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 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
113 actual_node
, &ir
->actual_parameters
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
115 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
117 if (sig_param
->data
.mode
== ir_var_function_out
||
118 sig_param
->data
.mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
127 if (ir
->return_deref
!= NULL
) {
128 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
130 if (strcmp(name
, var
->name
) == 0) {
136 return visit_continue_with_parent
;
139 bool variable_found()
145 const char *name
; /**< Find writes to a variable with this name. */
146 bool found
; /**< Was a write to the variable found? */
151 * Visitor that determines whether or not a variable is ever read.
153 class find_deref_visitor
: public ir_hierarchical_visitor
{
155 find_deref_visitor(const char *name
)
156 : name(name
), found(false)
161 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
163 if (strcmp(this->name
, ir
->var
->name
) == 0) {
168 return visit_continue
;
171 bool variable_found() const
177 const char *name
; /**< Find writes to a variable with this name. */
178 bool found
; /**< Was a write to the variable found? */
182 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
184 unsigned num_vertices
;
185 gl_shader_program
*prog
;
187 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
189 this->num_vertices
= num_vertices
;
193 virtual ~geom_array_resize_visitor()
198 virtual ir_visitor_status
visit(ir_variable
*var
)
200 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
201 return visit_continue
;
203 unsigned size
= var
->type
->length
;
205 /* Generate a link error if the shader has declared this array with an
208 if (size
&& size
!= this->num_vertices
) {
209 linker_error(this->prog
, "size of array %s declared as %u, "
210 "but number of input vertices is %u\n",
211 var
->name
, size
, this->num_vertices
);
212 return visit_continue
;
215 /* Generate a link error if the shader attempts to access an input
216 * array using an index too large for its actual size assigned at link
219 if (var
->data
.max_array_access
>= this->num_vertices
) {
220 linker_error(this->prog
, "geometry shader accesses element %i of "
221 "%s, but only %i input vertices\n",
222 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
223 return visit_continue
;
226 var
->type
= glsl_type::get_array_instance(var
->type
->element_type(),
228 var
->data
.max_array_access
= this->num_vertices
- 1;
230 return visit_continue
;
233 /* Dereferences of input variables need to be updated so that their type
234 * matches the newly assigned type of the variable they are accessing. */
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 /* Dereferences of 2D input arrays need to be updated so that their type
242 * matches the newly assigned type of the array they are accessing. */
243 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
245 const glsl_type
*const vt
= ir
->array
->type
;
247 ir
->type
= vt
->element_type();
248 return visit_continue
;
254 * Visitor that determines whether or not a shader uses ir_end_primitive.
256 class find_end_primitive_visitor
: public ir_hierarchical_visitor
{
258 find_end_primitive_visitor()
264 virtual ir_visitor_status
visit(ir_end_primitive
*)
270 bool end_primitive_found()
279 } /* anonymous namespace */
282 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
286 ralloc_strcat(&prog
->InfoLog
, "error: ");
288 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
291 prog
->LinkStatus
= false;
296 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
300 ralloc_strcat(&prog
->InfoLog
, "warning: ");
302 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
309 * Given a string identifying a program resource, break it into a base name
310 * and an optional array index in square brackets.
312 * If an array index is present, \c out_base_name_end is set to point to the
313 * "[" that precedes the array index, and the array index itself is returned
316 * If no array index is present (or if the array index is negative or
317 * mal-formed), \c out_base_name_end, is set to point to the null terminator
318 * at the end of the input string, and -1 is returned.
320 * Only the final array index is parsed; if the string contains other array
321 * indices (or structure field accesses), they are left in the base name.
323 * No attempt is made to check that the base name is properly formed;
324 * typically the caller will look up the base name in a hash table, so
325 * ill-formed base names simply turn into hash table lookup failures.
328 parse_program_resource_name(const GLchar
*name
,
329 const GLchar
**out_base_name_end
)
331 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
333 * "When an integer array element or block instance number is part of
334 * the name string, it will be specified in decimal form without a "+"
335 * or "-" sign or any extra leading zeroes. Additionally, the name
336 * string will not include white space anywhere in the string."
339 const size_t len
= strlen(name
);
340 *out_base_name_end
= name
+ len
;
342 if (len
== 0 || name
[len
-1] != ']')
345 /* Walk backwards over the string looking for a non-digit character. This
346 * had better be the opening bracket for an array index.
348 * Initially, i specifies the location of the ']'. Since the string may
349 * contain only the ']' charcater, walk backwards very carefully.
352 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
355 if ((i
== 0) || name
[i
-1] != '[')
358 long array_index
= strtol(&name
[i
], NULL
, 10);
362 *out_base_name_end
= name
+ (i
- 1);
368 link_invalidate_variable_locations(exec_list
*ir
)
370 foreach_list(node
, ir
) {
371 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
376 /* Only assign locations for variables that lack an explicit location.
377 * Explicit locations are set for all built-in variables, generic vertex
378 * shader inputs (via layout(location=...)), and generic fragment shader
379 * outputs (also via layout(location=...)).
381 if (!var
->data
.explicit_location
) {
382 var
->data
.location
= -1;
383 var
->data
.location_frac
= 0;
386 /* ir_variable::is_unmatched_generic_inout is used by the linker while
387 * connecting outputs from one stage to inputs of the next stage.
389 * There are two implicit assumptions here. First, we assume that any
390 * built-in variable (i.e., non-generic in or out) will have
391 * explicit_location set. Second, we assume that any generic in or out
392 * will not have explicit_location set.
394 * This second assumption will only be valid until
395 * GL_ARB_separate_shader_objects is supported. When that extension is
396 * implemented, this function will need some modifications.
398 if (!var
->data
.explicit_location
) {
399 var
->data
.is_unmatched_generic_inout
= 1;
401 var
->data
.is_unmatched_generic_inout
= 0;
408 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
410 * Also check for errors based on incorrect usage of gl_ClipVertex and
413 * Return false if an error was reported.
416 analyze_clip_usage(struct gl_shader_program
*prog
,
417 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
418 GLuint
*ClipDistanceArraySize
)
420 *ClipDistanceArraySize
= 0;
422 if (!prog
->IsES
&& prog
->Version
>= 130) {
423 /* From section 7.1 (Vertex Shader Special Variables) of the
426 * "It is an error for a shader to statically write both
427 * gl_ClipVertex and gl_ClipDistance."
429 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
430 * gl_ClipVertex nor gl_ClipDistance.
