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"
69 #include "glsl_parser_extras.h"
72 #include "program/hash_table.h"
74 #include "link_varyings.h"
75 #include "ir_optimization.h"
76 #include "ir_rvalue_visitor.h"
79 #include "main/shaderobj.h"
80 #include "main/enums.h"
83 void linker_error(gl_shader_program
*, const char *, ...);
88 * Visitor that determines whether or not a variable is ever written.
90 class find_assignment_visitor
: public ir_hierarchical_visitor
{
92 find_assignment_visitor(const char *name
)
93 : name(name
), found(false)
98 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
100 ir_variable
*const var
= ir
->lhs
->variable_referenced();
102 if (strcmp(name
, var
->name
) == 0) {
107 return visit_continue_with_parent
;
110 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
112 exec_list_iterator sig_iter
= ir
->callee
->parameters
.iterator();
113 foreach_iter(exec_list_iterator
, iter
, *ir
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
115 ir_variable
*sig_param
= (ir_variable
*)sig_iter
.get();
117 if (sig_param
->mode
== ir_var_function_out
||
118 sig_param
->mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
128 if (ir
->return_deref
!= NULL
) {
129 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
131 if (strcmp(name
, var
->name
) == 0) {
137 return visit_continue_with_parent
;
140 bool variable_found()
146 const char *name
; /**< Find writes to a variable with this name. */
147 bool found
; /**< Was a write to the variable found? */
152 * Visitor that determines whether or not a variable is ever read.
154 class find_deref_visitor
: public ir_hierarchical_visitor
{
156 find_deref_visitor(const char *name
)
157 : name(name
), found(false)
162 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
164 if (strcmp(this->name
, ir
->var
->name
) == 0) {
169 return visit_continue
;
172 bool variable_found() const
178 const char *name
; /**< Find writes to a variable with this name. */
179 bool found
; /**< Was a write to the variable found? */
183 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
185 unsigned num_vertices
;
186 gl_shader_program
*prog
;
188 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
190 this->num_vertices
= num_vertices
;
194 virtual ~geom_array_resize_visitor()
199 virtual ir_visitor_status
visit(ir_variable
*var
)
201 if (!var
->type
->is_array() || var
->mode
!= ir_var_shader_in
)
202 return visit_continue
;
204 unsigned size
= var
->type
->length
;
206 /* Generate a link error if the shader has declared this array with an
209 if (size
&& size
!= this->num_vertices
) {
210 linker_error(this->prog
, "size of array %s declared as %u, "
211 "but number of input vertices is %u\n",
212 var
->name
, size
, this->num_vertices
);
213 return visit_continue
;
216 /* Generate a link error if the shader attempts to access an input
217 * array using an index too large for its actual size assigned at link
220 if (var
->max_array_access
>= this->num_vertices
) {
221 linker_error(this->prog
, "geometry shader accesses element %i of "
222 "%s, but only %i input vertices\n",
223 var
->max_array_access
, var
->name
, this->num_vertices
);
224 return visit_continue
;
227 var
->type
= glsl_type::get_array_instance(var
->type
->element_type(),
229 var
->max_array_access
= this->num_vertices
- 1;
231 return visit_continue
;
234 /* Dereferences of input variables need to be updated so that their type
235 * matches the newly assigned type of the variable they are accessing. */
236 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
238 ir
->type
= ir
->var
->type
;
239 return visit_continue
;
242 /* Dereferences of 2D input arrays need to be updated so that their type
243 * matches the newly assigned type of the array they are accessing. */
244 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
246 const glsl_type
*const vt
= ir
->array
->type
;
248 ir
->type
= vt
->element_type();
249 return visit_continue
;
255 * Visitor that determines whether or not a shader uses ir_end_primitive.
257 class find_end_primitive_visitor
: public ir_hierarchical_visitor
{
259 find_end_primitive_visitor()
265 virtual ir_visitor_status
visit(ir_end_primitive
*)
271 bool end_primitive_found()
280 } /* anonymous namespace */
283 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
287 ralloc_strcat(&prog
->InfoLog
, "error: ");
289 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
292 prog
->LinkStatus
= false;
297 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
301 ralloc_strcat(&prog
->InfoLog
, "error: ");
303 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
310 * Given a string identifying a program resource, break it into a base name
311 * and an optional array index in square brackets.
313 * If an array index is present, \c out_base_name_end is set to point to the
314 * "[" that precedes the array index, and the array index itself is returned
317 * If no array index is present (or if the array index is negative or
318 * mal-formed), \c out_base_name_end, is set to point to the null terminator
319 * at the end of the input string, and -1 is returned.
321 * Only the final array index is parsed; if the string contains other array
322 * indices (or structure field accesses), they are left in the base name.
324 * No attempt is made to check that the base name is properly formed;
325 * typically the caller will look up the base name in a hash table, so
326 * ill-formed base names simply turn into hash table lookup failures.
329 parse_program_resource_name(const GLchar
*name
,
330 const GLchar
**out_base_name_end
)
332 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
334 * "When an integer array element or block instance number is part of
335 * the name string, it will be specified in decimal form without a "+"
336 * or "-" sign or any extra leading zeroes. Additionally, the name
337 * string will not include white space anywhere in the string."
340 const size_t len
= strlen(name
);
341 *out_base_name_end
= name
+ len
;
343 if (len
== 0 || name
[len
-1] != ']')
346 /* Walk backwards over the string looking for a non-digit character. This
347 * had better be the opening bracket for an array index.
349 * Initially, i specifies the location of the ']'. Since the string may
350 * contain only the ']' charcater, walk backwards very carefully.
353 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
356 if ((i
== 0) || name
[i
-1] != '[')
359 long array_index
= strtol(&name
[i
], NULL
, 10);
363 *out_base_name_end
= name
+ (i
- 1);
369 link_invalidate_variable_locations(exec_list
*ir
)
371 foreach_list(node
, ir
) {
372 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
377 /* Only assign locations for variables that lack an explicit location.
378 * Explicit locations are set for all built-in variables, generic vertex
379 * shader inputs (via layout(location=...)), and generic fragment shader
380 * outputs (also via layout(location=...)).
382 if (!var
->explicit_location
) {
384 var
->location_frac
= 0;
387 /* ir_variable::is_unmatched_generic_inout is used by the linker while
388 * connecting outputs from one stage to inputs of the next stage.
390 * There are two implicit assumptions here. First, we assume that any
391 * built-in variable (i.e., non-generic in or out) will have
392 * explicit_location set. Second, we assume that any generic in or out
393 * will not have explicit_location set.
395 * This second assumption will only be valid until
396 * GL_ARB_separate_shader_objects is supported. When that extension is
397 * implemented, this function will need some modifications.
399 if (!var
->explicit_location
) {
400 var
->is_unmatched_generic_inout
= 1;
402 var
->is_unmatched_generic_inout
= 0;
409 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
411 * Also check for errors based on incorrect usage of gl_ClipVertex and
414 * Return false if an error was reported.
