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
;
253 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
255 unsigned num_vertices
;
256 gl_shader_program
*prog
;
258 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
260 this->num_vertices
= num_vertices
;
264 virtual ~tess_eval_array_resize_visitor()
269 virtual ir_visitor_status
visit(ir_variable
*var
)
271 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
272 return visit_continue
;
274 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
276 var
->data
.max_array_access
= this->num_vertices
- 1;
278 return visit_continue
;
281 /* Dereferences of input variables need to be updated so that their type
282 * matches the newly assigned type of the variable they are accessing. */
283 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
285 ir
->type
= ir
->var
->type
;
286 return visit_continue
;
289 /* Dereferences of 2D input arrays need to be updated so that their type
290 * matches the newly assigned type of the array they are accessing. */
291 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
293 const glsl_type
*const vt
= ir
->array
->type
;
295 ir
->type
= vt
->fields
.array
;
296 return visit_continue
;
301 * Visitor that determines the highest stream id to which a (geometry) shader
302 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
304 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
306 find_emit_vertex_visitor(int max_allowed
)
307 : max_stream_allowed(max_allowed
),
308 invalid_stream_id(0),
309 invalid_stream_id_from_emit_vertex(false),
310 end_primitive_found(false),
311 uses_non_zero_stream(false)
316 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
318 int stream_id
= ir
->stream_id();
321 invalid_stream_id
= stream_id
;
322 invalid_stream_id_from_emit_vertex
= true;
326 if (stream_id
> max_stream_allowed
) {
327 invalid_stream_id
= stream_id
;
328 invalid_stream_id_from_emit_vertex
= true;
333 uses_non_zero_stream
= true;
335 return visit_continue
;
338 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
340 end_primitive_found
= true;
342 int stream_id
= ir
->stream_id();
345 invalid_stream_id
= stream_id
;
346 invalid_stream_id_from_emit_vertex
= false;
350 if (stream_id
> max_stream_allowed
) {
351 invalid_stream_id
= stream_id
;
352 invalid_stream_id_from_emit_vertex
= false;
357 uses_non_zero_stream
= true;
359 return visit_continue
;
364 return invalid_stream_id
!= 0;
367 const char *error_func()
369 return invalid_stream_id_from_emit_vertex
?
370 "EmitStreamVertex" : "EndStreamPrimitive";
375 return invalid_stream_id
;
380 return uses_non_zero_stream
;
383 bool uses_end_primitive()
385 return end_primitive_found
;
389 int max_stream_allowed
;
390 int invalid_stream_id
;
391 bool invalid_stream_id_from_emit_vertex
;
392 bool end_primitive_found
;
393 bool uses_non_zero_stream
;
396 /* Class that finds array derefs and check if indexes are dynamic. */
397 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
400 dynamic_sampler_array_indexing_visitor() :
401 dynamic_sampler_array_indexing(false)
405 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
407 if (!ir
->variable_referenced())
408 return visit_continue
;
410 if (!ir
->variable_referenced()->type
->contains_sampler())
411 return visit_continue
;
413 if (!ir
->array_index
->constant_expression_value()) {
414 dynamic_sampler_array_indexing
= true;
417 return visit_continue
;
420 bool uses_dynamic_sampler_array_indexing()
422 return dynamic_sampler_array_indexing
;
426 bool dynamic_sampler_array_indexing
;
429 } /* anonymous namespace */
432 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
436 ralloc_strcat(&prog
->InfoLog
, "error: ");
438 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
441 prog
->LinkStatus
= false;
446 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
450 ralloc_strcat(&prog
->InfoLog
, "warning: ");
452 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
459 * Given a string identifying a program resource, break it into a base name
460 * and an optional array index in square brackets.
462 * If an array index is present, \c out_base_name_end is set to point to the
463 * "[" that precedes the array index, and the array index itself is returned
466 * If no array index is present (or if the array index is negative or
467 * mal-formed), \c out_base_name_end, is set to point to the null terminator
468 * at the end of the input string, and -1 is returned.
470 * Only the final array index is parsed; if the string contains other array
471 * indices (or structure field accesses), they are left in the base name.
473 * No attempt is made to check that the base name is properly formed;
474 * typically the caller will look up the base name in a hash table, so
475 * ill-formed base names simply turn into hash table lookup failures.
478 parse_program_resource_name(const GLchar
*name
,
479 const GLchar
**out_base_name_end
)
481 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
483 * "When an integer array element or block instance number is part of
484 * the name string, it will be specified in decimal form without a "+"
485 * or "-" sign or any extra leading zeroes. Additionally, the name
486 * string will not include white space anywhere in the string."
489 const size_t len
= strlen(name
);
490 *out_base_name_end
= name
+ len
;
492 if (len
== 0 || name
[len
-1] != ']')
495 /* Walk backwards over the string looking for a non-digit character. This
496 * had better be the opening bracket for an array index.
498 * Initially, i specifies the location of the ']'. Since the string may
499 * contain only the ']' charcater, walk backwards very carefully.
502 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
505 if ((i
== 0) || name
[i
-1] != '[')
508 long array_index
= strtol(&name
[i
], NULL
, 10);
512 /* Check for leading zero */
513 if (name
[i
] == '0' && name
[i
+1] != ']')
516 *out_base_name_end
= name
+ (i
- 1);
522 link_invalidate_variable_locations(exec_list
*ir
)
524 foreach_in_list(ir_instruction
, node
, ir
) {
525 ir_variable
*const var
= node
->as_variable();
530 /* Only assign locations for variables that lack an explicit location.
531 * Explicit locations are set for all built-in variables, generic vertex
532 * shader inputs (via layout(location=...)), and generic fragment shader
533 * outputs (also via layout(location=...)).
535 if (!var
->data
.explicit_location
) {
536 var
->data
.location
= -1;
537 var
->data
.location_frac
= 0;
540 /* ir_variable::is_unmatched_generic_inout is used by the linker while
541 * connecting outputs from one stage to inputs of the next stage.
543 * There are two implicit assumptions here. First, we assume that any
544 * built-in variable (i.e., non-generic in or out) will have
545 * explicit_location set. Second, we assume that any generic in or out
546 * will not have explicit_location set.
548 * This second assumption will only be valid until
549 * GL_ARB_separate_shader_objects is supported. When that extension is
550 * implemented, this function will need some modifications.
552 if (!var
->data
.explicit_location
) {
553 var
->data
.is_unmatched_generic_inout
= 1;
555 var
->data
.is_unmatched_generic_inout
= 0;
562 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
564 * Also check for errors based on incorrect usage of gl_ClipVertex and
567 * Return false if an error was reported.
570 analyze_clip_usage(struct gl_shader_program
*prog
,
571 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
572 GLuint
*ClipDistanceArraySize
)
574 *ClipDistanceArraySize
= 0;
576 if (!prog
->IsES
&& prog
->Version
>= 130) {
577 /* From section 7.1 (Vertex Shader Special Variables) of the
580 * "It is an error for a shader to statically write both
581 * gl_ClipVertex and gl_ClipDistance."
583 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
584 * gl_ClipVertex nor gl_ClipDistance.
586 find_assignment_visitor
clip_vertex("gl_ClipVertex");
587 find_assignment_visitor
clip_distance("gl_ClipDistance");
589 clip_vertex
.run(shader
->ir
);
590 clip_distance
.run(shader
->ir
);
591 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
592 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
593 "and `gl_ClipDistance'\n",
594 _mesa_shader_stage_to_string(shader
->Stage
));
597 *UsesClipDistance
= clip_distance
.variable_found();
598 ir_variable
*clip_distance_var
=
599 shader
->symbols
->get_variable("gl_ClipDistance");
600 if (clip_distance_var
)
601 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
603 *UsesClipDistance
= false;
609 * Verify that a vertex shader executable meets all semantic requirements.
611 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
614 * \param shader Vertex shader executable to be verified
617 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
618 struct gl_shader
*shader
)
623 /* From the GLSL 1.10 spec, page 48:
625 * "The variable gl_Position is available only in the vertex
626 * language and is intended for writing the homogeneous vertex
627 * position. All executions of a well-formed vertex shader
628 * executable must write a value into this variable. [...] The
629 * variable gl_Position is available only in the vertex
630 * language and is intended for writing the homogeneous vertex
631 * position. All executions of a well-formed vertex shader
632 * executable must write a value into this variable."
634 * while in GLSL 1.40 this text is changed to:
636 * "The variable gl_Position is available only in the vertex
637 * language and is intended for writing the homogeneous vertex
638 * position. It can be written at any time during shader
639 * execution. It may also be read back by a vertex shader
640 * after being written. This value will be used by primitive
641 * assembly, clipping, culling, and other fixed functionality
642 * operations, if present, that operate on primitives after
643 * vertex processing has occurred. Its value is undefined if
644 * the vertex shader executable does not write gl_Position."
646 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
647 * gl_Position is not an error.
649 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
650 find_assignment_visitor
find("gl_Position");
651 find
.run(shader
->ir
);
652 if (!find
.variable_found()) {
655 "vertex shader does not write to `gl_Position'."
656 "It's value is undefined. \n");
659 "vertex shader does not write to `gl_Position'. \n");
665 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
666 &prog
->Vert
.ClipDistanceArraySize
);
671 * Verify that a fragment shader executable meets all semantic requirements
673 * \param shader Fragment shader executable to be verified
676 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
677 struct gl_shader
*shader
)
682 find_assignment_visitor
frag_color("gl_FragColor");
683 find_assignment_visitor
frag_data("gl_FragData");
685 frag_color
.run(shader
->ir
);
686 frag_data
.run(shader
->ir
);
688 if (frag_color
.variable_found() && frag_data
.variable_found()) {
689 linker_error(prog
, "fragment shader writes to both "
690 "`gl_FragColor' and `gl_FragData'\n");
695 * Verify that a geometry shader executable meets all semantic requirements
697 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
698 * prog->Geom.ClipDistanceArraySize as a side effect.
