2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
68 #include "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/prog_instruction.h"
75 #include "program/program.h"
76 #include "util/mesa-sha1.h"
78 #include "string_to_uint_map.h"
80 #include "link_varyings.h"
81 #include "ir_optimization.h"
82 #include "ir_rvalue_visitor.h"
83 #include "ir_uniform.h"
84 #include "builtin_functions.h"
85 #include "shader_cache.h"
87 #include "main/shaderobj.h"
88 #include "main/enums.h"
93 struct find_variable
{
97 find_variable(const char *name
) : name(name
), found(false) {}
101 * Visitor that determines whether or not a variable is ever written.
103 * Use \ref find_assignments for convenience.
105 class find_assignment_visitor
: public ir_hierarchical_visitor
{
107 find_assignment_visitor(unsigned num_vars
,
108 find_variable
* const *vars
)
109 : num_variables(num_vars
), num_found(0), variables(vars
)
113 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
115 ir_variable
*const var
= ir
->lhs
->variable_referenced();
117 return check_variable_name(var
->name
);
120 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
122 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
123 actual_node
, &ir
->actual_parameters
) {
124 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
125 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
127 if (sig_param
->data
.mode
== ir_var_function_out
||
128 sig_param
->data
.mode
== ir_var_function_inout
) {
129 ir_variable
*var
= param_rval
->variable_referenced();
130 if (var
&& check_variable_name(var
->name
) == visit_stop
)
135 if (ir
->return_deref
!= NULL
) {
136 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
138 if (check_variable_name(var
->name
) == visit_stop
)
142 return visit_continue_with_parent
;
146 ir_visitor_status
check_variable_name(const char *name
)
148 for (unsigned i
= 0; i
< num_variables
; ++i
) {
149 if (strcmp(variables
[i
]->name
, name
) == 0) {
150 if (!variables
[i
]->found
) {
151 variables
[i
]->found
= true;
153 assert(num_found
< num_variables
);
154 if (++num_found
== num_variables
)
161 return visit_continue_with_parent
;
165 unsigned num_variables
; /**< Number of variables to find */
166 unsigned num_found
; /**< Number of variables already found */
167 find_variable
* const *variables
; /**< Variables to find */
171 * Determine whether or not any of NULL-terminated list of variables is ever
175 find_assignments(exec_list
*ir
, find_variable
* const *vars
)
177 unsigned num_variables
= 0;
179 for (find_variable
* const *v
= vars
; *v
; ++v
)
182 find_assignment_visitor
visitor(num_variables
, vars
);
187 * Determine whether or not the given variable is ever written to.
190 find_assignments(exec_list
*ir
, find_variable
*var
)
192 find_assignment_visitor
visitor(1, &var
);
197 * Visitor that determines whether or not a variable is ever read.
199 class find_deref_visitor
: public ir_hierarchical_visitor
{
201 find_deref_visitor(const char *name
)
202 : name(name
), found(false)
207 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
209 if (strcmp(this->name
, ir
->var
->name
) == 0) {
214 return visit_continue
;
217 bool variable_found() const
223 const char *name
; /**< Find writes to a variable with this name. */
224 bool found
; /**< Was a write to the variable found? */
229 * A visitor helper that provides methods for updating the types of
230 * ir_dereferences. Classes that update variable types (say, updating
231 * array sizes) will want to use this so that dereference types stay in sync.
233 class deref_type_updater
: public ir_hierarchical_visitor
{
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
243 const glsl_type
*const vt
= ir
->array
->type
;
245 ir
->type
= vt
->fields
.array
;
246 return visit_continue
;
249 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
251 ir
->type
= ir
->record
->type
->fields
.structure
[ir
->field_idx
].type
;
252 return visit_continue
;
257 class array_resize_visitor
: public deref_type_updater
{
259 unsigned num_vertices
;
260 gl_shader_program
*prog
;
261 gl_shader_stage stage
;
263 array_resize_visitor(unsigned num_vertices
,
264 gl_shader_program
*prog
,
265 gl_shader_stage stage
)
267 this->num_vertices
= num_vertices
;
272 virtual ~array_resize_visitor()
277 virtual ir_visitor_status
visit(ir_variable
*var
)
279 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
281 return visit_continue
;
283 unsigned size
= var
->type
->length
;
285 if (stage
== MESA_SHADER_GEOMETRY
) {
286 /* Generate a link error if the shader has declared this array with
289 if (!var
->data
.implicit_sized_array
&&
290 size
&& size
!= this->num_vertices
) {
291 linker_error(this->prog
, "size of array %s declared as %u, "
292 "but number of input vertices is %u\n",
293 var
->name
, size
, this->num_vertices
);
294 return visit_continue
;
297 /* Generate a link error if the shader attempts to access an input
298 * array using an index too large for its actual size assigned at
301 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
302 linker_error(this->prog
, "%s shader accesses element %i of "
303 "%s, but only %i input vertices\n",
304 _mesa_shader_stage_to_string(this->stage
),
305 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
306 return visit_continue
;
310 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
312 var
->data
.max_array_access
= this->num_vertices
- 1;
314 return visit_continue
;
319 * Visitor that determines the highest stream id to which a (geometry) shader
320 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
322 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
324 find_emit_vertex_visitor(int max_allowed
)
325 : max_stream_allowed(max_allowed
),
326 invalid_stream_id(0),
327 invalid_stream_id_from_emit_vertex(false),
328 end_primitive_found(false),
329 uses_non_zero_stream(false)
334 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
336 int stream_id
= ir
->stream_id();
339 invalid_stream_id
= stream_id
;
340 invalid_stream_id_from_emit_vertex
= true;
344 if (stream_id
> max_stream_allowed
) {
345 invalid_stream_id
= stream_id
;
346 invalid_stream_id_from_emit_vertex
= true;
351 uses_non_zero_stream
= true;
353 return visit_continue
;
356 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
358 end_primitive_found
= true;
360 int stream_id
= ir
->stream_id();
363 invalid_stream_id
= stream_id
;
364 invalid_stream_id_from_emit_vertex
= false;
368 if (stream_id
> max_stream_allowed
) {
369 invalid_stream_id
= stream_id
;
370 invalid_stream_id_from_emit_vertex
= false;
375 uses_non_zero_stream
= true;
377 return visit_continue
;
382 return invalid_stream_id
!= 0;
385 const char *error_func()
387 return invalid_stream_id_from_emit_vertex
?
388 "EmitStreamVertex" : "EndStreamPrimitive";
393 return invalid_stream_id
;
398 return uses_non_zero_stream
;
401 bool uses_end_primitive()
403 return end_primitive_found
;
407 int max_stream_allowed
;
408 int invalid_stream_id
;
409 bool invalid_stream_id_from_emit_vertex
;
410 bool end_primitive_found
;
411 bool uses_non_zero_stream
;
414 /* Class that finds array derefs and check if indexes are dynamic. */
415 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
418 dynamic_sampler_array_indexing_visitor() :
419 dynamic_sampler_array_indexing(false)
423 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
425 if (!ir
->variable_referenced())
426 return visit_continue
;
428 if (!ir
->variable_referenced()->type
->contains_sampler())
429 return visit_continue
;
431 if (!ir
->array_index
->constant_expression_value(ralloc_parent(ir
))) {
432 dynamic_sampler_array_indexing
= true;
435 return visit_continue
;
438 bool uses_dynamic_sampler_array_indexing()
440 return dynamic_sampler_array_indexing
;
444 bool dynamic_sampler_array_indexing
;
447 } /* anonymous namespace */
450 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
454 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
456 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
459 prog
->data
->LinkStatus
= LINKING_FAILURE
;
464 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
468 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
470 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
477 * Given a string identifying a program resource, break it into a base name
478 * and an optional array index in square brackets.
480 * If an array index is present, \c out_base_name_end is set to point to the
481 * "[" that precedes the array index, and the array index itself is returned
484 * If no array index is present (or if the array index is negative or
485 * mal-formed), \c out_base_name_end, is set to point to the null terminator
486 * at the end of the input string, and -1 is returned.
488 * Only the final array index is parsed; if the string contains other array
489 * indices (or structure field accesses), they are left in the base name.
491 * No attempt is made to check that the base name is properly formed;
492 * typically the caller will look up the base name in a hash table, so
493 * ill-formed base names simply turn into hash table lookup failures.
496 parse_program_resource_name(const GLchar
*name
,
497 const GLchar
**out_base_name_end
)
499 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
501 * "When an integer array element or block instance number is part of
502 * the name string, it will be specified in decimal form without a "+"
503 * or "-" sign or any extra leading zeroes. Additionally, the name
504 * string will not include white space anywhere in the string."
507 const size_t len
= strlen(name
);
508 *out_base_name_end
= name
+ len
;
510 if (len
== 0 || name
[len
-1] != ']')
513 /* Walk backwards over the string looking for a non-digit character. This
514 * had better be the opening bracket for an array index.
516 * Initially, i specifies the location of the ']'. Since the string may
517 * contain only the ']' charcater, walk backwards very carefully.
520 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
523 if ((i
== 0) || name
[i
-1] != '[')
526 long array_index
= strtol(&name
[i
], NULL
, 10);
530 /* Check for leading zero */
531 if (name
[i
] == '0' && name
[i
+1] != ']')
534 *out_base_name_end
= name
+ (i
- 1);
540 link_invalidate_variable_locations(exec_list
*ir
)
542 foreach_in_list(ir_instruction
, node
, ir
) {
543 ir_variable
*const var
= node
->as_variable();
548 /* Only assign locations for variables that lack an explicit location.
549 * Explicit locations are set for all built-in variables, generic vertex
550 * shader inputs (via layout(location=...)), and generic fragment shader
551 * outputs (also via layout(location=...)).
553 if (!var
->data
.explicit_location
) {
554 var
->data
.location
= -1;
555 var
->data
.location_frac
= 0;
558 /* ir_variable::is_unmatched_generic_inout is used by the linker while
559 * connecting outputs from one stage to inputs of the next stage.
561 if (var
->data
.explicit_location
&&
562 var
->data
.location
< VARYING_SLOT_VAR0
) {
563 var
->data
.is_unmatched_generic_inout
= 0;
565 var
->data
.is_unmatched_generic_inout
= 1;
572 * Set clip_distance_array_size based and cull_distance_array_size on the given
575 * Also check for errors based on incorrect usage of gl_ClipVertex and
576 * gl_ClipDistance and gl_CullDistance.
577 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
578 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
580 * Return false if an error was reported.
583 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
584 struct gl_linked_shader
*shader
,
585 struct gl_context
*ctx
,
586 GLuint
*clip_distance_array_size
,
587 GLuint
*cull_distance_array_size
)
589 *clip_distance_array_size
= 0;
590 *cull_distance_array_size
= 0;
592 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
593 /* From section 7.1 (Vertex Shader Special Variables) of the
596 * "It is an error for a shader to statically write both
597 * gl_ClipVertex and gl_ClipDistance."
599 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
600 * gl_ClipVertex nor gl_ClipDistance. However with
601 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
603 find_variable
gl_ClipDistance("gl_ClipDistance");
604 find_variable
gl_CullDistance("gl_CullDistance");
605 find_variable
gl_ClipVertex("gl_ClipVertex");
606 find_variable
* const variables
[] = {
609 !prog
->IsES
? &gl_ClipVertex
: NULL
,
612 find_assignments(shader
->ir
, variables
);
614 /* From the ARB_cull_distance spec:
616 * It is a compile-time or link-time error for the set of shaders forming
617 * a program to statically read or write both gl_ClipVertex and either
618 * gl_ClipDistance or gl_CullDistance.
620 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
624 if (gl_ClipVertex
.found
&& gl_ClipDistance
.found
) {
625 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
626 "and `gl_ClipDistance'\n",
627 _mesa_shader_stage_to_string(shader
->Stage
));
630 if (gl_ClipVertex
.found
&& gl_CullDistance
.found
) {
631 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
632 "and `gl_CullDistance'\n",
633 _mesa_shader_stage_to_string(shader
->Stage
));
638 if (gl_ClipDistance
.found
) {
639 ir_variable
*clip_distance_var
=
640 shader
->symbols
->get_variable("gl_ClipDistance");
641 assert(clip_distance_var
);
642 *clip_distance_array_size
= clip_distance_var
->type
->length
;
644 if (gl_CullDistance
.found
) {
645 ir_variable
*cull_distance_var
=
646 shader
->symbols
->get_variable("gl_CullDistance");
647 assert(cull_distance_var
);
648 *cull_distance_array_size
= cull_distance_var
->type
->length
;
650 /* From the ARB_cull_distance spec:
652 * It is a compile-time or link-time error for the set of shaders forming
653 * a program to have the sum of the sizes of the gl_ClipDistance and
654 * gl_CullDistance arrays to be larger than
655 * gl_MaxCombinedClipAndCullDistances.
657 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
658 ctx
->Const
.MaxClipPlanes
) {
659 linker_error(prog
, "%s shader: the combined size of "
660 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
662 "gl_MaxCombinedClipAndCullDistances (%u)",
663 _mesa_shader_stage_to_string(shader
->Stage
),
664 ctx
->Const
.MaxClipPlanes
);
671 * Verify that a vertex shader executable meets all semantic requirements.
673 * Also sets info.clip_distance_array_size and
674 * info.cull_distance_array_size as a side effect.
676 * \param shader Vertex shader executable to be verified
679 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
680 struct gl_linked_shader
*shader
,
681 struct gl_context
*ctx
)
686 /* From the GLSL 1.10 spec, page 48:
688 * "The variable gl_Position is available only in the vertex
689 * language and is intended for writing the homogeneous vertex
690 * position. All executions of a well-formed vertex shader
691 * executable must write a value into this variable. [...] The
692 * variable gl_Position is available only in the vertex
693 * language and is intended for writing the homogeneous vertex
694 * position. All executions of a well-formed vertex shader
695 * executable must write a value into this variable."
697 * while in GLSL 1.40 this text is changed to:
699 * "The variable gl_Position is available only in the vertex
700 * language and is intended for writing the homogeneous vertex
701 * position. It can be written at any time during shader
702 * execution. It may also be read back by a vertex shader
703 * after being written. This value will be used by primitive
704 * assembly, clipping, culling, and other fixed functionality
705 * operations, if present, that operate on primitives after
706 * vertex processing has occurred. Its value is undefined if
707 * the vertex shader executable does not write gl_Position."
709 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
710 * gl_Position is not an error.
712 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
713 find_variable
gl_Position("gl_Position");
714 find_assignments(shader
->ir
, &gl_Position
);
715 if (!gl_Position
.found
) {
718 "vertex shader does not write to `gl_Position'. "
719 "Its value is undefined. \n");
722 "vertex shader does not write to `gl_Position'. \n");
728 analyze_clip_cull_usage(prog
, shader
, ctx
,
729 &shader
->Program
->info
.clip_distance_array_size
,
730 &shader
->Program
->info
.cull_distance_array_size
);
734 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
735 struct gl_linked_shader
*shader
,
736 struct gl_context
*ctx
)
741 analyze_clip_cull_usage(prog
, shader
, ctx
,
742 &shader
->Program
->info
.clip_distance_array_size
,
743 &shader
->Program
->info
.cull_distance_array_size
);
748 * Verify that a fragment shader executable meets all semantic requirements
750 * \param shader Fragment shader executable to be verified
753 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
754 struct gl_linked_shader
*shader
)
759 find_variable
gl_FragColor("gl_FragColor");
760 find_variable
gl_FragData("gl_FragData");
761 find_variable
* const variables
[] = { &gl_FragColor
, &gl_FragData
, NULL
};
762 find_assignments(shader
->ir
, variables
);
764 if (gl_FragColor
.found
&& gl_FragData
.found
) {
765 linker_error(prog
, "fragment shader writes to both "
766 "`gl_FragColor' and `gl_FragData'\n");
771 * Verify that a geometry shader executable meets all semantic requirements
773 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
774 * info.cull_distance_array_size as a side effect.
