2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
68 #include "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/prog_instruction.h"
75 #include "program/program.h"
76 #include "util/mesa-sha1.h"
78 #include "string_to_uint_map.h"
80 #include "link_varyings.h"
81 #include "ir_optimization.h"
82 #include "ir_rvalue_visitor.h"
83 #include "ir_uniform.h"
84 #include "builtin_functions.h"
85 #include "shader_cache.h"
87 #include "main/shaderobj.h"
88 #include "main/enums.h"
93 struct find_variable
{
97 find_variable(const char *name
) : name(name
), found(false) {}
101 * Visitor that determines whether or not a variable is ever written.
103 * Use \ref find_assignments for convenience.
105 class find_assignment_visitor
: public ir_hierarchical_visitor
{
107 find_assignment_visitor(unsigned num_vars
,
108 find_variable
* const *vars
)
109 : num_variables(num_vars
), num_found(0), variables(vars
)
113 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
115 ir_variable
*const var
= ir
->lhs
->variable_referenced();
117 return check_variable_name(var
->name
);
120 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
122 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
123 actual_node
, &ir
->actual_parameters
) {
124 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
125 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
127 if (sig_param
->data
.mode
== ir_var_function_out
||
128 sig_param
->data
.mode
== ir_var_function_inout
) {
129 ir_variable
*var
= param_rval
->variable_referenced();
130 if (var
&& check_variable_name(var
->name
) == visit_stop
)
135 if (ir
->return_deref
!= NULL
) {
136 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
138 if (check_variable_name(var
->name
) == visit_stop
)
142 return visit_continue_with_parent
;
146 ir_visitor_status
check_variable_name(const char *name
)
148 for (unsigned i
= 0; i
< num_variables
; ++i
) {
149 if (strcmp(variables
[i
]->name
, name
) == 0) {
150 if (!variables
[i
]->found
) {
151 variables
[i
]->found
= true;
153 assert(num_found
< num_variables
);
154 if (++num_found
== num_variables
)
161 return visit_continue_with_parent
;
165 unsigned num_variables
; /**< Number of variables to find */
166 unsigned num_found
; /**< Number of variables already found */
167 find_variable
* const *variables
; /**< Variables to find */
171 * Determine whether or not any of NULL-terminated list of variables is ever
175 find_assignments(exec_list
*ir
, find_variable
* const *vars
)
177 unsigned num_variables
= 0;
179 for (find_variable
* const *v
= vars
; *v
; ++v
)
182 find_assignment_visitor
visitor(num_variables
, vars
);
187 * Determine whether or not the given variable is ever written to.
190 find_assignments(exec_list
*ir
, find_variable
*var
)
192 find_assignment_visitor
visitor(1, &var
);
197 * Visitor that determines whether or not a variable is ever read.
199 class find_deref_visitor
: public ir_hierarchical_visitor
{
201 find_deref_visitor(const char *name
)
202 : name(name
), found(false)
207 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
209 if (strcmp(this->name
, ir
->var
->name
) == 0) {
214 return visit_continue
;
217 bool variable_found() const
223 const char *name
; /**< Find writes to a variable with this name. */
224 bool found
; /**< Was a write to the variable found? */
229 * A visitor helper that provides methods for updating the types of
230 * ir_dereferences. Classes that update variable types (say, updating
231 * array sizes) will want to use this so that dereference types stay in sync.
233 class deref_type_updater
: public ir_hierarchical_visitor
{
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
243 const glsl_type
*const vt
= ir
->array
->type
;
245 ir
->type
= vt
->fields
.array
;
246 return visit_continue
;
249 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
251 ir
->type
= ir
->record
->type
->fields
.structure
[ir
->field_idx
].type
;
252 return visit_continue
;
257 class array_resize_visitor
: public deref_type_updater
{
259 unsigned num_vertices
;
260 gl_shader_program
*prog
;
261 gl_shader_stage stage
;
263 array_resize_visitor(unsigned num_vertices
,
264 gl_shader_program
*prog
,
265 gl_shader_stage stage
)
267 this->num_vertices
= num_vertices
;
272 virtual ~array_resize_visitor()
277 virtual ir_visitor_status
visit(ir_variable
*var
)
279 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
281 return visit_continue
;
283 unsigned size
= var
->type
->length
;
285 if (stage
== MESA_SHADER_GEOMETRY
) {
286 /* Generate a link error if the shader has declared this array with
289 if (!var
->data
.implicit_sized_array
&&
290 size
&& size
!= this->num_vertices
) {
291 linker_error(this->prog
, "size of array %s declared as %u, "
292 "but number of input vertices is %u\n",
293 var
->name
, size
, this->num_vertices
);
294 return visit_continue
;
297 /* Generate a link error if the shader attempts to access an input
298 * array using an index too large for its actual size assigned at
301 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
302 linker_error(this->prog
, "%s shader accesses element %i of "
303 "%s, but only %i input vertices\n",
304 _mesa_shader_stage_to_string(this->stage
),
305 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
306 return visit_continue
;
310 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
312 var
->data
.max_array_access
= this->num_vertices
- 1;
314 return visit_continue
;
319 * Visitor that determines the highest stream id to which a (geometry) shader
320 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
322 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
324 find_emit_vertex_visitor(int max_allowed
)
325 : max_stream_allowed(max_allowed
),
326 invalid_stream_id(0),
327 invalid_stream_id_from_emit_vertex(false),
328 end_primitive_found(false),
329 uses_non_zero_stream(false)
334 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
336 int stream_id
= ir
->stream_id();
339 invalid_stream_id
= stream_id
;
340 invalid_stream_id_from_emit_vertex
= true;
344 if (stream_id
> max_stream_allowed
) {
345 invalid_stream_id
= stream_id
;
346 invalid_stream_id_from_emit_vertex
= true;
351 uses_non_zero_stream
= true;
353 return visit_continue
;
356 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
358 end_primitive_found
= true;
360 int stream_id
= ir
->stream_id();
363 invalid_stream_id
= stream_id
;
364 invalid_stream_id_from_emit_vertex
= false;
368 if (stream_id
> max_stream_allowed
) {
369 invalid_stream_id
= stream_id
;
370 invalid_stream_id_from_emit_vertex
= false;
375 uses_non_zero_stream
= true;
377 return visit_continue
;
382 return invalid_stream_id
!= 0;
385 const char *error_func()
387 return invalid_stream_id_from_emit_vertex
?
388 "EmitStreamVertex" : "EndStreamPrimitive";
393 return invalid_stream_id
;
398 return uses_non_zero_stream
;
401 bool uses_end_primitive()
403 return end_primitive_found
;
407 int max_stream_allowed
;
408 int invalid_stream_id
;
409 bool invalid_stream_id_from_emit_vertex
;
410 bool end_primitive_found
;
411 bool uses_non_zero_stream
;
414 /* Class that finds array derefs and check if indexes are dynamic. */
415 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
418 dynamic_sampler_array_indexing_visitor() :
419 dynamic_sampler_array_indexing(false)
423 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
425 if (!ir
->variable_referenced())
426 return visit_continue
;
428 if (!ir
->variable_referenced()->type
->contains_sampler())
429 return visit_continue
;
431 if (!ir
->array_index
->constant_expression_value(ralloc_parent(ir
))) {
432 dynamic_sampler_array_indexing
= true;
435 return visit_continue
;
438 bool uses_dynamic_sampler_array_indexing()
440 return dynamic_sampler_array_indexing
;
444 bool dynamic_sampler_array_indexing
;
447 } /* anonymous namespace */
450 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
454 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
456 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
459 prog
->data
->LinkStatus
= LINKING_FAILURE
;
464 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
468 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
470 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
477 * Given a string identifying a program resource, break it into a base name
478 * and an optional array index in square brackets.
480 * If an array index is present, \c out_base_name_end is set to point to the
481 * "[" that precedes the array index, and the array index itself is returned
484 * If no array index is present (or if the array index is negative or
485 * mal-formed), \c out_base_name_end, is set to point to the null terminator
486 * at the end of the input string, and -1 is returned.
488 * Only the final array index is parsed; if the string contains other array
489 * indices (or structure field accesses), they are left in the base name.
491 * No attempt is made to check that the base name is properly formed;
492 * typically the caller will look up the base name in a hash table, so
493 * ill-formed base names simply turn into hash table lookup failures.
496 parse_program_resource_name(const GLchar
*name
,
497 const GLchar
**out_base_name_end
)
499 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
501 * "When an integer array element or block instance number is part of
502 * the name string, it will be specified in decimal form without a "+"
503 * or "-" sign or any extra leading zeroes. Additionally, the name
504 * string will not include white space anywhere in the string."
507 const size_t len
= strlen(name
);
508 *out_base_name_end
= name
+ len
;
510 if (len
== 0 || name
[len
-1] != ']')
513 /* Walk backwards over the string looking for a non-digit character. This
514 * had better be the opening bracket for an array index.
516 * Initially, i specifies the location of the ']'. Since the string may
517 * contain only the ']' charcater, walk backwards very carefully.