432 find_assignment_visitor
clip_vertex("gl_ClipVertex");
433 find_assignment_visitor
clip_distance("gl_ClipDistance");
435 clip_vertex
.run(shader
->ir
);
436 clip_distance
.run(shader
->ir
);
437 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
438 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
439 "and `gl_ClipDistance'\n",
440 _mesa_shader_stage_to_string(shader
->Stage
));
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(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(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
->data
.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
->data
.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
;
612 } else if (var
->type
->is_record()
613 && existing
->type
->is_record()
614 && existing
->type
->record_compare(var
->type
)) {
615 existing
->type
= var
->type
;
617 linker_error(prog
, "%s `%s' declared as type "
618 "`%s' and type `%s'\n",
620 var
->name
, var
->type
->name
,
621 existing
->type
->name
);
626 if (var
->data
.explicit_location
) {
627 if (existing
->data
.explicit_location
628 && (var
->data
.location
!= existing
->data
.location
)) {
629 linker_error(prog
, "explicit locations for %s "
630 "`%s' have differing values\n",
631 mode_string(var
), var
->name
);
635 existing
->data
.location
= var
->data
.location
;
636 existing
->data
.explicit_location
= true;
639 /* From the GLSL 4.20 specification:
640 * "A link error will result if two compilation units in a program
641 * specify different integer-constant bindings for the same
642 * opaque-uniform name. However, it is not an error to specify a
643 * binding on some but not all declarations for the same name"
645 if (var
->data
.explicit_binding
) {
646 if (existing
->data
.explicit_binding
&&
647 var
->data
.binding
!= existing
->data
.binding
) {
648 linker_error(prog
, "explicit bindings for %s "
649 "`%s' have differing values\n",
650 mode_string(var
), var
->name
);
654 existing
->data
.binding
= var
->data
.binding
;
655 existing
->data
.explicit_binding
= true;
658 if (var
->type
->contains_atomic() &&
659 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
660 linker_error(prog
, "offset specifications for %s "
661 "`%s' have differing values\n",
662 mode_string(var
), var
->name
);
666 /* Validate layout qualifiers for gl_FragDepth.
668 * From the AMD/ARB_conservative_depth specs:
670 * "If gl_FragDepth is redeclared in any fragment shader in a
671 * program, it must be redeclared in all fragment shaders in
672 * that program that have static assignments to
673 * gl_FragDepth. All redeclarations of gl_FragDepth in all
674 * fragment shaders in a single program must have the same set
677 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
678 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
679 bool layout_differs
=
680 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
682 if (layout_declared
&& layout_differs
) {
684 "All redeclarations of gl_FragDepth in all "
685 "fragment shaders in a single program must have "
686 "the same set of qualifiers.");
689 if (var
->data
.used
&& layout_differs
) {
691 "If gl_FragDepth is redeclared with a layout "
692 "qualifier in any fragment shader, it must be "
693 "redeclared with the same layout qualifier in "
694 "all fragment shaders that have assignments to "
699 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
701 * "If a shared global has multiple initializers, the
702 * initializers must all be constant expressions, and they
703 * must all have the same value. Otherwise, a link error will
704 * result. (A shared global having only one initializer does
705 * not require that initializer to be a constant expression.)"
707 * Previous to 4.20 the GLSL spec simply said that initializers
708 * must have the same value. In this case of non-constant
709 * initializers, this was impossible to determine. As a result,
710 * no vendor actually implemented that behavior. The 4.20
711 * behavior matches the implemented behavior of at least one other
712 * vendor, so we'll implement that for all GLSL versions.
714 if (var
->constant_initializer
!= NULL
) {
715 if (existing
->constant_initializer
!= NULL
) {
716 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
717 linker_error(prog
, "initializers for %s "
718 "`%s' have differing values\n",
719 mode_string(var
), var
->name
);
723 /* If the first-seen instance of a particular uniform did not
724 * have an initializer but a later instance does, copy the
725 * initializer to the version stored in the symbol table.
727 /* FINISHME: This is wrong. The constant_value field should
728 * FINISHME: not be modified! Imagine a case where a shader
729 * FINISHME: without an initializer is linked in two different
730 * FINISHME: programs with shaders that have differing
731 * FINISHME: initializers. Linking with the first will
732 * FINISHME: modify the shader, and linking with the second
733 * FINISHME: will fail.
735 existing
->constant_initializer
=
736 var
->constant_initializer
->clone(ralloc_parent(existing
),
741 if (var
->data
.has_initializer
) {
742 if (existing
->data
.has_initializer
743 && (var
->constant_initializer
== NULL
744 || existing
->constant_initializer
== NULL
)) {
746 "shared global variable `%s' has multiple "
747 "non-constant initializers.\n",
752 /* Some instance had an initializer, so keep track of that. In
753 * this location, all sorts of initializers (constant or
754 * otherwise) will propagate the existence to the variable
755 * stored in the symbol table.
757 existing
->data
.has_initializer
= true;
760 if (existing
->data
.invariant
!= var
->data
.invariant
) {
761 linker_error(prog
, "declarations for %s `%s' have "
762 "mismatching invariant qualifiers\n",
763 mode_string(var
), var
->name
);
766 if (existing
->data
.centroid
!= var
->data
.centroid
) {
767 linker_error(prog
, "declarations for %s `%s' have "
768 "mismatching centroid qualifiers\n",
769 mode_string(var
), var
->name
);
772 if (existing
->data
.sample
!= var
->data
.sample
) {
773 linker_error(prog
, "declarations for %s `%s` have "
774 "mismatching sample qualifiers\n",
775 mode_string(var
), var
->name
);
779 variables
.add_variable(var
);
786 * Perform validation of uniforms used across multiple shader stages
789 cross_validate_uniforms(struct gl_shader_program
*prog
)
791 cross_validate_globals(prog
, prog
->_LinkedShaders
,
792 MESA_SHADER_STAGES
, true);
796 * Accumulates the array of prog->UniformBlocks and checks that all
797 * definitons of blocks agree on their contents.
800 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
802 unsigned max_num_uniform_blocks
= 0;
803 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
804 if (prog
->_LinkedShaders
[i
])
805 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
808 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
809 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
811 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
812 max_num_uniform_blocks
);
813 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
814 prog
->UniformBlockStageIndex
[i
][j
] = -1;
819 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
820 int index
= link_cross_validate_uniform_block(prog
,
821 &prog
->UniformBlocks
,
822 &prog
->NumUniformBlocks
,
823 &sh
->UniformBlocks
[j
]);
826 linker_error(prog
, "uniform block `%s' has mismatching definitions",
827 sh
->UniformBlocks
[j
].Name
);
831 prog
->UniformBlockStageIndex
[i
][index
] = j
;
840 * Populates a shaders symbol table with all global declarations
843 populate_symbol_table(gl_shader
*sh
)
845 sh
->symbols
= new(sh
) glsl_symbol_table
;
847 foreach_list(node
, sh
->ir
) {
848 ir_instruction
*const inst
= (ir_instruction
*) node
;
852 if ((func
= inst
->as_function()) != NULL
) {
853 sh
->symbols
->add_function(func
);
854 } else if ((var
= inst
->as_variable()) != NULL
) {
855 sh
->symbols
->add_variable(var
);
862 * Remap variables referenced in an instruction tree
864 * This is used when instruction trees are cloned from one shader and placed in
865 * another. These trees will contain references to \c ir_variable nodes that
866 * do not exist in the target shader. This function finds these \c ir_variable
867 * references and replaces the references with matching variables in the target
870 * If there is no matching variable in the target shader, a clone of the
871 * \c ir_variable is made and added to the target shader. The new variable is
872 * added to \b both the instruction stream and the symbol table.
874 * \param inst IR tree that is to be processed.