417 analyze_clip_usage(const char *shader_type
, struct gl_shader_program
*prog
,
418 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
419 GLuint
*ClipDistanceArraySize
)
421 *ClipDistanceArraySize
= 0;
423 if (!prog
->IsES
&& prog
->Version
>= 130) {
424 /* From section 7.1 (Vertex Shader Special Variables) of the
427 * "It is an error for a shader to statically write both
428 * gl_ClipVertex and gl_ClipDistance."
430 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
431 * gl_ClipVertex nor gl_ClipDistance.
433 find_assignment_visitor
clip_vertex("gl_ClipVertex");
434 find_assignment_visitor
clip_distance("gl_ClipDistance");
436 clip_vertex
.run(shader
->ir
);
437 clip_distance
.run(shader
->ir
);
438 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
439 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
440 "and `gl_ClipDistance'\n", shader_type
);
443 *UsesClipDistance
= clip_distance
.variable_found();
444 ir_variable
*clip_distance_var
=
445 shader
->symbols
->get_variable("gl_ClipDistance");
446 if (clip_distance_var
)
447 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
449 *UsesClipDistance
= false;
455 * Verify that a vertex shader executable meets all semantic requirements.
457 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
460 * \param shader Vertex shader executable to be verified
463 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
464 struct gl_shader
*shader
)
469 /* From the GLSL 1.10 spec, page 48:
471 * "The variable gl_Position is available only in the vertex
472 * language and is intended for writing the homogeneous vertex
473 * position. All executions of a well-formed vertex shader
474 * executable must write a value into this variable. [...] The
475 * variable gl_Position is available only in the vertex
476 * language and is intended for writing the homogeneous vertex
477 * position. All executions of a well-formed vertex shader
478 * executable must write a value into this variable."
480 * while in GLSL 1.40 this text is changed to:
482 * "The variable gl_Position is available only in the vertex
483 * language and is intended for writing the homogeneous vertex
484 * position. It can be written at any time during shader
485 * execution. It may also be read back by a vertex shader
486 * after being written. This value will be used by primitive
487 * assembly, clipping, culling, and other fixed functionality
488 * operations, if present, that operate on primitives after
489 * vertex processing has occurred. Its value is undefined if
490 * the vertex shader executable does not write gl_Position."
492 * GLSL ES 3.00 is similar to GLSL 1.40--failing to write to gl_Position is
495 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
496 find_assignment_visitor
find("gl_Position");
497 find
.run(shader
->ir
);
498 if (!find
.variable_found()) {
499 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
504 analyze_clip_usage("vertex", prog
, shader
, &prog
->Vert
.UsesClipDistance
,
505 &prog
->Vert
.ClipDistanceArraySize
);
510 * Verify that a fragment shader executable meets all semantic requirements
512 * \param shader Fragment shader executable to be verified
515 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
516 struct gl_shader
*shader
)
521 find_assignment_visitor
frag_color("gl_FragColor");
522 find_assignment_visitor
frag_data("gl_FragData");
524 frag_color
.run(shader
->ir
);
525 frag_data
.run(shader
->ir
);
527 if (frag_color
.variable_found() && frag_data
.variable_found()) {
528 linker_error(prog
, "fragment shader writes to both "
529 "`gl_FragColor' and `gl_FragData'\n");
534 * Verify that a geometry shader executable meets all semantic requirements
536 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
537 * prog->Geom.ClipDistanceArraySize as a side effect.
539 * \param shader Geometry shader executable to be verified
542 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
543 struct gl_shader
*shader
)
548 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
549 prog
->Geom
.VerticesIn
= num_vertices
;
551 analyze_clip_usage("geometry", prog
, shader
, &prog
->Geom
.UsesClipDistance
,
552 &prog
->Geom
.ClipDistanceArraySize
);
554 find_end_primitive_visitor end_primitive
;
555 end_primitive
.run(shader
->ir
);
556 prog
->Geom
.UsesEndPrimitive
= end_primitive
.end_primitive_found();
561 * Perform validation of global variables used across multiple shaders
564 cross_validate_globals(struct gl_shader_program
*prog
,
565 struct gl_shader
**shader_list
,
566 unsigned num_shaders
,
569 /* Examine all of the uniforms in all of the shaders and cross validate
572 glsl_symbol_table variables
;
573 for (unsigned i
= 0; i
< num_shaders
; i
++) {
574 if (shader_list
[i
] == NULL
)
577 foreach_list(node
, shader_list
[i
]->ir
) {
578 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
583 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
586 /* Don't cross validate temporaries that are at global scope. These
587 * will eventually get pulled into the shaders 'main'.
589 if (var
->mode
== ir_var_temporary
)
592 /* If a global with this name has already been seen, verify that the
593 * new instance has the same type. In addition, if the globals have
594 * initializers, the values of the initializers must be the same.
596 ir_variable
*const existing
= variables
.get_variable(var
->name
);
597 if (existing
!= NULL
) {
598 if (var
->type
!= existing
->type
) {
599 /* Consider the types to be "the same" if both types are arrays
600 * of the same type and one of the arrays is implicitly sized.
601 * In addition, set the type of the linked variable to the
602 * explicitly sized array.
604 if (var
->type
->is_array()
605 && existing
->type
->is_array()
606 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
607 && ((var
->type
->length
== 0)
608 || (existing
->type
->length
== 0))) {
609 if (var
->type
->length
!= 0) {
610 existing
->type
= var
->type
;
613 linker_error(prog
, "%s `%s' declared as type "
614 "`%s' and type `%s'\n",
616 var
->name
, var
->type
->name
,
617 existing
->type
->name
);
622 if (var
->explicit_location
) {
623 if (existing
->explicit_location
624 && (var
->location
!= existing
->location
)) {
625 linker_error(prog
, "explicit locations for %s "
626 "`%s' have differing values\n",
627 mode_string(var
), var
->name
);
631 existing
->location
= var
->location
;
632 existing
->explicit_location
= true;
635 /* From the GLSL 4.20 specification:
636 * "A link error will result if two compilation units in a program
637 * specify different integer-constant bindings for the same
638 * opaque-uniform name. However, it is not an error to specify a
639 * binding on some but not all declarations for the same name"
641 if (var
->explicit_binding
) {
642 if (existing
->explicit_binding
&&
643 var
->binding
!= existing
->binding
) {
644 linker_error(prog
, "explicit bindings for %s "
645 "`%s' have differing values\n",
646 mode_string(var
), var
->name
);
650 existing
->binding
= var
->binding
;
651 existing
->explicit_binding
= true;
654 if (var
->type
->contains_atomic() &&
655 var
->atomic
.offset
!= existing
->atomic
.offset
) {
656 linker_error(prog
, "offset specifications for %s "
657 "`%s' have differing values\n",
658 mode_string(var
), var
->name
);
662 /* Validate layout qualifiers for gl_FragDepth.
