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 "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/prog_instruction.h"
75 #include "program/program.h"
76 #include "util/mesa-sha1.h"
78 #include "string_to_uint_map.h"
80 #include "link_varyings.h"
81 #include "ir_optimization.h"
82 #include "ir_rvalue_visitor.h"
83 #include "ir_uniform.h"
84 #include "builtin_functions.h"
85 #include "shader_cache.h"
87 #include "main/shaderobj.h"
88 #include "main/enums.h"
93 struct find_variable
{
97 find_variable(const char *name
) : name(name
), found(false) {}
101 * Visitor that determines whether or not a variable is ever written.
103 * Use \ref find_assignments for convenience.
105 class find_assignment_visitor
: public ir_hierarchical_visitor
{
107 find_assignment_visitor(unsigned num_vars
,
108 find_variable
* const *vars
)
109 : num_variables(num_vars
), num_found(0), variables(vars
)
113 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
115 ir_variable
*const var
= ir
->lhs
->variable_referenced();
117 return check_variable_name(var
->name
);
120 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
122 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
123 actual_node
, &ir
->actual_parameters
) {
124 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
125 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
127 if (sig_param
->data
.mode
== ir_var_function_out
||
128 sig_param
->data
.mode
== ir_var_function_inout
) {
129 ir_variable
*var
= param_rval
->variable_referenced();
130 if (var
&& check_variable_name(var
->name
) == visit_stop
)
135 if (ir
->return_deref
!= NULL
) {
136 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
138 if (check_variable_name(var
->name
) == visit_stop
)
142 return visit_continue_with_parent
;
146 ir_visitor_status
check_variable_name(const char *name
)
148 for (unsigned i
= 0; i
< num_variables
; ++i
) {
149 if (strcmp(variables
[i
]->name
, name
) == 0) {
150 if (!variables
[i
]->found
) {
151 variables
[i
]->found
= true;
153 assert(num_found
< num_variables
);
154 if (++num_found
== num_variables
)
161 return visit_continue_with_parent
;
165 unsigned num_variables
; /**< Number of variables to find */
166 unsigned num_found
; /**< Number of variables already found */
167 find_variable
* const *variables
; /**< Variables to find */
171 * Determine whether or not any of NULL-terminated list of variables is ever
175 find_assignments(exec_list
*ir
, find_variable
* const *vars
)
177 unsigned num_variables
= 0;
179 for (find_variable
* const *v
= vars
; *v
; ++v
)
182 find_assignment_visitor
visitor(num_variables
, vars
);
187 * Determine whether or not the given variable is ever written to.
190 find_assignments(exec_list
*ir
, find_variable
*var
)
192 find_assignment_visitor
visitor(1, &var
);
197 * Visitor that determines whether or not a variable is ever read.
199 class find_deref_visitor
: public ir_hierarchical_visitor
{
201 find_deref_visitor(const char *name
)
202 : name(name
), found(false)
207 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
209 if (strcmp(this->name
, ir
->var
->name
) == 0) {
214 return visit_continue
;
217 bool variable_found() const
223 const char *name
; /**< Find writes to a variable with this name. */
224 bool found
; /**< Was a write to the variable found? */
229 * A visitor helper that provides methods for updating the types of
230 * ir_dereferences. Classes that update variable types (say, updating
231 * array sizes) will want to use this so that dereference types stay in sync.
233 class deref_type_updater
: public ir_hierarchical_visitor
{
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
243 const glsl_type
*const vt
= ir
->array
->type
;
245 ir
->type
= vt
->fields
.array
;
246 return visit_continue
;
249 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
251 ir
->type
= ir
->record
->type
->fields
.structure
[ir
->field_idx
].type
;
252 return visit_continue
;
257 class array_resize_visitor
: public deref_type_updater
{
259 unsigned num_vertices
;
260 gl_shader_program
*prog
;
261 gl_shader_stage stage
;
263 array_resize_visitor(unsigned num_vertices
,
264 gl_shader_program
*prog
,
265 gl_shader_stage stage
)
267 this->num_vertices
= num_vertices
;
272 virtual ~array_resize_visitor()
277 virtual ir_visitor_status
visit(ir_variable
*var
)
279 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
281 return visit_continue
;
283 unsigned size
= var
->type
->length
;
285 if (stage
== MESA_SHADER_GEOMETRY
) {
286 /* Generate a link error if the shader has declared this array with
289 if (!var
->data
.implicit_sized_array
&&
290 size
&& size
!= this->num_vertices
) {
291 linker_error(this->prog
, "size of array %s declared as %u, "
292 "but number of input vertices is %u\n",
293 var
->name
, size
, this->num_vertices
);
294 return visit_continue
;
297 /* Generate a link error if the shader attempts to access an input
298 * array using an index too large for its actual size assigned at
301 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
302 linker_error(this->prog
, "%s shader accesses element %i of "
303 "%s, but only %i input vertices\n",
304 _mesa_shader_stage_to_string(this->stage
),
305 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
306 return visit_continue
;
310 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
312 var
->data
.max_array_access
= this->num_vertices
- 1;
314 return visit_continue
;
319 * Visitor that determines the highest stream id to which a (geometry) shader
320 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
322 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
324 find_emit_vertex_visitor(int max_allowed
)
325 : max_stream_allowed(max_allowed
),
326 invalid_stream_id(0),
327 invalid_stream_id_from_emit_vertex(false),
328 end_primitive_found(false),
329 uses_non_zero_stream(false)
334 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
336 int stream_id
= ir
->stream_id();
339 invalid_stream_id
= stream_id
;
340 invalid_stream_id_from_emit_vertex
= true;
344 if (stream_id
> max_stream_allowed
) {
345 invalid_stream_id
= stream_id
;
346 invalid_stream_id_from_emit_vertex
= true;
351 uses_non_zero_stream
= true;
353 return visit_continue
;
356 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
358 end_primitive_found
= true;
360 int stream_id
= ir
->stream_id();
363 invalid_stream_id
= stream_id
;
364 invalid_stream_id_from_emit_vertex
= false;
368 if (stream_id
> max_stream_allowed
) {
369 invalid_stream_id
= stream_id
;
370 invalid_stream_id_from_emit_vertex
= false;
375 uses_non_zero_stream
= true;
377 return visit_continue
;
382 return invalid_stream_id
!= 0;
385 const char *error_func()
387 return invalid_stream_id_from_emit_vertex
?
388 "EmitStreamVertex" : "EndStreamPrimitive";
393 return invalid_stream_id
;
398 return uses_non_zero_stream
;
401 bool uses_end_primitive()
403 return end_primitive_found
;
407 int max_stream_allowed
;
408 int invalid_stream_id
;
409 bool invalid_stream_id_from_emit_vertex
;
410 bool end_primitive_found
;
411 bool uses_non_zero_stream
;
414 /* Class that finds array derefs and check if indexes are dynamic. */
415 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
418 dynamic_sampler_array_indexing_visitor() :
419 dynamic_sampler_array_indexing(false)
423 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
425 if (!ir
->variable_referenced())
426 return visit_continue
;
428 if (!ir
->variable_referenced()->type
->contains_sampler())
429 return visit_continue
;
431 if (!ir
->array_index
->constant_expression_value(ralloc_parent(ir
))) {
432 dynamic_sampler_array_indexing
= true;
435 return visit_continue
;
438 bool uses_dynamic_sampler_array_indexing()
440 return dynamic_sampler_array_indexing
;
444 bool dynamic_sampler_array_indexing
;
447 } /* anonymous namespace */
450 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
454 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
456 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
459 prog
->data
->LinkStatus
= linking_failure
;
464 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
468 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
470 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
477 * Given a string identifying a program resource, break it into a base name
478 * and an optional array index in square brackets.
480 * If an array index is present, \c out_base_name_end is set to point to the
481 * "[" that precedes the array index, and the array index itself is returned
484 * If no array index is present (or if the array index is negative or
485 * mal-formed), \c out_base_name_end, is set to point to the null terminator
486 * at the end of the input string, and -1 is returned.
488 * Only the final array index is parsed; if the string contains other array
489 * indices (or structure field accesses), they are left in the base name.
491 * No attempt is made to check that the base name is properly formed;
492 * typically the caller will look up the base name in a hash table, so
493 * ill-formed base names simply turn into hash table lookup failures.
496 parse_program_resource_name(const GLchar
*name
,
497 const GLchar
**out_base_name_end
)
499 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
501 * "When an integer array element or block instance number is part of
502 * the name string, it will be specified in decimal form without a "+"
503 * or "-" sign or any extra leading zeroes. Additionally, the name
504 * string will not include white space anywhere in the string."
507 const size_t len
= strlen(name
);
508 *out_base_name_end
= name
+ len
;
510 if (len
== 0 || name
[len
-1] != ']')
513 /* Walk backwards over the string looking for a non-digit character. This
514 * had better be the opening bracket for an array index.
516 * Initially, i specifies the location of the ']'. Since the string may
517 * contain only the ']' charcater, walk backwards very carefully.
520 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
523 if ((i
== 0) || name
[i
-1] != '[')
526 long array_index
= strtol(&name
[i
], NULL
, 10);
530 /* Check for leading zero */
531 if (name
[i
] == '0' && name
[i
+1] != ']')
534 *out_base_name_end
= name
+ (i
- 1);
540 link_invalidate_variable_locations(exec_list
*ir
)
542 foreach_in_list(ir_instruction
, node
, ir
) {
543 ir_variable
*const var
= node
->as_variable();
548 /* Only assign locations for variables that lack an explicit location.
549 * Explicit locations are set for all built-in variables, generic vertex
550 * shader inputs (via layout(location=...)), and generic fragment shader
551 * outputs (also via layout(location=...)).
553 if (!var
->data
.explicit_location
) {
554 var
->data
.location
= -1;
555 var
->data
.location_frac
= 0;
558 /* ir_variable::is_unmatched_generic_inout is used by the linker while
559 * connecting outputs from one stage to inputs of the next stage.
561 if (var
->data
.explicit_location
&&
562 var
->data
.location
< VARYING_SLOT_VAR0
) {
563 var
->data
.is_unmatched_generic_inout
= 0;
565 var
->data
.is_unmatched_generic_inout
= 1;
572 * Set clip_distance_array_size based and cull_distance_array_size on the given
575 * Also check for errors based on incorrect usage of gl_ClipVertex and
576 * gl_ClipDistance and gl_CullDistance.
577 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
578 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
580 * Return false if an error was reported.
583 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
584 struct gl_linked_shader
*shader
,
585 struct gl_context
*ctx
,
586 GLuint
*clip_distance_array_size
,
587 GLuint
*cull_distance_array_size
)
589 *clip_distance_array_size
= 0;
590 *cull_distance_array_size
= 0;
592 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
593 /* From section 7.1 (Vertex Shader Special Variables) of the
596 * "It is an error for a shader to statically write both
597 * gl_ClipVertex and gl_ClipDistance."
599 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
600 * gl_ClipVertex nor gl_ClipDistance. However with
601 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
603 find_variable
gl_ClipDistance("gl_ClipDistance");
604 find_variable
gl_CullDistance("gl_CullDistance");
605 find_variable
gl_ClipVertex("gl_ClipVertex");
606 find_variable
* const variables
[] = {
609 !prog
->IsES
? &gl_ClipVertex
: NULL
,
612 find_assignments(shader
->ir
, variables
);
614 /* From the ARB_cull_distance spec:
616 * It is a compile-time or link-time error for the set of shaders forming
617 * a program to statically read or write both gl_ClipVertex and either
618 * gl_ClipDistance or gl_CullDistance.
620 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
624 if (gl_ClipVertex
.found
&& gl_ClipDistance
.found
) {
625 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
626 "and `gl_ClipDistance'\n",
627 _mesa_shader_stage_to_string(shader
->Stage
));
630 if (gl_ClipVertex
.found
&& gl_CullDistance
.found
) {
631 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
632 "and `gl_CullDistance'\n",
633 _mesa_shader_stage_to_string(shader
->Stage
));
638 if (gl_ClipDistance
.found
) {
639 ir_variable
*clip_distance_var
=
640 shader
->symbols
->get_variable("gl_ClipDistance");
641 assert(clip_distance_var
);
642 *clip_distance_array_size
= clip_distance_var
->type
->length
;
644 if (gl_CullDistance
.found
) {
645 ir_variable
*cull_distance_var
=
646 shader
->symbols
->get_variable("gl_CullDistance");
647 assert(cull_distance_var
);
648 *cull_distance_array_size
= cull_distance_var
->type
->length
;
650 /* From the ARB_cull_distance spec:
652 * It is a compile-time or link-time error for the set of shaders forming
653 * a program to have the sum of the sizes of the gl_ClipDistance and
654 * gl_CullDistance arrays to be larger than
655 * gl_MaxCombinedClipAndCullDistances.
657 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
658 ctx
->Const
.MaxClipPlanes
) {
659 linker_error(prog
, "%s shader: the combined size of "
660 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
662 "gl_MaxCombinedClipAndCullDistances (%u)",
663 _mesa_shader_stage_to_string(shader
->Stage
),
664 ctx
->Const
.MaxClipPlanes
);
671 * Verify that a vertex shader executable meets all semantic requirements.
673 * Also sets info.clip_distance_array_size and
674 * info.cull_distance_array_size as a side effect.
676 * \param shader Vertex shader executable to be verified
679 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
680 struct gl_linked_shader
*shader
,
681 struct gl_context
*ctx
)
686 /* From the GLSL 1.10 spec, page 48:
688 * "The variable gl_Position is available only in the vertex
689 * language and is intended for writing the homogeneous vertex
690 * position. All executions of a well-formed vertex shader
691 * executable must write a value into this variable. [...] The
692 * variable gl_Position is available only in the vertex
693 * language and is intended for writing the homogeneous vertex
694 * position. All executions of a well-formed vertex shader
695 * executable must write a value into this variable."
697 * while in GLSL 1.40 this text is changed to:
699 * "The variable gl_Position is available only in the vertex
700 * language and is intended for writing the homogeneous vertex
701 * position. It can be written at any time during shader
702 * execution. It may also be read back by a vertex shader
703 * after being written. This value will be used by primitive
704 * assembly, clipping, culling, and other fixed functionality
705 * operations, if present, that operate on primitives after
706 * vertex processing has occurred. Its value is undefined if
707 * the vertex shader executable does not write gl_Position."
709 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
710 * gl_Position is not an error.