776 * \param shader Geometry shader executable to be verified
779 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
780 struct gl_linked_shader
*shader
,
781 struct gl_context
*ctx
)
786 unsigned num_vertices
=
787 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
788 prog
->Geom
.VerticesIn
= num_vertices
;
790 analyze_clip_cull_usage(prog
, shader
, ctx
,
791 &shader
->Program
->info
.clip_distance_array_size
,
792 &shader
->Program
->info
.cull_distance_array_size
);
796 * Check if geometry shaders emit to non-zero streams and do corresponding
800 validate_geometry_shader_emissions(struct gl_context
*ctx
,
801 struct gl_shader_program
*prog
)
803 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
806 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
807 emit_vertex
.run(sh
->ir
);
808 if (emit_vertex
.error()) {
809 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
810 "stream parameter are in the range [0, %d].\n",
811 emit_vertex
.error_func(),
812 emit_vertex
.error_stream(),
813 ctx
->Const
.MaxVertexStreams
- 1);
815 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
816 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
818 /* From the ARB_gpu_shader5 spec:
820 * "Multiple vertex streams are supported only if the output primitive
821 * type is declared to be "points". A program will fail to link if it
822 * contains a geometry shader calling EmitStreamVertex() or
823 * EndStreamPrimitive() if its output primitive type is not "points".
825 * However, in the same spec:
827 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
828 * with <stream> set to zero."
832 * "The function EndPrimitive() is equivalent to calling
833 * EndStreamPrimitive() with <stream> set to zero."
835 * Since we can call EmitVertex() and EndPrimitive() when we output
836 * primitives other than points, calling EmitStreamVertex(0) or
837 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
838 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
839 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
842 if (prog
->Geom
.UsesStreams
&&
843 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
844 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
845 "with n>0 requires point output\n");
851 validate_intrastage_arrays(struct gl_shader_program
*prog
,
852 ir_variable
*const var
,
853 ir_variable
*const existing
)
855 /* Consider the types to be "the same" if both types are arrays
856 * of the same type and one of the arrays is implicitly sized.
857 * In addition, set the type of the linked variable to the
858 * explicitly sized array.
860 if (var
->type
->is_array() && existing
->type
->is_array()) {
861 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
862 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
863 if (var
->type
->length
!= 0) {
864 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
865 linker_error(prog
, "%s `%s' declared as type "
866 "`%s' but outermost dimension has an index"
869 var
->name
, var
->type
->name
,
870 existing
->data
.max_array_access
);
872 existing
->type
= var
->type
;
874 } else if (existing
->type
->length
!= 0) {
875 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
876 !existing
->data
.from_ssbo_unsized_array
) {
877 linker_error(prog
, "%s `%s' declared as type "
878 "`%s' but outermost dimension has an index"
881 var
->name
, existing
->type
->name
,
882 var
->data
.max_array_access
);
893 * Perform validation of global variables used across multiple shaders
896 cross_validate_globals(struct gl_shader_program
*prog
,
897 struct exec_list
*ir
, glsl_symbol_table
*variables
,
900 foreach_in_list(ir_instruction
, node
, ir
) {
901 ir_variable
*const var
= node
->as_variable();
906 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
909 /* don't cross validate subroutine uniforms */
910 if (var
->type
->contains_subroutine())
913 /* Don't cross validate interface instances. These are only relevant
914 * inside a shader. The cross validation is done at the Interface Block
917 if (var
->is_interface_instance())
920 /* Don't cross validate temporaries that are at global scope. These
921 * will eventually get pulled into the shaders 'main'.
923 if (var
->data
.mode
== ir_var_temporary
)
926 /* If a global with this name has already been seen, verify that the
927 * new instance has the same type. In addition, if the globals have
928 * initializers, the values of the initializers must be the same.
930 ir_variable
*const existing
= variables
->get_variable(var
->name
);
931 if (existing
!= NULL
) {
932 /* Check if types match. */
933 if (var
->type
!= existing
->type
) {
934 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
935 /* If it is an unsized array in a Shader Storage Block,
936 * two different shaders can access to different elements.
937 * Because of that, they might be converted to different
938 * sized arrays, then check that they are compatible but
939 * ignore the array size.
941 if (!(var
->data
.mode
== ir_var_shader_storage
&&
942 var
->data
.from_ssbo_unsized_array
&&
943 existing
->data
.mode
== ir_var_shader_storage
&&
944 existing
->data
.from_ssbo_unsized_array
&&
945 var
->type
->gl_type
== existing
->type
->gl_type
)) {
946 linker_error(prog
, "%s `%s' declared as type "
947 "`%s' and type `%s'\n",
949 var
->name
, var
->type
->name
,
950 existing
->type
->name
);
956 if (var
->data
.explicit_location
) {
957 if (existing
->data
.explicit_location
958 && (var
->data
.location
!= existing
->data
.location
)) {
959 linker_error(prog
, "explicit locations for %s "
960 "`%s' have differing values\n",
961 mode_string(var
), var
->name
);
965 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
966 linker_error(prog
, "explicit components for %s `%s' have "
967 "differing values\n", mode_string(var
), var
->name
);
971 existing
->data
.location
= var
->data
.location
;
972 existing
->data
.explicit_location
= true;
974 /* Check if uniform with implicit location was marked explicit
975 * by earlier shader stage. If so, mark it explicit in this stage
976 * too to make sure later processing does not treat it as
979 if (existing
->data
.explicit_location
) {
980 var
->data
.location
= existing
->data
.location
;
981 var
->data
.explicit_location
= true;
985 /* From the GLSL 4.20 specification:
986 * "A link error will result if two compilation units in a program
987 * specify different integer-constant bindings for the same
988 * opaque-uniform name. However, it is not an error to specify a
989 * binding on some but not all declarations for the same name"
991 if (var
->data
.explicit_binding
) {
992 if (existing
->data
.explicit_binding
&&
993 var
->data
.binding
!= existing
->data
.binding
) {
994 linker_error(prog
, "explicit bindings for %s "
995 "`%s' have differing values\n",
996 mode_string(var
), var
->name
);
1000 existing
->data
.binding
= var
->data
.binding
;
1001 existing
->data
.explicit_binding
= true;
1004 if (var
->type
->contains_atomic() &&
1005 var
->data
.offset
!= existing
->data
.offset
) {
1006 linker_error(prog
, "offset specifications for %s "
1007 "`%s' have differing values\n",
1008 mode_string(var
), var
->name
);
1012 /* Validate layout qualifiers for gl_FragDepth.
1014 * From the AMD/ARB_conservative_depth specs:
1016 * "If gl_FragDepth is redeclared in any fragment shader in a
1017 * program, it must be redeclared in all fragment shaders in
1018 * that program that have static assignments to
1019 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1020 * fragment shaders in a single program must have the same set
1023 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1024 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1025 bool layout_differs
=
1026 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1028 if (layout_declared
&& layout_differs
) {
1030 "All redeclarations of gl_FragDepth in all "
1031 "fragment shaders in a single program must have "
1032 "the same set of qualifiers.\n");
1035 if (var
->data
.used
&& layout_differs
) {
1037 "If gl_FragDepth is redeclared with a layout "
1038 "qualifier in any fragment shader, it must be "
1039 "redeclared with the same layout qualifier in "
1040 "all fragment shaders that have assignments to "
1045 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1047 * "If a shared global has multiple initializers, the
1048 * initializers must all be constant expressions, and they
1049 * must all have the same value. Otherwise, a link error will
1050 * result. (A shared global having only one initializer does
1051 * not require that initializer to be a constant expression.)"
1053 * Previous to 4.20 the GLSL spec simply said that initializers
1054 * must have the same value. In this case of non-constant
1055 * initializers, this was impossible to determine. As a result,
1056 * no vendor actually implemented that behavior. The 4.20
1057 * behavior matches the implemented behavior of at least one other
1058 * vendor, so we'll implement that for all GLSL versions.
1060 if (var
->constant_initializer
!= NULL
) {
1061 if (existing
->constant_initializer
!= NULL
) {
1062 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1063 linker_error(prog
, "initializers for %s "
1064 "`%s' have differing values\n",
1065 mode_string(var
), var
->name
);
1069 /* If the first-seen instance of a particular uniform did
1070 * not have an initializer but a later instance does,
1071 * replace the former with the later.
1073 variables
->replace_variable(existing
->name
, var
);
1077 if (var
->data
.has_initializer
) {
1078 if (existing
->data
.has_initializer
1079 && (var
->constant_initializer
== NULL
1080 || existing
->constant_initializer
== NULL
)) {
1082 "shared global variable `%s' has multiple "
1083 "non-constant initializers.\n",
1089 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1090 linker_error(prog
, "declarations for %s `%s' have "
1091 "mismatching invariant qualifiers\n",
1092 mode_string(var
), var
->name
);
1095 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1096 linker_error(prog
, "declarations for %s `%s' have "
1097 "mismatching centroid qualifiers\n",
1098 mode_string(var
), var
->name
);
1101 if (existing
->data
.sample
!= var
->data
.sample
) {
1102 linker_error(prog
, "declarations for %s `%s` have "
1103 "mismatching sample qualifiers\n",
1104 mode_string(var
), var
->name
);
1107 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1108 linker_error(prog
, "declarations for %s `%s` have "
1109 "mismatching image format qualifiers\n",
1110 mode_string(var
), var
->name
);
1114 /* In OpenGL GLSL 4.20 spec, section 4.3.9, page 57:
1116 * "It is a link-time error if any particular shader interface
1119 * - two different blocks, each having no instance name, and each
1120 * having a member of the same name, or
1122 * - a variable outside a block, and a block with no instance name,
1123 * where the variable has the same name as a member in the block."
1125 if (var
->data
.mode
== existing
->data
.mode
&&
1126 var
->get_interface_type() != existing
->get_interface_type()) {
1127 linker_error(prog
, "declarations for %s `%s` are in "
1129 mode_string(var
), var
->name
,
1130 existing
->get_interface_type() ?
1131 existing
->get_interface_type()->name
: "outside a block",
1132 var
->get_interface_type() ?
1133 var
->get_interface_type()->name
: "outside a block");
1137 /* Only in GLSL ES 3.10, the precision qualifier should not match
1138 * between block members defined in matched block names within a
1141 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1142 * member should match.
1144 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1145 !var
->get_interface_type()) &&
1146 existing
->data
.precision
!= var
->data
.precision
) {
1147 if ((existing
->data
.used
&& var
->data
.used
) || prog
->data
->Version
>= 300) {
1148 linker_error(prog
, "declarations for %s `%s` have "
1149 "mismatching precision qualifiers\n",
1150 mode_string(var
), var
->name
);
1153 linker_warning(prog
, "declarations for %s `%s` have "
1154 "mismatching precision qualifiers\n",
1155 mode_string(var
), var
->name
);
1159 variables
->add_variable(var
);
1165 * Perform validation of uniforms used across multiple shader stages
1168 cross_validate_uniforms(struct gl_shader_program
*prog
)
1170 glsl_symbol_table variables
;
1171 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1172 if (prog
->_LinkedShaders
[i
] == NULL
)
1175 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1181 * Accumulates the array of buffer blocks and checks that all definitions of
1182 * blocks agree on their contents.
1185 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1188 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1189 struct gl_uniform_block
*blks
= NULL
;
1190 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1191 &prog
->data
->NumUniformBlocks
;
1193 unsigned max_num_buffer_blocks
= 0;
1194 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1195 if (prog
->_LinkedShaders
[i
]) {
1196 if (validate_ssbo
) {
1197 max_num_buffer_blocks
+=
1198 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1200 max_num_buffer_blocks
+=
1201 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1206 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1207 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1209 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1210 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1211 InterfaceBlockStageIndex
[i
][j
] = -1;
1216 unsigned sh_num_blocks
;
1217 struct gl_uniform_block
**sh_blks
;
1218 if (validate_ssbo
) {
1219 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1220 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1222 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1223 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1226 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1227 int index
= link_cross_validate_uniform_block(prog
->data
, &blks
,
1228 num_blks
, sh_blks
[j
]);
1231 linker_error(prog
, "buffer block `%s' has mismatching "
1232 "definitions\n", sh_blks
[j
]->Name
);
1234 for (unsigned k
= 0; k
<= i
; k
++) {
1235 delete[] InterfaceBlockStageIndex
[k
];
1238 /* Reset the block count. This will help avoid various segfaults
1239 * from api calls that assume the array exists due to the count
1246 InterfaceBlockStageIndex
[i
][index
] = j
;
1250 /* Update per stage block pointers to point to the program list.
1251 * FIXME: We should be able to free the per stage blocks here.
1253 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1254 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1255 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1257 if (stage_index
!= -1) {
1258 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1260 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1261 sh
->Program
->sh
.ShaderStorageBlocks
:
1262 sh
->Program
->sh
.UniformBlocks
;
1264 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1265 sh_blks
[stage_index
] = &blks
[j
];
1270 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1271 delete[] InterfaceBlockStageIndex
[i
];
1275 prog
->data
->ShaderStorageBlocks
= blks
;
1277 prog
->data
->UniformBlocks
= blks
;
1284 * Populates a shaders symbol table with all global declarations
1287 populate_symbol_table(gl_linked_shader
*sh
, glsl_symbol_table
*symbols
)
1289 sh
->symbols
= new(sh
) glsl_symbol_table
;
1291 _mesa_glsl_copy_symbols_from_table(sh
->ir
, symbols
, sh
->symbols
);
1296 * Remap variables referenced in an instruction tree
1298 * This is used when instruction trees are cloned from one shader and placed in
1299 * another. These trees will contain references to \c ir_variable nodes that
1300 * do not exist in the target shader. This function finds these \c ir_variable
1301 * references and replaces the references with matching variables in the target
1304 * If there is no matching variable in the target shader, a clone of the
1305 * \c ir_variable is made and added to the target shader. The new variable is
1306 * added to \b both the instruction stream and the symbol table.
1308 * \param inst IR tree that is to be processed.
1309 * \param symbols Symbol table containing global scope symbols in the
1311 * \param instructions Instruction stream where new variable declarations
1315 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1318 class remap_visitor
: public ir_hierarchical_visitor
{
1320 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1322 this->target
= target
;
1323 this->symbols
= target
->symbols
;
1324 this->instructions
= target
->ir
;
1325 this->temps
= temps
;
1328 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1330 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1331 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1332 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1334 assert(var
!= NULL
);
1336 return visit_continue
;
1339 ir_variable
*const existing
=
1340 this->symbols
->get_variable(ir
->var
->name
);
1341 if (existing
!= NULL
)
1344 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1346 this->symbols
->add_variable(copy
);
1347 this->instructions
->push_head(copy
);
1351 return visit_continue
;
1355 struct gl_linked_shader
*target
;
1356 glsl_symbol_table
*symbols
;
1357 exec_list
*instructions
;
1361 remap_visitor
v(target
, temps
);
1368 * Move non-declarations from one instruction stream to another
1370 * The intended usage pattern of this function is to pass the pointer to the
1371 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1372 * pointer) for \c last and \c false for \c make_copies on the first
1373 * call. Successive calls pass the return value of the previous call for
1374 * \c last and \c true for \c make_copies.
