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 "glsl_symbol_table.h"
70 #include "glsl_parser_extras.h"
73 #include "program/prog_instruction.h"
74 #include "program/program.h"
75 #include "util/mesa-sha1.h"
77 #include "string_to_uint_map.h"
79 #include "linker_util.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"
86 #include "util/u_string.h"
87 #include "util/u_math.h"
89 #include "main/imports.h"
90 #include "main/shaderobj.h"
91 #include "main/enums.h"
92 #include "main/mtypes.h"
97 struct find_variable
{
101 find_variable(const char *name
) : name(name
), found(false) {}
105 * Visitor that determines whether or not a variable is ever written.
107 * Use \ref find_assignments for convenience.
109 class find_assignment_visitor
: public ir_hierarchical_visitor
{
111 find_assignment_visitor(unsigned num_vars
,
112 find_variable
* const *vars
)
113 : num_variables(num_vars
), num_found(0), variables(vars
)
117 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
119 ir_variable
*const var
= ir
->lhs
->variable_referenced();
121 return check_variable_name(var
->name
);
124 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
126 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
127 actual_node
, &ir
->actual_parameters
) {
128 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
129 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
131 if (sig_param
->data
.mode
== ir_var_function_out
||
132 sig_param
->data
.mode
== ir_var_function_inout
) {
133 ir_variable
*var
= param_rval
->variable_referenced();
134 if (var
&& check_variable_name(var
->name
) == visit_stop
)
139 if (ir
->return_deref
!= NULL
) {
140 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
142 if (check_variable_name(var
->name
) == visit_stop
)
146 return visit_continue_with_parent
;
150 ir_visitor_status
check_variable_name(const char *name
)
152 for (unsigned i
= 0; i
< num_variables
; ++i
) {
153 if (strcmp(variables
[i
]->name
, name
) == 0) {
154 if (!variables
[i
]->found
) {
155 variables
[i
]->found
= true;
157 assert(num_found
< num_variables
);
158 if (++num_found
== num_variables
)
165 return visit_continue_with_parent
;
169 unsigned num_variables
; /**< Number of variables to find */
170 unsigned num_found
; /**< Number of variables already found */
171 find_variable
* const *variables
; /**< Variables to find */
175 * Determine whether or not any of NULL-terminated list of variables is ever
179 find_assignments(exec_list
*ir
, find_variable
* const *vars
)
181 unsigned num_variables
= 0;
183 for (find_variable
* const *v
= vars
; *v
; ++v
)
186 find_assignment_visitor
visitor(num_variables
, vars
);
191 * Determine whether or not the given variable is ever written to.
194 find_assignments(exec_list
*ir
, find_variable
*var
)
196 find_assignment_visitor
visitor(1, &var
);
201 * Visitor that determines whether or not a variable is ever read.
203 class find_deref_visitor
: public ir_hierarchical_visitor
{
205 find_deref_visitor(const char *name
)
206 : name(name
), found(false)
211 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
213 if (strcmp(this->name
, ir
->var
->name
) == 0) {
218 return visit_continue
;
221 bool variable_found() const
227 const char *name
; /**< Find writes to a variable with this name. */
228 bool found
; /**< Was a write to the variable found? */
233 * A visitor helper that provides methods for updating the types of
234 * ir_dereferences. Classes that update variable types (say, updating
235 * array sizes) will want to use this so that dereference types stay in sync.
237 class deref_type_updater
: public ir_hierarchical_visitor
{
239 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
241 ir
->type
= ir
->var
->type
;
242 return visit_continue
;
245 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
247 const glsl_type
*const vt
= ir
->array
->type
;
249 ir
->type
= vt
->fields
.array
;
250 return visit_continue
;
253 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
255 ir
->type
= ir
->record
->type
->fields
.structure
[ir
->field_idx
].type
;
256 return visit_continue
;
261 class array_resize_visitor
: public deref_type_updater
{
263 unsigned num_vertices
;
264 gl_shader_program
*prog
;
265 gl_shader_stage stage
;
267 array_resize_visitor(unsigned num_vertices
,
268 gl_shader_program
*prog
,
269 gl_shader_stage stage
)
271 this->num_vertices
= num_vertices
;
276 virtual ~array_resize_visitor()
281 virtual ir_visitor_status
visit(ir_variable
*var
)
283 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
285 return visit_continue
;
287 unsigned size
= var
->type
->length
;
289 if (stage
== MESA_SHADER_GEOMETRY
) {
290 /* Generate a link error if the shader has declared this array with
293 if (!var
->data
.implicit_sized_array
&&
294 size
&& size
!= this->num_vertices
) {
295 linker_error(this->prog
, "size of array %s declared as %u, "
296 "but number of input vertices is %u\n",
297 var
->name
, size
, this->num_vertices
);
298 return visit_continue
;
301 /* Generate a link error if the shader attempts to access an input
302 * array using an index too large for its actual size assigned at
305 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
306 linker_error(this->prog
, "%s shader accesses element %i of "
307 "%s, but only %i input vertices\n",
308 _mesa_shader_stage_to_string(this->stage
),
309 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
310 return visit_continue
;
314 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
316 var
->data
.max_array_access
= this->num_vertices
- 1;
318 return visit_continue
;
323 * Visitor that determines the highest stream id to which a (geometry) shader
324 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
326 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
328 find_emit_vertex_visitor(int max_allowed
)
329 : max_stream_allowed(max_allowed
),
330 invalid_stream_id(0),
331 invalid_stream_id_from_emit_vertex(false),
332 end_primitive_found(false),
333 uses_non_zero_stream(false)
338 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
340 int stream_id
= ir
->stream_id();
343 invalid_stream_id
= stream_id
;
344 invalid_stream_id_from_emit_vertex
= true;
348 if (stream_id
> max_stream_allowed
) {
349 invalid_stream_id
= stream_id
;
350 invalid_stream_id_from_emit_vertex
= true;
355 uses_non_zero_stream
= true;
357 return visit_continue
;
360 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
362 end_primitive_found
= true;
364 int stream_id
= ir
->stream_id();
367 invalid_stream_id
= stream_id
;
368 invalid_stream_id_from_emit_vertex
= false;
372 if (stream_id
> max_stream_allowed
) {
373 invalid_stream_id
= stream_id
;
374 invalid_stream_id_from_emit_vertex
= false;
379 uses_non_zero_stream
= true;
381 return visit_continue
;
386 return invalid_stream_id
!= 0;
389 const char *error_func()
391 return invalid_stream_id_from_emit_vertex
?
392 "EmitStreamVertex" : "EndStreamPrimitive";
397 return invalid_stream_id
;
402 return uses_non_zero_stream
;
405 bool uses_end_primitive()
407 return end_primitive_found
;
411 int max_stream_allowed
;
412 int invalid_stream_id
;
413 bool invalid_stream_id_from_emit_vertex
;
414 bool end_primitive_found
;
415 bool uses_non_zero_stream
;
418 /* Class that finds array derefs and check if indexes are dynamic. */
419 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
422 dynamic_sampler_array_indexing_visitor() :
423 dynamic_sampler_array_indexing(false)
427 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
429 if (!ir
->variable_referenced())
430 return visit_continue
;
432 if (!ir
->variable_referenced()->type
->contains_sampler())
433 return visit_continue
;
435 if (!ir
->array_index
->constant_expression_value(ralloc_parent(ir
))) {
436 dynamic_sampler_array_indexing
= true;
439 return visit_continue
;
442 bool uses_dynamic_sampler_array_indexing()
444 return dynamic_sampler_array_indexing
;
448 bool dynamic_sampler_array_indexing
;
451 } /* anonymous namespace */
454 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
458 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
460 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
463 prog
->data
->LinkStatus
= LINKING_FAILURE
;
468 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
472 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
474 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
481 * Given a string identifying a program resource, break it into a base name
482 * and an optional array index in square brackets.
484 * If an array index is present, \c out_base_name_end is set to point to the
485 * "[" that precedes the array index, and the array index itself is returned
488 * If no array index is present (or if the array index is negative or
489 * mal-formed), \c out_base_name_end, is set to point to the null terminator
490 * at the end of the input string, and -1 is returned.
492 * Only the final array index is parsed; if the string contains other array
493 * indices (or structure field accesses), they are left in the base name.
495 * No attempt is made to check that the base name is properly formed;
496 * typically the caller will look up the base name in a hash table, so
497 * ill-formed base names simply turn into hash table lookup failures.
500 parse_program_resource_name(const GLchar
*name
,
501 const GLchar
**out_base_name_end
)
503 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
505 * "When an integer array element or block instance number is part of
506 * the name string, it will be specified in decimal form without a "+"
507 * or "-" sign or any extra leading zeroes. Additionally, the name
508 * string will not include white space anywhere in the string."
511 const size_t len
= strlen(name
);
512 *out_base_name_end
= name
+ len
;
514 if (len
== 0 || name
[len
-1] != ']')
517 /* Walk backwards over the string looking for a non-digit character. This
518 * had better be the opening bracket for an array index.
520 * Initially, i specifies the location of the ']'. Since the string may
521 * contain only the ']' charcater, walk backwards very carefully.
524 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
527 if ((i
== 0) || name
[i
-1] != '[')
530 long array_index
= strtol(&name
[i
], NULL
, 10);
534 /* Check for leading zero */
535 if (name
[i
] == '0' && name
[i
+1] != ']')
538 *out_base_name_end
= name
+ (i
- 1);
544 link_invalidate_variable_locations(exec_list
*ir
)
546 foreach_in_list(ir_instruction
, node
, ir
) {
547 ir_variable
*const var
= node
->as_variable();
552 /* Only assign locations for variables that lack an explicit location.
553 * Explicit locations are set for all built-in variables, generic vertex
554 * shader inputs (via layout(location=...)), and generic fragment shader
555 * outputs (also via layout(location=...)).
557 if (!var
->data
.explicit_location
) {
558 var
->data
.location
= -1;
559 var
->data
.location_frac
= 0;
562 /* ir_variable::is_unmatched_generic_inout is used by the linker while
563 * connecting outputs from one stage to inputs of the next stage.
565 if (var
->data
.explicit_location
&&
566 var
->data
.location
< VARYING_SLOT_VAR0
) {
567 var
->data
.is_unmatched_generic_inout
= 0;
569 var
->data
.is_unmatched_generic_inout
= 1;
576 * Set clip_distance_array_size based and cull_distance_array_size on the given
579 * Also check for errors based on incorrect usage of gl_ClipVertex and
580 * gl_ClipDistance and gl_CullDistance.
581 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
582 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
584 * Return false if an error was reported.
587 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
588 struct gl_linked_shader
*shader
,
589 struct gl_context
*ctx
,
590 GLuint
*clip_distance_array_size
,
591 GLuint
*cull_distance_array_size
)
593 *clip_distance_array_size
= 0;
594 *cull_distance_array_size
= 0;
596 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
597 /* From section 7.1 (Vertex Shader Special Variables) of the
600 * "It is an error for a shader to statically write both
601 * gl_ClipVertex and gl_ClipDistance."
603 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
604 * gl_ClipVertex nor gl_ClipDistance. However with
605 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
607 find_variable
gl_ClipDistance("gl_ClipDistance");
608 find_variable
gl_CullDistance("gl_CullDistance");
609 find_variable
gl_ClipVertex("gl_ClipVertex");
610 find_variable
* const variables
[] = {
613 !prog
->IsES
? &gl_ClipVertex
: NULL
,
616 find_assignments(shader
->ir
, variables
);
618 /* From the ARB_cull_distance spec:
620 * It is a compile-time or link-time error for the set of shaders forming
621 * a program to statically read or write both gl_ClipVertex and either
622 * gl_ClipDistance or gl_CullDistance.
624 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
628 if (gl_ClipVertex
.found
&& gl_ClipDistance
.found
) {
629 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
630 "and `gl_ClipDistance'\n",
631 _mesa_shader_stage_to_string(shader
->Stage
));
634 if (gl_ClipVertex
.found
&& gl_CullDistance
.found
) {
635 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
636 "and `gl_CullDistance'\n",
637 _mesa_shader_stage_to_string(shader
->Stage
));
642 if (gl_ClipDistance
.found
) {
643 ir_variable
*clip_distance_var
=
644 shader
->symbols
->get_variable("gl_ClipDistance");
645 assert(clip_distance_var
);
646 *clip_distance_array_size
= clip_distance_var
->type
->length
;
648 if (gl_CullDistance
.found
) {
649 ir_variable
*cull_distance_var
=
650 shader
->symbols
->get_variable("gl_CullDistance");
651 assert(cull_distance_var
);
652 *cull_distance_array_size
= cull_distance_var
->type
->length
;
654 /* From the ARB_cull_distance spec:
656 * It is a compile-time or link-time error for the set of shaders forming
657 * a program to have the sum of the sizes of the gl_ClipDistance and
658 * gl_CullDistance arrays to be larger than
659 * gl_MaxCombinedClipAndCullDistances.
661 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
662 ctx
->Const
.MaxClipPlanes
) {
663 linker_error(prog
, "%s shader: the combined size of "
664 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
666 "gl_MaxCombinedClipAndCullDistances (%u)",
667 _mesa_shader_stage_to_string(shader
->Stage
),
668 ctx
->Const
.MaxClipPlanes
);
675 * Verify that a vertex shader executable meets all semantic requirements.
677 * Also sets info.clip_distance_array_size and
678 * info.cull_distance_array_size as a side effect.
680 * \param shader Vertex shader executable to be verified
683 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
684 struct gl_linked_shader
*shader
,
685 struct gl_context
*ctx
)
690 /* From the GLSL 1.10 spec, page 48:
692 * "The 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. [...] The
696 * variable gl_Position is available only in the vertex
697 * language and is intended for writing the homogeneous vertex
698 * position. All executions of a well-formed vertex shader
699 * executable must write a value into this variable."
701 * while in GLSL 1.40 this text is changed to:
703 * "The variable gl_Position is available only in the vertex
704 * language and is intended for writing the homogeneous vertex
705 * position. It can be written at any time during shader
706 * execution. It may also be read back by a vertex shader
707 * after being written. This value will be used by primitive
708 * assembly, clipping, culling, and other fixed functionality
709 * operations, if present, that operate on primitives after
710 * vertex processing has occurred. Its value is undefined if
711 * the vertex shader executable does not write gl_Position."
713 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
714 * gl_Position is not an error.
716 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
717 find_variable
gl_Position("gl_Position");
718 find_assignments(shader
->ir
, &gl_Position
);
719 if (!gl_Position
.found
) {
722 "vertex shader does not write to `gl_Position'. "
723 "Its value is undefined. \n");
726 "vertex shader does not write to `gl_Position'. \n");
732 analyze_clip_cull_usage(prog
, shader
, ctx
,
733 &shader
->Program
->info
.clip_distance_array_size
,
734 &shader
->Program
->info
.cull_distance_array_size
);
738 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
739 struct gl_linked_shader
*shader
,
740 struct gl_context
*ctx
)
745 analyze_clip_cull_usage(prog
, shader
, ctx
,
746 &shader
->Program
->info
.clip_distance_array_size
,
747 &shader
->Program
->info
.cull_distance_array_size
);
752 * Verify that a fragment shader executable meets all semantic requirements
754 * \param shader Fragment shader executable to be verified
757 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
758 struct gl_linked_shader
*shader
)
763 find_variable
gl_FragColor("gl_FragColor");
764 find_variable
gl_FragData("gl_FragData");
765 find_variable
* const variables
[] = { &gl_FragColor
, &gl_FragData
, NULL
};
766 find_assignments(shader
->ir
, variables
);
768 if (gl_FragColor
.found
&& gl_FragData
.found
) {
769 linker_error(prog
, "fragment shader writes to both "
770 "`gl_FragColor' and `gl_FragData'\n");
775 * Verify that a geometry shader executable meets all semantic requirements
777 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
778 * info.cull_distance_array_size as a side effect.