712 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
713 find_variable
gl_Position("gl_Position");
714 find_assignments(shader
->ir
, &gl_Position
);
715 if (!gl_Position
.found
) {
718 "vertex shader does not write to `gl_Position'. "
719 "Its value is undefined. \n");
722 "vertex shader does not write to `gl_Position'. \n");
728 analyze_clip_cull_usage(prog
, shader
, ctx
,
729 &shader
->Program
->info
.clip_distance_array_size
,
730 &shader
->Program
->info
.cull_distance_array_size
);
734 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
735 struct gl_linked_shader
*shader
,
736 struct gl_context
*ctx
)
741 analyze_clip_cull_usage(prog
, shader
, ctx
,
742 &shader
->Program
->info
.clip_distance_array_size
,
743 &shader
->Program
->info
.cull_distance_array_size
);
748 * Verify that a fragment shader executable meets all semantic requirements
750 * \param shader Fragment shader executable to be verified
753 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
754 struct gl_linked_shader
*shader
)
759 find_variable
gl_FragColor("gl_FragColor");
760 find_variable
gl_FragData("gl_FragData");
761 find_variable
* const variables
[] = { &gl_FragColor
, &gl_FragData
, NULL
};
762 find_assignments(shader
->ir
, variables
);
764 if (gl_FragColor
.found
&& gl_FragData
.found
) {
765 linker_error(prog
, "fragment shader writes to both "
766 "`gl_FragColor' and `gl_FragData'\n");
771 * Verify that a geometry shader executable meets all semantic requirements
773 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
774 * info.cull_distance_array_size as a side effect.
776 * \param shader Geometry shader executable to be verified
779 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
780 struct gl_linked_shader
*shader
,
781 struct gl_context
*ctx
)
786 unsigned num_vertices
=
787 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
788 prog
->Geom
.VerticesIn
= num_vertices
;
790 analyze_clip_cull_usage(prog
, shader
, ctx
,
791 &shader
->Program
->info
.clip_distance_array_size
,
792 &shader
->Program
->info
.cull_distance_array_size
);
796 * Check if geometry shaders emit to non-zero streams and do corresponding
800 validate_geometry_shader_emissions(struct gl_context
*ctx
,
801 struct gl_shader_program
*prog
)
803 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
806 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
807 emit_vertex
.run(sh
->ir
);
808 if (emit_vertex
.error()) {
809 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
810 "stream parameter are in the range [0, %d].\n",
811 emit_vertex
.error_func(),
812 emit_vertex
.error_stream(),
813 ctx
->Const
.MaxVertexStreams
- 1);
815 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
816 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
818 /* From the ARB_gpu_shader5 spec:
820 * "Multiple vertex streams are supported only if the output primitive
821 * type is declared to be "points". A program will fail to link if it
822 * contains a geometry shader calling EmitStreamVertex() or
823 * EndStreamPrimitive() if its output primitive type is not "points".
825 * However, in the same spec:
827 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
828 * with <stream> set to zero."
832 * "The function EndPrimitive() is equivalent to calling
833 * EndStreamPrimitive() with <stream> set to zero."
835 * Since we can call EmitVertex() and EndPrimitive() when we output
836 * primitives other than points, calling EmitStreamVertex(0) or
837 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
838 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
839 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
842 if (prog
->Geom
.UsesStreams
&&
843 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
844 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
845 "with n>0 requires point output\n");
851 validate_intrastage_arrays(struct gl_shader_program
*prog
,
852 ir_variable
*const var
,
853 ir_variable
*const existing
)
855 /* Consider the types to be "the same" if both types are arrays
856 * of the same type and one of the arrays is implicitly sized.
857 * In addition, set the type of the linked variable to the
858 * explicitly sized array.
860 if (var
->type
->is_array() && existing
->type
->is_array()) {
861 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
862 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
863 if (var
->type
->length
!= 0) {
864 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
865 linker_error(prog
, "%s `%s' declared as type "
866 "`%s' but outermost dimension has an index"
869 var
->name
, var
->type
->name
,
870 existing
->data
.max_array_access
);
872 existing
->type
= var
->type
;
874 } else if (existing
->type
->length
!= 0) {
875 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
876 !existing
->data
.from_ssbo_unsized_array
) {
877 linker_error(prog
, "%s `%s' declared as type "
878 "`%s' but outermost dimension has an index"
881 var
->name
, existing
->type
->name
,
882 var
->data
.max_array_access
);
893 * Perform validation of global variables used across multiple shaders
896 cross_validate_globals(struct gl_shader_program
*prog
,
897 struct exec_list
*ir
, glsl_symbol_table
*variables
,
900 foreach_in_list(ir_instruction
, node
, ir
) {
901 ir_variable
*const var
= node
->as_variable();
906 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
909 /* don't cross validate subroutine uniforms */
910 if (var
->type
->contains_subroutine())
913 /* Don't cross validate interface instances. These are only relevant
914 * inside a shader. The cross validation is done at the Interface Block
917 if (var
->is_interface_instance())
920 /* Don't cross validate temporaries that are at global scope. These
921 * will eventually get pulled into the shaders 'main'.
923 if (var
->data
.mode
== ir_var_temporary
)
926 /* If a global with this name has already been seen, verify that the
927 * new instance has the same type. In addition, if the globals have
928 * initializers, the values of the initializers must be the same.
930 ir_variable
*const existing
= variables
->get_variable(var
->name
);
931 if (existing
!= NULL
) {
932 /* Check if types match. */
933 if (var
->type
!= existing
->type
) {
934 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
935 /* If it is an unsized array in a Shader Storage Block,
936 * two different shaders can access to different elements.
937 * Because of that, they might be converted to different
938 * sized arrays, then check that they are compatible but
939 * ignore the array size.
941 if (!(var
->data
.mode
== ir_var_shader_storage
&&
942 var
->data
.from_ssbo_unsized_array
&&
943 existing
->data
.mode
== ir_var_shader_storage
&&
944 existing
->data
.from_ssbo_unsized_array
&&
945 var
->type
->gl_type
== existing
->type
->gl_type
)) {
946 linker_error(prog
, "%s `%s' declared as type "
947 "`%s' and type `%s'\n",
949 var
->name
, var
->type
->name
,
950 existing
->type
->name
);
956 if (var
->data
.explicit_location
) {
957 if (existing
->data
.explicit_location
958 && (var
->data
.location
!= existing
->data
.location
)) {
959 linker_error(prog
, "explicit locations for %s "
960 "`%s' have differing values\n",
961 mode_string(var
), var
->name
);
965 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
966 linker_error(prog
, "explicit components for %s `%s' have "
967 "differing values\n", mode_string(var
), var
->name
);
971 existing
->data
.location
= var
->data
.location
;
972 existing
->data
.explicit_location
= true;
974 /* Check if uniform with implicit location was marked explicit
975 * by earlier shader stage. If so, mark it explicit in this stage
976 * too to make sure later processing does not treat it as
979 if (existing
->data
.explicit_location
) {
980 var
->data
.location
= existing
->data
.location
;
981 var
->data
.explicit_location
= true;
985 /* From the GLSL 4.20 specification:
986 * "A link error will result if two compilation units in a program
987 * specify different integer-constant bindings for the same
988 * opaque-uniform name. However, it is not an error to specify a
989 * binding on some but not all declarations for the same name"
991 if (var
->data
.explicit_binding
) {
992 if (existing
->data
.explicit_binding
&&
993 var
->data
.binding
!= existing
->data
.binding
) {
994 linker_error(prog
, "explicit bindings for %s "
995 "`%s' have differing values\n",
996 mode_string(var
), var
->name
);
1000 existing
->data
.binding
= var
->data
.binding
;
1001 existing
->data
.explicit_binding
= true;
1004 if (var
->type
->contains_atomic() &&
1005 var
->data
.offset
!= existing
->data
.offset
) {
1006 linker_error(prog
, "offset specifications for %s "
1007 "`%s' have differing values\n",
1008 mode_string(var
), var
->name
);
1012 /* Validate layout qualifiers for gl_FragDepth.
1014 * From the AMD/ARB_conservative_depth specs:
1016 * "If gl_FragDepth is redeclared in any fragment shader in a
1017 * program, it must be redeclared in all fragment shaders in
1018 * that program that have static assignments to
1019 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1020 * fragment shaders in a single program must have the same set
1023 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1024 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1025 bool layout_differs
=
1026 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1028 if (layout_declared
&& layout_differs
) {
1030 "All redeclarations of gl_FragDepth in all "
1031 "fragment shaders in a single program must have "
1032 "the same set of qualifiers.\n");
1035 if (var
->data
.used
&& layout_differs
) {
1037 "If gl_FragDepth is redeclared with a layout "
1038 "qualifier in any fragment shader, it must be "
1039 "redeclared with the same layout qualifier in "
1040 "all fragment shaders that have assignments to "
1045 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1047 * "If a shared global has multiple initializers, the
1048 * initializers must all be constant expressions, and they
1049 * must all have the same value. Otherwise, a link error will
1050 * result. (A shared global having only one initializer does
1051 * not require that initializer to be a constant expression.)"
1053 * Previous to 4.20 the GLSL spec simply said that initializers
1054 * must have the same value. In this case of non-constant
1055 * initializers, this was impossible to determine. As a result,
1056 * no vendor actually implemented that behavior. The 4.20
1057 * behavior matches the implemented behavior of at least one other
1058 * vendor, so we'll implement that for all GLSL versions.
1060 if (var
->constant_initializer
!= NULL
) {
1061 if (existing
->constant_initializer
!= NULL
) {
1062 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1063 linker_error(prog
, "initializers for %s "
1064 "`%s' have differing values\n",
1065 mode_string(var
), var
->name
);
1069 /* If the first-seen instance of a particular uniform did
1070 * not have an initializer but a later instance does,
1071 * replace the former with the later.
1073 variables
->replace_variable(existing
->name
, var
);
1077 if (var
->data
.has_initializer
) {
1078 if (existing
->data
.has_initializer
1079 && (var
->constant_initializer
== NULL
1080 || existing
->constant_initializer
== NULL
)) {
1082 "shared global variable `%s' has multiple "
1083 "non-constant initializers.\n",
1089 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1090 linker_error(prog
, "declarations for %s `%s' have "
1091 "mismatching invariant qualifiers\n",
1092 mode_string(var
), var
->name
);
1095 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1096 linker_error(prog
, "declarations for %s `%s' have "
1097 "mismatching centroid qualifiers\n",
1098 mode_string(var
), var
->name
);
1101 if (existing
->data
.sample
!= var
->data
.sample
) {
1102 linker_error(prog
, "declarations for %s `%s` have "
1103 "mismatching sample qualifiers\n",
1104 mode_string(var
), var
->name
);
1107 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1108 linker_error(prog
, "declarations for %s `%s` have "
1109 "mismatching image format qualifiers\n",
1110 mode_string(var
), var
->name
);
1114 /* Only in GLSL ES 3.10, the precision qualifier should not match
1115 * between block members defined in matched block names within a
1118 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1119 * member should match.
1121 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1122 !var
->get_interface_type()) &&
1123 existing
->data
.precision
!= var
->data
.precision
) {
1124 linker_error(prog
, "declarations for %s `%s` have "
1125 "mismatching precision qualifiers\n",
1126 mode_string(var
), var
->name
);
1130 variables
->add_variable(var
);
1136 * Perform validation of uniforms used across multiple shader stages
1139 cross_validate_uniforms(struct gl_shader_program
*prog
)
1141 glsl_symbol_table variables
;
1142 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1143 if (prog
->_LinkedShaders
[i
] == NULL
)
1146 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1152 * Accumulates the array of buffer blocks and checks that all definitions of
1153 * blocks agree on their contents.
1156 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1159 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1160 struct gl_uniform_block
*blks
= NULL
;
1161 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1162 &prog
->data
->NumUniformBlocks
;
1164 unsigned max_num_buffer_blocks
= 0;
1165 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1166 if (prog
->_LinkedShaders
[i
]) {
1167 if (validate_ssbo
) {
1168 max_num_buffer_blocks
+=
1169 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1171 max_num_buffer_blocks
+=
1172 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1177 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1178 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1180 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1181 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1182 InterfaceBlockStageIndex
[i
][j
] = -1;
1187 unsigned sh_num_blocks
;
1188 struct gl_uniform_block
**sh_blks
;
1189 if (validate_ssbo
) {
1190 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1191 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1193 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1194 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1197 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1198 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1202 linker_error(prog
, "buffer block `%s' has mismatching "
1203 "definitions\n", sh_blks
[j
]->Name
);
1205 for (unsigned k
= 0; k
<= i
; k
++) {
1206 delete[] InterfaceBlockStageIndex
[k
];
1209 /* Reset the block count. This will help avoid various segfaults
1210 * from api calls that assume the array exists due to the count
1217 InterfaceBlockStageIndex
[i
][index
] = j
;
1221 /* Update per stage block pointers to point to the program list.
1222 * FIXME: We should be able to free the per stage blocks here.
1224 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1225 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1226 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1228 if (stage_index
!= -1) {
1229 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1231 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1232 sh
->Program
->sh
.ShaderStorageBlocks
:
1233 sh
->Program
->sh
.UniformBlocks
;
1235 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1236 sh_blks
[stage_index
] = &blks
[j
];
1241 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1242 delete[] InterfaceBlockStageIndex
[i
];
1246 prog
->data
->ShaderStorageBlocks
= blks
;
1248 prog
->data
->UniformBlocks
= blks
;
1255 * Populates a shaders symbol table with all global declarations
1258 populate_symbol_table(gl_linked_shader
*sh
, glsl_symbol_table
*symbols
)
1260 sh
->symbols
= new(sh
) glsl_symbol_table
;
1262 _mesa_glsl_copy_symbols_from_table(sh
->ir
, symbols
, sh
->symbols
);
1267 * Remap variables referenced in an instruction tree
1269 * This is used when instruction trees are cloned from one shader and placed in
1270 * another. These trees will contain references to \c ir_variable nodes that
1271 * do not exist in the target shader. This function finds these \c ir_variable
1272 * references and replaces the references with matching variables in the target
1275 * If there is no matching variable in the target shader, a clone of the
1276 * \c ir_variable is made and added to the target shader. The new variable is
1277 * added to \b both the instruction stream and the symbol table.
1279 * \param inst IR tree that is to be processed.
1280 * \param symbols Symbol table containing global scope symbols in the
1282 * \param instructions Instruction stream where new variable declarations
1286 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1289 class remap_visitor
: public ir_hierarchical_visitor
{
1291 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1293 this->target
= target
;
1294 this->symbols
= target
->symbols
;
1295 this->instructions
= target
->ir
;
1296 this->temps
= temps
;
1299 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1301 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1302 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1303 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1305 assert(var
!= NULL
);
1307 return visit_continue
;
1310 ir_variable
*const existing
=
1311 this->symbols
->get_variable(ir
->var
->name
);
1312 if (existing
!= NULL
)
1315 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1317 this->symbols
->add_variable(copy
);
1318 this->instructions
->push_head(copy
);
1322 return visit_continue
;
1326 struct gl_linked_shader
*target
;
1327 glsl_symbol_table
*symbols
;
1328 exec_list
*instructions
;
1332 remap_visitor
v(target
, temps
);
1339 * Move non-declarations from one instruction stream to another
1341 * The intended usage pattern of this function is to pass the pointer to the
1342 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1343 * pointer) for \c last and \c false for \c make_copies on the first
1344 * call. Successive calls pass the return value of the previous call for
1345 * \c last and \c true for \c make_copies.
1347 * \param instructions Source instruction stream
1348 * \param last Instruction after which new instructions should be
1349 * inserted in the target instruction stream
1350 * \param make_copies Flag selecting whether instructions in \c instructions
1351 * should be copied (via \c ir_instruction::clone) into the
1352 * target list or moved.