700 * \param shader Geometry shader executable to be verified
703 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
704 struct gl_shader
*shader
)
709 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
710 prog
->Geom
.VerticesIn
= num_vertices
;
712 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
713 &prog
->Geom
.ClipDistanceArraySize
);
717 * Check if geometry shaders emit to non-zero streams and do corresponding
721 validate_geometry_shader_emissions(struct gl_context
*ctx
,
722 struct gl_shader_program
*prog
)
724 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
725 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
726 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
727 if (emit_vertex
.error()) {
728 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
729 "stream parameter are in the range [0, %d].\n",
730 emit_vertex
.error_func(),
731 emit_vertex
.error_stream(),
732 ctx
->Const
.MaxVertexStreams
- 1);
734 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
735 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
737 /* From the ARB_gpu_shader5 spec:
739 * "Multiple vertex streams are supported only if the output primitive
740 * type is declared to be "points". A program will fail to link if it
741 * contains a geometry shader calling EmitStreamVertex() or
742 * EndStreamPrimitive() if its output primitive type is not "points".
744 * However, in the same spec:
746 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
747 * with <stream> set to zero."
751 * "The function EndPrimitive() is equivalent to calling
752 * EndStreamPrimitive() with <stream> set to zero."
754 * Since we can call EmitVertex() and EndPrimitive() when we output
755 * primitives other than points, calling EmitStreamVertex(0) or
756 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
757 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
758 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
761 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
762 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
763 "with n>0 requires point output\n");
769 validate_intrastage_arrays(struct gl_shader_program
*prog
,
770 ir_variable
*const var
,
771 ir_variable
*const existing
)
773 /* Consider the types to be "the same" if both types are arrays
774 * of the same type and one of the arrays is implicitly sized.
775 * In addition, set the type of the linked variable to the
776 * explicitly sized array.
778 if (var
->type
->is_array() && existing
->type
->is_array() &&
779 (var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
780 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
781 if (var
->type
->length
!= 0) {
782 if (var
->type
->length
<= existing
->data
.max_array_access
) {
783 linker_error(prog
, "%s `%s' declared as type "
784 "`%s' but outermost dimension has an index"
787 var
->name
, var
->type
->name
,
788 existing
->data
.max_array_access
);
790 existing
->type
= var
->type
;
792 } else if (existing
->type
->length
!= 0) {
793 if(existing
->type
->length
<= var
->data
.max_array_access
) {
794 linker_error(prog
, "%s `%s' declared as type "
795 "`%s' but outermost dimension has an index"
798 var
->name
, existing
->type
->name
,
799 var
->data
.max_array_access
);
809 * Perform validation of global variables used across multiple shaders
812 cross_validate_globals(struct gl_shader_program
*prog
,
813 struct gl_shader
**shader_list
,
814 unsigned num_shaders
,
817 /* Examine all of the uniforms in all of the shaders and cross validate
820 glsl_symbol_table variables
;
821 for (unsigned i
= 0; i
< num_shaders
; i
++) {
822 if (shader_list
[i
] == NULL
)
825 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
826 ir_variable
*const var
= node
->as_variable();
831 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
834 /* Don't cross validate temporaries that are at global scope. These
835 * will eventually get pulled into the shaders 'main'.
837 if (var
->data
.mode
== ir_var_temporary
)
840 /* If a global with this name has already been seen, verify that the
841 * new instance has the same type. In addition, if the globals have
842 * initializers, the values of the initializers must be the same.
844 ir_variable
*const existing
= variables
.get_variable(var
->name
);
845 if (existing
!= NULL
) {
846 /* Check if types match. Interface blocks have some special
847 * rules so we handle those elsewhere.
849 if (var
->type
!= existing
->type
&&
850 !var
->is_interface_instance()) {
851 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
852 if (var
->type
->is_record() && existing
->type
->is_record()
853 && existing
->type
->record_compare(var
->type
)) {
854 existing
->type
= var
->type
;
856 linker_error(prog
, "%s `%s' declared as type "
857 "`%s' and type `%s'\n",
859 var
->name
, var
->type
->name
,
860 existing
->type
->name
);
866 if (var
->data
.explicit_location
) {
867 if (existing
->data
.explicit_location
868 && (var
->data
.location
!= existing
->data
.location
)) {
869 linker_error(prog
, "explicit locations for %s "
870 "`%s' have differing values\n",
871 mode_string(var
), var
->name
);
875 existing
->data
.location
= var
->data
.location
;
876 existing
->data
.explicit_location
= true;
879 /* From the GLSL 4.20 specification:
880 * "A link error will result if two compilation units in a program
881 * specify different integer-constant bindings for the same
882 * opaque-uniform name. However, it is not an error to specify a
883 * binding on some but not all declarations for the same name"
885 if (var
->data
.explicit_binding
) {
886 if (existing
->data
.explicit_binding
&&
887 var
->data
.binding
!= existing
->data
.binding
) {
888 linker_error(prog
, "explicit bindings for %s "
889 "`%s' have differing values\n",
890 mode_string(var
), var
->name
);
894 existing
->data
.binding
= var
->data
.binding
;
895 existing
->data
.explicit_binding
= true;
898 if (var
->type
->contains_atomic() &&
899 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
900 linker_error(prog
, "offset specifications for %s "
901 "`%s' have differing values\n",
902 mode_string(var
), var
->name
);
906 /* Validate layout qualifiers for gl_FragDepth.
908 * From the AMD/ARB_conservative_depth specs:
910 * "If gl_FragDepth is redeclared in any fragment shader in a
911 * program, it must be redeclared in all fragment shaders in
912 * that program that have static assignments to
913 * gl_FragDepth. All redeclarations of gl_FragDepth in all
914 * fragment shaders in a single program must have the same set
917 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
918 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
919 bool layout_differs
=
920 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
922 if (layout_declared
&& layout_differs
) {
924 "All redeclarations of gl_FragDepth in all "
925 "fragment shaders in a single program must have "
926 "the same set of qualifiers.\n");
929 if (var
->data
.used
&& layout_differs
) {
931 "If gl_FragDepth is redeclared with a layout "
932 "qualifier in any fragment shader, it must be "
933 "redeclared with the same layout qualifier in "
934 "all fragment shaders that have assignments to "
939 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
941 * "If a shared global has multiple initializers, the
942 * initializers must all be constant expressions, and they
943 * must all have the same value. Otherwise, a link error will
944 * result. (A shared global having only one initializer does
945 * not require that initializer to be a constant expression.)"
947 * Previous to 4.20 the GLSL spec simply said that initializers
948 * must have the same value. In this case of non-constant
949 * initializers, this was impossible to determine. As a result,
950 * no vendor actually implemented that behavior. The 4.20
951 * behavior matches the implemented behavior of at least one other
952 * vendor, so we'll implement that for all GLSL versions.
954 if (var
->constant_initializer
!= NULL
) {
955 if (existing
->constant_initializer
!= NULL
) {
956 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
957 linker_error(prog
, "initializers for %s "
958 "`%s' have differing values\n",
959 mode_string(var
), var
->name
);
963 /* If the first-seen instance of a particular uniform did not
964 * have an initializer but a later instance does, copy the
965 * initializer to the version stored in the symbol table.
967 /* FINISHME: This is wrong. The constant_value field should
968 * FINISHME: not be modified! Imagine a case where a shader
969 * FINISHME: without an initializer is linked in two different
970 * FINISHME: programs with shaders that have differing
971 * FINISHME: initializers. Linking with the first will
972 * FINISHME: modify the shader, and linking with the second
973 * FINISHME: will fail.
975 existing
->constant_initializer
=
976 var
->constant_initializer
->clone(ralloc_parent(existing
),
981 if (var
->data
.has_initializer
) {
982 if (existing
->data
.has_initializer
983 && (var
->constant_initializer
== NULL
984 || existing
->constant_initializer
== NULL
)) {
986 "shared global variable `%s' has multiple "
987 "non-constant initializers.\n",
992 /* Some instance had an initializer, so keep track of that. In
993 * this location, all sorts of initializers (constant or
994 * otherwise) will propagate the existence to the variable
995 * stored in the symbol table.
997 existing
->data
.has_initializer
= true;
1000 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1001 linker_error(prog
, "declarations for %s `%s' have "
1002 "mismatching invariant qualifiers\n",
1003 mode_string(var
), var
->name
);
1006 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1007 linker_error(prog
, "declarations for %s `%s' have "
1008 "mismatching centroid qualifiers\n",
1009 mode_string(var
), var
->name
);
1012 if (existing
->data
.sample
!= var
->data
.sample
) {
1013 linker_error(prog
, "declarations for %s `%s` have "
1014 "mismatching sample qualifiers\n",
1015 mode_string(var
), var
->name
);
1019 variables
.add_variable(var
);
1026 * Perform validation of uniforms used across multiple shader stages
1029 cross_validate_uniforms(struct gl_shader_program
*prog
)
1031 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1032 MESA_SHADER_STAGES
, true);
1036 * Accumulates the array of prog->UniformBlocks and checks that all
1037 * definitons of blocks agree on their contents.
1040 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1042 unsigned max_num_uniform_blocks
= 0;
1043 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1044 if (prog
->_LinkedShaders
[i
])
1045 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1048 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1049 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1051 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
1052 max_num_uniform_blocks
);
1053 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1054 prog
->UniformBlockStageIndex
[i
][j
] = -1;
1059 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
1060 int index
= link_cross_validate_uniform_block(prog
,
1061 &prog
->UniformBlocks
,
1062 &prog
->NumUniformBlocks
,
1063 &sh
->UniformBlocks
[j
]);
1066 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1067 sh
->UniformBlocks
[j
].Name
);
1071 prog
->UniformBlockStageIndex
[i
][index
] = j
;
1080 * Populates a shaders symbol table with all global declarations
1083 populate_symbol_table(gl_shader
*sh
)
1085 sh
->symbols
= new(sh
) glsl_symbol_table
;
1087 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1091 if ((func
= inst
->as_function()) != NULL
) {
1092 sh
->symbols
->add_function(func
);
1093 } else if ((var
= inst
->as_variable()) != NULL
) {
1094 if (var
->data
.mode
!= ir_var_temporary
)
1095 sh
->symbols
->add_variable(var
);
1102 * Remap variables referenced in an instruction tree
1104 * This is used when instruction trees are cloned from one shader and placed in
1105 * another. These trees will contain references to \c ir_variable nodes that
1106 * do not exist in the target shader. This function finds these \c ir_variable
1107 * references and replaces the references with matching variables in the target
1110 * If there is no matching variable in the target shader, a clone of the
1111 * \c ir_variable is made and added to the target shader. The new variable is
1112 * added to \b both the instruction stream and the symbol table.