780 * \param shader Geometry shader executable to be verified
783 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
784 struct gl_linked_shader
*shader
,
785 struct gl_context
*ctx
)
790 unsigned num_vertices
=
791 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
792 prog
->Geom
.VerticesIn
= num_vertices
;
794 analyze_clip_cull_usage(prog
, shader
, ctx
,
795 &shader
->Program
->info
.clip_distance_array_size
,
796 &shader
->Program
->info
.cull_distance_array_size
);
800 * Check if geometry shaders emit to non-zero streams and do corresponding
804 validate_geometry_shader_emissions(struct gl_context
*ctx
,
805 struct gl_shader_program
*prog
)
807 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
810 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
811 emit_vertex
.run(sh
->ir
);
812 if (emit_vertex
.error()) {
813 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
814 "stream parameter are in the range [0, %d].\n",
815 emit_vertex
.error_func(),
816 emit_vertex
.error_stream(),
817 ctx
->Const
.MaxVertexStreams
- 1);
819 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
820 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
822 /* From the ARB_gpu_shader5 spec:
824 * "Multiple vertex streams are supported only if the output primitive
825 * type is declared to be "points". A program will fail to link if it
826 * contains a geometry shader calling EmitStreamVertex() or
827 * EndStreamPrimitive() if its output primitive type is not "points".
829 * However, in the same spec:
831 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
832 * with <stream> set to zero."
836 * "The function EndPrimitive() is equivalent to calling
837 * EndStreamPrimitive() with <stream> set to zero."
839 * Since we can call EmitVertex() and EndPrimitive() when we output
840 * primitives other than points, calling EmitStreamVertex(0) or
841 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
842 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
843 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
846 if (prog
->Geom
.UsesStreams
&&
847 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
848 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
849 "with n>0 requires point output\n");
855 validate_intrastage_arrays(struct gl_shader_program
*prog
,
856 ir_variable
*const var
,
857 ir_variable
*const existing
)
859 /* Consider the types to be "the same" if both types are arrays
860 * of the same type and one of the arrays is implicitly sized.
861 * In addition, set the type of the linked variable to the
862 * explicitly sized array.
864 if (var
->type
->is_array() && existing
->type
->is_array()) {
865 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
866 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
867 if (var
->type
->length
!= 0) {
868 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
869 linker_error(prog
, "%s `%s' declared as type "
870 "`%s' but outermost dimension has an index"
873 var
->name
, var
->type
->name
,
874 existing
->data
.max_array_access
);
876 existing
->type
= var
->type
;
878 } else if (existing
->type
->length
!= 0) {
879 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
880 !existing
->data
.from_ssbo_unsized_array
) {
881 linker_error(prog
, "%s `%s' declared as type "
882 "`%s' but outermost dimension has an index"
885 var
->name
, existing
->type
->name
,
886 var
->data
.max_array_access
);
897 * Perform validation of global variables used across multiple shaders
900 cross_validate_globals(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
901 struct exec_list
*ir
, glsl_symbol_table
*variables
,
904 foreach_in_list(ir_instruction
, node
, ir
) {
905 ir_variable
*const var
= node
->as_variable();
910 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
913 /* don't cross validate subroutine uniforms */
914 if (var
->type
->contains_subroutine())
917 /* Don't cross validate interface instances. These are only relevant
918 * inside a shader. The cross validation is done at the Interface Block
921 if (var
->is_interface_instance())
924 /* Don't cross validate temporaries that are at global scope. These
925 * will eventually get pulled into the shaders 'main'.
927 if (var
->data
.mode
== ir_var_temporary
)
930 /* If a global with this name has already been seen, verify that the
931 * new instance has the same type. In addition, if the globals have
932 * initializers, the values of the initializers must be the same.
934 ir_variable
*const existing
= variables
->get_variable(var
->name
);
935 if (existing
!= NULL
) {
936 /* Check if types match. */
937 if (var
->type
!= existing
->type
) {
938 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
939 /* If it is an unsized array in a Shader Storage Block,
940 * two different shaders can access to different elements.
941 * Because of that, they might be converted to different
942 * sized arrays, then check that they are compatible but
943 * ignore the array size.
945 if (!(var
->data
.mode
== ir_var_shader_storage
&&
946 var
->data
.from_ssbo_unsized_array
&&
947 existing
->data
.mode
== ir_var_shader_storage
&&
948 existing
->data
.from_ssbo_unsized_array
&&
949 var
->type
->gl_type
== existing
->type
->gl_type
)) {
950 linker_error(prog
, "%s `%s' declared as type "
951 "`%s' and type `%s'\n",
953 var
->name
, var
->type
->name
,
954 existing
->type
->name
);
960 if (var
->data
.explicit_location
) {
961 if (existing
->data
.explicit_location
962 && (var
->data
.location
!= existing
->data
.location
)) {
963 linker_error(prog
, "explicit locations for %s "
964 "`%s' have differing values\n",
965 mode_string(var
), var
->name
);
969 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
970 linker_error(prog
, "explicit components for %s `%s' have "
971 "differing values\n", mode_string(var
), var
->name
);
975 existing
->data
.location
= var
->data
.location
;
976 existing
->data
.explicit_location
= true;
978 /* Check if uniform with implicit location was marked explicit
979 * by earlier shader stage. If so, mark it explicit in this stage
980 * too to make sure later processing does not treat it as
983 if (existing
->data
.explicit_location
) {
984 var
->data
.location
= existing
->data
.location
;
985 var
->data
.explicit_location
= true;
989 /* From the GLSL 4.20 specification:
990 * "A link error will result if two compilation units in a program
991 * specify different integer-constant bindings for the same
992 * opaque-uniform name. However, it is not an error to specify a
993 * binding on some but not all declarations for the same name"
995 if (var
->data
.explicit_binding
) {
996 if (existing
->data
.explicit_binding
&&
997 var
->data
.binding
!= existing
->data
.binding
) {
998 linker_error(prog
, "explicit bindings for %s "
999 "`%s' have differing values\n",
1000 mode_string(var
), var
->name
);
1004 existing
->data
.binding
= var
->data
.binding
;
1005 existing
->data
.explicit_binding
= true;
1008 if (var
->type
->contains_atomic() &&
1009 var
->data
.offset
!= existing
->data
.offset
) {
1010 linker_error(prog
, "offset specifications for %s "
1011 "`%s' have differing values\n",
1012 mode_string(var
), var
->name
);
1016 /* Validate layout qualifiers for gl_FragDepth.
1018 * From the AMD/ARB_conservative_depth specs:
1020 * "If gl_FragDepth is redeclared in any fragment shader in a
1021 * program, it must be redeclared in all fragment shaders in
1022 * that program that have static assignments to
1023 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1024 * fragment shaders in a single program must have the same set
1027 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1028 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1029 bool layout_differs
=
1030 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1032 if (layout_declared
&& layout_differs
) {
1034 "All redeclarations of gl_FragDepth in all "
1035 "fragment shaders in a single program must have "
1036 "the same set of qualifiers.\n");
1039 if (var
->data
.used
&& layout_differs
) {
1041 "If gl_FragDepth is redeclared with a layout "
1042 "qualifier in any fragment shader, it must be "
1043 "redeclared with the same layout qualifier in "
1044 "all fragment shaders that have assignments to "
1049 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1051 * "If a shared global has multiple initializers, the
1052 * initializers must all be constant expressions, and they
1053 * must all have the same value. Otherwise, a link error will
1054 * result. (A shared global having only one initializer does
1055 * not require that initializer to be a constant expression.)"
1057 * Previous to 4.20 the GLSL spec simply said that initializers
1058 * must have the same value. In this case of non-constant
1059 * initializers, this was impossible to determine. As a result,
1060 * no vendor actually implemented that behavior. The 4.20
1061 * behavior matches the implemented behavior of at least one other
1062 * vendor, so we'll implement that for all GLSL versions.
1064 if (var
->constant_initializer
!= NULL
) {
1065 if (existing
->constant_initializer
!= NULL
) {
1066 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1067 linker_error(prog
, "initializers for %s "
1068 "`%s' have differing values\n",
1069 mode_string(var
), var
->name
);
1073 /* If the first-seen instance of a particular uniform did
1074 * not have an initializer but a later instance does,
1075 * replace the former with the later.
1077 variables
->replace_variable(existing
->name
, var
);
1081 if (var
->data
.has_initializer
) {
1082 if (existing
->data
.has_initializer
1083 && (var
->constant_initializer
== NULL
1084 || existing
->constant_initializer
== NULL
)) {
1086 "shared global variable `%s' has multiple "
1087 "non-constant initializers.\n",
1093 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1094 linker_error(prog
, "declarations for %s `%s' have "
1095 "mismatching invariant qualifiers\n",
1096 mode_string(var
), var
->name
);
1099 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1100 linker_error(prog
, "declarations for %s `%s' have "
1101 "mismatching centroid qualifiers\n",
1102 mode_string(var
), var
->name
);
1105 if (existing
->data
.sample
!= var
->data
.sample
) {
1106 linker_error(prog
, "declarations for %s `%s` have "
1107 "mismatching sample qualifiers\n",
1108 mode_string(var
), var
->name
);
1111 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1112 linker_error(prog
, "declarations for %s `%s` have "
1113 "mismatching image format qualifiers\n",
1114 mode_string(var
), var
->name
);
1118 /* Check the precision qualifier matches for uniform variables on
1121 if (!ctx
->Const
.AllowGLSLRelaxedES
&&
1122 prog
->IsES
&& !var
->get_interface_type() &&
1123 existing
->data
.precision
!= var
->data
.precision
) {
1124 if ((existing
->data
.used
&& var
->data
.used
) || prog
->data
->Version
>= 300) {
1125 linker_error(prog
, "declarations for %s `%s` have "
1126 "mismatching precision qualifiers\n",
1127 mode_string(var
), var
->name
);
1130 linker_warning(prog
, "declarations for %s `%s` have "
1131 "mismatching precision qualifiers\n",
1132 mode_string(var
), var
->name
);
1136 /* In OpenGL GLSL 3.20 spec, section 4.3.9:
1138 * "It is a link-time error if any particular shader interface
1141 * - two different blocks, each having no instance name, and each
1142 * having a member of the same name, or
1144 * - a variable outside a block, and a block with no instance name,
1145 * where the variable has the same name as a member in the block."
1147 const glsl_type
*var_itype
= var
->get_interface_type();
1148 const glsl_type
*existing_itype
= existing
->get_interface_type();
1149 if (var_itype
!= existing_itype
) {
1150 if (!var_itype
|| !existing_itype
) {
1151 linker_error(prog
, "declarations for %s `%s` are inside block "
1152 "`%s` and outside a block",
1153 mode_string(var
), var
->name
,
1154 var_itype
? var_itype
->name
: existing_itype
->name
);
1156 } else if (strcmp(var_itype
->name
, existing_itype
->name
) != 0) {
1157 linker_error(prog
, "declarations for %s `%s` are inside blocks "
1159 mode_string(var
), var
->name
,
1160 existing_itype
->name
,
1166 variables
->add_variable(var
);
1172 * Perform validation of uniforms used across multiple shader stages
1175 cross_validate_uniforms(struct gl_context
*ctx
,
1176 struct gl_shader_program
*prog
)
1178 glsl_symbol_table variables
;
1179 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1180 if (prog
->_LinkedShaders
[i
] == NULL
)
1183 cross_validate_globals(ctx
, prog
, prog
->_LinkedShaders
[i
]->ir
,
1189 * Accumulates the array of buffer blocks and checks that all definitions of
1190 * blocks agree on their contents.
1193 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1196 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1197 struct gl_uniform_block
*blks
= NULL
;
1198 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1199 &prog
->data
->NumUniformBlocks
;
1201 unsigned max_num_buffer_blocks
= 0;
1202 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1203 if (prog
->_LinkedShaders
[i
]) {
1204 if (validate_ssbo
) {
1205 max_num_buffer_blocks
+=
1206 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1208 max_num_buffer_blocks
+=
1209 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1214 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1215 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1217 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1218 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1219 InterfaceBlockStageIndex
[i
][j
] = -1;
1224 unsigned sh_num_blocks
;
1225 struct gl_uniform_block
**sh_blks
;
1226 if (validate_ssbo
) {
1227 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1228 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1230 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1231 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1234 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1235 int index
= link_cross_validate_uniform_block(prog
->data
, &blks
,
1236 num_blks
, sh_blks
[j
]);
1239 linker_error(prog
, "buffer block `%s' has mismatching "
1240 "definitions\n", sh_blks
[j
]->Name
);
1242 for (unsigned k
= 0; k
<= i
; k
++) {
1243 delete[] InterfaceBlockStageIndex
[k
];
1246 /* Reset the block count. This will help avoid various segfaults
1247 * from api calls that assume the array exists due to the count
1254 InterfaceBlockStageIndex
[i
][index
] = j
;
1258 /* Update per stage block pointers to point to the program list.
1259 * FIXME: We should be able to free the per stage blocks here.
1261 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1262 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1263 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1265 if (stage_index
!= -1) {
1266 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1268 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1269 sh
->Program
->sh
.ShaderStorageBlocks
:
1270 sh
->Program
->sh
.UniformBlocks
;
1272 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1273 sh_blks
[stage_index
] = &blks
[j
];
1278 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1279 delete[] InterfaceBlockStageIndex
[i
];
1283 prog
->data
->ShaderStorageBlocks
= blks
;
1285 prog
->data
->UniformBlocks
= blks
;
1291 * Verifies the invariance of built-in special variables.
1294 validate_invariant_builtins(struct gl_shader_program
*prog
,
1295 const gl_linked_shader
*vert
,
1296 const gl_linked_shader
*frag
)
1298 const ir_variable
*var_vert
;
1299 const ir_variable
*var_frag
;
1305 * From OpenGL ES Shading Language 1.0 specification
1306 * (4.6.4 Invariance and Linkage):
1307 * "The invariance of varyings that are declared in both the vertex and
1308 * fragment shaders must match. For the built-in special variables,
1309 * gl_FragCoord can only be declared invariant if and only if
1310 * gl_Position is declared invariant. Similarly gl_PointCoord can only
1311 * be declared invariant if and only if gl_PointSize is declared
1312 * invariant. It is an error to declare gl_FrontFacing as invariant.
1313 * The invariance of gl_FrontFacing is the same as the invariance of
1316 var_frag
= frag
->symbols
->get_variable("gl_FragCoord");
1317 if (var_frag
&& var_frag
->data
.invariant
) {
1318 var_vert
= vert
->symbols
->get_variable("gl_Position");
1319 if (var_vert
&& !var_vert
->data
.invariant
) {
1321 "fragment shader built-in `%s' has invariant qualifier, "
1322 "but vertex shader built-in `%s' lacks invariant qualifier\n",
1323 var_frag
->name
, var_vert
->name
);
1328 var_frag
= frag
->symbols
->get_variable("gl_PointCoord");
1329 if (var_frag
&& var_frag
->data
.invariant
) {
1330 var_vert
= vert
->symbols
->get_variable("gl_PointSize");
1331 if (var_vert
&& !var_vert
->data
.invariant
) {
1333 "fragment shader built-in `%s' has invariant qualifier, "
1334 "but vertex shader built-in `%s' lacks invariant qualifier\n",
1335 var_frag
->name
, var_vert
->name
);
1340 var_frag
= frag
->symbols
->get_variable("gl_FrontFacing");
1341 if (var_frag
&& var_frag
->data
.invariant
) {
1343 "fragment shader built-in `%s' can not be declared as invariant\n",
1352 * Populates a shaders symbol table with all global declarations
1355 populate_symbol_table(gl_linked_shader
*sh
, glsl_symbol_table
*symbols
)
1357 sh
->symbols
= new(sh
) glsl_symbol_table
;
1359 _mesa_glsl_copy_symbols_from_table(sh
->ir
, symbols
, sh
->symbols
);
1364 * Remap variables referenced in an instruction tree
1366 * This is used when instruction trees are cloned from one shader and placed in
1367 * another. These trees will contain references to \c ir_variable nodes that
1368 * do not exist in the target shader. This function finds these \c ir_variable
1369 * references and replaces the references with matching variables in the target
1372 * If there is no matching variable in the target shader, a clone of the
1373 * \c ir_variable is made and added to the target shader. The new variable is
1374 * added to \b both the instruction stream and the symbol table.
1376 * \param inst IR tree that is to be processed.