875 * \param symbols Symbol table containing global scope symbols in the
877 * \param instructions Instruction stream where new variable declarations
881 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
884 class remap_visitor
: public ir_hierarchical_visitor
{
886 remap_visitor(struct gl_shader
*target
,
889 this->target
= target
;
890 this->symbols
= target
->symbols
;
891 this->instructions
= target
->ir
;
895 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
897 if (ir
->var
->data
.mode
== ir_var_temporary
) {
898 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
902 return visit_continue
;
905 ir_variable
*const existing
=
906 this->symbols
->get_variable(ir
->var
->name
);
907 if (existing
!= NULL
)
910 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
912 this->symbols
->add_variable(copy
);
913 this->instructions
->push_head(copy
);
917 return visit_continue
;
921 struct gl_shader
*target
;
922 glsl_symbol_table
*symbols
;
923 exec_list
*instructions
;
927 remap_visitor
v(target
, temps
);
934 * Move non-declarations from one instruction stream to another
936 * The intended usage pattern of this function is to pass the pointer to the
937 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
938 * pointer) for \c last and \c false for \c make_copies on the first
939 * call. Successive calls pass the return value of the previous call for
940 * \c last and \c true for \c make_copies.
942 * \param instructions Source instruction stream
943 * \param last Instruction after which new instructions should be
944 * inserted in the target instruction stream
945 * \param make_copies Flag selecting whether instructions in \c instructions
946 * should be copied (via \c ir_instruction::clone) into the
947 * target list or moved.
950 * The new "last" instruction in the target instruction stream. This pointer
951 * is suitable for use as the \c last parameter of a later call to this
955 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
956 bool make_copies
, gl_shader
*target
)
958 hash_table
*temps
= NULL
;
961 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
962 hash_table_pointer_compare
);
964 foreach_list_safe(node
, instructions
) {
965 ir_instruction
*inst
= (ir_instruction
*) node
;
967 if (inst
->as_function())
970 ir_variable
*var
= inst
->as_variable();
971 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
974 assert(inst
->as_assignment()
976 || inst
->as_if() /* for initializers with the ?: operator */
977 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
980 inst
= inst
->clone(target
, NULL
);
983 hash_table_insert(temps
, inst
, var
);
985 remap_variables(inst
, target
, temps
);
990 last
->insert_after(inst
);
995 hash_table_dtor(temps
);
1001 * Get the function signature for main from a shader
1003 static ir_function_signature
*
1004 get_main_function_signature(gl_shader
*sh
)
1006 ir_function
*const f
= sh
->symbols
->get_function("main");
1008 exec_list void_parameters
;
1010 /* Look for the 'void main()' signature and ensure that it's defined.
1011 * This keeps the linker from accidentally pick a shader that just
1012 * contains a prototype for main.
1014 * We don't have to check for multiple definitions of main (in multiple
1015 * shaders) because that would have already been caught above.
1017 ir_function_signature
*sig
= f
->matching_signature(NULL
, &void_parameters
);
1018 if ((sig
!= NULL
) && sig
->is_defined
) {
1028 * This class is only used in link_intrastage_shaders() below but declaring
1029 * it inside that function leads to compiler warnings with some versions of
1032 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1034 array_sizing_visitor()
1035 : mem_ctx(ralloc_context(NULL
)),
1036 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1037 hash_table_pointer_compare
))
1041 ~array_sizing_visitor()
1043 hash_table_dtor(this->unnamed_interfaces
);
1044 ralloc_free(this->mem_ctx
);
1047 virtual ir_visitor_status
visit(ir_variable
*var
)
1049 fixup_type(&var
->type
, var
->data
.max_array_access
);
1050 if (var
->type
->is_interface()) {
1051 if (interface_contains_unsized_arrays(var
->type
)) {
1052 const glsl_type
*new_type
=
1053 resize_interface_members(var
->type
, var
->max_ifc_array_access
);
1054 var
->type
= new_type
;
1055 var
->change_interface_type(new_type
);
1057 } else if (var
->type
->is_array() &&
1058 var
->type
->fields
.array
->is_interface()) {
1059 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1060 const glsl_type
*new_type
=
1061 resize_interface_members(var
->type
->fields
.array
,
1062 var
->max_ifc_array_access
);
1063 var
->change_interface_type(new_type
);
1065 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1067 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1068 /* Store a pointer to the variable in the unnamed_interfaces
1071 ir_variable
**interface_vars
= (ir_variable
**)
1072 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1073 if (interface_vars
== NULL
) {
1074 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1076 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1079 unsigned index
= ifc_type
->field_index(var
->name
);
1080 assert(index
< ifc_type
->length
);
1081 assert(interface_vars
[index
] == NULL
);
1082 interface_vars
[index
] = var
;
1084 return visit_continue
;
1088 * For each unnamed interface block that was discovered while running the
1089 * visitor, adjust the interface type to reflect the newly assigned array
1090 * sizes, and fix up the ir_variable nodes to point to the new interface
1093 void fixup_unnamed_interface_types()
1095 hash_table_call_foreach(this->unnamed_interfaces
,
1096 fixup_unnamed_interface_type
, NULL
);
1101 * If the type pointed to by \c type represents an unsized array, replace
1102 * it with a sized array whose size is determined by max_array_access.
1104 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1106 if ((*type
)->is_unsized_array()) {
1107 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1108 max_array_access
+ 1);
1109 assert(*type
!= NULL
);
1114 * Determine whether the given interface type contains unsized arrays (if
1115 * it doesn't, array_sizing_visitor doesn't need to process it).
1117 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1119 for (unsigned i
= 0; i
< type
->length
; i
++) {
1120 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1121 if (elem_type
->is_unsized_array())
1128 * Create a new interface type based on the given type, with unsized arrays
1129 * replaced by sized arrays whose size is determined by
1130 * max_ifc_array_access.
1132 static const glsl_type
*
1133 resize_interface_members(const glsl_type
*type
,
1134 const unsigned *max_ifc_array_access
)
1136 unsigned num_fields
= type
->length
;
1137 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1138 memcpy(fields
, type
->fields
.structure
,
1139 num_fields
* sizeof(*fields
));
1140 for (unsigned i
= 0; i
< num_fields
; i
++) {
1141 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1143 glsl_interface_packing packing
=
1144 (glsl_interface_packing
) type
->interface_packing
;
1145 const glsl_type
*new_ifc_type
=
1146 glsl_type::get_interface_instance(fields
, num_fields
,
1147 packing
, type
->name
);
1149 return new_ifc_type
;
1152 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1155 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1156 ir_variable
**interface_vars
= (ir_variable
**) data
;
1157 unsigned num_fields
= ifc_type
->length
;
1158 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1159 memcpy(fields
, ifc_type
->fields
.structure
,
1160 num_fields
* sizeof(*fields
));
1161 bool interface_type_changed
= false;
1162 for (unsigned i
= 0; i
< num_fields
; i
++) {
1163 if (interface_vars
[i
] != NULL
&&
1164 fields
[i
].type
!= interface_vars
[i
]->type
) {
1165 fields
[i
].type
= interface_vars
[i
]->type
;
1166 interface_type_changed
= true;
1169 if (!interface_type_changed
) {
1173 glsl_interface_packing packing
=
1174 (glsl_interface_packing
) ifc_type
->interface_packing
;
1175 const glsl_type
*new_ifc_type
=
1176 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1179 for (unsigned i
= 0; i
< num_fields
; i
++) {
1180 if (interface_vars
[i
] != NULL
)
1181 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1186 * Memory context used to allocate the data in \c unnamed_interfaces.