664 * From the AMD/ARB_conservative_depth specs:
666 * "If gl_FragDepth is redeclared in any fragment shader in a
667 * program, it must be redeclared in all fragment shaders in
668 * that program that have static assignments to
669 * gl_FragDepth. All redeclarations of gl_FragDepth in all
670 * fragment shaders in a single program must have the same set
673 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
674 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
675 bool layout_differs
=
676 var
->depth_layout
!= existing
->depth_layout
;
678 if (layout_declared
&& layout_differs
) {
680 "All redeclarations of gl_FragDepth in all "
681 "fragment shaders in a single program must have "
682 "the same set of qualifiers.");
685 if (var
->used
&& layout_differs
) {
687 "If gl_FragDepth is redeclared with a layout "
688 "qualifier in any fragment shader, it must be "
689 "redeclared with the same layout qualifier in "
690 "all fragment shaders that have assignments to "
695 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
697 * "If a shared global has multiple initializers, the
698 * initializers must all be constant expressions, and they
699 * must all have the same value. Otherwise, a link error will
700 * result. (A shared global having only one initializer does
701 * not require that initializer to be a constant expression.)"
703 * Previous to 4.20 the GLSL spec simply said that initializers
704 * must have the same value. In this case of non-constant
705 * initializers, this was impossible to determine. As a result,
706 * no vendor actually implemented that behavior. The 4.20
707 * behavior matches the implemented behavior of at least one other
708 * vendor, so we'll implement that for all GLSL versions.
710 if (var
->constant_initializer
!= NULL
) {
711 if (existing
->constant_initializer
!= NULL
) {
712 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
713 linker_error(prog
, "initializers for %s "
714 "`%s' have differing values\n",
715 mode_string(var
), var
->name
);
719 /* If the first-seen instance of a particular uniform did not
720 * have an initializer but a later instance does, copy the
721 * initializer to the version stored in the symbol table.
723 /* FINISHME: This is wrong. The constant_value field should
724 * FINISHME: not be modified! Imagine a case where a shader
725 * FINISHME: without an initializer is linked in two different
726 * FINISHME: programs with shaders that have differing
727 * FINISHME: initializers. Linking with the first will
728 * FINISHME: modify the shader, and linking with the second
729 * FINISHME: will fail.
731 existing
->constant_initializer
=
732 var
->constant_initializer
->clone(ralloc_parent(existing
),
737 if (var
->has_initializer
) {
738 if (existing
->has_initializer
739 && (var
->constant_initializer
== NULL
740 || existing
->constant_initializer
== NULL
)) {
742 "shared global variable `%s' has multiple "
743 "non-constant initializers.\n",
748 /* Some instance had an initializer, so keep track of that. In
749 * this location, all sorts of initializers (constant or
750 * otherwise) will propagate the existence to the variable
751 * stored in the symbol table.
753 existing
->has_initializer
= true;
756 if (existing
->invariant
!= var
->invariant
) {
757 linker_error(prog
, "declarations for %s `%s' have "
758 "mismatching invariant qualifiers\n",
759 mode_string(var
), var
->name
);
762 if (existing
->centroid
!= var
->centroid
) {
763 linker_error(prog
, "declarations for %s `%s' have "
764 "mismatching centroid qualifiers\n",
765 mode_string(var
), var
->name
);
769 variables
.add_variable(var
);
776 * Perform validation of uniforms used across multiple shader stages
779 cross_validate_uniforms(struct gl_shader_program
*prog
)
781 cross_validate_globals(prog
, prog
->_LinkedShaders
,
782 MESA_SHADER_TYPES
, true);
786 * Accumulates the array of prog->UniformBlocks and checks that all
787 * definitons of blocks agree on their contents.
790 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
792 unsigned max_num_uniform_blocks
= 0;
793 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
794 if (prog
->_LinkedShaders
[i
])
795 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
798 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
799 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
801 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
802 max_num_uniform_blocks
);
803 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
804 prog
->UniformBlockStageIndex
[i
][j
] = -1;
809 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
810 int index
= link_cross_validate_uniform_block(prog
,
811 &prog
->UniformBlocks
,
812 &prog
->NumUniformBlocks
,
813 &sh
->UniformBlocks
[j
]);
816 linker_error(prog
, "uniform block `%s' has mismatching definitions",
817 sh
->UniformBlocks
[j
].Name
);
821 prog
->UniformBlockStageIndex
[i
][index
] = j
;
830 * Populates a shaders symbol table with all global declarations
833 populate_symbol_table(gl_shader
*sh
)
835 sh
->symbols
= new(sh
) glsl_symbol_table
;
837 foreach_list(node
, sh
->ir
) {
838 ir_instruction
*const inst
= (ir_instruction
*) node
;
842 if ((func
= inst
->as_function()) != NULL
) {
843 sh
->symbols
->add_function(func
);
844 } else if ((var
= inst
->as_variable()) != NULL
) {
845 sh
->symbols
->add_variable(var
);
852 * Remap variables referenced in an instruction tree
854 * This is used when instruction trees are cloned from one shader and placed in
855 * another. These trees will contain references to \c ir_variable nodes that
856 * do not exist in the target shader. This function finds these \c ir_variable
857 * references and replaces the references with matching variables in the target
860 * If there is no matching variable in the target shader, a clone of the
861 * \c ir_variable is made and added to the target shader. The new variable is
862 * added to \b both the instruction stream and the symbol table.
864 * \param inst IR tree that is to be processed.
865 * \param symbols Symbol table containing global scope symbols in the
867 * \param instructions Instruction stream where new variable declarations
871 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
874 class remap_visitor
: public ir_hierarchical_visitor
{
876 remap_visitor(struct gl_shader
*target
,
879 this->target
= target
;
880 this->symbols
= target
->symbols
;
881 this->instructions
= target
->ir
;
885 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
887 if (ir
->var
->mode
== ir_var_temporary
) {
888 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
892 return visit_continue
;
895 ir_variable
*const existing
=
896 this->symbols
->get_variable(ir
->var
->name
);
897 if (existing
!= NULL
)
900 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
902 this->symbols
->add_variable(copy
);
903 this->instructions
->push_head(copy
);
907 return visit_continue
;
911 struct gl_shader
*target
;
912 glsl_symbol_table
*symbols
;
913 exec_list
*instructions
;
917 remap_visitor
v(target
, temps
);
924 * Move non-declarations from one instruction stream to another
926 * The intended usage pattern of this function is to pass the pointer to the
927 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
928 * pointer) for \c last and \c false for \c make_copies on the first
929 * call. Successive calls pass the return value of the previous call for
930 * \c last and \c true for \c make_copies.
932 * \param instructions Source instruction stream
933 * \param last Instruction after which new instructions should be
934 * inserted in the target instruction stream
935 * \param make_copies Flag selecting whether instructions in \c instructions
936 * should be copied (via \c ir_instruction::clone) into the
937 * target list or moved.