1376 * \param instructions Source instruction stream
1377 * \param last Instruction after which new instructions should be
1378 * inserted in the target instruction stream
1379 * \param make_copies Flag selecting whether instructions in \c instructions
1380 * should be copied (via \c ir_instruction::clone) into the
1381 * target list or moved.
1384 * The new "last" instruction in the target instruction stream. This pointer
1385 * is suitable for use as the \c last parameter of a later call to this
1389 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1390 bool make_copies
, gl_linked_shader
*target
)
1392 hash_table
*temps
= NULL
;
1395 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1396 _mesa_key_pointer_equal
);
1398 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1399 if (inst
->as_function())
1402 ir_variable
*var
= inst
->as_variable();
1403 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1406 assert(inst
->as_assignment()
1408 || inst
->as_if() /* for initializers with the ?: operator */
1409 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1412 inst
= inst
->clone(target
, NULL
);
1415 _mesa_hash_table_insert(temps
, var
, inst
);
1417 remap_variables(inst
, target
, temps
);
1422 last
->insert_after(inst
);
1427 _mesa_hash_table_destroy(temps
, NULL
);
1434 * This class is only used in link_intrastage_shaders() below but declaring
1435 * it inside that function leads to compiler warnings with some versions of
1438 class array_sizing_visitor
: public deref_type_updater
{
1440 array_sizing_visitor()
1441 : mem_ctx(ralloc_context(NULL
)),
1442 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1443 _mesa_key_pointer_equal
))
1447 ~array_sizing_visitor()
1449 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1450 ralloc_free(this->mem_ctx
);
1453 virtual ir_visitor_status
visit(ir_variable
*var
)
1455 const glsl_type
*type_without_array
;
1456 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1457 fixup_type(&var
->type
, var
->data
.max_array_access
,
1458 var
->data
.from_ssbo_unsized_array
,
1459 &implicit_sized_array
);
1460 var
->data
.implicit_sized_array
= implicit_sized_array
;
1461 type_without_array
= var
->type
->without_array();
1462 if (var
->type
->is_interface()) {
1463 if (interface_contains_unsized_arrays(var
->type
)) {
1464 const glsl_type
*new_type
=
1465 resize_interface_members(var
->type
,
1466 var
->get_max_ifc_array_access(),
1467 var
->is_in_shader_storage_block());
1468 var
->type
= new_type
;
1469 var
->change_interface_type(new_type
);
1471 } else if (type_without_array
->is_interface()) {
1472 if (interface_contains_unsized_arrays(type_without_array
)) {
1473 const glsl_type
*new_type
=
1474 resize_interface_members(type_without_array
,
1475 var
->get_max_ifc_array_access(),
1476 var
->is_in_shader_storage_block());
1477 var
->change_interface_type(new_type
);
1478 var
->type
= update_interface_members_array(var
->type
, new_type
);
1480 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1481 /* Store a pointer to the variable in the unnamed_interfaces
1485 _mesa_hash_table_search(this->unnamed_interfaces
,
1488 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1490 if (interface_vars
== NULL
) {
1491 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1493 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1496 unsigned index
= ifc_type
->field_index(var
->name
);
1497 assert(index
< ifc_type
->length
);
1498 assert(interface_vars
[index
] == NULL
);
1499 interface_vars
[index
] = var
;
1501 return visit_continue
;
1505 * For each unnamed interface block that was discovered while running the
1506 * visitor, adjust the interface type to reflect the newly assigned array
1507 * sizes, and fix up the ir_variable nodes to point to the new interface
1510 void fixup_unnamed_interface_types()
1512 hash_table_call_foreach(this->unnamed_interfaces
,
1513 fixup_unnamed_interface_type
, NULL
);
1518 * If the type pointed to by \c type represents an unsized array, replace
1519 * it with a sized array whose size is determined by max_array_access.
1521 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1522 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1524 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1525 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1526 max_array_access
+ 1);
1527 *implicit_sized
= true;
1528 assert(*type
!= NULL
);
1532 static const glsl_type
*
1533 update_interface_members_array(const glsl_type
*type
,
1534 const glsl_type
*new_interface_type
)
1536 const glsl_type
*element_type
= type
->fields
.array
;
1537 if (element_type
->is_array()) {
1538 const glsl_type
*new_array_type
=
1539 update_interface_members_array(element_type
, new_interface_type
);
1540 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1542 return glsl_type::get_array_instance(new_interface_type
,
1548 * Determine whether the given interface type contains unsized arrays (if
1549 * it doesn't, array_sizing_visitor doesn't need to process it).
1551 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1553 for (unsigned i
= 0; i
< type
->length
; i
++) {
1554 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1555 if (elem_type
->is_unsized_array())
1562 * Create a new interface type based on the given type, with unsized arrays
1563 * replaced by sized arrays whose size is determined by
1564 * max_ifc_array_access.
1566 static const glsl_type
*
1567 resize_interface_members(const glsl_type
*type
,
1568 const int *max_ifc_array_access
,
1571 unsigned num_fields
= type
->length
;
1572 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1573 memcpy(fields
, type
->fields
.structure
,
1574 num_fields
* sizeof(*fields
));
1575 for (unsigned i
= 0; i
< num_fields
; i
++) {
1576 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1577 /* If SSBO last member is unsized array, we don't replace it by a sized
1580 if (is_ssbo
&& i
== (num_fields
- 1))
1581 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1582 true, &implicit_sized_array
);
1584 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1585 false, &implicit_sized_array
);
1586 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1588 glsl_interface_packing packing
=
1589 (glsl_interface_packing
) type
->interface_packing
;
1590 bool row_major
= (bool) type
->interface_row_major
;
1591 const glsl_type
*new_ifc_type
=
1592 glsl_type::get_interface_instance(fields
, num_fields
,
1593 packing
, row_major
, type
->name
);
1595 return new_ifc_type
;
1598 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1601 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1602 ir_variable
**interface_vars
= (ir_variable
**) data
;
1603 unsigned num_fields
= ifc_type
->length
;
1604 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1605 memcpy(fields
, ifc_type
->fields
.structure
,
1606 num_fields
* sizeof(*fields
));
1607 bool interface_type_changed
= false;
1608 for (unsigned i
= 0; i
< num_fields
; i
++) {
1609 if (interface_vars
[i
] != NULL
&&
1610 fields
[i
].type
!= interface_vars
[i
]->type
) {
1611 fields
[i
].type
= interface_vars
[i
]->type
;
1612 interface_type_changed
= true;
1615 if (!interface_type_changed
) {
1619 glsl_interface_packing packing
=
1620 (glsl_interface_packing
) ifc_type
->interface_packing
;
1621 bool row_major
= (bool) ifc_type
->interface_row_major
;
1622 const glsl_type
*new_ifc_type
=
1623 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1624 row_major
, ifc_type
->name
);
1626 for (unsigned i
= 0; i
< num_fields
; i
++) {
1627 if (interface_vars
[i
] != NULL
)
1628 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1633 * Memory context used to allocate the data in \c unnamed_interfaces.
1638 * Hash table from const glsl_type * to an array of ir_variable *'s
1639 * pointing to the ir_variables constituting each unnamed interface block.
1641 hash_table
*unnamed_interfaces
;
1645 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1646 struct gl_shader_program
*prog
)
1648 /* We will validate doubles at a later stage */
1649 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1650 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1651 "multiple of 4 or if its applied to a type that is "
1652 "or contains a double a multiple of 8.",
1653 prog
->TransformFeedback
.BufferStride
[idx
]);
1657 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1658 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1659 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1660 "limit has been exceeded.");
1668 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1672 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1673 struct gl_shader_program
*prog
,
1674 struct gl_shader
**shader_list
,
1675 unsigned num_shaders
)
1677 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1678 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1681 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1682 struct gl_shader
*shader
= shader_list
[i
];
1684 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1685 if (shader
->TransformFeedbackBufferStride
[j
]) {
1686 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1687 prog
->TransformFeedback
.BufferStride
[j
] =
1688 shader
->TransformFeedbackBufferStride
[j
];
1689 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1691 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1692 shader
->TransformFeedbackBufferStride
[j
]){
1694 "intrastage shaders defined with conflicting "
1695 "xfb_stride for buffer %d (%d and %d)\n", j
,
1696 prog
->TransformFeedback
.BufferStride
[j
],
1697 shader
->TransformFeedbackBufferStride
[j
]);
1706 * Check for conflicting bindless/bound sampler/image layout qualifiers at
1710 link_bindless_layout_qualifiers(struct gl_shader_program
*prog
,
1711 struct gl_shader
**shader_list
,
1712 unsigned num_shaders
)
1714 bool bindless_sampler
, bindless_image
;
1715 bool bound_sampler
, bound_image
;
1717 bindless_sampler
= bindless_image
= false;
1718 bound_sampler
= bound_image
= false;
1720 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1721 struct gl_shader
*shader
= shader_list
[i
];
1723 if (shader
->bindless_sampler
)
1724 bindless_sampler
= true;
1725 if (shader
->bindless_image
)
1726 bindless_image
= true;
1727 if (shader
->bound_sampler
)
1728 bound_sampler
= true;
1729 if (shader
->bound_image
)
1732 if ((bindless_sampler
&& bound_sampler
) ||
1733 (bindless_image
&& bound_image
)) {
1734 /* From section 4.4.6 of the ARB_bindless_texture spec:
1736 * "If both bindless_sampler and bound_sampler, or bindless_image
1737 * and bound_image, are declared at global scope in any
1738 * compilation unit, a link- time error will be generated."
1740 linker_error(prog
, "both bindless_sampler and bound_sampler, or "
1741 "bindless_image and bound_image, can't be declared at "
1748 * Performs the cross-validation of tessellation control shader vertices and
1749 * layout qualifiers for the attached tessellation control shaders,
1750 * and propagates them to the linked TCS and linked shader program.
1753 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1754 struct gl_program
*gl_prog
,
1755 struct gl_shader
**shader_list
,
1756 unsigned num_shaders
)
1758 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1761 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1763 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1765 * "All tessellation control shader layout declarations in a program
1766 * must specify the same output patch vertex count. There must be at
1767 * least one layout qualifier specifying an output patch vertex count
1768 * in any program containing tessellation control shaders; however,
1769 * such a declaration is not required in all tessellation control
1773 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1774 struct gl_shader
*shader
= shader_list
[i
];
1776 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1777 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1778 gl_prog
->info
.tess
.tcs_vertices_out
!=
1779 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1780 linker_error(prog
, "tessellation control shader defined with "
1781 "conflicting output vertex count (%d and %d)\n",
1782 gl_prog
->info
.tess
.tcs_vertices_out
,
1783 shader
->info
.TessCtrl
.VerticesOut
);
1786 gl_prog
->info
.tess
.tcs_vertices_out
=
1787 shader
->info
.TessCtrl
.VerticesOut
;
1791 /* Just do the intrastage -> interstage propagation right now,
1792 * since we already know we're in the right type of shader program
1795 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1796 linker_error(prog
, "tessellation control shader didn't declare "
1797 "vertices out layout qualifier\n");
1804 * Performs the cross-validation of tessellation evaluation shader
1805 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1806 * for the attached tessellation evaluation shaders, and propagates them
1807 * to the linked TES and linked shader program.
1810 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1811 struct gl_program
*gl_prog
,
1812 struct gl_shader
**shader_list
,
1813 unsigned num_shaders
)
1815 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1818 int point_mode
= -1;
1819 unsigned vertex_order
= 0;
1821 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1822 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1824 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1826 * "At least one tessellation evaluation shader (compilation unit) in
1827 * a program must declare a primitive mode in its input layout.
1828 * Declaration vertex spacing, ordering, and point mode identifiers is
1829 * optional. It is not required that all tessellation evaluation
1830 * shaders in a program declare a primitive mode. If spacing or
1831 * vertex ordering declarations are omitted, the tessellation
1832 * primitive generator will use equal spacing or counter-clockwise
1833 * vertex ordering, respectively. If a point mode declaration is
1834 * omitted, the tessellation primitive generator will produce lines or
1835 * triangles according to the primitive mode."
1838 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1839 struct gl_shader
*shader
= shader_list
[i
];
1841 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1842 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1843 gl_prog
->info
.tess
.primitive_mode
!=
1844 shader
->info
.TessEval
.PrimitiveMode
) {
1845 linker_error(prog
, "tessellation evaluation shader defined with "
1846 "conflicting input primitive modes.\n");
1849 gl_prog
->info
.tess
.primitive_mode
=
1850 shader
->info
.TessEval
.PrimitiveMode
;
1853 if (shader
->info
.TessEval
.Spacing
!= 0) {
1854 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1855 shader
->info
.TessEval
.Spacing
) {
1856 linker_error(prog
, "tessellation evaluation shader defined with "
1857 "conflicting vertex spacing.\n");
1860 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1863 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1864 if (vertex_order
!= 0 &&
1865 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1866 linker_error(prog
, "tessellation evaluation shader defined with "
1867 "conflicting ordering.\n");
1870 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1873 if (shader
->info
.TessEval
.PointMode
!= -1) {
1874 if (point_mode
!= -1 &&
1875 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1876 linker_error(prog
, "tessellation evaluation shader defined with "
1877 "conflicting point modes.\n");
1880 point_mode
= shader
->info
.TessEval
.PointMode
;
1885 /* Just do the intrastage -> interstage propagation right now,
1886 * since we already know we're in the right type of shader program
1889 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1891 "tessellation evaluation shader didn't declare input "
1892 "primitive modes.\n");
1896 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1897 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1899 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1900 gl_prog
->info
.tess
.ccw
= true;
1902 gl_prog
->info
.tess
.ccw
= false;
1905 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1906 gl_prog
->info
.tess
.point_mode
= false;
1908 gl_prog
->info
.tess
.point_mode
= true;
1913 * Performs the cross-validation of layout qualifiers specified in
1914 * redeclaration of gl_FragCoord for the attached fragment shaders,
1915 * and propagates them to the linked FS and linked shader program.
1918 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1919 struct gl_linked_shader
*linked_shader
,
1920 struct gl_shader
**shader_list
,
1921 unsigned num_shaders
)
1923 bool redeclares_gl_fragcoord
= false;
1924 bool uses_gl_fragcoord
= false;
1925 bool origin_upper_left
= false;
1926 bool pixel_center_integer
= false;
1928 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1929 (prog
->data
->Version
< 150 &&
1930 !prog
->ARB_fragment_coord_conventions_enable
))
1933 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1934 struct gl_shader
*shader
= shader_list
[i
];
1935 /* From the GLSL 1.50 spec, page 39:
1937 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1938 * it must be redeclared in all the fragment shaders in that program
1939 * that have a static use gl_FragCoord."
1941 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1942 shader
->uses_gl_fragcoord
)
1943 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1944 uses_gl_fragcoord
)) {
1945 linker_error(prog
, "fragment shader defined with conflicting "
1946 "layout qualifiers for gl_FragCoord\n");
1949 /* From the GLSL 1.50 spec, page 39:
1951 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1952 * single program must have the same set of qualifiers."
1954 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1955 (shader
->origin_upper_left
!= origin_upper_left
||
1956 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1957 linker_error(prog
, "fragment shader defined with conflicting "
1958 "layout qualifiers for gl_FragCoord\n");
1961 /* Update the linked shader state. Note that uses_gl_fragcoord should
1962 * accumulate the results. The other values should replace. If there
1963 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1964 * are already known to be the same.
