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>
68 #include "main/core.h"
69 #include "glsl_symbol_table.h"
70 #include "glsl_parser_extras.h"
73 #include "program/hash_table.h"
75 #include "link_varyings.h"
76 #include "ir_optimization.h"
77 #include "ir_rvalue_visitor.h"
78 #include "ir_uniform.h"
80 #include "main/shaderobj.h"
81 #include "main/enums.h"
84 void linker_error(gl_shader_program
*, const char *, ...);
89 * Visitor that determines whether or not a variable is ever written.
91 class find_assignment_visitor
: public ir_hierarchical_visitor
{
93 find_assignment_visitor(const char *name
)
94 : name(name
), found(false)
99 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
101 ir_variable
*const var
= ir
->lhs
->variable_referenced();
103 if (strcmp(name
, var
->name
) == 0) {
108 return visit_continue_with_parent
;
111 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
113 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
114 actual_node
, &ir
->actual_parameters
) {
115 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
116 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
118 if (sig_param
->data
.mode
== ir_var_function_out
||
119 sig_param
->data
.mode
== ir_var_function_inout
) {
120 ir_variable
*var
= param_rval
->variable_referenced();
121 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
->data
.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
->data
.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
->data
.max_array_access
, var
->name
, this->num_vertices
);
224 return visit_continue
;
227 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
229 var
->data
.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
->fields
.array
;
249 return visit_continue
;
254 * Visitor that determines the highest stream id to which a (geometry) shader
255 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
257 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
259 find_emit_vertex_visitor(int max_allowed
)
260 : max_stream_allowed(max_allowed
),
261 invalid_stream_id(0),
262 invalid_stream_id_from_emit_vertex(false),
263 end_primitive_found(false),
264 uses_non_zero_stream(false)
269 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
271 int stream_id
= ir
->stream_id();
274 invalid_stream_id
= stream_id
;
275 invalid_stream_id_from_emit_vertex
= true;
279 if (stream_id
> max_stream_allowed
) {
280 invalid_stream_id
= stream_id
;
281 invalid_stream_id_from_emit_vertex
= true;
286 uses_non_zero_stream
= true;
288 return visit_continue
;
291 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
293 end_primitive_found
= true;
295 int stream_id
= ir
->stream_id();
298 invalid_stream_id
= stream_id
;
299 invalid_stream_id_from_emit_vertex
= false;
303 if (stream_id
> max_stream_allowed
) {
304 invalid_stream_id
= stream_id
;
305 invalid_stream_id_from_emit_vertex
= false;
310 uses_non_zero_stream
= true;
312 return visit_continue
;
317 return invalid_stream_id
!= 0;
320 const char *error_func()
322 return invalid_stream_id_from_emit_vertex
?
323 "EmitStreamVertex" : "EndStreamPrimitive";
328 return invalid_stream_id
;
333 return uses_non_zero_stream
;
336 bool uses_end_primitive()
338 return end_primitive_found
;
342 int max_stream_allowed
;
343 int invalid_stream_id
;
344 bool invalid_stream_id_from_emit_vertex
;
345 bool end_primitive_found
;
346 bool uses_non_zero_stream
;
349 /* Class that finds array derefs and check if indexes are dynamic. */
350 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
353 dynamic_sampler_array_indexing_visitor() :
354 dynamic_sampler_array_indexing(false)
358 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
360 if (!ir
->variable_referenced())
361 return visit_continue
;
363 if (!ir
->variable_referenced()->type
->contains_sampler())
364 return visit_continue
;
366 if (!ir
->array_index
->constant_expression_value()) {
367 dynamic_sampler_array_indexing
= true;
370 return visit_continue
;
373 bool uses_dynamic_sampler_array_indexing()
375 return dynamic_sampler_array_indexing
;
379 bool dynamic_sampler_array_indexing
;
382 } /* anonymous namespace */
385 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
389 ralloc_strcat(&prog
->InfoLog
, "error: ");
391 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
394 prog
->LinkStatus
= false;
399 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
403 ralloc_strcat(&prog
->InfoLog
, "warning: ");
405 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
412 * Given a string identifying a program resource, break it into a base name
413 * and an optional array index in square brackets.
415 * If an array index is present, \c out_base_name_end is set to point to the
416 * "[" that precedes the array index, and the array index itself is returned
419 * If no array index is present (or if the array index is negative or
420 * mal-formed), \c out_base_name_end, is set to point to the null terminator
421 * at the end of the input string, and -1 is returned.
423 * Only the final array index is parsed; if the string contains other array
424 * indices (or structure field accesses), they are left in the base name.
426 * No attempt is made to check that the base name is properly formed;
427 * typically the caller will look up the base name in a hash table, so
428 * ill-formed base names simply turn into hash table lookup failures.
431 parse_program_resource_name(const GLchar
*name
,
432 const GLchar
**out_base_name_end
)
434 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
436 * "When an integer array element or block instance number is part of
437 * the name string, it will be specified in decimal form without a "+"
438 * or "-" sign or any extra leading zeroes. Additionally, the name
439 * string will not include white space anywhere in the string."
442 const size_t len
= strlen(name
);
443 *out_base_name_end
= name
+ len
;
445 if (len
== 0 || name
[len
-1] != ']')
448 /* Walk backwards over the string looking for a non-digit character. This
449 * had better be the opening bracket for an array index.
451 * Initially, i specifies the location of the ']'. Since the string may
452 * contain only the ']' charcater, walk backwards very carefully.
455 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
458 if ((i
== 0) || name
[i
-1] != '[')
461 long array_index
= strtol(&name
[i
], NULL
, 10);
465 *out_base_name_end
= name
+ (i
- 1);
471 link_invalidate_variable_locations(exec_list
*ir
)
473 foreach_in_list(ir_instruction
, node
, ir
) {
474 ir_variable
*const var
= node
->as_variable();
479 /* Only assign locations for variables that lack an explicit location.
480 * Explicit locations are set for all built-in variables, generic vertex
481 * shader inputs (via layout(location=...)), and generic fragment shader
482 * outputs (also via layout(location=...)).
484 if (!var
->data
.explicit_location
) {
485 var
->data
.location
= -1;
486 var
->data
.location_frac
= 0;
489 /* ir_variable::is_unmatched_generic_inout is used by the linker while
490 * connecting outputs from one stage to inputs of the next stage.
492 * There are two implicit assumptions here. First, we assume that any
493 * built-in variable (i.e., non-generic in or out) will have
494 * explicit_location set. Second, we assume that any generic in or out
495 * will not have explicit_location set.
497 * This second assumption will only be valid until
498 * GL_ARB_separate_shader_objects is supported. When that extension is
499 * implemented, this function will need some modifications.
501 if (!var
->data
.explicit_location
) {
502 var
->data
.is_unmatched_generic_inout
= 1;
504 var
->data
.is_unmatched_generic_inout
= 0;
511 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
513 * Also check for errors based on incorrect usage of gl_ClipVertex and
516 * Return false if an error was reported.
519 analyze_clip_usage(struct gl_shader_program
*prog
,
520 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
521 GLuint
*ClipDistanceArraySize
)
523 *ClipDistanceArraySize
= 0;
525 if (!prog
->IsES
&& prog
->Version
>= 130) {
526 /* From section 7.1 (Vertex Shader Special Variables) of the
529 * "It is an error for a shader to statically write both
530 * gl_ClipVertex and gl_ClipDistance."
532 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
533 * gl_ClipVertex nor gl_ClipDistance.
535 find_assignment_visitor
clip_vertex("gl_ClipVertex");
536 find_assignment_visitor
clip_distance("gl_ClipDistance");
538 clip_vertex
.run(shader
->ir
);
539 clip_distance
.run(shader
->ir
);
540 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
541 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
542 "and `gl_ClipDistance'\n",
543 _mesa_shader_stage_to_string(shader
->Stage
));
546 *UsesClipDistance
= clip_distance
.variable_found();
547 ir_variable
*clip_distance_var
=
548 shader
->symbols
->get_variable("gl_ClipDistance");
549 if (clip_distance_var
)
550 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
552 *UsesClipDistance
= false;
558 * Verify that a vertex shader executable meets all semantic requirements.
560 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
563 * \param shader Vertex shader executable to be verified
566 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
567 struct gl_shader
*shader
)
572 /* From the GLSL 1.10 spec, page 48:
574 * "The variable gl_Position is available only in the vertex
575 * language and is intended for writing the homogeneous vertex
576 * position. All executions of a well-formed vertex shader
577 * executable must write a value into this variable. [...] The
578 * variable gl_Position is available only in the vertex
579 * language and is intended for writing the homogeneous vertex
580 * position. All executions of a well-formed vertex shader
581 * executable must write a value into this variable."
583 * while in GLSL 1.40 this text is changed to:
585 * "The variable gl_Position is available only in the vertex
586 * language and is intended for writing the homogeneous vertex
587 * position. It can be written at any time during shader
588 * execution. It may also be read back by a vertex shader
589 * after being written. This value will be used by primitive
590 * assembly, clipping, culling, and other fixed functionality
591 * operations, if present, that operate on primitives after
592 * vertex processing has occurred. Its value is undefined if
593 * the vertex shader executable does not write gl_Position."
595 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
596 * gl_Position is not an error.
