2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
67 #include "main/core.h"
68 #include "glsl_symbol_table.h"
69 #include "glsl_parser_extras.h"
72 #include "program/hash_table.h"
74 #include "link_varyings.h"
75 #include "ir_optimization.h"
76 #include "ir_rvalue_visitor.h"
77 #include "ir_uniform.h"
79 #include "main/shaderobj.h"
80 #include "main/enums.h"
83 void linker_error(gl_shader_program
*, const char *, ...);
88 * Visitor that determines whether or not a variable is ever written.
90 class find_assignment_visitor
: public ir_hierarchical_visitor
{
92 find_assignment_visitor(const char *name
)
93 : name(name
), found(false)
98 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
100 ir_variable
*const var
= ir
->lhs
->variable_referenced();
102 if (strcmp(name
, var
->name
) == 0) {
107 return visit_continue_with_parent
;
110 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
112 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
113 actual_node
, &ir
->actual_parameters
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
115 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
117 if (sig_param
->data
.mode
== ir_var_function_out
||
118 sig_param
->data
.mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
127 if (ir
->return_deref
!= NULL
) {
128 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
130 if (strcmp(name
, var
->name
) == 0) {
136 return visit_continue_with_parent
;
139 bool variable_found()
145 const char *name
; /**< Find writes to a variable with this name. */
146 bool found
; /**< Was a write to the variable found? */
151 * Visitor that determines whether or not a variable is ever read.
153 class find_deref_visitor
: public ir_hierarchical_visitor
{
155 find_deref_visitor(const char *name
)
156 : name(name
), found(false)
161 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
163 if (strcmp(this->name
, ir
->var
->name
) == 0) {
168 return visit_continue
;
171 bool variable_found() const
177 const char *name
; /**< Find writes to a variable with this name. */
178 bool found
; /**< Was a write to the variable found? */
182 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
184 unsigned num_vertices
;
185 gl_shader_program
*prog
;
187 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
189 this->num_vertices
= num_vertices
;
193 virtual ~geom_array_resize_visitor()
198 virtual ir_visitor_status
visit(ir_variable
*var
)
200 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
201 return visit_continue
;
203 unsigned size
= var
->type
->length
;
205 /* Generate a link error if the shader has declared this array with an
208 if (size
&& size
!= this->num_vertices
) {
209 linker_error(this->prog
, "size of array %s declared as %u, "
210 "but number of input vertices is %u\n",
211 var
->name
, size
, this->num_vertices
);
212 return visit_continue
;
215 /* Generate a link error if the shader attempts to access an input
216 * array using an index too large for its actual size assigned at link
219 if (var
->data
.max_array_access
>= this->num_vertices
) {
220 linker_error(this->prog
, "geometry shader accesses element %i of "
221 "%s, but only %i input vertices\n",
222 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
223 return visit_continue
;
226 var
->type
= glsl_type::get_array_instance(var
->type
->element_type(),
228 var
->data
.max_array_access
= this->num_vertices
- 1;
230 return visit_continue
;
233 /* Dereferences of input variables need to be updated so that their type
234 * matches the newly assigned type of the variable they are accessing. */
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 /* Dereferences of 2D input arrays need to be updated so that their type
242 * matches the newly assigned type of the array they are accessing. */
243 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
245 const glsl_type
*const vt
= ir
->array
->type
;
247 ir
->type
= vt
->element_type();
248 return visit_continue
;
253 * Visitor that determines the highest stream id to which a (geometry) shader
254 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
256 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
258 find_emit_vertex_visitor(int max_allowed
)
259 : max_stream_allowed(max_allowed
),
260 invalid_stream_id(0),
261 invalid_stream_id_from_emit_vertex(false),
262 end_primitive_found(false),
263 uses_non_zero_stream(false)
268 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
270 int stream_id
= ir
->stream_id();
273 invalid_stream_id
= stream_id
;
274 invalid_stream_id_from_emit_vertex
= true;
278 if (stream_id
> max_stream_allowed
) {
279 invalid_stream_id
= stream_id
;
280 invalid_stream_id_from_emit_vertex
= true;
285 uses_non_zero_stream
= true;
287 return visit_continue
;
290 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
292 end_primitive_found
= true;
294 int stream_id
= ir
->stream_id();
297 invalid_stream_id
= stream_id
;
298 invalid_stream_id_from_emit_vertex
= false;
302 if (stream_id
> max_stream_allowed
) {
303 invalid_stream_id
= stream_id
;
304 invalid_stream_id_from_emit_vertex
= false;
309 uses_non_zero_stream
= true;
311 return visit_continue
;
316 return invalid_stream_id
!= 0;
319 const char *error_func()
321 return invalid_stream_id_from_emit_vertex
?
322 "EmitStreamVertex" : "EndStreamPrimitive";
327 return invalid_stream_id
;
332 return uses_non_zero_stream
;
335 bool uses_end_primitive()
337 return end_primitive_found
;
341 int max_stream_allowed
;
342 int invalid_stream_id
;
343 bool invalid_stream_id_from_emit_vertex
;
344 bool end_primitive_found
;
345 bool uses_non_zero_stream
;
348 } /* anonymous namespace */
351 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
355 ralloc_strcat(&prog
->InfoLog
, "error: ");
357 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
360 prog
->LinkStatus
= false;
365 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
369 ralloc_strcat(&prog
->InfoLog
, "warning: ");
371 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
378 * Given a string identifying a program resource, break it into a base name
379 * and an optional array index in square brackets.
381 * If an array index is present, \c out_base_name_end is set to point to the
382 * "[" that precedes the array index, and the array index itself is returned
385 * If no array index is present (or if the array index is negative or
386 * mal-formed), \c out_base_name_end, is set to point to the null terminator
387 * at the end of the input string, and -1 is returned.
389 * Only the final array index is parsed; if the string contains other array
390 * indices (or structure field accesses), they are left in the base name.
392 * No attempt is made to check that the base name is properly formed;
393 * typically the caller will look up the base name in a hash table, so
394 * ill-formed base names simply turn into hash table lookup failures.
397 parse_program_resource_name(const GLchar
*name
,
398 const GLchar
**out_base_name_end
)
400 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
402 * "When an integer array element or block instance number is part of
403 * the name string, it will be specified in decimal form without a "+"
404 * or "-" sign or any extra leading zeroes. Additionally, the name
405 * string will not include white space anywhere in the string."
408 const size_t len
= strlen(name
);
409 *out_base_name_end
= name
+ len
;
411 if (len
== 0 || name
[len
-1] != ']')
414 /* Walk backwards over the string looking for a non-digit character. This
415 * had better be the opening bracket for an array index.
417 * Initially, i specifies the location of the ']'. Since the string may
418 * contain only the ']' charcater, walk backwards very carefully.
421 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
424 if ((i
== 0) || name
[i
-1] != '[')
427 long array_index
= strtol(&name
[i
], NULL
, 10);
431 *out_base_name_end
= name
+ (i
- 1);
437 link_invalidate_variable_locations(exec_list
*ir
)
439 foreach_in_list(ir_instruction
, node
, ir
) {
440 ir_variable
*const var
= node
->as_variable();
445 /* Only assign locations for variables that lack an explicit location.
446 * Explicit locations are set for all built-in variables, generic vertex
447 * shader inputs (via layout(location=...)), and generic fragment shader
448 * outputs (also via layout(location=...)).
450 if (!var
->data
.explicit_location
) {
451 var
->data
.location
= -1;
452 var
->data
.location_frac
= 0;
455 /* ir_variable::is_unmatched_generic_inout is used by the linker while
456 * connecting outputs from one stage to inputs of the next stage.
458 * There are two implicit assumptions here. First, we assume that any
459 * built-in variable (i.e., non-generic in or out) will have
460 * explicit_location set. Second, we assume that any generic in or out
461 * will not have explicit_location set.
463 * This second assumption will only be valid until
464 * GL_ARB_separate_shader_objects is supported. When that extension is
465 * implemented, this function will need some modifications.
467 if (!var
->data
.explicit_location
) {
468 var
->data
.is_unmatched_generic_inout
= 1;
470 var
->data
.is_unmatched_generic_inout
= 0;
477 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
479 * Also check for errors based on incorrect usage of gl_ClipVertex and
482 * Return false if an error was reported.