1966 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1967 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1968 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1969 origin_upper_left
= shader
->origin_upper_left
;
1970 pixel_center_integer
= shader
->pixel_center_integer
;
1973 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1974 shader
->EarlyFragmentTests
|| shader
->PostDepthCoverage
;
1975 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1976 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1977 shader
->PostDepthCoverage
;
1979 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1984 * Performs the cross-validation of geometry shader max_vertices and
1985 * primitive type layout qualifiers for the attached geometry shaders,
1986 * and propagates them to the linked GS and linked shader program.
1989 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1990 struct gl_program
*gl_prog
,
1991 struct gl_shader
**shader_list
,
1992 unsigned num_shaders
)
1994 /* No in/out qualifiers defined for anything but GLSL 1.50+
1995 * geometry shaders so far.
1997 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
1998 prog
->data
->Version
< 150)
2001 int vertices_out
= -1;
2003 gl_prog
->info
.gs
.invocations
= 0;
2004 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
2005 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
2007 /* From the GLSL 1.50 spec, page 46:
2009 * "All geometry shader output layout declarations in a program
2010 * must declare the same layout and same value for
2011 * max_vertices. There must be at least one geometry output
2012 * layout declaration somewhere in a program, but not all
2013 * geometry shaders (compilation units) are required to
2017 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2018 struct gl_shader
*shader
= shader_list
[i
];
2020 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
2021 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
2022 gl_prog
->info
.gs
.input_primitive
!=
2023 shader
->info
.Geom
.InputType
) {
2024 linker_error(prog
, "geometry shader defined with conflicting "
2028 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
2031 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
2032 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
2033 gl_prog
->info
.gs
.output_primitive
!=
2034 shader
->info
.Geom
.OutputType
) {
2035 linker_error(prog
, "geometry shader defined with conflicting "
2039 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
2042 if (shader
->info
.Geom
.VerticesOut
!= -1) {
2043 if (vertices_out
!= -1 &&
2044 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
2045 linker_error(prog
, "geometry shader defined with conflicting "
2046 "output vertex count (%d and %d)\n",
2047 vertices_out
, shader
->info
.Geom
.VerticesOut
);
2050 vertices_out
= shader
->info
.Geom
.VerticesOut
;
2053 if (shader
->info
.Geom
.Invocations
!= 0) {
2054 if (gl_prog
->info
.gs
.invocations
!= 0 &&
2055 gl_prog
->info
.gs
.invocations
!=
2056 (unsigned) shader
->info
.Geom
.Invocations
) {
2057 linker_error(prog
, "geometry shader defined with conflicting "
2058 "invocation count (%d and %d)\n",
2059 gl_prog
->info
.gs
.invocations
,
2060 shader
->info
.Geom
.Invocations
);
2063 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
2067 /* Just do the intrastage -> interstage propagation right now,
2068 * since we already know we're in the right type of shader program
2071 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
2073 "geometry shader didn't declare primitive input type\n");
2077 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
2079 "geometry shader didn't declare primitive output type\n");
2083 if (vertices_out
== -1) {
2085 "geometry shader didn't declare max_vertices\n");
2088 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2091 if (gl_prog
->info
.gs
.invocations
== 0)
2092 gl_prog
->info
.gs
.invocations
= 1;
2097 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2098 * qualifiers for the attached compute shaders, and propagate them to the
2099 * linked CS and linked shader program.
2102 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2103 struct gl_program
*gl_prog
,
2104 struct gl_shader
**shader_list
,
2105 unsigned num_shaders
)
2107 /* This function is called for all shader stages, but it only has an effect
2108 * for compute shaders.
2110 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2113 for (int i
= 0; i
< 3; i
++)
2114 gl_prog
->info
.cs
.local_size
[i
] = 0;
2116 gl_prog
->info
.cs
.local_size_variable
= false;
2118 /* From the ARB_compute_shader spec, in the section describing local size
2121 * If multiple compute shaders attached to a single program object
2122 * declare local work-group size, the declarations must be identical;
2123 * otherwise a link-time error results. Furthermore, if a program
2124 * object contains any compute shaders, at least one must contain an
2125 * input layout qualifier specifying the local work sizes of the
2126 * program, or a link-time error will occur.
2128 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2129 struct gl_shader
*shader
= shader_list
[sh
];
2131 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2132 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2133 for (int i
= 0; i
< 3; i
++) {
2134 if (gl_prog
->info
.cs
.local_size
[i
] !=
2135 shader
->info
.Comp
.LocalSize
[i
]) {
2136 linker_error(prog
, "compute shader defined with conflicting "
2142 for (int i
= 0; i
< 3; i
++) {
2143 gl_prog
->info
.cs
.local_size
[i
] =
2144 shader
->info
.Comp
.LocalSize
[i
];
2146 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2147 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2148 /* The ARB_compute_variable_group_size spec says:
2150 * If one compute shader attached to a program declares a
2151 * variable local group size and a second compute shader
2152 * attached to the same program declares a fixed local group
2153 * size, a link-time error results.
2155 linker_error(prog
, "compute shader defined with both fixed and "
2156 "variable local group size\n");
2159 gl_prog
->info
.cs
.local_size_variable
= true;
2163 /* Just do the intrastage -> interstage propagation right now,
2164 * since we already know we're in the right type of shader program
2167 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2168 !gl_prog
->info
.cs
.local_size_variable
) {
2169 linker_error(prog
, "compute shader must contain a fixed or a variable "
2170 "local group size\n");
2177 * Combine a group of shaders for a single stage to generate a linked shader
2180 * If this function is supplied a single shader, it is cloned, and the new
2181 * shader is returned.
2183 struct gl_linked_shader
*
2184 link_intrastage_shaders(void *mem_ctx
,
2185 struct gl_context
*ctx
,
2186 struct gl_shader_program
*prog
,
2187 struct gl_shader
**shader_list
,
2188 unsigned num_shaders
,
2189 bool allow_missing_main
)
2191 struct gl_uniform_block
*ubo_blocks
= NULL
;
2192 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2193 unsigned num_ubo_blocks
= 0;
2194 unsigned num_ssbo_blocks
= 0;
2196 /* Check that global variables defined in multiple shaders are consistent.
2198 glsl_symbol_table variables
;
2199 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2200 if (shader_list
[i
] == NULL
)
2202 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2205 if (!prog
->data
->LinkStatus
)
2208 /* Check that interface blocks defined in multiple shaders are consistent.
2210 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2212 if (!prog
->data
->LinkStatus
)
2215 /* Check that there is only a single definition of each function signature
2216 * across all shaders.
2218 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2219 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2220 ir_function
*const f
= node
->as_function();
2225 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2226 ir_function
*const other
=
2227 shader_list
[j
]->symbols
->get_function(f
->name
);
2229 /* If the other shader has no function (and therefore no function
2230 * signatures) with the same name, skip to the next shader.
2235 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2236 if (!sig
->is_defined
)
2239 ir_function_signature
*other_sig
=
2240 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2242 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2243 linker_error(prog
, "function `%s' is multiply defined\n",
2252 /* Find the shader that defines main, and make a clone of it.
2254 * Starting with the clone, search for undefined references. If one is
2255 * found, find the shader that defines it. Clone the reference and add
2256 * it to the shader. Repeat until there are no undefined references or
2257 * until a reference cannot be resolved.
2259 gl_shader
*main
= NULL
;
2260 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2261 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2262 main
= shader_list
[i
];
2267 if (main
== NULL
&& allow_missing_main
)
2268 main
= shader_list
[0];
2271 linker_error(prog
, "%s shader lacks `main'\n",
2272 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2276 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2277 linked
->Stage
= shader_list
[0]->Stage
;
2279 /* Create program and attach it to the linked shader */
2280 struct gl_program
*gl_prog
=
2281 ctx
->Driver
.NewProgram(ctx
,
2282 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2285 prog
->data
->LinkStatus
= LINKING_FAILURE
;
2286 _mesa_delete_linked_shader(ctx
, linked
);
2290 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2292 /* Don't use _mesa_reference_program() just take ownership */
2293 linked
->Program
= gl_prog
;
2295 linked
->ir
= new(linked
) exec_list
;
2296 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2298 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2299 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2300 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2301 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2302 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2304 if (linked
->Stage
!= MESA_SHADER_FRAGMENT
)
2305 link_xfb_stride_layout_qualifiers(ctx
, prog
, shader_list
, num_shaders
);
2307 link_bindless_layout_qualifiers(prog
, shader_list
, num_shaders
);
2309 populate_symbol_table(linked
, shader_list
[0]->symbols
);
2311 /* The pointer to the main function in the final linked shader (i.e., the
2312 * copy of the original shader that contained the main function).
2314 ir_function_signature
*const main_sig
=
2315 _mesa_get_main_function_signature(linked
->symbols
);
2317 /* Move any instructions other than variable declarations or function
2318 * declarations into main.
2320 if (main_sig
!= NULL
) {
2321 exec_node
*insertion_point
=
2322 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2325 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2326 if (shader_list
[i
] == main
)
2329 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2330 insertion_point
, true, linked
);
2334 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2335 _mesa_delete_linked_shader(ctx
, linked
);
2339 /* Make a pass over all variable declarations to ensure that arrays with
2340 * unspecified sizes have a size specified. The size is inferred from the
2341 * max_array_access field.
2343 array_sizing_visitor v
;
2345 v
.fixup_unnamed_interface_types();
2347 /* Link up uniform blocks defined within this stage. */
2348 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2349 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2351 if (!prog
->data
->LinkStatus
) {
2352 _mesa_delete_linked_shader(ctx
, linked
);
2356 /* Copy ubo blocks to linked shader list */
2357 linked
->Program
->sh
.UniformBlocks
=
2358 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2359 ralloc_steal(linked
, ubo_blocks
);
2360 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2361 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2363 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2365 /* Copy ssbo blocks to linked shader list */
2366 linked
->Program
->sh
.ShaderStorageBlocks
=
2367 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2368 ralloc_steal(linked
, ssbo_blocks
);
2369 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2370 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2372 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2374 /* At this point linked should contain all of the linked IR, so
2375 * validate it to make sure nothing went wrong.
2377 validate_ir_tree(linked
->ir
);
2379 /* Set the size of geometry shader input arrays */
2380 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2381 unsigned num_vertices
=
2382 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2383 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2384 MESA_SHADER_GEOMETRY
);
2385 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2386 ir
->accept(&input_resize_visitor
);
2390 if (ctx
->Const
.VertexID_is_zero_based
)
2391 lower_vertex_id(linked
);
2393 if (ctx
->Const
.LowerCsDerivedVariables
)
2394 lower_cs_derived(linked
);
2397 /* Compute the source checksum. */
2398 linked
->SourceChecksum
= 0;
2399 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2400 if (shader_list
[i
] == NULL
)
2402 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2410 * Update the sizes of linked shader uniform arrays to the maximum
2413 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2415 * If one or more elements of an array are active,
2416 * GetActiveUniform will return the name of the array in name,
2417 * subject to the restrictions listed above. The type of the array
2418 * is returned in type. The size parameter contains the highest
2419 * array element index used, plus one. The compiler or linker
2420 * determines the highest index used. There will be only one
2421 * active uniform reported by the GL per uniform array.
2425 update_array_sizes(struct gl_shader_program
*prog
)
2427 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2428 if (prog
->_LinkedShaders
[i
] == NULL
)
2431 bool types_were_updated
= false;
2433 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2434 ir_variable
*const var
= node
->as_variable();
2436 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2437 !var
->type
->is_array())
2440 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2441 * will not be eliminated. Since we always do std140, just
2442 * don't resize arrays in UBOs.
2444 * Atomic counters are supposed to get deterministic
2445 * locations assigned based on the declaration ordering and
2446 * sizes, array compaction would mess that up.
2448 * Subroutine uniforms are not removed.
2450 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2451 var
->type
->contains_subroutine() || var
->constant_initializer
)
2454 int size
= var
->data
.max_array_access
;
2455 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2456 if (prog
->_LinkedShaders
[j
] == NULL
)
2459 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2460 ir_variable
*other_var
= node2
->as_variable();
2464 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2465 other_var
->data
.max_array_access
> size
) {
2466 size
= other_var
->data
.max_array_access
;
2471 if (size
+ 1 != (int)var
->type
->length
) {
2472 /* If this is a built-in uniform (i.e., it's backed by some
2473 * fixed-function state), adjust the number of state slots to
2474 * match the new array size. The number of slots per array entry
2475 * is not known. It seems safe to assume that the total number of
2476 * slots is an integer multiple of the number of array elements.
2477 * Determine the number of slots per array element by dividing by
2478 * the old (total) size.
2480 const unsigned num_slots
= var
->get_num_state_slots();
2481 if (num_slots
> 0) {
2482 var
->set_num_state_slots((size
+ 1)
2483 * (num_slots
/ var
->type
->length
));
2486 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2488 types_were_updated
= true;
2492 /* Update the types of dereferences in case we changed any. */
2493 if (types_were_updated
) {
2494 deref_type_updater v
;
2495 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2501 * Resize tessellation evaluation per-vertex inputs to the size of
2502 * tessellation control per-vertex outputs.
2505 resize_tes_inputs(struct gl_context
*ctx
,
2506 struct gl_shader_program
*prog
)
2508 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2511 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2512 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2514 /* If no control shader is present, then the TES inputs are statically
2515 * sized to MaxPatchVertices; the actual size of the arrays won't be
2516 * known until draw time.
2518 const int num_vertices
= tcs
2519 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2520 : ctx
->Const
.MaxPatchVertices
;
2522 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2523 MESA_SHADER_TESS_EVAL
);
2524 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2525 ir
->accept(&input_resize_visitor
);
2529 /* Convert the gl_PatchVerticesIn system value into a constant, since
2530 * the value is known at this point.
2532 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2533 ir_variable
*var
= ir
->as_variable();
2534 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2535 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2536 void *mem_ctx
= ralloc_parent(var
);
2537 var
->data
.location
= 0;
2538 var
->data
.explicit_location
= false;
2539 var
->data
.mode
= ir_var_auto
;
2540 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2547 * Find a contiguous set of available bits in a bitmask.
2549 * \param used_mask Bits representing used (1) and unused (0) locations
2550 * \param needed_count Number of contiguous bits needed.
2553 * Base location of the available bits on success or -1 on failure.
2556 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2558 unsigned needed_mask
= (1 << needed_count
) - 1;
2559 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2561 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2562 * cannot optimize possibly infinite loops" for the loop below.
2564 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2567 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2568 if ((needed_mask
& ~used_mask
) == needed_mask
)
2578 #define SAFE_MASK_FROM_INDEX(i) (((i) >= 32) ? ~0 : ((1 << (i)) - 1))
2581 * Assign locations for either VS inputs or FS outputs
2583 * \param mem_ctx Temporary ralloc context used for linking
2584 * \param prog Shader program whose variables need locations assigned
2585 * \param constants Driver specific constant values for the program.
2586 * \param target_index Selector for the program target to receive location
2587 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2588 * \c MESA_SHADER_FRAGMENT.
2591 * If locations are successfully assigned, true is returned. Otherwise an
2592 * error is emitted to the shader link log and false is returned.
2595 assign_attribute_or_color_locations(void *mem_ctx
,
2596 gl_shader_program
*prog
,
2597 struct gl_constants
*constants
,
2598 unsigned target_index
)
2600 /* Maximum number of generic locations. This corresponds to either the
2601 * maximum number of draw buffers or the maximum number of generic
2604 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2605 constants
->Program
[target_index
].MaxAttribs
:
2606 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2608 /* Mark invalid locations as being used.