1355 * The new "last" instruction in the target instruction stream. This pointer
1356 * is suitable for use as the \c last parameter of a later call to this
1360 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1361 bool make_copies
, gl_linked_shader
*target
)
1363 hash_table
*temps
= NULL
;
1366 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1367 _mesa_key_pointer_equal
);
1369 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1370 if (inst
->as_function())
1373 ir_variable
*var
= inst
->as_variable();
1374 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1377 assert(inst
->as_assignment()
1379 || inst
->as_if() /* for initializers with the ?: operator */
1380 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1383 inst
= inst
->clone(target
, NULL
);
1386 _mesa_hash_table_insert(temps
, var
, inst
);
1388 remap_variables(inst
, target
, temps
);
1393 last
->insert_after(inst
);
1398 _mesa_hash_table_destroy(temps
, NULL
);
1405 * This class is only used in link_intrastage_shaders() below but declaring
1406 * it inside that function leads to compiler warnings with some versions of
1409 class array_sizing_visitor
: public deref_type_updater
{
1411 array_sizing_visitor()
1412 : mem_ctx(ralloc_context(NULL
)),
1413 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1414 _mesa_key_pointer_equal
))
1418 ~array_sizing_visitor()
1420 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1421 ralloc_free(this->mem_ctx
);
1424 virtual ir_visitor_status
visit(ir_variable
*var
)
1426 const glsl_type
*type_without_array
;
1427 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1428 fixup_type(&var
->type
, var
->data
.max_array_access
,
1429 var
->data
.from_ssbo_unsized_array
,
1430 &implicit_sized_array
);
1431 var
->data
.implicit_sized_array
= implicit_sized_array
;
1432 type_without_array
= var
->type
->without_array();
1433 if (var
->type
->is_interface()) {
1434 if (interface_contains_unsized_arrays(var
->type
)) {
1435 const glsl_type
*new_type
=
1436 resize_interface_members(var
->type
,
1437 var
->get_max_ifc_array_access(),
1438 var
->is_in_shader_storage_block());
1439 var
->type
= new_type
;
1440 var
->change_interface_type(new_type
);
1442 } else if (type_without_array
->is_interface()) {
1443 if (interface_contains_unsized_arrays(type_without_array
)) {
1444 const glsl_type
*new_type
=
1445 resize_interface_members(type_without_array
,
1446 var
->get_max_ifc_array_access(),
1447 var
->is_in_shader_storage_block());
1448 var
->change_interface_type(new_type
);
1449 var
->type
= update_interface_members_array(var
->type
, new_type
);
1451 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1452 /* Store a pointer to the variable in the unnamed_interfaces
1456 _mesa_hash_table_search(this->unnamed_interfaces
,
1459 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1461 if (interface_vars
== NULL
) {
1462 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1464 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1467 unsigned index
= ifc_type
->field_index(var
->name
);
1468 assert(index
< ifc_type
->length
);
1469 assert(interface_vars
[index
] == NULL
);
1470 interface_vars
[index
] = var
;
1472 return visit_continue
;
1476 * For each unnamed interface block that was discovered while running the
1477 * visitor, adjust the interface type to reflect the newly assigned array
1478 * sizes, and fix up the ir_variable nodes to point to the new interface
1481 void fixup_unnamed_interface_types()
1483 hash_table_call_foreach(this->unnamed_interfaces
,
1484 fixup_unnamed_interface_type
, NULL
);
1489 * If the type pointed to by \c type represents an unsized array, replace
1490 * it with a sized array whose size is determined by max_array_access.
1492 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1493 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1495 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1496 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1497 max_array_access
+ 1);
1498 *implicit_sized
= true;
1499 assert(*type
!= NULL
);
1503 static const glsl_type
*
1504 update_interface_members_array(const glsl_type
*type
,
1505 const glsl_type
*new_interface_type
)
1507 const glsl_type
*element_type
= type
->fields
.array
;
1508 if (element_type
->is_array()) {
1509 const glsl_type
*new_array_type
=
1510 update_interface_members_array(element_type
, new_interface_type
);
1511 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1513 return glsl_type::get_array_instance(new_interface_type
,
1519 * Determine whether the given interface type contains unsized arrays (if
1520 * it doesn't, array_sizing_visitor doesn't need to process it).
1522 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1524 for (unsigned i
= 0; i
< type
->length
; i
++) {
1525 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1526 if (elem_type
->is_unsized_array())
1533 * Create a new interface type based on the given type, with unsized arrays
1534 * replaced by sized arrays whose size is determined by
1535 * max_ifc_array_access.
1537 static const glsl_type
*
1538 resize_interface_members(const glsl_type
*type
,
1539 const int *max_ifc_array_access
,
1542 unsigned num_fields
= type
->length
;
1543 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1544 memcpy(fields
, type
->fields
.structure
,
1545 num_fields
* sizeof(*fields
));
1546 for (unsigned i
= 0; i
< num_fields
; i
++) {
1547 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1548 /* If SSBO last member is unsized array, we don't replace it by a sized
1551 if (is_ssbo
&& i
== (num_fields
- 1))
1552 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1553 true, &implicit_sized_array
);
1555 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1556 false, &implicit_sized_array
);
1557 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1559 glsl_interface_packing packing
=
1560 (glsl_interface_packing
) type
->interface_packing
;
1561 bool row_major
= (bool) type
->interface_row_major
;
1562 const glsl_type
*new_ifc_type
=
1563 glsl_type::get_interface_instance(fields
, num_fields
,
1564 packing
, row_major
, type
->name
);
1566 return new_ifc_type
;
1569 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1572 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1573 ir_variable
**interface_vars
= (ir_variable
**) data
;
1574 unsigned num_fields
= ifc_type
->length
;
1575 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1576 memcpy(fields
, ifc_type
->fields
.structure
,
1577 num_fields
* sizeof(*fields
));
1578 bool interface_type_changed
= false;
1579 for (unsigned i
= 0; i
< num_fields
; i
++) {
1580 if (interface_vars
[i
] != NULL
&&
1581 fields
[i
].type
!= interface_vars
[i
]->type
) {
1582 fields
[i
].type
= interface_vars
[i
]->type
;
1583 interface_type_changed
= true;
1586 if (!interface_type_changed
) {
1590 glsl_interface_packing packing
=
1591 (glsl_interface_packing
) ifc_type
->interface_packing
;
1592 bool row_major
= (bool) ifc_type
->interface_row_major
;
1593 const glsl_type
*new_ifc_type
=
1594 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1595 row_major
, ifc_type
->name
);
1597 for (unsigned i
= 0; i
< num_fields
; i
++) {
1598 if (interface_vars
[i
] != NULL
)
1599 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1604 * Memory context used to allocate the data in \c unnamed_interfaces.
1609 * Hash table from const glsl_type * to an array of ir_variable *'s
1610 * pointing to the ir_variables constituting each unnamed interface block.
1612 hash_table
*unnamed_interfaces
;
1616 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1617 struct gl_shader_program
*prog
)
1619 /* We will validate doubles at a later stage */
1620 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1621 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1622 "multiple of 4 or if its applied to a type that is "
1623 "or contains a double a multiple of 8.",
1624 prog
->TransformFeedback
.BufferStride
[idx
]);
1628 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1629 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1630 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1631 "limit has been exceeded.");
1639 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1643 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1644 struct gl_shader_program
*prog
,
1645 struct gl_shader
**shader_list
,
1646 unsigned num_shaders
)
1648 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1649 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1652 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1653 struct gl_shader
*shader
= shader_list
[i
];
1655 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1656 if (shader
->TransformFeedbackBufferStride
[j
]) {
1657 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1658 prog
->TransformFeedback
.BufferStride
[j
] =
1659 shader
->TransformFeedbackBufferStride
[j
];
1660 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1662 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1663 shader
->TransformFeedbackBufferStride
[j
]){
1665 "intrastage shaders defined with conflicting "
1666 "xfb_stride for buffer %d (%d and %d)\n", j
,
1667 prog
->TransformFeedback
.BufferStride
[j
],
1668 shader
->TransformFeedbackBufferStride
[j
]);
1677 * Check for conflicting bindless/bound sampler/image layout qualifiers at
1681 link_bindless_layout_qualifiers(struct gl_shader_program
*prog
,
1682 struct gl_shader
**shader_list
,
1683 unsigned num_shaders
)
1685 bool bindless_sampler
, bindless_image
;
1686 bool bound_sampler
, bound_image
;
1688 bindless_sampler
= bindless_image
= false;
1689 bound_sampler
= bound_image
= false;
1691 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1692 struct gl_shader
*shader
= shader_list
[i
];
1694 if (shader
->bindless_sampler
)
1695 bindless_sampler
= true;
1696 if (shader
->bindless_image
)
1697 bindless_image
= true;
1698 if (shader
->bound_sampler
)
1699 bound_sampler
= true;
1700 if (shader
->bound_image
)
1703 if ((bindless_sampler
&& bound_sampler
) ||
1704 (bindless_image
&& bound_image
)) {
1705 /* From section 4.4.6 of the ARB_bindless_texture spec:
1707 * "If both bindless_sampler and bound_sampler, or bindless_image
1708 * and bound_image, are declared at global scope in any
1709 * compilation unit, a link- time error will be generated."
1711 linker_error(prog
, "both bindless_sampler and bound_sampler, or "
1712 "bindless_image and bound_image, can't be declared at "
1719 * Performs the cross-validation of tessellation control shader vertices and
1720 * layout qualifiers for the attached tessellation control shaders,
1721 * and propagates them to the linked TCS and linked shader program.
1724 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1725 struct gl_program
*gl_prog
,
1726 struct gl_shader
**shader_list
,
1727 unsigned num_shaders
)
1729 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1732 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1734 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1736 * "All tessellation control shader layout declarations in a program
1737 * must specify the same output patch vertex count. There must be at
1738 * least one layout qualifier specifying an output patch vertex count
1739 * in any program containing tessellation control shaders; however,
1740 * such a declaration is not required in all tessellation control
1744 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1745 struct gl_shader
*shader
= shader_list
[i
];
1747 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1748 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1749 gl_prog
->info
.tess
.tcs_vertices_out
!=
1750 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1751 linker_error(prog
, "tessellation control shader defined with "
1752 "conflicting output vertex count (%d and %d)\n",
1753 gl_prog
->info
.tess
.tcs_vertices_out
,
1754 shader
->info
.TessCtrl
.VerticesOut
);
1757 gl_prog
->info
.tess
.tcs_vertices_out
=
1758 shader
->info
.TessCtrl
.VerticesOut
;
1762 /* Just do the intrastage -> interstage propagation right now,
1763 * since we already know we're in the right type of shader program
1766 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1767 linker_error(prog
, "tessellation control shader didn't declare "
1768 "vertices out layout qualifier\n");
1775 * Performs the cross-validation of tessellation evaluation shader
1776 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1777 * for the attached tessellation evaluation shaders, and propagates them
1778 * to the linked TES and linked shader program.
1781 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1782 struct gl_program
*gl_prog
,
1783 struct gl_shader
**shader_list
,
1784 unsigned num_shaders
)
1786 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1789 int point_mode
= -1;
1790 unsigned vertex_order
= 0;
1792 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1793 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1795 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1797 * "At least one tessellation evaluation shader (compilation unit) in
1798 * a program must declare a primitive mode in its input layout.
1799 * Declaration vertex spacing, ordering, and point mode identifiers is
1800 * optional. It is not required that all tessellation evaluation
1801 * shaders in a program declare a primitive mode. If spacing or
1802 * vertex ordering declarations are omitted, the tessellation
1803 * primitive generator will use equal spacing or counter-clockwise
1804 * vertex ordering, respectively. If a point mode declaration is
1805 * omitted, the tessellation primitive generator will produce lines or
1806 * triangles according to the primitive mode."
1809 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1810 struct gl_shader
*shader
= shader_list
[i
];
1812 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1813 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1814 gl_prog
->info
.tess
.primitive_mode
!=
1815 shader
->info
.TessEval
.PrimitiveMode
) {
1816 linker_error(prog
, "tessellation evaluation shader defined with "
1817 "conflicting input primitive modes.\n");
1820 gl_prog
->info
.tess
.primitive_mode
=
1821 shader
->info
.TessEval
.PrimitiveMode
;
1824 if (shader
->info
.TessEval
.Spacing
!= 0) {
1825 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1826 shader
->info
.TessEval
.Spacing
) {
1827 linker_error(prog
, "tessellation evaluation shader defined with "
1828 "conflicting vertex spacing.\n");
1831 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1834 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1835 if (vertex_order
!= 0 &&
1836 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1837 linker_error(prog
, "tessellation evaluation shader defined with "
1838 "conflicting ordering.\n");
1841 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1844 if (shader
->info
.TessEval
.PointMode
!= -1) {
1845 if (point_mode
!= -1 &&
1846 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1847 linker_error(prog
, "tessellation evaluation shader defined with "
1848 "conflicting point modes.\n");
1851 point_mode
= shader
->info
.TessEval
.PointMode
;
1856 /* Just do the intrastage -> interstage propagation right now,
1857 * since we already know we're in the right type of shader program
1860 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1862 "tessellation evaluation shader didn't declare input "
1863 "primitive modes.\n");
1867 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1868 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1870 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1871 gl_prog
->info
.tess
.ccw
= true;
1873 gl_prog
->info
.tess
.ccw
= false;
1876 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1877 gl_prog
->info
.tess
.point_mode
= false;
1879 gl_prog
->info
.tess
.point_mode
= true;
1884 * Performs the cross-validation of layout qualifiers specified in
1885 * redeclaration of gl_FragCoord for the attached fragment shaders,
1886 * and propagates them to the linked FS and linked shader program.
1889 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1890 struct gl_linked_shader
*linked_shader
,
1891 struct gl_shader
**shader_list
,
1892 unsigned num_shaders
)
1894 bool redeclares_gl_fragcoord
= false;
1895 bool uses_gl_fragcoord
= false;
1896 bool origin_upper_left
= false;
1897 bool pixel_center_integer
= false;
1899 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1900 (prog
->data
->Version
< 150 &&
1901 !prog
->ARB_fragment_coord_conventions_enable
))
1904 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1905 struct gl_shader
*shader
= shader_list
[i
];
1906 /* From the GLSL 1.50 spec, page 39:
1908 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1909 * it must be redeclared in all the fragment shaders in that program
1910 * that have a static use gl_FragCoord."
1912 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1913 shader
->uses_gl_fragcoord
)
1914 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1915 uses_gl_fragcoord
)) {
1916 linker_error(prog
, "fragment shader defined with conflicting "
1917 "layout qualifiers for gl_FragCoord\n");
1920 /* From the GLSL 1.50 spec, page 39:
1922 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1923 * single program must have the same set of qualifiers."
1925 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1926 (shader
->origin_upper_left
!= origin_upper_left
||
1927 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1928 linker_error(prog
, "fragment shader defined with conflicting "
1929 "layout qualifiers for gl_FragCoord\n");
1932 /* Update the linked shader state. Note that uses_gl_fragcoord should
1933 * accumulate the results. The other values should replace. If there
1934 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1935 * are already known to be the same.