485 analyze_clip_usage(struct gl_shader_program
*prog
,
486 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
487 GLuint
*ClipDistanceArraySize
)
489 *ClipDistanceArraySize
= 0;
491 if (!prog
->IsES
&& prog
->Version
>= 130) {
492 /* From section 7.1 (Vertex Shader Special Variables) of the
495 * "It is an error for a shader to statically write both
496 * gl_ClipVertex and gl_ClipDistance."
498 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
499 * gl_ClipVertex nor gl_ClipDistance.
501 find_assignment_visitor
clip_vertex("gl_ClipVertex");
502 find_assignment_visitor
clip_distance("gl_ClipDistance");
504 clip_vertex
.run(shader
->ir
);
505 clip_distance
.run(shader
->ir
);
506 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
507 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
508 "and `gl_ClipDistance'\n",
509 _mesa_shader_stage_to_string(shader
->Stage
));
512 *UsesClipDistance
= clip_distance
.variable_found();
513 ir_variable
*clip_distance_var
=
514 shader
->symbols
->get_variable("gl_ClipDistance");
515 if (clip_distance_var
)
516 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
518 *UsesClipDistance
= false;
524 * Verify that a vertex shader executable meets all semantic requirements.
526 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
529 * \param shader Vertex shader executable to be verified
532 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
533 struct gl_shader
*shader
)
538 /* From the GLSL 1.10 spec, page 48:
540 * "The variable gl_Position is available only in the vertex
541 * language and is intended for writing the homogeneous vertex
542 * position. All executions of a well-formed vertex shader
543 * executable must write a value into this variable. [...] The
544 * variable gl_Position is available only in the vertex
545 * language and is intended for writing the homogeneous vertex
546 * position. All executions of a well-formed vertex shader
547 * executable must write a value into this variable."
549 * while in GLSL 1.40 this text is changed to:
551 * "The variable gl_Position is available only in the vertex
552 * language and is intended for writing the homogeneous vertex
553 * position. It can be written at any time during shader
554 * execution. It may also be read back by a vertex shader
555 * after being written. This value will be used by primitive
556 * assembly, clipping, culling, and other fixed functionality
557 * operations, if present, that operate on primitives after
558 * vertex processing has occurred. Its value is undefined if
559 * the vertex shader executable does not write gl_Position."
561 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
562 * gl_Position is not an error.
564 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
565 find_assignment_visitor
find("gl_Position");
566 find
.run(shader
->ir
);
567 if (!find
.variable_found()) {
570 "vertex shader does not write to `gl_Position'."
571 "It's value is undefined. \n");
574 "vertex shader does not write to `gl_Position'. \n");
580 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
581 &prog
->Vert
.ClipDistanceArraySize
);
586 * Verify that a fragment shader executable meets all semantic requirements
588 * \param shader Fragment shader executable to be verified
591 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
592 struct gl_shader
*shader
)
597 find_assignment_visitor
frag_color("gl_FragColor");
598 find_assignment_visitor
frag_data("gl_FragData");
600 frag_color
.run(shader
->ir
);
601 frag_data
.run(shader
->ir
);
603 if (frag_color
.variable_found() && frag_data
.variable_found()) {
604 linker_error(prog
, "fragment shader writes to both "
605 "`gl_FragColor' and `gl_FragData'\n");
610 * Verify that a geometry shader executable meets all semantic requirements
612 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
613 * prog->Geom.ClipDistanceArraySize as a side effect.
615 * \param shader Geometry shader executable to be verified
618 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
619 struct gl_shader
*shader
)
624 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
625 prog
->Geom
.VerticesIn
= num_vertices
;
627 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
628 &prog
->Geom
.ClipDistanceArraySize
);
632 * Check if geometry shaders emit to non-zero streams and do corresponding
636 validate_geometry_shader_emissions(struct gl_context
*ctx
,
637 struct gl_shader_program
*prog
)
639 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
640 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
641 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
642 if (emit_vertex
.error()) {
643 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
644 "stream parameter are in the range [0, %d].\n",
645 emit_vertex
.error_func(),
646 emit_vertex
.error_stream(),
647 ctx
->Const
.MaxVertexStreams
- 1);
649 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
650 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
652 /* From the ARB_gpu_shader5 spec:
654 * "Multiple vertex streams are supported only if the output primitive
655 * type is declared to be "points". A program will fail to link if it
656 * contains a geometry shader calling EmitStreamVertex() or
657 * EndStreamPrimitive() if its output primitive type is not "points".
659 * However, in the same spec:
661 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
662 * with <stream> set to zero."
666 * "The function EndPrimitive() is equivalent to calling
667 * EndStreamPrimitive() with <stream> set to zero."
669 * Since we can call EmitVertex() and EndPrimitive() when we output
670 * primitives other than points, calling EmitStreamVertex(0) or
671 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
672 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
673 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
676 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
677 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
678 "with n>0 requires point output\n");
685 * Perform validation of global variables used across multiple shaders
688 cross_validate_globals(struct gl_shader_program
*prog
,
689 struct gl_shader
**shader_list
,
690 unsigned num_shaders
,
693 /* Examine all of the uniforms in all of the shaders and cross validate
696 glsl_symbol_table variables
;
697 for (unsigned i
= 0; i
< num_shaders
; i
++) {
698 if (shader_list
[i
] == NULL
)
701 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
702 ir_variable
*const var
= node
->as_variable();
707 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
))
710 /* Don't cross validate temporaries that are at global scope. These
711 * will eventually get pulled into the shaders 'main'.
713 if (var
->data
.mode
== ir_var_temporary
)
716 /* If a global with this name has already been seen, verify that the
717 * new instance has the same type. In addition, if the globals have
718 * initializers, the values of the initializers must be the same.
720 ir_variable
*const existing
= variables
.get_variable(var
->name
);
721 if (existing
!= NULL
) {
722 if (var
->type
!= existing
->type
) {
723 /* Consider the types to be "the same" if both types are arrays
724 * of the same type and one of the arrays is implicitly sized.
725 * In addition, set the type of the linked variable to the
726 * explicitly sized array.
728 if (var
->type
->is_array()
729 && existing
->type
->is_array()
730 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
731 && ((var
->type
->length
== 0)
732 || (existing
->type
->length
== 0))) {
733 if (var
->type
->length
!= 0) {
734 if (var
->type
->length
<= existing
->data
.max_array_access
) {
735 linker_error(prog
, "%s `%s' declared as type "
736 "`%s' but outermost dimension has an index"
739 var
->name
, var
->type
->name
,
740 existing
->data
.max_array_access
);
743 existing
->type
= var
->type
;
744 } else if (existing
->type
->length
!= 0
745 && existing
->type
->length
<=
746 var
->data
.max_array_access
) {
747 linker_error(prog
, "%s `%s' declared as type "
748 "`%s' but outermost dimension has an index"
751 var
->name
, existing
->type
->name
,
752 var
->data
.max_array_access
);
755 } else if (var
->type
->is_record()
756 && existing
->type
->is_record()
757 && existing
->type
->record_compare(var
->type
)) {
758 existing
->type
= var
->type
;
760 linker_error(prog
, "%s `%s' declared as type "
761 "`%s' and type `%s'\n",
763 var
->name
, var
->type
->name
,
764 existing
->type
->name
);
769 if (var
->data
.explicit_location
) {
770 if (existing
->data
.explicit_location
771 && (var
->data
.location
!= existing
->data
.location
)) {
772 linker_error(prog
, "explicit locations for %s "
773 "`%s' have differing values\n",
774 mode_string(var
), var
->name
);
778 existing
->data
.location
= var
->data
.location
;
779 existing
->data
.explicit_location
= true;
782 /* From the GLSL 4.20 specification:
783 * "A link error will result if two compilation units in a program
784 * specify different integer-constant bindings for the same
785 * opaque-uniform name. However, it is not an error to specify a
786 * binding on some but not all declarations for the same name"
788 if (var
->data
.explicit_binding
) {
789 if (existing
->data
.explicit_binding
&&
790 var
->data
.binding
!= existing
->data
.binding
) {
791 linker_error(prog
, "explicit bindings for %s "
792 "`%s' have differing values\n",
793 mode_string(var
), var
->name
);
797 existing
->data
.binding
= var
->data
.binding
;
798 existing
->data
.explicit_binding
= true;
801 if (var
->type
->contains_atomic() &&
802 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
803 linker_error(prog
, "offset specifications for %s "
804 "`%s' have differing values\n",
805 mode_string(var
), var
->name
);
809 /* Validate layout qualifiers for gl_FragDepth.