1377 * \param symbols Symbol table containing global scope symbols in the
1379 * \param instructions Instruction stream where new variable declarations
1383 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1386 class remap_visitor
: public ir_hierarchical_visitor
{
1388 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1390 this->target
= target
;
1391 this->symbols
= target
->symbols
;
1392 this->instructions
= target
->ir
;
1393 this->temps
= temps
;
1396 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1398 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1399 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1400 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1402 assert(var
!= NULL
);
1404 return visit_continue
;
1407 ir_variable
*const existing
=
1408 this->symbols
->get_variable(ir
->var
->name
);
1409 if (existing
!= NULL
)
1412 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1414 this->symbols
->add_variable(copy
);
1415 this->instructions
->push_head(copy
);
1419 return visit_continue
;
1423 struct gl_linked_shader
*target
;
1424 glsl_symbol_table
*symbols
;
1425 exec_list
*instructions
;
1429 remap_visitor
v(target
, temps
);
1436 * Move non-declarations from one instruction stream to another
1438 * The intended usage pattern of this function is to pass the pointer to the
1439 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1440 * pointer) for \c last and \c false for \c make_copies on the first
1441 * call. Successive calls pass the return value of the previous call for
1442 * \c last and \c true for \c make_copies.
1444 * \param instructions Source instruction stream
1445 * \param last Instruction after which new instructions should be
1446 * inserted in the target instruction stream
1447 * \param make_copies Flag selecting whether instructions in \c instructions
1448 * should be copied (via \c ir_instruction::clone) into the
1449 * target list or moved.
1452 * The new "last" instruction in the target instruction stream. This pointer
1453 * is suitable for use as the \c last parameter of a later call to this
1457 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1458 bool make_copies
, gl_linked_shader
*target
)
1460 hash_table
*temps
= NULL
;
1463 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1464 _mesa_key_pointer_equal
);
1466 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1467 if (inst
->as_function())
1470 ir_variable
*var
= inst
->as_variable();
1471 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1474 assert(inst
->as_assignment()
1476 || inst
->as_if() /* for initializers with the ?: operator */
1477 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1480 inst
= inst
->clone(target
, NULL
);
1483 _mesa_hash_table_insert(temps
, var
, inst
);
1485 remap_variables(inst
, target
, temps
);
1490 last
->insert_after(inst
);
1495 _mesa_hash_table_destroy(temps
, NULL
);
1502 * This class is only used in link_intrastage_shaders() below but declaring
1503 * it inside that function leads to compiler warnings with some versions of
1506 class array_sizing_visitor
: public deref_type_updater
{
1508 array_sizing_visitor()
1509 : mem_ctx(ralloc_context(NULL
)),
1510 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1511 _mesa_key_pointer_equal
))
1515 ~array_sizing_visitor()
1517 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1518 ralloc_free(this->mem_ctx
);
1521 virtual ir_visitor_status
visit(ir_variable
*var
)
1523 const glsl_type
*type_without_array
;
1524 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1525 fixup_type(&var
->type
, var
->data
.max_array_access
,
1526 var
->data
.from_ssbo_unsized_array
,
1527 &implicit_sized_array
);
1528 var
->data
.implicit_sized_array
= implicit_sized_array
;
1529 type_without_array
= var
->type
->without_array();
1530 if (var
->type
->is_interface()) {
1531 if (interface_contains_unsized_arrays(var
->type
)) {
1532 const glsl_type
*new_type
=
1533 resize_interface_members(var
->type
,
1534 var
->get_max_ifc_array_access(),
1535 var
->is_in_shader_storage_block());
1536 var
->type
= new_type
;
1537 var
->change_interface_type(new_type
);
1539 } else if (type_without_array
->is_interface()) {
1540 if (interface_contains_unsized_arrays(type_without_array
)) {
1541 const glsl_type
*new_type
=
1542 resize_interface_members(type_without_array
,
1543 var
->get_max_ifc_array_access(),
1544 var
->is_in_shader_storage_block());
1545 var
->change_interface_type(new_type
);
1546 var
->type
= update_interface_members_array(var
->type
, new_type
);
1548 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1549 /* Store a pointer to the variable in the unnamed_interfaces
1553 _mesa_hash_table_search(this->unnamed_interfaces
,
1556 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1558 if (interface_vars
== NULL
) {
1559 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1561 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1564 unsigned index
= ifc_type
->field_index(var
->name
);
1565 assert(index
< ifc_type
->length
);
1566 assert(interface_vars
[index
] == NULL
);
1567 interface_vars
[index
] = var
;
1569 return visit_continue
;
1573 * For each unnamed interface block that was discovered while running the
1574 * visitor, adjust the interface type to reflect the newly assigned array
1575 * sizes, and fix up the ir_variable nodes to point to the new interface
1578 void fixup_unnamed_interface_types()
1580 hash_table_call_foreach(this->unnamed_interfaces
,
1581 fixup_unnamed_interface_type
, NULL
);
1586 * If the type pointed to by \c type represents an unsized array, replace
1587 * it with a sized array whose size is determined by max_array_access.
1589 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1590 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1592 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1593 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1594 max_array_access
+ 1);
1595 *implicit_sized
= true;
1596 assert(*type
!= NULL
);
1600 static const glsl_type
*
1601 update_interface_members_array(const glsl_type
*type
,
1602 const glsl_type
*new_interface_type
)
1604 const glsl_type
*element_type
= type
->fields
.array
;
1605 if (element_type
->is_array()) {
1606 const glsl_type
*new_array_type
=
1607 update_interface_members_array(element_type
, new_interface_type
);
1608 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1610 return glsl_type::get_array_instance(new_interface_type
,
1616 * Determine whether the given interface type contains unsized arrays (if
1617 * it doesn't, array_sizing_visitor doesn't need to process it).
1619 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1621 for (unsigned i
= 0; i
< type
->length
; i
++) {
1622 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1623 if (elem_type
->is_unsized_array())
1630 * Create a new interface type based on the given type, with unsized arrays
1631 * replaced by sized arrays whose size is determined by
1632 * max_ifc_array_access.
1634 static const glsl_type
*
1635 resize_interface_members(const glsl_type
*type
,
1636 const int *max_ifc_array_access
,
1639 unsigned num_fields
= type
->length
;
1640 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1641 memcpy(fields
, type
->fields
.structure
,
1642 num_fields
* sizeof(*fields
));
1643 for (unsigned i
= 0; i
< num_fields
; i
++) {
1644 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1645 /* If SSBO last member is unsized array, we don't replace it by a sized
1648 if (is_ssbo
&& i
== (num_fields
- 1))
1649 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1650 true, &implicit_sized_array
);
1652 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1653 false, &implicit_sized_array
);
1654 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1656 glsl_interface_packing packing
=
1657 (glsl_interface_packing
) type
->interface_packing
;
1658 bool row_major
= (bool) type
->interface_row_major
;
1659 const glsl_type
*new_ifc_type
=
1660 glsl_type::get_interface_instance(fields
, num_fields
,
1661 packing
, row_major
, type
->name
);
1663 return new_ifc_type
;
1666 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1669 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1670 ir_variable
**interface_vars
= (ir_variable
**) data
;
1671 unsigned num_fields
= ifc_type
->length
;
1672 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1673 memcpy(fields
, ifc_type
->fields
.structure
,
1674 num_fields
* sizeof(*fields
));
1675 bool interface_type_changed
= false;
1676 for (unsigned i
= 0; i
< num_fields
; i
++) {
1677 if (interface_vars
[i
] != NULL
&&
1678 fields
[i
].type
!= interface_vars
[i
]->type
) {
1679 fields
[i
].type
= interface_vars
[i
]->type
;
1680 interface_type_changed
= true;
1683 if (!interface_type_changed
) {
1687 glsl_interface_packing packing
=
1688 (glsl_interface_packing
) ifc_type
->interface_packing
;
1689 bool row_major
= (bool) ifc_type
->interface_row_major
;
1690 const glsl_type
*new_ifc_type
=
1691 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1692 row_major
, ifc_type
->name
);
1694 for (unsigned i
= 0; i
< num_fields
; i
++) {
1695 if (interface_vars
[i
] != NULL
)
1696 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1701 * Memory context used to allocate the data in \c unnamed_interfaces.
1706 * Hash table from const glsl_type * to an array of ir_variable *'s
1707 * pointing to the ir_variables constituting each unnamed interface block.
1709 hash_table
*unnamed_interfaces
;
1713 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1714 struct gl_shader_program
*prog
)
1716 /* We will validate doubles at a later stage */
1717 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1718 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1719 "multiple of 4 or if its applied to a type that is "
1720 "or contains a double a multiple of 8.",
1721 prog
->TransformFeedback
.BufferStride
[idx
]);
1725 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1726 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1727 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1728 "limit has been exceeded.");
1736 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1740 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1741 struct gl_shader_program
*prog
,
1742 struct gl_shader
**shader_list
,
1743 unsigned num_shaders
)
1745 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1746 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1749 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1750 struct gl_shader
*shader
= shader_list
[i
];
1752 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1753 if (shader
->TransformFeedbackBufferStride
[j
]) {
1754 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1755 prog
->TransformFeedback
.BufferStride
[j
] =
1756 shader
->TransformFeedbackBufferStride
[j
];
1757 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1759 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1760 shader
->TransformFeedbackBufferStride
[j
]){
1762 "intrastage shaders defined with conflicting "
1763 "xfb_stride for buffer %d (%d and %d)\n", j
,
1764 prog
->TransformFeedback
.BufferStride
[j
],
1765 shader
->TransformFeedbackBufferStride
[j
]);
1774 * Check for conflicting bindless/bound sampler/image layout qualifiers at
1778 link_bindless_layout_qualifiers(struct gl_shader_program
*prog
,
1779 struct gl_shader
**shader_list
,
1780 unsigned num_shaders
)
1782 bool bindless_sampler
, bindless_image
;
1783 bool bound_sampler
, bound_image
;
1785 bindless_sampler
= bindless_image
= false;
1786 bound_sampler
= bound_image
= false;
1788 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1789 struct gl_shader
*shader
= shader_list
[i
];
1791 if (shader
->bindless_sampler
)
1792 bindless_sampler
= true;
1793 if (shader
->bindless_image
)
1794 bindless_image
= true;
1795 if (shader
->bound_sampler
)
1796 bound_sampler
= true;
1797 if (shader
->bound_image
)
1800 if ((bindless_sampler
&& bound_sampler
) ||
1801 (bindless_image
&& bound_image
)) {
1802 /* From section 4.4.6 of the ARB_bindless_texture spec:
1804 * "If both bindless_sampler and bound_sampler, or bindless_image
1805 * and bound_image, are declared at global scope in any
1806 * compilation unit, a link- time error will be generated."
1808 linker_error(prog
, "both bindless_sampler and bound_sampler, or "
1809 "bindless_image and bound_image, can't be declared at "
1816 * Performs the cross-validation of tessellation control shader vertices and
1817 * layout qualifiers for the attached tessellation control shaders,
1818 * and propagates them to the linked TCS and linked shader program.
1821 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1822 struct gl_program
*gl_prog
,
1823 struct gl_shader
**shader_list
,
1824 unsigned num_shaders
)
1826 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1829 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1831 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1833 * "All tessellation control shader layout declarations in a program
1834 * must specify the same output patch vertex count. There must be at
1835 * least one layout qualifier specifying an output patch vertex count
1836 * in any program containing tessellation control shaders; however,
1837 * such a declaration is not required in all tessellation control
1841 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1842 struct gl_shader
*shader
= shader_list
[i
];
1844 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1845 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1846 gl_prog
->info
.tess
.tcs_vertices_out
!=
1847 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1848 linker_error(prog
, "tessellation control shader defined with "
1849 "conflicting output vertex count (%d and %d)\n",
1850 gl_prog
->info
.tess
.tcs_vertices_out
,
1851 shader
->info
.TessCtrl
.VerticesOut
);
1854 gl_prog
->info
.tess
.tcs_vertices_out
=
1855 shader
->info
.TessCtrl
.VerticesOut
;
1859 /* Just do the intrastage -> interstage propagation right now,
1860 * since we already know we're in the right type of shader program
1863 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1864 linker_error(prog
, "tessellation control shader didn't declare "
1865 "vertices out layout qualifier\n");
1872 * Performs the cross-validation of tessellation evaluation shader
1873 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1874 * for the attached tessellation evaluation shaders, and propagates them
1875 * to the linked TES and linked shader program.
1878 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1879 struct gl_program
*gl_prog
,
1880 struct gl_shader
**shader_list
,
1881 unsigned num_shaders
)
1883 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1886 int point_mode
= -1;
1887 unsigned vertex_order
= 0;
1889 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1890 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1892 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1894 * "At least one tessellation evaluation shader (compilation unit) in
1895 * a program must declare a primitive mode in its input layout.
1896 * Declaration vertex spacing, ordering, and point mode identifiers is
1897 * optional. It is not required that all tessellation evaluation
1898 * shaders in a program declare a primitive mode. If spacing or
1899 * vertex ordering declarations are omitted, the tessellation
1900 * primitive generator will use equal spacing or counter-clockwise
1901 * vertex ordering, respectively. If a point mode declaration is
1902 * omitted, the tessellation primitive generator will produce lines or
1903 * triangles according to the primitive mode."
1906 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1907 struct gl_shader
*shader
= shader_list
[i
];
1909 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1910 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1911 gl_prog
->info
.tess
.primitive_mode
!=
1912 shader
->info
.TessEval
.PrimitiveMode
) {
1913 linker_error(prog
, "tessellation evaluation shader defined with "
1914 "conflicting input primitive modes.\n");
1917 gl_prog
->info
.tess
.primitive_mode
=
1918 shader
->info
.TessEval
.PrimitiveMode
;
1921 if (shader
->info
.TessEval
.Spacing
!= 0) {
1922 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1923 shader
->info
.TessEval
.Spacing
) {
1924 linker_error(prog
, "tessellation evaluation shader defined with "
1925 "conflicting vertex spacing.\n");
1928 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1931 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1932 if (vertex_order
!= 0 &&
1933 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1934 linker_error(prog
, "tessellation evaluation shader defined with "
1935 "conflicting ordering.\n");
1938 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1941 if (shader
->info
.TessEval
.PointMode
!= -1) {
1942 if (point_mode
!= -1 &&
1943 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1944 linker_error(prog
, "tessellation evaluation shader defined with "
1945 "conflicting point modes.\n");
1948 point_mode
= shader
->info
.TessEval
.PointMode
;
1953 /* Just do the intrastage -> interstage propagation right now,
1954 * since we already know we're in the right type of shader program
1957 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1959 "tessellation evaluation shader didn't declare input "
1960 "primitive modes.\n");
1964 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1965 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1967 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1968 gl_prog
->info
.tess
.ccw
= true;
1970 gl_prog
->info
.tess
.ccw
= false;
1973 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1974 gl_prog
->info
.tess
.point_mode
= false;
1976 gl_prog
->info
.tess
.point_mode
= true;
1981 * Performs the cross-validation of layout qualifiers specified in
1982 * redeclaration of gl_FragCoord for the attached fragment shaders,
1983 * and propagates them to the linked FS and linked shader program.
1986 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1987 struct gl_linked_shader
*linked_shader
,
1988 struct gl_shader
**shader_list
,
1989 unsigned num_shaders
)
1991 bool redeclares_gl_fragcoord
= false;
1992 bool uses_gl_fragcoord
= false;
1993 bool origin_upper_left
= false;
1994 bool pixel_center_integer
= false;
1996 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1997 (prog
->data
->Version
< 150 &&
1998 !prog
->ARB_fragment_coord_conventions_enable
))
2001 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2002 struct gl_shader
*shader
= shader_list
[i
];
2003 /* From the GLSL 1.50 spec, page 39:
2005 * "If gl_FragCoord is redeclared in any fragment shader in a program,
2006 * it must be redeclared in all the fragment shaders in that program
2007 * that have a static use gl_FragCoord."
2009 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
2010 shader
->uses_gl_fragcoord
)
2011 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
2012 uses_gl_fragcoord
)) {
2013 linker_error(prog
, "fragment shader defined with conflicting "
2014 "layout qualifiers for gl_FragCoord\n");
2017 /* From the GLSL 1.50 spec, page 39:
2019 * "All redeclarations of gl_FragCoord in all fragment shaders in a
2020 * single program must have the same set of qualifiers."