940 * The new "last" instruction in the target instruction stream. This pointer
941 * is suitable for use as the \c last parameter of a later call to this
945 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
946 bool make_copies
, gl_shader
*target
)
948 hash_table
*temps
= NULL
;
951 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
952 hash_table_pointer_compare
);
954 foreach_list_safe(node
, instructions
) {
955 ir_instruction
*inst
= (ir_instruction
*) node
;
957 if (inst
->as_function())
960 ir_variable
*var
= inst
->as_variable();
961 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
964 assert(inst
->as_assignment()
966 || inst
->as_if() /* for initializers with the ?: operator */
967 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
970 inst
= inst
->clone(target
, NULL
);
973 hash_table_insert(temps
, inst
, var
);
975 remap_variables(inst
, target
, temps
);
980 last
->insert_after(inst
);
985 hash_table_dtor(temps
);
991 * Get the function signature for main from a shader
993 static ir_function_signature
*
994 get_main_function_signature(gl_shader
*sh
)
996 ir_function
*const f
= sh
->symbols
->get_function("main");
998 exec_list void_parameters
;
1000 /* Look for the 'void main()' signature and ensure that it's defined.
1001 * This keeps the linker from accidentally pick a shader that just
1002 * contains a prototype for main.
1004 * We don't have to check for multiple definitions of main (in multiple
1005 * shaders) because that would have already been caught above.
1007 ir_function_signature
*sig
= f
->matching_signature(NULL
, &void_parameters
);
1008 if ((sig
!= NULL
) && sig
->is_defined
) {
1018 * This class is only used in link_intrastage_shaders() below but declaring
1019 * it inside that function leads to compiler warnings with some versions of
1022 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1024 array_sizing_visitor()
1025 : mem_ctx(ralloc_context(NULL
)),
1026 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1027 hash_table_pointer_compare
))
1031 ~array_sizing_visitor()
1033 hash_table_dtor(this->unnamed_interfaces
);
1034 ralloc_free(this->mem_ctx
);
1037 virtual ir_visitor_status
visit(ir_variable
*var
)
1039 fixup_type(&var
->type
, var
->max_array_access
);
1040 if (var
->type
->is_interface()) {
1041 if (interface_contains_unsized_arrays(var
->type
)) {
1042 const glsl_type
*new_type
=
1043 resize_interface_members(var
->type
, var
->max_ifc_array_access
);
1044 var
->type
= new_type
;
1045 var
->change_interface_type(new_type
);
1047 } else if (var
->type
->is_array() &&
1048 var
->type
->fields
.array
->is_interface()) {
1049 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1050 const glsl_type
*new_type
=
1051 resize_interface_members(var
->type
->fields
.array
,
1052 var
->max_ifc_array_access
);
1053 var
->change_interface_type(new_type
);
1055 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1057 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1058 /* Store a pointer to the variable in the unnamed_interfaces
1061 ir_variable
**interface_vars
= (ir_variable
**)
1062 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1063 if (interface_vars
== NULL
) {
1064 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1066 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1069 unsigned index
= ifc_type
->field_index(var
->name
);
1070 assert(index
< ifc_type
->length
);
1071 assert(interface_vars
[index
] == NULL
);
1072 interface_vars
[index
] = var
;
1074 return visit_continue
;
1078 * For each unnamed interface block that was discovered while running the
1079 * visitor, adjust the interface type to reflect the newly assigned array
1080 * sizes, and fix up the ir_variable nodes to point to the new interface
1083 void fixup_unnamed_interface_types()
1085 hash_table_call_foreach(this->unnamed_interfaces
,
1086 fixup_unnamed_interface_type
, NULL
);
1091 * If the type pointed to by \c type represents an unsized array, replace
1092 * it with a sized array whose size is determined by max_array_access.
1094 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1096 if ((*type
)->is_unsized_array()) {
1097 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1098 max_array_access
+ 1);
1099 assert(*type
!= NULL
);
1104 * Determine whether the given interface type contains unsized arrays (if
1105 * it doesn't, array_sizing_visitor doesn't need to process it).
1107 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1109 for (unsigned i
= 0; i
< type
->length
; i
++) {
1110 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1111 if (elem_type
->is_unsized_array())
1118 * Create a new interface type based on the given type, with unsized arrays
1119 * replaced by sized arrays whose size is determined by
1120 * max_ifc_array_access.
1122 static const glsl_type
*
1123 resize_interface_members(const glsl_type
*type
,
1124 const unsigned *max_ifc_array_access
)
1126 unsigned num_fields
= type
->length
;
1127 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1128 memcpy(fields
, type
->fields
.structure
,
1129 num_fields
* sizeof(*fields
));
1130 for (unsigned i
= 0; i
< num_fields
; i
++) {
1131 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1133 glsl_interface_packing packing
=
1134 (glsl_interface_packing
) type
->interface_packing
;
1135 const glsl_type
*new_ifc_type
=
1136 glsl_type::get_interface_instance(fields
, num_fields
,
1137 packing
, type
->name
);
1139 return new_ifc_type
;
1142 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1145 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1146 ir_variable
**interface_vars
= (ir_variable
**) data
;
1147 unsigned num_fields
= ifc_type
->length
;
1148 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1149 memcpy(fields
, ifc_type
->fields
.structure
,
1150 num_fields
* sizeof(*fields
));
1151 bool interface_type_changed
= false;
1152 for (unsigned i
= 0; i
< num_fields
; i
++) {
1153 if (interface_vars
[i
] != NULL
&&
1154 fields
[i
].type
!= interface_vars
[i
]->type
) {
1155 fields
[i
].type
= interface_vars
[i
]->type
;
1156 interface_type_changed
= true;
1159 if (!interface_type_changed
) {
1163 glsl_interface_packing packing
=
1164 (glsl_interface_packing
) ifc_type
->interface_packing
;
1165 const glsl_type
*new_ifc_type
=
1166 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1169 for (unsigned i
= 0; i
< num_fields
; i
++) {
1170 if (interface_vars
[i
] != NULL
)
1171 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1176 * Memory context used to allocate the data in \c unnamed_interfaces.
1181 * Hash table from const glsl_type * to an array of ir_variable *'s
1182 * pointing to the ir_variables constituting each unnamed interface block.
1184 hash_table
*unnamed_interfaces
;
1188 * Performs the cross-validation of geometry shader max_vertices and
1189 * primitive type layout qualifiers for the attached geometry shaders,
1190 * and propagates them to the linked GS and linked shader program.
1193 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1194 struct gl_shader
*linked_shader
,
1195 struct gl_shader
**shader_list
,
1196 unsigned num_shaders
)
1198 linked_shader
->Geom
.VerticesOut
= 0;
1199 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1200 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1202 /* No in/out qualifiers defined for anything but GLSL 1.50+
1203 * geometry shaders so far.