811 * From the AMD/ARB_conservative_depth specs:
813 * "If gl_FragDepth is redeclared in any fragment shader in a
814 * program, it must be redeclared in all fragment shaders in
815 * that program that have static assignments to
816 * gl_FragDepth. All redeclarations of gl_FragDepth in all
817 * fragment shaders in a single program must have the same set
820 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
821 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
822 bool layout_differs
=
823 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
825 if (layout_declared
&& layout_differs
) {
827 "All redeclarations of gl_FragDepth in all "
828 "fragment shaders in a single program must have "
829 "the same set of qualifiers.\n");
832 if (var
->data
.used
&& layout_differs
) {
834 "If gl_FragDepth is redeclared with a layout "
835 "qualifier in any fragment shader, it must be "
836 "redeclared with the same layout qualifier in "
837 "all fragment shaders that have assignments to "
842 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
844 * "If a shared global has multiple initializers, the
845 * initializers must all be constant expressions, and they
846 * must all have the same value. Otherwise, a link error will
847 * result. (A shared global having only one initializer does
848 * not require that initializer to be a constant expression.)"
850 * Previous to 4.20 the GLSL spec simply said that initializers
851 * must have the same value. In this case of non-constant
852 * initializers, this was impossible to determine. As a result,
853 * no vendor actually implemented that behavior. The 4.20
854 * behavior matches the implemented behavior of at least one other
855 * vendor, so we'll implement that for all GLSL versions.
857 if (var
->constant_initializer
!= NULL
) {
858 if (existing
->constant_initializer
!= NULL
) {
859 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
860 linker_error(prog
, "initializers for %s "
861 "`%s' have differing values\n",
862 mode_string(var
), var
->name
);
866 /* If the first-seen instance of a particular uniform did not
867 * have an initializer but a later instance does, copy the
868 * initializer to the version stored in the symbol table.
870 /* FINISHME: This is wrong. The constant_value field should
871 * FINISHME: not be modified! Imagine a case where a shader
872 * FINISHME: without an initializer is linked in two different
873 * FINISHME: programs with shaders that have differing
874 * FINISHME: initializers. Linking with the first will
875 * FINISHME: modify the shader, and linking with the second
876 * FINISHME: will fail.
878 existing
->constant_initializer
=
879 var
->constant_initializer
->clone(ralloc_parent(existing
),
884 if (var
->data
.has_initializer
) {
885 if (existing
->data
.has_initializer
886 && (var
->constant_initializer
== NULL
887 || existing
->constant_initializer
== NULL
)) {
889 "shared global variable `%s' has multiple "
890 "non-constant initializers.\n",
895 /* Some instance had an initializer, so keep track of that. In
896 * this location, all sorts of initializers (constant or
897 * otherwise) will propagate the existence to the variable
898 * stored in the symbol table.
900 existing
->data
.has_initializer
= true;
903 if (existing
->data
.invariant
!= var
->data
.invariant
) {
904 linker_error(prog
, "declarations for %s `%s' have "
905 "mismatching invariant qualifiers\n",
906 mode_string(var
), var
->name
);
909 if (existing
->data
.centroid
!= var
->data
.centroid
) {
910 linker_error(prog
, "declarations for %s `%s' have "
911 "mismatching centroid qualifiers\n",
912 mode_string(var
), var
->name
);
915 if (existing
->data
.sample
!= var
->data
.sample
) {
916 linker_error(prog
, "declarations for %s `%s` have "
917 "mismatching sample qualifiers\n",
918 mode_string(var
), var
->name
);
922 variables
.add_variable(var
);
929 * Perform validation of uniforms used across multiple shader stages
932 cross_validate_uniforms(struct gl_shader_program
*prog
)
934 cross_validate_globals(prog
, prog
->_LinkedShaders
,
935 MESA_SHADER_STAGES
, true);
939 * Accumulates the array of prog->UniformBlocks and checks that all
940 * definitons of blocks agree on their contents.
943 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
945 unsigned max_num_uniform_blocks
= 0;
946 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
947 if (prog
->_LinkedShaders
[i
])
948 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
951 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
952 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
954 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
955 max_num_uniform_blocks
);
956 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
957 prog
->UniformBlockStageIndex
[i
][j
] = -1;
962 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
963 int index
= link_cross_validate_uniform_block(prog
,
964 &prog
->UniformBlocks
,
965 &prog
->NumUniformBlocks
,
966 &sh
->UniformBlocks
[j
]);
969 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
970 sh
->UniformBlocks
[j
].Name
);
974 prog
->UniformBlockStageIndex
[i
][index
] = j
;
983 * Populates a shaders symbol table with all global declarations
986 populate_symbol_table(gl_shader
*sh
)
988 sh
->symbols
= new(sh
) glsl_symbol_table
;
990 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
994 if ((func
= inst
->as_function()) != NULL
) {
995 sh
->symbols
->add_function(func
);
996 } else if ((var
= inst
->as_variable()) != NULL
) {
997 if (var
->data
.mode
!= ir_var_temporary
)
998 sh
->symbols
->add_variable(var
);
1005 * Remap variables referenced in an instruction tree
1007 * This is used when instruction trees are cloned from one shader and placed in
1008 * another. These trees will contain references to \c ir_variable nodes that
1009 * do not exist in the target shader. This function finds these \c ir_variable
1010 * references and replaces the references with matching variables in the target
1013 * If there is no matching variable in the target shader, a clone of the
1014 * \c ir_variable is made and added to the target shader. The new variable is
1015 * added to \b both the instruction stream and the symbol table.
1017 * \param inst IR tree that is to be processed.
1018 * \param symbols Symbol table containing global scope symbols in the
1020 * \param instructions Instruction stream where new variable declarations
1024 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1027 class remap_visitor
: public ir_hierarchical_visitor
{
1029 remap_visitor(struct gl_shader
*target
,
1032 this->target
= target
;
1033 this->symbols
= target
->symbols
;
1034 this->instructions
= target
->ir
;
1035 this->temps
= temps
;
1038 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1040 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1041 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1043 assert(var
!= NULL
);
1045 return visit_continue
;
1048 ir_variable
*const existing
=
1049 this->symbols
->get_variable(ir
->var
->name
);
1050 if (existing
!= NULL
)
1053 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1055 this->symbols
->add_variable(copy
);
1056 this->instructions
->push_head(copy
);
1060 return visit_continue
;
1064 struct gl_shader
*target
;
1065 glsl_symbol_table
*symbols
;
1066 exec_list
*instructions
;
1070 remap_visitor
v(target
, temps
);
1077 * Move non-declarations from one instruction stream to another
1079 * The intended usage pattern of this function is to pass the pointer to the
1080 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1081 * pointer) for \c last and \c false for \c make_copies on the first
1082 * call. Successive calls pass the return value of the previous call for
1083 * \c last and \c true for \c make_copies.
1085 * \param instructions Source instruction stream
1086 * \param last Instruction after which new instructions should be
1087 * inserted in the target instruction stream
1088 * \param make_copies Flag selecting whether instructions in \c instructions
1089 * should be copied (via \c ir_instruction::clone) into the
1090 * target list or moved.
1093 * The new "last" instruction in the target instruction stream. This pointer
1094 * is suitable for use as the \c last parameter of a later call to this
1098 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1099 bool make_copies
, gl_shader
*target
)
1101 hash_table
*temps
= NULL
;
1104 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1105 hash_table_pointer_compare
);
1107 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1108 if (inst
->as_function())
1111 ir_variable
*var
= inst
->as_variable();
1112 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1115 assert(inst
->as_assignment()
1117 || inst
->as_if() /* for initializers with the ?: operator */
1118 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1121 inst
= inst
->clone(target
, NULL
);
1124 hash_table_insert(temps
, inst
, var
);
1126 remap_variables(inst
, target
, temps
);
1131 last
->insert_after(inst
);
1136 hash_table_dtor(temps
);
1142 * Get the function signature for main from a shader
1144 ir_function_signature
*
1145 link_get_main_function_signature(gl_shader
*sh
)
1147 ir_function
*const f
= sh
->symbols
->get_function("main");
1149 exec_list void_parameters
;
1151 /* Look for the 'void main()' signature and ensure that it's defined.
1152 * This keeps the linker from accidentally pick a shader that just
1153 * contains a prototype for main.
1155 * We don't have to check for multiple definitions of main (in multiple
1156 * shaders) because that would have already been caught above.
