL
)
1719 if (!cross_validate_outputs_to_inputs(prog
,
1720 prog
->_LinkedShaders
[prev
],
1721 prog
->_LinkedShaders
[i
]))
1727 prog
->LinkStatus
= true;
1730 /* Do common optimization before assigning storage for attributes,
1731 * uniforms, and varyings. Later optimization could possibly make
1732 * some of that unused.
1734 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1735 if (prog
->_LinkedShaders
[i
] == NULL
)
1738 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
1739 if (!prog
->LinkStatus
)
1742 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
)
1743 lower_clip_distance(prog
->_LinkedShaders
[i
]->ir
);
1745 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, 32))
1749 /* FINISHME: The value of the max_attribute_index parameter is
1750 * FINISHME: implementation dependent based on the value of
1751 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1752 * FINISHME: at least 16, so hardcode 16 for now.
1754 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
1758 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, ctx
->Const
.MaxDrawBuffers
)) {
1763 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1764 if (prog
->_LinkedShaders
[prev
] != NULL
)
1768 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1769 if (prog
->_LinkedShaders
[i
] == NULL
)
1772 if (!assign_varying_locations(ctx
, prog
,
1773 prog
->_LinkedShaders
[prev
],
1774 prog
->_LinkedShaders
[i
])) {
1781 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1782 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1785 /* Eliminate code that is now dead due to unused vertex outputs being
1788 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
, false))
1792 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
1793 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
1795 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1796 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
1797 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
1799 /* Eliminate code that is now dead due to unused geometry outputs being
1802 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
, false))
1806 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1807 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1809 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1811 /* Eliminate code that is now dead due to unused fragment inputs being
1812 * demoted. This shouldn't actually do anything other than remove
1813 * declarations of the (now unused) global variables.
1815 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
, false))
1819 update_array_sizes(prog
);
1820 assign_uniform_locations(prog
);
1822 /* OpenGL ES requires that a vertex shader and a fragment shader both be
1823 * present in a linked program. By checking for use of shading language
1824 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
1826 if (!prog
->InternalSeparateShader
&&
1827 (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100)) {
1828 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
1829 linker_error(prog
, "program lacks a vertex shader\n");
1830 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1831 linker_error(prog
, "program lacks a fragment shader\n");
1835 /* FINISHME: Assign fragment shader output locations. */
1838 free(vert_shader_list
);
1840 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1841 if (prog
->_LinkedShaders
[i
] == NULL
)
1844 /* Retain any live IR, but trash the rest. */
1845 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
1847 /* The symbol table in the linked shaders may contain references to
1848 * variables that were removed (e.g., unused uniforms). Since it may
1849 * contain junk, there is no possible valid use. Delete it and set the
1852 delete prog
->_LinkedShaders
[i
]->symbols
;
1853 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
1856 ralloc_free(mem_ctx
);