2610 unsigned used_locations
= ~SAFE_MASK_FROM_INDEX(max_index
);
2611 unsigned double_storage_locations
= 0;
2613 assert((target_index
== MESA_SHADER_VERTEX
)
2614 || (target_index
== MESA_SHADER_FRAGMENT
));
2616 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2620 /* Operate in a total of four passes.
2622 * 1. Invalidate the location assignments for all vertex shader inputs.
2624 * 2. Assign locations for inputs that have user-defined (via
2625 * glBindVertexAttribLocation) locations and outputs that have
2626 * user-defined locations (via glBindFragDataLocation).
2628 * 3. Sort the attributes without assigned locations by number of slots
2629 * required in decreasing order. Fragmentation caused by attribute
2630 * locations assigned by the application may prevent large attributes
2631 * from having enough contiguous space.
2633 * 4. Assign locations to any inputs without assigned locations.
2636 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2637 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2639 const enum ir_variable_mode direction
=
2640 (target_index
== MESA_SHADER_VERTEX
)
2641 ? ir_var_shader_in
: ir_var_shader_out
;
2644 /* Temporary storage for the set of attributes that need locations assigned.
2650 /* Used below in the call to qsort. */
2651 static int compare(const void *a
, const void *b
)
2653 const temp_attr
*const l
= (const temp_attr
*) a
;
2654 const temp_attr
*const r
= (const temp_attr
*) b
;
2656 /* Reversed because we want a descending order sort below. */
2657 return r
->slots
- l
->slots
;
2660 assert(max_index
<= 32);
2662 /* Temporary array for the set of attributes that have locations assigned,
2663 * for the purpose of checking overlapping slots/components of (non-ES)
2664 * fragment shader outputs.
2666 ir_variable
*assigned
[12 * 4]; /* (max # of FS outputs) * # components */
2667 unsigned assigned_attr
= 0;
2669 unsigned num_attr
= 0;
2671 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2672 ir_variable
*const var
= node
->as_variable();
2674 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2677 if (var
->data
.explicit_location
) {
2678 var
->data
.is_unmatched_generic_inout
= 0;
2679 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2680 || (var
->data
.location
< 0)) {
2682 "invalid explicit location %d specified for `%s'\n",
2683 (var
->data
.location
< 0)
2684 ? var
->data
.location
2685 : var
->data
.location
- generic_base
,
2689 } else if (target_index
== MESA_SHADER_VERTEX
) {
2692 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2693 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2694 var
->data
.location
= binding
;
2695 var
->data
.is_unmatched_generic_inout
= 0;
2697 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2700 const char *name
= var
->name
;
2701 const glsl_type
*type
= var
->type
;
2704 /* Check if there's a binding for the variable name */
2705 if (prog
->FragDataBindings
->get(binding
, name
)) {
2706 assert(binding
>= FRAG_RESULT_DATA0
);
2707 var
->data
.location
= binding
;
2708 var
->data
.is_unmatched_generic_inout
= 0;
2710 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2711 var
->data
.index
= index
;
2716 /* If not, but it's an array type, look for name[0] */
2717 if (type
->is_array()) {
2718 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2719 type
= type
->fields
.array
;
2727 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2730 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2732 * "Output binding assignments will cause LinkProgram to fail:
2734 * If the program has an active output assigned to a location greater
2735 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2736 * an active output assigned an index greater than or equal to one;"
2738 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2739 var
->data
.location
- generic_base
>=
2740 (int) constants
->MaxDualSourceDrawBuffers
) {
2742 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2743 "with index %u for %s\n",
2744 var
->data
.location
- generic_base
, var
->data
.index
,
2749 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2751 /* If the variable is not a built-in and has a location statically
2752 * assigned in the shader (presumably via a layout qualifier), make sure
2753 * that it doesn't collide with other assigned locations. Otherwise,
2754 * add it to the list of variables that need linker-assigned locations.
2756 if (var
->data
.location
!= -1) {
2757 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2758 /* From page 61 of the OpenGL 4.0 spec:
2760 * "LinkProgram will fail if the attribute bindings assigned
2761 * by BindAttribLocation do not leave not enough space to
2762 * assign a location for an active matrix attribute or an
2763 * active attribute array, both of which require multiple
2764 * contiguous generic attributes."
2766 * I think above text prohibits the aliasing of explicit and
2767 * automatic assignments. But, aliasing is allowed in manual
2768 * assignments of attribute locations. See below comments for
2771 * From OpenGL 4.0 spec, page 61:
2773 * "It is possible for an application to bind more than one
2774 * attribute name to the same location. This is referred to as
2775 * aliasing. This will only work if only one of the aliased
2776 * attributes is active in the executable program, or if no
2777 * path through the shader consumes more than one attribute of
2778 * a set of attributes aliased to the same location. A link
2779 * error can occur if the linker determines that every path
2780 * through the shader consumes multiple aliased attributes,
2781 * but implementations are not required to generate an error
2784 * From GLSL 4.30 spec, page 54:
2786 * "A program will fail to link if any two non-vertex shader
2787 * input variables are assigned to the same location. For
2788 * vertex shaders, multiple input variables may be assigned
2789 * to the same location using either layout qualifiers or via
2790 * the OpenGL API. However, such aliasing is intended only to
2791 * support vertex shaders where each execution path accesses
2792 * at most one input per each location. Implementations are
2793 * permitted, but not required, to generate link-time errors
2794 * if they detect that every path through the vertex shader
2795 * executable accesses multiple inputs assigned to any single
2796 * location. For all shader types, a program will fail to link
2797 * if explicit location assignments leave the linker unable
2798 * to find space for other variables without explicit
2801 * From OpenGL ES 3.0 spec, page 56:
2803 * "Binding more than one attribute name to the same location
2804 * is referred to as aliasing, and is not permitted in OpenGL
2805 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2806 * fail when this condition exists. However, aliasing is
2807 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2808 * This will only work if only one of the aliased attributes
2809 * is active in the executable program, or if no path through
2810 * the shader consumes more than one attribute of a set of
2811 * attributes aliased to the same location. A link error can
2812 * occur if the linker determines that every path through the
2813 * shader consumes multiple aliased attributes, but implemen-
2814 * tations are not required to generate an error in this case."
2816 * After looking at above references from OpenGL, OpenGL ES and
2817 * GLSL specifications, we allow aliasing of vertex input variables
2818 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2820 * NOTE: This is not required by the spec but its worth mentioning
2821 * here that we're not doing anything to make sure that no path
2822 * through the vertex shader executable accesses multiple inputs
2823 * assigned to any single location.
2826 /* Mask representing the contiguous slots that will be used by
2829 const unsigned attr
= var
->data
.location
- generic_base
;
2830 const unsigned use_mask
= (1 << slots
) - 1;
2831 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2832 ? "vertex shader input" : "fragment shader output";
2834 /* Generate a link error if the requested locations for this
2835 * attribute exceed the maximum allowed attribute location.
2837 if (attr
+ slots
> max_index
) {
2839 "insufficient contiguous locations "
2840 "available for %s `%s' %d %d %d\n", string
,
2841 var
->name
, used_locations
, use_mask
, attr
);
2845 /* Generate a link error if the set of bits requested for this
2846 * attribute overlaps any previously allocated bits.
2848 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2849 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2850 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2853 * "Additionally, for fragment shader outputs, if two
2854 * variables are placed within the same location, they
2855 * must have the same underlying type (floating-point or
2856 * integer). No component aliasing of output variables or
2857 * members is allowed.
2859 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2860 unsigned assigned_slots
=
2861 assigned
[i
]->type
->count_attribute_slots(false);
2862 unsigned assig_attr
=
2863 assigned
[i
]->data
.location
- generic_base
;
2864 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2866 if ((assigned_use_mask
<< assig_attr
) &
2867 (use_mask
<< attr
)) {
2869 const glsl_type
*assigned_type
=
2870 assigned
[i
]->type
->without_array();
2871 const glsl_type
*type
= var
->type
->without_array();
2872 if (assigned_type
->base_type
!= type
->base_type
) {
2873 linker_error(prog
, "types do not match for aliased"
2874 " %ss %s and %s\n", string
,
2875 assigned
[i
]->name
, var
->name
);
2879 unsigned assigned_component_mask
=
2880 ((1 << assigned_type
->vector_elements
) - 1) <<
2881 assigned
[i
]->data
.location_frac
;
2882 unsigned component_mask
=
2883 ((1 << type
->vector_elements
) - 1) <<
2884 var
->data
.location_frac
;
2885 if (assigned_component_mask
& component_mask
) {
2886 linker_error(prog
, "overlapping component is "
2887 "assigned to %ss %s and %s "
2889 string
, assigned
[i
]->name
, var
->name
,
2890 var
->data
.location_frac
);
2895 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2896 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2897 linker_error(prog
, "overlapping location is assigned "
2898 "to %s `%s' %d %d %d\n", string
, var
->name
,
2899 used_locations
, use_mask
, attr
);
2902 linker_warning(prog
, "overlapping location is assigned "
2903 "to %s `%s' %d %d %d\n", string
, var
->name
,
2904 used_locations
, use_mask
, attr
);
2908 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2909 /* Only track assigned variables for non-ES fragment shaders
2910 * to avoid overflowing the array.
2912 * At most one variable per fragment output component should
2915 assert(assigned_attr
< ARRAY_SIZE(assigned
));
2916 assigned
[assigned_attr
] = var
;
2920 used_locations
|= (use_mask
<< attr
);
2922 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2924 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2925 * active attribute variables may fail to link, unless
2926 * device-dependent optimizations are able to make the program
2927 * fit within available hardware resources. For the purposes
2928 * of this test, attribute variables of the type dvec3, dvec4,
2929 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2930 * count as consuming twice as many attributes as equivalent
2931 * single-precision types. While these types use the same number
2932 * of generic attributes as their single-precision equivalents,
2933 * implementations are permitted to consume two single-precision
2934 * vectors of internal storage for each three- or four-component
2935 * double-precision vector."
2937 * Mark this attribute slot as taking up twice as much space
2938 * so we can count it properly against limits. According to
2939 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2940 * is optional behavior, but it seems preferable.
2942 if (var
->type
->without_array()->is_dual_slot())
2943 double_storage_locations
|= (use_mask
<< attr
);
2949 if (num_attr
>= max_index
) {
2950 linker_error(prog
, "too many %s (max %u)",
2951 target_index
== MESA_SHADER_VERTEX
?
2952 "vertex shader inputs" : "fragment shader outputs",
2956 to_assign
[num_attr
].slots
= slots
;
2957 to_assign
[num_attr
].var
= var
;
2961 if (target_index
== MESA_SHADER_VERTEX
) {
2962 unsigned total_attribs_size
=
2963 _mesa_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
2964 _mesa_bitcount(double_storage_locations
);
2965 if (total_attribs_size
> max_index
) {
2967 "attempt to use %d vertex attribute slots only %d available ",
2968 total_attribs_size
, max_index
);
2973 /* If all of the attributes were assigned locations by the application (or
2974 * are built-in attributes with fixed locations), return early. This should
2975 * be the common case.
2980 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2982 if (target_index
== MESA_SHADER_VERTEX
) {
2983 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2984 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2985 * reserved to prevent it from being automatically allocated below.
2987 find_deref_visitor
find("gl_Vertex");
2989 if (find
.variable_found())
2990 used_locations
|= (1 << 0);
2993 for (unsigned i
= 0; i
< num_attr
; i
++) {
2994 /* Mask representing the contiguous slots that will be used by this
2997 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2999 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
3002 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
3003 ? "vertex shader input" : "fragment shader output";
3006 "insufficient contiguous locations "
3007 "available for %s `%s'\n",
3008 string
, to_assign
[i
].var
->name
);
3012 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
3013 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
3014 used_locations
|= (use_mask
<< location
);
3016 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
3017 double_storage_locations
|= (use_mask
<< location
);
3020 /* Now that we have all the locations, from the GL 4.5 core spec, section
3021 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
3022 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
3023 * as equivalent single-precision types.
3025 if (target_index
== MESA_SHADER_VERTEX
) {
3026 unsigned total_attribs_size
=
3027 _mesa_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
3028 _mesa_bitcount(double_storage_locations
);
3029 if (total_attribs_size
> max_index
) {
3031 "attempt to use %d vertex attribute slots only %d available ",
3032 total_attribs_size
, max_index
);
3041 * Match explicit locations of outputs to inputs and deactivate the
3042 * unmatch flag if found so we don't optimise them away.
3045 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
3046 gl_linked_shader
*consumer
)
3048 glsl_symbol_table parameters
;
3049 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
3052 /* Find all shader outputs in the "producer" stage.
3054 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
3055 ir_variable
*const var
= node
->as_variable();
3057 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
3060 if (var
->data
.explicit_location
&&
3061 var
->data
.location
>= VARYING_SLOT_VAR0
) {
3062 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
3063 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
3064 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
3068 /* Match inputs to outputs */
3069 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
3070 ir_variable
*const input
= node
->as_variable();
3072 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
3075 ir_variable
*output
= NULL
;
3076 if (input
->data
.explicit_location
3077 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
3078 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
3079 [input
->data
.location_frac
];
3081 if (output
!= NULL
){
3082 input
->data
.is_unmatched_generic_inout
= 0;
3083 output
->data
.is_unmatched_generic_inout
= 0;
3090 * Store the gl_FragDepth layout in the gl_shader_program struct.
3093 store_fragdepth_layout(struct gl_shader_program
*prog
)
3095 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3099 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3101 /* We don't look up the gl_FragDepth symbol directly because if
3102 * gl_FragDepth is not used in the shader, it's removed from the IR.
3103 * However, the symbol won't be removed from the symbol table.