1205 if (linked_shader
->Type
!= GL_GEOMETRY_SHADER
|| prog
->Version
< 150)
1208 /* From the GLSL 1.50 spec, page 46:
1210 * "All geometry shader output layout declarations in a program
1211 * must declare the same layout and same value for
1212 * max_vertices. There must be at least one geometry output
1213 * layout declaration somewhere in a program, but not all
1214 * geometry shaders (compilation units) are required to
1218 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1219 struct gl_shader
*shader
= shader_list
[i
];
1221 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1222 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1223 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1224 linker_error(prog
, "geometry shader defined with conflicting "
1228 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1231 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1232 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1233 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1234 linker_error(prog
, "geometry shader defined with conflicting "
1238 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1241 if (shader
->Geom
.VerticesOut
!= 0) {
1242 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1243 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1244 linker_error(prog
, "geometry shader defined with conflicting "
1245 "output vertex count (%d and %d)\n",
1246 linked_shader
->Geom
.VerticesOut
,
1247 shader
->Geom
.VerticesOut
);
1250 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1254 /* Just do the intrastage -> interstage propagation right now,
1255 * since we already know we're in the right type of shader program
1258 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1260 "geometry shader didn't declare primitive input type\n");
1263 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1265 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1267 "geometry shader didn't declare primitive output type\n");
1270 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1272 if (linked_shader
->Geom
.VerticesOut
== 0) {
1274 "geometry shader didn't declare max_vertices\n");
1277 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1281 * Combine a group of shaders for a single stage to generate a linked shader
1284 * If this function is supplied a single shader, it is cloned, and the new
1285 * shader is returned.
1287 static struct gl_shader
*
1288 link_intrastage_shaders(void *mem_ctx
,
1289 struct gl_context
*ctx
,
1290 struct gl_shader_program
*prog
,
1291 struct gl_shader
**shader_list
,
1292 unsigned num_shaders
)
1294 struct gl_uniform_block
*uniform_blocks
= NULL
;
1296 /* Check that global variables defined in multiple shaders are consistent.
1298 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1299 if (!prog
->LinkStatus
)
1302 /* Check that interface blocks defined in multiple shaders are consistent.
1304 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1306 if (!prog
->LinkStatus
)
1309 /* Link up uniform blocks defined within this stage. */
1310 const unsigned num_uniform_blocks
=
1311 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1314 /* Check that there is only a single definition of each function signature
1315 * across all shaders.
1317 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1318 foreach_list(node
, shader_list
[i
]->ir
) {
1319 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
1324 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1325 ir_function
*const other
=
1326 shader_list
[j
]->symbols
->get_function(f
->name
);
1328 /* If the other shader has no function (and therefore no function
1329 * signatures) with the same name, skip to the next shader.
1334 foreach_iter (exec_list_iterator
, iter
, *f
) {
1335 ir_function_signature
*sig
=
1336 (ir_function_signature
*) iter
.get();
1338 if (!sig
->is_defined
|| sig
->is_builtin())
1341 ir_function_signature
*other_sig
=
1342 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1344 if ((other_sig
!= NULL
) && other_sig
->is_defined
1345 && !other_sig
->is_builtin()) {
1346 linker_error(prog
, "function `%s' is multiply defined",
1355 /* Find the shader that defines main, and make a clone of it.
1357 * Starting with the clone, search for undefined references. If one is
1358 * found, find the shader that defines it. Clone the reference and add
1359 * it to the shader. Repeat until there are no undefined references or
1360 * until a reference cannot be resolved.
1362 gl_shader
*main
= NULL
;
1363 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1364 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
1365 main
= shader_list
[i
];
1371 linker_error(prog
, "%s shader lacks `main'\n",
1372 _mesa_glsl_shader_target_name(shader_list
[0]->Type
));
1376 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1377 linked
->ir
= new(linked
) exec_list
;
1378 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1380 linked
->UniformBlocks
= uniform_blocks
;
1381 linked
->NumUniformBlocks
= num_uniform_blocks
;
1382 ralloc_steal(linked
, linked
->UniformBlocks
);
1384 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1386 populate_symbol_table(linked
);
1388 /* The a pointer to the main function in the final linked shader (i.e., the
1389 * copy of the original shader that contained the main function).
1391 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1393 /* Move any instructions other than variable declarations or function
1394 * declarations into main.
1396 exec_node
*insertion_point
=
1397 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1400 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1401 if (shader_list
[i
] == main
)
1404 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1405 insertion_point
, true, linked
);
1408 /* Resolve initializers for global variables in the linked shader.
1410 unsigned num_linking_shaders
= num_shaders
;
1411 for (unsigned i
= 0; i
< num_shaders
; i
++)
1412 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
1414 gl_shader
**linking_shaders
=
1415 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
1417 memcpy(linking_shaders
, shader_list
,
1418 sizeof(linking_shaders
[0]) * num_shaders
);
1420 unsigned idx
= num_shaders
;
1421 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1422 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
1423 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
1424 idx
+= shader_list
[i
]->num_builtins_to_link
;
1427 assert(idx
== num_linking_shaders
);
1429 if (!link_function_calls(prog
, linked
, linking_shaders
,
1430 num_linking_shaders
)) {
1431 ctx
->Driver
.DeleteShader(ctx
, linked
);
1432 free(linking_shaders
);
1436 free(linking_shaders
);
1438 /* At this point linked should contain all of the linked IR, so
1439 * validate it to make sure nothing went wrong.
1441 validate_ir_tree(linked
->ir
);
1443 /* Set the size of geometry shader input arrays */
1444 if (linked
->Type
== GL_GEOMETRY_SHADER
) {
1445 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1446 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1447 foreach_iter(exec_list_iterator
, iter
, *linked
->ir
) {
1448 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
1449 ir
->accept(&input_resize_visitor
);
1453 /* Make a pass over all variable declarations to ensure that arrays with
1454 * unspecified sizes have a size specified. The size is inferred from the
1455 * max_array_access field.
1457 array_sizing_visitor v
;
1459 v
.fixup_unnamed_interface_types();
1465 * Update the sizes of linked shader uniform arrays to the maximum
1468 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1470 * If one or more elements of an array are active,
1471 * GetActiveUniform will return the name of the array in name,
1472 * subject to the restrictions listed above. The type of the array
1473 * is returned in type. The size parameter contains the highest
1474 * array element index used, plus one. The compiler or linker
1475 * determines the highest index used. There will be only one
1476 * active uniform reported by the GL per uniform array.
1480 update_array_sizes(struct gl_shader_program
*prog
)
1482 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1483 if (prog
->_LinkedShaders
[i
] == NULL
)
1486 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1487 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1489 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
) ||
1490 !var
->type
->is_array())
1493 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1494 * will not be eliminated. Since we always do std140, just
1495 * don't resize arrays in UBOs.
1497 * Atomic counters are supposed to get deterministic
1498 * locations assigned based on the declaration ordering and
1499 * sizes, array compaction would mess that up.
1501 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1504 unsigned int size
= var
->max_array_access
;
1505 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1506 if (prog
->_LinkedShaders
[j
] == NULL
)
1509 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1510 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1514 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1515 other_var
->max_array_access
> size
) {
1516 size
= other_var
->max_array_access
;
1521 if (size
+ 1 != var
->type
->length
) {
1522 /* If this is a built-in uniform (i.e., it's backed by some
1523 * fixed-function state), adjust the number of state slots to
1524 * match the new array size. The number of slots per array entry
1525 * is not known. It seems safe to assume that the total number of
1526 * slots is an integer multiple of the number of array elements.
1527 * Determine the number of slots per array element by dividing by
1528 * the old (total) size.