2022 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
2023 (shader
->origin_upper_left
!= origin_upper_left
||
2024 shader
->pixel_center_integer
!= pixel_center_integer
)) {
2025 linker_error(prog
, "fragment shader defined with conflicting "
2026 "layout qualifiers for gl_FragCoord\n");
2029 /* Update the linked shader state. Note that uses_gl_fragcoord should
2030 * accumulate the results. The other values should replace. If there
2031 * are multiple redeclarations, all the fields except uses_gl_fragcoord
2032 * are already known to be the same.
2034 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
2035 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
2036 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
2037 origin_upper_left
= shader
->origin_upper_left
;
2038 pixel_center_integer
= shader
->pixel_center_integer
;
2041 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
2042 shader
->EarlyFragmentTests
|| shader
->PostDepthCoverage
;
2043 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
2044 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
2045 shader
->PostDepthCoverage
;
2046 linked_shader
->Program
->info
.fs
.pixel_interlock_ordered
|=
2047 shader
->PixelInterlockOrdered
;
2048 linked_shader
->Program
->info
.fs
.pixel_interlock_unordered
|=
2049 shader
->PixelInterlockUnordered
;
2050 linked_shader
->Program
->info
.fs
.sample_interlock_ordered
|=
2051 shader
->SampleInterlockOrdered
;
2052 linked_shader
->Program
->info
.fs
.sample_interlock_unordered
|=
2053 shader
->SampleInterlockUnordered
;
2055 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
2060 * Performs the cross-validation of geometry shader max_vertices and
2061 * primitive type layout qualifiers for the attached geometry shaders,
2062 * and propagates them to the linked GS and linked shader program.
2065 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
2066 struct gl_program
*gl_prog
,
2067 struct gl_shader
**shader_list
,
2068 unsigned num_shaders
)
2070 /* No in/out qualifiers defined for anything but GLSL 1.50+
2071 * geometry shaders so far.
2073 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
2074 prog
->data
->Version
< 150)
2077 int vertices_out
= -1;
2079 gl_prog
->info
.gs
.invocations
= 0;
2080 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
2081 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
2083 /* From the GLSL 1.50 spec, page 46:
2085 * "All geometry shader output layout declarations in a program
2086 * must declare the same layout and same value for
2087 * max_vertices. There must be at least one geometry output
2088 * layout declaration somewhere in a program, but not all
2089 * geometry shaders (compilation units) are required to
2093 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2094 struct gl_shader
*shader
= shader_list
[i
];
2096 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
2097 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
2098 gl_prog
->info
.gs
.input_primitive
!=
2099 shader
->info
.Geom
.InputType
) {
2100 linker_error(prog
, "geometry shader defined with conflicting "
2104 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
2107 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
2108 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
2109 gl_prog
->info
.gs
.output_primitive
!=
2110 shader
->info
.Geom
.OutputType
) {
2111 linker_error(prog
, "geometry shader defined with conflicting "
2115 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
2118 if (shader
->info
.Geom
.VerticesOut
!= -1) {
2119 if (vertices_out
!= -1 &&
2120 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
2121 linker_error(prog
, "geometry shader defined with conflicting "
2122 "output vertex count (%d and %d)\n",
2123 vertices_out
, shader
->info
.Geom
.VerticesOut
);
2126 vertices_out
= shader
->info
.Geom
.VerticesOut
;
2129 if (shader
->info
.Geom
.Invocations
!= 0) {
2130 if (gl_prog
->info
.gs
.invocations
!= 0 &&
2131 gl_prog
->info
.gs
.invocations
!=
2132 (unsigned) shader
->info
.Geom
.Invocations
) {
2133 linker_error(prog
, "geometry shader defined with conflicting "
2134 "invocation count (%d and %d)\n",
2135 gl_prog
->info
.gs
.invocations
,
2136 shader
->info
.Geom
.Invocations
);
2139 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
2143 /* Just do the intrastage -> interstage propagation right now,
2144 * since we already know we're in the right type of shader program
2147 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
2149 "geometry shader didn't declare primitive input type\n");
2153 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
2155 "geometry shader didn't declare primitive output type\n");
2159 if (vertices_out
== -1) {
2161 "geometry shader didn't declare max_vertices\n");
2164 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2167 if (gl_prog
->info
.gs
.invocations
== 0)
2168 gl_prog
->info
.gs
.invocations
= 1;
2173 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2174 * qualifiers for the attached compute shaders, and propagate them to the
2175 * linked CS and linked shader program.
2178 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2179 struct gl_program
*gl_prog
,
2180 struct gl_shader
**shader_list
,
2181 unsigned num_shaders
)
2183 /* This function is called for all shader stages, but it only has an effect
2184 * for compute shaders.
2186 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2189 for (int i
= 0; i
< 3; i
++)
2190 gl_prog
->info
.cs
.local_size
[i
] = 0;
2192 gl_prog
->info
.cs
.local_size_variable
= false;
2194 /* From the ARB_compute_shader spec, in the section describing local size
2197 * If multiple compute shaders attached to a single program object
2198 * declare local work-group size, the declarations must be identical;
2199 * otherwise a link-time error results. Furthermore, if a program
2200 * object contains any compute shaders, at least one must contain an
2201 * input layout qualifier specifying the local work sizes of the
2202 * program, or a link-time error will occur.
2204 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2205 struct gl_shader
*shader
= shader_list
[sh
];
2207 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2208 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2209 for (int i
= 0; i
< 3; i
++) {
2210 if (gl_prog
->info
.cs
.local_size
[i
] !=
2211 shader
->info
.Comp
.LocalSize
[i
]) {
2212 linker_error(prog
, "compute shader defined with conflicting "
2218 for (int i
= 0; i
< 3; i
++) {
2219 gl_prog
->info
.cs
.local_size
[i
] =
2220 shader
->info
.Comp
.LocalSize
[i
];
2222 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2223 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2224 /* The ARB_compute_variable_group_size spec says:
2226 * If one compute shader attached to a program declares a
2227 * variable local group size and a second compute shader
2228 * attached to the same program declares a fixed local group
2229 * size, a link-time error results.
2231 linker_error(prog
, "compute shader defined with both fixed and "
2232 "variable local group size\n");
2235 gl_prog
->info
.cs
.local_size_variable
= true;
2239 /* Just do the intrastage -> interstage propagation right now,
2240 * since we already know we're in the right type of shader program
2243 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2244 !gl_prog
->info
.cs
.local_size_variable
) {
2245 linker_error(prog
, "compute shader must contain a fixed or a variable "
2246 "local group size\n");
2252 * Link all out variables on a single stage which are not
2253 * directly used in a shader with the main function.
2256 link_output_variables(struct gl_linked_shader
*linked_shader
,
2257 struct gl_shader
**shader_list
,
2258 unsigned num_shaders
)
2260 struct glsl_symbol_table
*symbols
= linked_shader
->symbols
;
2262 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2264 /* Skip shader object with main function */
2265 if (shader_list
[i
]->symbols
->get_function("main"))
2268 foreach_in_list(ir_instruction
, ir
, shader_list
[i
]->ir
) {
2269 if (ir
->ir_type
!= ir_type_variable
)
2272 ir_variable
*var
= (ir_variable
*) ir
;
2274 if (var
->data
.mode
== ir_var_shader_out
&&
2275 !symbols
->get_variable(var
->name
)) {
2276 var
= var
->clone(linked_shader
, NULL
);
2277 symbols
->add_variable(var
);
2278 linked_shader
->ir
->push_head(var
);
2288 * Combine a group of shaders for a single stage to generate a linked shader
2291 * If this function is supplied a single shader, it is cloned, and the new
2292 * shader is returned.
2294 struct gl_linked_shader
*
2295 link_intrastage_shaders(void *mem_ctx
,
2296 struct gl_context
*ctx
,
2297 struct gl_shader_program
*prog
,
2298 struct gl_shader
**shader_list
,
2299 unsigned num_shaders
,
2300 bool allow_missing_main
)
2302 struct gl_uniform_block
*ubo_blocks
= NULL
;
2303 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2304 unsigned num_ubo_blocks
= 0;
2305 unsigned num_ssbo_blocks
= 0;
2307 /* Check that global variables defined in multiple shaders are consistent.
2309 glsl_symbol_table variables
;
2310 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2311 if (shader_list
[i
] == NULL
)
2313 cross_validate_globals(ctx
, prog
, shader_list
[i
]->ir
, &variables
,
2317 if (!prog
->data
->LinkStatus
)
2320 /* Check that interface blocks defined in multiple shaders are consistent.
2322 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2324 if (!prog
->data
->LinkStatus
)
2327 /* Check that there is only a single definition of each function signature
2328 * across all shaders.
2330 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2331 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2332 ir_function
*const f
= node
->as_function();
2337 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2338 ir_function
*const other
=
2339 shader_list
[j
]->symbols
->get_function(f
->name
);
2341 /* If the other shader has no function (and therefore no function
2342 * signatures) with the same name, skip to the next shader.
2347 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2348 if (!sig
->is_defined
)
2351 ir_function_signature
*other_sig
=
2352 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2354 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2355 linker_error(prog
, "function `%s' is multiply defined\n",
2364 /* Find the shader that defines main, and make a clone of it.
2366 * Starting with the clone, search for undefined references. If one is
2367 * found, find the shader that defines it. Clone the reference and add
2368 * it to the shader. Repeat until there are no undefined references or
2369 * until a reference cannot be resolved.
2371 gl_shader
*main
= NULL
;
2372 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2373 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2374 main
= shader_list
[i
];
2379 if (main
== NULL
&& allow_missing_main
)
2380 main
= shader_list
[0];
2383 linker_error(prog
, "%s shader lacks `main'\n",
2384 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2388 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2389 linked
->Stage
= shader_list
[0]->Stage
;
2391 /* Create program and attach it to the linked shader */
2392 struct gl_program
*gl_prog
=
2393 ctx
->Driver
.NewProgram(ctx
,
2394 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2397 prog
->data
->LinkStatus
= LINKING_FAILURE
;
2398 _mesa_delete_linked_shader(ctx
, linked
);
2402 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2404 /* Don't use _mesa_reference_program() just take ownership */
2405 linked
->Program
= gl_prog
;
2407 linked
->ir
= new(linked
) exec_list
;
2408 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2410 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2411 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2412 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2413 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2414 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2416 if (linked
->Stage
!= MESA_SHADER_FRAGMENT
)
2417 link_xfb_stride_layout_qualifiers(ctx
, prog
, shader_list
, num_shaders
);
2419 link_bindless_layout_qualifiers(prog
, shader_list
, num_shaders
);
2421 populate_symbol_table(linked
, shader_list
[0]->symbols
);
2423 /* The pointer to the main function in the final linked shader (i.e., the
2424 * copy of the original shader that contained the main function).
2426 ir_function_signature
*const main_sig
=
2427 _mesa_get_main_function_signature(linked
->symbols
);
2429 /* Move any instructions other than variable declarations or function
2430 * declarations into main.
2432 if (main_sig
!= NULL
) {
2433 exec_node
*insertion_point
=
2434 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2437 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2438 if (shader_list
[i
] == main
)
2441 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2442 insertion_point
, true, linked
);
2446 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2447 _mesa_delete_linked_shader(ctx
, linked
);
2451 if (linked
->Stage
!= MESA_SHADER_FRAGMENT
)
2452 link_output_variables(linked
, shader_list
, num_shaders
);
2454 /* Make a pass over all variable declarations to ensure that arrays with
2455 * unspecified sizes have a size specified. The size is inferred from the
2456 * max_array_access field.
2458 array_sizing_visitor v
;
2460 v
.fixup_unnamed_interface_types();
2462 /* Link up uniform blocks defined within this stage. */
2463 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2464 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2466 if (!prog
->data
->LinkStatus
) {
2467 _mesa_delete_linked_shader(ctx
, linked
);
2471 /* Copy ubo blocks to linked shader list */
2472 linked
->Program
->sh
.UniformBlocks
=
2473 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2474 ralloc_steal(linked
, ubo_blocks
);
2475 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2476 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2478 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2480 /* Copy ssbo blocks to linked shader list */
2481 linked
->Program
->sh
.ShaderStorageBlocks
=
2482 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2483 ralloc_steal(linked
, ssbo_blocks
);
2484 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2485 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2487 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2489 /* At this point linked should contain all of the linked IR, so
2490 * validate it to make sure nothing went wrong.
2492 validate_ir_tree(linked
->ir
);
2494 /* Set the size of geometry shader input arrays */
2495 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2496 unsigned num_vertices
=
2497 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2498 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2499 MESA_SHADER_GEOMETRY
);
2500 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2501 ir
->accept(&input_resize_visitor
);
2505 if (ctx
->Const
.VertexID_is_zero_based
)
2506 lower_vertex_id(linked
);
2508 if (ctx
->Const
.LowerCsDerivedVariables
)
2509 lower_cs_derived(linked
);
2512 /* Compute the source checksum. */
2513 linked
->SourceChecksum
= 0;
2514 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2515 if (shader_list
[i
] == NULL
)
2517 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2525 * Update the sizes of linked shader uniform arrays to the maximum
2528 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2530 * If one or more elements of an array are active,
2531 * GetActiveUniform will return the name of the array in name,
2532 * subject to the restrictions listed above. The type of the array
2533 * is returned in type. The size parameter contains the highest
2534 * array element index used, plus one. The compiler or linker
2535 * determines the highest index used. There will be only one
2536 * active uniform reported by the GL per uniform array.
2540 update_array_sizes(struct gl_shader_program
*prog
)
2542 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2543 if (prog
->_LinkedShaders
[i
] == NULL
)
2546 bool types_were_updated
= false;
2548 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2549 ir_variable
*const var
= node
->as_variable();
2551 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2552 !var
->type
->is_array())
2555 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2556 * will not be eliminated. Since we always do std140, just
2557 * don't resize arrays in UBOs.
2559 * Atomic counters are supposed to get deterministic
2560 * locations assigned based on the declaration ordering and
2561 * sizes, array compaction would mess that up.
2563 * Subroutine uniforms are not removed.
2565 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2566 var
->type
->contains_subroutine() || var
->constant_initializer
)
2569 int size
= var
->data
.max_array_access
;
2570 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2571 if (prog
->_LinkedShaders
[j
] == NULL
)
2574 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2575 ir_variable
*other_var
= node2
->as_variable();
2579 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2580 other_var
->data
.max_array_access
> size
) {
2581 size
= other_var
->data
.max_array_access
;
2586 if (size
+ 1 != (int)var
->type
->length
) {
2587 /* If this is a built-in uniform (i.e., it's backed by some
2588 * fixed-function state), adjust the number of state slots to
2589 * match the new array size. The number of slots per array entry
2590 * is not known. It seems safe to assume that the total number of
2591 * slots is an integer multiple of the number of array elements.
2592 * Determine the number of slots per array element by dividing by
2593 * the old (total) size.
2595 const unsigned num_slots
= var
->get_num_state_slots();
2596 if (num_slots
> 0) {
2597 var
->set_num_state_slots((size
+ 1)
2598 * (num_slots
/ var
->type
->length
));
2601 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2603 types_were_updated
= true;
2607 /* Update the types of dereferences in case we changed any. */
2608 if (types_were_updated
) {
2609 deref_type_updater v
;
2610 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2616 * Resize tessellation evaluation per-vertex inputs to the size of
2617 * tessellation control per-vertex outputs.
2620 resize_tes_inputs(struct gl_context
*ctx
,
2621 struct gl_shader_program
*prog
)
2623 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2626 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2627 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2629 /* If no control shader is present, then the TES inputs are statically
2630 * sized to MaxPatchVertices; the actual size of the arrays won't be
2631 * known until draw time.
2633 const int num_vertices
= tcs
2634 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2635 : ctx
->Const
.MaxPatchVertices
;
2637 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2638 MESA_SHADER_TESS_EVAL
);
2639 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2640 ir
->accept(&input_resize_visitor
);
2644 /* Convert the gl_PatchVerticesIn system value into a constant, since
2645 * the value is known at this point.
2647 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2648 ir_variable
*var
= ir
->as_variable();
2649 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2650 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2651 void *mem_ctx
= ralloc_parent(var
);
2652 var
->data
.location
= 0;
2653 var
->data
.explicit_location
= false;
2654 var
->data
.mode
= ir_var_auto
;
2655 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2662 * Find a contiguous set of available bits in a bitmask.