1158 ir_function_signature
*sig
=
1159 f
->matching_signature(NULL
, &void_parameters
, false);
1160 if ((sig
!= NULL
) && sig
->is_defined
) {
1170 * This class is only used in link_intrastage_shaders() below but declaring
1171 * it inside that function leads to compiler warnings with some versions of
1174 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1176 array_sizing_visitor()
1177 : mem_ctx(ralloc_context(NULL
)),
1178 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1179 hash_table_pointer_compare
))
1183 ~array_sizing_visitor()
1185 hash_table_dtor(this->unnamed_interfaces
);
1186 ralloc_free(this->mem_ctx
);
1189 virtual ir_visitor_status
visit(ir_variable
*var
)
1191 fixup_type(&var
->type
, var
->data
.max_array_access
);
1192 if (var
->type
->is_interface()) {
1193 if (interface_contains_unsized_arrays(var
->type
)) {
1194 const glsl_type
*new_type
=
1195 resize_interface_members(var
->type
,
1196 var
->get_max_ifc_array_access());
1197 var
->type
= new_type
;
1198 var
->change_interface_type(new_type
);
1200 } else if (var
->type
->is_array() &&
1201 var
->type
->fields
.array
->is_interface()) {
1202 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1203 const glsl_type
*new_type
=
1204 resize_interface_members(var
->type
->fields
.array
,
1205 var
->get_max_ifc_array_access());
1206 var
->change_interface_type(new_type
);
1208 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1210 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1211 /* Store a pointer to the variable in the unnamed_interfaces
1214 ir_variable
**interface_vars
= (ir_variable
**)
1215 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1216 if (interface_vars
== NULL
) {
1217 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1219 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1222 unsigned index
= ifc_type
->field_index(var
->name
);
1223 assert(index
< ifc_type
->length
);
1224 assert(interface_vars
[index
] == NULL
);
1225 interface_vars
[index
] = var
;
1227 return visit_continue
;
1231 * For each unnamed interface block that was discovered while running the
1232 * visitor, adjust the interface type to reflect the newly assigned array
1233 * sizes, and fix up the ir_variable nodes to point to the new interface
1236 void fixup_unnamed_interface_types()
1238 hash_table_call_foreach(this->unnamed_interfaces
,
1239 fixup_unnamed_interface_type
, NULL
);
1244 * If the type pointed to by \c type represents an unsized array, replace
1245 * it with a sized array whose size is determined by max_array_access.
1247 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1249 if ((*type
)->is_unsized_array()) {
1250 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1251 max_array_access
+ 1);
1252 assert(*type
!= NULL
);
1257 * Determine whether the given interface type contains unsized arrays (if
1258 * it doesn't, array_sizing_visitor doesn't need to process it).
1260 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1262 for (unsigned i
= 0; i
< type
->length
; i
++) {
1263 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1264 if (elem_type
->is_unsized_array())
1271 * Create a new interface type based on the given type, with unsized arrays
1272 * replaced by sized arrays whose size is determined by
1273 * max_ifc_array_access.
1275 static const glsl_type
*
1276 resize_interface_members(const glsl_type
*type
,
1277 const unsigned *max_ifc_array_access
)
1279 unsigned num_fields
= type
->length
;
1280 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1281 memcpy(fields
, type
->fields
.structure
,
1282 num_fields
* sizeof(*fields
));
1283 for (unsigned i
= 0; i
< num_fields
; i
++) {
1284 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1286 glsl_interface_packing packing
=
1287 (glsl_interface_packing
) type
->interface_packing
;
1288 const glsl_type
*new_ifc_type
=
1289 glsl_type::get_interface_instance(fields
, num_fields
,
1290 packing
, type
->name
);
1292 return new_ifc_type
;
1295 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1298 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1299 ir_variable
**interface_vars
= (ir_variable
**) data
;
1300 unsigned num_fields
= ifc_type
->length
;
1301 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1302 memcpy(fields
, ifc_type
->fields
.structure
,
1303 num_fields
* sizeof(*fields
));
1304 bool interface_type_changed
= false;
1305 for (unsigned i
= 0; i
< num_fields
; i
++) {
1306 if (interface_vars
[i
] != NULL
&&
1307 fields
[i
].type
!= interface_vars
[i
]->type
) {
1308 fields
[i
].type
= interface_vars
[i
]->type
;
1309 interface_type_changed
= true;
1312 if (!interface_type_changed
) {
1316 glsl_interface_packing packing
=
1317 (glsl_interface_packing
) ifc_type
->interface_packing
;
1318 const glsl_type
*new_ifc_type
=
1319 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1322 for (unsigned i
= 0; i
< num_fields
; i
++) {
1323 if (interface_vars
[i
] != NULL
)
1324 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1329 * Memory context used to allocate the data in \c unnamed_interfaces.
1334 * Hash table from const glsl_type * to an array of ir_variable *'s
1335 * pointing to the ir_variables constituting each unnamed interface block.
1337 hash_table
*unnamed_interfaces
;
1341 * Performs the cross-validation of layout qualifiers specified in
1342 * redeclaration of gl_FragCoord for the attached fragment shaders,
1343 * and propagates them to the linked FS and linked shader program.
1346 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1347 struct gl_shader
*linked_shader
,
1348 struct gl_shader
**shader_list
,
1349 unsigned num_shaders
)
1351 linked_shader
->redeclares_gl_fragcoord
= false;
1352 linked_shader
->uses_gl_fragcoord
= false;
1353 linked_shader
->origin_upper_left
= false;
1354 linked_shader
->pixel_center_integer
= false;
1356 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1357 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1360 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1361 struct gl_shader
*shader
= shader_list
[i
];
1362 /* From the GLSL 1.50 spec, page 39:
1364 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1365 * it must be redeclared in all the fragment shaders in that program
1366 * that have a static use gl_FragCoord."
1368 * Exclude the case when one of the 'linked_shader' or 'shader' redeclares
1369 * gl_FragCoord with no layout qualifiers but the other one doesn't
1370 * redeclare it. If we strictly follow GLSL 1.50 spec's language, it
1371 * should be a link error. But, generating link error for this case will
1372 * be a wrong behaviour which spec didn't intend to do and it could also
1373 * break some applications.
1375 if ((linked_shader
->redeclares_gl_fragcoord
1376 && !shader
->redeclares_gl_fragcoord
1377 && shader
->uses_gl_fragcoord
1378 && (linked_shader
->origin_upper_left
1379 || linked_shader
->pixel_center_integer
))
1380 || (shader
->redeclares_gl_fragcoord
1381 && !linked_shader
->redeclares_gl_fragcoord
1382 && linked_shader
->uses_gl_fragcoord
1383 && (shader
->origin_upper_left
1384 || shader
->pixel_center_integer
))) {
1385 linker_error(prog
, "fragment shader defined with conflicting "
1386 "layout qualifiers for gl_FragCoord\n");
1389 /* From the GLSL 1.50 spec, page 39:
1391 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1392 * single program must have the same set of qualifiers."
1394 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1395 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1396 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1397 linker_error(prog
, "fragment shader defined with conflicting "
1398 "layout qualifiers for gl_FragCoord\n");
1401 /* Update the linked shader state. Note that uses_gl_fragcoord should
1402 * accumulate the results. The other values should replace. If there
1403 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1404 * are already known to be the same.
1406 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1407 linked_shader
->redeclares_gl_fragcoord
=
1408 shader
->redeclares_gl_fragcoord
;
1409 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1410 || shader
->uses_gl_fragcoord
;
1411 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1412 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1418 * Performs the cross-validation of geometry shader max_vertices and
1419 * primitive type layout qualifiers for the attached geometry shaders,
1420 * and propagates them to the linked GS and linked shader program.
1423 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1424 struct gl_shader
*linked_shader
,
1425 struct gl_shader
**shader_list
,
1426 unsigned num_shaders
)
1428 linked_shader
->Geom
.VerticesOut
= 0;
1429 linked_shader
->Geom
.Invocations
= 0;
1430 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1431 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1433 /* No in/out qualifiers defined for anything but GLSL 1.50+
1434 * geometry shaders so far.