1191 * Hash table from const glsl_type * to an array of ir_variable *'s
1192 * pointing to the ir_variables constituting each unnamed interface block.
1194 hash_table
*unnamed_interfaces
;
1198 * Performs the cross-validation of layout qualifiers specified in
1199 * redeclaration of gl_FragCoord for the attached fragment shaders,
1200 * and propagates them to the linked FS and linked shader program.
1203 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1204 struct gl_shader
*linked_shader
,
1205 struct gl_shader
**shader_list
,
1206 unsigned num_shaders
)
1208 linked_shader
->redeclares_gl_fragcoord
= false;
1209 linked_shader
->uses_gl_fragcoord
= false;
1210 linked_shader
->origin_upper_left
= false;
1211 linked_shader
->pixel_center_integer
= false;
1213 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1214 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1217 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1218 struct gl_shader
*shader
= shader_list
[i
];
1219 /* From the GLSL 1.50 spec, page 39:
1221 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1222 * it must be redeclared in all the fragment shaders in that program
1223 * that have a static use gl_FragCoord."
1225 * Exclude the case when one of the 'linked_shader' or 'shader' redeclares
1226 * gl_FragCoord with no layout qualifiers but the other one doesn't
1227 * redeclare it. If we strictly follow GLSL 1.50 spec's language, it
1228 * should be a link error. But, generating link error for this case will
1229 * be a wrong behaviour which spec didn't intend to do and it could also
1230 * break some applications.
1232 if ((linked_shader
->redeclares_gl_fragcoord
1233 && !shader
->redeclares_gl_fragcoord
1234 && shader
->uses_gl_fragcoord
1235 && (linked_shader
->origin_upper_left
1236 || linked_shader
->pixel_center_integer
))
1237 || (shader
->redeclares_gl_fragcoord
1238 && !linked_shader
->redeclares_gl_fragcoord
1239 && linked_shader
->uses_gl_fragcoord
1240 && (shader
->origin_upper_left
1241 || shader
->pixel_center_integer
))) {
1242 linker_error(prog
, "fragment shader defined with conflicting "
1243 "layout qualifiers for gl_FragCoord\n");
1246 /* From the GLSL 1.50 spec, page 39:
1248 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1249 * single program must have the same set of qualifiers."
1251 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1252 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1253 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1254 linker_error(prog
, "fragment shader defined with conflicting "
1255 "layout qualifiers for gl_FragCoord\n");
1258 /* Update the linked shader state. Note that uses_gl_fragcoord should
1259 * accumulate the results. The other values should replace. If there
1260 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1261 * are already known to be the same.
1263 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1264 linked_shader
->redeclares_gl_fragcoord
=
1265 shader
->redeclares_gl_fragcoord
;
1266 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1267 || shader
->uses_gl_fragcoord
;
1268 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1269 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1275 * Performs the cross-validation of geometry shader max_vertices and
1276 * primitive type layout qualifiers for the attached geometry shaders,
1277 * and propagates them to the linked GS and linked shader program.
1280 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1281 struct gl_shader
*linked_shader
,
1282 struct gl_shader
**shader_list
,
1283 unsigned num_shaders
)
1285 linked_shader
->Geom
.VerticesOut
= 0;
1286 linked_shader
->Geom
.Invocations
= 0;
1287 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1288 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1290 /* No in/out qualifiers defined for anything but GLSL 1.50+
1291 * geometry shaders so far.
1293 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1296 /* From the GLSL 1.50 spec, page 46:
1298 * "All geometry shader output layout declarations in a program
1299 * must declare the same layout and same value for
1300 * max_vertices. There must be at least one geometry output
1301 * layout declaration somewhere in a program, but not all
1302 * geometry shaders (compilation units) are required to
1306 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1307 struct gl_shader
*shader
= shader_list
[i
];
1309 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1310 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1311 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1312 linker_error(prog
, "geometry shader defined with conflicting "
1316 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1319 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1320 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1321 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1322 linker_error(prog
, "geometry shader defined with conflicting "
1326 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1329 if (shader
->Geom
.VerticesOut
!= 0) {
1330 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1331 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1332 linker_error(prog
, "geometry shader defined with conflicting "
1333 "output vertex count (%d and %d)\n",
1334 linked_shader
->Geom
.VerticesOut
,
1335 shader
->Geom
.VerticesOut
);
1338 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1341 if (shader
->Geom
.Invocations
!= 0) {
1342 if (linked_shader
->Geom
.Invocations
!= 0 &&
1343 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1344 linker_error(prog
, "geometry shader defined with conflicting "
1345 "invocation count (%d and %d)\n",
1346 linked_shader
->Geom
.Invocations
,
1347 shader
->Geom
.Invocations
);
1350 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1354 /* Just do the intrastage -> interstage propagation right now,
1355 * since we already know we're in the right type of shader program
1358 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1360 "geometry shader didn't declare primitive input type\n");
1363 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1365 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1367 "geometry shader didn't declare primitive output type\n");
1370 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1372 if (linked_shader
->Geom
.VerticesOut
== 0) {
1374 "geometry shader didn't declare max_vertices\n");
1377 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1379 if (linked_shader
->Geom
.Invocations
== 0)
1380 linked_shader
->Geom
.Invocations
= 1;
1382 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1387 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1388 * qualifiers for the attached compute shaders, and propagate them to the
1389 * linked CS and linked shader program.
1392 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1393 struct gl_shader
*linked_shader
,
1394 struct gl_shader
**shader_list
,
1395 unsigned num_shaders
)
1397 for (int i
= 0; i
< 3; i
++)
1398 linked_shader
->Comp
.LocalSize
[i
] = 0;
1400 /* This function is called for all shader stages, but it only has an effect
1401 * for compute shaders.
1403 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1406 /* From the ARB_compute_shader spec, in the section describing local size
1409 * If multiple compute shaders attached to a single program object
1410 * declare local work-group size, the declarations must be identical;
1411 * otherwise a link-time error results. Furthermore, if a program
1412 * object contains any compute shaders, at least one must contain an
1413 * input layout qualifier specifying the local work sizes of the
1414 * program, or a link-time error will occur.
1416 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1417 struct gl_shader
*shader
= shader_list
[sh
];
1419 if (shader
->Comp
.LocalSize
[0] != 0) {
1420 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1421 for (int i
= 0; i
< 3; i
++) {
1422 if (linked_shader
->Comp
.LocalSize
[i
] !=
1423 shader
->Comp
.LocalSize
[i
]) {
1424 linker_error(prog
, "compute shader defined with conflicting "
1430 for (int i
= 0; i
< 3; i
++)
1431 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1435 /* Just do the intrastage -> interstage propagation right now,
1436 * since we already know we're in the right type of shader program
1439 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1440 linker_error(prog
, "compute shader didn't declare local size\n");
1443 for (int i
= 0; i
< 3; i
++)
1444 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1449 * Combine a group of shaders for a single stage to generate a linked shader
1452 * If this function is supplied a single shader, it is cloned, and the new
1453 * shader is returned.