520 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
523 if ((i
== 0) || name
[i
-1] != '[')
526 long array_index
= strtol(&name
[i
], NULL
, 10);
530 /* Check for leading zero */
531 if (name
[i
] == '0' && name
[i
+1] != ']')
534 *out_base_name_end
= name
+ (i
- 1);
540 link_invalidate_variable_locations(exec_list
*ir
)
542 foreach_in_list(ir_instruction
, node
, ir
) {
543 ir_variable
*const var
= node
->as_variable();
548 /* Only assign locations for variables that lack an explicit location.
549 * Explicit locations are set for all built-in variables, generic vertex
550 * shader inputs (via layout(location=...)), and generic fragment shader
551 * outputs (also via layout(location=...)).
553 if (!var
->data
.explicit_location
) {
554 var
->data
.location
= -1;
555 var
->data
.location_frac
= 0;
558 /* ir_variable::is_unmatched_generic_inout is used by the linker while
559 * connecting outputs from one stage to inputs of the next stage.
561 if (var
->data
.explicit_location
&&
562 var
->data
.location
< VARYING_SLOT_VAR0
) {
563 var
->data
.is_unmatched_generic_inout
= 0;
565 var
->data
.is_unmatched_generic_inout
= 1;
572 * Set clip_distance_array_size based and cull_distance_array_size on the given
575 * Also check for errors based on incorrect usage of gl_ClipVertex and
576 * gl_ClipDistance and gl_CullDistance.
577 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
578 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
580 * Return false if an error was reported.
583 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
584 struct gl_linked_shader
*shader
,
585 struct gl_context
*ctx
,
586 GLuint
*clip_distance_array_size
,
587 GLuint
*cull_distance_array_size
)
589 *clip_distance_array_size
= 0;
590 *cull_distance_array_size
= 0;
592 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
593 /* From section 7.1 (Vertex Shader Special Variables) of the
596 * "It is an error for a shader to statically write both
597 * gl_ClipVertex and gl_ClipDistance."
599 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
600 * gl_ClipVertex nor gl_ClipDistance. However with
601 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
603 find_variable
gl_ClipDistance("gl_ClipDistance");
604 find_variable
gl_CullDistance("gl_CullDistance");
605 find_variable
gl_ClipVertex("gl_ClipVertex");
606 find_variable
* const variables
[] = {
609 !prog
->IsES
? &gl_ClipVertex
: NULL
,
612 find_assignments(shader
->ir
, variables
);
614 /* From the ARB_cull_distance spec:
616 * It is a compile-time or link-time error for the set of shaders forming
617 * a program to statically read or write both gl_ClipVertex and either
618 * gl_ClipDistance or gl_CullDistance.
620 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
624 if (gl_ClipVertex
.found
&& gl_ClipDistance
.found
) {
625 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
626 "and `gl_ClipDistance'\n",
627 _mesa_shader_stage_to_string(shader
->Stage
));
630 if (gl_ClipVertex
.found
&& gl_CullDistance
.found
) {
631 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
632 "and `gl_CullDistance'\n",
633 _mesa_shader_stage_to_string(shader
->Stage
));
638 if (gl_ClipDistance
.found
) {
639 ir_variable
*clip_distance_var
=
640 shader
->symbols
->get_variable("gl_ClipDistance");
641 assert(clip_distance_var
);
642 *clip_distance_array_size
= clip_distance_var
->type
->length
;
644 if (gl_CullDistance
.found
) {
645 ir_variable
*cull_distance_var
=
646 shader
->symbols
->get_variable("gl_CullDistance");
647 assert(cull_distance_var
);
648 *cull_distance_array_size
= cull_distance_var
->type
->length
;
650 /* From the ARB_cull_distance spec:
652 * It is a compile-time or link-time error for the set of shaders forming
653 * a program to have the sum of the sizes of the gl_ClipDistance and
654 * gl_CullDistance arrays to be larger than
655 * gl_MaxCombinedClipAndCullDistances.
657 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
658 ctx
->Const
.MaxClipPlanes
) {
659 linker_error(prog
, "%s shader: the combined size of "
660 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
662 "gl_MaxCombinedClipAndCullDistances (%u)",
663 _mesa_shader_stage_to_string(shader
->Stage
),
664 ctx
->Const
.MaxClipPlanes
);
671 * Verify that a vertex shader executable meets all semantic requirements.
673 * Also sets info.clip_distance_array_size and
674 * info.cull_distance_array_size as a side effect.
676 * \param shader Vertex shader executable to be verified
679 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
680 struct gl_linked_shader
*shader
,
681 struct gl_context
*ctx
)
686 /* From the GLSL 1.10 spec, page 48:
688 * "The variable gl_Position is available only in the vertex
689 * language and is intended for writing the homogeneous vertex
690 * position. All executions of a well-formed vertex shader
691 * executable must write a value into this variable. [...] The
692 * variable gl_Position is available only in the vertex
693 * language and is intended for writing the homogeneous vertex
694 * position. All executions of a well-formed vertex shader
695 * executable must write a value into this variable."
697 * while in GLSL 1.40 this text is changed to:
699 * "The variable gl_Position is available only in the vertex
700 * language and is intended for writing the homogeneous vertex
701 * position. It can be written at any time during shader
702 * execution. It may also be read back by a vertex shader
703 * after being written. This value will be used by primitive
704 * assembly, clipping, culling, and other fixed functionality
705 * operations, if present, that operate on primitives after
706 * vertex processing has occurred. Its value is undefined if
707 * the vertex shader executable does not write gl_Position."
709 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
710 * gl_Position is not an error.
712 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
713 find_variable
gl_Position("gl_Position");
714 find_assignments(shader
->ir
, &gl_Position
);
715 if (!gl_Position
.found
) {
718 "vertex shader does not write to `gl_Position'. "
719 "Its value is undefined. \n");
722 "vertex shader does not write to `gl_Position'. \n");
728 analyze_clip_cull_usage(prog
, shader
, ctx
,
729 &shader
->Program
->info
.clip_distance_array_size
,
730 &shader
->Program
->info
.cull_distance_array_size
);
734 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
735 struct gl_linked_shader
*shader
,
736 struct gl_context
*ctx
)
741 analyze_clip_cull_usage(prog
, shader
, ctx
,
742 &shader
->Program
->info
.clip_distance_array_size
,
743 &shader
->Program
->info
.cull_distance_array_size
);
748 * Verify that a fragment shader executable meets all semantic requirements
750 * \param shader Fragment shader executable to be verified
753 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
754 struct gl_linked_shader
*shader
)
759 find_variable
gl_FragColor("gl_FragColor");
760 find_variable
gl_FragData("gl_FragData");
761 find_variable
* const variables
[] = { &gl_FragColor
, &gl_FragData
, NULL
};
762 find_assignments(shader
->ir
, variables
);
764 if (gl_FragColor
.found
&& gl_FragData
.found
) {
765 linker_error(prog
, "fragment shader writes to both "
766 "`gl_FragColor' and `gl_FragData'\n");
771 * Verify that a geometry shader executable meets all semantic requirements
773 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
774 * info.cull_distance_array_size as a side effect.
776 * \param shader Geometry shader executable to be verified
779 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
780 struct gl_linked_shader
*shader
,
781 struct gl_context
*ctx
)
786 unsigned num_vertices
=
787 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
788 prog
->Geom
.VerticesIn
= num_vertices
;
790 analyze_clip_cull_usage(prog
, shader
, ctx
,
791 &shader
->Program
->info
.clip_distance_array_size
,
792 &shader
->Program
->info
.cull_distance_array_size
);
796 * Check if geometry shaders emit to non-zero streams and do corresponding
800 validate_geometry_shader_emissions(struct gl_context
*ctx
,
801 struct gl_shader_program
*prog
)
803 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
806 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
807 emit_vertex
.run(sh
->ir
);
808 if (emit_vertex
.error()) {
809 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
810 "stream parameter are in the range [0, %d].\n",
811 emit_vertex
.error_func(),
812 emit_vertex
.error_stream(),
813 ctx
->Const
.MaxVertexStreams
- 1);
815 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
816 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
818 /* From the ARB_gpu_shader5 spec:
820 * "Multiple vertex streams are supported only if the output primitive
821 * type is declared to be "points". A program will fail to link if it
822 * contains a geometry shader calling EmitStreamVertex() or
823 * EndStreamPrimitive() if its output primitive type is not "points".
825 * However, in the same spec:
827 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
828 * with <stream> set to zero."
832 * "The function EndPrimitive() is equivalent to calling
833 * EndStreamPrimitive() with <stream> set to zero."
835 * Since we can call EmitVertex() and EndPrimitive() when we output
836 * primitives other than points, calling EmitStreamVertex(0) or
837 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
838 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
839 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
842 if (prog
->Geom
.UsesStreams
&&
843 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
844 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
845 "with n>0 requires point output\n");
851 validate_intrastage_arrays(struct gl_shader_program
*prog
,
852 ir_variable
*const var
,
853 ir_variable
*const existing
)
855 /* Consider the types to be "the same" if both types are arrays
856 * of the same type and one of the arrays is implicitly sized.
857 * In addition, set the type of the linked variable to the
858 * explicitly sized array.