1114 * \param inst IR tree that is to be processed.
1115 * \param symbols Symbol table containing global scope symbols in the
1117 * \param instructions Instruction stream where new variable declarations
1121 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1124 class remap_visitor
: public ir_hierarchical_visitor
{
1126 remap_visitor(struct gl_shader
*target
,
1129 this->target
= target
;
1130 this->symbols
= target
->symbols
;
1131 this->instructions
= target
->ir
;
1132 this->temps
= temps
;
1135 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1137 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1138 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1140 assert(var
!= NULL
);
1142 return visit_continue
;
1145 ir_variable
*const existing
=
1146 this->symbols
->get_variable(ir
->var
->name
);
1147 if (existing
!= NULL
)
1150 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1152 this->symbols
->add_variable(copy
);
1153 this->instructions
->push_head(copy
);
1157 return visit_continue
;
1161 struct gl_shader
*target
;
1162 glsl_symbol_table
*symbols
;
1163 exec_list
*instructions
;
1167 remap_visitor
v(target
, temps
);
1174 * Move non-declarations from one instruction stream to another
1176 * The intended usage pattern of this function is to pass the pointer to the
1177 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1178 * pointer) for \c last and \c false for \c make_copies on the first
1179 * call. Successive calls pass the return value of the previous call for
1180 * \c last and \c true for \c make_copies.
1182 * \param instructions Source instruction stream
1183 * \param last Instruction after which new instructions should be
1184 * inserted in the target instruction stream
1185 * \param make_copies Flag selecting whether instructions in \c instructions
1186 * should be copied (via \c ir_instruction::clone) into the
1187 * target list or moved.
1190 * The new "last" instruction in the target instruction stream. This pointer
1191 * is suitable for use as the \c last parameter of a later call to this
1195 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1196 bool make_copies
, gl_shader
*target
)
1198 hash_table
*temps
= NULL
;
1201 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1202 hash_table_pointer_compare
);
1204 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1205 if (inst
->as_function())
1208 ir_variable
*var
= inst
->as_variable();
1209 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1212 assert(inst
->as_assignment()
1214 || inst
->as_if() /* for initializers with the ?: operator */
1215 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1218 inst
= inst
->clone(target
, NULL
);
1221 hash_table_insert(temps
, inst
, var
);
1223 remap_variables(inst
, target
, temps
);
1228 last
->insert_after(inst
);
1233 hash_table_dtor(temps
);
1239 * Get the function signature for main from a shader
1241 ir_function_signature
*
1242 link_get_main_function_signature(gl_shader
*sh
)
1244 ir_function
*const f
= sh
->symbols
->get_function("main");
1246 exec_list void_parameters
;
1248 /* Look for the 'void main()' signature and ensure that it's defined.
1249 * This keeps the linker from accidentally pick a shader that just
1250 * contains a prototype for main.
1252 * We don't have to check for multiple definitions of main (in multiple
1253 * shaders) because that would have already been caught above.
1255 ir_function_signature
*sig
=
1256 f
->matching_signature(NULL
, &void_parameters
, false);
1257 if ((sig
!= NULL
) && sig
->is_defined
) {
1267 * This class is only used in link_intrastage_shaders() below but declaring
1268 * it inside that function leads to compiler warnings with some versions of
1271 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1273 array_sizing_visitor()
1274 : mem_ctx(ralloc_context(NULL
)),
1275 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1276 hash_table_pointer_compare
))
1280 ~array_sizing_visitor()
1282 hash_table_dtor(this->unnamed_interfaces
);
1283 ralloc_free(this->mem_ctx
);
1286 virtual ir_visitor_status
visit(ir_variable
*var
)
1288 fixup_type(&var
->type
, var
->data
.max_array_access
);
1289 if (var
->type
->is_interface()) {
1290 if (interface_contains_unsized_arrays(var
->type
)) {
1291 const glsl_type
*new_type
=
1292 resize_interface_members(var
->type
,
1293 var
->get_max_ifc_array_access());
1294 var
->type
= new_type
;
1295 var
->change_interface_type(new_type
);
1297 } else if (var
->type
->is_array() &&
1298 var
->type
->fields
.array
->is_interface()) {
1299 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1300 const glsl_type
*new_type
=
1301 resize_interface_members(var
->type
->fields
.array
,
1302 var
->get_max_ifc_array_access());
1303 var
->change_interface_type(new_type
);
1304 var
->type
= update_interface_members_array(var
->type
, new_type
);
1306 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1307 /* Store a pointer to the variable in the unnamed_interfaces
1310 ir_variable
**interface_vars
= (ir_variable
**)
1311 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1312 if (interface_vars
== NULL
) {
1313 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1315 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1318 unsigned index
= ifc_type
->field_index(var
->name
);
1319 assert(index
< ifc_type
->length
);
1320 assert(interface_vars
[index
] == NULL
);
1321 interface_vars
[index
] = var
;
1323 return visit_continue
;
1327 * For each unnamed interface block that was discovered while running the
1328 * visitor, adjust the interface type to reflect the newly assigned array
1329 * sizes, and fix up the ir_variable nodes to point to the new interface
1332 void fixup_unnamed_interface_types()
1334 hash_table_call_foreach(this->unnamed_interfaces
,
1335 fixup_unnamed_interface_type
, NULL
);
1340 * If the type pointed to by \c type represents an unsized array, replace
1341 * it with a sized array whose size is determined by max_array_access.
1343 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1345 if ((*type
)->is_unsized_array()) {
1346 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1347 max_array_access
+ 1);
1348 assert(*type
!= NULL
);
1352 static const glsl_type
*
1353 update_interface_members_array(const glsl_type
*type
,
1354 const glsl_type
*new_interface_type
)
1356 const glsl_type
*element_type
= type
->fields
.array
;
1357 if (element_type
->is_array()) {
1358 const glsl_type
*new_array_type
=
1359 update_interface_members_array(element_type
, new_interface_type
);
1360 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1362 return glsl_type::get_array_instance(new_interface_type
,
1368 * Determine whether the given interface type contains unsized arrays (if
1369 * it doesn't, array_sizing_visitor doesn't need to process it).
1371 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1373 for (unsigned i
= 0; i
< type
->length
; i
++) {
1374 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1375 if (elem_type
->is_unsized_array())
1382 * Create a new interface type based on the given type, with unsized arrays
1383 * replaced by sized arrays whose size is determined by
1384 * max_ifc_array_access.
1386 static const glsl_type
*
1387 resize_interface_members(const glsl_type
*type
,
1388 const unsigned *max_ifc_array_access
)
1390 unsigned num_fields
= type
->length
;
1391 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1392 memcpy(fields
, type
->fields
.structure
,
1393 num_fields
* sizeof(*fields
));
1394 for (unsigned i
= 0; i
< num_fields
; i
++) {
1395 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1397 glsl_interface_packing packing
=
1398 (glsl_interface_packing
) type
->interface_packing
;
1399 const glsl_type
*new_ifc_type
=
1400 glsl_type::get_interface_instance(fields
, num_fields
,
1401 packing
, type
->name
);
1403 return new_ifc_type
;
1406 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1409 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1410 ir_variable
**interface_vars
= (ir_variable
**) data
;
1411 unsigned num_fields
= ifc_type
->length
;
1412 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1413 memcpy(fields
, ifc_type
->fields
.structure
,
1414 num_fields
* sizeof(*fields
));
1415 bool interface_type_changed
= false;
1416 for (unsigned i
= 0; i
< num_fields
; i
++) {
1417 if (interface_vars
[i
] != NULL
&&
1418 fields
[i
].type
!= interface_vars
[i
]->type
) {
1419 fields
[i
].type
= interface_vars
[i
]->type
;
1420 interface_type_changed
= true;
1423 if (!interface_type_changed
) {
1427 glsl_interface_packing packing
=
1428 (glsl_interface_packing
) ifc_type
->interface_packing
;
1429 const glsl_type
*new_ifc_type
=
1430 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1433 for (unsigned i
= 0; i
< num_fields
; i
++) {
1434 if (interface_vars
[i
] != NULL
)
1435 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1440 * Memory context used to allocate the data in \c unnamed_interfaces.
1445 * Hash table from const glsl_type * to an array of ir_variable *'s
1446 * pointing to the ir_variables constituting each unnamed interface block.
1448 hash_table
*unnamed_interfaces
;
1453 * Performs the cross-validation of tessellation control shader vertices and
1454 * layout qualifiers for the attached tessellation control shaders,
1455 * and propagates them to the linked TCS and linked shader program.
1458 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1459 struct gl_shader
*linked_shader
,
1460 struct gl_shader
**shader_list
,
1461 unsigned num_shaders
)
1463 linked_shader
->TessCtrl
.VerticesOut
= 0;
1465 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1468 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1470 * "All tessellation control shader layout declarations in a program
1471 * must specify the same output patch vertex count. There must be at
1472 * least one layout qualifier specifying an output patch vertex count
1473 * in any program containing tessellation control shaders; however,
1474 * such a declaration is not required in all tessellation control
1478 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1479 struct gl_shader
*shader
= shader_list
[i
];
1481 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1482 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1483 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1484 linker_error(prog
, "tessellation control shader defined with "
1485 "conflicting output vertex count (%d and %d)\n",
1486 linked_shader
->TessCtrl
.VerticesOut
,
1487 shader
->TessCtrl
.VerticesOut
);
1490 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1494 /* Just do the intrastage -> interstage propagation right now,
1495 * since we already know we're in the right type of shader program
1498 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1499 linker_error(prog
, "tessellation control shader didn't declare "
1500 "vertices out layout qualifier\n");
1503 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1508 * Performs the cross-validation of tessellation evaluation shader
1509 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1510 * for the attached tessellation evaluation shaders, and propagates them
1511 * to the linked TES and linked shader program.