598 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
599 find_assignment_visitor
find("gl_Position");
600 find
.run(shader
->ir
);
601 if (!find
.variable_found()) {
604 "vertex shader does not write to `gl_Position'."
605 "It's value is undefined. \n");
608 "vertex shader does not write to `gl_Position'. \n");
614 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
615 &prog
->Vert
.ClipDistanceArraySize
);
620 * Verify that a fragment shader executable meets all semantic requirements
622 * \param shader Fragment shader executable to be verified
625 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
626 struct gl_shader
*shader
)
631 find_assignment_visitor
frag_color("gl_FragColor");
632 find_assignment_visitor
frag_data("gl_FragData");
634 frag_color
.run(shader
->ir
);
635 frag_data
.run(shader
->ir
);
637 if (frag_color
.variable_found() && frag_data
.variable_found()) {
638 linker_error(prog
, "fragment shader writes to both "
639 "`gl_FragColor' and `gl_FragData'\n");
644 * Verify that a geometry shader executable meets all semantic requirements
646 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
647 * prog->Geom.ClipDistanceArraySize as a side effect.
649 * \param shader Geometry shader executable to be verified
652 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
653 struct gl_shader
*shader
)
658 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
659 prog
->Geom
.VerticesIn
= num_vertices
;
661 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
662 &prog
->Geom
.ClipDistanceArraySize
);
666 * Check if geometry shaders emit to non-zero streams and do corresponding
670 validate_geometry_shader_emissions(struct gl_context
*ctx
,
671 struct gl_shader_program
*prog
)
673 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
674 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
675 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
676 if (emit_vertex
.error()) {
677 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
678 "stream parameter are in the range [0, %d].\n",
679 emit_vertex
.error_func(),
680 emit_vertex
.error_stream(),
681 ctx
->Const
.MaxVertexStreams
- 1);
683 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
684 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
686 /* From the ARB_gpu_shader5 spec:
688 * "Multiple vertex streams are supported only if the output primitive
689 * type is declared to be "points". A program will fail to link if it
690 * contains a geometry shader calling EmitStreamVertex() or
691 * EndStreamPrimitive() if its output primitive type is not "points".
693 * However, in the same spec:
695 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
696 * with <stream> set to zero."
700 * "The function EndPrimitive() is equivalent to calling
701 * EndStreamPrimitive() with <stream> set to zero."
703 * Since we can call EmitVertex() and EndPrimitive() when we output
704 * primitives other than points, calling EmitStreamVertex(0) or
705 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
706 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
707 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
710 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
711 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
712 "with n>0 requires point output\n");
718 validate_intrastage_arrays(struct gl_shader_program
*prog
,
719 ir_variable
*const var
,
720 ir_variable
*const existing
)
722 /* Consider the types to be "the same" if both types are arrays
723 * of the same type and one of the arrays is implicitly sized.
724 * In addition, set the type of the linked variable to the
725 * explicitly sized array.
727 if (var
->type
->is_array() && existing
->type
->is_array() &&
728 (var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
729 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
730 if (var
->type
->length
!= 0) {
731 if (var
->type
->length
<= existing
->data
.max_array_access
) {
732 linker_error(prog
, "%s `%s' declared as type "
733 "`%s' but outermost dimension has an index"
736 var
->name
, var
->type
->name
,
737 existing
->data
.max_array_access
);
739 existing
->type
= var
->type
;
741 } else if (existing
->type
->length
!= 0) {
742 if(existing
->type
->length
<= var
->data
.max_array_access
) {
743 linker_error(prog
, "%s `%s' declared as type "
744 "`%s' but outermost dimension has an index"
747 var
->name
, existing
->type
->name
,
748 var
->data
.max_array_access
);
758 * Perform validation of global variables used across multiple shaders
761 cross_validate_globals(struct gl_shader_program
*prog
,
762 struct gl_shader
**shader_list
,
763 unsigned num_shaders
,
766 /* Examine all of the uniforms in all of the shaders and cross validate
769 glsl_symbol_table variables
;
770 for (unsigned i
= 0; i
< num_shaders
; i
++) {
771 if (shader_list
[i
] == NULL
)
774 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
775 ir_variable
*const var
= node
->as_variable();
780 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
))
783 /* Don't cross validate temporaries that are at global scope. These
784 * will eventually get pulled into the shaders 'main'.
786 if (var
->data
.mode
== ir_var_temporary
)
789 /* If a global with this name has already been seen, verify that the
790 * new instance has the same type. In addition, if the globals have
791 * initializers, the values of the initializers must be the same.
793 ir_variable
*const existing
= variables
.get_variable(var
->name
);
794 if (existing
!= NULL
) {
795 /* Check if types match. Interface blocks have some special
796 * rules so we handle those elsewhere.
798 if (var
->type
!= existing
->type
&&
799 !var
->is_interface_instance()) {
800 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
801 if (var
->type
->is_record() && existing
->type
->is_record()
802 && existing
->type
->record_compare(var
->type
)) {
803 existing
->type
= var
->type
;
805 linker_error(prog
, "%s `%s' declared as type "
806 "`%s' and type `%s'\n",
808 var
->name
, var
->type
->name
,
809 existing
->type
->name
);
815 if (var
->data
.explicit_location
) {
816 if (existing
->data
.explicit_location
817 && (var
->data
.location
!= existing
->data
.location
)) {
818 linker_error(prog
, "explicit locations for %s "
819 "`%s' have differing values\n",
820 mode_string(var
), var
->name
);
824 existing
->data
.location
= var
->data
.location
;
825 existing
->data
.explicit_location
= true;
828 /* From the GLSL 4.20 specification:
829 * "A link error will result if two compilation units in a program
830 * specify different integer-constant bindings for the same
831 * opaque-uniform name. However, it is not an error to specify a
832 * binding on some but not all declarations for the same name"
834 if (var
->data
.explicit_binding
) {
835 if (existing
->data
.explicit_binding
&&
836 var
->data
.binding
!= existing
->data
.binding
) {
837 linker_error(prog
, "explicit bindings for %s "
838 "`%s' have differing values\n",
839 mode_string(var
), var
->name
);
843 existing
->data
.binding
= var
->data
.binding
;
844 existing
->data
.explicit_binding
= true;
847 if (var
->type
->contains_atomic() &&
848 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
849 linker_error(prog
, "offset specifications for %s "
850 "`%s' have differing values\n",
851 mode_string(var
), var
->name
);
855 /* Validate layout qualifiers for gl_FragDepth.
857 * From the AMD/ARB_conservative_depth specs:
859 * "If gl_FragDepth is redeclared in any fragment shader in a
860 * program, it must be redeclared in all fragment shaders in
861 * that program that have static assignments to
862 * gl_FragDepth. All redeclarations of gl_FragDepth in all
863 * fragment shaders in a single program must have the same set
866 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
867 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
868 bool layout_differs
=
869 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
871 if (layout_declared
&& layout_differs
) {
873 "All redeclarations of gl_FragDepth in all "
874 "fragment shaders in a single program must have "
875 "the same set of qualifiers.\n");
878 if (var
->data
.used
&& layout_differs
) {
880 "If gl_FragDepth is redeclared with a layout "
881 "qualifier in any fragment shader, it must be "
882 "redeclared with the same layout qualifier in "
883 "all fragment shaders that have assignments to "
888 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
890 * "If a shared global has multiple initializers, the
891 * initializers must all be constant expressions, and they
892 * must all have the same value. Otherwise, a link error will
893 * result. (A shared global having only one initializer does
894 * not require that initializer to be a constant expression.)"
896 * Previous to 4.20 the GLSL spec simply said that initializers
897 * must have the same value. In this case of non-constant
898 * initializers, this was impossible to determine. As a result,
899 * no vendor actually implemented that behavior. The 4.20
900 * behavior matches the implemented behavior of at least one other
901 * vendor, so we'll implement that for all GLSL versions.
903 if (var
->constant_initializer
!= NULL
) {
904 if (existing
->constant_initializer
!= NULL
) {
905 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
906 linker_error(prog
, "initializers for %s "
907 "`%s' have differing values\n",
908 mode_string(var
), var
->name
);
912 /* If the first-seen instance of a particular uniform did not
913 * have an initializer but a later instance does, copy the
914 * initializer to the version stored in the symbol table.
916 /* FINISHME: This is wrong. The constant_value field should
917 * FINISHME: not be modified! Imagine a case where a shader
918 * FINISHME: without an initializer is linked in two different
919 * FINISHME: programs with shaders that have differing
920 * FINISHME: initializers. Linking with the first will
921 * FINISHME: modify the shader, and linking with the second
922 * FINISHME: will fail.
924 existing
->constant_initializer
=
925 var
->constant_initializer
->clone(ralloc_parent(existing
),
930 if (var
->data
.has_initializer
) {
931 if (existing
->data
.has_initializer
932 && (var
->constant_initializer
== NULL
933 || existing
->constant_initializer
== NULL
)) {
935 "shared global variable `%s' has multiple "
936 "non-constant initializers.\n",
941 /* Some instance had an initializer, so keep track of that. In
942 * this location, all sorts of initializers (constant or
943 * otherwise) will propagate the existence to the variable
944 * stored in the symbol table.