2664 * \param used_mask Bits representing used (1) and unused (0) locations
2665 * \param needed_count Number of contiguous bits needed.
2668 * Base location of the available bits on success or -1 on failure.
2671 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2673 unsigned needed_mask
= (1 << needed_count
) - 1;
2674 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2676 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2677 * cannot optimize possibly infinite loops" for the loop below.
2679 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2682 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2683 if ((needed_mask
& ~used_mask
) == needed_mask
)
2693 #define SAFE_MASK_FROM_INDEX(i) (((i) >= 32) ? ~0 : ((1 << (i)) - 1))
2696 * Assign locations for either VS inputs or FS outputs.
2698 * \param mem_ctx Temporary ralloc context used for linking.
2699 * \param prog Shader program whose variables need locations
2701 * \param constants Driver specific constant values for the program.
2702 * \param target_index Selector for the program target to receive location
2703 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2704 * \c MESA_SHADER_FRAGMENT.
2705 * \param do_assignment Whether we are actually marking the assignment or we
2706 * are just doing a dry-run checking.
2709 * If locations are (or can be, in case of dry-running) successfully assigned,
2710 * true is returned. Otherwise an error is emitted to the shader link log and
2711 * false is returned.
2714 assign_attribute_or_color_locations(void *mem_ctx
,
2715 gl_shader_program
*prog
,
2716 struct gl_constants
*constants
,
2717 unsigned target_index
,
2720 /* Maximum number of generic locations. This corresponds to either the
2721 * maximum number of draw buffers or the maximum number of generic
2724 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2725 constants
->Program
[target_index
].MaxAttribs
:
2726 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2728 /* Mark invalid locations as being used.
2730 unsigned used_locations
= ~SAFE_MASK_FROM_INDEX(max_index
);
2731 unsigned double_storage_locations
= 0;
2733 assert((target_index
== MESA_SHADER_VERTEX
)
2734 || (target_index
== MESA_SHADER_FRAGMENT
));
2736 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2740 /* Operate in a total of four passes.
2742 * 1. Invalidate the location assignments for all vertex shader inputs.
2744 * 2. Assign locations for inputs that have user-defined (via
2745 * glBindVertexAttribLocation) locations and outputs that have
2746 * user-defined locations (via glBindFragDataLocation).
2748 * 3. Sort the attributes without assigned locations by number of slots
2749 * required in decreasing order. Fragmentation caused by attribute
2750 * locations assigned by the application may prevent large attributes
2751 * from having enough contiguous space.
2753 * 4. Assign locations to any inputs without assigned locations.
2756 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2757 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2759 const enum ir_variable_mode direction
=
2760 (target_index
== MESA_SHADER_VERTEX
)
2761 ? ir_var_shader_in
: ir_var_shader_out
;
2764 /* Temporary storage for the set of attributes that need locations assigned.
2770 /* Used below in the call to qsort. */
2771 static int compare(const void *a
, const void *b
)
2773 const temp_attr
*const l
= (const temp_attr
*) a
;
2774 const temp_attr
*const r
= (const temp_attr
*) b
;
2776 /* Reversed because we want a descending order sort below. */
2777 return r
->slots
- l
->slots
;
2780 assert(max_index
<= 32);
2782 /* Temporary array for the set of attributes that have locations assigned,
2783 * for the purpose of checking overlapping slots/components of (non-ES)
2784 * fragment shader outputs.
2786 ir_variable
*assigned
[12 * 4]; /* (max # of FS outputs) * # components */
2787 unsigned assigned_attr
= 0;
2789 unsigned num_attr
= 0;
2791 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2792 ir_variable
*const var
= node
->as_variable();
2794 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2797 if (var
->data
.explicit_location
) {
2798 var
->data
.is_unmatched_generic_inout
= 0;
2799 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2800 || (var
->data
.location
< 0)) {
2802 "invalid explicit location %d specified for `%s'\n",
2803 (var
->data
.location
< 0)
2804 ? var
->data
.location
2805 : var
->data
.location
- generic_base
,
2809 } else if (target_index
== MESA_SHADER_VERTEX
) {
2812 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2813 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2814 var
->data
.location
= binding
;
2815 var
->data
.is_unmatched_generic_inout
= 0;
2817 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2820 const char *name
= var
->name
;
2821 const glsl_type
*type
= var
->type
;
2824 /* Check if there's a binding for the variable name */
2825 if (prog
->FragDataBindings
->get(binding
, name
)) {
2826 assert(binding
>= FRAG_RESULT_DATA0
);
2827 var
->data
.location
= binding
;
2828 var
->data
.is_unmatched_generic_inout
= 0;
2830 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2831 var
->data
.index
= index
;
2836 /* If not, but it's an array type, look for name[0] */
2837 if (type
->is_array()) {
2838 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2839 type
= type
->fields
.array
;
2847 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2850 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2852 * "Output binding assignments will cause LinkProgram to fail:
2854 * If the program has an active output assigned to a location greater
2855 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2856 * an active output assigned an index greater than or equal to one;"
2858 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2859 var
->data
.location
- generic_base
>=
2860 (int) constants
->MaxDualSourceDrawBuffers
) {
2862 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2863 "with index %u for %s\n",
2864 var
->data
.location
- generic_base
, var
->data
.index
,
2869 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2871 /* If the variable is not a built-in and has a location statically
2872 * assigned in the shader (presumably via a layout qualifier), make sure
2873 * that it doesn't collide with other assigned locations. Otherwise,
2874 * add it to the list of variables that need linker-assigned locations.
2876 if (var
->data
.location
!= -1) {
2877 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2878 /* From page 61 of the OpenGL 4.0 spec:
2880 * "LinkProgram will fail if the attribute bindings assigned
2881 * by BindAttribLocation do not leave not enough space to
2882 * assign a location for an active matrix attribute or an
2883 * active attribute array, both of which require multiple
2884 * contiguous generic attributes."
2886 * I think above text prohibits the aliasing of explicit and
2887 * automatic assignments. But, aliasing is allowed in manual
2888 * assignments of attribute locations. See below comments for
2891 * From OpenGL 4.0 spec, page 61:
2893 * "It is possible for an application to bind more than one
2894 * attribute name to the same location. This is referred to as
2895 * aliasing. This will only work if only one of the aliased
2896 * attributes is active in the executable program, or if no
2897 * path through the shader consumes more than one attribute of
2898 * a set of attributes aliased to the same location. A link
2899 * error can occur if the linker determines that every path
2900 * through the shader consumes multiple aliased attributes,
2901 * but implementations are not required to generate an error
2904 * From GLSL 4.30 spec, page 54:
2906 * "A program will fail to link if any two non-vertex shader
2907 * input variables are assigned to the same location. For
2908 * vertex shaders, multiple input variables may be assigned
2909 * to the same location using either layout qualifiers or via
2910 * the OpenGL API. However, such aliasing is intended only to
2911 * support vertex shaders where each execution path accesses
2912 * at most one input per each location. Implementations are
2913 * permitted, but not required, to generate link-time errors
2914 * if they detect that every path through the vertex shader
2915 * executable accesses multiple inputs assigned to any single
2916 * location. For all shader types, a program will fail to link
2917 * if explicit location assignments leave the linker unable
2918 * to find space for other variables without explicit
2921 * From OpenGL ES 3.0 spec, page 56:
2923 * "Binding more than one attribute name to the same location
2924 * is referred to as aliasing, and is not permitted in OpenGL
2925 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2926 * fail when this condition exists. However, aliasing is
2927 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2928 * This will only work if only one of the aliased attributes
2929 * is active in the executable program, or if no path through
2930 * the shader consumes more than one attribute of a set of
2931 * attributes aliased to the same location. A link error can
2932 * occur if the linker determines that every path through the
2933 * shader consumes multiple aliased attributes, but implemen-
2934 * tations are not required to generate an error in this case."
2936 * After looking at above references from OpenGL, OpenGL ES and
2937 * GLSL specifications, we allow aliasing of vertex input variables
2938 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2940 * NOTE: This is not required by the spec but its worth mentioning
2941 * here that we're not doing anything to make sure that no path
2942 * through the vertex shader executable accesses multiple inputs
2943 * assigned to any single location.
2946 /* Mask representing the contiguous slots that will be used by
2949 const unsigned attr
= var
->data
.location
- generic_base
;
2950 const unsigned use_mask
= (1 << slots
) - 1;
2951 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2952 ? "vertex shader input" : "fragment shader output";
2954 /* Generate a link error if the requested locations for this
2955 * attribute exceed the maximum allowed attribute location.
2957 if (attr
+ slots
> max_index
) {
2959 "insufficient contiguous locations "
2960 "available for %s `%s' %d %d %d\n", string
,
2961 var
->name
, used_locations
, use_mask
, attr
);
2965 /* Generate a link error if the set of bits requested for this
2966 * attribute overlaps any previously allocated bits.
2968 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2969 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2970 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2973 * "Additionally, for fragment shader outputs, if two
2974 * variables are placed within the same location, they
2975 * must have the same underlying type (floating-point or
2976 * integer). No component aliasing of output variables or
2977 * members is allowed.
2979 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2980 unsigned assigned_slots
=
2981 assigned
[i
]->type
->count_attribute_slots(false);
2982 unsigned assig_attr
=
2983 assigned
[i
]->data
.location
- generic_base
;
2984 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2986 if ((assigned_use_mask
<< assig_attr
) &
2987 (use_mask
<< attr
)) {
2989 const glsl_type
*assigned_type
=
2990 assigned
[i
]->type
->without_array();
2991 const glsl_type
*type
= var
->type
->without_array();
2992 if (assigned_type
->base_type
!= type
->base_type
) {
2993 linker_error(prog
, "types do not match for aliased"
2994 " %ss %s and %s\n", string
,
2995 assigned
[i
]->name
, var
->name
);
2999 unsigned assigned_component_mask
=
3000 ((1 << assigned_type
->vector_elements
) - 1) <<
3001 assigned
[i
]->data
.location_frac
;
3002 unsigned component_mask
=
3003 ((1 << type
->vector_elements
) - 1) <<
3004 var
->data
.location_frac
;
3005 if (assigned_component_mask
& component_mask
) {
3006 linker_error(prog
, "overlapping component is "
3007 "assigned to %ss %s and %s "
3009 string
, assigned
[i
]->name
, var
->name
,
3010 var
->data
.location_frac
);
3015 } else if (target_index
== MESA_SHADER_FRAGMENT
||
3016 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
3017 linker_error(prog
, "overlapping location is assigned "
3018 "to %s `%s' %d %d %d\n", string
, var
->name
,
3019 used_locations
, use_mask
, attr
);
3022 linker_warning(prog
, "overlapping location is assigned "
3023 "to %s `%s' %d %d %d\n", string
, var
->name
,
3024 used_locations
, use_mask
, attr
);
3028 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
3029 /* Only track assigned variables for non-ES fragment shaders
3030 * to avoid overflowing the array.
3032 * At most one variable per fragment output component should
3035 assert(assigned_attr
< ARRAY_SIZE(assigned
));
3036 assigned
[assigned_attr
] = var
;
3040 used_locations
|= (use_mask
<< attr
);
3042 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
3044 * "A program with more than the value of MAX_VERTEX_ATTRIBS
3045 * active attribute variables may fail to link, unless
3046 * device-dependent optimizations are able to make the program
3047 * fit within available hardware resources. For the purposes
3048 * of this test, attribute variables of the type dvec3, dvec4,
3049 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
3050 * count as consuming twice as many attributes as equivalent
3051 * single-precision types. While these types use the same number
3052 * of generic attributes as their single-precision equivalents,
3053 * implementations are permitted to consume two single-precision
3054 * vectors of internal storage for each three- or four-component
3055 * double-precision vector."
3057 * Mark this attribute slot as taking up twice as much space
3058 * so we can count it properly against limits. According to
3059 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
3060 * is optional behavior, but it seems preferable.
3062 if (var
->type
->without_array()->is_dual_slot())
3063 double_storage_locations
|= (use_mask
<< attr
);
3069 if (num_attr
>= max_index
) {
3070 linker_error(prog
, "too many %s (max %u)",
3071 target_index
== MESA_SHADER_VERTEX
?
3072 "vertex shader inputs" : "fragment shader outputs",
3076 to_assign
[num_attr
].slots
= slots
;
3077 to_assign
[num_attr
].var
= var
;
3084 if (target_index
== MESA_SHADER_VERTEX
) {
3085 unsigned total_attribs_size
=
3086 util_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
3087 util_bitcount(double_storage_locations
);
3088 if (total_attribs_size
> max_index
) {
3090 "attempt to use %d vertex attribute slots only %d available ",
3091 total_attribs_size
, max_index
);
3096 /* If all of the attributes were assigned locations by the application (or
3097 * are built-in attributes with fixed locations), return early. This should
3098 * be the common case.
3103 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
3105 if (target_index
== MESA_SHADER_VERTEX
) {
3106 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
3107 * only be explicitly assigned by via glBindAttribLocation. Mark it as
3108 * reserved to prevent it from being automatically allocated below.
3110 find_deref_visitor
find("gl_Vertex");
3112 if (find
.variable_found())
3113 used_locations
|= (1 << 0);
3116 for (unsigned i
= 0; i
< num_attr
; i
++) {
3117 /* Mask representing the contiguous slots that will be used by this
3120 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
3122 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
3125 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
3126 ? "vertex shader input" : "fragment shader output";
3129 "insufficient contiguous locations "
3130 "available for %s `%s'\n",
3131 string
, to_assign
[i
].var
->name
);
3135 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
3136 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
3137 used_locations
|= (use_mask
<< location
);
3139 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
3140 double_storage_locations
|= (use_mask
<< location
);
3143 /* Now that we have all the locations, from the GL 4.5 core spec, section
3144 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
3145 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
3146 * as equivalent single-precision types.
3148 if (target_index
== MESA_SHADER_VERTEX
) {
3149 unsigned total_attribs_size
=
3150 util_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
3151 util_bitcount(double_storage_locations
);
3152 if (total_attribs_size
> max_index
) {
3154 "attempt to use %d vertex attribute slots only %d available ",
3155 total_attribs_size
, max_index
);
3164 * Match explicit locations of outputs to inputs and deactivate the
3165 * unmatch flag if found so we don't optimise them away.
3168 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
3169 gl_linked_shader
*consumer
)
3171 glsl_symbol_table parameters
;
3172 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
3175 /* Find all shader outputs in the "producer" stage.
3177 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
3178 ir_variable
*const var
= node
->as_variable();
3180 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
3183 if (var
->data
.explicit_location
&&
3184 var
->data
.location
>= VARYING_SLOT_VAR0
) {
3185 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
3186 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
3187 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
3191 /* Match inputs to outputs */
3192 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
3193 ir_variable
*const input
= node
->as_variable();
3195 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
3198 ir_variable
*output
= NULL
;
3199 if (input
->data
.explicit_location
3200 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
3201 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
3202 [input
->data
.location_frac
];
3204 if (output
!= NULL
){
3205 input
->data
.is_unmatched_generic_inout
= 0;
3206 output
->data
.is_unmatched_generic_inout
= 0;
3213 * Store the gl_FragDepth layout in the gl_shader_program struct.
3216 store_fragdepth_layout(struct gl_shader_program
*prog
)
3218 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3222 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3224 /* We don't look up the gl_FragDepth symbol directly because if
3225 * gl_FragDepth is not used in the shader, it's removed from the IR.
3226 * However, the symbol won't be removed from the symbol table.