860 if (var
->type
->is_array() && existing
->type
->is_array()) {
861 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
862 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
863 if (var
->type
->length
!= 0) {
864 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
865 linker_error(prog
, "%s `%s' declared as type "
866 "`%s' but outermost dimension has an index"
869 var
->name
, var
->type
->name
,
870 existing
->data
.max_array_access
);
872 existing
->type
= var
->type
;
874 } else if (existing
->type
->length
!= 0) {
875 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
876 !existing
->data
.from_ssbo_unsized_array
) {
877 linker_error(prog
, "%s `%s' declared as type "
878 "`%s' but outermost dimension has an index"
881 var
->name
, existing
->type
->name
,
882 var
->data
.max_array_access
);
893 * Perform validation of global variables used across multiple shaders
896 cross_validate_globals(struct gl_shader_program
*prog
,
897 struct exec_list
*ir
, glsl_symbol_table
*variables
,
900 foreach_in_list(ir_instruction
, node
, ir
) {
901 ir_variable
*const var
= node
->as_variable();
906 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
909 /* don't cross validate subroutine uniforms */
910 if (var
->type
->contains_subroutine())
913 /* Don't cross validate interface instances. These are only relevant
914 * inside a shader. The cross validation is done at the Interface Block
917 if (var
->is_interface_instance())
920 /* Don't cross validate temporaries that are at global scope. These
921 * will eventually get pulled into the shaders 'main'.
923 if (var
->data
.mode
== ir_var_temporary
)
926 /* If a global with this name has already been seen, verify that the
927 * new instance has the same type. In addition, if the globals have
928 * initializers, the values of the initializers must be the same.
930 ir_variable
*const existing
= variables
->get_variable(var
->name
);
931 if (existing
!= NULL
) {
932 /* Check if types match. */
933 if (var
->type
!= existing
->type
) {
934 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
935 /* If it is an unsized array in a Shader Storage Block,
936 * two different shaders can access to different elements.
937 * Because of that, they might be converted to different
938 * sized arrays, then check that they are compatible but
939 * ignore the array size.
941 if (!(var
->data
.mode
== ir_var_shader_storage
&&
942 var
->data
.from_ssbo_unsized_array
&&
943 existing
->data
.mode
== ir_var_shader_storage
&&
944 existing
->data
.from_ssbo_unsized_array
&&
945 var
->type
->gl_type
== existing
->type
->gl_type
)) {
946 linker_error(prog
, "%s `%s' declared as type "
947 "`%s' and type `%s'\n",
949 var
->name
, var
->type
->name
,
950 existing
->type
->name
);
956 if (var
->data
.explicit_location
) {
957 if (existing
->data
.explicit_location
958 && (var
->data
.location
!= existing
->data
.location
)) {
959 linker_error(prog
, "explicit locations for %s "
960 "`%s' have differing values\n",
961 mode_string(var
), var
->name
);
965 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
966 linker_error(prog
, "explicit components for %s `%s' have "
967 "differing values\n", mode_string(var
), var
->name
);
971 existing
->data
.location
= var
->data
.location
;
972 existing
->data
.explicit_location
= true;
974 /* Check if uniform with implicit location was marked explicit
975 * by earlier shader stage. If so, mark it explicit in this stage
976 * too to make sure later processing does not treat it as
979 if (existing
->data
.explicit_location
) {
980 var
->data
.location
= existing
->data
.location
;
981 var
->data
.explicit_location
= true;
985 /* From the GLSL 4.20 specification:
986 * "A link error will result if two compilation units in a program
987 * specify different integer-constant bindings for the same
988 * opaque-uniform name. However, it is not an error to specify a
989 * binding on some but not all declarations for the same name"
991 if (var
->data
.explicit_binding
) {
992 if (existing
->data
.explicit_binding
&&
993 var
->data
.binding
!= existing
->data
.binding
) {
994 linker_error(prog
, "explicit bindings for %s "
995 "`%s' have differing values\n",
996 mode_string(var
), var
->name
);
1000 existing
->data
.binding
= var
->data
.binding
;
1001 existing
->data
.explicit_binding
= true;
1004 if (var
->type
->contains_atomic() &&
1005 var
->data
.offset
!= existing
->data
.offset
) {
1006 linker_error(prog
, "offset specifications for %s "
1007 "`%s' have differing values\n",
1008 mode_string(var
), var
->name
);
1012 /* Validate layout qualifiers for gl_FragDepth.
1014 * From the AMD/ARB_conservative_depth specs:
1016 * "If gl_FragDepth is redeclared in any fragment shader in a
1017 * program, it must be redeclared in all fragment shaders in
1018 * that program that have static assignments to
1019 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1020 * fragment shaders in a single program must have the same set
1023 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1024 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1025 bool layout_differs
=
1026 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1028 if (layout_declared
&& layout_differs
) {
1030 "All redeclarations of gl_FragDepth in all "
1031 "fragment shaders in a single program must have "
1032 "the same set of qualifiers.\n");
1035 if (var
->data
.used
&& layout_differs
) {
1037 "If gl_FragDepth is redeclared with a layout "
1038 "qualifier in any fragment shader, it must be "
1039 "redeclared with the same layout qualifier in "
1040 "all fragment shaders that have assignments to "
1045 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1047 * "If a shared global has multiple initializers, the
1048 * initializers must all be constant expressions, and they
1049 * must all have the same value. Otherwise, a link error will
1050 * result. (A shared global having only one initializer does
1051 * not require that initializer to be a constant expression.)"
1053 * Previous to 4.20 the GLSL spec simply said that initializers
1054 * must have the same value. In this case of non-constant
1055 * initializers, this was impossible to determine. As a result,
1056 * no vendor actually implemented that behavior. The 4.20
1057 * behavior matches the implemented behavior of at least one other
1058 * vendor, so we'll implement that for all GLSL versions.
1060 if (var
->constant_initializer
!= NULL
) {
1061 if (existing
->constant_initializer
!= NULL
) {
1062 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1063 linker_error(prog
, "initializers for %s "
1064 "`%s' have differing values\n",
1065 mode_string(var
), var
->name
);
1069 /* If the first-seen instance of a particular uniform did
1070 * not have an initializer but a later instance does,
1071 * replace the former with the later.
1073 variables
->replace_variable(existing
->name
, var
);
1077 if (var
->data
.has_initializer
) {
1078 if (existing
->data
.has_initializer
1079 && (var
->constant_initializer
== NULL
1080 || existing
->constant_initializer
== NULL
)) {
1082 "shared global variable `%s' has multiple "
1083 "non-constant initializers.\n",
1089 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1090 linker_error(prog
, "declarations for %s `%s' have "
1091 "mismatching invariant qualifiers\n",
1092 mode_string(var
), var
->name
);
1095 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1096 linker_error(prog
, "declarations for %s `%s' have "
1097 "mismatching centroid qualifiers\n",
1098 mode_string(var
), var
->name
);
1101 if (existing
->data
.sample
!= var
->data
.sample
) {
1102 linker_error(prog
, "declarations for %s `%s` have "
1103 "mismatching sample qualifiers\n",
1104 mode_string(var
), var
->name
);
1107 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1108 linker_error(prog
, "declarations for %s `%s` have "
1109 "mismatching image format qualifiers\n",
1110 mode_string(var
), var
->name
);
1114 /* Only in GLSL ES 3.10, the precision qualifier should not match
1115 * between block members defined in matched block names within a
1118 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1119 * member should match.
1121 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1122 !var
->get_interface_type()) &&
1123 existing
->data
.precision
!= var
->data
.precision
) {
1124 if ((existing
->data
.used
&& var
->data
.used
) || prog
->data
->Version
>= 300) {
1125 linker_error(prog
, "declarations for %s `%s` have "
1126 "mismatching precision qualifiers\n",
1127 mode_string(var
), var
->name
);
1130 linker_warning(prog
, "declarations for %s `%s` have "
1131 "mismatching precision qualifiers\n",
1132 mode_string(var
), var
->name
);
1136 /* In OpenGL GLSL 3.20 spec, section 4.3.9:
1138 * "It is a link-time error if any particular shader interface
1141 * - two different blocks, each having no instance name, and each
1142 * having a member of the same name, or
1144 * - a variable outside a block, and a block with no instance name,
1145 * where the variable has the same name as a member in the block."
1147 const glsl_type
*var_itype
= var
->get_interface_type();
1148 const glsl_type
*existing_itype
= existing
->get_interface_type();
1149 if (var_itype
!= existing_itype
) {
1150 if (!var_itype
|| !existing_itype
) {
1151 linker_error(prog
, "declarations for %s `%s` are inside block "
1152 "`%s` and outside a block",
1153 mode_string(var
), var
->name
,
1154 var_itype
? var_itype
->name
: existing_itype
->name
);
1156 } else if (strcmp(var_itype
->name
, existing_itype
->name
) != 0) {
1157 linker_error(prog
, "declarations for %s `%s` are inside blocks "
1159 mode_string(var
), var
->name
,
1160 existing_itype
->name
,
1166 variables
->add_variable(var
);
1172 * Perform validation of uniforms used across multiple shader stages
1175 cross_validate_uniforms(struct gl_shader_program
*prog
)
1177 glsl_symbol_table variables
;
1178 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1179 if (prog
->_LinkedShaders
[i
] == NULL
)
1182 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1188 * Accumulates the array of buffer blocks and checks that all definitions of
1189 * blocks agree on their contents.