946 existing
->data
.has_initializer
= true;
949 if (existing
->data
.invariant
!= var
->data
.invariant
) {
950 linker_error(prog
, "declarations for %s `%s' have "
951 "mismatching invariant qualifiers\n",
952 mode_string(var
), var
->name
);
955 if (existing
->data
.centroid
!= var
->data
.centroid
) {
956 linker_error(prog
, "declarations for %s `%s' have "
957 "mismatching centroid qualifiers\n",
958 mode_string(var
), var
->name
);
961 if (existing
->data
.sample
!= var
->data
.sample
) {
962 linker_error(prog
, "declarations for %s `%s` have "
963 "mismatching sample qualifiers\n",
964 mode_string(var
), var
->name
);
968 variables
.add_variable(var
);
975 * Perform validation of uniforms used across multiple shader stages
978 cross_validate_uniforms(struct gl_shader_program
*prog
)
980 cross_validate_globals(prog
, prog
->_LinkedShaders
,
981 MESA_SHADER_STAGES
, true);
985 * Accumulates the array of prog->UniformBlocks and checks that all
986 * definitons of blocks agree on their contents.
989 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
991 unsigned max_num_uniform_blocks
= 0;
992 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
993 if (prog
->_LinkedShaders
[i
])
994 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
997 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
998 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1000 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
1001 max_num_uniform_blocks
);
1002 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1003 prog
->UniformBlockStageIndex
[i
][j
] = -1;
1008 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
1009 int index
= link_cross_validate_uniform_block(prog
,
1010 &prog
->UniformBlocks
,
1011 &prog
->NumUniformBlocks
,
1012 &sh
->UniformBlocks
[j
]);
1015 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1016 sh
->UniformBlocks
[j
].Name
);
1020 prog
->UniformBlockStageIndex
[i
][index
] = j
;
1029 * Populates a shaders symbol table with all global declarations
1032 populate_symbol_table(gl_shader
*sh
)
1034 sh
->symbols
= new(sh
) glsl_symbol_table
;
1036 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1040 if ((func
= inst
->as_function()) != NULL
) {
1041 sh
->symbols
->add_function(func
);
1042 } else if ((var
= inst
->as_variable()) != NULL
) {
1043 if (var
->data
.mode
!= ir_var_temporary
)
1044 sh
->symbols
->add_variable(var
);
1051 * Remap variables referenced in an instruction tree
1053 * This is used when instruction trees are cloned from one shader and placed in
1054 * another. These trees will contain references to \c ir_variable nodes that
1055 * do not exist in the target shader. This function finds these \c ir_variable
1056 * references and replaces the references with matching variables in the target
1059 * If there is no matching variable in the target shader, a clone of the
1060 * \c ir_variable is made and added to the target shader. The new variable is
1061 * added to \b both the instruction stream and the symbol table.
1063 * \param inst IR tree that is to be processed.
1064 * \param symbols Symbol table containing global scope symbols in the
1066 * \param instructions Instruction stream where new variable declarations
1070 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1073 class remap_visitor
: public ir_hierarchical_visitor
{
1075 remap_visitor(struct gl_shader
*target
,
1078 this->target
= target
;
1079 this->symbols
= target
->symbols
;
1080 this->instructions
= target
->ir
;
1081 this->temps
= temps
;
1084 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1086 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1087 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1089 assert(var
!= NULL
);
1091 return visit_continue
;
1094 ir_variable
*const existing
=
1095 this->symbols
->get_variable(ir
->var
->name
);
1096 if (existing
!= NULL
)
1099 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1101 this->symbols
->add_variable(copy
);
1102 this->instructions
->push_head(copy
);
1106 return visit_continue
;
1110 struct gl_shader
*target
;
1111 glsl_symbol_table
*symbols
;
1112 exec_list
*instructions
;
1116 remap_visitor
v(target
, temps
);
1123 * Move non-declarations from one instruction stream to another
1125 * The intended usage pattern of this function is to pass the pointer to the
1126 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1127 * pointer) for \c last and \c false for \c make_copies on the first
1128 * call. Successive calls pass the return value of the previous call for
1129 * \c last and \c true for \c make_copies.
1131 * \param instructions Source instruction stream
1132 * \param last Instruction after which new instructions should be
1133 * inserted in the target instruction stream
1134 * \param make_copies Flag selecting whether instructions in \c instructions
1135 * should be copied (via \c ir_instruction::clone) into the
1136 * target list or moved.
1139 * The new "last" instruction in the target instruction stream. This pointer
1140 * is suitable for use as the \c last parameter of a later call to this
1144 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1145 bool make_copies
, gl_shader
*target
)
1147 hash_table
*temps
= NULL
;
1150 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1151 hash_table_pointer_compare
);
1153 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1154 if (inst
->as_function())
1157 ir_variable
*var
= inst
->as_variable();
1158 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1161 assert(inst
->as_assignment()
1163 || inst
->as_if() /* for initializers with the ?: operator */
1164 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1167 inst
= inst
->clone(target
, NULL
);
1170 hash_table_insert(temps
, inst
, var
);
1172 remap_variables(inst
, target
, temps
);
1177 last
->insert_after(inst
);
1182 hash_table_dtor(temps
);
1188 * Get the function signature for main from a shader
1190 ir_function_signature
*
1191 link_get_main_function_signature(gl_shader
*sh
)
1193 ir_function
*const f
= sh
->symbols
->get_function("main");
1195 exec_list void_parameters
;
1197 /* Look for the 'void main()' signature and ensure that it's defined.
1198 * This keeps the linker from accidentally pick a shader that just
1199 * contains a prototype for main.
1201 * We don't have to check for multiple definitions of main (in multiple
1202 * shaders) because that would have already been caught above.
1204 ir_function_signature
*sig
=
1205 f
->matching_signature(NULL
, &void_parameters
, false);
1206 if ((sig
!= NULL
) && sig
->is_defined
) {
1216 * This class is only used in link_intrastage_shaders() below but declaring
1217 * it inside that function leads to compiler warnings with some versions of
1220 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1222 array_sizing_visitor()
1223 : mem_ctx(ralloc_context(NULL
)),
1224 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1225 hash_table_pointer_compare
))
1229 ~array_sizing_visitor()
1231 hash_table_dtor(this->unnamed_interfaces
);
1232 ralloc_free(this->mem_ctx
);
1235 virtual ir_visitor_status
visit(ir_variable
*var
)
1237 fixup_type(&var
->type
, var
->data
.max_array_access
);
1238 if (var
->type
->is_interface()) {
1239 if (interface_contains_unsized_arrays(var
->type
)) {
1240 const glsl_type
*new_type
=
1241 resize_interface_members(var
->type
,
1242 var
->get_max_ifc_array_access());
1243 var
->type
= new_type
;
1244 var
->change_interface_type(new_type
);
1246 } else if (var
->type
->is_array() &&
1247 var
->type
->fields
.array
->is_interface()) {
1248 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1249 const glsl_type
*new_type
=
1250 resize_interface_members(var
->type
->fields
.array
,
1251 var
->get_max_ifc_array_access());
1252 var
->change_interface_type(new_type
);
1253 var
->type
= update_interface_members_array(var
->type
, new_type
);
1255 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1256 /* Store a pointer to the variable in the unnamed_interfaces
1259 ir_variable
**interface_vars
= (ir_variable
**)
1260 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1261 if (interface_vars
== NULL
) {
1262 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1264 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1267 unsigned index
= ifc_type
->field_index(var
->name
);
1268 assert(index
< ifc_type
->length
);
1269 assert(interface_vars
[index
] == NULL
);
1270 interface_vars
[index
] = var
;
1272 return visit_continue
;
1276 * For each unnamed interface block that was discovered while running the
1277 * visitor, adjust the interface type to reflect the newly assigned array
1278 * sizes, and fix up the ir_variable nodes to point to the new interface
1281 void fixup_unnamed_interface_types()
1283 hash_table_call_foreach(this->unnamed_interfaces
,
1284 fixup_unnamed_interface_type
, NULL
);
1289 * If the type pointed to by \c type represents an unsized array, replace
1290 * it with a sized array whose size is determined by max_array_access.
1292 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1294 if ((*type
)->is_unsized_array()) {
1295 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1296 max_array_access
+ 1);
1297 assert(*type
!= NULL
);
1301 static const glsl_type
*
1302 update_interface_members_array(const glsl_type
*type
,
1303 const glsl_type
*new_interface_type
)
1305 const glsl_type
*element_type
= type
->fields
.array
;
1306 if (element_type
->is_array()) {
1307 const glsl_type
*new_array_type
=
1308 update_interface_members_array(element_type
, new_interface_type
);
1309 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1311 return glsl_type::get_array_instance(new_interface_type
,
1317 * Determine whether the given interface type contains unsized arrays (if
1318 * it doesn't, array_sizing_visitor doesn't need to process it).
1320 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1322 for (unsigned i
= 0; i
< type
->length
; i
++) {
1323 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1324 if (elem_type
->is_unsized_array())
1331 * Create a new interface type based on the given type, with unsized arrays
1332 * replaced by sized arrays whose size is determined by
1333 * max_ifc_array_access.