1436 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1439 /* From the GLSL 1.50 spec, page 46:
1441 * "All geometry shader output layout declarations in a program
1442 * must declare the same layout and same value for
1443 * max_vertices. There must be at least one geometry output
1444 * layout declaration somewhere in a program, but not all
1445 * geometry shaders (compilation units) are required to
1449 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1450 struct gl_shader
*shader
= shader_list
[i
];
1452 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1453 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1454 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1455 linker_error(prog
, "geometry shader defined with conflicting "
1459 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1462 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1463 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1464 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1465 linker_error(prog
, "geometry shader defined with conflicting "
1469 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1472 if (shader
->Geom
.VerticesOut
!= 0) {
1473 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1474 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1475 linker_error(prog
, "geometry shader defined with conflicting "
1476 "output vertex count (%d and %d)\n",
1477 linked_shader
->Geom
.VerticesOut
,
1478 shader
->Geom
.VerticesOut
);
1481 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1484 if (shader
->Geom
.Invocations
!= 0) {
1485 if (linked_shader
->Geom
.Invocations
!= 0 &&
1486 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1487 linker_error(prog
, "geometry shader defined with conflicting "
1488 "invocation count (%d and %d)\n",
1489 linked_shader
->Geom
.Invocations
,
1490 shader
->Geom
.Invocations
);
1493 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1497 /* Just do the intrastage -> interstage propagation right now,
1498 * since we already know we're in the right type of shader program
1501 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1503 "geometry shader didn't declare primitive input type\n");
1506 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1508 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1510 "geometry shader didn't declare primitive output type\n");
1513 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1515 if (linked_shader
->Geom
.VerticesOut
== 0) {
1517 "geometry shader didn't declare max_vertices\n");
1520 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1522 if (linked_shader
->Geom
.Invocations
== 0)
1523 linked_shader
->Geom
.Invocations
= 1;
1525 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1530 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1531 * qualifiers for the attached compute shaders, and propagate them to the
1532 * linked CS and linked shader program.
1535 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1536 struct gl_shader
*linked_shader
,
1537 struct gl_shader
**shader_list
,
1538 unsigned num_shaders
)
1540 for (int i
= 0; i
< 3; i
++)
1541 linked_shader
->Comp
.LocalSize
[i
] = 0;
1543 /* This function is called for all shader stages, but it only has an effect
1544 * for compute shaders.
1546 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1549 /* From the ARB_compute_shader spec, in the section describing local size
1552 * If multiple compute shaders attached to a single program object
1553 * declare local work-group size, the declarations must be identical;
1554 * otherwise a link-time error results. Furthermore, if a program
1555 * object contains any compute shaders, at least one must contain an
1556 * input layout qualifier specifying the local work sizes of the
1557 * program, or a link-time error will occur.
1559 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1560 struct gl_shader
*shader
= shader_list
[sh
];
1562 if (shader
->Comp
.LocalSize
[0] != 0) {
1563 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1564 for (int i
= 0; i
< 3; i
++) {
1565 if (linked_shader
->Comp
.LocalSize
[i
] !=
1566 shader
->Comp
.LocalSize
[i
]) {
1567 linker_error(prog
, "compute shader defined with conflicting "
1573 for (int i
= 0; i
< 3; i
++)
1574 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1578 /* Just do the intrastage -> interstage propagation right now,
1579 * since we already know we're in the right type of shader program
1582 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1583 linker_error(prog
, "compute shader didn't declare local size\n");
1586 for (int i
= 0; i
< 3; i
++)
1587 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1592 * Combine a group of shaders for a single stage to generate a linked shader
1595 * If this function is supplied a single shader, it is cloned, and the new
1596 * shader is returned.
1598 static struct gl_shader
*
1599 link_intrastage_shaders(void *mem_ctx
,
1600 struct gl_context
*ctx
,
1601 struct gl_shader_program
*prog
,
1602 struct gl_shader
**shader_list
,
1603 unsigned num_shaders
)
1605 struct gl_uniform_block
*uniform_blocks
= NULL
;
1607 /* Check that global variables defined in multiple shaders are consistent.
1609 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1610 if (!prog
->LinkStatus
)
1613 /* Check that interface blocks defined in multiple shaders are consistent.
1615 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1617 if (!prog
->LinkStatus
)
1620 /* Link up uniform blocks defined within this stage. */
1621 const unsigned num_uniform_blocks
=
1622 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1624 if (!prog
->LinkStatus
)
1627 /* Check that there is only a single definition of each function signature
1628 * across all shaders.
1630 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1631 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1632 ir_function
*const f
= node
->as_function();
1637 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1638 ir_function
*const other
=
1639 shader_list
[j
]->symbols
->get_function(f
->name
);
1641 /* If the other shader has no function (and therefore no function
1642 * signatures) with the same name, skip to the next shader.
1647 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1648 if (!sig
->is_defined
|| sig
->is_builtin())
1651 ir_function_signature
*other_sig
=
1652 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1654 if ((other_sig
!= NULL
) && other_sig
->is_defined
1655 && !other_sig
->is_builtin()) {
1656 linker_error(prog
, "function `%s' is multiply defined\n",
1665 /* Find the shader that defines main, and make a clone of it.
1667 * Starting with the clone, search for undefined references. If one is
1668 * found, find the shader that defines it. Clone the reference and add
1669 * it to the shader. Repeat until there are no undefined references or
1670 * until a reference cannot be resolved.
1672 gl_shader
*main
= NULL
;
1673 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1674 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1675 main
= shader_list
[i
];
1681 linker_error(prog
, "%s shader lacks `main'\n",
1682 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1686 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1687 linked
->ir
= new(linked
) exec_list
;
1688 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1690 linked
->UniformBlocks
= uniform_blocks
;
1691 linked
->NumUniformBlocks
= num_uniform_blocks
;
1692 ralloc_steal(linked
, linked
->UniformBlocks
);
1694 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1695 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1696 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1698 populate_symbol_table(linked
);
1700 /* The pointer to the main function in the final linked shader (i.e., the
1701 * copy of the original shader that contained the main function).
1703 ir_function_signature
*const main_sig
=
1704 link_get_main_function_signature(linked
);
1706 /* Move any instructions other than variable declarations or function
1707 * declarations into main.
1709 exec_node
*insertion_point
=
1710 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1713 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1714 if (shader_list
[i
] == main
)
1717 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1718 insertion_point
, true, linked
);
1721 /* Check if any shader needs built-in functions. */
1722 bool need_builtins
= false;
1723 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1724 if (shader_list
[i
]->uses_builtin_functions
) {
1725 need_builtins
= true;
1731 if (need_builtins
) {
1732 /* Make a temporary array one larger than shader_list, which will hold
1733 * the built-in function shader as well.
1735 gl_shader
**linking_shaders
= (gl_shader
**)
1736 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1738 ok
= linking_shaders
!= NULL
;
1741 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1742 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1744 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1746 free(linking_shaders
);
1748 _mesa_error_no_memory(__func__
);
1751 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1756 ctx
->Driver
.DeleteShader(ctx
, linked
);
1760 /* At this point linked should contain all of the linked IR, so
1761 * validate it to make sure nothing went wrong.
1763 validate_ir_tree(linked
->ir
);
1765 /* Set the size of geometry shader input arrays */
1766 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1767 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1768 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1769 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
1770 ir
->accept(&input_resize_visitor
);
1774 if (ctx
->Const
.VertexID_is_zero_based
)
1775 lower_vertex_id(linked
);
1777 /* Make a pass over all variable declarations to ensure that arrays with
1778 * unspecified sizes have a size specified. The size is inferred from the
1779 * max_array_access field.
1781 array_sizing_visitor v
;
1783 v
.fixup_unnamed_interface_types();
1789 * Update the sizes of linked shader uniform arrays to the maximum
1792 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1794 * If one or more elements of an array are active,
1795 * GetActiveUniform will return the name of the array in name,
1796 * subject to the restrictions listed above. The type of the array
1797 * is returned in type. The size parameter contains the highest
1798 * array element index used, plus one. The compiler or linker
1799 * determines the highest index used. There will be only one
1800 * active uniform reported by the GL per uniform array.
1804 update_array_sizes(struct gl_shader_program
*prog
)
1806 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1807 if (prog
->_LinkedShaders
[i
] == NULL
)
1810 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
1811 ir_variable
*const var
= node
->as_variable();
1813 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1814 !var
->type
->is_array())
1817 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1818 * will not be eliminated. Since we always do std140, just
1819 * don't resize arrays in UBOs.
1821 * Atomic counters are supposed to get deterministic
1822 * locations assigned based on the declaration ordering and
1823 * sizes, array compaction would mess that up.