1455 static struct gl_shader
*
1456 link_intrastage_shaders(void *mem_ctx
,
1457 struct gl_context
*ctx
,
1458 struct gl_shader_program
*prog
,
1459 struct gl_shader
**shader_list
,
1460 unsigned num_shaders
)
1462 struct gl_uniform_block
*uniform_blocks
= NULL
;
1464 /* Check that global variables defined in multiple shaders are consistent.
1466 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1467 if (!prog
->LinkStatus
)
1470 /* Check that interface blocks defined in multiple shaders are consistent.
1472 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1474 if (!prog
->LinkStatus
)
1477 /* Link up uniform blocks defined within this stage. */
1478 const unsigned num_uniform_blocks
=
1479 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1482 /* Check that there is only a single definition of each function signature
1483 * across all shaders.
1485 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1486 foreach_list(node
, shader_list
[i
]->ir
) {
1487 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
1492 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1493 ir_function
*const other
=
1494 shader_list
[j
]->symbols
->get_function(f
->name
);
1496 /* If the other shader has no function (and therefore no function
1497 * signatures) with the same name, skip to the next shader.
1502 foreach_list(n
, &f
->signatures
) {
1503 ir_function_signature
*sig
= (ir_function_signature
*) n
;
1505 if (!sig
->is_defined
|| sig
->is_builtin())
1508 ir_function_signature
*other_sig
=
1509 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1511 if ((other_sig
!= NULL
) && other_sig
->is_defined
1512 && !other_sig
->is_builtin()) {
1513 linker_error(prog
, "function `%s' is multiply defined",
1522 /* Find the shader that defines main, and make a clone of it.
1524 * Starting with the clone, search for undefined references. If one is
1525 * found, find the shader that defines it. Clone the reference and add
1526 * it to the shader. Repeat until there are no undefined references or
1527 * until a reference cannot be resolved.
1529 gl_shader
*main
= NULL
;
1530 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1531 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
1532 main
= shader_list
[i
];
1538 linker_error(prog
, "%s shader lacks `main'\n",
1539 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1543 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1544 linked
->ir
= new(linked
) exec_list
;
1545 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1547 linked
->UniformBlocks
= uniform_blocks
;
1548 linked
->NumUniformBlocks
= num_uniform_blocks
;
1549 ralloc_steal(linked
, linked
->UniformBlocks
);
1551 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1552 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1553 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1555 populate_symbol_table(linked
);
1557 /* The a pointer to the main function in the final linked shader (i.e., the
1558 * copy of the original shader that contained the main function).
1560 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1562 /* Move any instructions other than variable declarations or function
1563 * declarations into main.
1565 exec_node
*insertion_point
=
1566 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1569 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1570 if (shader_list
[i
] == main
)
1573 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1574 insertion_point
, true, linked
);
1577 /* Check if any shader needs built-in functions. */
1578 bool need_builtins
= false;
1579 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1580 if (shader_list
[i
]->uses_builtin_functions
) {
1581 need_builtins
= true;
1587 if (need_builtins
) {
1588 /* Make a temporary array one larger than shader_list, which will hold
1589 * the built-in function shader as well.
1591 gl_shader
**linking_shaders
= (gl_shader
**)
1592 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1593 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1594 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1596 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1598 free(linking_shaders
);
1600 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1605 ctx
->Driver
.DeleteShader(ctx
, linked
);
1609 /* At this point linked should contain all of the linked IR, so
1610 * validate it to make sure nothing went wrong.
1612 validate_ir_tree(linked
->ir
);
1614 /* Set the size of geometry shader input arrays */
1615 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1616 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1617 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1618 foreach_list(n
, linked
->ir
) {
1619 ir_instruction
*ir
= (ir_instruction
*) n
;
1620 ir
->accept(&input_resize_visitor
);
1624 /* Make a pass over all variable declarations to ensure that arrays with
1625 * unspecified sizes have a size specified. The size is inferred from the
1626 * max_array_access field.
1628 array_sizing_visitor v
;
1630 v
.fixup_unnamed_interface_types();
1636 * Update the sizes of linked shader uniform arrays to the maximum
1639 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1641 * If one or more elements of an array are active,
1642 * GetActiveUniform will return the name of the array in name,
1643 * subject to the restrictions listed above. The type of the array
1644 * is returned in type. The size parameter contains the highest
1645 * array element index used, plus one. The compiler or linker
1646 * determines the highest index used. There will be only one
1647 * active uniform reported by the GL per uniform array.
1651 update_array_sizes(struct gl_shader_program
*prog
)
1653 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1654 if (prog
->_LinkedShaders
[i
] == NULL
)
1657 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1658 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1660 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1661 !var
->type
->is_array())
1664 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1665 * will not be eliminated. Since we always do std140, just
1666 * don't resize arrays in UBOs.
1668 * Atomic counters are supposed to get deterministic
1669 * locations assigned based on the declaration ordering and
1670 * sizes, array compaction would mess that up.
1672 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1675 unsigned int size
= var
->data
.max_array_access
;
1676 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1677 if (prog
->_LinkedShaders
[j
] == NULL
)
1680 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1681 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1685 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1686 other_var
->data
.max_array_access
> size
) {
1687 size
= other_var
->data
.max_array_access
;
1692 if (size
+ 1 != var
->type
->length
) {
1693 /* If this is a built-in uniform (i.e., it's backed by some
1694 * fixed-function state), adjust the number of state slots to
1695 * match the new array size. The number of slots per array entry
1696 * is not known. It seems safe to assume that the total number of
1697 * slots is an integer multiple of the number of array elements.
1698 * Determine the number of slots per array element by dividing by
1699 * the old (total) size.
1701 if (var
->num_state_slots
> 0) {
1702 var
->num_state_slots
= (size
+ 1)
1703 * (var
->num_state_slots
/ var
->type
->length
);
1706 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1708 /* FINISHME: We should update the types of array
1709 * dereferences of this variable now.
1717 * Find a contiguous set of available bits in a bitmask.
1719 * \param used_mask Bits representing used (1) and unused (0) locations
1720 * \param needed_count Number of contiguous bits needed.
1723 * Base location of the available bits on success or -1 on failure.
1726 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1728 unsigned needed_mask
= (1 << needed_count
) - 1;
1729 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1731 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1732 * cannot optimize possibly infinite loops" for the loop below.
1734 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1737 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1738 if ((needed_mask
& ~used_mask
) == needed_mask
)
1749 * Assign locations for either VS inputs for FS outputs
1751 * \param prog Shader program whose variables need locations assigned
1752 * \param target_index Selector for the program target to receive location
1753 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1754 * \c MESA_SHADER_FRAGMENT.
1755 * \param max_index Maximum number of generic locations. This corresponds
1756 * to either the maximum number of draw buffers or the
1757 * maximum number of generic attributes.