1335 static const glsl_type
*
1336 resize_interface_members(const glsl_type
*type
,
1337 const unsigned *max_ifc_array_access
)
1339 unsigned num_fields
= type
->length
;
1340 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1341 memcpy(fields
, type
->fields
.structure
,
1342 num_fields
* sizeof(*fields
));
1343 for (unsigned i
= 0; i
< num_fields
; i
++) {
1344 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1346 glsl_interface_packing packing
=
1347 (glsl_interface_packing
) type
->interface_packing
;
1348 const glsl_type
*new_ifc_type
=
1349 glsl_type::get_interface_instance(fields
, num_fields
,
1350 packing
, type
->name
);
1352 return new_ifc_type
;
1355 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1358 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1359 ir_variable
**interface_vars
= (ir_variable
**) data
;
1360 unsigned num_fields
= ifc_type
->length
;
1361 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1362 memcpy(fields
, ifc_type
->fields
.structure
,
1363 num_fields
* sizeof(*fields
));
1364 bool interface_type_changed
= false;
1365 for (unsigned i
= 0; i
< num_fields
; i
++) {
1366 if (interface_vars
[i
] != NULL
&&
1367 fields
[i
].type
!= interface_vars
[i
]->type
) {
1368 fields
[i
].type
= interface_vars
[i
]->type
;
1369 interface_type_changed
= true;
1372 if (!interface_type_changed
) {
1376 glsl_interface_packing packing
=
1377 (glsl_interface_packing
) ifc_type
->interface_packing
;
1378 const glsl_type
*new_ifc_type
=
1379 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1382 for (unsigned i
= 0; i
< num_fields
; i
++) {
1383 if (interface_vars
[i
] != NULL
)
1384 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1389 * Memory context used to allocate the data in \c unnamed_interfaces.
1394 * Hash table from const glsl_type * to an array of ir_variable *'s
1395 * pointing to the ir_variables constituting each unnamed interface block.
1397 hash_table
*unnamed_interfaces
;
1401 * Performs the cross-validation of layout qualifiers specified in
1402 * redeclaration of gl_FragCoord for the attached fragment shaders,
1403 * and propagates them to the linked FS and linked shader program.
1406 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1407 struct gl_shader
*linked_shader
,
1408 struct gl_shader
**shader_list
,
1409 unsigned num_shaders
)
1411 linked_shader
->redeclares_gl_fragcoord
= false;
1412 linked_shader
->uses_gl_fragcoord
= false;
1413 linked_shader
->origin_upper_left
= false;
1414 linked_shader
->pixel_center_integer
= false;
1416 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1417 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1420 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1421 struct gl_shader
*shader
= shader_list
[i
];
1422 /* From the GLSL 1.50 spec, page 39:
1424 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1425 * it must be redeclared in all the fragment shaders in that program
1426 * that have a static use gl_FragCoord."
1428 if ((linked_shader
->redeclares_gl_fragcoord
1429 && !shader
->redeclares_gl_fragcoord
1430 && shader
->uses_gl_fragcoord
)
1431 || (shader
->redeclares_gl_fragcoord
1432 && !linked_shader
->redeclares_gl_fragcoord
1433 && linked_shader
->uses_gl_fragcoord
)) {
1434 linker_error(prog
, "fragment shader defined with conflicting "
1435 "layout qualifiers for gl_FragCoord\n");
1438 /* From the GLSL 1.50 spec, page 39:
1440 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1441 * single program must have the same set of qualifiers."
1443 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1444 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1445 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1446 linker_error(prog
, "fragment shader defined with conflicting "
1447 "layout qualifiers for gl_FragCoord\n");
1450 /* Update the linked shader state. Note that uses_gl_fragcoord should
1451 * accumulate the results. The other values should replace. If there
1452 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1453 * are already known to be the same.
1455 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1456 linked_shader
->redeclares_gl_fragcoord
=
1457 shader
->redeclares_gl_fragcoord
;
1458 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1459 || shader
->uses_gl_fragcoord
;
1460 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1461 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1464 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1469 * Performs the cross-validation of geometry shader max_vertices and
1470 * primitive type layout qualifiers for the attached geometry shaders,
1471 * and propagates them to the linked GS and linked shader program.
1474 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1475 struct gl_shader
*linked_shader
,
1476 struct gl_shader
**shader_list
,
1477 unsigned num_shaders
)
1479 linked_shader
->Geom
.VerticesOut
= 0;
1480 linked_shader
->Geom
.Invocations
= 0;
1481 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1482 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1484 /* No in/out qualifiers defined for anything but GLSL 1.50+
1485 * geometry shaders so far.
1487 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1490 /* From the GLSL 1.50 spec, page 46:
1492 * "All geometry shader output layout declarations in a program
1493 * must declare the same layout and same value for
1494 * max_vertices. There must be at least one geometry output
1495 * layout declaration somewhere in a program, but not all
1496 * geometry shaders (compilation units) are required to
1500 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1501 struct gl_shader
*shader
= shader_list
[i
];
1503 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1504 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1505 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1506 linker_error(prog
, "geometry shader defined with conflicting "
1510 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1513 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1514 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1515 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1516 linker_error(prog
, "geometry shader defined with conflicting "
1520 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1523 if (shader
->Geom
.VerticesOut
!= 0) {
1524 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1525 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1526 linker_error(prog
, "geometry shader defined with conflicting "
1527 "output vertex count (%d and %d)\n",
1528 linked_shader
->Geom
.VerticesOut
,
1529 shader
->Geom
.VerticesOut
);
1532 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1535 if (shader
->Geom
.Invocations
!= 0) {
1536 if (linked_shader
->Geom
.Invocations
!= 0 &&
1537 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1538 linker_error(prog
, "geometry shader defined with conflicting "
1539 "invocation count (%d and %d)\n",
1540 linked_shader
->Geom
.Invocations
,
1541 shader
->Geom
.Invocations
);
1544 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1548 /* Just do the intrastage -> interstage propagation right now,
1549 * since we already know we're in the right type of shader program
1552 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1554 "geometry shader didn't declare primitive input type\n");
1557 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1559 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1561 "geometry shader didn't declare primitive output type\n");
1564 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1566 if (linked_shader
->Geom
.VerticesOut
== 0) {
1568 "geometry shader didn't declare max_vertices\n");
1571 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1573 if (linked_shader
->Geom
.Invocations
== 0)
1574 linked_shader
->Geom
.Invocations
= 1;
1576 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1581 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1582 * qualifiers for the attached compute shaders, and propagate them to the
1583 * linked CS and linked shader program.
1586 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1587 struct gl_shader
*linked_shader
,
1588 struct gl_shader
**shader_list
,
1589 unsigned num_shaders
)
1591 for (int i
= 0; i
< 3; i
++)
1592 linked_shader
->Comp
.LocalSize
[i
] = 0;
1594 /* This function is called for all shader stages, but it only has an effect
1595 * for compute shaders.
1597 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1600 /* From the ARB_compute_shader spec, in the section describing local size
1603 * If multiple compute shaders attached to a single program object
1604 * declare local work-group size, the declarations must be identical;
1605 * otherwise a link-time error results. Furthermore, if a program
1606 * object contains any compute shaders, at least one must contain an
1607 * input layout qualifier specifying the local work sizes of the
1608 * program, or a link-time error will occur.
1610 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1611 struct gl_shader
*shader
= shader_list
[sh
];
1613 if (shader
->Comp
.LocalSize
[0] != 0) {
1614 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1615 for (int i
= 0; i
< 3; i
++) {
1616 if (linked_shader
->Comp
.LocalSize
[i
] !=
1617 shader
->Comp
.LocalSize
[i
]) {
1618 linker_error(prog
, "compute shader defined with conflicting "
1624 for (int i
= 0; i
< 3; i
++)
1625 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1629 /* Just do the intrastage -> interstage propagation right now,
1630 * since we already know we're in the right type of shader program
1633 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1634 linker_error(prog
, "compute shader didn't declare local size\n");
1637 for (int i
= 0; i
< 3; i
++)
1638 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1643 * Combine a group of shaders for a single stage to generate a linked shader
1646 * If this function is supplied a single shader, it is cloned, and the new
1647 * shader is returned.
1649 static struct gl_shader
*
1650 link_intrastage_shaders(void *mem_ctx
,
1651 struct gl_context
*ctx
,
1652 struct gl_shader_program
*prog
,
1653 struct gl_shader
**shader_list
,
1654 unsigned num_shaders
)
1656 struct gl_uniform_block
*uniform_blocks
= NULL
;
1658 /* Check that global variables defined in multiple shaders are consistent.
1660 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1661 if (!prog
->LinkStatus
)
1664 /* Check that interface blocks defined in multiple shaders are consistent.
1666 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1668 if (!prog
->LinkStatus
)
1671 /* Link up uniform blocks defined within this stage. */
1672 const unsigned num_uniform_blocks
=
1673 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1675 if (!prog
->LinkStatus
)
1678 /* Check that there is only a single definition of each function signature
1679 * across all shaders.
1681 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1682 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1683 ir_function
*const f
= node
->as_function();
1688 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1689 ir_function
*const other
=
1690 shader_list
[j
]->symbols
->get_function(f
->name
);
1692 /* If the other shader has no function (and therefore no function
1693 * signatures) with the same name, skip to the next shader.
1698 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1699 if (!sig
->is_defined
|| sig
->is_builtin())
1702 ir_function_signature
*other_sig
=
1703 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1705 if ((other_sig
!= NULL
) && other_sig
->is_defined
1706 && !other_sig
->is_builtin()) {
1707 linker_error(prog
, "function `%s' is multiply defined\n",
1716 /* Find the shader that defines main, and make a clone of it.