1192 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1195 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1196 struct gl_uniform_block
*blks
= NULL
;
1197 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1198 &prog
->data
->NumUniformBlocks
;
1200 unsigned max_num_buffer_blocks
= 0;
1201 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1202 if (prog
->_LinkedShaders
[i
]) {
1203 if (validate_ssbo
) {
1204 max_num_buffer_blocks
+=
1205 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1207 max_num_buffer_blocks
+=
1208 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1213 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1214 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1216 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1217 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1218 InterfaceBlockStageIndex
[i
][j
] = -1;
1223 unsigned sh_num_blocks
;
1224 struct gl_uniform_block
**sh_blks
;
1225 if (validate_ssbo
) {
1226 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1227 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1229 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1230 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1233 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1234 int index
= link_cross_validate_uniform_block(prog
->data
, &blks
,
1235 num_blks
, sh_blks
[j
]);
1238 linker_error(prog
, "buffer block `%s' has mismatching "
1239 "definitions\n", sh_blks
[j
]->Name
);
1241 for (unsigned k
= 0; k
<= i
; k
++) {
1242 delete[] InterfaceBlockStageIndex
[k
];
1245 /* Reset the block count. This will help avoid various segfaults
1246 * from api calls that assume the array exists due to the count
1253 InterfaceBlockStageIndex
[i
][index
] = j
;
1257 /* Update per stage block pointers to point to the program list.
1258 * FIXME: We should be able to free the per stage blocks here.
1260 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1261 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1262 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1264 if (stage_index
!= -1) {
1265 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1267 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1268 sh
->Program
->sh
.ShaderStorageBlocks
:
1269 sh
->Program
->sh
.UniformBlocks
;
1271 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1272 sh_blks
[stage_index
] = &blks
[j
];
1277 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1278 delete[] InterfaceBlockStageIndex
[i
];
1282 prog
->data
->ShaderStorageBlocks
= blks
;
1284 prog
->data
->UniformBlocks
= blks
;
1291 * Populates a shaders symbol table with all global declarations
1294 populate_symbol_table(gl_linked_shader
*sh
, glsl_symbol_table
*symbols
)
1296 sh
->symbols
= new(sh
) glsl_symbol_table
;
1298 _mesa_glsl_copy_symbols_from_table(sh
->ir
, symbols
, sh
->symbols
);
1303 * Remap variables referenced in an instruction tree
1305 * This is used when instruction trees are cloned from one shader and placed in
1306 * another. These trees will contain references to \c ir_variable nodes that
1307 * do not exist in the target shader. This function finds these \c ir_variable
1308 * references and replaces the references with matching variables in the target
1311 * If there is no matching variable in the target shader, a clone of the
1312 * \c ir_variable is made and added to the target shader. The new variable is
1313 * added to \b both the instruction stream and the symbol table.
1315 * \param inst IR tree that is to be processed.
1316 * \param symbols Symbol table containing global scope symbols in the
1318 * \param instructions Instruction stream where new variable declarations
1322 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1325 class remap_visitor
: public ir_hierarchical_visitor
{
1327 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1329 this->target
= target
;
1330 this->symbols
= target
->symbols
;
1331 this->instructions
= target
->ir
;
1332 this->temps
= temps
;
1335 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1337 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1338 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1339 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1341 assert(var
!= NULL
);
1343 return visit_continue
;
1346 ir_variable
*const existing
=
1347 this->symbols
->get_variable(ir
->var
->name
);
1348 if (existing
!= NULL
)
1351 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1353 this->symbols
->add_variable(copy
);
1354 this->instructions
->push_head(copy
);
1358 return visit_continue
;
1362 struct gl_linked_shader
*target
;
1363 glsl_symbol_table
*symbols
;
1364 exec_list
*instructions
;
1368 remap_visitor
v(target
, temps
);
1375 * Move non-declarations from one instruction stream to another
1377 * The intended usage pattern of this function is to pass the pointer to the
1378 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1379 * pointer) for \c last and \c false for \c make_copies on the first
1380 * call. Successive calls pass the return value of the previous call for
1381 * \c last and \c true for \c make_copies.
1383 * \param instructions Source instruction stream
1384 * \param last Instruction after which new instructions should be
1385 * inserted in the target instruction stream
1386 * \param make_copies Flag selecting whether instructions in \c instructions
1387 * should be copied (via \c ir_instruction::clone) into the
1388 * target list or moved.
1391 * The new "last" instruction in the target instruction stream. This pointer
1392 * is suitable for use as the \c last parameter of a later call to this
1396 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1397 bool make_copies
, gl_linked_shader
*target
)
1399 hash_table
*temps
= NULL
;
1402 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1403 _mesa_key_pointer_equal
);
1405 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1406 if (inst
->as_function())
1409 ir_variable
*var
= inst
->as_variable();
1410 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1413 assert(inst
->as_assignment()
1415 || inst
->as_if() /* for initializers with the ?: operator */
1416 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1419 inst
= inst
->clone(target
, NULL
);
1422 _mesa_hash_table_insert(temps
, var
, inst
);
1424 remap_variables(inst
, target
, temps
);
1429 last
->insert_after(inst
);
1434 _mesa_hash_table_destroy(temps
, NULL
);
1441 * This class is only used in link_intrastage_shaders() below but declaring
1442 * it inside that function leads to compiler warnings with some versions of
1445 class array_sizing_visitor
: public deref_type_updater
{
1447 array_sizing_visitor()
1448 : mem_ctx(ralloc_context(NULL
)),
1449 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1450 _mesa_key_pointer_equal
))
1454 ~array_sizing_visitor()
1456 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1457 ralloc_free(this->mem_ctx
);
1460 virtual ir_visitor_status
visit(ir_variable
*var
)
1462 const glsl_type
*type_without_array
;
1463 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1464 fixup_type(&var
->type
, var
->data
.max_array_access
,
1465 var
->data
.from_ssbo_unsized_array
,
1466 &implicit_sized_array
);
1467 var
->data
.implicit_sized_array
= implicit_sized_array
;
1468 type_without_array
= var
->type
->without_array();
1469 if (var
->type
->is_interface()) {
1470 if (interface_contains_unsized_arrays(var
->type
)) {
1471 const glsl_type
*new_type
=
1472 resize_interface_members(var
->type
,
1473 var
->get_max_ifc_array_access(),
1474 var
->is_in_shader_storage_block());
1475 var
->type
= new_type
;
1476 var
->change_interface_type(new_type
);
1478 } else if (type_without_array
->is_interface()) {
1479 if (interface_contains_unsized_arrays(type_without_array
)) {
1480 const glsl_type
*new_type
=
1481 resize_interface_members(type_without_array
,
1482 var
->get_max_ifc_array_access(),
1483 var
->is_in_shader_storage_block());
1484 var
->change_interface_type(new_type
);
1485 var
->type
= update_interface_members_array(var
->type
, new_type
);
1487 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1488 /* Store a pointer to the variable in the unnamed_interfaces
1492 _mesa_hash_table_search(this->unnamed_interfaces
,
1495 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1497 if (interface_vars
== NULL
) {
1498 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1500 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1503 unsigned index
= ifc_type
->field_index(var
->name
);
1504 assert(index
< ifc_type
->length
);
1505 assert(interface_vars
[index
] == NULL
);
1506 interface_vars
[index
] = var
;
1508 return visit_continue
;
1512 * For each unnamed interface block that was discovered while running the
1513 * visitor, adjust the interface type to reflect the newly assigned array
1514 * sizes, and fix up the ir_variable nodes to point to the new interface
1517 void fixup_unnamed_interface_types()
1519 hash_table_call_foreach(this->unnamed_interfaces
,
1520 fixup_unnamed_interface_type
, NULL
);
1525 * If the type pointed to by \c type represents an unsized array, replace
1526 * it with a sized array whose size is determined by max_array_access.
1528 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1529 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1531 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1532 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1533 max_array_access
+ 1);
1534 *implicit_sized
= true;
1535 assert(*type
!= NULL
);
1539 static const glsl_type
*
1540 update_interface_members_array(const glsl_type
*type
,
1541 const glsl_type
*new_interface_type
)
1543 const glsl_type
*element_type
= type
->fields
.array
;
1544 if (element_type
->is_array()) {
1545 const glsl_type
*new_array_type
=
1546 update_interface_members_array(element_type
, new_interface_type
);
1547 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1549 return glsl_type::get_array_instance(new_interface_type
,
1555 * Determine whether the given interface type contains unsized arrays (if
1556 * it doesn't, array_sizing_visitor doesn't need to process it).
1558 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1560 for (unsigned i
= 0; i
< type
->length
; i
++) {
1561 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1562 if (elem_type
->is_unsized_array())
1569 * Create a new interface type based on the given type, with unsized arrays
1570 * replaced by sized arrays whose size is determined by
1571 * max_ifc_array_access.