1530 if (var
->num_state_slots
> 0) {
1531 var
->num_state_slots
= (size
+ 1)
1532 * (var
->num_state_slots
/ var
->type
->length
);
1535 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1537 /* FINISHME: We should update the types of array
1538 * dereferences of this variable now.
1546 * Find a contiguous set of available bits in a bitmask.
1548 * \param used_mask Bits representing used (1) and unused (0) locations
1549 * \param needed_count Number of contiguous bits needed.
1552 * Base location of the available bits on success or -1 on failure.
1555 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1557 unsigned needed_mask
= (1 << needed_count
) - 1;
1558 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1560 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1561 * cannot optimize possibly infinite loops" for the loop below.
1563 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1566 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1567 if ((needed_mask
& ~used_mask
) == needed_mask
)
1578 * Assign locations for either VS inputs for FS outputs
1580 * \param prog Shader program whose variables need locations assigned
1581 * \param target_index Selector for the program target to receive location
1582 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1583 * \c MESA_SHADER_FRAGMENT.
1584 * \param max_index Maximum number of generic locations. This corresponds
1585 * to either the maximum number of draw buffers or the
1586 * maximum number of generic attributes.
1589 * If locations are successfully assigned, true is returned. Otherwise an
1590 * error is emitted to the shader link log and false is returned.
1593 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1594 unsigned target_index
,
1597 /* Mark invalid locations as being used.
1599 unsigned used_locations
= (max_index
>= 32)
1600 ? ~0 : ~((1 << max_index
) - 1);
1602 assert((target_index
== MESA_SHADER_VERTEX
)
1603 || (target_index
== MESA_SHADER_FRAGMENT
));
1605 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1609 /* Operate in a total of four passes.
1611 * 1. Invalidate the location assignments for all vertex shader inputs.
1613 * 2. Assign locations for inputs that have user-defined (via
1614 * glBindVertexAttribLocation) locations and outputs that have
1615 * user-defined locations (via glBindFragDataLocation).
1617 * 3. Sort the attributes without assigned locations by number of slots
1618 * required in decreasing order. Fragmentation caused by attribute
1619 * locations assigned by the application may prevent large attributes
1620 * from having enough contiguous space.
1622 * 4. Assign locations to any inputs without assigned locations.
1625 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1626 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1628 const enum ir_variable_mode direction
=
1629 (target_index
== MESA_SHADER_VERTEX
)
1630 ? ir_var_shader_in
: ir_var_shader_out
;
1633 /* Temporary storage for the set of attributes that need locations assigned.
1639 /* Used below in the call to qsort. */
1640 static int compare(const void *a
, const void *b
)
1642 const temp_attr
*const l
= (const temp_attr
*) a
;
1643 const temp_attr
*const r
= (const temp_attr
*) b
;
1645 /* Reversed because we want a descending order sort below. */
1646 return r
->slots
- l
->slots
;
1650 unsigned num_attr
= 0;
1652 foreach_list(node
, sh
->ir
) {
1653 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1655 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1658 if (var
->explicit_location
) {
1659 if ((var
->location
>= (int)(max_index
+ generic_base
))
1660 || (var
->location
< 0)) {
1662 "invalid explicit location %d specified for `%s'\n",
1664 ? var
->location
: var
->location
- generic_base
,
1668 } else if (target_index
== MESA_SHADER_VERTEX
) {
1671 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1672 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1673 var
->location
= binding
;
1674 var
->is_unmatched_generic_inout
= 0;
1676 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1680 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1681 assert(binding
>= FRAG_RESULT_DATA0
);
1682 var
->location
= binding
;
1683 var
->is_unmatched_generic_inout
= 0;
1685 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1691 /* If the variable is not a built-in and has a location statically
1692 * assigned in the shader (presumably via a layout qualifier), make sure
1693 * that it doesn't collide with other assigned locations. Otherwise,
1694 * add it to the list of variables that need linker-assigned locations.
1696 const unsigned slots
= var
->type
->count_attribute_slots();
1697 if (var
->location
!= -1) {
1698 if (var
->location
>= generic_base
&& var
->index
< 1) {
1699 /* From page 61 of the OpenGL 4.0 spec:
1701 * "LinkProgram will fail if the attribute bindings assigned
1702 * by BindAttribLocation do not leave not enough space to
1703 * assign a location for an active matrix attribute or an
1704 * active attribute array, both of which require multiple
1705 * contiguous generic attributes."
1707 * Previous versions of the spec contain similar language but omit
1708 * the bit about attribute arrays.
1710 * Page 61 of the OpenGL 4.0 spec also says:
1712 * "It is possible for an application to bind more than one
1713 * attribute name to the same location. This is referred to as
1714 * aliasing. This will only work if only one of the aliased
1715 * attributes is active in the executable program, or if no
1716 * path through the shader consumes more than one attribute of
1717 * a set of attributes aliased to the same location. A link
1718 * error can occur if the linker determines that every path
1719 * through the shader consumes multiple aliased attributes,
1720 * but implementations are not required to generate an error
1723 * These two paragraphs are either somewhat contradictory, or I
1724 * don't fully understand one or both of them.
1726 /* FINISHME: The code as currently written does not support
1727 * FINISHME: attribute location aliasing (see comment above).
1729 /* Mask representing the contiguous slots that will be used by
1732 const unsigned attr
= var
->location
- generic_base
;
1733 const unsigned use_mask
= (1 << slots
) - 1;
1735 /* Generate a link error if the set of bits requested for this
1736 * attribute overlaps any previously allocated bits.
1738 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1739 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1740 ? "vertex shader input" : "fragment shader output";
1742 "insufficient contiguous locations "
1743 "available for %s `%s' %d %d %d", string
,
1744 var
->name
, used_locations
, use_mask
, attr
);
1748 used_locations
|= (use_mask
<< attr
);
1754 to_assign
[num_attr
].slots
= slots
;
1755 to_assign
[num_attr
].var
= var
;
1759 /* If all of the attributes were assigned locations by the application (or
1760 * are built-in attributes with fixed locations), return early. This should
1761 * be the common case.
1766 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1768 if (target_index
== MESA_SHADER_VERTEX
) {
1769 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1770 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1771 * reserved to prevent it from being automatically allocated below.
1773 find_deref_visitor
find("gl_Vertex");
1775 if (find
.variable_found())
1776 used_locations
|= (1 << 0);
1779 for (unsigned i
= 0; i
< num_attr
; i
++) {
1780 /* Mask representing the contiguous slots that will be used by this
1783 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1785 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1788 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1789 ? "vertex shader input" : "fragment shader output";
1792 "insufficient contiguous locations "
1793 "available for %s `%s'",
1794 string
, to_assign
[i
].var
->name
);
1798 to_assign
[i
].var
->location
= generic_base
+ location
;
1799 to_assign
[i
].var
->is_unmatched_generic_inout
= 0;
1800 used_locations
|= (use_mask
<< location
);
1808 * Demote shader inputs and outputs that are not used in other stages
1811 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1813 foreach_list(node
, sh
->ir
) {
1814 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1816 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1819 /* A shader 'in' or 'out' variable is only really an input or output if
1820 * its value is used by other shader stages. This will cause the variable
1821 * to have a location assigned.