1718 * Starting with the clone, search for undefined references. If one is
1719 * found, find the shader that defines it. Clone the reference and add
1720 * it to the shader. Repeat until there are no undefined references or
1721 * until a reference cannot be resolved.
1723 gl_shader
*main
= NULL
;
1724 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1725 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1726 main
= shader_list
[i
];
1732 linker_error(prog
, "%s shader lacks `main'\n",
1733 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1737 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1738 linked
->ir
= new(linked
) exec_list
;
1739 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1741 linked
->UniformBlocks
= uniform_blocks
;
1742 linked
->NumUniformBlocks
= num_uniform_blocks
;
1743 ralloc_steal(linked
, linked
->UniformBlocks
);
1745 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1746 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1747 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1749 populate_symbol_table(linked
);
1751 /* The pointer to the main function in the final linked shader (i.e., the
1752 * copy of the original shader that contained the main function).
1754 ir_function_signature
*const main_sig
=
1755 link_get_main_function_signature(linked
);
1757 /* Move any instructions other than variable declarations or function
1758 * declarations into main.
1760 exec_node
*insertion_point
=
1761 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1764 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1765 if (shader_list
[i
] == main
)
1768 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1769 insertion_point
, true, linked
);
1772 /* Check if any shader needs built-in functions. */
1773 bool need_builtins
= false;
1774 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1775 if (shader_list
[i
]->uses_builtin_functions
) {
1776 need_builtins
= true;
1782 if (need_builtins
) {
1783 /* Make a temporary array one larger than shader_list, which will hold
1784 * the built-in function shader as well.
1786 gl_shader
**linking_shaders
= (gl_shader
**)
1787 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1789 ok
= linking_shaders
!= NULL
;
1792 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1793 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1795 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1797 free(linking_shaders
);
1799 _mesa_error_no_memory(__func__
);
1802 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1807 ctx
->Driver
.DeleteShader(ctx
, linked
);
1811 /* At this point linked should contain all of the linked IR, so
1812 * validate it to make sure nothing went wrong.
1814 validate_ir_tree(linked
->ir
);
1816 /* Set the size of geometry shader input arrays */
1817 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1818 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1819 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1820 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
1821 ir
->accept(&input_resize_visitor
);
1825 if (ctx
->Const
.VertexID_is_zero_based
)
1826 lower_vertex_id(linked
);
1828 /* Make a pass over all variable declarations to ensure that arrays with
1829 * unspecified sizes have a size specified. The size is inferred from the
1830 * max_array_access field.
1832 array_sizing_visitor v
;
1834 v
.fixup_unnamed_interface_types();
1840 * Update the sizes of linked shader uniform arrays to the maximum
1843 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1845 * If one or more elements of an array are active,
1846 * GetActiveUniform will return the name of the array in name,
1847 * subject to the restrictions listed above. The type of the array
1848 * is returned in type. The size parameter contains the highest
1849 * array element index used, plus one. The compiler or linker
1850 * determines the highest index used. There will be only one
1851 * active uniform reported by the GL per uniform array.
1855 update_array_sizes(struct gl_shader_program
*prog
)
1857 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1858 if (prog
->_LinkedShaders
[i
] == NULL
)
1861 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
1862 ir_variable
*const var
= node
->as_variable();
1864 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1865 !var
->type
->is_array())
1868 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1869 * will not be eliminated. Since we always do std140, just
1870 * don't resize arrays in UBOs.
1872 * Atomic counters are supposed to get deterministic
1873 * locations assigned based on the declaration ordering and
1874 * sizes, array compaction would mess that up.
1876 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1879 unsigned int size
= var
->data
.max_array_access
;
1880 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1881 if (prog
->_LinkedShaders
[j
] == NULL
)
1884 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
1885 ir_variable
*other_var
= node2
->as_variable();
1889 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1890 other_var
->data
.max_array_access
> size
) {
1891 size
= other_var
->data
.max_array_access
;
1896 if (size
+ 1 != var
->type
->length
) {
1897 /* If this is a built-in uniform (i.e., it's backed by some
1898 * fixed-function state), adjust the number of state slots to
1899 * match the new array size. The number of slots per array entry
1900 * is not known. It seems safe to assume that the total number of
1901 * slots is an integer multiple of the number of array elements.
1902 * Determine the number of slots per array element by dividing by
1903 * the old (total) size.
1905 const unsigned num_slots
= var
->get_num_state_slots();
1906 if (num_slots
> 0) {
1907 var
->set_num_state_slots((size
+ 1)
1908 * (num_slots
/ var
->type
->length
));
1911 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1913 /* FINISHME: We should update the types of array
1914 * dereferences of this variable now.
1922 * Find a contiguous set of available bits in a bitmask.
1924 * \param used_mask Bits representing used (1) and unused (0) locations
1925 * \param needed_count Number of contiguous bits needed.
1928 * Base location of the available bits on success or -1 on failure.
1931 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1933 unsigned needed_mask
= (1 << needed_count
) - 1;
1934 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1936 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1937 * cannot optimize possibly infinite loops" for the loop below.
1939 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1942 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1943 if ((needed_mask
& ~used_mask
) == needed_mask
)
1954 * Assign locations for either VS inputs or FS outputs
1956 * \param prog Shader program whose variables need locations assigned
1957 * \param target_index Selector for the program target to receive location
1958 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1959 * \c MESA_SHADER_FRAGMENT.
1960 * \param max_index Maximum number of generic locations. This corresponds
1961 * to either the maximum number of draw buffers or the
1962 * maximum number of generic attributes.
1965 * If locations are successfully assigned, true is returned. Otherwise an
1966 * error is emitted to the shader link log and false is returned.
1969 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1970 unsigned target_index
,
1973 /* Mark invalid locations as being used.
1975 unsigned used_locations
= (max_index
>= 32)
1976 ? ~0 : ~((1 << max_index
) - 1);
1978 assert((target_index
== MESA_SHADER_VERTEX
)
1979 || (target_index
== MESA_SHADER_FRAGMENT
));
1981 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1985 /* Operate in a total of four passes.
1987 * 1. Invalidate the location assignments for all vertex shader inputs.
1989 * 2. Assign locations for inputs that have user-defined (via
1990 * glBindVertexAttribLocation) locations and outputs that have
1991 * user-defined locations (via glBindFragDataLocation).
1993 * 3. Sort the attributes without assigned locations by number of slots
1994 * required in decreasing order. Fragmentation caused by attribute
1995 * locations assigned by the application may prevent large attributes
1996 * from having enough contiguous space.
1998 * 4. Assign locations to any inputs without assigned locations.
2001 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2002 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2004 const enum ir_variable_mode direction
=
2005 (target_index
== MESA_SHADER_VERTEX
)
2006 ? ir_var_shader_in
: ir_var_shader_out
;
2009 /* Temporary storage for the set of attributes that need locations assigned.
2015 /* Used below in the call to qsort. */
2016 static int compare(const void *a
, const void *b
)
2018 const temp_attr
*const l
= (const temp_attr
*) a
;
2019 const temp_attr
*const r
= (const temp_attr
*) b
;
2021 /* Reversed because we want a descending order sort below. */
2022 return r
->slots
- l
->slots
;
2026 unsigned num_attr
= 0;
2027 unsigned total_attribs_size
= 0;
2029 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2030 ir_variable
*const var
= node
->as_variable();
2032 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2035 if (var
->data
.explicit_location
) {
2036 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2037 || (var
->data
.location
< 0)) {
2039 "invalid explicit location %d specified for `%s'\n",
2040 (var
->data
.location
< 0)
2041 ? var
->data
.location
2042 : var
->data
.location
- generic_base
,
2046 } else if (target_index
== MESA_SHADER_VERTEX
) {
2049 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2050 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2051 var
->data
.location
= binding
;
2052 var
->data
.is_unmatched_generic_inout
= 0;
2054 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2058 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2059 assert(binding
>= FRAG_RESULT_DATA0
);
2060 var
->data
.location
= binding
;
2061 var
->data
.is_unmatched_generic_inout
= 0;
2063 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2064 var
->data
.index
= index
;
2069 const unsigned slots
= var
->type
->count_attribute_slots();
2071 /* From GL4.5 core spec, section 11.1.1 (Vertex Attributes):
2073 * "A program with more than the value of MAX_VERTEX_ATTRIBS active
2074 * attribute variables may fail to link, unless device-dependent
2075 * optimizations are able to make the program fit within available
2076 * hardware resources. For the purposes of this test, attribute variables
2077 * of the type dvec3, dvec4, dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3,
2078 * and dmat4 may count as consuming twice as many attributes as equivalent
2079 * single-precision types. While these types use the same number of
2080 * generic attributes as their single-precision equivalents,
2081 * implementations are permitted to consume two single-precision vectors
2082 * of internal storage for each three- or four-component double-precision
2084 * Until someone has a good reason in Mesa, enforce that now.