1573 static const glsl_type
*
1574 resize_interface_members(const glsl_type
*type
,
1575 const int *max_ifc_array_access
,
1578 unsigned num_fields
= type
->length
;
1579 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1580 memcpy(fields
, type
->fields
.structure
,
1581 num_fields
* sizeof(*fields
));
1582 for (unsigned i
= 0; i
< num_fields
; i
++) {
1583 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1584 /* If SSBO last member is unsized array, we don't replace it by a sized
1587 if (is_ssbo
&& i
== (num_fields
- 1))
1588 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1589 true, &implicit_sized_array
);
1591 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1592 false, &implicit_sized_array
);
1593 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1595 glsl_interface_packing packing
=
1596 (glsl_interface_packing
) type
->interface_packing
;
1597 bool row_major
= (bool) type
->interface_row_major
;
1598 const glsl_type
*new_ifc_type
=
1599 glsl_type::get_interface_instance(fields
, num_fields
,
1600 packing
, row_major
, type
->name
);
1602 return new_ifc_type
;
1605 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1608 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1609 ir_variable
**interface_vars
= (ir_variable
**) data
;
1610 unsigned num_fields
= ifc_type
->length
;
1611 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1612 memcpy(fields
, ifc_type
->fields
.structure
,
1613 num_fields
* sizeof(*fields
));
1614 bool interface_type_changed
= false;
1615 for (unsigned i
= 0; i
< num_fields
; i
++) {
1616 if (interface_vars
[i
] != NULL
&&
1617 fields
[i
].type
!= interface_vars
[i
]->type
) {
1618 fields
[i
].type
= interface_vars
[i
]->type
;
1619 interface_type_changed
= true;
1622 if (!interface_type_changed
) {
1626 glsl_interface_packing packing
=
1627 (glsl_interface_packing
) ifc_type
->interface_packing
;
1628 bool row_major
= (bool) ifc_type
->interface_row_major
;
1629 const glsl_type
*new_ifc_type
=
1630 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1631 row_major
, ifc_type
->name
);
1633 for (unsigned i
= 0; i
< num_fields
; i
++) {
1634 if (interface_vars
[i
] != NULL
)
1635 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1640 * Memory context used to allocate the data in \c unnamed_interfaces.
1645 * Hash table from const glsl_type * to an array of ir_variable *'s
1646 * pointing to the ir_variables constituting each unnamed interface block.
1648 hash_table
*unnamed_interfaces
;
1652 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1653 struct gl_shader_program
*prog
)
1655 /* We will validate doubles at a later stage */
1656 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1657 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1658 "multiple of 4 or if its applied to a type that is "
1659 "or contains a double a multiple of 8.",
1660 prog
->TransformFeedback
.BufferStride
[idx
]);
1664 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1665 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1666 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1667 "limit has been exceeded.");
1675 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1679 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1680 struct gl_shader_program
*prog
,
1681 struct gl_shader
**shader_list
,
1682 unsigned num_shaders
)
1684 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1685 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1688 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1689 struct gl_shader
*shader
= shader_list
[i
];
1691 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1692 if (shader
->TransformFeedbackBufferStride
[j
]) {
1693 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1694 prog
->TransformFeedback
.BufferStride
[j
] =
1695 shader
->TransformFeedbackBufferStride
[j
];
1696 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1698 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1699 shader
->TransformFeedbackBufferStride
[j
]){
1701 "intrastage shaders defined with conflicting "
1702 "xfb_stride for buffer %d (%d and %d)\n", j
,
1703 prog
->TransformFeedback
.BufferStride
[j
],
1704 shader
->TransformFeedbackBufferStride
[j
]);
1713 * Check for conflicting bindless/bound sampler/image layout qualifiers at
1717 link_bindless_layout_qualifiers(struct gl_shader_program
*prog
,
1718 struct gl_shader
**shader_list
,
1719 unsigned num_shaders
)
1721 bool bindless_sampler
, bindless_image
;
1722 bool bound_sampler
, bound_image
;
1724 bindless_sampler
= bindless_image
= false;
1725 bound_sampler
= bound_image
= false;
1727 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1728 struct gl_shader
*shader
= shader_list
[i
];
1730 if (shader
->bindless_sampler
)
1731 bindless_sampler
= true;
1732 if (shader
->bindless_image
)
1733 bindless_image
= true;
1734 if (shader
->bound_sampler
)
1735 bound_sampler
= true;
1736 if (shader
->bound_image
)
1739 if ((bindless_sampler
&& bound_sampler
) ||
1740 (bindless_image
&& bound_image
)) {
1741 /* From section 4.4.6 of the ARB_bindless_texture spec:
1743 * "If both bindless_sampler and bound_sampler, or bindless_image
1744 * and bound_image, are declared at global scope in any
1745 * compilation unit, a link- time error will be generated."
1747 linker_error(prog
, "both bindless_sampler and bound_sampler, or "
1748 "bindless_image and bound_image, can't be declared at "
1755 * Performs the cross-validation of tessellation control shader vertices and
1756 * layout qualifiers for the attached tessellation control shaders,
1757 * and propagates them to the linked TCS and linked shader program.
1760 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1761 struct gl_program
*gl_prog
,
1762 struct gl_shader
**shader_list
,
1763 unsigned num_shaders
)
1765 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1768 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1770 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1772 * "All tessellation control shader layout declarations in a program
1773 * must specify the same output patch vertex count. There must be at
1774 * least one layout qualifier specifying an output patch vertex count
1775 * in any program containing tessellation control shaders; however,
1776 * such a declaration is not required in all tessellation control
1780 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1781 struct gl_shader
*shader
= shader_list
[i
];
1783 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1784 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1785 gl_prog
->info
.tess
.tcs_vertices_out
!=
1786 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1787 linker_error(prog
, "tessellation control shader defined with "
1788 "conflicting output vertex count (%d and %d)\n",
1789 gl_prog
->info
.tess
.tcs_vertices_out
,
1790 shader
->info
.TessCtrl
.VerticesOut
);
1793 gl_prog
->info
.tess
.tcs_vertices_out
=
1794 shader
->info
.TessCtrl
.VerticesOut
;
1798 /* Just do the intrastage -> interstage propagation right now,
1799 * since we already know we're in the right type of shader program
1802 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1803 linker_error(prog
, "tessellation control shader didn't declare "
1804 "vertices out layout qualifier\n");
1811 * Performs the cross-validation of tessellation evaluation shader
1812 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1813 * for the attached tessellation evaluation shaders, and propagates them
1814 * to the linked TES and linked shader program.
1817 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1818 struct gl_program
*gl_prog
,
1819 struct gl_shader
**shader_list
,
1820 unsigned num_shaders
)
1822 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1825 int point_mode
= -1;
1826 unsigned vertex_order
= 0;
1828 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1829 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1831 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1833 * "At least one tessellation evaluation shader (compilation unit) in
1834 * a program must declare a primitive mode in its input layout.
1835 * Declaration vertex spacing, ordering, and point mode identifiers is
1836 * optional. It is not required that all tessellation evaluation
1837 * shaders in a program declare a primitive mode. If spacing or
1838 * vertex ordering declarations are omitted, the tessellation
1839 * primitive generator will use equal spacing or counter-clockwise
1840 * vertex ordering, respectively. If a point mode declaration is
1841 * omitted, the tessellation primitive generator will produce lines or
1842 * triangles according to the primitive mode."
1845 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1846 struct gl_shader
*shader
= shader_list
[i
];
1848 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1849 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1850 gl_prog
->info
.tess
.primitive_mode
!=
1851 shader
->info
.TessEval
.PrimitiveMode
) {
1852 linker_error(prog
, "tessellation evaluation shader defined with "
1853 "conflicting input primitive modes.\n");
1856 gl_prog
->info
.tess
.primitive_mode
=
1857 shader
->info
.TessEval
.PrimitiveMode
;
1860 if (shader
->info
.TessEval
.Spacing
!= 0) {
1861 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1862 shader
->info
.TessEval
.Spacing
) {
1863 linker_error(prog
, "tessellation evaluation shader defined with "
1864 "conflicting vertex spacing.\n");
1867 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1870 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1871 if (vertex_order
!= 0 &&
1872 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1873 linker_error(prog
, "tessellation evaluation shader defined with "
1874 "conflicting ordering.\n");
1877 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1880 if (shader
->info
.TessEval
.PointMode
!= -1) {
1881 if (point_mode
!= -1 &&
1882 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1883 linker_error(prog
, "tessellation evaluation shader defined with "
1884 "conflicting point modes.\n");
1887 point_mode
= shader
->info
.TessEval
.PointMode
;
1892 /* Just do the intrastage -> interstage propagation right now,
1893 * since we already know we're in the right type of shader program
1896 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1898 "tessellation evaluation shader didn't declare input "
1899 "primitive modes.\n");
1903 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1904 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1906 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1907 gl_prog
->info
.tess
.ccw
= true;
1909 gl_prog
->info
.tess
.ccw
= false;
1912 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1913 gl_prog
->info
.tess
.point_mode
= false;
1915 gl_prog
->info
.tess
.point_mode
= true;
1920 * Performs the cross-validation of layout qualifiers specified in
1921 * redeclaration of gl_FragCoord for the attached fragment shaders,
1922 * and propagates them to the linked FS and linked shader program.