3228 * We're only interested in the cases where the variable is NOT removed
3231 foreach_in_list(ir_instruction
, node
, ir
) {
3232 ir_variable
*const var
= node
->as_variable();
3234 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3238 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3239 switch (var
->data
.depth_layout
) {
3240 case ir_depth_layout_none
:
3241 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3243 case ir_depth_layout_any
:
3244 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3246 case ir_depth_layout_greater
:
3247 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3249 case ir_depth_layout_less
:
3250 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3252 case ir_depth_layout_unchanged
:
3253 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3264 * Validate the resources used by a program versus the implementation limits
3267 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3269 unsigned total_uniform_blocks
= 0;
3270 unsigned total_shader_storage_blocks
= 0;
3272 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3273 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3278 if (sh
->Program
->info
.num_textures
>
3279 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3280 linker_error(prog
, "Too many %s shader texture samplers\n",
3281 _mesa_shader_stage_to_string(i
));
3284 if (sh
->num_uniform_components
>
3285 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3286 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3287 linker_warning(prog
, "Too many %s shader default uniform block "
3288 "components, but the driver will try to optimize "
3289 "them out; this is non-portable out-of-spec "
3291 _mesa_shader_stage_to_string(i
));
3293 linker_error(prog
, "Too many %s shader default uniform block "
3295 _mesa_shader_stage_to_string(i
));
3299 if (sh
->num_combined_uniform_components
>
3300 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3301 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3302 linker_warning(prog
, "Too many %s shader uniform components, "
3303 "but the driver will try to optimize them out; "
3304 "this is non-portable out-of-spec behavior\n",
3305 _mesa_shader_stage_to_string(i
));
3307 linker_error(prog
, "Too many %s shader uniform components\n",
3308 _mesa_shader_stage_to_string(i
));
3312 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3313 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3315 const unsigned max_uniform_blocks
=
3316 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3317 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3318 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3319 _mesa_shader_stage_to_string(i
),
3320 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3323 const unsigned max_shader_storage_blocks
=
3324 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3325 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3326 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3327 _mesa_shader_stage_to_string(i
),
3328 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3332 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3333 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3334 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3337 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3338 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3339 total_shader_storage_blocks
,
3340 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3343 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3344 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3345 ctx
->Const
.MaxUniformBlockSize
) {
3346 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3347 prog
->data
->UniformBlocks
[i
].Name
,
3348 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3349 ctx
->Const
.MaxUniformBlockSize
);
3353 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3354 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3355 ctx
->Const
.MaxShaderStorageBlockSize
) {
3356 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3357 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3358 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3359 ctx
->Const
.MaxShaderStorageBlockSize
);
3365 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3367 unsigned mask
= prog
->data
->linked_stages
;
3369 const int i
= u_bit_scan(&mask
);
3370 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3372 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3373 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3376 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3382 if (p
->sh
.NumSubroutineFunctions
== 0) {
3383 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3386 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3387 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3388 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3389 if (fn
->types
[k
] == uni
->type
) {
3395 uni
->num_compatible_subroutines
= count
;
3401 check_subroutine_resources(struct gl_shader_program
*prog
)
3403 unsigned mask
= prog
->data
->linked_stages
;
3405 const int i
= u_bit_scan(&mask
);
3406 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3408 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3409 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3410 _mesa_shader_stage_to_string(i
));
3415 * Validate shader image resources.
3418 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3420 unsigned total_image_units
= 0;
3421 unsigned fragment_outputs
= 0;
3422 unsigned total_shader_storage_blocks
= 0;
3424 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3427 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3428 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3431 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3432 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3433 _mesa_shader_stage_to_string(i
),
3434 sh
->Program
->info
.num_images
,
3435 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3437 total_image_units
+= sh
->Program
->info
.num_images
;
3438 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3440 if (i
== MESA_SHADER_FRAGMENT
) {
3441 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3442 ir_variable
*var
= node
->as_variable();
3443 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3444 /* since there are no double fs outputs - pass false */
3445 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3451 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3452 linker_error(prog
, "Too many combined image uniforms\n");
3454 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3455 ctx
->Const
.MaxCombinedShaderOutputResources
)
3456 linker_error(prog
, "Too many combined image uniforms, shader storage "
3457 " buffers and fragment outputs\n");
3462 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3463 * for a variable, checks for overlaps between other uniforms using explicit
3467 reserve_explicit_locations(struct gl_shader_program
*prog
,
3468 string_to_uint_map
*map
, ir_variable
*var
)
3470 unsigned slots
= var
->type
->uniform_locations();
3471 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3472 unsigned return_value
= slots
;
3474 /* Resize remap table if locations do not fit in the current one. */
3475 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3476 prog
->UniformRemapTable
=
3477 reralloc(prog
, prog
->UniformRemapTable
,
3478 gl_uniform_storage
*,
3481 if (!prog
->UniformRemapTable
) {
3482 linker_error(prog
, "Out of memory during linking.\n");
3486 /* Initialize allocated space. */
3487 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3488 prog
->UniformRemapTable
[i
] = NULL
;
3490 prog
->NumUniformRemapTable
= max_loc
+ 1;
3493 for (unsigned i
= 0; i
< slots
; i
++) {
3494 unsigned loc
= var
->data
.location
+ i
;
3496 /* Check if location is already used. */
3497 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3499 /* Possibly same uniform from a different stage, this is ok. */
3501 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3506 /* ARB_explicit_uniform_location specification states:
3508 * "No two default-block uniform variables in the program can have
3509 * the same location, even if they are unused, otherwise a compiler
3510 * or linker error will be generated."
3513 "location qualifier for uniform %s overlaps "
3514 "previously used location\n",
3519 /* Initialize location as inactive before optimization
3520 * rounds and location assignment.
3522 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3525 /* Note, base location used for arrays. */
3526 map
->put(var
->data
.location
, var
->name
);
3528 return return_value
;
3532 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3533 struct gl_program
*p
,
3536 unsigned slots
= var
->type
->uniform_locations();
3537 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3539 /* Resize remap table if locations do not fit in the current one. */
3540 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3541 p
->sh
.SubroutineUniformRemapTable
=
3542 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3543 gl_uniform_storage
*,
3546 if (!p
->sh
.SubroutineUniformRemapTable
) {
3547 linker_error(prog
, "Out of memory during linking.\n");
3551 /* Initialize allocated space. */
3552 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3553 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3555 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3558 for (unsigned i
= 0; i
< slots
; i
++) {
3559 unsigned loc
= var
->data
.location
+ i
;
3561 /* Check if location is already used. */
3562 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3564 /* ARB_explicit_uniform_location specification states:
3565 * "No two subroutine uniform variables can have the same location
3566 * in the same shader stage, otherwise a compiler or linker error
3567 * will be generated."
3570 "location qualifier for uniform %s overlaps "
3571 "previously used location\n",
3576 /* Initialize location as inactive before optimization
3577 * rounds and location assignment.
3579 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3585 * Check and reserve all explicit uniform locations, called before
3586 * any optimizations happen to handle also inactive uniforms and
3587 * inactive array elements that may get trimmed away.
3590 check_explicit_uniform_locations(struct gl_context
*ctx
,
3591 struct gl_shader_program
*prog
)
3593 prog
->NumExplicitUniformLocations
= 0;
3595 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3598 /* This map is used to detect if overlapping explicit locations
3599 * occur with the same uniform (from different stage) or a different one.
3601 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3604 linker_error(prog
, "Out of memory during linking.\n");
3608 unsigned entries_total
= 0;
3609 unsigned mask
= prog
->data
->linked_stages
;
3611 const int i
= u_bit_scan(&mask
);
3612 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3614 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3615 ir_variable
*var
= node
->as_variable();
3616 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3619 if (var
->data
.explicit_location
) {
3621 if (var
->type
->without_array()->is_subroutine())
3622 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3624 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3628 entries_total
+= slots
;
3639 link_util_update_empty_uniform_locations(prog
);
3642 prog
->NumExplicitUniformLocations
= entries_total
;
3646 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3647 GLenum type
, const char *name
)
3649 bool found_interface
= false;
3650 unsigned block_name_len
= 0;
3651 const char *block_name_dot
= strchr(name
, '.');
3653 /* These rules only apply to buffer variables. So we return
3654 * true for the rest of types.
3656 if (type
!= GL_BUFFER_VARIABLE
)
3659 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3660 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3661 block_name_len
= strlen(block_name
);
3663 const char *block_square_bracket
= strchr(block_name
, '[');
3664 if (block_square_bracket
) {
3665 /* The block is part of an array of named interfaces,
3666 * for the name comparison we ignore the "[x]" part.
3668 block_name_len
-= strlen(block_square_bracket
);
3671 if (block_name_dot
) {
3672 /* Check if the variable name starts with the interface
3673 * name. The interface name (if present) should have the
3674 * length than the interface block name we are comparing to.
3676 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3677 if (len
!= block_name_len
)
3681 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3682 found_interface
= true;
3687 /* We remove the interface name from the buffer variable name,
3688 * including the dot that follows it.
3690 if (found_interface
)
3691 name
= name
+ block_name_len
+ 1;
3693 /* The ARB_program_interface_query spec says:
3695 * "For an active shader storage block member declared as an array, an
3696 * entry will be generated only for the first array element, regardless
3697 * of its type. For arrays of aggregate types, the enumeration rules
3698 * are applied recursively for the single enumerated array element."
3700 const char *struct_first_dot
= strchr(name
, '.');
3701 const char *first_square_bracket
= strchr(name
, '[');
3703 /* The buffer variable is on top level and it is not an array */
3704 if (!first_square_bracket
) {
3706 /* The shader storage block member is a struct, then generate the entry */
3707 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3710 /* Shader storage block member is an array, only generate an entry for the
3711 * first array element.
3713 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3720 /* Function checks if a variable var is a packed varying and
3721 * if given name is part of packed varying's list.
3723 * If a variable is a packed varying, it has a name like
3724 * 'packed:a,b,c' where a, b and c are separate variables.
3727 included_in_packed_varying(ir_variable
*var
, const char *name
)
3729 if (strncmp(var
->name
, "packed:", 7) != 0)
3732 char *list
= strdup(var
->name
+ 7);
3737 char *token
= strtok_r(list
, ",", &saveptr
);
3739 if (strcmp(token
, name
) == 0) {
3743 token
= strtok_r(NULL
, ",", &saveptr
);
3750 * Function builds a stage reference bitmask from variable name.
3753 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3758 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3759 * used for reference mask in gl_program_resource will need to be changed.
3761 assert(MESA_SHADER_STAGES
< 8);
3763 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3764 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3768 /* Shader symbol table may contain variables that have
3769 * been optimized away. Search IR for the variable instead.
3771 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3772 ir_variable
*var
= node
->as_variable();
3774 unsigned baselen
= strlen(var
->name
);
3776 if (included_in_packed_varying(var
, name
)) {
3781 /* Type needs to match if specified, otherwise we might
3782 * pick a variable with same name but different interface.
3784 if (var
->data
.mode
!= mode
)
3787 if (strncmp(var
->name
, name
, baselen
) == 0) {
3788 /* Check for exact name matches but also check for arrays and
3791 if (name
[baselen
] == '\0' ||
3792 name
[baselen
] == '[' ||
3793 name
[baselen
] == '.') {
3805 * Create gl_shader_variable from ir_variable class.
3807 static gl_shader_variable
*
3808 create_shader_variable(struct gl_shader_program
*shProg
,
3809 const ir_variable
*in
,
3810 const char *name
, const glsl_type
*type
,
3811 const glsl_type
*interface_type
,
3812 bool use_implicit_location
, int location
,
3813 const glsl_type
*outermost_struct_type
)
3815 /* Allocate zero-initialized memory to ensure that bitfield padding
3818 gl_shader_variable
*out
= rzalloc(shProg
, struct gl_shader_variable
);
3822 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3823 * expect to see gl_VertexID in the program resource list. Pretend.
3825 if (in
->data
.mode
== ir_var_system_value
&&
3826 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3827 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3828 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3829 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3830 (in
->data
.mode
== ir_var_system_value
&&
3831 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3832 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3833 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3834 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3835 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3836 (in
->data
.mode
== ir_var_system_value
&&
3837 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3838 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3839 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3841 out
->name
= ralloc_strdup(shProg
, name
);
3847 /* The ARB_program_interface_query spec says:
3849 * "Not all active variables are assigned valid locations; the
3850 * following variables will have an effective location of -1:
3852 * * uniforms declared as atomic counters;
3854 * * members of a uniform block;
3856 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3858 * * inputs or outputs not declared with a "location" layout
3859 * qualifier, except for vertex shader inputs and fragment shader
3862 if (in
->type
->is_atomic_uint() || is_gl_identifier(in
->name
) ||
3863 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3866 out
->location
= location
;
3870 out
->outermost_struct_type
= outermost_struct_type
;
3871 out
->interface_type
= interface_type
;
3872 out
->component
= in
->data
.location_frac
;
3873 out
->index
= in
->data
.index
;
3874 out
->patch
= in
->data
.patch
;
3875 out
->mode
= in
->data
.mode
;
3876 out
->interpolation
= in
->data
.interpolation
;
3877 out
->explicit_location
= in
->data
.explicit_location
;
3878 out
->precision
= in
->data
.precision
;
3884 add_shader_variable(const struct gl_context
*ctx
,
3885 struct gl_shader_program
*shProg
,
3886 struct set
*resource_set
,
3887 unsigned stage_mask
,
3888 GLenum programInterface
, ir_variable
*var
,
3889 const char *name
, const glsl_type
*type
,
3890 bool use_implicit_location
, int location
,
3891 bool inouts_share_location
,
3892 const glsl_type
*outermost_struct_type
= NULL
)
3894 const glsl_type
*interface_type
= var
->get_interface_type();
3896 if (outermost_struct_type
== NULL
) {
3897 if (var
->data
.from_named_ifc_block
) {
3898 const char *interface_name
= interface_type
->name
;
3900 if (interface_type
->is_array()) {
3901 /* Issue #16 of the ARB_program_interface_query spec says:
3903 * "* If a variable is a member of an interface block without an
3904 * instance name, it is enumerated using just the variable name.
3906 * * If a variable is a member of an interface block with an
3907 * instance name, it is enumerated as "BlockName.Member", where
3908 * "BlockName" is the name of the interface block (not the
3909 * instance name) and "Member" is the name of the variable."
3911 * In particular, it indicates that it should be "BlockName",
3912 * not "BlockName[array length]". The conformance suite and
3913 * dEQP both require this behavior.
3915 * Here, we unwrap the extra array level added by named interface
3916 * block array lowering so we have the correct variable type. We
3917 * also unwrap the interface type when constructing the name.
3919 * We leave interface_type the same so that ES 3.x SSO pipeline
3920 * validation can enforce the rules requiring array length to
3921 * match on interface blocks.
3923 type
= type
->fields
.array
;
3925 interface_name
= interface_type
->fields
.array
->name
;
3928 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3932 switch (type
->base_type
) {
3933 case GLSL_TYPE_STRUCT
: {
3934 /* The ARB_program_interface_query spec says:
3936 * "For an active variable declared as a structure, a separate entry
3937 * will be generated for each active structure member. The name of
3938 * each entry is formed by concatenating the name of the structure,
3939 * the "." character, and the name of the structure member. If a
3940 * structure member to enumerate is itself a structure or array,
3941 * these enumeration rules are applied recursively."
3943 if (outermost_struct_type
== NULL
)
3944 outermost_struct_type
= type
;
3946 unsigned field_location
= location
;
3947 for (unsigned i
= 0; i
< type
->length
; i
++) {
3948 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3949 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3950 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3951 stage_mask
, programInterface
,
3952 var
, field_name
, field
->type
,
3953 use_implicit_location
, field_location
,
3954 false, outermost_struct_type
))
3957 field_location
+= field
->type
->count_attribute_slots(false);
3962 case GLSL_TYPE_ARRAY
: {
3963 /* The ARB_program_interface_query spec says:
3965 * "For an active variable declared as an array of basic types, a
3966 * single entry will be generated, with its name string formed by
3967 * concatenating the name of the array and the string "[0]"."
3969 * "For an active variable declared as an array of an aggregate data
3970 * type (structures or arrays), a separate entry will be generated
3971 * for each active array element, unless noted immediately below.
3972 * The name of each entry is formed by concatenating the name of
3973 * the array, the "[" character, an integer identifying the element
3974 * number, and the "]" character. These enumeration rules are
3975 * applied recursively, treating each enumerated array element as a
3976 * separate active variable."