1514 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1515 struct gl_shader
*linked_shader
,
1516 struct gl_shader
**shader_list
,
1517 unsigned num_shaders
)
1519 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1520 linked_shader
->TessEval
.Spacing
= 0;
1521 linked_shader
->TessEval
.VertexOrder
= 0;
1522 linked_shader
->TessEval
.PointMode
= -1;
1524 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1527 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1529 * "At least one tessellation evaluation shader (compilation unit) in
1530 * a program must declare a primitive mode in its input layout.
1531 * Declaration vertex spacing, ordering, and point mode identifiers is
1532 * optional. It is not required that all tessellation evaluation
1533 * shaders in a program declare a primitive mode. If spacing or
1534 * vertex ordering declarations are omitted, the tessellation
1535 * primitive generator will use equal spacing or counter-clockwise
1536 * vertex ordering, respectively. If a point mode declaration is
1537 * omitted, the tessellation primitive generator will produce lines or
1538 * triangles according to the primitive mode."
1541 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1542 struct gl_shader
*shader
= shader_list
[i
];
1544 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1545 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1546 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1547 linker_error(prog
, "tessellation evaluation shader defined with "
1548 "conflicting input primitive modes.\n");
1551 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1554 if (shader
->TessEval
.Spacing
!= 0) {
1555 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1556 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1557 linker_error(prog
, "tessellation evaluation shader defined with "
1558 "conflicting vertex spacing.\n");
1561 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1564 if (shader
->TessEval
.VertexOrder
!= 0) {
1565 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1566 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1567 linker_error(prog
, "tessellation evaluation shader defined with "
1568 "conflicting ordering.\n");
1571 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1574 if (shader
->TessEval
.PointMode
!= -1) {
1575 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1576 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1577 linker_error(prog
, "tessellation evaluation shader defined with "
1578 "conflicting point modes.\n");
1581 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1586 /* Just do the intrastage -> interstage propagation right now,
1587 * since we already know we're in the right type of shader program
1590 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1592 "tessellation evaluation shader didn't declare input "
1593 "primitive modes.\n");
1596 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1598 if (linked_shader
->TessEval
.Spacing
== 0)
1599 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1600 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1602 if (linked_shader
->TessEval
.VertexOrder
== 0)
1603 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1604 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1606 if (linked_shader
->TessEval
.PointMode
== -1)
1607 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1608 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1613 * Performs the cross-validation of layout qualifiers specified in
1614 * redeclaration of gl_FragCoord for the attached fragment shaders,
1615 * and propagates them to the linked FS and linked shader program.
1618 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1619 struct gl_shader
*linked_shader
,
1620 struct gl_shader
**shader_list
,
1621 unsigned num_shaders
)
1623 linked_shader
->redeclares_gl_fragcoord
= false;
1624 linked_shader
->uses_gl_fragcoord
= false;
1625 linked_shader
->origin_upper_left
= false;
1626 linked_shader
->pixel_center_integer
= false;
1628 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1629 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1632 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1633 struct gl_shader
*shader
= shader_list
[i
];
1634 /* From the GLSL 1.50 spec, page 39:
1636 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1637 * it must be redeclared in all the fragment shaders in that program
1638 * that have a static use gl_FragCoord."
1640 if ((linked_shader
->redeclares_gl_fragcoord
1641 && !shader
->redeclares_gl_fragcoord
1642 && shader
->uses_gl_fragcoord
)
1643 || (shader
->redeclares_gl_fragcoord
1644 && !linked_shader
->redeclares_gl_fragcoord
1645 && linked_shader
->uses_gl_fragcoord
)) {
1646 linker_error(prog
, "fragment shader defined with conflicting "
1647 "layout qualifiers for gl_FragCoord\n");
1650 /* From the GLSL 1.50 spec, page 39:
1652 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1653 * single program must have the same set of qualifiers."
1655 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1656 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1657 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1658 linker_error(prog
, "fragment shader defined with conflicting "
1659 "layout qualifiers for gl_FragCoord\n");
1662 /* Update the linked shader state. Note that uses_gl_fragcoord should
1663 * accumulate the results. The other values should replace. If there
1664 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1665 * are already known to be the same.
1667 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1668 linked_shader
->redeclares_gl_fragcoord
=
1669 shader
->redeclares_gl_fragcoord
;
1670 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1671 || shader
->uses_gl_fragcoord
;
1672 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1673 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1676 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1681 * Performs the cross-validation of geometry shader max_vertices and
1682 * primitive type layout qualifiers for the attached geometry shaders,
1683 * and propagates them to the linked GS and linked shader program.
1686 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1687 struct gl_shader
*linked_shader
,
1688 struct gl_shader
**shader_list
,
1689 unsigned num_shaders
)
1691 linked_shader
->Geom
.VerticesOut
= 0;
1692 linked_shader
->Geom
.Invocations
= 0;
1693 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1694 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1696 /* No in/out qualifiers defined for anything but GLSL 1.50+
1697 * geometry shaders so far.
1699 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1702 /* From the GLSL 1.50 spec, page 46:
1704 * "All geometry shader output layout declarations in a program
1705 * must declare the same layout and same value for
1706 * max_vertices. There must be at least one geometry output
1707 * layout declaration somewhere in a program, but not all
1708 * geometry shaders (compilation units) are required to
1712 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1713 struct gl_shader
*shader
= shader_list
[i
];
1715 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1716 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1717 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1718 linker_error(prog
, "geometry shader defined with conflicting "
1722 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1725 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1726 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1727 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1728 linker_error(prog
, "geometry shader defined with conflicting "
1732 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1735 if (shader
->Geom
.VerticesOut
!= 0) {
1736 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1737 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1738 linker_error(prog
, "geometry shader defined with conflicting "
1739 "output vertex count (%d and %d)\n",
1740 linked_shader
->Geom
.VerticesOut
,
1741 shader
->Geom
.VerticesOut
);
1744 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1747 if (shader
->Geom
.Invocations
!= 0) {
1748 if (linked_shader
->Geom
.Invocations
!= 0 &&
1749 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1750 linker_error(prog
, "geometry shader defined with conflicting "
1751 "invocation count (%d and %d)\n",
1752 linked_shader
->Geom
.Invocations
,
1753 shader
->Geom
.Invocations
);
1756 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1760 /* Just do the intrastage -> interstage propagation right now,
1761 * since we already know we're in the right type of shader program
1764 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1766 "geometry shader didn't declare primitive input type\n");
1769 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1771 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1773 "geometry shader didn't declare primitive output type\n");
1776 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1778 if (linked_shader
->Geom
.VerticesOut
== 0) {
1780 "geometry shader didn't declare max_vertices\n");
1783 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1785 if (linked_shader
->Geom
.Invocations
== 0)
1786 linked_shader
->Geom
.Invocations
= 1;
1788 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1793 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1794 * qualifiers for the attached compute shaders, and propagate them to the
1795 * linked CS and linked shader program.
1798 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1799 struct gl_shader
*linked_shader
,
1800 struct gl_shader
**shader_list
,
1801 unsigned num_shaders
)
1803 for (int i
= 0; i
< 3; i
++)
1804 linked_shader
->Comp
.LocalSize
[i
] = 0;
1806 /* This function is called for all shader stages, but it only has an effect
1807 * for compute shaders.
1809 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1812 /* From the ARB_compute_shader spec, in the section describing local size
1815 * If multiple compute shaders attached to a single program object
1816 * declare local work-group size, the declarations must be identical;
1817 * otherwise a link-time error results. Furthermore, if a program
1818 * object contains any compute shaders, at least one must contain an
1819 * input layout qualifier specifying the local work sizes of the
1820 * program, or a link-time error will occur.
1822 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1823 struct gl_shader
*shader
= shader_list
[sh
];
1825 if (shader
->Comp
.LocalSize
[0] != 0) {
1826 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1827 for (int i
= 0; i
< 3; i
++) {
1828 if (linked_shader
->Comp
.LocalSize
[i
] !=
1829 shader
->Comp
.LocalSize
[i
]) {
1830 linker_error(prog
, "compute shader defined with conflicting "
1836 for (int i
= 0; i
< 3; i
++)
1837 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1841 /* Just do the intrastage -> interstage propagation right now,
1842 * since we already know we're in the right type of shader program
1845 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1846 linker_error(prog
, "compute shader didn't declare local size\n");
1849 for (int i
= 0; i
< 3; i
++)
1850 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1855 * Combine a group of shaders for a single stage to generate a linked shader
1858 * If this function is supplied a single shader, it is cloned, and the new
1859 * shader is returned.
1861 static struct gl_shader
*
1862 link_intrastage_shaders(void *mem_ctx
,
1863 struct gl_context
*ctx
,
1864 struct gl_shader_program
*prog
,
1865 struct gl_shader
**shader_list
,
1866 unsigned num_shaders
)
1868 struct gl_uniform_block
*uniform_blocks
= NULL
;
1870 /* Check that global variables defined in multiple shaders are consistent.
1872 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1873 if (!prog
->LinkStatus
)
1876 /* Check that interface blocks defined in multiple shaders are consistent.
1878 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1880 if (!prog
->LinkStatus
)
1883 /* Link up uniform blocks defined within this stage. */
1884 const unsigned num_uniform_blocks
=
1885 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1887 if (!prog
->LinkStatus
)
1890 /* Check that there is only a single definition of each function signature
1891 * across all shaders.