2086 if (target_index
== MESA_SHADER_VERTEX
) {
2087 total_attribs_size
+= slots
;
2088 if (var
->type
->without_array() == glsl_type::dvec3_type
||
2089 var
->type
->without_array() == glsl_type::dvec4_type
||
2090 var
->type
->without_array() == glsl_type::dmat2x3_type
||
2091 var
->type
->without_array() == glsl_type::dmat2x4_type
||
2092 var
->type
->without_array() == glsl_type::dmat3_type
||
2093 var
->type
->without_array() == glsl_type::dmat3x4_type
||
2094 var
->type
->without_array() == glsl_type::dmat4x3_type
||
2095 var
->type
->without_array() == glsl_type::dmat4_type
)
2096 total_attribs_size
+= slots
;
2099 /* If the variable is not a built-in and has a location statically
2100 * assigned in the shader (presumably via a layout qualifier), make sure
2101 * that it doesn't collide with other assigned locations. Otherwise,
2102 * add it to the list of variables that need linker-assigned locations.
2104 if (var
->data
.location
!= -1) {
2105 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2106 /* From page 61 of the OpenGL 4.0 spec:
2108 * "LinkProgram will fail if the attribute bindings assigned
2109 * by BindAttribLocation do not leave not enough space to
2110 * assign a location for an active matrix attribute or an
2111 * active attribute array, both of which require multiple
2112 * contiguous generic attributes."
2114 * I think above text prohibits the aliasing of explicit and
2115 * automatic assignments. But, aliasing is allowed in manual
2116 * assignments of attribute locations. See below comments for
2119 * From OpenGL 4.0 spec, page 61:
2121 * "It is possible for an application to bind more than one
2122 * attribute name to the same location. This is referred to as
2123 * aliasing. This will only work if only one of the aliased
2124 * attributes is active in the executable program, or if no
2125 * path through the shader consumes more than one attribute of
2126 * a set of attributes aliased to the same location. A link
2127 * error can occur if the linker determines that every path
2128 * through the shader consumes multiple aliased attributes,
2129 * but implementations are not required to generate an error
2132 * From GLSL 4.30 spec, page 54:
2134 * "A program will fail to link if any two non-vertex shader
2135 * input variables are assigned to the same location. For
2136 * vertex shaders, multiple input variables may be assigned
2137 * to the same location using either layout qualifiers or via
2138 * the OpenGL API. However, such aliasing is intended only to
2139 * support vertex shaders where each execution path accesses
2140 * at most one input per each location. Implementations are
2141 * permitted, but not required, to generate link-time errors
2142 * if they detect that every path through the vertex shader
2143 * executable accesses multiple inputs assigned to any single
2144 * location. For all shader types, a program will fail to link
2145 * if explicit location assignments leave the linker unable
2146 * to find space for other variables without explicit
2149 * From OpenGL ES 3.0 spec, page 56:
2151 * "Binding more than one attribute name to the same location
2152 * is referred to as aliasing, and is not permitted in OpenGL
2153 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2154 * fail when this condition exists. However, aliasing is
2155 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2156 * This will only work if only one of the aliased attributes
2157 * is active in the executable program, or if no path through
2158 * the shader consumes more than one attribute of a set of
2159 * attributes aliased to the same location. A link error can
2160 * occur if the linker determines that every path through the
2161 * shader consumes multiple aliased attributes, but implemen-
2162 * tations are not required to generate an error in this case."
2164 * After looking at above references from OpenGL, OpenGL ES and
2165 * GLSL specifications, we allow aliasing of vertex input variables
2166 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2168 * NOTE: This is not required by the spec but its worth mentioning
2169 * here that we're not doing anything to make sure that no path
2170 * through the vertex shader executable accesses multiple inputs
2171 * assigned to any single location.
2174 /* Mask representing the contiguous slots that will be used by
2177 const unsigned attr
= var
->data
.location
- generic_base
;
2178 const unsigned use_mask
= (1 << slots
) - 1;
2179 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2180 ? "vertex shader input" : "fragment shader output";
2182 /* Generate a link error if the requested locations for this
2183 * attribute exceed the maximum allowed attribute location.
2185 if (attr
+ slots
> max_index
) {
2187 "insufficient contiguous locations "
2188 "available for %s `%s' %d %d %d\n", string
,
2189 var
->name
, used_locations
, use_mask
, attr
);
2193 /* Generate a link error if the set of bits requested for this
2194 * attribute overlaps any previously allocated bits.
2196 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2197 if (target_index
== MESA_SHADER_FRAGMENT
||
2198 (prog
->IsES
&& prog
->Version
>= 300)) {
2200 "overlapping location is assigned "
2201 "to %s `%s' %d %d %d\n", string
,
2202 var
->name
, used_locations
, use_mask
, attr
);
2205 linker_warning(prog
,
2206 "overlapping location is assigned "
2207 "to %s `%s' %d %d %d\n", string
,
2208 var
->name
, used_locations
, use_mask
, attr
);
2212 used_locations
|= (use_mask
<< attr
);
2218 to_assign
[num_attr
].slots
= slots
;
2219 to_assign
[num_attr
].var
= var
;
2223 if (target_index
== MESA_SHADER_VERTEX
) {
2224 if (total_attribs_size
> max_index
) {
2226 "attempt to use %d vertex attribute slots only %d available ",
2227 total_attribs_size
, max_index
);
2232 /* If all of the attributes were assigned locations by the application (or
2233 * are built-in attributes with fixed locations), return early. This should
2234 * be the common case.
2239 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2241 if (target_index
== MESA_SHADER_VERTEX
) {
2242 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2243 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2244 * reserved to prevent it from being automatically allocated below.
2246 find_deref_visitor
find("gl_Vertex");
2248 if (find
.variable_found())
2249 used_locations
|= (1 << 0);
2252 for (unsigned i
= 0; i
< num_attr
; i
++) {
2253 /* Mask representing the contiguous slots that will be used by this
2256 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2258 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2261 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2262 ? "vertex shader input" : "fragment shader output";
2265 "insufficient contiguous locations "
2266 "available for %s `%s'\n",
2267 string
, to_assign
[i
].var
->name
);
2271 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2272 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2273 used_locations
|= (use_mask
<< location
);
2281 * Demote shader inputs and outputs that are not used in other stages
2284 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2286 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2287 ir_variable
*const var
= node
->as_variable();
2289 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2292 /* A shader 'in' or 'out' variable is only really an input or output if
2293 * its value is used by other shader stages. This will cause the variable
2294 * to have a location assigned.
2296 if (var
->data
.is_unmatched_generic_inout
) {
2297 assert(var
->data
.mode
!= ir_var_temporary
);
2298 var
->data
.mode
= ir_var_auto
;
2305 * Store the gl_FragDepth layout in the gl_shader_program struct.
2308 store_fragdepth_layout(struct gl_shader_program
*prog
)
2310 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2314 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2316 /* We don't look up the gl_FragDepth symbol directly because if
2317 * gl_FragDepth is not used in the shader, it's removed from the IR.
2318 * However, the symbol won't be removed from the symbol table.
2320 * We're only interested in the cases where the variable is NOT removed
2323 foreach_in_list(ir_instruction
, node
, ir
) {
2324 ir_variable
*const var
= node
->as_variable();
2326 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2330 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2331 switch (var
->data
.depth_layout
) {
2332 case ir_depth_layout_none
:
2333 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2335 case ir_depth_layout_any
:
2336 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2338 case ir_depth_layout_greater
:
2339 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2341 case ir_depth_layout_less
:
2342 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2344 case ir_depth_layout_unchanged
:
2345 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2356 * Validate the resources used by a program versus the implementation limits
2359 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2361 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2362 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2367 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2368 linker_error(prog
, "Too many %s shader texture samplers\n",
2369 _mesa_shader_stage_to_string(i
));
2372 if (sh
->num_uniform_components
>
2373 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2374 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2375 linker_warning(prog
, "Too many %s shader default uniform block "
2376 "components, but the driver will try to optimize "
2377 "them out; this is non-portable out-of-spec "
2379 _mesa_shader_stage_to_string(i
));
2381 linker_error(prog
, "Too many %s shader default uniform block "
2383 _mesa_shader_stage_to_string(i
));
2387 if (sh
->num_combined_uniform_components
>
2388 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2389 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2390 linker_warning(prog
, "Too many %s shader uniform components, "
2391 "but the driver will try to optimize them out; "
2392 "this is non-portable out-of-spec behavior\n",
2393 _mesa_shader_stage_to_string(i
));
2395 linker_error(prog
, "Too many %s shader uniform components\n",
2396 _mesa_shader_stage_to_string(i
));
2401 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2402 unsigned total_uniform_blocks
= 0;
2404 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2405 if (prog
->UniformBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2406 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2407 prog
->UniformBlocks
[i
].Name
,
2408 prog
->UniformBlocks
[i
].UniformBufferSize
,
2409 ctx
->Const
.MaxUniformBlockSize
);
2412 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2413 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2415 total_uniform_blocks
++;
2419 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2420 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2421 prog
->NumUniformBlocks
,
2422 ctx
->Const
.MaxCombinedUniformBlocks
);
2424 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2425 const unsigned max_uniform_blocks
=
2426 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2427 if (blocks
[i
] > max_uniform_blocks
) {
2428 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2429 _mesa_shader_stage_to_string(i
),
2431 max_uniform_blocks
);
2440 * Validate shader image resources.