1937 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1938 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1939 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1940 origin_upper_left
= shader
->origin_upper_left
;
1941 pixel_center_integer
= shader
->pixel_center_integer
;
1944 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1945 shader
->EarlyFragmentTests
|| shader
->PostDepthCoverage
;
1946 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1947 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1948 shader
->PostDepthCoverage
;
1950 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1955 * Performs the cross-validation of geometry shader max_vertices and
1956 * primitive type layout qualifiers for the attached geometry shaders,
1957 * and propagates them to the linked GS and linked shader program.
1960 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1961 struct gl_program
*gl_prog
,
1962 struct gl_shader
**shader_list
,
1963 unsigned num_shaders
)
1965 /* No in/out qualifiers defined for anything but GLSL 1.50+
1966 * geometry shaders so far.
1968 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
1969 prog
->data
->Version
< 150)
1972 int vertices_out
= -1;
1974 gl_prog
->info
.gs
.invocations
= 0;
1975 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
1976 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
1978 /* From the GLSL 1.50 spec, page 46:
1980 * "All geometry shader output layout declarations in a program
1981 * must declare the same layout and same value for
1982 * max_vertices. There must be at least one geometry output
1983 * layout declaration somewhere in a program, but not all
1984 * geometry shaders (compilation units) are required to
1988 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1989 struct gl_shader
*shader
= shader_list
[i
];
1991 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1992 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
1993 gl_prog
->info
.gs
.input_primitive
!=
1994 shader
->info
.Geom
.InputType
) {
1995 linker_error(prog
, "geometry shader defined with conflicting "
1999 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
2002 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
2003 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
2004 gl_prog
->info
.gs
.output_primitive
!=
2005 shader
->info
.Geom
.OutputType
) {
2006 linker_error(prog
, "geometry shader defined with conflicting "
2010 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
2013 if (shader
->info
.Geom
.VerticesOut
!= -1) {
2014 if (vertices_out
!= -1 &&
2015 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
2016 linker_error(prog
, "geometry shader defined with conflicting "
2017 "output vertex count (%d and %d)\n",
2018 vertices_out
, shader
->info
.Geom
.VerticesOut
);
2021 vertices_out
= shader
->info
.Geom
.VerticesOut
;
2024 if (shader
->info
.Geom
.Invocations
!= 0) {
2025 if (gl_prog
->info
.gs
.invocations
!= 0 &&
2026 gl_prog
->info
.gs
.invocations
!=
2027 (unsigned) shader
->info
.Geom
.Invocations
) {
2028 linker_error(prog
, "geometry shader defined with conflicting "
2029 "invocation count (%d and %d)\n",
2030 gl_prog
->info
.gs
.invocations
,
2031 shader
->info
.Geom
.Invocations
);
2034 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
2038 /* Just do the intrastage -> interstage propagation right now,
2039 * since we already know we're in the right type of shader program
2042 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
2044 "geometry shader didn't declare primitive input type\n");
2048 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
2050 "geometry shader didn't declare primitive output type\n");
2054 if (vertices_out
== -1) {
2056 "geometry shader didn't declare max_vertices\n");
2059 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2062 if (gl_prog
->info
.gs
.invocations
== 0)
2063 gl_prog
->info
.gs
.invocations
= 1;
2068 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2069 * qualifiers for the attached compute shaders, and propagate them to the
2070 * linked CS and linked shader program.
2073 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2074 struct gl_program
*gl_prog
,
2075 struct gl_shader
**shader_list
,
2076 unsigned num_shaders
)
2078 /* This function is called for all shader stages, but it only has an effect
2079 * for compute shaders.
2081 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2084 for (int i
= 0; i
< 3; i
++)
2085 gl_prog
->info
.cs
.local_size
[i
] = 0;
2087 gl_prog
->info
.cs
.local_size_variable
= false;
2089 /* From the ARB_compute_shader spec, in the section describing local size
2092 * If multiple compute shaders attached to a single program object
2093 * declare local work-group size, the declarations must be identical;
2094 * otherwise a link-time error results. Furthermore, if a program
2095 * object contains any compute shaders, at least one must contain an
2096 * input layout qualifier specifying the local work sizes of the
2097 * program, or a link-time error will occur.
2099 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2100 struct gl_shader
*shader
= shader_list
[sh
];
2102 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2103 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2104 for (int i
= 0; i
< 3; i
++) {
2105 if (gl_prog
->info
.cs
.local_size
[i
] !=
2106 shader
->info
.Comp
.LocalSize
[i
]) {
2107 linker_error(prog
, "compute shader defined with conflicting "
2113 for (int i
= 0; i
< 3; i
++) {
2114 gl_prog
->info
.cs
.local_size
[i
] =
2115 shader
->info
.Comp
.LocalSize
[i
];
2117 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2118 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2119 /* The ARB_compute_variable_group_size spec says:
2121 * If one compute shader attached to a program declares a
2122 * variable local group size and a second compute shader
2123 * attached to the same program declares a fixed local group
2124 * size, a link-time error results.
2126 linker_error(prog
, "compute shader defined with both fixed and "
2127 "variable local group size\n");
2130 gl_prog
->info
.cs
.local_size_variable
= true;
2134 /* Just do the intrastage -> interstage propagation right now,
2135 * since we already know we're in the right type of shader program
2138 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2139 !gl_prog
->info
.cs
.local_size_variable
) {
2140 linker_error(prog
, "compute shader must contain a fixed or a variable "
2141 "local group size\n");
2148 * Combine a group of shaders for a single stage to generate a linked shader
2151 * If this function is supplied a single shader, it is cloned, and the new
2152 * shader is returned.
2154 struct gl_linked_shader
*
2155 link_intrastage_shaders(void *mem_ctx
,
2156 struct gl_context
*ctx
,
2157 struct gl_shader_program
*prog
,
2158 struct gl_shader
**shader_list
,
2159 unsigned num_shaders
,
2160 bool allow_missing_main
)
2162 struct gl_uniform_block
*ubo_blocks
= NULL
;
2163 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2164 unsigned num_ubo_blocks
= 0;
2165 unsigned num_ssbo_blocks
= 0;
2167 /* Check that global variables defined in multiple shaders are consistent.
2169 glsl_symbol_table variables
;
2170 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2171 if (shader_list
[i
] == NULL
)
2173 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2176 if (!prog
->data
->LinkStatus
)
2179 /* Check that interface blocks defined in multiple shaders are consistent.
2181 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2183 if (!prog
->data
->LinkStatus
)
2186 /* Check that there is only a single definition of each function signature
2187 * across all shaders.
2189 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2190 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2191 ir_function
*const f
= node
->as_function();
2196 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2197 ir_function
*const other
=
2198 shader_list
[j
]->symbols
->get_function(f
->name
);
2200 /* If the other shader has no function (and therefore no function
2201 * signatures) with the same name, skip to the next shader.
2206 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2207 if (!sig
->is_defined
)
2210 ir_function_signature
*other_sig
=
2211 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2213 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2214 linker_error(prog
, "function `%s' is multiply defined\n",
2223 /* Find the shader that defines main, and make a clone of it.
2225 * Starting with the clone, search for undefined references. If one is
2226 * found, find the shader that defines it. Clone the reference and add
2227 * it to the shader. Repeat until there are no undefined references or
2228 * until a reference cannot be resolved.
2230 gl_shader
*main
= NULL
;
2231 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2232 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2233 main
= shader_list
[i
];
2238 if (main
== NULL
&& allow_missing_main
)
2239 main
= shader_list
[0];
2242 linker_error(prog
, "%s shader lacks `main'\n",
2243 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2247 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2248 linked
->Stage
= shader_list
[0]->Stage
;
2250 /* Create program and attach it to the linked shader */
2251 struct gl_program
*gl_prog
=
2252 ctx
->Driver
.NewProgram(ctx
,
2253 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2256 prog
->data
->LinkStatus
= linking_failure
;
2257 _mesa_delete_linked_shader(ctx
, linked
);
2261 if (!prog
->data
->cache_fallback
)
2262 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2264 /* Don't use _mesa_reference_program() just take ownership */
2265 linked
->Program
= gl_prog
;
2267 linked
->ir
= new(linked
) exec_list
;
2268 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2270 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2271 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2272 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2273 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2274 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2276 if (linked
->Stage
!= MESA_SHADER_FRAGMENT
)
2277 link_xfb_stride_layout_qualifiers(ctx
, prog
, shader_list
, num_shaders
);
2279 link_bindless_layout_qualifiers(prog
, shader_list
, num_shaders
);
2281 populate_symbol_table(linked
, shader_list
[0]->symbols
);
2283 /* The pointer to the main function in the final linked shader (i.e., the
2284 * copy of the original shader that contained the main function).
2286 ir_function_signature
*const main_sig
=
2287 _mesa_get_main_function_signature(linked
->symbols
);
2289 /* Move any instructions other than variable declarations or function
2290 * declarations into main.
2292 if (main_sig
!= NULL
) {
2293 exec_node
*insertion_point
=
2294 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2297 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2298 if (shader_list
[i
] == main
)
2301 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2302 insertion_point
, true, linked
);
2306 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2307 _mesa_delete_linked_shader(ctx
, linked
);
2311 /* Make a pass over all variable declarations to ensure that arrays with
2312 * unspecified sizes have a size specified. The size is inferred from the
2313 * max_array_access field.
2315 array_sizing_visitor v
;
2317 v
.fixup_unnamed_interface_types();
2319 if (!prog
->data
->cache_fallback
) {
2320 /* Link up uniform blocks defined within this stage. */
2321 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2322 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2324 if (!prog
->data
->LinkStatus
) {
2325 _mesa_delete_linked_shader(ctx
, linked
);
2329 /* Copy ubo blocks to linked shader list */
2330 linked
->Program
->sh
.UniformBlocks
=
2331 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2332 ralloc_steal(linked
, ubo_blocks
);
2333 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2334 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2336 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2338 /* Copy ssbo blocks to linked shader list */
2339 linked
->Program
->sh
.ShaderStorageBlocks
=
2340 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2341 ralloc_steal(linked
, ssbo_blocks
);
2342 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2343 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2345 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2348 /* At this point linked should contain all of the linked IR, so
2349 * validate it to make sure nothing went wrong.
2351 validate_ir_tree(linked
->ir
);
2353 /* Set the size of geometry shader input arrays */
2354 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2355 unsigned num_vertices
=
2356 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2357 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2358 MESA_SHADER_GEOMETRY
);
2359 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2360 ir
->accept(&input_resize_visitor
);
2364 if (ctx
->Const
.VertexID_is_zero_based
)
2365 lower_vertex_id(linked
);
2367 if (ctx
->Const
.LowerCsDerivedVariables
)
2368 lower_cs_derived(linked
);
2371 /* Compute the source checksum. */
2372 linked
->SourceChecksum
= 0;
2373 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2374 if (shader_list
[i
] == NULL
)
2376 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2384 * Update the sizes of linked shader uniform arrays to the maximum
2387 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2389 * If one or more elements of an array are active,
2390 * GetActiveUniform will return the name of the array in name,
2391 * subject to the restrictions listed above. The type of the array
2392 * is returned in type. The size parameter contains the highest
2393 * array element index used, plus one. The compiler or linker
2394 * determines the highest index used. There will be only one
2395 * active uniform reported by the GL per uniform array.
2399 update_array_sizes(struct gl_shader_program
*prog
)
2401 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2402 if (prog
->_LinkedShaders
[i
] == NULL
)
2405 bool types_were_updated
= false;
2407 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2408 ir_variable
*const var
= node
->as_variable();
2410 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2411 !var
->type
->is_array())
2414 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2415 * will not be eliminated. Since we always do std140, just
2416 * don't resize arrays in UBOs.
2418 * Atomic counters are supposed to get deterministic
2419 * locations assigned based on the declaration ordering and
2420 * sizes, array compaction would mess that up.
2422 * Subroutine uniforms are not removed.
2424 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2425 var
->type
->contains_subroutine() || var
->constant_initializer
)
2428 int size
= var
->data
.max_array_access
;
2429 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2430 if (prog
->_LinkedShaders
[j
] == NULL
)
2433 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2434 ir_variable
*other_var
= node2
->as_variable();
2438 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2439 other_var
->data
.max_array_access
> size
) {
2440 size
= other_var
->data
.max_array_access
;
2445 if (size
+ 1 != (int)var
->type
->length
) {
2446 /* If this is a built-in uniform (i.e., it's backed by some
2447 * fixed-function state), adjust the number of state slots to
2448 * match the new array size. The number of slots per array entry
2449 * is not known. It seems safe to assume that the total number of
2450 * slots is an integer multiple of the number of array elements.
2451 * Determine the number of slots per array element by dividing by
2452 * the old (total) size.
2454 const unsigned num_slots
= var
->get_num_state_slots();
2455 if (num_slots
> 0) {
2456 var
->set_num_state_slots((size
+ 1)
2457 * (num_slots
/ var
->type
->length
));
2460 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2462 types_were_updated
= true;
2466 /* Update the types of dereferences in case we changed any. */
2467 if (types_were_updated
) {
2468 deref_type_updater v
;
2469 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2475 * Resize tessellation evaluation per-vertex inputs to the size of
2476 * tessellation control per-vertex outputs.
2479 resize_tes_inputs(struct gl_context
*ctx
,
2480 struct gl_shader_program
*prog
)
2482 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2485 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2486 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2488 /* If no control shader is present, then the TES inputs are statically
2489 * sized to MaxPatchVertices; the actual size of the arrays won't be
2490 * known until draw time.
2492 const int num_vertices
= tcs
2493 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2494 : ctx
->Const
.MaxPatchVertices
;
2496 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2497 MESA_SHADER_TESS_EVAL
);
2498 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2499 ir
->accept(&input_resize_visitor
);
2502 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2503 /* Convert the gl_PatchVerticesIn system value into a constant, since
2504 * the value is known at this point.
2506 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2507 ir_variable
*var
= ir
->as_variable();
2508 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2509 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2510 void *mem_ctx
= ralloc_parent(var
);
2511 var
->data
.location
= 0;
2512 var
->data
.explicit_location
= false;
2514 var
->data
.mode
= ir_var_auto
;
2515 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2517 var
->data
.mode
= ir_var_uniform
;
2518 var
->data
.how_declared
= ir_var_hidden
;
2519 var
->allocate_state_slots(1);
2520 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2521 slot0
->swizzle
= SWIZZLE_XXXX
;
2522 slot0
->tokens
[0] = STATE_INTERNAL
;
2523 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2524 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2525 slot0
->tokens
[i
] = 0;
2533 * Find a contiguous set of available bits in a bitmask.
2535 * \param used_mask Bits representing used (1) and unused (0) locations
2536 * \param needed_count Number of contiguous bits needed.
2539 * Base location of the available bits on success or -1 on failure.
2542 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2544 unsigned needed_mask
= (1 << needed_count
) - 1;
2545 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2547 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2548 * cannot optimize possibly infinite loops" for the loop below.