1823 if (var
->is_unmatched_generic_inout
) {
1824 var
->mode
= ir_var_auto
;
1831 * Store the gl_FragDepth layout in the gl_shader_program struct.
1834 store_fragdepth_layout(struct gl_shader_program
*prog
)
1836 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1840 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
1842 /* We don't look up the gl_FragDepth symbol directly because if
1843 * gl_FragDepth is not used in the shader, it's removed from the IR.
1844 * However, the symbol won't be removed from the symbol table.
1846 * We're only interested in the cases where the variable is NOT removed
1849 foreach_list(node
, ir
) {
1850 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1852 if (var
== NULL
|| var
->mode
!= ir_var_shader_out
) {
1856 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1857 switch (var
->depth_layout
) {
1858 case ir_depth_layout_none
:
1859 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
1861 case ir_depth_layout_any
:
1862 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
1864 case ir_depth_layout_greater
:
1865 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
1867 case ir_depth_layout_less
:
1868 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
1870 case ir_depth_layout_unchanged
:
1871 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
1882 * Validate the resources used by a program versus the implementation limits
1885 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1887 static const char *const shader_names
[MESA_SHADER_TYPES
] = {
1888 "vertex", "geometry", "fragment"
1891 const unsigned max_samplers
[MESA_SHADER_TYPES
] = {
1892 ctx
->Const
.VertexProgram
.MaxTextureImageUnits
,
1893 ctx
->Const
.GeometryProgram
.MaxTextureImageUnits
,
1894 ctx
->Const
.FragmentProgram
.MaxTextureImageUnits
1897 const unsigned max_default_uniform_components
[MESA_SHADER_TYPES
] = {
1898 ctx
->Const
.VertexProgram
.MaxUniformComponents
,
1899 ctx
->Const
.GeometryProgram
.MaxUniformComponents
,
1900 ctx
->Const
.FragmentProgram
.MaxUniformComponents
1903 const unsigned max_combined_uniform_components
[MESA_SHADER_TYPES
] = {
1904 ctx
->Const
.VertexProgram
.MaxCombinedUniformComponents
,
1905 ctx
->Const
.GeometryProgram
.MaxCombinedUniformComponents
,
1906 ctx
->Const
.FragmentProgram
.MaxCombinedUniformComponents
1909 const unsigned max_uniform_blocks
[MESA_SHADER_TYPES
] = {
1910 ctx
->Const
.VertexProgram
.MaxUniformBlocks
,
1911 ctx
->Const
.GeometryProgram
.MaxUniformBlocks
,
1912 ctx
->Const
.FragmentProgram
.MaxUniformBlocks
1915 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1916 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1921 if (sh
->num_samplers
> max_samplers
[i
]) {
1922 linker_error(prog
, "Too many %s shader texture samplers",
1926 if (sh
->num_uniform_components
> max_default_uniform_components
[i
]) {
1927 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
1928 linker_warning(prog
, "Too many %s shader default uniform block "
1929 "components, but the driver will try to optimize "
1930 "them out; this is non-portable out-of-spec "
1934 linker_error(prog
, "Too many %s shader default uniform block "
1940 if (sh
->num_combined_uniform_components
>
1941 max_combined_uniform_components
[i
]) {
1942 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
1943 linker_warning(prog
, "Too many %s shader uniform components, "
1944 "but the driver will try to optimize them out; "
1945 "this is non-portable out-of-spec behavior\n",
1948 linker_error(prog
, "Too many %s shader uniform components",
1954 unsigned blocks
[MESA_SHADER_TYPES
] = {0};
1955 unsigned total_uniform_blocks
= 0;
1957 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
1958 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1959 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
1961 total_uniform_blocks
++;
1965 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
1966 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
1967 prog
->NumUniformBlocks
,
1968 ctx
->Const
.MaxCombinedUniformBlocks
);
1970 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1971 if (blocks
[i
] > max_uniform_blocks
[i
]) {
1972 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
1975 max_uniform_blocks
[i
]);
1984 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1986 tfeedback_decl
*tfeedback_decls
= NULL
;
1987 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
1989 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1991 prog
->LinkStatus
= true; /* All error paths will set this to false */
1992 prog
->Validated
= false;
1993 prog
->_Used
= false;
1995 ralloc_free(prog
->InfoLog
);
1996 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1998 ralloc_free(prog
->UniformBlocks
);
1999 prog
->UniformBlocks
= NULL
;
2000 prog
->NumUniformBlocks
= 0;
2001 for (int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2002 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2003 prog
->UniformBlockStageIndex
[i
] = NULL
;
2006 ralloc_free(prog
->AtomicBuffers
);
2007 prog
->AtomicBuffers
= NULL
;
2008 prog
->NumAtomicBuffers
= 0;
2010 /* Separate the shaders into groups based on their type.
2012 struct gl_shader
**vert_shader_list
;
2013 unsigned num_vert_shaders
= 0;
2014 struct gl_shader
**frag_shader_list
;
2015 unsigned num_frag_shaders
= 0;
2016 struct gl_shader
**geom_shader_list
;
2017 unsigned num_geom_shaders
= 0;
2019 vert_shader_list
= (struct gl_shader
**)
2020 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2021 frag_shader_list
= (struct gl_shader
**)
2022 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2023 geom_shader_list
= (struct gl_shader
**)
2024 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2026 unsigned min_version
= UINT_MAX
;
2027 unsigned max_version
= 0;
2028 const bool is_es_prog
=
2029 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2030 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2031 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2032 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2034 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2035 linker_error(prog
, "all shaders must use same shading "
2036 "language version\n");
2040 switch (prog
->Shaders
[i
]->Type
) {
2041 case GL_VERTEX_SHADER
:
2042 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
2045 case GL_FRAGMENT_SHADER
:
2046 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
2049 case GL_GEOMETRY_SHADER
:
2050 geom_shader_list
[num_geom_shaders
] = prog
->Shaders
[i
];
2056 /* In desktop GLSL, different shader versions may be linked together. In
2057 * GLSL ES, all shader versions must be the same.
2059 if (is_es_prog
&& min_version
!= max_version
) {
2060 linker_error(prog
, "all shaders must use same shading "
2061 "language version\n");
2065 prog
->Version
= max_version
;
2066 prog
->IsES
= is_es_prog
;
2068 /* Geometry shaders have to be linked with vertex shaders.
2070 if (num_geom_shaders
> 0 && num_vert_shaders
== 0) {
2071 linker_error(prog
, "Geometry shader must be linked with "
2076 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2077 if (prog
->_LinkedShaders
[i
] != NULL
)
2078 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2080 prog
->_LinkedShaders
[i
] = NULL
;
2083 /* Link all shaders for a particular stage and validate the result.