3105 * We're only interested in the cases where the variable is NOT removed
3108 foreach_in_list(ir_instruction
, node
, ir
) {
3109 ir_variable
*const var
= node
->as_variable();
3111 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3115 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3116 switch (var
->data
.depth_layout
) {
3117 case ir_depth_layout_none
:
3118 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3120 case ir_depth_layout_any
:
3121 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3123 case ir_depth_layout_greater
:
3124 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3126 case ir_depth_layout_less
:
3127 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3129 case ir_depth_layout_unchanged
:
3130 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3141 * Validate the resources used by a program versus the implementation limits
3144 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3146 unsigned total_uniform_blocks
= 0;
3147 unsigned total_shader_storage_blocks
= 0;
3149 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3150 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3155 if (sh
->Program
->info
.num_textures
>
3156 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3157 linker_error(prog
, "Too many %s shader texture samplers\n",
3158 _mesa_shader_stage_to_string(i
));
3161 if (sh
->num_uniform_components
>
3162 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3163 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3164 linker_warning(prog
, "Too many %s shader default uniform block "
3165 "components, but the driver will try to optimize "
3166 "them out; this is non-portable out-of-spec "
3168 _mesa_shader_stage_to_string(i
));
3170 linker_error(prog
, "Too many %s shader default uniform block "
3172 _mesa_shader_stage_to_string(i
));
3176 if (sh
->num_combined_uniform_components
>
3177 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3178 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3179 linker_warning(prog
, "Too many %s shader uniform components, "
3180 "but the driver will try to optimize them out; "
3181 "this is non-portable out-of-spec behavior\n",
3182 _mesa_shader_stage_to_string(i
));
3184 linker_error(prog
, "Too many %s shader uniform components\n",
3185 _mesa_shader_stage_to_string(i
));
3189 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3190 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3192 const unsigned max_uniform_blocks
=
3193 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3194 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3195 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3196 _mesa_shader_stage_to_string(i
),
3197 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3200 const unsigned max_shader_storage_blocks
=
3201 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3202 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3203 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3204 _mesa_shader_stage_to_string(i
),
3205 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3209 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3210 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3211 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3214 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3215 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3216 total_shader_storage_blocks
,
3217 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3220 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3221 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3222 ctx
->Const
.MaxUniformBlockSize
) {
3223 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3224 prog
->data
->UniformBlocks
[i
].Name
,
3225 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3226 ctx
->Const
.MaxUniformBlockSize
);
3230 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3231 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3232 ctx
->Const
.MaxShaderStorageBlockSize
) {
3233 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3234 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3235 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3236 ctx
->Const
.MaxShaderStorageBlockSize
);
3242 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3244 unsigned mask
= prog
->data
->linked_stages
;
3246 const int i
= u_bit_scan(&mask
);
3247 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3249 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3250 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3253 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3259 if (p
->sh
.NumSubroutineFunctions
== 0) {
3260 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3263 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3264 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3265 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3266 if (fn
->types
[k
] == uni
->type
) {
3272 uni
->num_compatible_subroutines
= count
;
3278 check_subroutine_resources(struct gl_shader_program
*prog
)
3280 unsigned mask
= prog
->data
->linked_stages
;
3282 const int i
= u_bit_scan(&mask
);
3283 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3285 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3286 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3287 _mesa_shader_stage_to_string(i
));
3292 * Validate shader image resources.
3295 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3297 unsigned total_image_units
= 0;
3298 unsigned fragment_outputs
= 0;
3299 unsigned total_shader_storage_blocks
= 0;
3301 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3304 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3305 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3308 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3309 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3310 _mesa_shader_stage_to_string(i
),
3311 sh
->Program
->info
.num_images
,
3312 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3314 total_image_units
+= sh
->Program
->info
.num_images
;
3315 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3317 if (i
== MESA_SHADER_FRAGMENT
) {
3318 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3319 ir_variable
*var
= node
->as_variable();
3320 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3321 /* since there are no double fs outputs - pass false */
3322 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3328 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3329 linker_error(prog
, "Too many combined image uniforms\n");
3331 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3332 ctx
->Const
.MaxCombinedShaderOutputResources
)
3333 linker_error(prog
, "Too many combined image uniforms, shader storage "
3334 " buffers and fragment outputs\n");
3339 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3340 * for a variable, checks for overlaps between other uniforms using explicit
3344 reserve_explicit_locations(struct gl_shader_program
*prog
,
3345 string_to_uint_map
*map
, ir_variable
*var
)
3347 unsigned slots
= var
->type
->uniform_locations();
3348 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3349 unsigned return_value
= slots
;
3351 /* Resize remap table if locations do not fit in the current one. */
3352 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3353 prog
->UniformRemapTable
=
3354 reralloc(prog
, prog
->UniformRemapTable
,
3355 gl_uniform_storage
*,
3358 if (!prog
->UniformRemapTable
) {
3359 linker_error(prog
, "Out of memory during linking.\n");
3363 /* Initialize allocated space. */
3364 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3365 prog
->UniformRemapTable
[i
] = NULL
;
3367 prog
->NumUniformRemapTable
= max_loc
+ 1;
3370 for (unsigned i
= 0; i
< slots
; i
++) {
3371 unsigned loc
= var
->data
.location
+ i
;
3373 /* Check if location is already used. */
3374 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3376 /* Possibly same uniform from a different stage, this is ok. */
3378 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3383 /* ARB_explicit_uniform_location specification states:
3385 * "No two default-block uniform variables in the program can have
3386 * the same location, even if they are unused, otherwise a compiler
3387 * or linker error will be generated."
3390 "location qualifier for uniform %s overlaps "
3391 "previously used location\n",
3396 /* Initialize location as inactive before optimization
3397 * rounds and location assignment.
3399 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3402 /* Note, base location used for arrays. */
3403 map
->put(var
->data
.location
, var
->name
);
3405 return return_value
;
3409 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3410 struct gl_program
*p
,
3413 unsigned slots
= var
->type
->uniform_locations();
3414 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3416 /* Resize remap table if locations do not fit in the current one. */
3417 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3418 p
->sh
.SubroutineUniformRemapTable
=
3419 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3420 gl_uniform_storage
*,
3423 if (!p
->sh
.SubroutineUniformRemapTable
) {
3424 linker_error(prog
, "Out of memory during linking.\n");
3428 /* Initialize allocated space. */
3429 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3430 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3432 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3435 for (unsigned i
= 0; i
< slots
; i
++) {
3436 unsigned loc
= var
->data
.location
+ i
;
3438 /* Check if location is already used. */
3439 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3441 /* ARB_explicit_uniform_location specification states:
3442 * "No two subroutine uniform variables can have the same location
3443 * in the same shader stage, otherwise a compiler or linker error
3444 * will be generated."
3447 "location qualifier for uniform %s overlaps "
3448 "previously used location\n",
3453 /* Initialize location as inactive before optimization
3454 * rounds and location assignment.
3456 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3462 * Check and reserve all explicit uniform locations, called before
3463 * any optimizations happen to handle also inactive uniforms and
3464 * inactive array elements that may get trimmed away.
3467 check_explicit_uniform_locations(struct gl_context
*ctx
,
3468 struct gl_shader_program
*prog
)
3470 prog
->NumExplicitUniformLocations
= 0;
3472 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3475 /* This map is used to detect if overlapping explicit locations
3476 * occur with the same uniform (from different stage) or a different one.
3478 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3481 linker_error(prog
, "Out of memory during linking.\n");
3485 unsigned entries_total
= 0;
3486 unsigned mask
= prog
->data
->linked_stages
;
3488 const int i
= u_bit_scan(&mask
);
3489 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3491 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3492 ir_variable
*var
= node
->as_variable();
3493 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3496 if (var
->data
.explicit_location
) {
3498 if (var
->type
->without_array()->is_subroutine())
3499 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3501 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3505 entries_total
+= slots
;
3516 struct empty_uniform_block
*current_block
= NULL
;
3518 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3519 /* We found empty space in UniformRemapTable. */
3520 if (prog
->UniformRemapTable
[i
] == NULL
) {
3521 /* We've found the beginning of a new continous block of empty slots */
3522 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3523 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3524 current_block
->start
= i
;
3525 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3526 ¤t_block
->link
);
3529 /* The current block continues, so we simply increment its slots */
3530 current_block
->slots
++;
3535 prog
->NumExplicitUniformLocations
= entries_total
;
3539 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3540 GLenum type
, const char *name
)
3542 bool found_interface
= false;
3543 unsigned block_name_len
= 0;
3544 const char *block_name_dot
= strchr(name
, '.');
3546 /* These rules only apply to buffer variables. So we return
3547 * true for the rest of types.
3549 if (type
!= GL_BUFFER_VARIABLE
)
3552 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3553 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3554 block_name_len
= strlen(block_name
);
3556 const char *block_square_bracket
= strchr(block_name
, '[');
3557 if (block_square_bracket
) {
3558 /* The block is part of an array of named interfaces,
3559 * for the name comparison we ignore the "[x]" part.
3561 block_name_len
-= strlen(block_square_bracket
);
3564 if (block_name_dot
) {
3565 /* Check if the variable name starts with the interface
3566 * name. The interface name (if present) should have the
3567 * length than the interface block name we are comparing to.
3569 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3570 if (len
!= block_name_len
)
3574 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3575 found_interface
= true;
3580 /* We remove the interface name from the buffer variable name,
3581 * including the dot that follows it.
3583 if (found_interface
)
3584 name
= name
+ block_name_len
+ 1;
3586 /* The ARB_program_interface_query spec says:
3588 * "For an active shader storage block member declared as an array, an
3589 * entry will be generated only for the first array element, regardless
3590 * of its type. For arrays of aggregate types, the enumeration rules
3591 * are applied recursively for the single enumerated array element."
3593 const char *struct_first_dot
= strchr(name
, '.');
3594 const char *first_square_bracket
= strchr(name
, '[');
3596 /* The buffer variable is on top level and it is not an array */
3597 if (!first_square_bracket
) {
3599 /* The shader storage block member is a struct, then generate the entry */
3600 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3603 /* Shader storage block member is an array, only generate an entry for the
3604 * first array element.
3606 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3614 add_program_resource(struct gl_shader_program
*prog
,
3615 struct set
*resource_set
,
3616 GLenum type
, const void *data
, uint8_t stages
)
3620 /* If resource already exists, do not add it again. */
3621 if (_mesa_set_search(resource_set
, data
))
3624 prog
->data
->ProgramResourceList
=
3625 reralloc(prog
->data
,
3626 prog
->data
->ProgramResourceList
,
3627 gl_program_resource
,
3628 prog
->data
->NumProgramResourceList
+ 1);
3630 if (!prog
->data
->ProgramResourceList
) {
3631 linker_error(prog
, "Out of memory during linking.\n");
3635 struct gl_program_resource
*res
=
3636 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3640 res
->StageReferences
= stages
;
3642 prog
->data
->NumProgramResourceList
++;
3644 _mesa_set_add(resource_set
, data
);
3649 /* Function checks if a variable var is a packed varying and
3650 * if given name is part of packed varying's list.
3652 * If a variable is a packed varying, it has a name like
3653 * 'packed:a,b,c' where a, b and c are separate variables.
3656 included_in_packed_varying(ir_variable
*var
, const char *name
)
3658 if (strncmp(var
->name
, "packed:", 7) != 0)
3661 char *list
= strdup(var
->name
+ 7);
3666 char *token
= strtok_r(list
, ",", &saveptr
);
3668 if (strcmp(token
, name
) == 0) {
3672 token
= strtok_r(NULL
, ",", &saveptr
);
3679 * Function builds a stage reference bitmask from variable name.
3682 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3687 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3688 * used for reference mask in gl_program_resource will need to be changed.
3690 assert(MESA_SHADER_STAGES
< 8);
3692 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3693 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3697 /* Shader symbol table may contain variables that have
3698 * been optimized away. Search IR for the variable instead.
3700 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3701 ir_variable
*var
= node
->as_variable();
3703 unsigned baselen
= strlen(var
->name
);
3705 if (included_in_packed_varying(var
, name
)) {
3710 /* Type needs to match if specified, otherwise we might
3711 * pick a variable with same name but different interface.
3713 if (var
->data
.mode
!= mode
)
3716 if (strncmp(var
->name
, name
, baselen
) == 0) {
3717 /* Check for exact name matches but also check for arrays and
3720 if (name
[baselen
] == '\0' ||
3721 name
[baselen
] == '[' ||
3722 name
[baselen
] == '.') {
3734 * Create gl_shader_variable from ir_variable class.
3736 static gl_shader_variable
*
3737 create_shader_variable(struct gl_shader_program
*shProg
,
3738 const ir_variable
*in
,
3739 const char *name
, const glsl_type
*type
,
3740 const glsl_type
*interface_type
,
3741 bool use_implicit_location
, int location
,
3742 const glsl_type
*outermost_struct_type
)
3744 /* Allocate zero-initialized memory to ensure that bitfield padding
3747 gl_shader_variable
*out
= rzalloc(shProg
, struct gl_shader_variable
);
3751 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3752 * expect to see gl_VertexID in the program resource list. Pretend.
3754 if (in
->data
.mode
== ir_var_system_value
&&
3755 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3756 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3757 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3758 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3759 (in
->data
.mode
== ir_var_system_value
&&
3760 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3761 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3762 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3763 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3764 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3765 (in
->data
.mode
== ir_var_system_value
&&
3766 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3767 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3768 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3770 out
->name
= ralloc_strdup(shProg
, name
);
3776 /* The ARB_program_interface_query spec says:
3778 * "Not all active variables are assigned valid locations; the
3779 * following variables will have an effective location of -1:
3781 * * uniforms declared as atomic counters;
3783 * * members of a uniform block;
3785 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3787 * * inputs or outputs not declared with a "location" layout
3788 * qualifier, except for vertex shader inputs and fragment shader
3791 if (in
->type
->is_atomic_uint() || is_gl_identifier(in
->name
) ||
3792 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3795 out
->location
= location
;
3799 out
->outermost_struct_type
= outermost_struct_type
;
3800 out
->interface_type
= interface_type
;
3801 out
->component
= in
->data
.location_frac
;
3802 out
->index
= in
->data
.index
;
3803 out
->patch
= in
->data
.patch
;
3804 out
->mode
= in
->data
.mode
;
3805 out
->interpolation
= in
->data
.interpolation
;
3806 out
->explicit_location
= in
->data
.explicit_location
;
3807 out
->precision
= in
->data
.precision
;
3813 add_shader_variable(const struct gl_context
*ctx
,
3814 struct gl_shader_program
*shProg
,
3815 struct set
*resource_set
,
3816 unsigned stage_mask
,
3817 GLenum programInterface
, ir_variable
*var
,
3818 const char *name
, const glsl_type
*type
,
3819 bool use_implicit_location
, int location
,
3820 bool inouts_share_location
,
3821 const glsl_type
*outermost_struct_type
= NULL
)
3823 const glsl_type
*interface_type
= var
->get_interface_type();
3825 if (outermost_struct_type
== NULL
) {
3826 if (var
->data
.from_named_ifc_block
) {
3827 const char *interface_name
= interface_type
->name
;
3829 if (interface_type
->is_array()) {
3830 /* Issue #16 of the ARB_program_interface_query spec says:
3832 * "* If a variable is a member of an interface block without an
3833 * instance name, it is enumerated using just the variable name.
3835 * * If a variable is a member of an interface block with an
3836 * instance name, it is enumerated as "BlockName.Member", where
3837 * "BlockName" is the name of the interface block (not the
3838 * instance name) and "Member" is the name of the variable."
3840 * In particular, it indicates that it should be "BlockName",
3841 * not "BlockName[array length]". The conformance suite and
3842 * dEQP both require this behavior.
3844 * Here, we unwrap the extra array level added by named interface
3845 * block array lowering so we have the correct variable type. We
3846 * also unwrap the interface type when constructing the name.
3848 * We leave interface_type the same so that ES 3.x SSO pipeline
3849 * validation can enforce the rules requiring array length to
3850 * match on interface blocks.
3852 type
= type
->fields
.array
;
3854 interface_name
= interface_type
->fields
.array
->name
;
3857 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3861 switch (type
->base_type
) {
3862 case GLSL_TYPE_STRUCT
: {
3863 /* The ARB_program_interface_query spec says:
3865 * "For an active variable declared as a structure, a separate entry
3866 * will be generated for each active structure member. The name of
3867 * each entry is formed by concatenating the name of the structure,
3868 * the "." character, and the name of the structure member. If a
3869 * structure member to enumerate is itself a structure or array,
3870 * these enumeration rules are applied recursively."
3872 if (outermost_struct_type
== NULL
)
3873 outermost_struct_type
= type
;
3875 unsigned field_location
= location
;
3876 for (unsigned i
= 0; i
< type
->length
; i
++) {
3877 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3878 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3879 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3880 stage_mask
, programInterface
,
3881 var
, field_name
, field
->type
,
3882 use_implicit_location
, field_location
,
3883 false, outermost_struct_type
))
3886 field_location
+= field
->type
->count_attribute_slots(false);
3891 case GLSL_TYPE_ARRAY
: {
3892 /* The ARB_program_interface_query spec says:
3894 * "For an active variable declared as an array of basic types, a
3895 * single entry will be generated, with its name string formed by
3896 * concatenating the name of the array and the string "[0]"."