2550 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2553 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2554 if ((needed_mask
& ~used_mask
) == needed_mask
)
2565 * Assign locations for either VS inputs or FS outputs
2567 * \param mem_ctx Temporary ralloc context used for linking
2568 * \param prog Shader program whose variables need locations assigned
2569 * \param constants Driver specific constant values for the program.
2570 * \param target_index Selector for the program target to receive location
2571 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2572 * \c MESA_SHADER_FRAGMENT.
2575 * If locations are successfully assigned, true is returned. Otherwise an
2576 * error is emitted to the shader link log and false is returned.
2579 assign_attribute_or_color_locations(void *mem_ctx
,
2580 gl_shader_program
*prog
,
2581 struct gl_constants
*constants
,
2582 unsigned target_index
)
2584 /* Maximum number of generic locations. This corresponds to either the
2585 * maximum number of draw buffers or the maximum number of generic
2588 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2589 constants
->Program
[target_index
].MaxAttribs
:
2590 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2592 /* Mark invalid locations as being used.
2594 unsigned used_locations
= (max_index
>= 32)
2595 ? ~0 : ~((1 << max_index
) - 1);
2596 unsigned double_storage_locations
= 0;
2598 assert((target_index
== MESA_SHADER_VERTEX
)
2599 || (target_index
== MESA_SHADER_FRAGMENT
));
2601 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2605 /* Operate in a total of four passes.
2607 * 1. Invalidate the location assignments for all vertex shader inputs.
2609 * 2. Assign locations for inputs that have user-defined (via
2610 * glBindVertexAttribLocation) locations and outputs that have
2611 * user-defined locations (via glBindFragDataLocation).
2613 * 3. Sort the attributes without assigned locations by number of slots
2614 * required in decreasing order. Fragmentation caused by attribute
2615 * locations assigned by the application may prevent large attributes
2616 * from having enough contiguous space.
2618 * 4. Assign locations to any inputs without assigned locations.
2621 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2622 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2624 const enum ir_variable_mode direction
=
2625 (target_index
== MESA_SHADER_VERTEX
)
2626 ? ir_var_shader_in
: ir_var_shader_out
;
2629 /* Temporary storage for the set of attributes that need locations assigned.
2635 /* Used below in the call to qsort. */
2636 static int compare(const void *a
, const void *b
)
2638 const temp_attr
*const l
= (const temp_attr
*) a
;
2639 const temp_attr
*const r
= (const temp_attr
*) b
;
2641 /* Reversed because we want a descending order sort below. */
2642 return r
->slots
- l
->slots
;
2645 assert(max_index
<= 32);
2647 /* Temporary array for the set of attributes that have locations assigned,
2648 * for the purpose of checking overlapping slots/components of (non-ES)
2649 * fragment shader outputs.
2651 ir_variable
*assigned
[12 * 4]; /* (max # of FS outputs) * # components */
2652 unsigned assigned_attr
= 0;
2654 unsigned num_attr
= 0;
2656 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2657 ir_variable
*const var
= node
->as_variable();
2659 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2662 if (var
->data
.explicit_location
) {
2663 var
->data
.is_unmatched_generic_inout
= 0;
2664 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2665 || (var
->data
.location
< 0)) {
2667 "invalid explicit location %d specified for `%s'\n",
2668 (var
->data
.location
< 0)
2669 ? var
->data
.location
2670 : var
->data
.location
- generic_base
,
2674 } else if (target_index
== MESA_SHADER_VERTEX
) {
2677 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2678 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2679 var
->data
.location
= binding
;
2680 var
->data
.is_unmatched_generic_inout
= 0;
2682 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2685 const char *name
= var
->name
;
2686 const glsl_type
*type
= var
->type
;
2689 /* Check if there's a binding for the variable name */
2690 if (prog
->FragDataBindings
->get(binding
, name
)) {
2691 assert(binding
>= FRAG_RESULT_DATA0
);
2692 var
->data
.location
= binding
;
2693 var
->data
.is_unmatched_generic_inout
= 0;
2695 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2696 var
->data
.index
= index
;
2701 /* If not, but it's an array type, look for name[0] */
2702 if (type
->is_array()) {
2703 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2704 type
= type
->fields
.array
;
2712 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2715 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2717 * "Output binding assignments will cause LinkProgram to fail:
2719 * If the program has an active output assigned to a location greater
2720 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2721 * an active output assigned an index greater than or equal to one;"
2723 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2724 var
->data
.location
- generic_base
>=
2725 (int) constants
->MaxDualSourceDrawBuffers
) {
2727 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2728 "with index %u for %s\n",
2729 var
->data
.location
- generic_base
, var
->data
.index
,
2734 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2736 /* If the variable is not a built-in and has a location statically
2737 * assigned in the shader (presumably via a layout qualifier), make sure
2738 * that it doesn't collide with other assigned locations. Otherwise,
2739 * add it to the list of variables that need linker-assigned locations.
2741 if (var
->data
.location
!= -1) {
2742 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2743 /* From page 61 of the OpenGL 4.0 spec:
2745 * "LinkProgram will fail if the attribute bindings assigned
2746 * by BindAttribLocation do not leave not enough space to
2747 * assign a location for an active matrix attribute or an
2748 * active attribute array, both of which require multiple
2749 * contiguous generic attributes."
2751 * I think above text prohibits the aliasing of explicit and
2752 * automatic assignments. But, aliasing is allowed in manual
2753 * assignments of attribute locations. See below comments for
2756 * From OpenGL 4.0 spec, page 61:
2758 * "It is possible for an application to bind more than one
2759 * attribute name to the same location. This is referred to as
2760 * aliasing. This will only work if only one of the aliased
2761 * attributes is active in the executable program, or if no
2762 * path through the shader consumes more than one attribute of
2763 * a set of attributes aliased to the same location. A link
2764 * error can occur if the linker determines that every path
2765 * through the shader consumes multiple aliased attributes,
2766 * but implementations are not required to generate an error
2769 * From GLSL 4.30 spec, page 54:
2771 * "A program will fail to link if any two non-vertex shader
2772 * input variables are assigned to the same location. For
2773 * vertex shaders, multiple input variables may be assigned
2774 * to the same location using either layout qualifiers or via
2775 * the OpenGL API. However, such aliasing is intended only to
2776 * support vertex shaders where each execution path accesses
2777 * at most one input per each location. Implementations are
2778 * permitted, but not required, to generate link-time errors
2779 * if they detect that every path through the vertex shader
2780 * executable accesses multiple inputs assigned to any single
2781 * location. For all shader types, a program will fail to link
2782 * if explicit location assignments leave the linker unable
2783 * to find space for other variables without explicit
2786 * From OpenGL ES 3.0 spec, page 56:
2788 * "Binding more than one attribute name to the same location
2789 * is referred to as aliasing, and is not permitted in OpenGL
2790 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2791 * fail when this condition exists. However, aliasing is
2792 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2793 * This will only work if only one of the aliased attributes
2794 * is active in the executable program, or if no path through
2795 * the shader consumes more than one attribute of a set of
2796 * attributes aliased to the same location. A link error can
2797 * occur if the linker determines that every path through the
2798 * shader consumes multiple aliased attributes, but implemen-
2799 * tations are not required to generate an error in this case."
2801 * After looking at above references from OpenGL, OpenGL ES and
2802 * GLSL specifications, we allow aliasing of vertex input variables
2803 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2805 * NOTE: This is not required by the spec but its worth mentioning
2806 * here that we're not doing anything to make sure that no path
2807 * through the vertex shader executable accesses multiple inputs
2808 * assigned to any single location.
2811 /* Mask representing the contiguous slots that will be used by
2814 const unsigned attr
= var
->data
.location
- generic_base
;
2815 const unsigned use_mask
= (1 << slots
) - 1;
2816 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2817 ? "vertex shader input" : "fragment shader output";
2819 /* Generate a link error if the requested locations for this
2820 * attribute exceed the maximum allowed attribute location.
2822 if (attr
+ slots
> max_index
) {
2824 "insufficient contiguous locations "
2825 "available for %s `%s' %d %d %d\n", string
,
2826 var
->name
, used_locations
, use_mask
, attr
);
2830 /* Generate a link error if the set of bits requested for this
2831 * attribute overlaps any previously allocated bits.
2833 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2834 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2835 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2838 * "Additionally, for fragment shader outputs, if two
2839 * variables are placed within the same location, they
2840 * must have the same underlying type (floating-point or
2841 * integer). No component aliasing of output variables or
2842 * members is allowed.
2844 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2845 unsigned assigned_slots
=
2846 assigned
[i
]->type
->count_attribute_slots(false);
2847 unsigned assig_attr
=
2848 assigned
[i
]->data
.location
- generic_base
;
2849 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2851 if ((assigned_use_mask
<< assig_attr
) &
2852 (use_mask
<< attr
)) {
2854 const glsl_type
*assigned_type
=
2855 assigned
[i
]->type
->without_array();
2856 const glsl_type
*type
= var
->type
->without_array();
2857 if (assigned_type
->base_type
!= type
->base_type
) {
2858 linker_error(prog
, "types do not match for aliased"
2859 " %ss %s and %s\n", string
,
2860 assigned
[i
]->name
, var
->name
);
2864 unsigned assigned_component_mask
=
2865 ((1 << assigned_type
->vector_elements
) - 1) <<
2866 assigned
[i
]->data
.location_frac
;
2867 unsigned component_mask
=
2868 ((1 << type
->vector_elements
) - 1) <<
2869 var
->data
.location_frac
;
2870 if (assigned_component_mask
& component_mask
) {
2871 linker_error(prog
, "overlapping component is "
2872 "assigned to %ss %s and %s "
2874 string
, assigned
[i
]->name
, var
->name
,
2875 var
->data
.location_frac
);
2880 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2881 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2882 linker_error(prog
, "overlapping location is assigned "
2883 "to %s `%s' %d %d %d\n", string
, var
->name
,
2884 used_locations
, use_mask
, attr
);
2887 linker_warning(prog
, "overlapping location is assigned "
2888 "to %s `%s' %d %d %d\n", string
, var
->name
,
2889 used_locations
, use_mask
, attr
);
2893 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2894 /* Only track assigned variables for non-ES fragment shaders
2895 * to avoid overflowing the array.
2897 * At most one variable per fragment output component should
2900 assert(assigned_attr
< ARRAY_SIZE(assigned
));
2901 assigned
[assigned_attr
] = var
;
2905 used_locations
|= (use_mask
<< attr
);
2907 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2909 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2910 * active attribute variables may fail to link, unless
2911 * device-dependent optimizations are able to make the program
2912 * fit within available hardware resources. For the purposes
2913 * of this test, attribute variables of the type dvec3, dvec4,
2914 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2915 * count as consuming twice as many attributes as equivalent
2916 * single-precision types. While these types use the same number
2917 * of generic attributes as their single-precision equivalents,
2918 * implementations are permitted to consume two single-precision
2919 * vectors of internal storage for each three- or four-component
2920 * double-precision vector."
2922 * Mark this attribute slot as taking up twice as much space
2923 * so we can count it properly against limits. According to
2924 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2925 * is optional behavior, but it seems preferable.
2927 if (var
->type
->without_array()->is_dual_slot())
2928 double_storage_locations
|= (use_mask
<< attr
);
2934 if (num_attr
>= max_index
) {
2935 linker_error(prog
, "too many %s (max %u)",
2936 target_index
== MESA_SHADER_VERTEX
?
2937 "vertex shader inputs" : "fragment shader outputs",
2941 to_assign
[num_attr
].slots
= slots
;
2942 to_assign
[num_attr
].var
= var
;
2946 if (target_index
== MESA_SHADER_VERTEX
) {
2947 unsigned total_attribs_size
=
2948 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2949 _mesa_bitcount(double_storage_locations
);
2950 if (total_attribs_size
> max_index
) {
2952 "attempt to use %d vertex attribute slots only %d available ",
2953 total_attribs_size
, max_index
);
2958 /* If all of the attributes were assigned locations by the application (or
2959 * are built-in attributes with fixed locations), return early. This should
2960 * be the common case.
2965 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2967 if (target_index
== MESA_SHADER_VERTEX
) {
2968 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2969 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2970 * reserved to prevent it from being automatically allocated below.
2972 find_deref_visitor
find("gl_Vertex");
2974 if (find
.variable_found())
2975 used_locations
|= (1 << 0);
2978 for (unsigned i
= 0; i
< num_attr
; i
++) {
2979 /* Mask representing the contiguous slots that will be used by this
2982 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2984 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2987 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2988 ? "vertex shader input" : "fragment shader output";
2991 "insufficient contiguous locations "
2992 "available for %s `%s'\n",
2993 string
, to_assign
[i
].var
->name
);
2997 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2998 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2999 used_locations
|= (use_mask
<< location
);
3001 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
3002 double_storage_locations
|= (use_mask
<< location
);
3005 /* Now that we have all the locations, from the GL 4.5 core spec, section
3006 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
3007 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
3008 * as equivalent single-precision types.
3010 if (target_index
== MESA_SHADER_VERTEX
) {
3011 unsigned total_attribs_size
=
3012 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
3013 _mesa_bitcount(double_storage_locations
);
3014 if (total_attribs_size
> max_index
) {
3016 "attempt to use %d vertex attribute slots only %d available ",
3017 total_attribs_size
, max_index
);
3026 * Match explicit locations of outputs to inputs and deactivate the
3027 * unmatch flag if found so we don't optimise them away.
3030 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
3031 gl_linked_shader
*consumer
)
3033 glsl_symbol_table parameters
;
3034 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
3037 /* Find all shader outputs in the "producer" stage.
3039 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
3040 ir_variable
*const var
= node
->as_variable();
3042 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
3045 if (var
->data
.explicit_location
&&
3046 var
->data
.location
>= VARYING_SLOT_VAR0
) {
3047 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
3048 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
3049 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
3053 /* Match inputs to outputs */
3054 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
3055 ir_variable
*const input
= node
->as_variable();
3057 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
3060 ir_variable
*output
= NULL
;
3061 if (input
->data
.explicit_location
3062 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
3063 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
3064 [input
->data
.location_frac
];
3066 if (output
!= NULL
){
3067 input
->data
.is_unmatched_generic_inout
= 0;
3068 output
->data
.is_unmatched_generic_inout
= 0;
3075 * Store the gl_FragDepth layout in the gl_shader_program struct.
3078 store_fragdepth_layout(struct gl_shader_program
*prog
)
3080 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3084 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3086 /* We don't look up the gl_FragDepth symbol directly because if
3087 * gl_FragDepth is not used in the shader, it's removed from the IR.
3088 * However, the symbol won't be removed from the symbol table.