1893 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1894 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1895 ir_function
*const f
= node
->as_function();
1900 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1901 ir_function
*const other
=
1902 shader_list
[j
]->symbols
->get_function(f
->name
);
1904 /* If the other shader has no function (and therefore no function
1905 * signatures) with the same name, skip to the next shader.
1910 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1911 if (!sig
->is_defined
|| sig
->is_builtin())
1914 ir_function_signature
*other_sig
=
1915 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1917 if ((other_sig
!= NULL
) && other_sig
->is_defined
1918 && !other_sig
->is_builtin()) {
1919 linker_error(prog
, "function `%s' is multiply defined\n",
1928 /* Find the shader that defines main, and make a clone of it.
1930 * Starting with the clone, search for undefined references. If one is
1931 * found, find the shader that defines it. Clone the reference and add
1932 * it to the shader. Repeat until there are no undefined references or
1933 * until a reference cannot be resolved.
1935 gl_shader
*main
= NULL
;
1936 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1937 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1938 main
= shader_list
[i
];
1944 linker_error(prog
, "%s shader lacks `main'\n",
1945 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1949 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1950 linked
->ir
= new(linked
) exec_list
;
1951 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1953 linked
->UniformBlocks
= uniform_blocks
;
1954 linked
->NumUniformBlocks
= num_uniform_blocks
;
1955 ralloc_steal(linked
, linked
->UniformBlocks
);
1957 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1958 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1959 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1960 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1961 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1963 populate_symbol_table(linked
);
1965 /* The pointer to the main function in the final linked shader (i.e., the
1966 * copy of the original shader that contained the main function).
1968 ir_function_signature
*const main_sig
=
1969 link_get_main_function_signature(linked
);
1971 /* Move any instructions other than variable declarations or function
1972 * declarations into main.
1974 exec_node
*insertion_point
=
1975 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1978 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1979 if (shader_list
[i
] == main
)
1982 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1983 insertion_point
, true, linked
);
1986 /* Check if any shader needs built-in functions. */
1987 bool need_builtins
= false;
1988 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1989 if (shader_list
[i
]->uses_builtin_functions
) {
1990 need_builtins
= true;
1996 if (need_builtins
) {
1997 /* Make a temporary array one larger than shader_list, which will hold
1998 * the built-in function shader as well.
2000 gl_shader
**linking_shaders
= (gl_shader
**)
2001 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2003 ok
= linking_shaders
!= NULL
;
2006 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2007 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2009 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2011 free(linking_shaders
);
2013 _mesa_error_no_memory(__func__
);
2016 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2021 ctx
->Driver
.DeleteShader(ctx
, linked
);
2025 /* At this point linked should contain all of the linked IR, so
2026 * validate it to make sure nothing went wrong.
2028 validate_ir_tree(linked
->ir
);
2030 /* Set the size of geometry shader input arrays */
2031 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2032 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2033 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2034 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2035 ir
->accept(&input_resize_visitor
);
2039 if (ctx
->Const
.VertexID_is_zero_based
)
2040 lower_vertex_id(linked
);
2042 /* Make a pass over all variable declarations to ensure that arrays with
2043 * unspecified sizes have a size specified. The size is inferred from the
2044 * max_array_access field.
2046 array_sizing_visitor v
;
2048 v
.fixup_unnamed_interface_types();
2054 * Update the sizes of linked shader uniform arrays to the maximum
2057 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2059 * If one or more elements of an array are active,
2060 * GetActiveUniform will return the name of the array in name,
2061 * subject to the restrictions listed above. The type of the array
2062 * is returned in type. The size parameter contains the highest
2063 * array element index used, plus one. The compiler or linker
2064 * determines the highest index used. There will be only one
2065 * active uniform reported by the GL per uniform array.
2069 update_array_sizes(struct gl_shader_program
*prog
)
2071 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2072 if (prog
->_LinkedShaders
[i
] == NULL
)
2075 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2076 ir_variable
*const var
= node
->as_variable();
2078 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2079 !var
->type
->is_array())
2082 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2083 * will not be eliminated. Since we always do std140, just
2084 * don't resize arrays in UBOs.
2086 * Atomic counters are supposed to get deterministic
2087 * locations assigned based on the declaration ordering and
2088 * sizes, array compaction would mess that up.
2090 if (var
->is_in_buffer_block() || var
->type
->contains_atomic())
2093 unsigned int size
= var
->data
.max_array_access
;
2094 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2095 if (prog
->_LinkedShaders
[j
] == NULL
)
2098 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2099 ir_variable
*other_var
= node2
->as_variable();
2103 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2104 other_var
->data
.max_array_access
> size
) {
2105 size
= other_var
->data
.max_array_access
;
2110 if (size
+ 1 != var
->type
->length
) {
2111 /* If this is a built-in uniform (i.e., it's backed by some
2112 * fixed-function state), adjust the number of state slots to
2113 * match the new array size. The number of slots per array entry
2114 * is not known. It seems safe to assume that the total number of
2115 * slots is an integer multiple of the number of array elements.
2116 * Determine the number of slots per array element by dividing by
2117 * the old (total) size.
2119 const unsigned num_slots
= var
->get_num_state_slots();
2120 if (num_slots
> 0) {
2121 var
->set_num_state_slots((size
+ 1)
2122 * (num_slots
/ var
->type
->length
));
2125 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2127 /* FINISHME: We should update the types of array
2128 * dereferences of this variable now.
2136 * Resize tessellation evaluation per-vertex inputs to the size of
2137 * tessellation control per-vertex outputs.
2140 resize_tes_inputs(struct gl_context
*ctx
,
2141 struct gl_shader_program
*prog
)
2143 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2146 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2147 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2149 /* If no control shader is present, then the TES inputs are statically
2150 * sized to MaxPatchVertices; the actual size of the arrays won't be
2151 * known until draw time.
2153 const int num_vertices
= tcs
2154 ? tcs
->TessCtrl
.VerticesOut
2155 : ctx
->Const
.MaxPatchVertices
;
2157 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2158 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2159 ir
->accept(&input_resize_visitor
);
2164 * Find a contiguous set of available bits in a bitmask.
2166 * \param used_mask Bits representing used (1) and unused (0) locations
2167 * \param needed_count Number of contiguous bits needed.
2170 * Base location of the available bits on success or -1 on failure.
2173 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2175 unsigned needed_mask
= (1 << needed_count
) - 1;
2176 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2178 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2179 * cannot optimize possibly infinite loops" for the loop below.
2181 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2184 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2185 if ((needed_mask
& ~used_mask
) == needed_mask
)
2196 * Assign locations for either VS inputs or FS outputs
2198 * \param prog Shader program whose variables need locations assigned
2199 * \param target_index Selector for the program target to receive location
2200 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2201 * \c MESA_SHADER_FRAGMENT.
2202 * \param max_index Maximum number of generic locations. This corresponds
2203 * to either the maximum number of draw buffers or the
2204 * maximum number of generic attributes.
2207 * If locations are successfully assigned, true is returned. Otherwise an
2208 * error is emitted to the shader link log and false is returned.
2211 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2212 unsigned target_index
,
2215 /* Mark invalid locations as being used.
2217 unsigned used_locations
= (max_index
>= 32)
2218 ? ~0 : ~((1 << max_index
) - 1);
2220 assert((target_index
== MESA_SHADER_VERTEX
)
2221 || (target_index
== MESA_SHADER_FRAGMENT
));
2223 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2227 /* Operate in a total of four passes.
2229 * 1. Invalidate the location assignments for all vertex shader inputs.
2231 * 2. Assign locations for inputs that have user-defined (via
2232 * glBindVertexAttribLocation) locations and outputs that have
2233 * user-defined locations (via glBindFragDataLocation).
2235 * 3. Sort the attributes without assigned locations by number of slots
2236 * required in decreasing order. Fragmentation caused by attribute
2237 * locations assigned by the application may prevent large attributes
2238 * from having enough contiguous space.
2240 * 4. Assign locations to any inputs without assigned locations.
2243 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2244 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2246 const enum ir_variable_mode direction
=
2247 (target_index
== MESA_SHADER_VERTEX
)
2248 ? ir_var_shader_in
: ir_var_shader_out
;
2251 /* Temporary storage for the set of attributes that need locations assigned.
2257 /* Used below in the call to qsort. */
2258 static int compare(const void *a
, const void *b
)
2260 const temp_attr
*const l
= (const temp_attr
*) a
;
2261 const temp_attr
*const r
= (const temp_attr
*) b
;
2263 /* Reversed because we want a descending order sort below. */
2264 return r
->slots
- l
->slots
;
2268 unsigned num_attr
= 0;
2269 unsigned total_attribs_size
= 0;
2271 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2272 ir_variable
*const var
= node
->as_variable();
2274 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2277 if (var
->data
.explicit_location
) {
2278 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2279 || (var
->data
.location
< 0)) {
2281 "invalid explicit location %d specified for `%s'\n",
2282 (var
->data
.location
< 0)
2283 ? var
->data
.location
2284 : var
->data
.location
- generic_base
,
2288 } else if (target_index
== MESA_SHADER_VERTEX
) {
2291 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2292 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2293 var
->data
.location
= binding
;
2294 var
->data
.is_unmatched_generic_inout
= 0;
2296 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2300 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2301 assert(binding
>= FRAG_RESULT_DATA0
);
2302 var
->data
.location
= binding
;
2303 var
->data
.is_unmatched_generic_inout
= 0;
2305 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2306 var
->data
.index
= index
;
2311 const unsigned slots
= var
->type
->count_attribute_slots();
2313 /* From GL4.5 core spec, section 11.1.1 (Vertex Attributes):
2315 * "A program with more than the value of MAX_VERTEX_ATTRIBS active
2316 * attribute variables may fail to link, unless device-dependent
2317 * optimizations are able to make the program fit within available
2318 * hardware resources. For the purposes of this test, attribute variables
2319 * of the type dvec3, dvec4, dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3,
2320 * and dmat4 may count as consuming twice as many attributes as equivalent
2321 * single-precision types. While these types use the same number of
2322 * generic attributes as their single-precision equivalents,
2323 * implementations are permitted to consume two single-precision vectors
2324 * of internal storage for each three- or four-component double-precision
2326 * Until someone has a good reason in Mesa, enforce that now.