2085 if (num_vert_shaders
> 0) {
2086 gl_shader
*const sh
=
2087 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
2090 if (!prog
->LinkStatus
)
2093 validate_vertex_shader_executable(prog
, sh
);
2094 if (!prog
->LinkStatus
)
2096 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2098 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2102 if (num_frag_shaders
> 0) {
2103 gl_shader
*const sh
=
2104 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
2107 if (!prog
->LinkStatus
)
2110 validate_fragment_shader_executable(prog
, sh
);
2111 if (!prog
->LinkStatus
)
2114 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
2118 if (num_geom_shaders
> 0) {
2119 gl_shader
*const sh
=
2120 link_intrastage_shaders(mem_ctx
, ctx
, prog
, geom_shader_list
,
2123 if (!prog
->LinkStatus
)
2126 validate_geometry_shader_executable(prog
, sh
);
2127 if (!prog
->LinkStatus
)
2129 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2131 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
],
2135 /* Here begins the inter-stage linking phase. Some initial validation is
2136 * performed, then locations are assigned for uniforms, attributes, and
2139 cross_validate_uniforms(prog
);
2140 if (!prog
->LinkStatus
)
2145 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
2146 if (prog
->_LinkedShaders
[prev
] != NULL
)
2150 /* Validate the inputs of each stage with the output of the preceding
2153 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
2154 if (prog
->_LinkedShaders
[i
] == NULL
)
2157 validate_interstage_interface_blocks(prog
, prog
->_LinkedShaders
[prev
],
2158 prog
->_LinkedShaders
[i
]);
2159 if (!prog
->LinkStatus
)
2162 cross_validate_outputs_to_inputs(prog
,
2163 prog
->_LinkedShaders
[prev
],
2164 prog
->_LinkedShaders
[i
]);
2165 if (!prog
->LinkStatus
)
2172 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2173 if (prog
->_LinkedShaders
[i
] != NULL
)
2174 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2177 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2178 * it before optimization because we want most of the checks to get
2179 * dropped thanks to constant propagation.
2181 * This rule also applies to GLSL ES 3.00.
2183 if (max_version
>= (is_es_prog
? 300 : 130)) {
2184 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2186 lower_discard_flow(sh
->ir
);
2190 if (!interstage_cross_validate_uniform_blocks(prog
))
2193 /* Do common optimization before assigning storage for attributes,
2194 * uniforms, and varyings. Later optimization could possibly make
2195 * some of that unused.
2197 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2198 if (prog
->_LinkedShaders
[i
] == NULL
)
2201 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2202 if (!prog
->LinkStatus
)
2205 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
) {
2206 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2209 unsigned max_unroll
= ctx
->ShaderCompilerOptions
[i
].MaxUnrollIterations
;
2211 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, max_unroll
, &ctx
->ShaderCompilerOptions
[i
]))
2215 /* Mark all generic shader inputs and outputs as unpaired. */
2216 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
2217 link_invalidate_variable_locations(
2218 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
);
2220 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
2221 link_invalidate_variable_locations(
2222 prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
2224 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
2225 link_invalidate_variable_locations(
2226 prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
);
2229 /* FINISHME: The value of the max_attribute_index parameter is
2230 * FINISHME: implementation dependent based on the value of
2231 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2232 * FINISHME: at least 16, so hardcode 16 for now.
2234 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2238 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2243 for (first
= 0; first
< MESA_SHADER_TYPES
; first
++) {
2244 if (prog
->_LinkedShaders
[first
] != NULL
)
2248 if (num_tfeedback_decls
!= 0) {
2249 /* From GL_EXT_transform_feedback:
2250 * A program will fail to link if:
2252 * * the <count> specified by TransformFeedbackVaryingsEXT is
2253 * non-zero, but the program object has no vertex or geometry
2256 if (first
== MESA_SHADER_FRAGMENT
) {
2257 linker_error(prog
, "Transform feedback varyings specified, but "
2258 "no vertex or geometry shader is present.");
2262 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2263 prog
->TransformFeedback
.NumVarying
);
2264 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2265 prog
->TransformFeedback
.VaryingNames
,
2270 /* Linking the stages in the opposite order (from fragment to vertex)
2271 * ensures that inter-shader outputs written to in an earlier stage are
2272 * eliminated if they are (transitively) not used in a later stage.
2275 for (last
= MESA_SHADER_TYPES
-1; last
>= 0; last
--) {
2276 if (prog
->_LinkedShaders
[last
] != NULL
)
2280 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2281 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2283 if (num_tfeedback_decls
!= 0) {
2284 /* There was no fragment shader, but we still have to assign varying
2285 * locations for use by transform feedback.
2287 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2289 num_tfeedback_decls
, tfeedback_decls
,
2294 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2295 num_tfeedback_decls
, tfeedback_decls
);
2297 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2299 /* Eliminate code that is now dead due to unused outputs being demoted.
2301 while (do_dead_code(sh
->ir
, false))
2304 else if (first
== MESA_SHADER_FRAGMENT
) {
2305 /* If the program only contains a fragment shader...
2307 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2309 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2310 num_tfeedback_decls
, tfeedback_decls
);
2312 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2314 while (do_dead_code(sh
->ir
, false))
2319 for (int i
= next
- 1; i
>= 0; i
--) {
2320 if (prog
->_LinkedShaders
[i
] == NULL
)
2323 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2324 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2325 unsigned gs_input_vertices
=
2326 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2328 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2329 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2330 tfeedback_decls
, gs_input_vertices
))
2333 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2334 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2337 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2338 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2340 /* Eliminate code that is now dead due to unused outputs being demoted.
2342 while (do_dead_code(sh_i
->ir
, false))
2344 while (do_dead_code(sh_next
->ir
, false))
2347 /* This must be done after all dead varyings are eliminated. */
2348 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2350 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2356 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2359 update_array_sizes(prog
);
2360 link_assign_uniform_locations(prog
);
2361 link_assign_atomic_counter_resources(ctx
, prog
);
2362 store_fragdepth_layout(prog
);
2364 check_resources(ctx
, prog
);
2365 link_check_atomic_counter_resources(ctx
, prog
);
2367 if (!prog
->LinkStatus
)
2370 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2371 * present in a linked program. By checking prog->IsES, we also
2372 * catch the GL_ARB_ES2_compatibility case.
2374 if (!prog
->InternalSeparateShader
&&
2375 (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)) {
2376 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2377 linker_error(prog
, "program lacks a vertex shader\n");
2378 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2379 linker_error(prog
, "program lacks a fragment shader\n");
2383 /* FINISHME: Assign fragment shader output locations. */
2386 free(vert_shader_list
);
2387 free(frag_shader_list
);
2388 free(geom_shader_list
);
2390 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2391 if (prog
->_LinkedShaders
[i
] == NULL
)
2394 /* Retain any live IR, but trash the rest. */
2395 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2397 /* The symbol table in the linked shaders may contain references to
2398 * variables that were removed (e.g., unused uniforms). Since it may
2399 * contain junk, there is no possible valid use. Delete it and set the
2402 delete prog
->_LinkedShaders
[i
]->symbols
;
2403 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2406 ralloc_free(mem_ctx
);