1825 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1828 unsigned int size
= var
->data
.max_array_access
;
1829 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1830 if (prog
->_LinkedShaders
[j
] == NULL
)
1833 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
1834 ir_variable
*other_var
= node2
->as_variable();
1838 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1839 other_var
->data
.max_array_access
> size
) {
1840 size
= other_var
->data
.max_array_access
;
1845 if (size
+ 1 != var
->type
->length
) {
1846 /* If this is a built-in uniform (i.e., it's backed by some
1847 * fixed-function state), adjust the number of state slots to
1848 * match the new array size. The number of slots per array entry
1849 * is not known. It seems safe to assume that the total number of
1850 * slots is an integer multiple of the number of array elements.
1851 * Determine the number of slots per array element by dividing by
1852 * the old (total) size.
1854 const unsigned num_slots
= var
->get_num_state_slots();
1855 if (num_slots
> 0) {
1856 var
->set_num_state_slots((size
+ 1)
1857 * (num_slots
/ var
->type
->length
));
1860 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1862 /* FINISHME: We should update the types of array
1863 * dereferences of this variable now.
1871 * Find a contiguous set of available bits in a bitmask.
1873 * \param used_mask Bits representing used (1) and unused (0) locations
1874 * \param needed_count Number of contiguous bits needed.
1877 * Base location of the available bits on success or -1 on failure.
1880 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1882 unsigned needed_mask
= (1 << needed_count
) - 1;
1883 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1885 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1886 * cannot optimize possibly infinite loops" for the loop below.
1888 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1891 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1892 if ((needed_mask
& ~used_mask
) == needed_mask
)
1903 * Assign locations for either VS inputs or FS outputs
1905 * \param prog Shader program whose variables need locations assigned
1906 * \param target_index Selector for the program target to receive location
1907 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1908 * \c MESA_SHADER_FRAGMENT.
1909 * \param max_index Maximum number of generic locations. This corresponds
1910 * to either the maximum number of draw buffers or the
1911 * maximum number of generic attributes.
1914 * If locations are successfully assigned, true is returned. Otherwise an
1915 * error is emitted to the shader link log and false is returned.
1918 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1919 unsigned target_index
,
1922 /* Mark invalid locations as being used.
1924 unsigned used_locations
= (max_index
>= 32)
1925 ? ~0 : ~((1 << max_index
) - 1);
1927 assert((target_index
== MESA_SHADER_VERTEX
)
1928 || (target_index
== MESA_SHADER_FRAGMENT
));
1930 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1934 /* Operate in a total of four passes.
1936 * 1. Invalidate the location assignments for all vertex shader inputs.
1938 * 2. Assign locations for inputs that have user-defined (via
1939 * glBindVertexAttribLocation) locations and outputs that have
1940 * user-defined locations (via glBindFragDataLocation).
1942 * 3. Sort the attributes without assigned locations by number of slots
1943 * required in decreasing order. Fragmentation caused by attribute
1944 * locations assigned by the application may prevent large attributes
1945 * from having enough contiguous space.
1947 * 4. Assign locations to any inputs without assigned locations.
1950 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1951 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1953 const enum ir_variable_mode direction
=
1954 (target_index
== MESA_SHADER_VERTEX
)
1955 ? ir_var_shader_in
: ir_var_shader_out
;
1958 /* Temporary storage for the set of attributes that need locations assigned.
1964 /* Used below in the call to qsort. */
1965 static int compare(const void *a
, const void *b
)
1967 const temp_attr
*const l
= (const temp_attr
*) a
;
1968 const temp_attr
*const r
= (const temp_attr
*) b
;
1970 /* Reversed because we want a descending order sort below. */
1971 return r
->slots
- l
->slots
;
1975 unsigned num_attr
= 0;
1977 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
1978 ir_variable
*const var
= node
->as_variable();
1980 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
1983 if (var
->data
.explicit_location
) {
1984 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
1985 || (var
->data
.location
< 0)) {
1987 "invalid explicit location %d specified for `%s'\n",
1988 (var
->data
.location
< 0)
1989 ? var
->data
.location
1990 : var
->data
.location
- generic_base
,
1994 } else if (target_index
== MESA_SHADER_VERTEX
) {
1997 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1998 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1999 var
->data
.location
= binding
;
2000 var
->data
.is_unmatched_generic_inout
= 0;
2002 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2006 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2007 assert(binding
>= FRAG_RESULT_DATA0
);
2008 var
->data
.location
= binding
;
2009 var
->data
.is_unmatched_generic_inout
= 0;
2011 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2012 var
->data
.index
= index
;
2017 /* If the variable is not a built-in and has a location statically
2018 * assigned in the shader (presumably via a layout qualifier), make sure
2019 * that it doesn't collide with other assigned locations. Otherwise,
2020 * add it to the list of variables that need linker-assigned locations.
2022 const unsigned slots
= var
->type
->count_attribute_slots();
2023 if (var
->data
.location
!= -1) {
2024 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2025 /* From page 61 of the OpenGL 4.0 spec:
2027 * "LinkProgram will fail if the attribute bindings assigned
2028 * by BindAttribLocation do not leave not enough space to
2029 * assign a location for an active matrix attribute or an
2030 * active attribute array, both of which require multiple
2031 * contiguous generic attributes."
2033 * I think above text prohibits the aliasing of explicit and
2034 * automatic assignments. But, aliasing is allowed in manual
2035 * assignments of attribute locations. See below comments for
2038 * From OpenGL 4.0 spec, page 61:
2040 * "It is possible for an application to bind more than one
2041 * attribute name to the same location. This is referred to as
2042 * aliasing. This will only work if only one of the aliased
2043 * attributes is active in the executable program, or if no
2044 * path through the shader consumes more than one attribute of
2045 * a set of attributes aliased to the same location. A link
2046 * error can occur if the linker determines that every path
2047 * through the shader consumes multiple aliased attributes,
2048 * but implementations are not required to generate an error
2051 * From GLSL 4.30 spec, page 54:
2053 * "A program will fail to link if any two non-vertex shader
2054 * input variables are assigned to the same location. For
2055 * vertex shaders, multiple input variables may be assigned
2056 * to the same location using either layout qualifiers or via
2057 * the OpenGL API. However, such aliasing is intended only to
2058 * support vertex shaders where each execution path accesses
2059 * at most one input per each location. Implementations are
2060 * permitted, but not required, to generate link-time errors
2061 * if they detect that every path through the vertex shader
2062 * executable accesses multiple inputs assigned to any single
2063 * location. For all shader types, a program will fail to link
2064 * if explicit location assignments leave the linker unable
2065 * to find space for other variables without explicit
2068 * From OpenGL ES 3.0 spec, page 56:
2070 * "Binding more than one attribute name to the same location
2071 * is referred to as aliasing, and is not permitted in OpenGL
2072 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2073 * fail when this condition exists. However, aliasing is
2074 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2075 * This will only work if only one of the aliased attributes
2076 * is active in the executable program, or if no path through
2077 * the shader consumes more than one attribute of a set of
2078 * attributes aliased to the same location. A link error can
2079 * occur if the linker determines that every path through the
2080 * shader consumes multiple aliased attributes, but implemen-
2081 * tations are not required to generate an error in this case."
2083 * After looking at above references from OpenGL, OpenGL ES and
2084 * GLSL specifications, we allow aliasing of vertex input variables
2085 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2087 * NOTE: This is not required by the spec but its worth mentioning
2088 * here that we're not doing anything to make sure that no path
2089 * through the vertex shader executable accesses multiple inputs
2090 * assigned to any single location.
2093 /* Mask representing the contiguous slots that will be used by
2096 const unsigned attr
= var
->data
.location
- generic_base
;
2097 const unsigned use_mask
= (1 << slots
) - 1;
2098 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2099 ? "vertex shader input" : "fragment shader output";
2101 /* Generate a link error if the requested locations for this
2102 * attribute exceed the maximum allowed attribute location.
2104 if (attr
+ slots
> max_index
) {
2106 "insufficient contiguous locations "
2107 "available for %s `%s' %d %d %d\n", string
,
2108 var
->name
, used_locations
, use_mask
, attr
);
2112 /* Generate a link error if the set of bits requested for this
2113 * attribute overlaps any previously allocated bits.
2115 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2116 if (target_index
== MESA_SHADER_FRAGMENT
||
2117 (prog
->IsES
&& prog
->Version
>= 300)) {
2119 "overlapping location is assigned "
2120 "to %s `%s' %d %d %d\n", string
,
2121 var
->name
, used_locations
, use_mask
, attr
);
2124 linker_warning(prog
,
2125 "overlapping location is assigned "
2126 "to %s `%s' %d %d %d\n", string
,
2127 var
->name
, used_locations
, use_mask
, attr
);
2131 used_locations
|= (use_mask
<< attr
);
2137 to_assign
[num_attr
].slots
= slots
;
2138 to_assign
[num_attr
].var
= var
;
2142 /* If all of the attributes were assigned locations by the application (or
2143 * are built-in attributes with fixed locations), return early. This should
2144 * be the common case.