3090 * We're only interested in the cases where the variable is NOT removed
3093 foreach_in_list(ir_instruction
, node
, ir
) {
3094 ir_variable
*const var
= node
->as_variable();
3096 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3100 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3101 switch (var
->data
.depth_layout
) {
3102 case ir_depth_layout_none
:
3103 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3105 case ir_depth_layout_any
:
3106 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3108 case ir_depth_layout_greater
:
3109 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3111 case ir_depth_layout_less
:
3112 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3114 case ir_depth_layout_unchanged
:
3115 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3126 * Validate the resources used by a program versus the implementation limits
3129 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3131 unsigned total_uniform_blocks
= 0;
3132 unsigned total_shader_storage_blocks
= 0;
3134 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3135 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3140 if (sh
->Program
->info
.num_textures
>
3141 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3142 linker_error(prog
, "Too many %s shader texture samplers\n",
3143 _mesa_shader_stage_to_string(i
));
3146 if (sh
->num_uniform_components
>
3147 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3148 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3149 linker_warning(prog
, "Too many %s shader default uniform block "
3150 "components, but the driver will try to optimize "
3151 "them out; this is non-portable out-of-spec "
3153 _mesa_shader_stage_to_string(i
));
3155 linker_error(prog
, "Too many %s shader default uniform block "
3157 _mesa_shader_stage_to_string(i
));
3161 if (sh
->num_combined_uniform_components
>
3162 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3163 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3164 linker_warning(prog
, "Too many %s shader uniform components, "
3165 "but the driver will try to optimize them out; "
3166 "this is non-portable out-of-spec behavior\n",
3167 _mesa_shader_stage_to_string(i
));
3169 linker_error(prog
, "Too many %s shader uniform components\n",
3170 _mesa_shader_stage_to_string(i
));
3174 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3175 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3177 const unsigned max_uniform_blocks
=
3178 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3179 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3180 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3181 _mesa_shader_stage_to_string(i
),
3182 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3185 const unsigned max_shader_storage_blocks
=
3186 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3187 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3188 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3189 _mesa_shader_stage_to_string(i
),
3190 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3194 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3195 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3196 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3199 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3200 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3201 total_shader_storage_blocks
,
3202 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3205 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3206 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3207 ctx
->Const
.MaxUniformBlockSize
) {
3208 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3209 prog
->data
->UniformBlocks
[i
].Name
,
3210 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3211 ctx
->Const
.MaxUniformBlockSize
);
3215 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3216 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3217 ctx
->Const
.MaxShaderStorageBlockSize
) {
3218 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3219 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3220 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3221 ctx
->Const
.MaxShaderStorageBlockSize
);
3227 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3229 unsigned mask
= prog
->data
->linked_stages
;
3231 const int i
= u_bit_scan(&mask
);
3232 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3234 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3235 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3238 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3244 if (p
->sh
.NumSubroutineFunctions
== 0) {
3245 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3248 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3249 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3250 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3251 if (fn
->types
[k
] == uni
->type
) {
3257 uni
->num_compatible_subroutines
= count
;
3263 check_subroutine_resources(struct gl_shader_program
*prog
)
3265 unsigned mask
= prog
->data
->linked_stages
;
3267 const int i
= u_bit_scan(&mask
);
3268 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3270 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3271 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3272 _mesa_shader_stage_to_string(i
));
3277 * Validate shader image resources.
3280 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3282 unsigned total_image_units
= 0;
3283 unsigned fragment_outputs
= 0;
3284 unsigned total_shader_storage_blocks
= 0;
3286 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3289 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3290 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3293 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3294 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3295 _mesa_shader_stage_to_string(i
),
3296 sh
->Program
->info
.num_images
,
3297 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3299 total_image_units
+= sh
->Program
->info
.num_images
;
3300 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3302 if (i
== MESA_SHADER_FRAGMENT
) {
3303 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3304 ir_variable
*var
= node
->as_variable();
3305 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3306 /* since there are no double fs outputs - pass false */
3307 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3313 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3314 linker_error(prog
, "Too many combined image uniforms\n");
3316 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3317 ctx
->Const
.MaxCombinedShaderOutputResources
)
3318 linker_error(prog
, "Too many combined image uniforms, shader storage "
3319 " buffers and fragment outputs\n");
3324 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3325 * for a variable, checks for overlaps between other uniforms using explicit
3329 reserve_explicit_locations(struct gl_shader_program
*prog
,
3330 string_to_uint_map
*map
, ir_variable
*var
)
3332 unsigned slots
= var
->type
->uniform_locations();
3333 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3334 unsigned return_value
= slots
;
3336 /* Resize remap table if locations do not fit in the current one. */
3337 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3338 prog
->UniformRemapTable
=
3339 reralloc(prog
, prog
->UniformRemapTable
,
3340 gl_uniform_storage
*,
3343 if (!prog
->UniformRemapTable
) {
3344 linker_error(prog
, "Out of memory during linking.\n");
3348 /* Initialize allocated space. */
3349 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3350 prog
->UniformRemapTable
[i
] = NULL
;
3352 prog
->NumUniformRemapTable
= max_loc
+ 1;
3355 for (unsigned i
= 0; i
< slots
; i
++) {
3356 unsigned loc
= var
->data
.location
+ i
;
3358 /* Check if location is already used. */
3359 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3361 /* Possibly same uniform from a different stage, this is ok. */
3363 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3368 /* ARB_explicit_uniform_location specification states:
3370 * "No two default-block uniform variables in the program can have
3371 * the same location, even if they are unused, otherwise a compiler
3372 * or linker error will be generated."
3375 "location qualifier for uniform %s overlaps "
3376 "previously used location\n",
3381 /* Initialize location as inactive before optimization
3382 * rounds and location assignment.
3384 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3387 /* Note, base location used for arrays. */
3388 map
->put(var
->data
.location
, var
->name
);
3390 return return_value
;
3394 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3395 struct gl_program
*p
,
3398 unsigned slots
= var
->type
->uniform_locations();
3399 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3401 /* Resize remap table if locations do not fit in the current one. */
3402 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3403 p
->sh
.SubroutineUniformRemapTable
=
3404 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3405 gl_uniform_storage
*,
3408 if (!p
->sh
.SubroutineUniformRemapTable
) {
3409 linker_error(prog
, "Out of memory during linking.\n");
3413 /* Initialize allocated space. */
3414 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3415 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3417 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3420 for (unsigned i
= 0; i
< slots
; i
++) {
3421 unsigned loc
= var
->data
.location
+ i
;
3423 /* Check if location is already used. */
3424 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3426 /* ARB_explicit_uniform_location specification states:
3427 * "No two subroutine uniform variables can have the same location
3428 * in the same shader stage, otherwise a compiler or linker error
3429 * will be generated."
3432 "location qualifier for uniform %s overlaps "
3433 "previously used location\n",
3438 /* Initialize location as inactive before optimization
3439 * rounds and location assignment.
3441 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3447 * Check and reserve all explicit uniform locations, called before
3448 * any optimizations happen to handle also inactive uniforms and
3449 * inactive array elements that may get trimmed away.
3452 check_explicit_uniform_locations(struct gl_context
*ctx
,
3453 struct gl_shader_program
*prog
)
3455 prog
->NumExplicitUniformLocations
= 0;
3457 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3460 /* This map is used to detect if overlapping explicit locations
3461 * occur with the same uniform (from different stage) or a different one.
3463 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3466 linker_error(prog
, "Out of memory during linking.\n");
3470 unsigned entries_total
= 0;
3471 unsigned mask
= prog
->data
->linked_stages
;
3473 const int i
= u_bit_scan(&mask
);
3474 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3476 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3477 ir_variable
*var
= node
->as_variable();
3478 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3481 if (var
->data
.explicit_location
) {
3483 if (var
->type
->without_array()->is_subroutine())
3484 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3486 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3490 entries_total
+= slots
;
3501 struct empty_uniform_block
*current_block
= NULL
;
3503 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3504 /* We found empty space in UniformRemapTable. */
3505 if (prog
->UniformRemapTable
[i
] == NULL
) {
3506 /* We've found the beginning of a new continous block of empty slots */
3507 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3508 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3509 current_block
->start
= i
;
3510 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3511 ¤t_block
->link
);
3514 /* The current block continues, so we simply increment its slots */
3515 current_block
->slots
++;
3520 prog
->NumExplicitUniformLocations
= entries_total
;
3524 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3525 GLenum type
, const char *name
)
3527 bool found_interface
= false;
3528 unsigned block_name_len
= 0;
3529 const char *block_name_dot
= strchr(name
, '.');
3531 /* These rules only apply to buffer variables. So we return
3532 * true for the rest of types.
3534 if (type
!= GL_BUFFER_VARIABLE
)
3537 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3538 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3539 block_name_len
= strlen(block_name
);
3541 const char *block_square_bracket
= strchr(block_name
, '[');
3542 if (block_square_bracket
) {
3543 /* The block is part of an array of named interfaces,
3544 * for the name comparison we ignore the "[x]" part.
3546 block_name_len
-= strlen(block_square_bracket
);
3549 if (block_name_dot
) {
3550 /* Check if the variable name starts with the interface
3551 * name. The interface name (if present) should have the
3552 * length than the interface block name we are comparing to.
3554 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3555 if (len
!= block_name_len
)
3559 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3560 found_interface
= true;
3565 /* We remove the interface name from the buffer variable name,
3566 * including the dot that follows it.
3568 if (found_interface
)
3569 name
= name
+ block_name_len
+ 1;
3571 /* The ARB_program_interface_query spec says:
3573 * "For an active shader storage block member declared as an array, an
3574 * entry will be generated only for the first array element, regardless
3575 * of its type. For arrays of aggregate types, the enumeration rules
3576 * are applied recursively for the single enumerated array element."
3578 const char *struct_first_dot
= strchr(name
, '.');
3579 const char *first_square_bracket
= strchr(name
, '[');
3581 /* The buffer variable is on top level and it is not an array */
3582 if (!first_square_bracket
) {
3584 /* The shader storage block member is a struct, then generate the entry */
3585 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3588 /* Shader storage block member is an array, only generate an entry for the
3589 * first array element.
3591 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3599 add_program_resource(struct gl_shader_program
*prog
,
3600 struct set
*resource_set
,
3601 GLenum type
, const void *data
, uint8_t stages
)
3605 /* If resource already exists, do not add it again. */
3606 if (_mesa_set_search(resource_set
, data
))
3609 prog
->data
->ProgramResourceList
=
3611 prog
->data
->ProgramResourceList
,
3612 gl_program_resource
,
3613 prog
->data
->NumProgramResourceList
+ 1);
3615 if (!prog
->data
->ProgramResourceList
) {
3616 linker_error(prog
, "Out of memory during linking.\n");
3620 struct gl_program_resource
*res
=
3621 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3625 res
->StageReferences
= stages
;
3627 prog
->data
->NumProgramResourceList
++;
3629 _mesa_set_add(resource_set
, data
);
3634 /* Function checks if a variable var is a packed varying and
3635 * if given name is part of packed varying's list.
3637 * If a variable is a packed varying, it has a name like
3638 * 'packed:a,b,c' where a, b and c are separate variables.
3641 included_in_packed_varying(ir_variable
*var
, const char *name
)
3643 if (strncmp(var
->name
, "packed:", 7) != 0)
3646 char *list
= strdup(var
->name
+ 7);
3651 char *token
= strtok_r(list
, ",", &saveptr
);
3653 if (strcmp(token
, name
) == 0) {
3657 token
= strtok_r(NULL
, ",", &saveptr
);
3664 * Function builds a stage reference bitmask from variable name.
3667 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3672 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3673 * used for reference mask in gl_program_resource will need to be changed.
3675 assert(MESA_SHADER_STAGES
< 8);
3677 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3678 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3682 /* Shader symbol table may contain variables that have
3683 * been optimized away. Search IR for the variable instead.
3685 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3686 ir_variable
*var
= node
->as_variable();
3688 unsigned baselen
= strlen(var
->name
);
3690 if (included_in_packed_varying(var
, name
)) {
3695 /* Type needs to match if specified, otherwise we might
3696 * pick a variable with same name but different interface.
3698 if (var
->data
.mode
!= mode
)
3701 if (strncmp(var
->name
, name
, baselen
) == 0) {
3702 /* Check for exact name matches but also check for arrays and
3705 if (name
[baselen
] == '\0' ||
3706 name
[baselen
] == '[' ||
3707 name
[baselen
] == '.') {
3719 * Create gl_shader_variable from ir_variable class.
3721 static gl_shader_variable
*
3722 create_shader_variable(struct gl_shader_program
*shProg
,
3723 const ir_variable
*in
,
3724 const char *name
, const glsl_type
*type
,
3725 const glsl_type
*interface_type
,
3726 bool use_implicit_location
, int location
,
3727 const glsl_type
*outermost_struct_type
)
3729 /* Allocate zero-initialized memory to ensure that bitfield padding
3732 gl_shader_variable
*out
= rzalloc(shProg
, struct gl_shader_variable
);
3736 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3737 * expect to see gl_VertexID in the program resource list. Pretend.
3739 if (in
->data
.mode
== ir_var_system_value
&&
3740 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3741 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3742 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3743 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3744 (in
->data
.mode
== ir_var_system_value
&&
3745 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3746 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3747 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3748 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3749 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3750 (in
->data
.mode
== ir_var_system_value
&&
3751 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3752 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3753 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3755 out
->name
= ralloc_strdup(shProg
, name
);
3761 /* The ARB_program_interface_query spec says:
3763 * "Not all active variables are assigned valid locations; the
3764 * following variables will have an effective location of -1:
3766 * * uniforms declared as atomic counters;
3768 * * members of a uniform block;
3770 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3772 * * inputs or outputs not declared with a "location" layout
3773 * qualifier, except for vertex shader inputs and fragment shader
3776 if (in
->type
->is_atomic_uint() || is_gl_identifier(in
->name
) ||
3777 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3780 out
->location
= location
;
3784 out
->outermost_struct_type
= outermost_struct_type
;
3785 out
->interface_type
= interface_type
;
3786 out
->component
= in
->data
.location_frac
;
3787 out
->index
= in
->data
.index
;
3788 out
->patch
= in
->data
.patch
;
3789 out
->mode
= in
->data
.mode
;
3790 out
->interpolation
= in
->data
.interpolation
;
3791 out
->explicit_location
= in
->data
.explicit_location
;
3792 out
->precision
= in
->data
.precision
;
3798 add_shader_variable(const struct gl_context
*ctx
,
3799 struct gl_shader_program
*shProg
,
3800 struct set
*resource_set
,
3801 unsigned stage_mask
,
3802 GLenum programInterface
, ir_variable
*var
,
3803 const char *name
, const glsl_type
*type
,
3804 bool use_implicit_location
, int location
,
3805 const glsl_type
*outermost_struct_type
= NULL
)
3807 const glsl_type
*interface_type
= var
->get_interface_type();
3809 if (outermost_struct_type
== NULL
) {
3810 if (var
->data
.from_named_ifc_block
) {
3811 const char *interface_name
= interface_type
->name
;
3813 if (interface_type
->is_array()) {
3814 /* Issue #16 of the ARB_program_interface_query spec says:
3816 * "* If a variable is a member of an interface block without an
3817 * instance name, it is enumerated using just the variable name.