1760 * If locations are successfully assigned, true is returned. Otherwise an
1761 * error is emitted to the shader link log and false is returned.
1764 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1765 unsigned target_index
,
1768 /* Mark invalid locations as being used.
1770 unsigned used_locations
= (max_index
>= 32)
1771 ? ~0 : ~((1 << max_index
) - 1);
1773 assert((target_index
== MESA_SHADER_VERTEX
)
1774 || (target_index
== MESA_SHADER_FRAGMENT
));
1776 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1780 /* Operate in a total of four passes.
1782 * 1. Invalidate the location assignments for all vertex shader inputs.
1784 * 2. Assign locations for inputs that have user-defined (via
1785 * glBindVertexAttribLocation) locations and outputs that have
1786 * user-defined locations (via glBindFragDataLocation).
1788 * 3. Sort the attributes without assigned locations by number of slots
1789 * required in decreasing order. Fragmentation caused by attribute
1790 * locations assigned by the application may prevent large attributes
1791 * from having enough contiguous space.
1793 * 4. Assign locations to any inputs without assigned locations.
1796 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1797 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1799 const enum ir_variable_mode direction
=
1800 (target_index
== MESA_SHADER_VERTEX
)
1801 ? ir_var_shader_in
: ir_var_shader_out
;
1804 /* Temporary storage for the set of attributes that need locations assigned.
1810 /* Used below in the call to qsort. */
1811 static int compare(const void *a
, const void *b
)
1813 const temp_attr
*const l
= (const temp_attr
*) a
;
1814 const temp_attr
*const r
= (const temp_attr
*) b
;
1816 /* Reversed because we want a descending order sort below. */
1817 return r
->slots
- l
->slots
;
1821 unsigned num_attr
= 0;
1823 foreach_list(node
, sh
->ir
) {
1824 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1826 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
1829 if (var
->data
.explicit_location
) {
1830 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
1831 || (var
->data
.location
< 0)) {
1833 "invalid explicit location %d specified for `%s'\n",
1834 (var
->data
.location
< 0)
1835 ? var
->data
.location
1836 : var
->data
.location
- generic_base
,
1840 } else if (target_index
== MESA_SHADER_VERTEX
) {
1843 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1844 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1845 var
->data
.location
= binding
;
1846 var
->data
.is_unmatched_generic_inout
= 0;
1848 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1852 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1853 assert(binding
>= FRAG_RESULT_DATA0
);
1854 var
->data
.location
= binding
;
1855 var
->data
.is_unmatched_generic_inout
= 0;
1857 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1858 var
->data
.index
= index
;
1863 /* If the variable is not a built-in and has a location statically
1864 * assigned in the shader (presumably via a layout qualifier), make sure
1865 * that it doesn't collide with other assigned locations. Otherwise,
1866 * add it to the list of variables that need linker-assigned locations.
1868 const unsigned slots
= var
->type
->count_attribute_slots();
1869 if (var
->data
.location
!= -1) {
1870 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
1871 /* From page 61 of the OpenGL 4.0 spec:
1873 * "LinkProgram will fail if the attribute bindings assigned
1874 * by BindAttribLocation do not leave not enough space to
1875 * assign a location for an active matrix attribute or an
1876 * active attribute array, both of which require multiple
1877 * contiguous generic attributes."
1879 * I think above text prohibits the aliasing of explicit and
1880 * automatic assignments. But, aliasing is allowed in manual
1881 * assignments of attribute locations. See below comments for
1884 * From OpenGL 4.0 spec, page 61:
1886 * "It is possible for an application to bind more than one
1887 * attribute name to the same location. This is referred to as
1888 * aliasing. This will only work if only one of the aliased
1889 * attributes is active in the executable program, or if no
1890 * path through the shader consumes more than one attribute of
1891 * a set of attributes aliased to the same location. A link
1892 * error can occur if the linker determines that every path
1893 * through the shader consumes multiple aliased attributes,
1894 * but implementations are not required to generate an error
1897 * From GLSL 4.30 spec, page 54:
1899 * "A program will fail to link if any two non-vertex shader
1900 * input variables are assigned to the same location. For
1901 * vertex shaders, multiple input variables may be assigned
1902 * to the same location using either layout qualifiers or via
1903 * the OpenGL API. However, such aliasing is intended only to
1904 * support vertex shaders where each execution path accesses
1905 * at most one input per each location. Implementations are
1906 * permitted, but not required, to generate link-time errors
1907 * if they detect that every path through the vertex shader
1908 * executable accesses multiple inputs assigned to any single
1909 * location. For all shader types, a program will fail to link
1910 * if explicit location assignments leave the linker unable
1911 * to find space for other variables without explicit
1914 * From OpenGL ES 3.0 spec, page 56:
1916 * "Binding more than one attribute name to the same location
1917 * is referred to as aliasing, and is not permitted in OpenGL
1918 * ES Shading Language 3.00 vertex shaders. LinkProgram will
1919 * fail when this condition exists. However, aliasing is
1920 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
1921 * This will only work if only one of the aliased attributes
1922 * is active in the executable program, or if no path through
1923 * the shader consumes more than one attribute of a set of
1924 * attributes aliased to the same location. A link error can
1925 * occur if the linker determines that every path through the
1926 * shader consumes multiple aliased attributes, but implemen-
1927 * tations are not required to generate an error in this case."
1929 * After looking at above references from OpenGL, OpenGL ES and
1930 * GLSL specifications, we allow aliasing of vertex input variables
1931 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
1933 * NOTE: This is not required by the spec but its worth mentioning
1934 * here that we're not doing anything to make sure that no path
1935 * through the vertex shader executable accesses multiple inputs
1936 * assigned to any single location.
1939 /* Mask representing the contiguous slots that will be used by
1942 const unsigned attr
= var
->data
.location
- generic_base
;
1943 const unsigned use_mask
= (1 << slots
) - 1;
1944 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1945 ? "vertex shader input" : "fragment shader output";
1947 /* Generate a link error if the requested locations for this
1948 * attribute exceed the maximum allowed attribute location.
1950 if (attr
+ slots
> max_index
) {
1952 "insufficient contiguous locations "
1953 "available for %s `%s' %d %d %d", string
,
1954 var
->name
, used_locations
, use_mask
, attr
);
1958 /* Generate a link error if the set of bits requested for this
1959 * attribute overlaps any previously allocated bits.
1961 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1962 if (target_index
== MESA_SHADER_FRAGMENT
||
1963 (prog
->IsES
&& prog
->Version
>= 300)) {
1965 "overlapping location is assigned "
1966 "to %s `%s' %d %d %d\n", string
,
1967 var
->name
, used_locations
, use_mask
, attr
);
1970 linker_warning(prog
,
1971 "overlapping location is assigned "
1972 "to %s `%s' %d %d %d\n", string
,
1973 var
->name
, used_locations
, use_mask
, attr
);
1977 used_locations
|= (use_mask
<< attr
);
1983 to_assign
[num_attr
].slots
= slots
;
1984 to_assign
[num_attr
].var
= var
;
1988 /* If all of the attributes were assigned locations by the application (or
1989 * are built-in attributes with fixed locations), return early. This should
1990 * be the common case.