3978 const struct glsl_type
*array_type
= type
->fields
.array
;
3979 if (array_type
->base_type
== GLSL_TYPE_STRUCT
||
3980 array_type
->base_type
== GLSL_TYPE_ARRAY
) {
3981 unsigned elem_location
= location
;
3982 unsigned stride
= inouts_share_location
? 0 :
3983 array_type
->count_attribute_slots(false);
3984 for (unsigned i
= 0; i
< type
->length
; i
++) {
3985 char *elem
= ralloc_asprintf(shProg
, "%s[%d]", name
, i
);
3986 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3987 stage_mask
, programInterface
,
3988 var
, elem
, array_type
,
3989 use_implicit_location
, elem_location
,
3990 false, outermost_struct_type
))
3992 elem_location
+= stride
;
4000 /* The ARB_program_interface_query spec says:
4002 * "For an active variable declared as a single instance of a basic
4003 * type, a single entry will be generated, using the variable name
4004 * from the shader source."
4006 gl_shader_variable
*sha_v
=
4007 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
4008 use_implicit_location
, location
,
4009 outermost_struct_type
);
4013 return link_util_add_program_resource(shProg
, resource_set
,
4014 programInterface
, sha_v
, stage_mask
);
4020 inout_has_same_location(const ir_variable
*var
, unsigned stage
)
4022 if (!var
->data
.patch
&&
4023 ((var
->data
.mode
== ir_var_shader_out
&&
4024 stage
== MESA_SHADER_TESS_CTRL
) ||
4025 (var
->data
.mode
== ir_var_shader_in
&&
4026 (stage
== MESA_SHADER_TESS_CTRL
|| stage
== MESA_SHADER_TESS_EVAL
||
4027 stage
== MESA_SHADER_GEOMETRY
))))
4034 add_interface_variables(const struct gl_context
*ctx
,
4035 struct gl_shader_program
*shProg
,
4036 struct set
*resource_set
,
4037 unsigned stage
, GLenum programInterface
)
4039 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
4041 foreach_in_list(ir_instruction
, node
, ir
) {
4042 ir_variable
*var
= node
->as_variable();
4044 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
4049 switch (var
->data
.mode
) {
4050 case ir_var_system_value
:
4051 case ir_var_shader_in
:
4052 if (programInterface
!= GL_PROGRAM_INPUT
)
4054 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
4055 : int(VARYING_SLOT_VAR0
);
4057 case ir_var_shader_out
:
4058 if (programInterface
!= GL_PROGRAM_OUTPUT
)
4060 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
4061 : int(VARYING_SLOT_VAR0
);
4067 if (var
->data
.patch
)
4068 loc_bias
= int(VARYING_SLOT_PATCH0
);
4070 /* Skip packed varyings, packed varyings are handled separately
4071 * by add_packed_varyings.
4073 if (strncmp(var
->name
, "packed:", 7) == 0)
4076 /* Skip fragdata arrays, these are handled separately
4077 * by add_fragdata_arrays.
4079 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
4082 const bool vs_input_or_fs_output
=
4083 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
4084 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
4086 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4087 1 << stage
, programInterface
,
4088 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
4089 var
->data
.location
- loc_bias
,
4090 inout_has_same_location(var
, stage
)))
4097 add_packed_varyings(const struct gl_context
*ctx
,
4098 struct gl_shader_program
*shProg
,
4099 struct set
*resource_set
,
4100 int stage
, GLenum type
)
4102 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
4105 if (!sh
|| !sh
->packed_varyings
)
4108 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
4109 ir_variable
*var
= node
->as_variable();
4111 switch (var
->data
.mode
) {
4112 case ir_var_shader_in
:
4113 iface
= GL_PROGRAM_INPUT
;
4115 case ir_var_shader_out
:
4116 iface
= GL_PROGRAM_OUTPUT
;
4119 unreachable("unexpected type");
4122 if (type
== iface
) {
4123 const int stage_mask
=
4124 build_stageref(shProg
, var
->name
, var
->data
.mode
);
4125 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4127 iface
, var
, var
->name
, var
->type
, false,
4128 var
->data
.location
- VARYING_SLOT_VAR0
,
4129 inout_has_same_location(var
, stage
)))
4138 add_fragdata_arrays(const struct gl_context
*ctx
,
4139 struct gl_shader_program
*shProg
,
4140 struct set
*resource_set
)
4142 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4144 if (!sh
|| !sh
->fragdata_arrays
)
4147 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
4148 ir_variable
*var
= node
->as_variable();
4150 assert(var
->data
.mode
== ir_var_shader_out
);
4152 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4153 1 << MESA_SHADER_FRAGMENT
,
4154 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
4155 true, var
->data
.location
- FRAG_RESULT_DATA0
,
4164 get_top_level_name(const char *name
)
4166 const char *first_dot
= strchr(name
, '.');
4167 const char *first_square_bracket
= strchr(name
, '[');
4170 /* The ARB_program_interface_query spec says:
4172 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4173 * the number of active array elements of the top-level shader storage
4174 * block member containing to the active variable is written to
4175 * <params>. If the top-level block member is not declared as an
4176 * array, the value one is written to <params>. If the top-level block
4177 * member is an array with no declared size, the value zero is written
4181 /* The buffer variable is on top level.*/
4182 if (!first_square_bracket
&& !first_dot
)
4183 name_size
= strlen(name
);
4184 else if ((!first_square_bracket
||
4185 (first_dot
&& first_dot
< first_square_bracket
)))
4186 name_size
= first_dot
- name
;
4188 name_size
= first_square_bracket
- name
;
4190 return strndup(name
, name_size
);
4194 get_var_name(const char *name
)
4196 const char *first_dot
= strchr(name
, '.');
4199 return strdup(name
);
4201 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4205 is_top_level_shader_storage_block_member(const char* name
,
4206 const char* interface_name
,
4207 const char* field_name
)
4209 bool result
= false;
4211 /* If the given variable is already a top-level shader storage
4212 * block member, then return array_size = 1.
4213 * We could have two possibilities: if we have an instanced
4214 * shader storage block or not instanced.
4216 * For the first, we check create a name as it was in top level and
4217 * compare it with the real name. If they are the same, then
4218 * the variable is already at top-level.
4220 * Full instanced name is: interface name + '.' + var name +
4223 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4224 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4225 if (!full_instanced_name
) {
4226 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4230 util_snprintf(full_instanced_name
, name_length
, "%s.%s",
4231 interface_name
, field_name
);
4233 /* Check if its top-level shader storage block member of an
4234 * instanced interface block, or of a unnamed interface block.
4236 if (strcmp(name
, full_instanced_name
) == 0 ||
4237 strcmp(name
, field_name
) == 0)
4240 free(full_instanced_name
);
4245 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4246 char *interface_name
, char *var_name
)
4248 /* The ARB_program_interface_query spec says:
4250 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4251 * the number of active array elements of the top-level shader storage
4252 * block member containing to the active variable is written to
4253 * <params>. If the top-level block member is not declared as an
4254 * array, the value one is written to <params>. If the top-level block
4255 * member is an array with no declared size, the value zero is written
4258 if (is_top_level_shader_storage_block_member(uni
->name
,
4262 else if (field
->type
->is_unsized_array())
4264 else if (field
->type
->is_array())
4265 return field
->type
->length
;
4271 get_array_stride(struct gl_context
*ctx
, struct gl_uniform_storage
*uni
,
4272 const glsl_type
*iface
, const glsl_struct_field
*field
,
4273 char *interface_name
, char *var_name
)
4275 /* The ARB_program_interface_query spec says:
4277 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4278 * identifying the stride between array elements of the top-level
4279 * shader storage block member containing the active variable is
4280 * written to <params>. For top-level block members declared as
4281 * arrays, the value written is the difference, in basic machine units,
4282 * between the offsets of the active variable for consecutive elements
4283 * in the top-level array. For top-level block members not declared as
4284 * an array, zero is written to <params>."
4286 if (field
->type
->is_array()) {
4287 const enum glsl_matrix_layout matrix_layout
=
4288 glsl_matrix_layout(field
->matrix_layout
);
4289 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4290 const glsl_type
*array_type
= field
->type
->fields
.array
;
4292 if (is_top_level_shader_storage_block_member(uni
->name
,
4297 if (GLSL_INTERFACE_PACKING_STD140
==
4299 get_internal_ifc_packing(ctx
->Const
.UseSTD430AsDefaultPacking
)) {
4300 if (array_type
->is_record() || array_type
->is_array())
4301 return glsl_align(array_type
->std140_size(row_major
), 16);
4303 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4305 return array_type
->std430_array_stride(row_major
);
4312 calculate_array_size_and_stride(struct gl_context
*ctx
,
4313 struct gl_shader_program
*shProg
,
4314 struct gl_uniform_storage
*uni
)
4316 int block_index
= uni
->block_index
;
4317 int array_size
= -1;
4318 int array_stride
= -1;
4319 char *var_name
= get_top_level_name(uni
->name
);
4320 char *interface_name
=
4321 get_top_level_name(uni
->is_shader_storage
?
4322 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4323 shProg
->data
->UniformBlocks
[block_index
].Name
);
4325 if (strcmp(var_name
, interface_name
) == 0) {
4326 /* Deal with instanced array of SSBOs */
4327 char *temp_name
= get_var_name(uni
->name
);
4329 linker_error(shProg
, "Out of memory during linking.\n");
4330 goto write_top_level_array_size_and_stride
;
4333 var_name
= get_top_level_name(temp_name
);
4336 linker_error(shProg
, "Out of memory during linking.\n");
4337 goto write_top_level_array_size_and_stride
;
4341 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4342 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4346 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4347 ir_variable
*var
= node
->as_variable();
4348 if (!var
|| !var
->get_interface_type() ||
4349 var
->data
.mode
!= ir_var_shader_storage
)
4352 const glsl_type
*iface
= var
->get_interface_type();
4354 if (strcmp(interface_name
, iface
->name
) != 0)
4357 for (unsigned i
= 0; i
< iface
->length
; i
++) {
4358 const glsl_struct_field
*field
= &iface
->fields
.structure
[i
];
4359 if (strcmp(field
->name
, var_name
) != 0)
4362 array_stride
= get_array_stride(ctx
, uni
, iface
, field
,
4363 interface_name
, var_name
);
4364 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4365 goto write_top_level_array_size_and_stride
;
4369 write_top_level_array_size_and_stride
:
4370 free(interface_name
);
4372 uni
->top_level_array_stride
= array_stride
;
4373 uni
->top_level_array_size
= array_size
;
4377 * Builds up a list of program resources that point to existing
4381 build_program_resource_list(struct gl_context
*ctx
,
4382 struct gl_shader_program
*shProg
)
4384 /* Rebuild resource list. */
4385 if (shProg
->data
->ProgramResourceList
) {
4386 ralloc_free(shProg
->data
->ProgramResourceList
);
4387 shProg
->data
->ProgramResourceList
= NULL
;
4388 shProg
->data
->NumProgramResourceList
= 0;
4391 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4393 /* Determine first input and final output stage. These are used to
4394 * detect which variables should be enumerated in the resource list
4395 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4397 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4398 if (!shProg
->_LinkedShaders
[i
])
4400 if (input_stage
== MESA_SHADER_STAGES
)
4405 /* Empty shader, no resources. */
4406 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4409 struct set
*resource_set
= _mesa_set_create(NULL
,
4411 _mesa_key_pointer_equal
);
4413 /* Program interface needs to expose varyings in case of SSO. */
4414 if (shProg
->SeparateShader
) {
4415 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4416 input_stage
, GL_PROGRAM_INPUT
))
4419 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4420 output_stage
, GL_PROGRAM_OUTPUT
))
4424 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4427 /* Add inputs and outputs to the resource list. */
4428 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4429 input_stage
, GL_PROGRAM_INPUT
))
4432 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4433 output_stage
, GL_PROGRAM_OUTPUT
))
4436 if (shProg
->last_vert_prog
) {
4437 struct gl_transform_feedback_info
*linked_xfb
=
4438 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4440 /* Add transform feedback varyings. */
4441 if (linked_xfb
->NumVarying
> 0) {
4442 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4443 if (!link_util_add_program_resource(shProg
, resource_set
,
4444 GL_TRANSFORM_FEEDBACK_VARYING
,
4445 &linked_xfb
->Varyings
[i
], 0))
4450 /* Add transform feedback buffers. */
4451 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4452 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4453 linked_xfb
->Buffers
[i
].Binding
= i
;
4454 if (!link_util_add_program_resource(shProg
, resource_set
,
4455 GL_TRANSFORM_FEEDBACK_BUFFER
,
4456 &linked_xfb
->Buffers
[i
], 0))
4462 /* Add uniforms from uniform storage. */
4463 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4464 /* Do not add uniforms internally used by Mesa. */
4465 if (shProg
->data
->UniformStorage
[i
].hidden
)
4469 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4472 /* Add stagereferences for uniforms in a uniform block. */
4473 bool is_shader_storage
=
4474 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4475 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4476 if (block_index
!= -1) {
4477 stageref
|= is_shader_storage
?
4478 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4479 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4482 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4483 if (!should_add_buffer_variable(shProg
, type
,
4484 shProg
->data
->UniformStorage
[i
].name
))
4487 if (is_shader_storage
) {
4488 calculate_array_size_and_stride(ctx
, shProg
,
4489 &shProg
->data
->UniformStorage
[i
]);
4492 if (!link_util_add_program_resource(shProg
, resource_set
, type
,
4493 &shProg
->data
->UniformStorage
[i
], stageref
))
4497 /* Add program uniform blocks. */
4498 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4499 if (!link_util_add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4500 &shProg
->data
->UniformBlocks
[i
], 0))
4504 /* Add program shader storage blocks. */
4505 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4506 if (!link_util_add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4507 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4511 /* Add atomic counter buffers. */
4512 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4513 if (!link_util_add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4514 &shProg
->data
->AtomicBuffers
[i
], 0))
4518 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4520 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4523 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4524 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4525 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4528 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4529 /* add shader subroutines */
4530 if (!link_util_add_program_resource(shProg
, resource_set
,
4531 type
, &shProg
->data
->UniformStorage
[i
], 0))
4536 unsigned mask
= shProg
->data
->linked_stages
;
4538 const int i
= u_bit_scan(&mask
);
4539 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4541 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4542 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4543 if (!link_util_add_program_resource(shProg
, resource_set
,
4544 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4549 _mesa_set_destroy(resource_set
, NULL
);
4553 * This check is done to make sure we allow only constant expression
4554 * indexing and "constant-index-expression" (indexing with an expression
4555 * that includes loop induction variable).
4558 validate_sampler_array_indexing(struct gl_context
*ctx
,
4559 struct gl_shader_program
*prog
)
4561 dynamic_sampler_array_indexing_visitor v
;
4562 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4563 if (prog
->_LinkedShaders
[i
] == NULL
)
4566 bool no_dynamic_indexing
=
4567 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4569 /* Search for array derefs in shader. */
4570 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4571 if (v
.uses_dynamic_sampler_array_indexing()) {
4572 const char *msg
= "sampler arrays indexed with non-constant "
4573 "expressions is forbidden in GLSL %s %u";
4574 /* Backend has indicated that it has no dynamic indexing support. */
4575 if (no_dynamic_indexing
) {
4576 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4577 prog
->data
->Version
);
4580 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4581 prog
->data
->Version
);
4589 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4591 unsigned mask
= prog
->data
->linked_stages
;
4593 const int i
= u_bit_scan(&mask
);
4594 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4596 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4597 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4598 ir_function
*fn
= node
->as_function();
4602 if (fn
->is_subroutine
)
4603 p
->sh
.NumSubroutineUniformTypes
++;
4605 if (!fn
->num_subroutine_types
)
4608 /* these should have been calculated earlier. */
4609 assert(fn
->subroutine_index
!= -1);
4610 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4611 linker_error(prog
, "Too many subroutine functions declared.\n");
4614 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4615 struct gl_subroutine_function
,
4616 p
->sh
.NumSubroutineFunctions
+ 1);
4617 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4618 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4619 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4620 ralloc_array(p
, const struct glsl_type
*,
4621 fn
->num_subroutine_types
);
4623 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4626 * "Each subroutine with an index qualifier in the shader must be
4627 * given a unique index, otherwise a compile or link error will be
4630 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4631 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4632 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4633 linker_error(prog
, "each subroutine index qualifier in the "
4634 "shader must be unique\n");
4638 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4639 fn
->subroutine_index
;
4641 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4642 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4644 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4645 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4646 p
->sh
.NumSubroutineFunctions
++;
4652 verify_subroutine_associated_funcs(struct gl_shader_program
*prog
)
4654 unsigned mask
= prog
->data
->linked_stages
;
4656 const int i
= u_bit_scan(&mask
);
4657 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4658 glsl_symbol_table
*symbols
= prog
->_LinkedShaders
[i
]->symbols
;
4660 /* Section 6.1.2 (Subroutines) of the GLSL 4.00 spec says:
4662 * "A program will fail to compile or link if any shader
4663 * or stage contains two or more functions with the same
4664 * name if the name is associated with a subroutine type."