3898 * "For an active variable declared as an array of an aggregate data
3899 * type (structures or arrays), a separate entry will be generated
3900 * for each active array element, unless noted immediately below.
3901 * The name of each entry is formed by concatenating the name of
3902 * the array, the "[" character, an integer identifying the element
3903 * number, and the "]" character. These enumeration rules are
3904 * applied recursively, treating each enumerated array element as a
3905 * separate active variable."
3907 const struct glsl_type
*array_type
= type
->fields
.array
;
3908 if (array_type
->base_type
== GLSL_TYPE_STRUCT
||
3909 array_type
->base_type
== GLSL_TYPE_ARRAY
) {
3910 unsigned elem_location
= location
;
3911 unsigned stride
= inouts_share_location
? 0 :
3912 array_type
->count_attribute_slots(false);
3913 for (unsigned i
= 0; i
< type
->length
; i
++) {
3914 char *elem
= ralloc_asprintf(shProg
, "%s[%d]", name
, i
);
3915 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3916 stage_mask
, programInterface
,
3917 var
, elem
, array_type
,
3918 use_implicit_location
, elem_location
,
3919 false, outermost_struct_type
))
3921 elem_location
+= stride
;
3929 /* The ARB_program_interface_query spec says:
3931 * "For an active variable declared as a single instance of a basic
3932 * type, a single entry will be generated, using the variable name
3933 * from the shader source."
3935 gl_shader_variable
*sha_v
=
3936 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3937 use_implicit_location
, location
,
3938 outermost_struct_type
);
3942 return add_program_resource(shProg
, resource_set
,
3943 programInterface
, sha_v
, stage_mask
);
3949 inout_has_same_location(const ir_variable
*var
, unsigned stage
)
3951 if (!var
->data
.patch
&&
3952 ((var
->data
.mode
== ir_var_shader_out
&&
3953 stage
== MESA_SHADER_TESS_CTRL
) ||
3954 (var
->data
.mode
== ir_var_shader_in
&&
3955 (stage
== MESA_SHADER_TESS_CTRL
|| stage
== MESA_SHADER_TESS_EVAL
||
3956 stage
== MESA_SHADER_GEOMETRY
))))
3963 add_interface_variables(const struct gl_context
*ctx
,
3964 struct gl_shader_program
*shProg
,
3965 struct set
*resource_set
,
3966 unsigned stage
, GLenum programInterface
)
3968 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3970 foreach_in_list(ir_instruction
, node
, ir
) {
3971 ir_variable
*var
= node
->as_variable();
3973 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3978 switch (var
->data
.mode
) {
3979 case ir_var_system_value
:
3980 case ir_var_shader_in
:
3981 if (programInterface
!= GL_PROGRAM_INPUT
)
3983 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3984 : int(VARYING_SLOT_VAR0
);
3986 case ir_var_shader_out
:
3987 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3989 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3990 : int(VARYING_SLOT_VAR0
);
3996 if (var
->data
.patch
)
3997 loc_bias
= int(VARYING_SLOT_PATCH0
);
3999 /* Skip packed varyings, packed varyings are handled separately
4000 * by add_packed_varyings.
4002 if (strncmp(var
->name
, "packed:", 7) == 0)
4005 /* Skip fragdata arrays, these are handled separately
4006 * by add_fragdata_arrays.
4008 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
4011 const bool vs_input_or_fs_output
=
4012 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
4013 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
4015 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4016 1 << stage
, programInterface
,
4017 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
4018 var
->data
.location
- loc_bias
,
4019 inout_has_same_location(var
, stage
)))
4026 add_packed_varyings(const struct gl_context
*ctx
,
4027 struct gl_shader_program
*shProg
,
4028 struct set
*resource_set
,
4029 int stage
, GLenum type
)
4031 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
4034 if (!sh
|| !sh
->packed_varyings
)
4037 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
4038 ir_variable
*var
= node
->as_variable();
4040 switch (var
->data
.mode
) {
4041 case ir_var_shader_in
:
4042 iface
= GL_PROGRAM_INPUT
;
4044 case ir_var_shader_out
:
4045 iface
= GL_PROGRAM_OUTPUT
;
4048 unreachable("unexpected type");
4051 if (type
== iface
) {
4052 const int stage_mask
=
4053 build_stageref(shProg
, var
->name
, var
->data
.mode
);
4054 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4056 iface
, var
, var
->name
, var
->type
, false,
4057 var
->data
.location
- VARYING_SLOT_VAR0
,
4058 inout_has_same_location(var
, stage
)))
4067 add_fragdata_arrays(const struct gl_context
*ctx
,
4068 struct gl_shader_program
*shProg
,
4069 struct set
*resource_set
)
4071 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4073 if (!sh
|| !sh
->fragdata_arrays
)
4076 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
4077 ir_variable
*var
= node
->as_variable();
4079 assert(var
->data
.mode
== ir_var_shader_out
);
4081 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4082 1 << MESA_SHADER_FRAGMENT
,
4083 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
4084 true, var
->data
.location
- FRAG_RESULT_DATA0
,
4093 get_top_level_name(const char *name
)
4095 const char *first_dot
= strchr(name
, '.');
4096 const char *first_square_bracket
= strchr(name
, '[');
4099 /* The ARB_program_interface_query spec says:
4101 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4102 * the number of active array elements of the top-level shader storage
4103 * block member containing to the active variable is written to
4104 * <params>. If the top-level block member is not declared as an
4105 * array, the value one is written to <params>. If the top-level block
4106 * member is an array with no declared size, the value zero is written
4110 /* The buffer variable is on top level.*/
4111 if (!first_square_bracket
&& !first_dot
)
4112 name_size
= strlen(name
);
4113 else if ((!first_square_bracket
||
4114 (first_dot
&& first_dot
< first_square_bracket
)))
4115 name_size
= first_dot
- name
;
4117 name_size
= first_square_bracket
- name
;
4119 return strndup(name
, name_size
);
4123 get_var_name(const char *name
)
4125 const char *first_dot
= strchr(name
, '.');
4128 return strdup(name
);
4130 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4134 is_top_level_shader_storage_block_member(const char* name
,
4135 const char* interface_name
,
4136 const char* field_name
)
4138 bool result
= false;
4140 /* If the given variable is already a top-level shader storage
4141 * block member, then return array_size = 1.
4142 * We could have two possibilities: if we have an instanced
4143 * shader storage block or not instanced.
4145 * For the first, we check create a name as it was in top level and
4146 * compare it with the real name. If they are the same, then
4147 * the variable is already at top-level.
4149 * Full instanced name is: interface name + '.' + var name +
4152 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4153 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4154 if (!full_instanced_name
) {
4155 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4159 snprintf(full_instanced_name
, name_length
, "%s.%s",
4160 interface_name
, field_name
);
4162 /* Check if its top-level shader storage block member of an
4163 * instanced interface block, or of a unnamed interface block.
4165 if (strcmp(name
, full_instanced_name
) == 0 ||
4166 strcmp(name
, field_name
) == 0)
4169 free(full_instanced_name
);
4174 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4175 char *interface_name
, char *var_name
)
4177 /* The ARB_program_interface_query spec says:
4179 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4180 * the number of active array elements of the top-level shader storage
4181 * block member containing to the active variable is written to
4182 * <params>. If the top-level block member is not declared as an
4183 * array, the value one is written to <params>. If the top-level block
4184 * member is an array with no declared size, the value zero is written
4187 if (is_top_level_shader_storage_block_member(uni
->name
,
4191 else if (field
->type
->is_unsized_array())
4193 else if (field
->type
->is_array())
4194 return field
->type
->length
;
4200 get_array_stride(struct gl_context
*ctx
, struct gl_uniform_storage
*uni
,
4201 const glsl_type
*interface
, const glsl_struct_field
*field
,
4202 char *interface_name
, char *var_name
)
4204 /* The ARB_program_interface_query spec says:
4206 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4207 * identifying the stride between array elements of the top-level
4208 * shader storage block member containing the active variable is
4209 * written to <params>. For top-level block members declared as
4210 * arrays, the value written is the difference, in basic machine units,
4211 * between the offsets of the active variable for consecutive elements
4212 * in the top-level array. For top-level block members not declared as
4213 * an array, zero is written to <params>."
4215 if (field
->type
->is_array()) {
4216 const enum glsl_matrix_layout matrix_layout
=
4217 glsl_matrix_layout(field
->matrix_layout
);
4218 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4219 const glsl_type
*array_type
= field
->type
->fields
.array
;
4221 if (is_top_level_shader_storage_block_member(uni
->name
,
4226 if (GLSL_INTERFACE_PACKING_STD140
==
4228 get_internal_ifc_packing(ctx
->Const
.UseSTD430AsDefaultPacking
)) {
4229 if (array_type
->is_record() || array_type
->is_array())
4230 return glsl_align(array_type
->std140_size(row_major
), 16);
4232 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4234 return array_type
->std430_array_stride(row_major
);
4241 calculate_array_size_and_stride(struct gl_context
*ctx
,
4242 struct gl_shader_program
*shProg
,
4243 struct gl_uniform_storage
*uni
)
4245 int block_index
= uni
->block_index
;
4246 int array_size
= -1;
4247 int array_stride
= -1;
4248 char *var_name
= get_top_level_name(uni
->name
);
4249 char *interface_name
=
4250 get_top_level_name(uni
->is_shader_storage
?
4251 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4252 shProg
->data
->UniformBlocks
[block_index
].Name
);
4254 if (strcmp(var_name
, interface_name
) == 0) {
4255 /* Deal with instanced array of SSBOs */
4256 char *temp_name
= get_var_name(uni
->name
);
4258 linker_error(shProg
, "Out of memory during linking.\n");
4259 goto write_top_level_array_size_and_stride
;
4262 var_name
= get_top_level_name(temp_name
);
4265 linker_error(shProg
, "Out of memory during linking.\n");
4266 goto write_top_level_array_size_and_stride
;
4270 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4271 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4275 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4276 ir_variable
*var
= node
->as_variable();
4277 if (!var
|| !var
->get_interface_type() ||
4278 var
->data
.mode
!= ir_var_shader_storage
)
4281 const glsl_type
*interface
= var
->get_interface_type();
4283 if (strcmp(interface_name
, interface
->name
) != 0)
4286 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4287 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4288 if (strcmp(field
->name
, var_name
) != 0)
4291 array_stride
= get_array_stride(ctx
, uni
, interface
, field
,
4292 interface_name
, var_name
);
4293 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4294 goto write_top_level_array_size_and_stride
;
4298 write_top_level_array_size_and_stride
:
4299 free(interface_name
);
4301 uni
->top_level_array_stride
= array_stride
;
4302 uni
->top_level_array_size
= array_size
;
4306 * Builds up a list of program resources that point to existing
4310 build_program_resource_list(struct gl_context
*ctx
,
4311 struct gl_shader_program
*shProg
)
4313 /* Rebuild resource list. */
4314 if (shProg
->data
->ProgramResourceList
) {
4315 ralloc_free(shProg
->data
->ProgramResourceList
);
4316 shProg
->data
->ProgramResourceList
= NULL
;
4317 shProg
->data
->NumProgramResourceList
= 0;
4320 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4322 /* Determine first input and final output stage. These are used to
4323 * detect which variables should be enumerated in the resource list
4324 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4326 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4327 if (!shProg
->_LinkedShaders
[i
])
4329 if (input_stage
== MESA_SHADER_STAGES
)
4334 /* Empty shader, no resources. */
4335 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4338 struct set
*resource_set
= _mesa_set_create(NULL
,
4340 _mesa_key_pointer_equal
);
4342 /* Program interface needs to expose varyings in case of SSO. */
4343 if (shProg
->SeparateShader
) {
4344 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4345 input_stage
, GL_PROGRAM_INPUT
))
4348 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4349 output_stage
, GL_PROGRAM_OUTPUT
))
4353 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4356 /* Add inputs and outputs to the resource list. */
4357 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4358 input_stage
, GL_PROGRAM_INPUT
))
4361 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4362 output_stage
, GL_PROGRAM_OUTPUT
))
4365 if (shProg
->last_vert_prog
) {
4366 struct gl_transform_feedback_info
*linked_xfb
=
4367 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4369 /* Add transform feedback varyings. */
4370 if (linked_xfb
->NumVarying
> 0) {
4371 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4372 if (!add_program_resource(shProg
, resource_set
,
4373 GL_TRANSFORM_FEEDBACK_VARYING
,
4374 &linked_xfb
->Varyings
[i
], 0))
4379 /* Add transform feedback buffers. */
4380 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4381 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4382 linked_xfb
->Buffers
[i
].Binding
= i
;
4383 if (!add_program_resource(shProg
, resource_set
,
4384 GL_TRANSFORM_FEEDBACK_BUFFER
,
4385 &linked_xfb
->Buffers
[i
], 0))
4391 /* Add uniforms from uniform storage. */
4392 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4393 /* Do not add uniforms internally used by Mesa. */
4394 if (shProg
->data
->UniformStorage
[i
].hidden
)
4398 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4401 /* Add stagereferences for uniforms in a uniform block. */
4402 bool is_shader_storage
=
4403 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4404 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4405 if (block_index
!= -1) {
4406 stageref
|= is_shader_storage
?
4407 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4408 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4411 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4412 if (!should_add_buffer_variable(shProg
, type
,
4413 shProg
->data
->UniformStorage
[i
].name
))
4416 if (is_shader_storage
) {
4417 calculate_array_size_and_stride(ctx
, shProg
,
4418 &shProg
->data
->UniformStorage
[i
]);
4421 if (!add_program_resource(shProg
, resource_set
, type
,
4422 &shProg
->data
->UniformStorage
[i
], stageref
))
4426 /* Add program uniform blocks. */
4427 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4428 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4429 &shProg
->data
->UniformBlocks
[i
], 0))
4433 /* Add program shader storage blocks. */
4434 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4435 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4436 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4440 /* Add atomic counter buffers. */
4441 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4442 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4443 &shProg
->data
->AtomicBuffers
[i
], 0))
4447 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4449 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4452 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4453 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4454 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4457 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4458 /* add shader subroutines */
4459 if (!add_program_resource(shProg
, resource_set
,
4460 type
, &shProg
->data
->UniformStorage
[i
], 0))
4465 unsigned mask
= shProg
->data
->linked_stages
;
4467 const int i
= u_bit_scan(&mask
);
4468 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4470 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4471 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4472 if (!add_program_resource(shProg
, resource_set
,
4473 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4478 _mesa_set_destroy(resource_set
, NULL
);
4482 * This check is done to make sure we allow only constant expression
4483 * indexing and "constant-index-expression" (indexing with an expression
4484 * that includes loop induction variable).