2149 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2151 if (target_index
== MESA_SHADER_VERTEX
) {
2152 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2153 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2154 * reserved to prevent it from being automatically allocated below.
2156 find_deref_visitor
find("gl_Vertex");
2158 if (find
.variable_found())
2159 used_locations
|= (1 << 0);
2162 for (unsigned i
= 0; i
< num_attr
; i
++) {
2163 /* Mask representing the contiguous slots that will be used by this
2166 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2168 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2171 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2172 ? "vertex shader input" : "fragment shader output";
2175 "insufficient contiguous locations "
2176 "available for %s `%s'\n",
2177 string
, to_assign
[i
].var
->name
);
2181 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2182 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2183 used_locations
|= (use_mask
<< location
);
2191 * Demote shader inputs and outputs that are not used in other stages
2194 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2196 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2197 ir_variable
*const var
= node
->as_variable();
2199 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2202 /* A shader 'in' or 'out' variable is only really an input or output if
2203 * its value is used by other shader stages. This will cause the variable
2204 * to have a location assigned.
2206 if (var
->data
.is_unmatched_generic_inout
) {
2207 assert(var
->data
.mode
!= ir_var_temporary
);
2208 var
->data
.mode
= ir_var_auto
;
2215 * Store the gl_FragDepth layout in the gl_shader_program struct.
2218 store_fragdepth_layout(struct gl_shader_program
*prog
)
2220 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2224 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2226 /* We don't look up the gl_FragDepth symbol directly because if
2227 * gl_FragDepth is not used in the shader, it's removed from the IR.
2228 * However, the symbol won't be removed from the symbol table.
2230 * We're only interested in the cases where the variable is NOT removed
2233 foreach_in_list(ir_instruction
, node
, ir
) {
2234 ir_variable
*const var
= node
->as_variable();
2236 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2240 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2241 switch (var
->data
.depth_layout
) {
2242 case ir_depth_layout_none
:
2243 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2245 case ir_depth_layout_any
:
2246 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2248 case ir_depth_layout_greater
:
2249 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2251 case ir_depth_layout_less
:
2252 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2254 case ir_depth_layout_unchanged
:
2255 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2266 * Validate the resources used by a program versus the implementation limits
2269 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2271 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2272 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2277 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2278 linker_error(prog
, "Too many %s shader texture samplers\n",
2279 _mesa_shader_stage_to_string(i
));
2282 if (sh
->num_uniform_components
>
2283 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2284 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2285 linker_warning(prog
, "Too many %s shader default uniform block "
2286 "components, but the driver will try to optimize "
2287 "them out; this is non-portable out-of-spec "
2289 _mesa_shader_stage_to_string(i
));
2291 linker_error(prog
, "Too many %s shader default uniform block "
2293 _mesa_shader_stage_to_string(i
));
2297 if (sh
->num_combined_uniform_components
>
2298 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2299 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2300 linker_warning(prog
, "Too many %s shader uniform components, "
2301 "but the driver will try to optimize them out; "
2302 "this is non-portable out-of-spec behavior\n",
2303 _mesa_shader_stage_to_string(i
));
2305 linker_error(prog
, "Too many %s shader uniform components\n",
2306 _mesa_shader_stage_to_string(i
));
2311 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2312 unsigned total_uniform_blocks
= 0;
2314 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2315 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2316 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2318 total_uniform_blocks
++;
2322 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2323 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2324 prog
->NumUniformBlocks
,
2325 ctx
->Const
.MaxCombinedUniformBlocks
);
2327 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2328 const unsigned max_uniform_blocks
=
2329 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2330 if (blocks
[i
] > max_uniform_blocks
) {
2331 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2332 _mesa_shader_stage_to_string(i
),
2334 max_uniform_blocks
);
2343 * Validate shader image resources.
2346 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2348 unsigned total_image_units
= 0;
2349 unsigned fragment_outputs
= 0;
2351 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2354 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2355 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2358 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2359 linker_error(prog
, "Too many %s shader image uniforms\n",
2360 _mesa_shader_stage_to_string(i
));
2362 total_image_units
+= sh
->NumImages
;
2364 if (i
== MESA_SHADER_FRAGMENT
) {
2365 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2366 ir_variable
*var
= node
->as_variable();
2367 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2368 fragment_outputs
+= var
->type
->count_attribute_slots();
2374 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2375 linker_error(prog
, "Too many combined image uniforms\n");
2377 if (total_image_units
+ fragment_outputs
>
2378 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2379 linker_error(prog
, "Too many combined image uniforms and fragment outputs\n");
2384 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2385 * for a variable, checks for overlaps between other uniforms using explicit
2389 reserve_explicit_locations(struct gl_shader_program
*prog
,
2390 string_to_uint_map
*map
, ir_variable
*var
)
2392 unsigned slots
= var
->type
->uniform_locations();
2393 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2395 /* Resize remap table if locations do not fit in the current one. */
2396 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2397 prog
->UniformRemapTable
=
2398 reralloc(prog
, prog
->UniformRemapTable
,
2399 gl_uniform_storage
*,
2402 if (!prog
->UniformRemapTable
) {
2403 linker_error(prog
, "Out of memory during linking.\n");
2407 /* Initialize allocated space. */
2408 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2409 prog
->UniformRemapTable
[i
] = NULL
;
2411 prog
->NumUniformRemapTable
= max_loc
+ 1;
2414 for (unsigned i
= 0; i
< slots
; i
++) {
2415 unsigned loc
= var
->data
.location
+ i
;
2417 /* Check if location is already used. */
2418 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2420 /* Possibly same uniform from a different stage, this is ok. */
2422 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2425 /* ARB_explicit_uniform_location specification states:
2427 * "No two default-block uniform variables in the program can have
2428 * the same location, even if they are unused, otherwise a compiler
2429 * or linker error will be generated."
2432 "location qualifier for uniform %s overlaps "
2433 "previously used location\n",
2438 /* Initialize location as inactive before optimization
2439 * rounds and location assignment.
2441 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2444 /* Note, base location used for arrays. */
2445 map
->put(var
->data
.location
, var
->name
);
2451 * Check and reserve all explicit uniform locations, called before
2452 * any optimizations happen to handle also inactive uniforms and
2453 * inactive array elements that may get trimmed away.
2456 check_explicit_uniform_locations(struct gl_context
*ctx
,
2457 struct gl_shader_program
*prog
)
2459 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2462 /* This map is used to detect if overlapping explicit locations
2463 * occur with the same uniform (from different stage) or a different one.
2465 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2468 linker_error(prog
, "Out of memory during linking.\n");
2472 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2473 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2478 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2479 ir_variable
*var
= node
->as_variable();
2480 if ((var
&& var
->data
.mode
== ir_var_uniform
) &&
2481 var
->data
.explicit_location
) {
2482 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2494 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2496 tfeedback_decl
*tfeedback_decls
= NULL
;
2497 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2499 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2501 prog
->LinkStatus
= true; /* All error paths will set this to false */
2502 prog
->Validated
= false;
2503 prog
->_Used
= false;
2505 prog
->ARB_fragment_coord_conventions_enable
= false;
2507 /* Separate the shaders into groups based on their type.
2509 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2510 unsigned num_shaders
[MESA_SHADER_STAGES
];
2512 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2513 shader_list
[i
] = (struct gl_shader
**)
2514 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2518 unsigned min_version
= UINT_MAX
;
2519 unsigned max_version
= 0;
2520 const bool is_es_prog
=
2521 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2522 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2523 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2524 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2526 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2527 linker_error(prog
, "all shaders must use same shading "
2528 "language version\n");
2532 prog
->ARB_fragment_coord_conventions_enable
|=
2533 prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
;
2535 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2536 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2537 num_shaders
[shader_type
]++;
2540 /* In desktop GLSL, different shader versions may be linked together. In
2541 * GLSL ES, all shader versions must be the same.
2543 if (is_es_prog
&& min_version
!= max_version
) {
2544 linker_error(prog
, "all shaders must use same shading "
2545 "language version\n");
2549 prog
->Version
= max_version
;
2550 prog
->IsES
= is_es_prog
;
2552 /* Geometry shaders have to be linked with vertex shaders.