3819 * * If a variable is a member of an interface block with an
3820 * instance name, it is enumerated as "BlockName.Member", where
3821 * "BlockName" is the name of the interface block (not the
3822 * instance name) and "Member" is the name of the variable."
3824 * In particular, it indicates that it should be "BlockName",
3825 * not "BlockName[array length]". The conformance suite and
3826 * dEQP both require this behavior.
3828 * Here, we unwrap the extra array level added by named interface
3829 * block array lowering so we have the correct variable type. We
3830 * also unwrap the interface type when constructing the name.
3832 * We leave interface_type the same so that ES 3.x SSO pipeline
3833 * validation can enforce the rules requiring array length to
3834 * match on interface blocks.
3836 type
= type
->fields
.array
;
3838 interface_name
= interface_type
->fields
.array
->name
;
3841 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3845 switch (type
->base_type
) {
3846 case GLSL_TYPE_STRUCT
: {
3847 /* The ARB_program_interface_query spec says:
3849 * "For an active variable declared as a structure, a separate entry
3850 * will be generated for each active structure member. The name of
3851 * each entry is formed by concatenating the name of the structure,
3852 * the "." character, and the name of the structure member. If a
3853 * structure member to enumerate is itself a structure or array,
3854 * these enumeration rules are applied recursively."
3856 if (outermost_struct_type
== NULL
)
3857 outermost_struct_type
= type
;
3859 unsigned field_location
= location
;
3860 for (unsigned i
= 0; i
< type
->length
; i
++) {
3861 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3862 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3863 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3864 stage_mask
, programInterface
,
3865 var
, field_name
, field
->type
,
3866 use_implicit_location
, field_location
,
3867 outermost_struct_type
))
3870 field_location
+= field
->type
->count_attribute_slots(false);
3876 /* The ARB_program_interface_query spec says:
3878 * "For an active variable declared as a single instance of a basic
3879 * type, a single entry will be generated, using the variable name
3880 * from the shader source."
3882 gl_shader_variable
*sha_v
=
3883 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3884 use_implicit_location
, location
,
3885 outermost_struct_type
);
3889 return add_program_resource(shProg
, resource_set
,
3890 programInterface
, sha_v
, stage_mask
);
3896 add_interface_variables(const struct gl_context
*ctx
,
3897 struct gl_shader_program
*shProg
,
3898 struct set
*resource_set
,
3899 unsigned stage
, GLenum programInterface
)
3901 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3903 foreach_in_list(ir_instruction
, node
, ir
) {
3904 ir_variable
*var
= node
->as_variable();
3906 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3911 switch (var
->data
.mode
) {
3912 case ir_var_system_value
:
3913 case ir_var_shader_in
:
3914 if (programInterface
!= GL_PROGRAM_INPUT
)
3916 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3917 : int(VARYING_SLOT_VAR0
);
3919 case ir_var_shader_out
:
3920 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3922 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3923 : int(VARYING_SLOT_VAR0
);
3929 if (var
->data
.patch
)
3930 loc_bias
= int(VARYING_SLOT_PATCH0
);
3932 /* Skip packed varyings, packed varyings are handled separately
3933 * by add_packed_varyings.
3935 if (strncmp(var
->name
, "packed:", 7) == 0)
3938 /* Skip fragdata arrays, these are handled separately
3939 * by add_fragdata_arrays.
3941 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3944 const bool vs_input_or_fs_output
=
3945 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3946 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3948 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3949 1 << stage
, programInterface
,
3950 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3951 var
->data
.location
- loc_bias
))
3958 add_packed_varyings(const struct gl_context
*ctx
,
3959 struct gl_shader_program
*shProg
,
3960 struct set
*resource_set
,
3961 int stage
, GLenum type
)
3963 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3966 if (!sh
|| !sh
->packed_varyings
)
3969 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3970 ir_variable
*var
= node
->as_variable();
3972 switch (var
->data
.mode
) {
3973 case ir_var_shader_in
:
3974 iface
= GL_PROGRAM_INPUT
;
3976 case ir_var_shader_out
:
3977 iface
= GL_PROGRAM_OUTPUT
;
3980 unreachable("unexpected type");
3983 if (type
== iface
) {
3984 const int stage_mask
=
3985 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3986 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3988 iface
, var
, var
->name
, var
->type
, false,
3989 var
->data
.location
- VARYING_SLOT_VAR0
))
3998 add_fragdata_arrays(const struct gl_context
*ctx
,
3999 struct gl_shader_program
*shProg
,
4000 struct set
*resource_set
)
4002 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4004 if (!sh
|| !sh
->fragdata_arrays
)
4007 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
4008 ir_variable
*var
= node
->as_variable();
4010 assert(var
->data
.mode
== ir_var_shader_out
);
4012 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4013 1 << MESA_SHADER_FRAGMENT
,
4014 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
4015 true, var
->data
.location
- FRAG_RESULT_DATA0
))
4023 get_top_level_name(const char *name
)
4025 const char *first_dot
= strchr(name
, '.');
4026 const char *first_square_bracket
= strchr(name
, '[');
4029 /* The ARB_program_interface_query spec says:
4031 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4032 * the number of active array elements of the top-level shader storage
4033 * block member containing to the active variable is written to
4034 * <params>. If the top-level block member is not declared as an
4035 * array, the value one is written to <params>. If the top-level block
4036 * member is an array with no declared size, the value zero is written
4040 /* The buffer variable is on top level.*/
4041 if (!first_square_bracket
&& !first_dot
)
4042 name_size
= strlen(name
);
4043 else if ((!first_square_bracket
||
4044 (first_dot
&& first_dot
< first_square_bracket
)))
4045 name_size
= first_dot
- name
;
4047 name_size
= first_square_bracket
- name
;
4049 return strndup(name
, name_size
);
4053 get_var_name(const char *name
)
4055 const char *first_dot
= strchr(name
, '.');
4058 return strdup(name
);
4060 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4064 is_top_level_shader_storage_block_member(const char* name
,
4065 const char* interface_name
,
4066 const char* field_name
)
4068 bool result
= false;
4070 /* If the given variable is already a top-level shader storage
4071 * block member, then return array_size = 1.
4072 * We could have two possibilities: if we have an instanced
4073 * shader storage block or not instanced.
4075 * For the first, we check create a name as it was in top level and
4076 * compare it with the real name. If they are the same, then
4077 * the variable is already at top-level.
4079 * Full instanced name is: interface name + '.' + var name +
4082 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4083 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4084 if (!full_instanced_name
) {
4085 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4089 snprintf(full_instanced_name
, name_length
, "%s.%s",
4090 interface_name
, field_name
);
4092 /* Check if its top-level shader storage block member of an
4093 * instanced interface block, or of a unnamed interface block.
4095 if (strcmp(name
, full_instanced_name
) == 0 ||
4096 strcmp(name
, field_name
) == 0)
4099 free(full_instanced_name
);
4104 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4105 char *interface_name
, char *var_name
)
4107 /* The ARB_program_interface_query spec says:
4109 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4110 * the number of active array elements of the top-level shader storage
4111 * block member containing to the active variable is written to
4112 * <params>. If the top-level block member is not declared as an
4113 * array, the value one is written to <params>. If the top-level block
4114 * member is an array with no declared size, the value zero is written
4117 if (is_top_level_shader_storage_block_member(uni
->name
,
4121 else if (field
->type
->is_unsized_array())
4123 else if (field
->type
->is_array())
4124 return field
->type
->length
;
4130 get_array_stride(struct gl_context
*ctx
, struct gl_uniform_storage
*uni
,
4131 const glsl_type
*interface
, const glsl_struct_field
*field
,
4132 char *interface_name
, char *var_name
)
4134 /* The ARB_program_interface_query spec says:
4136 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4137 * identifying the stride between array elements of the top-level
4138 * shader storage block member containing the active variable is
4139 * written to <params>. For top-level block members declared as
4140 * arrays, the value written is the difference, in basic machine units,
4141 * between the offsets of the active variable for consecutive elements
4142 * in the top-level array. For top-level block members not declared as
4143 * an array, zero is written to <params>."
4145 if (field
->type
->is_array()) {
4146 const enum glsl_matrix_layout matrix_layout
=
4147 glsl_matrix_layout(field
->matrix_layout
);
4148 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4149 const glsl_type
*array_type
= field
->type
->fields
.array
;
4151 if (is_top_level_shader_storage_block_member(uni
->name
,
4156 if (GLSL_INTERFACE_PACKING_STD140
==
4158 get_internal_ifc_packing(ctx
->Const
.UseSTD430AsDefaultPacking
)) {
4159 if (array_type
->is_record() || array_type
->is_array())
4160 return glsl_align(array_type
->std140_size(row_major
), 16);
4162 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4164 return array_type
->std430_array_stride(row_major
);
4171 calculate_array_size_and_stride(struct gl_context
*ctx
,
4172 struct gl_shader_program
*shProg
,
4173 struct gl_uniform_storage
*uni
)
4175 int block_index
= uni
->block_index
;
4176 int array_size
= -1;
4177 int array_stride
= -1;
4178 char *var_name
= get_top_level_name(uni
->name
);
4179 char *interface_name
=
4180 get_top_level_name(uni
->is_shader_storage
?
4181 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4182 shProg
->data
->UniformBlocks
[block_index
].Name
);
4184 if (strcmp(var_name
, interface_name
) == 0) {
4185 /* Deal with instanced array of SSBOs */
4186 char *temp_name
= get_var_name(uni
->name
);
4188 linker_error(shProg
, "Out of memory during linking.\n");
4189 goto write_top_level_array_size_and_stride
;
4192 var_name
= get_top_level_name(temp_name
);
4195 linker_error(shProg
, "Out of memory during linking.\n");
4196 goto write_top_level_array_size_and_stride
;
4200 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4201 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4205 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4206 ir_variable
*var
= node
->as_variable();
4207 if (!var
|| !var
->get_interface_type() ||
4208 var
->data
.mode
!= ir_var_shader_storage
)
4211 const glsl_type
*interface
= var
->get_interface_type();
4213 if (strcmp(interface_name
, interface
->name
) != 0)
4216 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4217 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4218 if (strcmp(field
->name
, var_name
) != 0)
4221 array_stride
= get_array_stride(ctx
, uni
, interface
, field
,
4222 interface_name
, var_name
);
4223 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4224 goto write_top_level_array_size_and_stride
;
4228 write_top_level_array_size_and_stride
:
4229 free(interface_name
);
4231 uni
->top_level_array_stride
= array_stride
;
4232 uni
->top_level_array_size
= array_size
;
4236 * Builds up a list of program resources that point to existing
4240 build_program_resource_list(struct gl_context
*ctx
,
4241 struct gl_shader_program
*shProg
)
4243 /* Rebuild resource list. */
4244 if (shProg
->data
->ProgramResourceList
) {
4245 ralloc_free(shProg
->data
->ProgramResourceList
);
4246 shProg
->data
->ProgramResourceList
= NULL
;
4247 shProg
->data
->NumProgramResourceList
= 0;
4250 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4252 /* Determine first input and final output stage. These are used to
4253 * detect which variables should be enumerated in the resource list
4254 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4256 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4257 if (!shProg
->_LinkedShaders
[i
])
4259 if (input_stage
== MESA_SHADER_STAGES
)
4264 /* Empty shader, no resources. */
4265 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4268 struct set
*resource_set
= _mesa_set_create(NULL
,
4270 _mesa_key_pointer_equal
);
4272 /* Program interface needs to expose varyings in case of SSO. */
4273 if (shProg
->SeparateShader
) {
4274 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4275 input_stage
, GL_PROGRAM_INPUT
))
4278 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4279 output_stage
, GL_PROGRAM_OUTPUT
))
4283 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4286 /* Add inputs and outputs to the resource list. */
4287 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4288 input_stage
, GL_PROGRAM_INPUT
))
4291 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4292 output_stage
, GL_PROGRAM_OUTPUT
))
4295 if (shProg
->last_vert_prog
) {
4296 struct gl_transform_feedback_info
*linked_xfb
=
4297 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4299 /* Add transform feedback varyings. */
4300 if (linked_xfb
->NumVarying
> 0) {
4301 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4302 if (!add_program_resource(shProg
, resource_set
,
4303 GL_TRANSFORM_FEEDBACK_VARYING
,
4304 &linked_xfb
->Varyings
[i
], 0))
4309 /* Add transform feedback buffers. */
4310 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4311 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4312 linked_xfb
->Buffers
[i
].Binding
= i
;
4313 if (!add_program_resource(shProg
, resource_set
,
4314 GL_TRANSFORM_FEEDBACK_BUFFER
,
4315 &linked_xfb
->Buffers
[i
], 0))
4321 /* Add uniforms from uniform storage. */
4322 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4323 /* Do not add uniforms internally used by Mesa. */
4324 if (shProg
->data
->UniformStorage
[i
].hidden
)
4328 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4331 /* Add stagereferences for uniforms in a uniform block. */
4332 bool is_shader_storage
=
4333 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4334 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4335 if (block_index
!= -1) {
4336 stageref
|= is_shader_storage
?
4337 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4338 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4341 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4342 if (!should_add_buffer_variable(shProg
, type
,
4343 shProg
->data
->UniformStorage
[i
].name
))
4346 if (is_shader_storage
) {
4347 calculate_array_size_and_stride(ctx
, shProg
,
4348 &shProg
->data
->UniformStorage
[i
]);
4351 if (!add_program_resource(shProg
, resource_set
, type
,
4352 &shProg
->data
->UniformStorage
[i
], stageref
))
4356 /* Add program uniform blocks. */
4357 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4358 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4359 &shProg
->data
->UniformBlocks
[i
], 0))
4363 /* Add program shader storage blocks. */
4364 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4365 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4366 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4370 /* Add atomic counter buffers. */
4371 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4372 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4373 &shProg
->data
->AtomicBuffers
[i
], 0))
4377 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4379 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4382 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4383 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4384 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4387 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4388 /* add shader subroutines */
4389 if (!add_program_resource(shProg
, resource_set
,
4390 type
, &shProg
->data
->UniformStorage
[i
], 0))
4395 unsigned mask
= shProg
->data
->linked_stages
;
4397 const int i
= u_bit_scan(&mask
);
4398 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4400 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4401 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4402 if (!add_program_resource(shProg
, resource_set
,
4403 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4408 _mesa_set_destroy(resource_set
, NULL
);
4412 * This check is done to make sure we allow only constant expression
4413 * indexing and "constant-index-expression" (indexing with an expression
4414 * that includes loop induction variable).