4487 validate_sampler_array_indexing(struct gl_context
*ctx
,
4488 struct gl_shader_program
*prog
)
4490 dynamic_sampler_array_indexing_visitor v
;
4491 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4492 if (prog
->_LinkedShaders
[i
] == NULL
)
4495 bool no_dynamic_indexing
=
4496 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4498 /* Search for array derefs in shader. */
4499 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4500 if (v
.uses_dynamic_sampler_array_indexing()) {
4501 const char *msg
= "sampler arrays indexed with non-constant "
4502 "expressions is forbidden in GLSL %s %u";
4503 /* Backend has indicated that it has no dynamic indexing support. */
4504 if (no_dynamic_indexing
) {
4505 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4506 prog
->data
->Version
);
4509 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4510 prog
->data
->Version
);
4518 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4520 unsigned mask
= prog
->data
->linked_stages
;
4522 const int i
= u_bit_scan(&mask
);
4523 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4525 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4526 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4527 ir_function
*fn
= node
->as_function();
4531 if (fn
->is_subroutine
)
4532 p
->sh
.NumSubroutineUniformTypes
++;
4534 if (!fn
->num_subroutine_types
)
4537 /* these should have been calculated earlier. */
4538 assert(fn
->subroutine_index
!= -1);
4539 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4540 linker_error(prog
, "Too many subroutine functions declared.\n");
4543 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4544 struct gl_subroutine_function
,
4545 p
->sh
.NumSubroutineFunctions
+ 1);
4546 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4547 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4548 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4549 ralloc_array(p
, const struct glsl_type
*,
4550 fn
->num_subroutine_types
);
4552 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4555 * "Each subroutine with an index qualifier in the shader must be
4556 * given a unique index, otherwise a compile or link error will be
4559 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4560 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4561 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4562 linker_error(prog
, "each subroutine index qualifier in the "
4563 "shader must be unique\n");
4567 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4568 fn
->subroutine_index
;
4570 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4571 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4573 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4574 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4575 p
->sh
.NumSubroutineFunctions
++;
4581 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4583 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4585 foreach_in_list(ir_instruction
, node
, ir
) {
4586 ir_variable
*const var
= node
->as_variable();
4588 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4591 /* Don't set always active on builtins that haven't been redeclared */
4592 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4595 var
->data
.always_active_io
= true;
4600 * When separate shader programs are enabled, only input/outputs between
4601 * the stages of a multi-stage separate program can be safely removed
4602 * from the shader interface. Other inputs/outputs must remain active.
4605 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4607 unsigned first
, last
;
4608 assert(prog
->SeparateShader
);
4610 first
= MESA_SHADER_STAGES
;
4613 /* Determine first and last stage. Excluding the compute stage */
4614 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4615 if (!prog
->_LinkedShaders
[i
])
4617 if (first
== MESA_SHADER_STAGES
)
4622 if (first
== MESA_SHADER_STAGES
)
4625 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4626 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4630 /* Prevent the removal of inputs to the first and outputs from the last
4631 * stage, unless they are the initial pipeline inputs or final pipeline
4632 * outputs, respectively.
4634 * The removal of IO between shaders in the same program is always
4637 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4638 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4639 if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4640 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4645 link_and_validate_uniforms(struct gl_context
*ctx
,
4646 struct gl_shader_program
*prog
)
4648 update_array_sizes(prog
);
4649 link_assign_uniform_locations(prog
, ctx
);
4651 link_assign_atomic_counter_resources(ctx
, prog
);
4652 link_calculate_subroutine_compat(prog
);
4653 check_resources(ctx
, prog
);
4654 check_subroutine_resources(prog
);
4655 check_image_resources(ctx
, prog
);
4656 link_check_atomic_counter_resources(ctx
, prog
);
4660 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4661 struct gl_context
*ctx
,
4662 struct gl_shader_program
*prog
, void *mem_ctx
)
4664 /* Mark all generic shader inputs and outputs as unpaired. */
4665 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4666 if (prog
->_LinkedShaders
[i
] != NULL
) {
4667 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4671 unsigned prev
= first
;
4672 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4673 if (prog
->_LinkedShaders
[i
] == NULL
)
4676 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4677 prog
->_LinkedShaders
[i
]);
4681 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4682 MESA_SHADER_VERTEX
)) {
4686 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4687 MESA_SHADER_FRAGMENT
)) {
4691 prog
->last_vert_prog
= NULL
;
4692 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4693 if (prog
->_LinkedShaders
[i
] == NULL
)
4696 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4700 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4703 link_and_validate_uniforms(ctx
, prog
);
4705 if (!prog
->data
->LinkStatus
)
4708 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4709 if (prog
->_LinkedShaders
[i
] == NULL
)
4712 const struct gl_shader_compiler_options
*options
=
4713 &ctx
->Const
.ShaderCompilerOptions
[i
];
4715 if (options
->LowerBufferInterfaceBlocks
)
4716 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4717 options
->ClampBlockIndicesToArrayBounds
,
4718 ctx
->Const
.UseSTD430AsDefaultPacking
);
4720 if (i
== MESA_SHADER_COMPUTE
)
4721 lower_shared_reference(ctx
, prog
, prog
->_LinkedShaders
[i
]);
4723 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4724 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4731 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4734 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4735 /* Run it just once. */
4736 do_common_optimization(ir
, true, false,
4737 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4738 ctx
->Const
.NativeIntegers
);
4740 /* Repeat it until it stops making changes. */
4741 while (do_common_optimization(ir
, true, false,
4742 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4743 ctx
->Const
.NativeIntegers
))
4749 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4751 prog
->data
->LinkStatus
= LINKING_SUCCESS
; /* All error paths will set this to false */
4752 prog
->data
->Validated
= false;
4754 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4756 * "Linking can fail for a variety of reasons as specified in the
4757 * OpenGL Shading Language Specification, as well as any of the
4758 * following reasons:
4760 * - No shader objects are attached to program."
4762 * The Compatibility Profile specification does not list the error. In
4763 * Compatibility Profile missing shader stages are replaced by
4764 * fixed-function. This applies to the case where all stages are
4767 if (prog
->NumShaders
== 0) {
4768 if (ctx
->API
!= API_OPENGL_COMPAT
)
4769 linker_error(prog
, "no shaders attached to the program\n");
4773 #ifdef ENABLE_SHADER_CACHE
4774 /* If transform feedback used on the program then compile all shaders. */
4775 bool skip_cache
= false;
4776 if (prog
->TransformFeedback
.NumVarying
> 0) {
4777 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4778 _mesa_glsl_compile_shader(ctx
, prog
->Shaders
[i
], false, false, true);
4783 if (!skip_cache
&& shader_cache_read_program_metadata(ctx
, prog
))
4787 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4789 prog
->ARB_fragment_coord_conventions_enable
= false;
4791 /* Separate the shaders into groups based on their type.
4793 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4794 unsigned num_shaders
[MESA_SHADER_STAGES
];
4796 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4797 shader_list
[i
] = (struct gl_shader
**)
4798 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4802 unsigned min_version
= UINT_MAX
;
4803 unsigned max_version
= 0;
4804 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4805 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4806 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4808 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4809 linker_error(prog
, "all shaders must use same shading "
4810 "language version\n");
4814 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4815 prog
->ARB_fragment_coord_conventions_enable
= true;
4818 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4819 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4820 num_shaders
[shader_type
]++;
4823 /* In desktop GLSL, different shader versions may be linked together. In
4824 * GLSL ES, all shader versions must be the same.
4826 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4827 linker_error(prog
, "all shaders must use same shading "
4828 "language version\n");
4832 prog
->data
->Version
= max_version
;
4833 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4835 /* Some shaders have to be linked with some other shaders present.
4837 if (!prog
->SeparateShader
) {
4838 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4839 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4840 linker_error(prog
, "Geometry shader must be linked with "
4844 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4845 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4846 linker_error(prog
, "Tessellation evaluation shader must be linked "
4847 "with vertex shader\n");
4850 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4851 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4852 linker_error(prog
, "Tessellation control shader must be linked with "
4857 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4859 * "Linking can fail for [...] any of the following reasons:
4861 * * program contains an object to form a tessellation control
4862 * shader [...] and [...] the program is not separable and
4863 * contains no object to form a tessellation evaluation shader"
4865 * The OpenGL spec is contradictory. It allows linking without a tess
4866 * eval shader, but that can only be used with transform feedback and
4867 * rasterization disabled. However, transform feedback isn't allowed
4868 * with GL_PATCHES, so it can't be used.
4870 * More investigation showed that the idea of transform feedback after
4871 * a tess control shader was dropped, because some hw vendors couldn't
4872 * support tessellation without a tess eval shader, but the linker
4873 * section wasn't updated to reflect that.
4875 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4878 * Do what's reasonable and always require a tess eval shader if a tess
4879 * control shader is present.
4881 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4882 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4883 linker_error(prog
, "Tessellation control shader must be linked with "
4884 "tessellation evaluation shader\n");
4889 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4890 num_shaders
[MESA_SHADER_TESS_CTRL
] == 0) {
4891 linker_error(prog
, "GLSL ES requires non-separable programs "
4892 "containing a tessellation evaluation shader to also "
4893 "be linked with a tessellation control shader\n");
4899 /* Compute shaders have additional restrictions. */
4900 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4901 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4902 linker_error(prog
, "Compute shaders may not be linked with any other "
4903 "type of shader\n");
4906 /* Link all shaders for a particular stage and validate the result.
4908 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4909 if (num_shaders
[stage
] > 0) {
4910 gl_linked_shader
*const sh
=
4911 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4912 num_shaders
[stage
], false);
4914 if (!prog
->data
->LinkStatus
) {
4916 _mesa_delete_linked_shader(ctx
, sh
);
4921 case MESA_SHADER_VERTEX
:
4922 validate_vertex_shader_executable(prog
, sh
, ctx
);
4924 case MESA_SHADER_TESS_CTRL
:
4925 /* nothing to be done */
4927 case MESA_SHADER_TESS_EVAL
:
4928 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4930 case MESA_SHADER_GEOMETRY
:
4931 validate_geometry_shader_executable(prog
, sh
, ctx
);
4933 case MESA_SHADER_FRAGMENT
:
4934 validate_fragment_shader_executable(prog
, sh
);
4937 if (!prog
->data
->LinkStatus
) {
4939 _mesa_delete_linked_shader(ctx
, sh
);
4943 prog
->_LinkedShaders
[stage
] = sh
;
4944 prog
->data
->linked_stages
|= 1 << stage
;
4948 /* Here begins the inter-stage linking phase. Some initial validation is
4949 * performed, then locations are assigned for uniforms, attributes, and
4952 cross_validate_uniforms(prog
);
4953 if (!prog
->data
->LinkStatus
)
4956 unsigned first
, last
, prev
;
4958 first
= MESA_SHADER_STAGES
;
4961 /* Determine first and last stage. */
4962 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4963 if (!prog
->_LinkedShaders
[i
])
4965 if (first
== MESA_SHADER_STAGES
)
4970 check_explicit_uniform_locations(ctx
, prog
);
4971 link_assign_subroutine_types(prog
);
4973 if (!prog
->data
->LinkStatus
)
4976 resize_tes_inputs(ctx
, prog
);
4978 /* Validate the inputs of each stage with the output of the preceding
4982 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4983 if (prog
->_LinkedShaders
[i
] == NULL
)
4986 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4987 prog
->_LinkedShaders
[i
]);
4988 if (!prog
->data
->LinkStatus
)
4991 cross_validate_outputs_to_inputs(ctx
, prog
,
4992 prog
->_LinkedShaders
[prev
],
4993 prog
->_LinkedShaders
[i
]);
4994 if (!prog
->data
->LinkStatus
)
5000 /* The cross validation of outputs/inputs above validates explicit locations
5001 * but for SSO programs we need to do this also for the inputs in the
5002 * first stage and outputs of the last stage included in the program, since
5003 * there is no cross validation for these.
5005 if (prog
->SeparateShader
)
5006 validate_sso_explicit_locations(ctx
, prog
,
5007 (gl_shader_stage
) first
,
5008 (gl_shader_stage
) last
);
5010 /* Cross-validate uniform blocks between shader stages */
5011 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
5012 if (!prog
->data
->LinkStatus
)
5015 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5016 if (prog
->_LinkedShaders
[i
] != NULL
)
5017 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
5020 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
5021 * it before optimization because we want most of the checks to get
5022 * dropped thanks to constant propagation.
5024 * This rule also applies to GLSL ES 3.00.
5026 if (max_version
>= (prog
->IsES
? 300 : 130)) {
5027 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
5029 lower_discard_flow(sh
->ir
);
5033 if (prog
->SeparateShader
)
5034 disable_varying_optimizations_for_sso(prog
);
5037 if (!interstage_cross_validate_uniform_blocks(prog
, false))
5041 if (!interstage_cross_validate_uniform_blocks(prog
, true))
5044 /* Do common optimization before assigning storage for attributes,
5045 * uniforms, and varyings. Later optimization could possibly make
5046 * some of that unused.
5048 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5049 if (prog
->_LinkedShaders
[i
] == NULL
)
5052 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
5053 if (!prog
->data
->LinkStatus
)
5056 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
5057 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
5060 if (ctx
->Const
.LowerTessLevel
) {
5061 lower_tess_level(prog
->_LinkedShaders
[i
]);
5064 /* Call opts before lowering const arrays to uniforms so we can const
5065 * propagate any elements accessed directly.
5067 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5069 /* Call opts after lowering const arrays to copy propagate things. */
5070 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
5071 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5073 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
5076 /* Validation for special cases where we allow sampler array indexing
5077 * with loop induction variable. This check emits a warning or error
5078 * depending if backend can handle dynamic indexing.
5080 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
5081 (prog
->IsES
&& prog
->data
->Version
< 300)) {
5082 if (!validate_sampler_array_indexing(ctx
, prog
))
5086 /* Check and validate stream emissions in geometry shaders */
5087 validate_geometry_shader_emissions(ctx
, prog
);
5089 store_fragdepth_layout(prog
);
5091 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
5094 /* Linking varyings can cause some extra, useless swizzles to be generated
5095 * due to packing and unpacking.
5097 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5098 if (prog
->_LinkedShaders
[i
] == NULL
)
5101 optimize_swizzles(prog
->_LinkedShaders
[i
]->ir
);
5104 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
5105 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
5106 * anything about shader linking when one of the shaders (vertex or
5107 * fragment shader) is absent. So, the extension shouldn't change the
5108 * behavior specified in GLSL specification.
5110 * From OpenGL ES 3.1 specification (7.3 Program Objects):
5111 * "Linking can fail for a variety of reasons as specified in the
5112 * OpenGL ES Shading Language Specification, as well as any of the
5113 * following reasons:
5117 * * program contains objects to form either a vertex shader or
5118 * fragment shader, and program is not separable, and does not
5119 * contain objects to form both a vertex shader and fragment
5122 * However, the only scenario in 3.1+ where we don't require them both is
5123 * when we have a compute shader. For example:
5125 * - No shaders is a link error.
5126 * - Geom or Tess without a Vertex shader is a link error which means we
5127 * always require a Vertex shader and hence a Fragment shader.
5128 * - Finally a Compute shader linked with any other stage is a link error.
5130 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
5131 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
5132 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
5133 linker_error(prog
, "program lacks a vertex shader\n");
5134 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5135 linker_error(prog
, "program lacks a fragment shader\n");
5140 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5141 free(shader_list
[i
]);
5142 if (prog
->_LinkedShaders
[i
] == NULL
)
5145 /* Do a final validation step to make sure that the IR wasn't
5146 * invalidated by any modifications performed after intrastage linking.
5148 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5150 /* Retain any live IR, but trash the rest. */
5151 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5153 /* The symbol table in the linked shaders may contain references to
5154 * variables that were removed (e.g., unused uniforms). Since it may
5155 * contain junk, there is no possible valid use. Delete it and set the
5158 delete prog
->_LinkedShaders
[i
]->symbols
;
5159 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5162 ralloc_free(mem_ctx
);