1925 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1926 struct gl_linked_shader
*linked_shader
,
1927 struct gl_shader
**shader_list
,
1928 unsigned num_shaders
)
1930 bool redeclares_gl_fragcoord
= false;
1931 bool uses_gl_fragcoord
= false;
1932 bool origin_upper_left
= false;
1933 bool pixel_center_integer
= false;
1935 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1936 (prog
->data
->Version
< 150 &&
1937 !prog
->ARB_fragment_coord_conventions_enable
))
1940 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1941 struct gl_shader
*shader
= shader_list
[i
];
1942 /* From the GLSL 1.50 spec, page 39:
1944 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1945 * it must be redeclared in all the fragment shaders in that program
1946 * that have a static use gl_FragCoord."
1948 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1949 shader
->uses_gl_fragcoord
)
1950 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1951 uses_gl_fragcoord
)) {
1952 linker_error(prog
, "fragment shader defined with conflicting "
1953 "layout qualifiers for gl_FragCoord\n");
1956 /* From the GLSL 1.50 spec, page 39:
1958 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1959 * single program must have the same set of qualifiers."
1961 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1962 (shader
->origin_upper_left
!= origin_upper_left
||
1963 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1964 linker_error(prog
, "fragment shader defined with conflicting "
1965 "layout qualifiers for gl_FragCoord\n");
1968 /* Update the linked shader state. Note that uses_gl_fragcoord should
1969 * accumulate the results. The other values should replace. If there
1970 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1971 * are already known to be the same.
1973 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1974 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1975 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1976 origin_upper_left
= shader
->origin_upper_left
;
1977 pixel_center_integer
= shader
->pixel_center_integer
;
1980 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1981 shader
->EarlyFragmentTests
|| shader
->PostDepthCoverage
;
1982 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1983 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1984 shader
->PostDepthCoverage
;
1986 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1991 * Performs the cross-validation of geometry shader max_vertices and
1992 * primitive type layout qualifiers for the attached geometry shaders,
1993 * and propagates them to the linked GS and linked shader program.
1996 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1997 struct gl_program
*gl_prog
,
1998 struct gl_shader
**shader_list
,
1999 unsigned num_shaders
)
2001 /* No in/out qualifiers defined for anything but GLSL 1.50+
2002 * geometry shaders so far.
2004 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
2005 prog
->data
->Version
< 150)
2008 int vertices_out
= -1;
2010 gl_prog
->info
.gs
.invocations
= 0;
2011 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
2012 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
2014 /* From the GLSL 1.50 spec, page 46:
2016 * "All geometry shader output layout declarations in a program
2017 * must declare the same layout and same value for
2018 * max_vertices. There must be at least one geometry output
2019 * layout declaration somewhere in a program, but not all
2020 * geometry shaders (compilation units) are required to
2024 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2025 struct gl_shader
*shader
= shader_list
[i
];
2027 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
2028 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
2029 gl_prog
->info
.gs
.input_primitive
!=
2030 shader
->info
.Geom
.InputType
) {
2031 linker_error(prog
, "geometry shader defined with conflicting "
2035 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
2038 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
2039 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
2040 gl_prog
->info
.gs
.output_primitive
!=
2041 shader
->info
.Geom
.OutputType
) {
2042 linker_error(prog
, "geometry shader defined with conflicting "
2046 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
2049 if (shader
->info
.Geom
.VerticesOut
!= -1) {
2050 if (vertices_out
!= -1 &&
2051 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
2052 linker_error(prog
, "geometry shader defined with conflicting "
2053 "output vertex count (%d and %d)\n",
2054 vertices_out
, shader
->info
.Geom
.VerticesOut
);
2057 vertices_out
= shader
->info
.Geom
.VerticesOut
;
2060 if (shader
->info
.Geom
.Invocations
!= 0) {
2061 if (gl_prog
->info
.gs
.invocations
!= 0 &&
2062 gl_prog
->info
.gs
.invocations
!=
2063 (unsigned) shader
->info
.Geom
.Invocations
) {
2064 linker_error(prog
, "geometry shader defined with conflicting "
2065 "invocation count (%d and %d)\n",
2066 gl_prog
->info
.gs
.invocations
,
2067 shader
->info
.Geom
.Invocations
);
2070 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
2074 /* Just do the intrastage -> interstage propagation right now,
2075 * since we already know we're in the right type of shader program
2078 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
2080 "geometry shader didn't declare primitive input type\n");
2084 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
2086 "geometry shader didn't declare primitive output type\n");
2090 if (vertices_out
== -1) {
2092 "geometry shader didn't declare max_vertices\n");
2095 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2098 if (gl_prog
->info
.gs
.invocations
== 0)
2099 gl_prog
->info
.gs
.invocations
= 1;
2104 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2105 * qualifiers for the attached compute shaders, and propagate them to the
2106 * linked CS and linked shader program.
2109 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2110 struct gl_program
*gl_prog
,
2111 struct gl_shader
**shader_list
,
2112 unsigned num_shaders
)
2114 /* This function is called for all shader stages, but it only has an effect
2115 * for compute shaders.
2117 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2120 for (int i
= 0; i
< 3; i
++)
2121 gl_prog
->info
.cs
.local_size
[i
] = 0;
2123 gl_prog
->info
.cs
.local_size_variable
= false;
2125 /* From the ARB_compute_shader spec, in the section describing local size
2128 * If multiple compute shaders attached to a single program object
2129 * declare local work-group size, the declarations must be identical;
2130 * otherwise a link-time error results. Furthermore, if a program
2131 * object contains any compute shaders, at least one must contain an
2132 * input layout qualifier specifying the local work sizes of the
2133 * program, or a link-time error will occur.
2135 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2136 struct gl_shader
*shader
= shader_list
[sh
];
2138 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2139 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2140 for (int i
= 0; i
< 3; i
++) {
2141 if (gl_prog
->info
.cs
.local_size
[i
] !=
2142 shader
->info
.Comp
.LocalSize
[i
]) {
2143 linker_error(prog
, "compute shader defined with conflicting "
2149 for (int i
= 0; i
< 3; i
++) {
2150 gl_prog
->info
.cs
.local_size
[i
] =
2151 shader
->info
.Comp
.LocalSize
[i
];
2153 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2154 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2155 /* The ARB_compute_variable_group_size spec says:
2157 * If one compute shader attached to a program declares a
2158 * variable local group size and a second compute shader
2159 * attached to the same program declares a fixed local group
2160 * size, a link-time error results.
2162 linker_error(prog
, "compute shader defined with both fixed and "
2163 "variable local group size\n");
2166 gl_prog
->info
.cs
.local_size_variable
= true;
2170 /* Just do the intrastage -> interstage propagation right now,
2171 * since we already know we're in the right type of shader program
2174 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2175 !gl_prog
->info
.cs
.local_size_variable
) {
2176 linker_error(prog
, "compute shader must contain a fixed or a variable "
2177 "local group size\n");
2184 * Combine a group of shaders for a single stage to generate a linked shader
2187 * If this function is supplied a single shader, it is cloned, and the new
2188 * shader is returned.
2190 struct gl_linked_shader
*
2191 link_intrastage_shaders(void *mem_ctx
,
2192 struct gl_context
*ctx
,
2193 struct gl_shader_program
*prog
,
2194 struct gl_shader
**shader_list
,
2195 unsigned num_shaders
,
2196 bool allow_missing_main
)
2198 struct gl_uniform_block
*ubo_blocks
= NULL
;
2199 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2200 unsigned num_ubo_blocks
= 0;
2201 unsigned num_ssbo_blocks
= 0;
2203 /* Check that global variables defined in multiple shaders are consistent.
2205 glsl_symbol_table variables
;
2206 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2207 if (shader_list
[i
] == NULL
)
2209 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2212 if (!prog
->data
->LinkStatus
)
2215 /* Check that interface blocks defined in multiple shaders are consistent.
2217 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2219 if (!prog
->data
->LinkStatus
)
2222 /* Check that there is only a single definition of each function signature
2223 * across all shaders.
2225 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2226 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2227 ir_function
*const f
= node
->as_function();
2232 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2233 ir_function
*const other
=
2234 shader_list
[j
]->symbols
->get_function(f
->name
);
2236 /* If the other shader has no function (and therefore no function
2237 * signatures) with the same name, skip to the next shader.
2242 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2243 if (!sig
->is_defined
)
2246 ir_function_signature
*other_sig
=
2247 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2249 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2250 linker_error(prog
, "function `%s' is multiply defined\n",
2259 /* Find the shader that defines main, and make a clone of it.
2261 * Starting with the clone, search for undefined references. If one is
2262 * found, find the shader that defines it. Clone the reference and add
2263 * it to the shader. Repeat until there are no undefined references or
2264 * until a reference cannot be resolved.