4417 validate_sampler_array_indexing(struct gl_context
*ctx
,
4418 struct gl_shader_program
*prog
)
4420 dynamic_sampler_array_indexing_visitor v
;
4421 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4422 if (prog
->_LinkedShaders
[i
] == NULL
)
4425 bool no_dynamic_indexing
=
4426 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4428 /* Search for array derefs in shader. */
4429 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4430 if (v
.uses_dynamic_sampler_array_indexing()) {
4431 const char *msg
= "sampler arrays indexed with non-constant "
4432 "expressions is forbidden in GLSL %s %u";
4433 /* Backend has indicated that it has no dynamic indexing support. */
4434 if (no_dynamic_indexing
) {
4435 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4436 prog
->data
->Version
);
4439 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4440 prog
->data
->Version
);
4448 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4450 unsigned mask
= prog
->data
->linked_stages
;
4452 const int i
= u_bit_scan(&mask
);
4453 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4455 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4456 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4457 ir_function
*fn
= node
->as_function();
4461 if (fn
->is_subroutine
)
4462 p
->sh
.NumSubroutineUniformTypes
++;
4464 if (!fn
->num_subroutine_types
)
4467 /* these should have been calculated earlier. */
4468 assert(fn
->subroutine_index
!= -1);
4469 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4470 linker_error(prog
, "Too many subroutine functions declared.\n");
4473 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4474 struct gl_subroutine_function
,
4475 p
->sh
.NumSubroutineFunctions
+ 1);
4476 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4477 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4478 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4479 ralloc_array(p
, const struct glsl_type
*,
4480 fn
->num_subroutine_types
);
4482 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4485 * "Each subroutine with an index qualifier in the shader must be
4486 * given a unique index, otherwise a compile or link error will be
4489 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4490 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4491 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4492 linker_error(prog
, "each subroutine index qualifier in the "
4493 "shader must be unique\n");
4497 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4498 fn
->subroutine_index
;
4500 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4501 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4503 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4504 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4505 p
->sh
.NumSubroutineFunctions
++;
4511 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4513 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4515 foreach_in_list(ir_instruction
, node
, ir
) {
4516 ir_variable
*const var
= node
->as_variable();
4518 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4521 /* Don't set always active on builtins that haven't been redeclared */
4522 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4525 var
->data
.always_active_io
= true;
4530 * When separate shader programs are enabled, only input/outputs between
4531 * the stages of a multi-stage separate program can be safely removed
4532 * from the shader interface. Other inputs/outputs must remain active.
4535 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4537 unsigned first
, last
;
4538 assert(prog
->SeparateShader
);
4540 first
= MESA_SHADER_STAGES
;
4543 /* Determine first and last stage. Excluding the compute stage */
4544 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4545 if (!prog
->_LinkedShaders
[i
])
4547 if (first
== MESA_SHADER_STAGES
)
4552 if (first
== MESA_SHADER_STAGES
)
4555 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4556 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4560 /* Prevent the removal of inputs to the first and outputs from the last
4561 * stage, unless they are the initial pipeline inputs or final pipeline
4562 * outputs, respectively.
4564 * The removal of IO between shaders in the same program is always
4567 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4568 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4569 if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4570 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4575 link_and_validate_uniforms(struct gl_context
*ctx
,
4576 struct gl_shader_program
*prog
)
4578 update_array_sizes(prog
);
4579 link_assign_uniform_locations(prog
, ctx
);
4581 if (!prog
->data
->cache_fallback
) {
4582 link_assign_atomic_counter_resources(ctx
, prog
);
4583 link_calculate_subroutine_compat(prog
);
4584 check_resources(ctx
, prog
);
4585 check_subroutine_resources(prog
);
4586 check_image_resources(ctx
, prog
);
4587 link_check_atomic_counter_resources(ctx
, prog
);
4592 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4593 struct gl_context
*ctx
,
4594 struct gl_shader_program
*prog
, void *mem_ctx
)
4596 /* Mark all generic shader inputs and outputs as unpaired. */
4597 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4598 if (prog
->_LinkedShaders
[i
] != NULL
) {
4599 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4603 unsigned prev
= first
;
4604 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4605 if (prog
->_LinkedShaders
[i
] == NULL
)
4608 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4609 prog
->_LinkedShaders
[i
]);
4613 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4614 MESA_SHADER_VERTEX
)) {
4618 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4619 MESA_SHADER_FRAGMENT
)) {
4623 prog
->last_vert_prog
= NULL
;
4624 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4625 if (prog
->_LinkedShaders
[i
] == NULL
)
4628 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4632 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4635 link_and_validate_uniforms(ctx
, prog
);
4637 if (!prog
->data
->LinkStatus
)
4640 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4641 if (prog
->_LinkedShaders
[i
] == NULL
)
4644 const struct gl_shader_compiler_options
*options
=
4645 &ctx
->Const
.ShaderCompilerOptions
[i
];
4647 if (options
->LowerBufferInterfaceBlocks
)
4648 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4649 options
->ClampBlockIndicesToArrayBounds
,
4650 ctx
->Const
.UseSTD430AsDefaultPacking
);
4652 if (i
== MESA_SHADER_COMPUTE
)
4653 lower_shared_reference(ctx
, prog
, prog
->_LinkedShaders
[i
]);
4655 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4656 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4663 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4666 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4667 /* Run it just once. */
4668 do_common_optimization(ir
, true, false,
4669 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4670 ctx
->Const
.NativeIntegers
);
4672 /* Repeat it until it stops making changes. */
4673 while (do_common_optimization(ir
, true, false,
4674 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4675 ctx
->Const
.NativeIntegers
))
4681 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4683 prog
->data
->LinkStatus
= linking_success
; /* All error paths will set this to false */
4684 prog
->data
->Validated
= false;
4686 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4688 * "Linking can fail for a variety of reasons as specified in the
4689 * OpenGL Shading Language Specification, as well as any of the
4690 * following reasons:
4692 * - No shader objects are attached to program."
4694 * The Compatibility Profile specification does not list the error. In
4695 * Compatibility Profile missing shader stages are replaced by
4696 * fixed-function. This applies to the case where all stages are
4699 if (prog
->NumShaders
== 0) {
4700 if (ctx
->API
!= API_OPENGL_COMPAT
)
4701 linker_error(prog
, "no shaders attached to the program\n");
4705 #ifdef ENABLE_SHADER_CACHE
4706 /* If transform feedback used on the program then compile all shaders. */
4707 bool skip_cache
= false;
4708 if (prog
->TransformFeedback
.NumVarying
> 0) {
4709 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4710 _mesa_glsl_compile_shader(ctx
, prog
->Shaders
[i
], false, false, true);
4715 if (!skip_cache
&& shader_cache_read_program_metadata(ctx
, prog
))
4719 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4721 prog
->ARB_fragment_coord_conventions_enable
= false;
4723 /* Separate the shaders into groups based on their type.
4725 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4726 unsigned num_shaders
[MESA_SHADER_STAGES
];
4728 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4729 shader_list
[i
] = (struct gl_shader
**)
4730 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4734 unsigned min_version
= UINT_MAX
;
4735 unsigned max_version
= 0;
4736 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4737 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4738 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4740 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4741 linker_error(prog
, "all shaders must use same shading "
4742 "language version\n");
4746 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4747 prog
->ARB_fragment_coord_conventions_enable
= true;
4750 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4751 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4752 num_shaders
[shader_type
]++;
4755 /* In desktop GLSL, different shader versions may be linked together. In
4756 * GLSL ES, all shader versions must be the same.
4758 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4759 linker_error(prog
, "all shaders must use same shading "
4760 "language version\n");
4764 prog
->data
->Version
= max_version
;
4765 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4767 /* Some shaders have to be linked with some other shaders present.
4769 if (!prog
->SeparateShader
) {
4770 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4771 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4772 linker_error(prog
, "Geometry shader must be linked with "
4776 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4777 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4778 linker_error(prog
, "Tessellation evaluation shader must be linked "
4779 "with vertex shader\n");
4782 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4783 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4784 linker_error(prog
, "Tessellation control shader must be linked with "
4789 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4791 * "Linking can fail for [...] any of the following reasons:
4793 * * program contains an object to form a tessellation control
4794 * shader [...] and [...] the program is not separable and
4795 * contains no object to form a tessellation evaluation shader"
4797 * The OpenGL spec is contradictory. It allows linking without a tess
4798 * eval shader, but that can only be used with transform feedback and
4799 * rasterization disabled. However, transform feedback isn't allowed
4800 * with GL_PATCHES, so it can't be used.
4802 * More investigation showed that the idea of transform feedback after
4803 * a tess control shader was dropped, because some hw vendors couldn't
4804 * support tessellation without a tess eval shader, but the linker
4805 * section wasn't updated to reflect that.
4807 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4810 * Do what's reasonable and always require a tess eval shader if a tess
4811 * control shader is present.
4813 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4814 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4815 linker_error(prog
, "Tessellation control shader must be linked with "
4816 "tessellation evaluation shader\n");
4821 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4822 num_shaders
[MESA_SHADER_TESS_CTRL
] == 0) {
4823 linker_error(prog
, "GLSL ES requires non-separable programs "
4824 "containing a tessellation evaluation shader to also "
4825 "be linked with a tessellation control shader\n");
4831 /* Compute shaders have additional restrictions. */
4832 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4833 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4834 linker_error(prog
, "Compute shaders may not be linked with any other "
4835 "type of shader\n");
4838 /* Link all shaders for a particular stage and validate the result.
4840 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4841 if (num_shaders
[stage
] > 0) {
4842 gl_linked_shader
*const sh
=
4843 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4844 num_shaders
[stage
], false);
4846 if (!prog
->data
->LinkStatus
) {
4848 _mesa_delete_linked_shader(ctx
, sh
);
4853 case MESA_SHADER_VERTEX
:
4854 validate_vertex_shader_executable(prog
, sh
, ctx
);
4856 case MESA_SHADER_TESS_CTRL
:
4857 /* nothing to be done */
4859 case MESA_SHADER_TESS_EVAL
:
4860 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4862 case MESA_SHADER_GEOMETRY
:
4863 validate_geometry_shader_executable(prog
, sh
, ctx
);
4865 case MESA_SHADER_FRAGMENT
:
4866 validate_fragment_shader_executable(prog
, sh
);
4869 if (!prog
->data
->LinkStatus
) {
4871 _mesa_delete_linked_shader(ctx
, sh
);
4875 prog
->_LinkedShaders
[stage
] = sh
;
4876 prog
->data
->linked_stages
|= 1 << stage
;
4880 /* Here begins the inter-stage linking phase. Some initial validation is
4881 * performed, then locations are assigned for uniforms, attributes, and
4884 cross_validate_uniforms(prog
);
4885 if (!prog
->data
->LinkStatus
)
4888 unsigned first
, last
, prev
;
4890 first
= MESA_SHADER_STAGES
;
4893 /* Determine first and last stage. */
4894 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4895 if (!prog
->_LinkedShaders
[i
])
4897 if (first
== MESA_SHADER_STAGES
)
4902 if (!prog
->data
->cache_fallback
) {
4903 check_explicit_uniform_locations(ctx
, prog
);
4904 link_assign_subroutine_types(prog
);
4907 if (!prog
->data
->LinkStatus
)
4910 resize_tes_inputs(ctx
, prog
);
4912 /* Validate the inputs of each stage with the output of the preceding
4916 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4917 if (prog
->_LinkedShaders
[i
] == NULL
)
4920 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4921 prog
->_LinkedShaders
[i
]);
4922 if (!prog
->data
->LinkStatus
)
4925 cross_validate_outputs_to_inputs(ctx
, prog
,
4926 prog
->_LinkedShaders
[prev
],
4927 prog
->_LinkedShaders
[i
]);
4928 if (!prog
->data
->LinkStatus
)
4934 /* The cross validation of outputs/inputs above validates explicit locations
4935 * but for SSO programs we need to do this also for the inputs in the
4936 * first stage and outputs of the last stage included in the program, since
4937 * there is no cross validation for these.
4939 if (prog
->SeparateShader
)
4940 validate_sso_explicit_locations(ctx
, prog
,
4941 (gl_shader_stage
) first
,
4942 (gl_shader_stage
) last
);
4944 /* Cross-validate uniform blocks between shader stages */
4945 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4946 if (!prog
->data
->LinkStatus
)
4949 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4950 if (prog
->_LinkedShaders
[i
] != NULL
)
4951 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4954 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4955 * it before optimization because we want most of the checks to get
4956 * dropped thanks to constant propagation.
4958 * This rule also applies to GLSL ES 3.00.
4960 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4961 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4963 lower_discard_flow(sh
->ir
);
4967 if (prog
->SeparateShader
)
4968 disable_varying_optimizations_for_sso(prog
);
4970 if (!prog
->data
->cache_fallback
) {
4972 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4976 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4980 /* Do common optimization before assigning storage for attributes,
4981 * uniforms, and varyings. Later optimization could possibly make
4982 * some of that unused.
4984 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4985 if (prog
->_LinkedShaders
[i
] == NULL
)
4988 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4989 if (!prog
->data
->LinkStatus
)
4992 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4993 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4996 if (ctx
->Const
.LowerTessLevel
) {
4997 lower_tess_level(prog
->_LinkedShaders
[i
]);
5000 /* Call opts before lowering const arrays to uniforms so we can const
5001 * propagate any elements accessed directly.
5003 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5005 /* Call opts after lowering const arrays to copy propagate things. */
5006 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
5007 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5009 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
5012 /* Validation for special cases where we allow sampler array indexing
5013 * with loop induction variable. This check emits a warning or error
5014 * depending if backend can handle dynamic indexing.
5016 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
5017 (prog
->IsES
&& prog
->data
->Version
< 300)) {
5018 if (!validate_sampler_array_indexing(ctx
, prog
))
5022 /* Check and validate stream emissions in geometry shaders */
5023 validate_geometry_shader_emissions(ctx
, prog
);
5025 store_fragdepth_layout(prog
);
5027 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
5030 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
5031 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
5032 * anything about shader linking when one of the shaders (vertex or
5033 * fragment shader) is absent. So, the extension shouldn't change the
5034 * behavior specified in GLSL specification.
5036 * From OpenGL ES 3.1 specification (7.3 Program Objects):
5037 * "Linking can fail for a variety of reasons as specified in the
5038 * OpenGL ES Shading Language Specification, as well as any of the
5039 * following reasons:
5043 * * program contains objects to form either a vertex shader or
5044 * fragment shader, and program is not separable, and does not
5045 * contain objects to form both a vertex shader and fragment
5048 * However, the only scenario in 3.1+ where we don't require them both is
5049 * when we have a compute shader. For example:
5051 * - No shaders is a link error.
5052 * - Geom or Tess without a Vertex shader is a link error which means we
5053 * always require a Vertex shader and hence a Fragment shader.
5054 * - Finally a Compute shader linked with any other stage is a link error.
5056 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
5057 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
5058 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
5059 linker_error(prog
, "program lacks a vertex shader\n");
5060 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5061 linker_error(prog
, "program lacks a fragment shader\n");
5066 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5067 free(shader_list
[i
]);
5068 if (prog
->_LinkedShaders
[i
] == NULL
)
5071 /* Do a final validation step to make sure that the IR wasn't
5072 * invalidated by any modifications performed after intrastage linking.
5074 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5076 /* Retain any live IR, but trash the rest. */
5077 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5079 /* The symbol table in the linked shaders may contain references to
5080 * variables that were removed (e.g., unused uniforms). Since it may
5081 * contain junk, there is no possible valid use. Delete it and set the
5084 delete prog
->_LinkedShaders
[i
]->symbols
;
5085 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5088 ralloc_free(mem_ctx
);