2443 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2445 unsigned total_image_units
= 0;
2446 unsigned fragment_outputs
= 0;
2448 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2451 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2452 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2455 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2456 linker_error(prog
, "Too many %s shader image uniforms\n",
2457 _mesa_shader_stage_to_string(i
));
2459 total_image_units
+= sh
->NumImages
;
2461 if (i
== MESA_SHADER_FRAGMENT
) {
2462 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2463 ir_variable
*var
= node
->as_variable();
2464 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2465 fragment_outputs
+= var
->type
->count_attribute_slots();
2471 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2472 linker_error(prog
, "Too many combined image uniforms\n");
2474 if (total_image_units
+ fragment_outputs
>
2475 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2476 linker_error(prog
, "Too many combined image uniforms and fragment outputs\n");
2481 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2482 * for a variable, checks for overlaps between other uniforms using explicit
2486 reserve_explicit_locations(struct gl_shader_program
*prog
,
2487 string_to_uint_map
*map
, ir_variable
*var
)
2489 unsigned slots
= var
->type
->uniform_locations();
2490 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2492 /* Resize remap table if locations do not fit in the current one. */
2493 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2494 prog
->UniformRemapTable
=
2495 reralloc(prog
, prog
->UniformRemapTable
,
2496 gl_uniform_storage
*,
2499 if (!prog
->UniformRemapTable
) {
2500 linker_error(prog
, "Out of memory during linking.\n");
2504 /* Initialize allocated space. */
2505 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2506 prog
->UniformRemapTable
[i
] = NULL
;
2508 prog
->NumUniformRemapTable
= max_loc
+ 1;
2511 for (unsigned i
= 0; i
< slots
; i
++) {
2512 unsigned loc
= var
->data
.location
+ i
;
2514 /* Check if location is already used. */
2515 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2517 /* Possibly same uniform from a different stage, this is ok. */
2519 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2522 /* ARB_explicit_uniform_location specification states:
2524 * "No two default-block uniform variables in the program can have
2525 * the same location, even if they are unused, otherwise a compiler
2526 * or linker error will be generated."
2529 "location qualifier for uniform %s overlaps "
2530 "previously used location\n",
2535 /* Initialize location as inactive before optimization
2536 * rounds and location assignment.
2538 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2541 /* Note, base location used for arrays. */
2542 map
->put(var
->data
.location
, var
->name
);
2548 * Check and reserve all explicit uniform locations, called before
2549 * any optimizations happen to handle also inactive uniforms and
2550 * inactive array elements that may get trimmed away.
2553 check_explicit_uniform_locations(struct gl_context
*ctx
,
2554 struct gl_shader_program
*prog
)
2556 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2559 /* This map is used to detect if overlapping explicit locations
2560 * occur with the same uniform (from different stage) or a different one.
2562 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2565 linker_error(prog
, "Out of memory during linking.\n");
2569 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2570 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2575 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2576 ir_variable
*var
= node
->as_variable();
2577 if ((var
&& var
->data
.mode
== ir_var_uniform
) &&
2578 var
->data
.explicit_location
) {
2579 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2591 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
2592 const void *data
, uint8_t stages
)
2596 /* If resource already exists, do not add it again. */
2597 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
2598 if (prog
->ProgramResourceList
[i
].Data
== data
)
2601 prog
->ProgramResourceList
=
2603 prog
->ProgramResourceList
,
2604 gl_program_resource
,
2605 prog
->NumProgramResourceList
+ 1);
2607 if (!prog
->ProgramResourceList
) {
2608 linker_error(prog
, "Out of memory during linking.\n");
2612 struct gl_program_resource
*res
=
2613 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
2617 res
->StageReferences
= stages
;
2619 prog
->NumProgramResourceList
++;
2625 * Function builds a stage reference bitmask from variable name.
2628 build_stageref(struct gl_shader_program
*shProg
, const char *name
)
2632 /* Note, that we assume MAX 8 stages, if there will be more stages, type
2633 * used for reference mask in gl_program_resource will need to be changed.
2635 assert(MESA_SHADER_STAGES
< 8);
2637 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2638 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
2642 /* Shader symbol table may contain variables that have
2643 * been optimized away. Search IR for the variable instead.
2645 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2646 ir_variable
*var
= node
->as_variable();
2647 if (var
&& strcmp(var
->name
, name
) == 0) {
2657 add_interface_variables(struct gl_shader_program
*shProg
,
2658 struct gl_shader
*sh
, GLenum programInterface
)
2660 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2661 ir_variable
*var
= node
->as_variable();
2667 switch (var
->data
.mode
) {
2668 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
2669 * "For GetActiveAttrib, all active vertex shader input variables
2670 * are enumerated, including the special built-in inputs gl_VertexID
2671 * and gl_InstanceID."
2673 case ir_var_system_value
:
2674 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
2675 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
2676 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
2678 /* Mark special built-in inputs referenced by the vertex stage so
2679 * that they are considered active by the shader queries.
2681 mask
= (1 << (MESA_SHADER_VERTEX
));
2683 case ir_var_shader_in
:
2684 if (programInterface
!= GL_PROGRAM_INPUT
)
2687 case ir_var_shader_out
:
2688 if (programInterface
!= GL_PROGRAM_OUTPUT
)
2695 if (!add_program_resource(shProg
, programInterface
, var
,
2696 build_stageref(shProg
, var
->name
) | mask
))
2703 * Builds up a list of program resources that point to existing
2707 build_program_resource_list(struct gl_context
*ctx
,
2708 struct gl_shader_program
*shProg
)
2710 /* Rebuild resource list. */
2711 if (shProg
->ProgramResourceList
) {
2712 ralloc_free(shProg
->ProgramResourceList
);
2713 shProg
->ProgramResourceList
= NULL
;
2714 shProg
->NumProgramResourceList
= 0;
2717 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
2719 /* Determine first input and final output stage. These are used to
2720 * detect which variables should be enumerated in the resource list
2721 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
2723 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2724 if (!shProg
->_LinkedShaders
[i
])
2726 if (input_stage
== MESA_SHADER_STAGES
)
2731 /* Empty shader, no resources. */
2732 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
2735 /* Add inputs and outputs to the resource list. */
2736 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
],
2740 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
],
2744 /* Add transform feedback varyings. */
2745 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
2746 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
2748 build_stageref(shProg
,
2749 shProg
->LinkedTransformFeedback
.Varyings
[i
].Name
);
2750 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
2751 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
2757 /* Add uniforms from uniform storage. */
2758 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
2759 /* Do not add uniforms internally used by Mesa. */
2760 if (shProg
->UniformStorage
[i
].hidden
)
2764 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
);
2766 /* Add stagereferences for uniforms in a uniform block. */
2767 int block_index
= shProg
->UniformStorage
[i
].block_index
;
2768 if (block_index
!= -1) {
2769 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2770 if (shProg
->UniformBlockStageIndex
[j
][block_index
] != -1)
2771 stageref
|= (1 << j
);
2775 if (!add_program_resource(shProg
, GL_UNIFORM
,
2776 &shProg
->UniformStorage
[i
], stageref
))
2780 /* Add program uniform blocks. */
2781 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
2782 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
2783 &shProg
->UniformBlocks
[i
], 0))
2787 /* Add atomic counter buffers. */
2788 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
2789 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
2790 &shProg
->AtomicBuffers
[i
], 0))
2794 /* TODO - following extensions will require more resource types:
2796 * GL_ARB_shader_storage_buffer_object
2797 * GL_ARB_shader_subroutine
2802 * This check is done to make sure we allow only constant expression
2803 * indexing and "constant-index-expression" (indexing with an expression
2804 * that includes loop induction variable).
2807 validate_sampler_array_indexing(struct gl_context
*ctx
,
2808 struct gl_shader_program
*prog
)
2810 dynamic_sampler_array_indexing_visitor v
;
2811 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2812 if (prog
->_LinkedShaders
[i
] == NULL
)
2815 bool no_dynamic_indexing
=
2816 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
2818 /* Search for array derefs in shader. */
2819 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2820 if (v
.uses_dynamic_sampler_array_indexing()) {
2821 const char *msg
= "sampler arrays indexed with non-constant "
2822 "expressions is forbidden in GLSL %s %u";
2823 /* Backend has indicated that it has no dynamic indexing support. */
2824 if (no_dynamic_indexing
) {
2825 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
2828 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
2837 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2839 tfeedback_decl
*tfeedback_decls
= NULL
;
2840 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2842 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2844 prog
->LinkStatus
= true; /* All error paths will set this to false */
2845 prog
->Validated
= false;
2846 prog
->_Used
= false;
2848 prog
->ARB_fragment_coord_conventions_enable
= false;
2850 /* Separate the shaders into groups based on their type.
2852 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2853 unsigned num_shaders
[MESA_SHADER_STAGES
];
2855 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2856 shader_list
[i
] = (struct gl_shader
**)
2857 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2861 unsigned min_version
= UINT_MAX
;
2862 unsigned max_version
= 0;
2863 const bool is_es_prog
=
2864 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2865 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2866 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2867 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2869 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2870 linker_error(prog
, "all shaders must use same shading "
2871 "language version\n");
2875 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
2876 prog
->ARB_fragment_coord_conventions_enable
= true;
2879 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2880 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2881 num_shaders
[shader_type
]++;
2884 /* In desktop GLSL, different shader versions may be linked together. In
2885 * GLSL ES, all shader versions must be the same.
2887 if (is_es_prog
&& min_version
!= max_version
) {
2888 linker_error(prog
, "all shaders must use same shading "
2889 "language version\n");
2893 prog
->Version
= max_version
;
2894 prog
->IsES
= is_es_prog
;
2896 /* Geometry shaders have to be linked with vertex shaders.