2266 gl_shader
*main
= NULL
;
2267 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2268 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2269 main
= shader_list
[i
];
2274 if (main
== NULL
&& allow_missing_main
)
2275 main
= shader_list
[0];
2278 linker_error(prog
, "%s shader lacks `main'\n",
2279 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2283 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2284 linked
->Stage
= shader_list
[0]->Stage
;
2286 /* Create program and attach it to the linked shader */
2287 struct gl_program
*gl_prog
=
2288 ctx
->Driver
.NewProgram(ctx
,
2289 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2292 prog
->data
->LinkStatus
= LINKING_FAILURE
;
2293 _mesa_delete_linked_shader(ctx
, linked
);
2297 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2299 /* Don't use _mesa_reference_program() just take ownership */
2300 linked
->Program
= gl_prog
;
2302 linked
->ir
= new(linked
) exec_list
;
2303 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2305 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2306 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2307 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2308 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2309 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2311 if (linked
->Stage
!= MESA_SHADER_FRAGMENT
)
2312 link_xfb_stride_layout_qualifiers(ctx
, prog
, shader_list
, num_shaders
);
2314 link_bindless_layout_qualifiers(prog
, shader_list
, num_shaders
);
2316 populate_symbol_table(linked
, shader_list
[0]->symbols
);
2318 /* The pointer to the main function in the final linked shader (i.e., the
2319 * copy of the original shader that contained the main function).
2321 ir_function_signature
*const main_sig
=
2322 _mesa_get_main_function_signature(linked
->symbols
);
2324 /* Move any instructions other than variable declarations or function
2325 * declarations into main.
2327 if (main_sig
!= NULL
) {
2328 exec_node
*insertion_point
=
2329 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2332 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2333 if (shader_list
[i
] == main
)
2336 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2337 insertion_point
, true, linked
);
2341 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2342 _mesa_delete_linked_shader(ctx
, linked
);
2346 /* Make a pass over all variable declarations to ensure that arrays with
2347 * unspecified sizes have a size specified. The size is inferred from the
2348 * max_array_access field.
2350 array_sizing_visitor v
;
2352 v
.fixup_unnamed_interface_types();
2354 /* Link up uniform blocks defined within this stage. */
2355 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2356 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2358 if (!prog
->data
->LinkStatus
) {
2359 _mesa_delete_linked_shader(ctx
, linked
);
2363 /* Copy ubo blocks to linked shader list */
2364 linked
->Program
->sh
.UniformBlocks
=
2365 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2366 ralloc_steal(linked
, ubo_blocks
);
2367 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2368 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2370 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2372 /* Copy ssbo blocks to linked shader list */
2373 linked
->Program
->sh
.ShaderStorageBlocks
=
2374 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2375 ralloc_steal(linked
, ssbo_blocks
);
2376 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2377 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2379 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2381 /* At this point linked should contain all of the linked IR, so
2382 * validate it to make sure nothing went wrong.
2384 validate_ir_tree(linked
->ir
);
2386 /* Set the size of geometry shader input arrays */
2387 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2388 unsigned num_vertices
=
2389 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2390 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2391 MESA_SHADER_GEOMETRY
);
2392 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2393 ir
->accept(&input_resize_visitor
);
2397 if (ctx
->Const
.VertexID_is_zero_based
)
2398 lower_vertex_id(linked
);
2400 if (ctx
->Const
.LowerCsDerivedVariables
)
2401 lower_cs_derived(linked
);
2404 /* Compute the source checksum. */
2405 linked
->SourceChecksum
= 0;
2406 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2407 if (shader_list
[i
] == NULL
)
2409 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2417 * Update the sizes of linked shader uniform arrays to the maximum
2420 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2422 * If one or more elements of an array are active,
2423 * GetActiveUniform will return the name of the array in name,
2424 * subject to the restrictions listed above. The type of the array
2425 * is returned in type. The size parameter contains the highest
2426 * array element index used, plus one. The compiler or linker
2427 * determines the highest index used. There will be only one
2428 * active uniform reported by the GL per uniform array.
2432 update_array_sizes(struct gl_shader_program
*prog
)
2434 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2435 if (prog
->_LinkedShaders
[i
] == NULL
)
2438 bool types_were_updated
= false;
2440 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2441 ir_variable
*const var
= node
->as_variable();
2443 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2444 !var
->type
->is_array())
2447 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2448 * will not be eliminated. Since we always do std140, just
2449 * don't resize arrays in UBOs.
2451 * Atomic counters are supposed to get deterministic
2452 * locations assigned based on the declaration ordering and
2453 * sizes, array compaction would mess that up.
2455 * Subroutine uniforms are not removed.
2457 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2458 var
->type
->contains_subroutine() || var
->constant_initializer
)
2461 int size
= var
->data
.max_array_access
;
2462 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2463 if (prog
->_LinkedShaders
[j
] == NULL
)
2466 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2467 ir_variable
*other_var
= node2
->as_variable();
2471 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2472 other_var
->data
.max_array_access
> size
) {
2473 size
= other_var
->data
.max_array_access
;
2478 if (size
+ 1 != (int)var
->type
->length
) {
2479 /* If this is a built-in uniform (i.e., it's backed by some
2480 * fixed-function state), adjust the number of state slots to
2481 * match the new array size. The number of slots per array entry
2482 * is not known. It seems safe to assume that the total number of
2483 * slots is an integer multiple of the number of array elements.
2484 * Determine the number of slots per array element by dividing by
2485 * the old (total) size.
2487 const unsigned num_slots
= var
->get_num_state_slots();
2488 if (num_slots
> 0) {
2489 var
->set_num_state_slots((size
+ 1)
2490 * (num_slots
/ var
->type
->length
));
2493 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2495 types_were_updated
= true;
2499 /* Update the types of dereferences in case we changed any. */
2500 if (types_were_updated
) {
2501 deref_type_updater v
;
2502 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2508 * Resize tessellation evaluation per-vertex inputs to the size of
2509 * tessellation control per-vertex outputs.
2512 resize_tes_inputs(struct gl_context
*ctx
,
2513 struct gl_shader_program
*prog
)
2515 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2518 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2519 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2521 /* If no control shader is present, then the TES inputs are statically
2522 * sized to MaxPatchVertices; the actual size of the arrays won't be
2523 * known until draw time.
2525 const int num_vertices
= tcs
2526 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2527 : ctx
->Const
.MaxPatchVertices
;
2529 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2530 MESA_SHADER_TESS_EVAL
);
2531 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2532 ir
->accept(&input_resize_visitor
);
2536 /* Convert the gl_PatchVerticesIn system value into a constant, since
2537 * the value is known at this point.
2539 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2540 ir_variable
*var
= ir
->as_variable();
2541 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2542 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2543 void *mem_ctx
= ralloc_parent(var
);
2544 var
->data
.location
= 0;
2545 var
->data
.explicit_location
= false;
2546 var
->data
.mode
= ir_var_auto
;
2547 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2554 * Find a contiguous set of available bits in a bitmask.
2556 * \param used_mask Bits representing used (1) and unused (0) locations
2557 * \param needed_count Number of contiguous bits needed.
2560 * Base location of the available bits on success or -1 on failure.
2563 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2565 unsigned needed_mask
= (1 << needed_count
) - 1;
2566 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2568 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2569 * cannot optimize possibly infinite loops" for the loop below.
2571 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2574 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2575 if ((needed_mask
& ~used_mask
) == needed_mask
)
2585 #define SAFE_MASK_FROM_INDEX(i) (((i) >= 32) ? ~0 : ((1 << (i)) - 1))
2588 * Assign locations for either VS inputs or FS outputs
2590 * \param mem_ctx Temporary ralloc context used for linking
2591 * \param prog Shader program whose variables need locations assigned
2592 * \param constants Driver specific constant values for the program.
2593 * \param target_index Selector for the program target to receive location
2594 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2595 * \c MESA_SHADER_FRAGMENT.
2598 * If locations are successfully assigned, true is returned. Otherwise an
2599 * error is emitted to the shader link log and false is returned.
2602 assign_attribute_or_color_locations(void *mem_ctx
,
2603 gl_shader_program
*prog
,
2604 struct gl_constants
*constants
,
2605 unsigned target_index
)
2607 /* Maximum number of generic locations. This corresponds to either the
2608 * maximum number of draw buffers or the maximum number of generic
2611 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2612 constants
->Program
[target_index
].MaxAttribs
:
2613 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2615 /* Mark invalid locations as being used.
2617 unsigned used_locations
= ~SAFE_MASK_FROM_INDEX(max_index
);
2618 unsigned double_storage_locations
= 0;
2620 assert((target_index
== MESA_SHADER_VERTEX
)
2621 || (target_index
== MESA_SHADER_FRAGMENT
));
2623 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2627 /* Operate in a total of four passes.
2629 * 1. Invalidate the location assignments for all vertex shader inputs.
2631 * 2. Assign locations for inputs that have user-defined (via
2632 * glBindVertexAttribLocation) locations and outputs that have
2633 * user-defined locations (via glBindFragDataLocation).
2635 * 3. Sort the attributes without assigned locations by number of slots
2636 * required in decreasing order. Fragmentation caused by attribute
2637 * locations assigned by the application may prevent large attributes
2638 * from having enough contiguous space.
2640 * 4. Assign locations to any inputs without assigned locations.
2643 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2644 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2646 const enum ir_variable_mode direction
=
2647 (target_index
== MESA_SHADER_VERTEX
)
2648 ? ir_var_shader_in
: ir_var_shader_out
;
2651 /* Temporary storage for the set of attributes that need locations assigned.