2328 if (target_index
== MESA_SHADER_VERTEX
) {
2329 total_attribs_size
+= slots
;
2330 if (var
->type
->without_array() == glsl_type::dvec3_type
||
2331 var
->type
->without_array() == glsl_type::dvec4_type
||
2332 var
->type
->without_array() == glsl_type::dmat2x3_type
||
2333 var
->type
->without_array() == glsl_type::dmat2x4_type
||
2334 var
->type
->without_array() == glsl_type::dmat3_type
||
2335 var
->type
->without_array() == glsl_type::dmat3x4_type
||
2336 var
->type
->without_array() == glsl_type::dmat4x3_type
||
2337 var
->type
->without_array() == glsl_type::dmat4_type
)
2338 total_attribs_size
+= slots
;
2341 /* If the variable is not a built-in and has a location statically
2342 * assigned in the shader (presumably via a layout qualifier), make sure
2343 * that it doesn't collide with other assigned locations. Otherwise,
2344 * add it to the list of variables that need linker-assigned locations.
2346 if (var
->data
.location
!= -1) {
2347 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2348 /* From page 61 of the OpenGL 4.0 spec:
2350 * "LinkProgram will fail if the attribute bindings assigned
2351 * by BindAttribLocation do not leave not enough space to
2352 * assign a location for an active matrix attribute or an
2353 * active attribute array, both of which require multiple
2354 * contiguous generic attributes."
2356 * I think above text prohibits the aliasing of explicit and
2357 * automatic assignments. But, aliasing is allowed in manual
2358 * assignments of attribute locations. See below comments for
2361 * From OpenGL 4.0 spec, page 61:
2363 * "It is possible for an application to bind more than one
2364 * attribute name to the same location. This is referred to as
2365 * aliasing. This will only work if only one of the aliased
2366 * attributes is active in the executable program, or if no
2367 * path through the shader consumes more than one attribute of
2368 * a set of attributes aliased to the same location. A link
2369 * error can occur if the linker determines that every path
2370 * through the shader consumes multiple aliased attributes,
2371 * but implementations are not required to generate an error
2374 * From GLSL 4.30 spec, page 54:
2376 * "A program will fail to link if any two non-vertex shader
2377 * input variables are assigned to the same location. For
2378 * vertex shaders, multiple input variables may be assigned
2379 * to the same location using either layout qualifiers or via
2380 * the OpenGL API. However, such aliasing is intended only to
2381 * support vertex shaders where each execution path accesses
2382 * at most one input per each location. Implementations are
2383 * permitted, but not required, to generate link-time errors
2384 * if they detect that every path through the vertex shader
2385 * executable accesses multiple inputs assigned to any single
2386 * location. For all shader types, a program will fail to link
2387 * if explicit location assignments leave the linker unable
2388 * to find space for other variables without explicit
2391 * From OpenGL ES 3.0 spec, page 56:
2393 * "Binding more than one attribute name to the same location
2394 * is referred to as aliasing, and is not permitted in OpenGL
2395 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2396 * fail when this condition exists. However, aliasing is
2397 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2398 * This will only work if only one of the aliased attributes
2399 * is active in the executable program, or if no path through
2400 * the shader consumes more than one attribute of a set of
2401 * attributes aliased to the same location. A link error can
2402 * occur if the linker determines that every path through the
2403 * shader consumes multiple aliased attributes, but implemen-
2404 * tations are not required to generate an error in this case."
2406 * After looking at above references from OpenGL, OpenGL ES and
2407 * GLSL specifications, we allow aliasing of vertex input variables
2408 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2410 * NOTE: This is not required by the spec but its worth mentioning
2411 * here that we're not doing anything to make sure that no path
2412 * through the vertex shader executable accesses multiple inputs
2413 * assigned to any single location.
2416 /* Mask representing the contiguous slots that will be used by
2419 const unsigned attr
= var
->data
.location
- generic_base
;
2420 const unsigned use_mask
= (1 << slots
) - 1;
2421 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2422 ? "vertex shader input" : "fragment shader output";
2424 /* Generate a link error if the requested locations for this
2425 * attribute exceed the maximum allowed attribute location.
2427 if (attr
+ slots
> max_index
) {
2429 "insufficient contiguous locations "
2430 "available for %s `%s' %d %d %d\n", string
,
2431 var
->name
, used_locations
, use_mask
, attr
);
2435 /* Generate a link error if the set of bits requested for this
2436 * attribute overlaps any previously allocated bits.
2438 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2439 if (target_index
== MESA_SHADER_FRAGMENT
||
2440 (prog
->IsES
&& prog
->Version
>= 300)) {
2442 "overlapping location is assigned "
2443 "to %s `%s' %d %d %d\n", string
,
2444 var
->name
, used_locations
, use_mask
, attr
);
2447 linker_warning(prog
,
2448 "overlapping location is assigned "
2449 "to %s `%s' %d %d %d\n", string
,
2450 var
->name
, used_locations
, use_mask
, attr
);
2454 used_locations
|= (use_mask
<< attr
);
2460 to_assign
[num_attr
].slots
= slots
;
2461 to_assign
[num_attr
].var
= var
;
2465 if (target_index
== MESA_SHADER_VERTEX
) {
2466 if (total_attribs_size
> max_index
) {
2468 "attempt to use %d vertex attribute slots only %d available ",
2469 total_attribs_size
, max_index
);
2474 /* If all of the attributes were assigned locations by the application (or
2475 * are built-in attributes with fixed locations), return early. This should
2476 * be the common case.
2481 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2483 if (target_index
== MESA_SHADER_VERTEX
) {
2484 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2485 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2486 * reserved to prevent it from being automatically allocated below.
2488 find_deref_visitor
find("gl_Vertex");
2490 if (find
.variable_found())
2491 used_locations
|= (1 << 0);
2494 for (unsigned i
= 0; i
< num_attr
; i
++) {
2495 /* Mask representing the contiguous slots that will be used by this
2498 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2500 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2503 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2504 ? "vertex shader input" : "fragment shader output";
2507 "insufficient contiguous locations "
2508 "available for %s `%s'\n",
2509 string
, to_assign
[i
].var
->name
);
2513 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2514 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2515 used_locations
|= (use_mask
<< location
);
2523 * Demote shader inputs and outputs that are not used in other stages
2526 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2528 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2529 ir_variable
*const var
= node
->as_variable();
2531 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2534 /* A shader 'in' or 'out' variable is only really an input or output if
2535 * its value is used by other shader stages. This will cause the variable
2536 * to have a location assigned.
2538 if (var
->data
.is_unmatched_generic_inout
) {
2539 assert(var
->data
.mode
!= ir_var_temporary
);
2540 var
->data
.mode
= ir_var_auto
;
2547 * Store the gl_FragDepth layout in the gl_shader_program struct.
2550 store_fragdepth_layout(struct gl_shader_program
*prog
)
2552 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2556 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2558 /* We don't look up the gl_FragDepth symbol directly because if
2559 * gl_FragDepth is not used in the shader, it's removed from the IR.
2560 * However, the symbol won't be removed from the symbol table.
2562 * We're only interested in the cases where the variable is NOT removed
2565 foreach_in_list(ir_instruction
, node
, ir
) {
2566 ir_variable
*const var
= node
->as_variable();
2568 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2572 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2573 switch (var
->data
.depth_layout
) {
2574 case ir_depth_layout_none
:
2575 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2577 case ir_depth_layout_any
:
2578 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2580 case ir_depth_layout_greater
:
2581 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2583 case ir_depth_layout_less
:
2584 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2586 case ir_depth_layout_unchanged
:
2587 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2598 * Validate the resources used by a program versus the implementation limits
2601 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2603 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2604 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2609 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2610 linker_error(prog
, "Too many %s shader texture samplers\n",
2611 _mesa_shader_stage_to_string(i
));
2614 if (sh
->num_uniform_components
>
2615 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2616 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2617 linker_warning(prog
, "Too many %s shader default uniform block "
2618 "components, but the driver will try to optimize "
2619 "them out; this is non-portable out-of-spec "
2621 _mesa_shader_stage_to_string(i
));
2623 linker_error(prog
, "Too many %s shader default uniform block "
2625 _mesa_shader_stage_to_string(i
));
2629 if (sh
->num_combined_uniform_components
>
2630 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2631 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2632 linker_warning(prog
, "Too many %s shader uniform components, "
2633 "but the driver will try to optimize them out; "
2634 "this is non-portable out-of-spec behavior\n",
2635 _mesa_shader_stage_to_string(i
));
2637 linker_error(prog
, "Too many %s shader uniform components\n",
2638 _mesa_shader_stage_to_string(i
));
2643 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2644 unsigned total_uniform_blocks
= 0;
2646 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2647 if (prog
->UniformBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2648 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2649 prog
->UniformBlocks
[i
].Name
,
2650 prog
->UniformBlocks
[i
].UniformBufferSize
,
2651 ctx
->Const
.MaxUniformBlockSize
);
2654 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2655 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2657 total_uniform_blocks
++;
2661 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2662 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2663 prog
->NumUniformBlocks
,
2664 ctx
->Const
.MaxCombinedUniformBlocks
);
2666 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2667 const unsigned max_uniform_blocks
=
2668 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2669 if (blocks
[i
] > max_uniform_blocks
) {
2670 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2671 _mesa_shader_stage_to_string(i
),
2673 max_uniform_blocks
);
2682 * Validate shader image resources.