2554 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2555 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2556 !prog
->SeparateShader
) {
2557 linker_error(prog
, "Geometry shader must be linked with "
2562 /* Compute shaders have additional restrictions. */
2563 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2564 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2565 linker_error(prog
, "Compute shaders may not be linked with any other "
2566 "type of shader\n");
2569 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2570 if (prog
->_LinkedShaders
[i
] != NULL
)
2571 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2573 prog
->_LinkedShaders
[i
] = NULL
;
2576 /* Link all shaders for a particular stage and validate the result.
2578 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2579 if (num_shaders
[stage
] > 0) {
2580 gl_shader
*const sh
=
2581 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2582 num_shaders
[stage
]);
2584 if (!prog
->LinkStatus
)
2588 case MESA_SHADER_VERTEX
:
2589 validate_vertex_shader_executable(prog
, sh
);
2591 case MESA_SHADER_GEOMETRY
:
2592 validate_geometry_shader_executable(prog
, sh
);
2594 case MESA_SHADER_FRAGMENT
:
2595 validate_fragment_shader_executable(prog
, sh
);
2598 if (!prog
->LinkStatus
)
2601 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2605 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2606 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2607 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2608 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2610 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2612 /* Here begins the inter-stage linking phase. Some initial validation is
2613 * performed, then locations are assigned for uniforms, attributes, and
2616 cross_validate_uniforms(prog
);
2617 if (!prog
->LinkStatus
)
2622 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2623 if (prog
->_LinkedShaders
[prev
] != NULL
)
2627 check_explicit_uniform_locations(ctx
, prog
);
2628 if (!prog
->LinkStatus
)
2631 /* Validate the inputs of each stage with the output of the preceding
2634 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2635 if (prog
->_LinkedShaders
[i
] == NULL
)
2638 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2639 prog
->_LinkedShaders
[i
]);
2640 if (!prog
->LinkStatus
)
2643 cross_validate_outputs_to_inputs(prog
,
2644 prog
->_LinkedShaders
[prev
],
2645 prog
->_LinkedShaders
[i
]);
2646 if (!prog
->LinkStatus
)
2652 /* Cross-validate uniform blocks between shader stages */
2653 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
2654 MESA_SHADER_STAGES
);
2655 if (!prog
->LinkStatus
)
2658 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2659 if (prog
->_LinkedShaders
[i
] != NULL
)
2660 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2663 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2664 * it before optimization because we want most of the checks to get
2665 * dropped thanks to constant propagation.
2667 * This rule also applies to GLSL ES 3.00.
2669 if (max_version
>= (is_es_prog
? 300 : 130)) {
2670 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2672 lower_discard_flow(sh
->ir
);
2676 if (!interstage_cross_validate_uniform_blocks(prog
))
2679 /* Do common optimization before assigning storage for attributes,
2680 * uniforms, and varyings. Later optimization could possibly make
2681 * some of that unused.
2683 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2684 if (prog
->_LinkedShaders
[i
] == NULL
)
2687 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2688 if (!prog
->LinkStatus
)
2691 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
2692 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2695 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
2696 &ctx
->Const
.ShaderCompilerOptions
[i
],
2697 ctx
->Const
.NativeIntegers
))
2700 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
2703 /* Check and validate stream emissions in geometry shaders */
2704 validate_geometry_shader_emissions(ctx
, prog
);
2706 /* Mark all generic shader inputs and outputs as unpaired. */
2707 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2708 if (prog
->_LinkedShaders
[i
] != NULL
) {
2709 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
2713 /* FINISHME: The value of the max_attribute_index parameter is
2714 * FINISHME: implementation dependent based on the value of
2715 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2716 * FINISHME: at least 16, so hardcode 16 for now.
2718 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2722 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2727 for (first
= 0; first
<= MESA_SHADER_FRAGMENT
; first
++) {
2728 if (prog
->_LinkedShaders
[first
] != NULL
)
2732 if (num_tfeedback_decls
!= 0) {
2733 /* From GL_EXT_transform_feedback:
2734 * A program will fail to link if:
2736 * * the <count> specified by TransformFeedbackVaryingsEXT is
2737 * non-zero, but the program object has no vertex or geometry
2740 if (first
== MESA_SHADER_FRAGMENT
) {
2741 linker_error(prog
, "Transform feedback varyings specified, but "
2742 "no vertex or geometry shader is present.\n");
2746 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2747 prog
->TransformFeedback
.NumVarying
);
2748 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2749 prog
->TransformFeedback
.VaryingNames
,
2754 /* Linking the stages in the opposite order (from fragment to vertex)
2755 * ensures that inter-shader outputs written to in an earlier stage are
2756 * eliminated if they are (transitively) not used in a later stage.
2759 for (last
= MESA_SHADER_FRAGMENT
; last
>= 0; last
--) {
2760 if (prog
->_LinkedShaders
[last
] != NULL
)
2764 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2765 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2767 if (first
== MESA_SHADER_GEOMETRY
) {
2768 /* There was no vertex shader, but we still have to assign varying
2769 * locations for use by geometry shader inputs in SSO.
2771 * If the shader is not separable (i.e., prog->SeparateShader is
2772 * false), linking will have already failed when first is
2773 * MESA_SHADER_GEOMETRY.
2775 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2777 num_tfeedback_decls
, tfeedback_decls
,
2778 prog
->Geom
.VerticesIn
))
2782 if (num_tfeedback_decls
!= 0 || prog
->SeparateShader
) {
2783 /* There was no fragment shader, but we still have to assign varying
2784 * locations for use by transform feedback.
2786 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2788 num_tfeedback_decls
, tfeedback_decls
,
2793 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2794 num_tfeedback_decls
, tfeedback_decls
);
2796 if (!prog
->SeparateShader
)
2797 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2799 /* Eliminate code that is now dead due to unused outputs being demoted.
2801 while (do_dead_code(sh
->ir
, false))
2804 else if (first
== MESA_SHADER_FRAGMENT
) {
2805 /* If the program only contains a fragment shader...
2807 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2809 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2810 num_tfeedback_decls
, tfeedback_decls
);
2812 if (prog
->SeparateShader
) {
2813 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2814 NULL
/* producer */,
2816 0 /* num_tfeedback_decls */,
2817 NULL
/* tfeedback_decls */,
2818 0 /* gs_input_vertices */))
2821 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2823 while (do_dead_code(sh
->ir
, false))
2828 for (int i
= next
- 1; i
>= 0; i
--) {
2829 if (prog
->_LinkedShaders
[i
] == NULL
)
2832 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2833 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2834 unsigned gs_input_vertices
=
2835 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2837 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2838 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2839 tfeedback_decls
, gs_input_vertices
))
2842 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2843 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2846 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2847 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2849 /* Eliminate code that is now dead due to unused outputs being demoted.
2851 while (do_dead_code(sh_i
->ir
, false))
2853 while (do_dead_code(sh_next
->ir
, false))
2856 /* This must be done after all dead varyings are eliminated. */
2857 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2859 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2865 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2868 update_array_sizes(prog
);
2869 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
2870 link_assign_atomic_counter_resources(ctx
, prog
);
2871 store_fragdepth_layout(prog
);
2873 check_resources(ctx
, prog
);
2874 check_image_resources(ctx
, prog
);
2875 link_check_atomic_counter_resources(ctx
, prog
);
2877 if (!prog
->LinkStatus
)
2880 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2881 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
2882 * anything about shader linking when one of the shaders (vertex or
2883 * fragment shader) is absent. So, the extension shouldn't change the
2884 * behavior specified in GLSL specification.
2886 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
2887 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2888 linker_error(prog
, "program lacks a vertex shader\n");
2889 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2890 linker_error(prog
, "program lacks a fragment shader\n");
2894 /* FINISHME: Assign fragment shader output locations. */
2897 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2898 free(shader_list
[i
]);
2899 if (prog
->_LinkedShaders
[i
] == NULL
)
2902 /* Do a final validation step to make sure that the IR wasn't
2903 * invalidated by any modifications performed after intrastage linking.
2905 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
2907 /* Retain any live IR, but trash the rest. */
2908 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2910 /* The symbol table in the linked shaders may contain references to
2911 * variables that were removed (e.g., unused uniforms). Since it may
2912 * contain junk, there is no possible valid use. Delete it and set the
2915 delete prog
->_LinkedShaders
[i
]->symbols
;
2916 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2919 ralloc_free(mem_ctx
);