4666 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4667 unsigned definitions
= 0;
4668 char *name
= p
->sh
.SubroutineFunctions
[j
].name
;
4669 ir_function
*fn
= symbols
->get_function(name
);
4671 /* Calculate number of function definitions with the same name */
4672 foreach_in_list(ir_function_signature
, sig
, &fn
->signatures
) {
4673 if (sig
->is_defined
) {
4674 if (++definitions
> 1) {
4675 linker_error(prog
, "%s shader contains two or more function "
4676 "definitions with name `%s', which is "
4677 "associated with a subroutine type.\n",
4678 _mesa_shader_stage_to_string(i
),
4690 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4692 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4694 foreach_in_list(ir_instruction
, node
, ir
) {
4695 ir_variable
*const var
= node
->as_variable();
4697 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4700 /* Don't set always active on builtins that haven't been redeclared */
4701 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4704 var
->data
.always_active_io
= true;
4709 * When separate shader programs are enabled, only input/outputs between
4710 * the stages of a multi-stage separate program can be safely removed
4711 * from the shader interface. Other inputs/outputs must remain active.
4714 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4716 unsigned first
, last
;
4717 assert(prog
->SeparateShader
);
4719 first
= MESA_SHADER_STAGES
;
4722 /* Determine first and last stage. Excluding the compute stage */
4723 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4724 if (!prog
->_LinkedShaders
[i
])
4726 if (first
== MESA_SHADER_STAGES
)
4731 if (first
== MESA_SHADER_STAGES
)
4734 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4735 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4739 /* Prevent the removal of inputs to the first and outputs from the last
4740 * stage, unless they are the initial pipeline inputs or final pipeline
4741 * outputs, respectively.
4743 * The removal of IO between shaders in the same program is always
4746 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4747 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4748 if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4749 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4754 link_and_validate_uniforms(struct gl_context
*ctx
,
4755 struct gl_shader_program
*prog
)
4757 update_array_sizes(prog
);
4758 link_assign_uniform_locations(prog
, ctx
);
4760 link_assign_atomic_counter_resources(ctx
, prog
);
4761 link_calculate_subroutine_compat(prog
);
4762 check_resources(ctx
, prog
);
4763 check_subroutine_resources(prog
);
4764 check_image_resources(ctx
, prog
);
4765 link_check_atomic_counter_resources(ctx
, prog
);
4769 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4770 struct gl_context
*ctx
,
4771 struct gl_shader_program
*prog
, void *mem_ctx
)
4773 /* Mark all generic shader inputs and outputs as unpaired. */
4774 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4775 if (prog
->_LinkedShaders
[i
] != NULL
) {
4776 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4780 unsigned prev
= first
;
4781 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4782 if (prog
->_LinkedShaders
[i
] == NULL
)
4785 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4786 prog
->_LinkedShaders
[i
]);
4790 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4791 MESA_SHADER_VERTEX
, true)) {
4795 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4796 MESA_SHADER_FRAGMENT
, true)) {
4800 prog
->last_vert_prog
= NULL
;
4801 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4802 if (prog
->_LinkedShaders
[i
] == NULL
)
4805 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4809 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4812 link_and_validate_uniforms(ctx
, prog
);
4814 if (!prog
->data
->LinkStatus
)
4817 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4818 if (prog
->_LinkedShaders
[i
] == NULL
)
4821 const struct gl_shader_compiler_options
*options
=
4822 &ctx
->Const
.ShaderCompilerOptions
[i
];
4824 if (options
->LowerBufferInterfaceBlocks
)
4825 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4826 options
->ClampBlockIndicesToArrayBounds
,
4827 ctx
->Const
.UseSTD430AsDefaultPacking
);
4829 if (i
== MESA_SHADER_COMPUTE
)
4830 lower_shared_reference(ctx
, prog
, prog
->_LinkedShaders
[i
]);
4832 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4833 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4840 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4843 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4844 /* Run it just once. */
4845 do_common_optimization(ir
, true, false,
4846 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4847 ctx
->Const
.NativeIntegers
);
4849 /* Repeat it until it stops making changes. */
4850 while (do_common_optimization(ir
, true, false,
4851 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4852 ctx
->Const
.NativeIntegers
))
4858 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4860 prog
->data
->LinkStatus
= LINKING_SUCCESS
; /* All error paths will set this to false */
4861 prog
->data
->Validated
= false;
4863 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4865 * "Linking can fail for a variety of reasons as specified in the
4866 * OpenGL Shading Language Specification, as well as any of the
4867 * following reasons:
4869 * - No shader objects are attached to program."
4871 * The Compatibility Profile specification does not list the error. In
4872 * Compatibility Profile missing shader stages are replaced by
4873 * fixed-function. This applies to the case where all stages are
4876 if (prog
->NumShaders
== 0) {
4877 if (ctx
->API
!= API_OPENGL_COMPAT
)
4878 linker_error(prog
, "no shaders attached to the program\n");
4882 #ifdef ENABLE_SHADER_CACHE
4883 if (shader_cache_read_program_metadata(ctx
, prog
))
4887 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4889 prog
->ARB_fragment_coord_conventions_enable
= false;
4891 /* Separate the shaders into groups based on their type.
4893 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4894 unsigned num_shaders
[MESA_SHADER_STAGES
];
4896 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4897 shader_list
[i
] = (struct gl_shader
**)
4898 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4902 unsigned min_version
= UINT_MAX
;
4903 unsigned max_version
= 0;
4904 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4905 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4906 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4908 if (!ctx
->Const
.AllowGLSLRelaxedES
&&
4909 prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4910 linker_error(prog
, "all shaders must use same shading "
4911 "language version\n");
4915 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4916 prog
->ARB_fragment_coord_conventions_enable
= true;
4919 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4920 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4921 num_shaders
[shader_type
]++;
4924 /* In desktop GLSL, different shader versions may be linked together. In
4925 * GLSL ES, all shader versions must be the same.
4927 if (!ctx
->Const
.AllowGLSLRelaxedES
&& prog
->Shaders
[0]->IsES
&&
4928 min_version
!= max_version
) {
4929 linker_error(prog
, "all shaders must use same shading "
4930 "language version\n");
4934 prog
->data
->Version
= max_version
;
4935 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4937 /* Some shaders have to be linked with some other shaders present.
4939 if (!prog
->SeparateShader
) {
4940 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4941 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4942 linker_error(prog
, "Geometry shader must be linked with "
4946 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4947 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4948 linker_error(prog
, "Tessellation evaluation shader must be linked "
4949 "with vertex shader\n");
4952 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4953 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4954 linker_error(prog
, "Tessellation control shader must be linked with "
4959 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4961 * "Linking can fail for [...] any of the following reasons:
4963 * * program contains an object to form a tessellation control
4964 * shader [...] and [...] the program is not separable and
4965 * contains no object to form a tessellation evaluation shader"
4967 * The OpenGL spec is contradictory. It allows linking without a tess
4968 * eval shader, but that can only be used with transform feedback and
4969 * rasterization disabled. However, transform feedback isn't allowed
4970 * with GL_PATCHES, so it can't be used.
4972 * More investigation showed that the idea of transform feedback after
4973 * a tess control shader was dropped, because some hw vendors couldn't
4974 * support tessellation without a tess eval shader, but the linker
4975 * section wasn't updated to reflect that.
4977 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4980 * Do what's reasonable and always require a tess eval shader if a tess
4981 * control shader is present.
4983 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4984 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4985 linker_error(prog
, "Tessellation control shader must be linked with "
4986 "tessellation evaluation shader\n");
4991 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4992 num_shaders
[MESA_SHADER_TESS_CTRL
] == 0) {
4993 linker_error(prog
, "GLSL ES requires non-separable programs "
4994 "containing a tessellation evaluation shader to also "
4995 "be linked with a tessellation control shader\n");
5001 /* Compute shaders have additional restrictions. */
5002 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
5003 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
5004 linker_error(prog
, "Compute shaders may not be linked with any other "
5005 "type of shader\n");
5008 /* Link all shaders for a particular stage and validate the result.
5010 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
5011 if (num_shaders
[stage
] > 0) {
5012 gl_linked_shader
*const sh
=
5013 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
5014 num_shaders
[stage
], false);
5016 if (!prog
->data
->LinkStatus
) {
5018 _mesa_delete_linked_shader(ctx
, sh
);
5023 case MESA_SHADER_VERTEX
:
5024 validate_vertex_shader_executable(prog
, sh
, ctx
);
5026 case MESA_SHADER_TESS_CTRL
:
5027 /* nothing to be done */
5029 case MESA_SHADER_TESS_EVAL
:
5030 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
5032 case MESA_SHADER_GEOMETRY
:
5033 validate_geometry_shader_executable(prog
, sh
, ctx
);
5035 case MESA_SHADER_FRAGMENT
:
5036 validate_fragment_shader_executable(prog
, sh
);
5039 if (!prog
->data
->LinkStatus
) {
5041 _mesa_delete_linked_shader(ctx
, sh
);
5045 prog
->_LinkedShaders
[stage
] = sh
;
5046 prog
->data
->linked_stages
|= 1 << stage
;
5050 /* Here begins the inter-stage linking phase. Some initial validation is
5051 * performed, then locations are assigned for uniforms, attributes, and
5054 cross_validate_uniforms(ctx
, prog
);
5055 if (!prog
->data
->LinkStatus
)
5058 unsigned first
, last
, prev
;
5060 first
= MESA_SHADER_STAGES
;
5063 /* Determine first and last stage. */
5064 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5065 if (!prog
->_LinkedShaders
[i
])
5067 if (first
== MESA_SHADER_STAGES
)
5072 check_explicit_uniform_locations(ctx
, prog
);
5073 link_assign_subroutine_types(prog
);
5074 verify_subroutine_associated_funcs(prog
);
5076 if (!prog
->data
->LinkStatus
)
5079 resize_tes_inputs(ctx
, prog
);
5081 /* Validate the inputs of each stage with the output of the preceding
5085 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
5086 if (prog
->_LinkedShaders
[i
] == NULL
)
5089 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
5090 prog
->_LinkedShaders
[i
]);
5091 if (!prog
->data
->LinkStatus
)
5094 cross_validate_outputs_to_inputs(ctx
, prog
,
5095 prog
->_LinkedShaders
[prev
],
5096 prog
->_LinkedShaders
[i
]);
5097 if (!prog
->data
->LinkStatus
)
5103 /* The cross validation of outputs/inputs above validates explicit locations
5104 * but for SSO programs we need to do this also for the inputs in the
5105 * first stage and outputs of the last stage included in the program, since
5106 * there is no cross validation for these.
5108 if (prog
->SeparateShader
)
5109 validate_sso_explicit_locations(ctx
, prog
,
5110 (gl_shader_stage
) first
,
5111 (gl_shader_stage
) last
);
5113 /* Cross-validate uniform blocks between shader stages */
5114 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
5115 if (!prog
->data
->LinkStatus
)
5118 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5119 if (prog
->_LinkedShaders
[i
] != NULL
)
5120 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
5123 if (prog
->IsES
&& prog
->data
->Version
== 100)
5124 if (!validate_invariant_builtins(prog
,
5125 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
5126 prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]))
5129 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
5130 * it before optimization because we want most of the checks to get
5131 * dropped thanks to constant propagation.
5133 * This rule also applies to GLSL ES 3.00.
5135 if (max_version
>= (prog
->IsES
? 300 : 130)) {
5136 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
5138 lower_discard_flow(sh
->ir
);
5142 if (prog
->SeparateShader
)
5143 disable_varying_optimizations_for_sso(prog
);
5146 if (!interstage_cross_validate_uniform_blocks(prog
, false))
5150 if (!interstage_cross_validate_uniform_blocks(prog
, true))
5153 /* Do common optimization before assigning storage for attributes,
5154 * uniforms, and varyings. Later optimization could possibly make
5155 * some of that unused.
5157 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5158 if (prog
->_LinkedShaders
[i
] == NULL
)
5161 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
5162 if (!prog
->data
->LinkStatus
)
5165 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
5166 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
5169 if (ctx
->Const
.LowerTessLevel
) {
5170 lower_tess_level(prog
->_LinkedShaders
[i
]);
5173 /* Section 13.46 (Vertex Attribute Aliasing) of the OpenGL ES 3.2
5174 * specification says:
5176 * "In general, the behavior of GLSL ES should not depend on compiler
5177 * optimizations which might be implementation-dependent. Name matching
5178 * rules in most languages, including C++ from which GLSL ES is derived,
5179 * are based on declarations rather than use.
5181 * RESOLUTION: The existence of aliasing is determined by declarations
5182 * present after preprocessing."
5184 * Because of this rule, we do a 'dry-run' of attribute assignment for
5185 * vertex shader inputs here.
5187 if (prog
->IsES
&& i
== MESA_SHADER_VERTEX
) {
5188 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
5189 MESA_SHADER_VERTEX
, false)) {
5194 /* Call opts before lowering const arrays to uniforms so we can const
5195 * propagate any elements accessed directly.
5197 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5199 /* Call opts after lowering const arrays to copy propagate things. */
5200 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
5201 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5203 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
5206 /* Validation for special cases where we allow sampler array indexing
5207 * with loop induction variable. This check emits a warning or error
5208 * depending if backend can handle dynamic indexing.
5210 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
5211 (prog
->IsES
&& prog
->data
->Version
< 300)) {
5212 if (!validate_sampler_array_indexing(ctx
, prog
))
5216 /* Check and validate stream emissions in geometry shaders */
5217 validate_geometry_shader_emissions(ctx
, prog
);
5219 store_fragdepth_layout(prog
);
5221 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
5224 /* Linking varyings can cause some extra, useless swizzles to be generated
5225 * due to packing and unpacking.
5227 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5228 if (prog
->_LinkedShaders
[i
] == NULL
)
5231 optimize_swizzles(prog
->_LinkedShaders
[i
]->ir
);
5234 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
5235 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
5236 * anything about shader linking when one of the shaders (vertex or
5237 * fragment shader) is absent. So, the extension shouldn't change the
5238 * behavior specified in GLSL specification.
5240 * From OpenGL ES 3.1 specification (7.3 Program Objects):
5241 * "Linking can fail for a variety of reasons as specified in the
5242 * OpenGL ES Shading Language Specification, as well as any of the
5243 * following reasons:
5247 * * program contains objects to form either a vertex shader or
5248 * fragment shader, and program is not separable, and does not
5249 * contain objects to form both a vertex shader and fragment
5252 * However, the only scenario in 3.1+ where we don't require them both is
5253 * when we have a compute shader. For example:
5255 * - No shaders is a link error.
5256 * - Geom or Tess without a Vertex shader is a link error which means we
5257 * always require a Vertex shader and hence a Fragment shader.
5258 * - Finally a Compute shader linked with any other stage is a link error.
5260 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
5261 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
5262 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
5263 linker_error(prog
, "program lacks a vertex shader\n");
5264 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5265 linker_error(prog
, "program lacks a fragment shader\n");
5270 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5271 free(shader_list
[i
]);
5272 if (prog
->_LinkedShaders
[i
] == NULL
)
5275 /* Do a final validation step to make sure that the IR wasn't
5276 * invalidated by any modifications performed after intrastage linking.
5278 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5280 /* Retain any live IR, but trash the rest. */
5281 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5283 /* The symbol table in the linked shaders may contain references to
5284 * variables that were removed (e.g., unused uniforms). Since it may
5285 * contain junk, there is no possible valid use. Delete it and set the
5288 delete prog
->_LinkedShaders
[i
]->symbols
;
5289 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5292 ralloc_free(mem_ctx
);