2685 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2687 unsigned total_image_units
= 0;
2688 unsigned fragment_outputs
= 0;
2690 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2693 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2694 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2697 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2698 linker_error(prog
, "Too many %s shader image uniforms\n",
2699 _mesa_shader_stage_to_string(i
));
2701 total_image_units
+= sh
->NumImages
;
2703 if (i
== MESA_SHADER_FRAGMENT
) {
2704 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2705 ir_variable
*var
= node
->as_variable();
2706 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2707 fragment_outputs
+= var
->type
->count_attribute_slots();
2713 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2714 linker_error(prog
, "Too many combined image uniforms\n");
2716 if (total_image_units
+ fragment_outputs
>
2717 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2718 linker_error(prog
, "Too many combined image uniforms and fragment outputs\n");
2723 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2724 * for a variable, checks for overlaps between other uniforms using explicit
2728 reserve_explicit_locations(struct gl_shader_program
*prog
,
2729 string_to_uint_map
*map
, ir_variable
*var
)
2731 unsigned slots
= var
->type
->uniform_locations();
2732 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2734 /* Resize remap table if locations do not fit in the current one. */
2735 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2736 prog
->UniformRemapTable
=
2737 reralloc(prog
, prog
->UniformRemapTable
,
2738 gl_uniform_storage
*,
2741 if (!prog
->UniformRemapTable
) {
2742 linker_error(prog
, "Out of memory during linking.\n");
2746 /* Initialize allocated space. */
2747 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2748 prog
->UniformRemapTable
[i
] = NULL
;
2750 prog
->NumUniformRemapTable
= max_loc
+ 1;
2753 for (unsigned i
= 0; i
< slots
; i
++) {
2754 unsigned loc
= var
->data
.location
+ i
;
2756 /* Check if location is already used. */
2757 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2759 /* Possibly same uniform from a different stage, this is ok. */
2761 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2764 /* ARB_explicit_uniform_location specification states:
2766 * "No two default-block uniform variables in the program can have
2767 * the same location, even if they are unused, otherwise a compiler
2768 * or linker error will be generated."
2771 "location qualifier for uniform %s overlaps "
2772 "previously used location\n",
2777 /* Initialize location as inactive before optimization
2778 * rounds and location assignment.
2780 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2783 /* Note, base location used for arrays. */
2784 map
->put(var
->data
.location
, var
->name
);
2790 * Check and reserve all explicit uniform locations, called before
2791 * any optimizations happen to handle also inactive uniforms and
2792 * inactive array elements that may get trimmed away.
2795 check_explicit_uniform_locations(struct gl_context
*ctx
,
2796 struct gl_shader_program
*prog
)
2798 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2801 /* This map is used to detect if overlapping explicit locations
2802 * occur with the same uniform (from different stage) or a different one.
2804 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2807 linker_error(prog
, "Out of memory during linking.\n");
2811 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2812 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2817 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2818 ir_variable
*var
= node
->as_variable();
2819 if (var
&& (var
->data
.mode
== ir_var_uniform
|| var
->data
.mode
== ir_var_shader_storage
) &&
2820 var
->data
.explicit_location
) {
2821 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2833 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
2834 const void *data
, uint8_t stages
)
2838 /* If resource already exists, do not add it again. */
2839 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
2840 if (prog
->ProgramResourceList
[i
].Data
== data
)
2843 prog
->ProgramResourceList
=
2845 prog
->ProgramResourceList
,
2846 gl_program_resource
,
2847 prog
->NumProgramResourceList
+ 1);
2849 if (!prog
->ProgramResourceList
) {
2850 linker_error(prog
, "Out of memory during linking.\n");
2854 struct gl_program_resource
*res
=
2855 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
2859 res
->StageReferences
= stages
;
2861 prog
->NumProgramResourceList
++;
2867 * Function builds a stage reference bitmask from variable name.
2870 build_stageref(struct gl_shader_program
*shProg
, const char *name
)
2874 /* Note, that we assume MAX 8 stages, if there will be more stages, type
2875 * used for reference mask in gl_program_resource will need to be changed.
2877 assert(MESA_SHADER_STAGES
< 8);
2879 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2880 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
2884 /* Shader symbol table may contain variables that have
2885 * been optimized away. Search IR for the variable instead.
2887 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2888 ir_variable
*var
= node
->as_variable();
2889 if (var
&& strcmp(var
->name
, name
) == 0) {
2899 add_interface_variables(struct gl_shader_program
*shProg
,
2900 struct gl_shader
*sh
, GLenum programInterface
)
2902 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2903 ir_variable
*var
= node
->as_variable();
2909 switch (var
->data
.mode
) {
2910 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
2911 * "For GetActiveAttrib, all active vertex shader input variables
2912 * are enumerated, including the special built-in inputs gl_VertexID
2913 * and gl_InstanceID."
2915 case ir_var_system_value
:
2916 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
2917 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
2918 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
2920 /* Mark special built-in inputs referenced by the vertex stage so
2921 * that they are considered active by the shader queries.
2923 mask
= (1 << (MESA_SHADER_VERTEX
));
2925 case ir_var_shader_in
:
2926 if (programInterface
!= GL_PROGRAM_INPUT
)
2929 case ir_var_shader_out
:
2930 if (programInterface
!= GL_PROGRAM_OUTPUT
)
2937 if (!add_program_resource(shProg
, programInterface
, var
,
2938 build_stageref(shProg
, var
->name
) | mask
))
2945 * Builds up a list of program resources that point to existing
2949 build_program_resource_list(struct gl_context
*ctx
,
2950 struct gl_shader_program
*shProg
)
2952 /* Rebuild resource list. */
2953 if (shProg
->ProgramResourceList
) {
2954 ralloc_free(shProg
->ProgramResourceList
);
2955 shProg
->ProgramResourceList
= NULL
;
2956 shProg
->NumProgramResourceList
= 0;
2959 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
2961 /* Determine first input and final output stage. These are used to
2962 * detect which variables should be enumerated in the resource list
2963 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
2965 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2966 if (!shProg
->_LinkedShaders
[i
])
2968 if (input_stage
== MESA_SHADER_STAGES
)
2973 /* Empty shader, no resources. */
2974 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
2977 /* Add inputs and outputs to the resource list. */
2978 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
],
2982 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
],
2986 /* Add transform feedback varyings. */
2987 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
2988 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
2990 build_stageref(shProg
,
2991 shProg
->LinkedTransformFeedback
.Varyings
[i
].Name
);
2992 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
2993 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
2999 /* Add uniforms from uniform storage. */
3000 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3001 /* Do not add uniforms internally used by Mesa. */
3002 if (shProg
->UniformStorage
[i
].hidden
)
3006 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
);
3008 /* Add stagereferences for uniforms in a uniform block. */
3009 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3010 if (block_index
!= -1) {
3011 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
3012 if (shProg
->UniformBlockStageIndex
[j
][block_index
] != -1)
3013 stageref
|= (1 << j
);
3017 if (!add_program_resource(shProg
, GL_UNIFORM
,
3018 &shProg
->UniformStorage
[i
], stageref
))
3022 /* Add program uniform blocks. */
3023 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
3024 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
3025 &shProg
->UniformBlocks
[i
], 0))
3029 /* Add atomic counter buffers. */
3030 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3031 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3032 &shProg
->AtomicBuffers
[i
], 0))
3036 /* TODO - following extensions will require more resource types:
3038 * GL_ARB_shader_storage_buffer_object
3039 * GL_ARB_shader_subroutine
3044 * This check is done to make sure we allow only constant expression
3045 * indexing and "constant-index-expression" (indexing with an expression
3046 * that includes loop induction variable).
3049 validate_sampler_array_indexing(struct gl_context
*ctx
,
3050 struct gl_shader_program
*prog
)
3052 dynamic_sampler_array_indexing_visitor v
;
3053 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3054 if (prog
->_LinkedShaders
[i
] == NULL
)
3057 bool no_dynamic_indexing
=
3058 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
3060 /* Search for array derefs in shader. */
3061 v
.run(prog
->_LinkedShaders
[i
]->ir
);
3062 if (v
.uses_dynamic_sampler_array_indexing()) {
3063 const char *msg
= "sampler arrays indexed with non-constant "
3064 "expressions is forbidden in GLSL %s %u";
3065 /* Backend has indicated that it has no dynamic indexing support. */
3066 if (no_dynamic_indexing
) {
3067 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3070 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3079 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3081 tfeedback_decl
*tfeedback_decls
= NULL
;
3082 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
3084 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
3086 prog
->LinkStatus
= true; /* All error paths will set this to false */
3087 prog
->Validated
= false;
3088 prog
->_Used
= false;
3090 prog
->ARB_fragment_coord_conventions_enable
= false;
3092 /* Separate the shaders into groups based on their type.
3094 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
3095 unsigned num_shaders
[MESA_SHADER_STAGES
];
3097 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3098 shader_list
[i
] = (struct gl_shader
**)
3099 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
3103 unsigned min_version
= UINT_MAX
;
3104 unsigned max_version
= 0;
3105 const bool is_es_prog
=
3106 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
3107 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
3108 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
3109 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
3111 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
3112 linker_error(prog
, "all shaders must use same shading "
3113 "language version\n");
3117 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
3118 prog
->ARB_fragment_coord_conventions_enable
= true;
3121 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
3122 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
3123 num_shaders
[shader_type
]++;
3126 /* In desktop GLSL, different shader versions may be linked together. In
3127 * GLSL ES, all shader versions must be the same.
3129 if (is_es_prog
&& min_version
!= max_version
) {
3130 linker_error(prog
, "all shaders must use same shading "
3131 "language version\n");
3135 prog
->Version
= max_version
;
3136 prog
->IsES
= is_es_prog
;
3138 /* Some shaders have to be linked with some other shaders present.
3140 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
3141 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3142 !prog
->SeparateShader
) {
3143 linker_error(prog
, "Geometry shader must be linked with "
3147 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
3148 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3149 !prog
->SeparateShader
) {
3150 linker_error(prog
, "Tessellation evaluation shader must be linked with "
3154 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3155 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3156 !prog
->SeparateShader
) {
3157 linker_error(prog
, "Tessellation control shader must be linked with "
3162 /* The spec is self-contradictory here. It allows linking without a tess
3163 * eval shader, but that can only be used with transform feedback and
3164 * rasterization disabled. However, transform feedback isn't allowed
3165 * with GL_PATCHES, so it can't be used.