2898 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2899 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2900 !prog
->SeparateShader
) {
2901 linker_error(prog
, "Geometry shader must be linked with "
2906 /* Compute shaders have additional restrictions. */
2907 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2908 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2909 linker_error(prog
, "Compute shaders may not be linked with any other "
2910 "type of shader\n");
2913 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2914 if (prog
->_LinkedShaders
[i
] != NULL
)
2915 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2917 prog
->_LinkedShaders
[i
] = NULL
;
2920 /* Link all shaders for a particular stage and validate the result.
2922 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2923 if (num_shaders
[stage
] > 0) {
2924 gl_shader
*const sh
=
2925 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2926 num_shaders
[stage
]);
2928 if (!prog
->LinkStatus
) {
2930 ctx
->Driver
.DeleteShader(ctx
, sh
);
2935 case MESA_SHADER_VERTEX
:
2936 validate_vertex_shader_executable(prog
, sh
);
2938 case MESA_SHADER_GEOMETRY
:
2939 validate_geometry_shader_executable(prog
, sh
);
2941 case MESA_SHADER_FRAGMENT
:
2942 validate_fragment_shader_executable(prog
, sh
);
2945 if (!prog
->LinkStatus
) {
2947 ctx
->Driver
.DeleteShader(ctx
, sh
);
2951 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2955 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2956 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2957 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2958 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2960 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2962 /* Here begins the inter-stage linking phase. Some initial validation is
2963 * performed, then locations are assigned for uniforms, attributes, and
2966 cross_validate_uniforms(prog
);
2967 if (!prog
->LinkStatus
)
2972 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2973 if (prog
->_LinkedShaders
[prev
] != NULL
)
2977 check_explicit_uniform_locations(ctx
, prog
);
2978 if (!prog
->LinkStatus
)
2981 /* Validate the inputs of each stage with the output of the preceding
2984 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2985 if (prog
->_LinkedShaders
[i
] == NULL
)
2988 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2989 prog
->_LinkedShaders
[i
]);
2990 if (!prog
->LinkStatus
)
2993 cross_validate_outputs_to_inputs(prog
,
2994 prog
->_LinkedShaders
[prev
],
2995 prog
->_LinkedShaders
[i
]);
2996 if (!prog
->LinkStatus
)
3002 /* Cross-validate uniform blocks between shader stages */
3003 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
3004 MESA_SHADER_STAGES
);
3005 if (!prog
->LinkStatus
)
3008 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3009 if (prog
->_LinkedShaders
[i
] != NULL
)
3010 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
3013 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
3014 * it before optimization because we want most of the checks to get
3015 * dropped thanks to constant propagation.
3017 * This rule also applies to GLSL ES 3.00.
3019 if (max_version
>= (is_es_prog
? 300 : 130)) {
3020 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3022 lower_discard_flow(sh
->ir
);
3026 if (!interstage_cross_validate_uniform_blocks(prog
))
3029 /* Do common optimization before assigning storage for attributes,
3030 * uniforms, and varyings. Later optimization could possibly make
3031 * some of that unused.
3033 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3034 if (prog
->_LinkedShaders
[i
] == NULL
)
3037 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
3038 if (!prog
->LinkStatus
)
3041 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
3042 lower_clip_distance(prog
->_LinkedShaders
[i
]);
3045 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
3046 &ctx
->Const
.ShaderCompilerOptions
[i
],
3047 ctx
->Const
.NativeIntegers
))
3050 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
3053 /* Validation for special cases where we allow sampler array indexing
3054 * with loop induction variable. This check emits a warning or error
3055 * depending if backend can handle dynamic indexing.
3057 if ((!prog
->IsES
&& prog
->Version
< 130) ||
3058 (prog
->IsES
&& prog
->Version
< 300)) {
3059 if (!validate_sampler_array_indexing(ctx
, prog
))
3063 /* Check and validate stream emissions in geometry shaders */
3064 validate_geometry_shader_emissions(ctx
, prog
);
3066 /* Mark all generic shader inputs and outputs as unpaired. */
3067 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3068 if (prog
->_LinkedShaders
[i
] != NULL
) {
3069 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
3073 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
,
3074 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAttribs
)) {
3078 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
3082 unsigned first
, last
;
3084 first
= MESA_SHADER_STAGES
;
3087 /* Determine first and last stage. */
3088 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3089 if (!prog
->_LinkedShaders
[i
])
3091 if (first
== MESA_SHADER_STAGES
)
3096 if (num_tfeedback_decls
!= 0) {
3097 /* From GL_EXT_transform_feedback:
3098 * A program will fail to link if:
3100 * * the <count> specified by TransformFeedbackVaryingsEXT is
3101 * non-zero, but the program object has no vertex or geometry
3104 if (first
== MESA_SHADER_FRAGMENT
) {
3105 linker_error(prog
, "Transform feedback varyings specified, but "
3106 "no vertex or geometry shader is present.\n");
3110 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
3111 prog
->TransformFeedback
.NumVarying
);
3112 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
3113 prog
->TransformFeedback
.VaryingNames
,
3118 /* Linking the stages in the opposite order (from fragment to vertex)
3119 * ensures that inter-shader outputs written to in an earlier stage are
3120 * eliminated if they are (transitively) not used in a later stage.
3124 if (first
< MESA_SHADER_FRAGMENT
) {
3125 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
3127 if (first
== MESA_SHADER_GEOMETRY
) {
3128 /* There was no vertex shader, but we still have to assign varying
3129 * locations for use by geometry shader inputs in SSO.
3131 * If the shader is not separable (i.e., prog->SeparateShader is
3132 * false), linking will have already failed when first is
3133 * MESA_SHADER_GEOMETRY.
3135 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3136 NULL
, prog
->_LinkedShaders
[first
],
3137 num_tfeedback_decls
, tfeedback_decls
,
3138 prog
->Geom
.VerticesIn
))
3142 if (last
!= MESA_SHADER_FRAGMENT
&&
3143 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
3144 /* There was no fragment shader, but we still have to assign varying
3145 * locations for use by transform feedback.
3147 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3149 num_tfeedback_decls
, tfeedback_decls
,
3154 do_dead_builtin_varyings(ctx
, sh
, NULL
,
3155 num_tfeedback_decls
, tfeedback_decls
);
3157 if (!prog
->SeparateShader
)
3158 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
3160 /* Eliminate code that is now dead due to unused outputs being demoted.
3162 while (do_dead_code(sh
->ir
, false))
3165 else if (first
== MESA_SHADER_FRAGMENT
) {
3166 /* If the program only contains a fragment shader...
3168 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
3170 do_dead_builtin_varyings(ctx
, NULL
, sh
,
3171 num_tfeedback_decls
, tfeedback_decls
);
3173 if (prog
->SeparateShader
) {
3174 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3175 NULL
/* producer */,
3177 0 /* num_tfeedback_decls */,
3178 NULL
/* tfeedback_decls */,
3179 0 /* gs_input_vertices */))
3182 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
3184 while (do_dead_code(sh
->ir
, false))
3189 for (int i
= next
- 1; i
>= 0; i
--) {
3190 if (prog
->_LinkedShaders
[i
] == NULL
)
3193 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
3194 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
3195 unsigned gs_input_vertices
=
3196 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
3198 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
3199 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3200 tfeedback_decls
, gs_input_vertices
))
3203 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
3204 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3207 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
3208 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
3210 /* Eliminate code that is now dead due to unused outputs being demoted.
3212 while (do_dead_code(sh_i
->ir
, false))
3214 while (do_dead_code(sh_next
->ir
, false))
3217 /* This must be done after all dead varyings are eliminated. */
3218 if (!check_against_output_limit(ctx
, prog
, sh_i
))
3220 if (!check_against_input_limit(ctx
, prog
, sh_next
))
3226 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
3229 update_array_sizes(prog
);
3230 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
3231 link_assign_atomic_counter_resources(ctx
, prog
);
3232 store_fragdepth_layout(prog
);
3234 check_resources(ctx
, prog
);
3235 check_image_resources(ctx
, prog
);
3236 link_check_atomic_counter_resources(ctx
, prog
);
3238 if (!prog
->LinkStatus
)
3241 /* OpenGL ES requires that a vertex shader and a fragment shader both be
3242 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
3243 * anything about shader linking when one of the shaders (vertex or
3244 * fragment shader) is absent. So, the extension shouldn't change the
3245 * behavior specified in GLSL specification.
3247 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
3248 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
3249 linker_error(prog
, "program lacks a vertex shader\n");
3250 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3251 linker_error(prog
, "program lacks a fragment shader\n");
3255 /* FINISHME: Assign fragment shader output locations. */
3258 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3259 free(shader_list
[i
]);
3260 if (prog
->_LinkedShaders
[i
] == NULL
)
3263 /* Do a final validation step to make sure that the IR wasn't
3264 * invalidated by any modifications performed after intrastage linking.
3266 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
3268 /* Retain any live IR, but trash the rest. */
3269 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
3271 /* The symbol table in the linked shaders may contain references to
3272 * variables that were removed (e.g., unused uniforms). Since it may
3273 * contain junk, there is no possible valid use. Delete it and set the
3276 delete prog
->_LinkedShaders
[i
]->symbols
;
3277 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
3280 ralloc_free(mem_ctx
);