1995 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1997 if (target_index
== MESA_SHADER_VERTEX
) {
1998 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1999 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2000 * reserved to prevent it from being automatically allocated below.
2002 find_deref_visitor
find("gl_Vertex");
2004 if (find
.variable_found())
2005 used_locations
|= (1 << 0);
2008 for (unsigned i
= 0; i
< num_attr
; i
++) {
2009 /* Mask representing the contiguous slots that will be used by this
2012 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2014 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2017 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2018 ? "vertex shader input" : "fragment shader output";
2021 "insufficient contiguous locations "
2022 "available for %s `%s'",
2023 string
, to_assign
[i
].var
->name
);
2027 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2028 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2029 used_locations
|= (use_mask
<< location
);
2037 * Demote shader inputs and outputs that are not used in other stages
2040 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2042 foreach_list(node
, sh
->ir
) {
2043 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
2045 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2048 /* A shader 'in' or 'out' variable is only really an input or output if
2049 * its value is used by other shader stages. This will cause the variable
2050 * to have a location assigned.
2052 if (var
->data
.is_unmatched_generic_inout
) {
2053 var
->data
.mode
= ir_var_auto
;
2060 * Store the gl_FragDepth layout in the gl_shader_program struct.
2063 store_fragdepth_layout(struct gl_shader_program
*prog
)
2065 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2069 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2071 /* We don't look up the gl_FragDepth symbol directly because if
2072 * gl_FragDepth is not used in the shader, it's removed from the IR.
2073 * However, the symbol won't be removed from the symbol table.
2075 * We're only interested in the cases where the variable is NOT removed
2078 foreach_list(node
, ir
) {
2079 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
2081 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2085 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2086 switch (var
->data
.depth_layout
) {
2087 case ir_depth_layout_none
:
2088 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2090 case ir_depth_layout_any
:
2091 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2093 case ir_depth_layout_greater
:
2094 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2096 case ir_depth_layout_less
:
2097 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2099 case ir_depth_layout_unchanged
:
2100 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2111 * Validate the resources used by a program versus the implementation limits
2114 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2116 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2117 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2122 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2123 linker_error(prog
, "Too many %s shader texture samplers",
2124 _mesa_shader_stage_to_string(i
));
2127 if (sh
->num_uniform_components
>
2128 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2129 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2130 linker_warning(prog
, "Too many %s shader default uniform block "
2131 "components, but the driver will try to optimize "
2132 "them out; this is non-portable out-of-spec "
2134 _mesa_shader_stage_to_string(i
));
2136 linker_error(prog
, "Too many %s shader default uniform block "
2138 _mesa_shader_stage_to_string(i
));
2142 if (sh
->num_combined_uniform_components
>
2143 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2144 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2145 linker_warning(prog
, "Too many %s shader uniform components, "
2146 "but the driver will try to optimize them out; "
2147 "this is non-portable out-of-spec behavior\n",
2148 _mesa_shader_stage_to_string(i
));
2150 linker_error(prog
, "Too many %s shader uniform components",
2151 _mesa_shader_stage_to_string(i
));
2156 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2157 unsigned total_uniform_blocks
= 0;
2159 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2160 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2161 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2163 total_uniform_blocks
++;
2167 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2168 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
2169 prog
->NumUniformBlocks
,
2170 ctx
->Const
.MaxCombinedUniformBlocks
);
2172 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2173 const unsigned max_uniform_blocks
=
2174 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2175 if (blocks
[i
] > max_uniform_blocks
) {
2176 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
2177 _mesa_shader_stage_to_string(i
),
2179 max_uniform_blocks
);
2188 * Validate shader image resources.
2191 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2193 unsigned total_image_units
= 0;
2194 unsigned fragment_outputs
= 0;
2196 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2199 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2200 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2203 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2204 linker_error(prog
, "Too many %s shader image uniforms",
2205 _mesa_shader_stage_to_string(i
));
2207 total_image_units
+= sh
->NumImages
;
2209 if (i
== MESA_SHADER_FRAGMENT
) {
2210 foreach_list(node
, sh
->ir
) {
2211 ir_variable
*var
= ((ir_instruction
*)node
)->as_variable();
2212 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2213 fragment_outputs
+= var
->type
->count_attribute_slots();
2219 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2220 linker_error(prog
, "Too many combined image uniforms");
2222 if (total_image_units
+ fragment_outputs
>
2223 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2224 linker_error(prog
, "Too many combined image uniforms and fragment outputs");
2228 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2230 tfeedback_decl
*tfeedback_decls
= NULL
;
2231 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2233 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2235 prog
->LinkStatus
= true; /* All error paths will set this to false */
2236 prog
->Validated
= false;
2237 prog
->_Used
= false;
2239 ralloc_free(prog
->InfoLog
);
2240 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2242 ralloc_free(prog
->UniformBlocks
);
2243 prog
->UniformBlocks
= NULL
;
2244 prog
->NumUniformBlocks
= 0;
2245 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2246 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2247 prog
->UniformBlockStageIndex
[i
] = NULL
;
2250 ralloc_free(prog
->AtomicBuffers
);
2251 prog
->AtomicBuffers
= NULL
;
2252 prog
->NumAtomicBuffers
= 0;
2253 prog
->ARB_fragment_coord_conventions_enable
= false;
2255 /* Separate the shaders into groups based on their type.
2257 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2258 unsigned num_shaders
[MESA_SHADER_STAGES
];
2260 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2261 shader_list
[i
] = (struct gl_shader
**)
2262 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2266 unsigned min_version
= UINT_MAX
;
2267 unsigned max_version
= 0;
2268 const bool is_es_prog
=
2269 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2270 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2271 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2272 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2274 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2275 linker_error(prog
, "all shaders must use same shading "
2276 "language version\n");
2280 prog
->ARB_fragment_coord_conventions_enable
|=
2281 prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
;
2283 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2284 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2285 num_shaders
[shader_type
]++;
2288 /* In desktop GLSL, different shader versions may be linked together. In
2289 * GLSL ES, all shader versions must be the same.
2291 if (is_es_prog
&& min_version
!= max_version
) {
2292 linker_error(prog
, "all shaders must use same shading "
2293 "language version\n");
2297 prog
->Version
= max_version
;
2298 prog
->IsES
= is_es_prog
;
2300 /* Geometry shaders have to be linked with vertex shaders.
2302 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2303 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2304 !prog
->SeparateShader
) {
2305 linker_error(prog
, "Geometry shader must be linked with "
2310 /* Compute shaders have additional restrictions. */
2311 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2312 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2313 linker_error(prog
, "Compute shaders may not be linked with any other "
2314 "type of shader\n");
2317 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2318 if (prog
->_LinkedShaders
[i
] != NULL
)
2319 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2321 prog
->_LinkedShaders
[i
] = NULL
;
2324 /* Link all shaders for a particular stage and validate the result.