3167 * More investigation showed that the idea of transform feedback after
3168 * a tess control shader was dropped, because some hw vendors couldn't
3169 * support tessellation without a tess eval shader, but the linker section
3170 * wasn't updated to reflect that.
3172 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
3175 * Do what's reasonable and always require a tess eval shader if a tess
3176 * control shader is present.
3178 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3179 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0 &&
3180 !prog
->SeparateShader
) {
3181 linker_error(prog
, "Tessellation control shader must be linked with "
3182 "tessellation evaluation shader\n");
3186 /* Compute shaders have additional restrictions. */
3187 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
3188 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
3189 linker_error(prog
, "Compute shaders may not be linked with any other "
3190 "type of shader\n");
3193 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3194 if (prog
->_LinkedShaders
[i
] != NULL
)
3195 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
3197 prog
->_LinkedShaders
[i
] = NULL
;
3200 /* Link all shaders for a particular stage and validate the result.
3202 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
3203 if (num_shaders
[stage
] > 0) {
3204 gl_shader
*const sh
=
3205 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
3206 num_shaders
[stage
]);
3208 if (!prog
->LinkStatus
) {
3210 ctx
->Driver
.DeleteShader(ctx
, sh
);
3215 case MESA_SHADER_VERTEX
:
3216 validate_vertex_shader_executable(prog
, sh
);
3218 case MESA_SHADER_GEOMETRY
:
3219 validate_geometry_shader_executable(prog
, sh
);
3221 case MESA_SHADER_FRAGMENT
:
3222 validate_fragment_shader_executable(prog
, sh
);
3225 if (!prog
->LinkStatus
) {
3227 ctx
->Driver
.DeleteShader(ctx
, sh
);
3231 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
3235 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
3236 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
3237 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
3238 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
3240 prog
->LastClipDistanceArraySize
= 0; /* Not used */
3242 /* Here begins the inter-stage linking phase. Some initial validation is
3243 * performed, then locations are assigned for uniforms, attributes, and
3246 cross_validate_uniforms(prog
);
3247 if (!prog
->LinkStatus
)
3252 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
3253 if (prog
->_LinkedShaders
[prev
] != NULL
)
3257 check_explicit_uniform_locations(ctx
, prog
);
3258 if (!prog
->LinkStatus
)
3261 resize_tes_inputs(ctx
, prog
);
3263 /* Validate the inputs of each stage with the output of the preceding
3266 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3267 if (prog
->_LinkedShaders
[i
] == NULL
)
3270 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
3271 prog
->_LinkedShaders
[i
]);
3272 if (!prog
->LinkStatus
)
3275 cross_validate_outputs_to_inputs(prog
,
3276 prog
->_LinkedShaders
[prev
],
3277 prog
->_LinkedShaders
[i
]);
3278 if (!prog
->LinkStatus
)
3284 /* Cross-validate uniform blocks between shader stages */
3285 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
3286 MESA_SHADER_STAGES
);
3287 if (!prog
->LinkStatus
)
3290 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3291 if (prog
->_LinkedShaders
[i
] != NULL
)
3292 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
3295 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
3296 * it before optimization because we want most of the checks to get
3297 * dropped thanks to constant propagation.
3299 * This rule also applies to GLSL ES 3.00.
3301 if (max_version
>= (is_es_prog
? 300 : 130)) {
3302 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3304 lower_discard_flow(sh
->ir
);
3308 if (!interstage_cross_validate_uniform_blocks(prog
))
3311 /* Do common optimization before assigning storage for attributes,
3312 * uniforms, and varyings. Later optimization could possibly make
3313 * some of that unused.
3315 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3316 if (prog
->_LinkedShaders
[i
] == NULL
)
3319 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
3320 if (!prog
->LinkStatus
)
3323 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
3324 lower_clip_distance(prog
->_LinkedShaders
[i
]);
3327 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
3328 &ctx
->Const
.ShaderCompilerOptions
[i
],
3329 ctx
->Const
.NativeIntegers
))
3332 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
3335 /* Validation for special cases where we allow sampler array indexing
3336 * with loop induction variable. This check emits a warning or error
3337 * depending if backend can handle dynamic indexing.
3339 if ((!prog
->IsES
&& prog
->Version
< 130) ||
3340 (prog
->IsES
&& prog
->Version
< 300)) {
3341 if (!validate_sampler_array_indexing(ctx
, prog
))
3345 /* Check and validate stream emissions in geometry shaders */
3346 validate_geometry_shader_emissions(ctx
, prog
);
3348 /* Mark all generic shader inputs and outputs as unpaired. */
3349 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3350 if (prog
->_LinkedShaders
[i
] != NULL
) {
3351 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
3355 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
,
3356 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAttribs
)) {
3360 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
3364 unsigned first
, last
;
3366 first
= MESA_SHADER_STAGES
;
3369 /* Determine first and last stage. */
3370 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3371 if (!prog
->_LinkedShaders
[i
])
3373 if (first
== MESA_SHADER_STAGES
)
3378 if (num_tfeedback_decls
!= 0) {
3379 /* From GL_EXT_transform_feedback:
3380 * A program will fail to link if:
3382 * * the <count> specified by TransformFeedbackVaryingsEXT is
3383 * non-zero, but the program object has no vertex or geometry
3386 if (first
== MESA_SHADER_FRAGMENT
) {
3387 linker_error(prog
, "Transform feedback varyings specified, but "
3388 "no vertex or geometry shader is present.\n");
3392 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
3393 prog
->TransformFeedback
.NumVarying
);
3394 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
3395 prog
->TransformFeedback
.VaryingNames
,
3400 /* Linking the stages in the opposite order (from fragment to vertex)
3401 * ensures that inter-shader outputs written to in an earlier stage are
3402 * eliminated if they are (transitively) not used in a later stage.
3406 if (first
< MESA_SHADER_FRAGMENT
) {
3407 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
3409 if (first
== MESA_SHADER_GEOMETRY
) {
3410 /* There was no vertex shader, but we still have to assign varying
3411 * locations for use by geometry shader inputs in SSO.
3413 * If the shader is not separable (i.e., prog->SeparateShader is
3414 * false), linking will have already failed when first is
3415 * MESA_SHADER_GEOMETRY.
3417 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3418 NULL
, prog
->_LinkedShaders
[first
],
3419 num_tfeedback_decls
, tfeedback_decls
,
3420 prog
->Geom
.VerticesIn
))
3424 if (last
!= MESA_SHADER_FRAGMENT
&&
3425 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
3426 /* There was no fragment shader, but we still have to assign varying
3427 * locations for use by transform feedback.
3429 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3431 num_tfeedback_decls
, tfeedback_decls
,
3436 do_dead_builtin_varyings(ctx
, sh
, NULL
,
3437 num_tfeedback_decls
, tfeedback_decls
);
3439 if (!prog
->SeparateShader
)
3440 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
3442 /* Eliminate code that is now dead due to unused outputs being demoted.
3444 while (do_dead_code(sh
->ir
, false))
3447 else if (first
== MESA_SHADER_FRAGMENT
) {
3448 /* If the program only contains a fragment shader...
3450 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
3452 do_dead_builtin_varyings(ctx
, NULL
, sh
,
3453 num_tfeedback_decls
, tfeedback_decls
);
3455 if (prog
->SeparateShader
) {
3456 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3457 NULL
/* producer */,
3459 0 /* num_tfeedback_decls */,
3460 NULL
/* tfeedback_decls */,
3461 0 /* gs_input_vertices */))
3464 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
3466 while (do_dead_code(sh
->ir
, false))
3471 for (int i
= next
- 1; i
>= 0; i
--) {
3472 if (prog
->_LinkedShaders
[i
] == NULL
)
3475 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
3476 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
3477 unsigned gs_input_vertices
=
3478 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
3480 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
3481 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3482 tfeedback_decls
, gs_input_vertices
))
3485 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
3486 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3489 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
3490 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
3492 /* Eliminate code that is now dead due to unused outputs being demoted.
3494 while (do_dead_code(sh_i
->ir
, false))
3496 while (do_dead_code(sh_next
->ir
, false))
3499 /* This must be done after all dead varyings are eliminated. */
3500 if (!check_against_output_limit(ctx
, prog
, sh_i
))
3502 if (!check_against_input_limit(ctx
, prog
, sh_next
))
3508 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
3511 update_array_sizes(prog
);
3512 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
3513 link_assign_atomic_counter_resources(ctx
, prog
);
3514 store_fragdepth_layout(prog
);
3516 check_resources(ctx
, prog
);
3517 check_image_resources(ctx
, prog
);
3518 link_check_atomic_counter_resources(ctx
, prog
);
3520 if (!prog
->LinkStatus
)
3523 /* OpenGL ES requires that a vertex shader and a fragment shader both be
3524 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
3525 * anything about shader linking when one of the shaders (vertex or
3526 * fragment shader) is absent. So, the extension shouldn't change the
3527 * behavior specified in GLSL specification.
3529 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
3530 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
3531 linker_error(prog
, "program lacks a vertex shader\n");
3532 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3533 linker_error(prog
, "program lacks a fragment shader\n");
3537 /* FINISHME: Assign fragment shader output locations. */
3540 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3541 free(shader_list
[i
]);
3542 if (prog
->_LinkedShaders
[i
] == NULL
)
3545 /* Do a final validation step to make sure that the IR wasn't
3546 * invalidated by any modifications performed after intrastage linking.
3548 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
3550 /* Retain any live IR, but trash the rest. */
3551 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
3553 /* The symbol table in the linked shaders may contain references to
3554 * variables that were removed (e.g., unused uniforms). Since it may
3555 * contain junk, there is no possible valid use. Delete it and set the
3558 delete prog
->_LinkedShaders
[i
]->symbols
;
3559 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
3562 ralloc_free(mem_ctx
);