2326 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2327 if (num_shaders
[stage
] > 0) {
2328 gl_shader
*const sh
=
2329 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2330 num_shaders
[stage
]);
2332 if (!prog
->LinkStatus
)
2336 case MESA_SHADER_VERTEX
:
2337 validate_vertex_shader_executable(prog
, sh
);
2339 case MESA_SHADER_GEOMETRY
:
2340 validate_geometry_shader_executable(prog
, sh
);
2342 case MESA_SHADER_FRAGMENT
:
2343 validate_fragment_shader_executable(prog
, sh
);
2346 if (!prog
->LinkStatus
)
2349 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2353 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2354 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2355 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2356 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2358 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2360 /* Here begins the inter-stage linking phase. Some initial validation is
2361 * performed, then locations are assigned for uniforms, attributes, and
2364 cross_validate_uniforms(prog
);
2365 if (!prog
->LinkStatus
)
2370 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2371 if (prog
->_LinkedShaders
[prev
] != NULL
)
2375 /* Validate the inputs of each stage with the output of the preceding
2378 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2379 if (prog
->_LinkedShaders
[i
] == NULL
)
2382 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2383 prog
->_LinkedShaders
[i
]);
2384 if (!prog
->LinkStatus
)
2387 cross_validate_outputs_to_inputs(prog
,
2388 prog
->_LinkedShaders
[prev
],
2389 prog
->_LinkedShaders
[i
]);
2390 if (!prog
->LinkStatus
)
2396 /* Cross-validate uniform blocks between shader stages */
2397 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
2398 MESA_SHADER_STAGES
);
2399 if (!prog
->LinkStatus
)
2402 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2403 if (prog
->_LinkedShaders
[i
] != NULL
)
2404 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2407 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2408 * it before optimization because we want most of the checks to get
2409 * dropped thanks to constant propagation.
2411 * This rule also applies to GLSL ES 3.00.
2413 if (max_version
>= (is_es_prog
? 300 : 130)) {
2414 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2416 lower_discard_flow(sh
->ir
);
2420 if (!interstage_cross_validate_uniform_blocks(prog
))
2423 /* Do common optimization before assigning storage for attributes,
2424 * uniforms, and varyings. Later optimization could possibly make
2425 * some of that unused.
2427 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2428 if (prog
->_LinkedShaders
[i
] == NULL
)
2431 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2432 if (!prog
->LinkStatus
)
2435 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
) {
2436 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2439 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
2440 &ctx
->ShaderCompilerOptions
[i
],
2441 ctx
->Const
.NativeIntegers
))
2445 /* Mark all generic shader inputs and outputs as unpaired. */
2446 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2447 if (prog
->_LinkedShaders
[i
] != NULL
) {
2448 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
2452 /* FINISHME: The value of the max_attribute_index parameter is
2453 * FINISHME: implementation dependent based on the value of
2454 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2455 * FINISHME: at least 16, so hardcode 16 for now.
2457 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2461 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2466 for (first
= 0; first
<= MESA_SHADER_FRAGMENT
; first
++) {
2467 if (prog
->_LinkedShaders
[first
] != NULL
)
2471 if (num_tfeedback_decls
!= 0) {
2472 /* From GL_EXT_transform_feedback:
2473 * A program will fail to link if:
2475 * * the <count> specified by TransformFeedbackVaryingsEXT is
2476 * non-zero, but the program object has no vertex or geometry
2479 if (first
== MESA_SHADER_FRAGMENT
) {
2480 linker_error(prog
, "Transform feedback varyings specified, but "
2481 "no vertex or geometry shader is present.");
2485 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2486 prog
->TransformFeedback
.NumVarying
);
2487 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2488 prog
->TransformFeedback
.VaryingNames
,
2493 /* Linking the stages in the opposite order (from fragment to vertex)
2494 * ensures that inter-shader outputs written to in an earlier stage are
2495 * eliminated if they are (transitively) not used in a later stage.
2498 for (last
= MESA_SHADER_FRAGMENT
; last
>= 0; last
--) {
2499 if (prog
->_LinkedShaders
[last
] != NULL
)
2503 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2504 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2506 if (num_tfeedback_decls
!= 0 || prog
->SeparateShader
) {
2507 /* There was no fragment shader, but we still have to assign varying
2508 * locations for use by transform feedback.
2510 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2512 num_tfeedback_decls
, tfeedback_decls
,
2517 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2518 num_tfeedback_decls
, tfeedback_decls
);
2520 if (!prog
->SeparateShader
)
2521 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2523 /* Eliminate code that is now dead due to unused outputs being demoted.
2525 while (do_dead_code(sh
->ir
, false))
2528 else if (first
== MESA_SHADER_FRAGMENT
) {
2529 /* If the program only contains a fragment shader...
2531 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2533 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2534 num_tfeedback_decls
, tfeedback_decls
);
2536 if (prog
->SeparateShader
) {
2537 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2538 NULL
/* producer */,
2540 0 /* num_tfeedback_decls */,
2541 NULL
/* tfeedback_decls */,
2542 0 /* gs_input_vertices */))
2545 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2547 while (do_dead_code(sh
->ir
, false))
2552 for (int i
= next
- 1; i
>= 0; i
--) {
2553 if (prog
->_LinkedShaders
[i
] == NULL
)
2556 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2557 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2558 unsigned gs_input_vertices
=
2559 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2561 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2562 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2563 tfeedback_decls
, gs_input_vertices
))
2566 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2567 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2570 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2571 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2573 /* Eliminate code that is now dead due to unused outputs being demoted.
2575 while (do_dead_code(sh_i
->ir
, false))
2577 while (do_dead_code(sh_next
->ir
, false))
2580 /* This must be done after all dead varyings are eliminated. */
2581 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2583 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2589 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2592 update_array_sizes(prog
);
2593 link_assign_uniform_locations(prog
);
2594 link_assign_atomic_counter_resources(ctx
, prog
);
2595 store_fragdepth_layout(prog
);
2597 check_resources(ctx
, prog
);
2598 check_image_resources(ctx
, prog
);
2599 link_check_atomic_counter_resources(ctx
, prog
);
2601 if (!prog
->LinkStatus
)
2604 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2605 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
2606 * anything about shader linking when one of the shaders (vertex or
2607 * fragment shader) is absent. So, the extension shouldn't change the
2608 * behavior specified in GLSL specification.
2610 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
2611 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2612 linker_error(prog
, "program lacks a vertex shader\n");
2613 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2614 linker_error(prog
, "program lacks a fragment shader\n");
2618 /* FINISHME: Assign fragment shader output locations. */
2621 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2622 free(shader_list
[i
]);
2623 if (prog
->_LinkedShaders
[i
] == NULL
)
2626 /* Do a final validation step to make sure that the IR wasn't
2627 * invalidated by any modifications performed after intrastage linking.
2629 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
2631 /* Retain any live IR, but trash the rest. */
2632 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2634 /* The symbol table in the linked shaders may contain references to
2635 * variables that were removed (e.g., unused uniforms). Since it may
2636 * contain junk, there is no possible valid use. Delete it and set the
2639 delete prog
->_LinkedShaders
[i
]->symbols
;
2640 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2643 ralloc_free(mem_ctx
);