2657 /* Used below in the call to qsort. */
2658 static int compare(const void *a
, const void *b
)
2660 const temp_attr
*const l
= (const temp_attr
*) a
;
2661 const temp_attr
*const r
= (const temp_attr
*) b
;
2663 /* Reversed because we want a descending order sort below. */
2664 return r
->slots
- l
->slots
;
2667 assert(max_index
<= 32);
2669 /* Temporary array for the set of attributes that have locations assigned,
2670 * for the purpose of checking overlapping slots/components of (non-ES)
2671 * fragment shader outputs.
2673 ir_variable
*assigned
[12 * 4]; /* (max # of FS outputs) * # components */
2674 unsigned assigned_attr
= 0;
2676 unsigned num_attr
= 0;
2678 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2679 ir_variable
*const var
= node
->as_variable();
2681 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2684 if (var
->data
.explicit_location
) {
2685 var
->data
.is_unmatched_generic_inout
= 0;
2686 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2687 || (var
->data
.location
< 0)) {
2689 "invalid explicit location %d specified for `%s'\n",
2690 (var
->data
.location
< 0)
2691 ? var
->data
.location
2692 : var
->data
.location
- generic_base
,
2696 } else if (target_index
== MESA_SHADER_VERTEX
) {
2699 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2700 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2701 var
->data
.location
= binding
;
2702 var
->data
.is_unmatched_generic_inout
= 0;
2704 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2707 const char *name
= var
->name
;
2708 const glsl_type
*type
= var
->type
;
2711 /* Check if there's a binding for the variable name */
2712 if (prog
->FragDataBindings
->get(binding
, name
)) {
2713 assert(binding
>= FRAG_RESULT_DATA0
);
2714 var
->data
.location
= binding
;
2715 var
->data
.is_unmatched_generic_inout
= 0;
2717 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2718 var
->data
.index
= index
;
2723 /* If not, but it's an array type, look for name[0] */
2724 if (type
->is_array()) {
2725 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2726 type
= type
->fields
.array
;
2734 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2737 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2739 * "Output binding assignments will cause LinkProgram to fail:
2741 * If the program has an active output assigned to a location greater
2742 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2743 * an active output assigned an index greater than or equal to one;"
2745 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2746 var
->data
.location
- generic_base
>=
2747 (int) constants
->MaxDualSourceDrawBuffers
) {
2749 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2750 "with index %u for %s\n",
2751 var
->data
.location
- generic_base
, var
->data
.index
,
2756 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2758 /* If the variable is not a built-in and has a location statically
2759 * assigned in the shader (presumably via a layout qualifier), make sure
2760 * that it doesn't collide with other assigned locations. Otherwise,
2761 * add it to the list of variables that need linker-assigned locations.
2763 if (var
->data
.location
!= -1) {
2764 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2765 /* From page 61 of the OpenGL 4.0 spec:
2767 * "LinkProgram will fail if the attribute bindings assigned
2768 * by BindAttribLocation do not leave not enough space to
2769 * assign a location for an active matrix attribute or an
2770 * active attribute array, both of which require multiple
2771 * contiguous generic attributes."
2773 * I think above text prohibits the aliasing of explicit and
2774 * automatic assignments. But, aliasing is allowed in manual
2775 * assignments of attribute locations. See below comments for
2778 * From OpenGL 4.0 spec, page 61:
2780 * "It is possible for an application to bind more than one
2781 * attribute name to the same location. This is referred to as
2782 * aliasing. This will only work if only one of the aliased
2783 * attributes is active in the executable program, or if no
2784 * path through the shader consumes more than one attribute of
2785 * a set of attributes aliased to the same location. A link
2786 * error can occur if the linker determines that every path
2787 * through the shader consumes multiple aliased attributes,
2788 * but implementations are not required to generate an error
2791 * From GLSL 4.30 spec, page 54:
2793 * "A program will fail to link if any two non-vertex shader
2794 * input variables are assigned to the same location. For
2795 * vertex shaders, multiple input variables may be assigned
2796 * to the same location using either layout qualifiers or via
2797 * the OpenGL API. However, such aliasing is intended only to
2798 * support vertex shaders where each execution path accesses
2799 * at most one input per each location. Implementations are
2800 * permitted, but not required, to generate link-time errors
2801 * if they detect that every path through the vertex shader
2802 * executable accesses multiple inputs assigned to any single
2803 * location. For all shader types, a program will fail to link
2804 * if explicit location assignments leave the linker unable
2805 * to find space for other variables without explicit
2808 * From OpenGL ES 3.0 spec, page 56:
2810 * "Binding more than one attribute name to the same location
2811 * is referred to as aliasing, and is not permitted in OpenGL
2812 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2813 * fail when this condition exists. However, aliasing is
2814 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2815 * This will only work if only one of the aliased attributes
2816 * is active in the executable program, or if no path through
2817 * the shader consumes more than one attribute of a set of
2818 * attributes aliased to the same location. A link error can
2819 * occur if the linker determines that every path through the
2820 * shader consumes multiple aliased attributes, but implemen-
2821 * tations are not required to generate an error in this case."
2823 * After looking at above references from OpenGL, OpenGL ES and
2824 * GLSL specifications, we allow aliasing of vertex input variables
2825 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2827 * NOTE: This is not required by the spec but its worth mentioning
2828 * here that we're not doing anything to make sure that no path
2829 * through the vertex shader executable accesses multiple inputs
2830 * assigned to any single location.
2833 /* Mask representing the contiguous slots that will be used by
2836 const unsigned attr
= var
->data
.location
- generic_base
;
2837 const unsigned use_mask
= (1 << slots
) - 1;
2838 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2839 ? "vertex shader input" : "fragment shader output";
2841 /* Generate a link error if the requested locations for this
2842 * attribute exceed the maximum allowed attribute location.
2844 if (attr
+ slots
> max_index
) {
2846 "insufficient contiguous locations "
2847 "available for %s `%s' %d %d %d\n", string
,
2848 var
->name
, used_locations
, use_mask
, attr
);
2852 /* Generate a link error if the set of bits requested for this
2853 * attribute overlaps any previously allocated bits.
2855 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2856 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2857 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2860 * "Additionally, for fragment shader outputs, if two
2861 * variables are placed within the same location, they
2862 * must have the same underlying type (floating-point or
2863 * integer). No component aliasing of output variables or
2864 * members is allowed.
2866 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2867 unsigned assigned_slots
=
2868 assigned
[i
]->type
->count_attribute_slots(false);
2869 unsigned assig_attr
=
2870 assigned
[i
]->data
.location
- generic_base
;
2871 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2873 if ((assigned_use_mask
<< assig_attr
) &
2874 (use_mask
<< attr
)) {
2876 const glsl_type
*assigned_type
=
2877 assigned
[i
]->type
->without_array();
2878 const glsl_type
*type
= var
->type
->without_array();
2879 if (assigned_type
->base_type
!= type
->base_type
) {
2880 linker_error(prog
, "types do not match for aliased"
2881 " %ss %s and %s\n", string
,
2882 assigned
[i
]->name
, var
->name
);
2886 unsigned assigned_component_mask
=
2887 ((1 << assigned_type
->vector_elements
) - 1) <<
2888 assigned
[i
]->data
.location_frac
;
2889 unsigned component_mask
=
2890 ((1 << type
->vector_elements
) - 1) <<
2891 var
->data
.location_frac
;
2892 if (assigned_component_mask
& component_mask
) {
2893 linker_error(prog
, "overlapping component is "
2894 "assigned to %ss %s and %s "
2896 string
, assigned
[i
]->name
, var
->name
,
2897 var
->data
.location_frac
);
2902 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2903 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2904 linker_error(prog
, "overlapping location is assigned "
2905 "to %s `%s' %d %d %d\n", string
, var
->name
,
2906 used_locations
, use_mask
, attr
);
2909 linker_warning(prog
, "overlapping location is assigned "
2910 "to %s `%s' %d %d %d\n", string
, var
->name
,
2911 used_locations
, use_mask
, attr
);
2915 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2916 /* Only track assigned variables for non-ES fragment shaders
2917 * to avoid overflowing the array.
2919 * At most one variable per fragment output component should
2922 assert(assigned_attr
< ARRAY_SIZE(assigned
));
2923 assigned
[assigned_attr
] = var
;
2927 used_locations
|= (use_mask
<< attr
);
2929 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2931 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2932 * active attribute variables may fail to link, unless
2933 * device-dependent optimizations are able to make the program
2934 * fit within available hardware resources. For the purposes
2935 * of this test, attribute variables of the type dvec3, dvec4,
2936 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2937 * count as consuming twice as many attributes as equivalent
2938 * single-precision types. While these types use the same number
2939 * of generic attributes as their single-precision equivalents,
2940 * implementations are permitted to consume two single-precision
2941 * vectors of internal storage for each three- or four-component
2942 * double-precision vector."
2944 * Mark this attribute slot as taking up twice as much space
2945 * so we can count it properly against limits. According to
2946 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2947 * is optional behavior, but it seems preferable.
2949 if (var
->type
->without_array()->is_dual_slot())
2950 double_storage_locations
|= (use_mask
<< attr
);
2956 if (num_attr
>= max_index
) {
2957 linker_error(prog
, "too many %s (max %u)",
2958 target_index
== MESA_SHADER_VERTEX
?
2959 "vertex shader inputs" : "fragment shader outputs",
2963 to_assign
[num_attr
].slots
= slots
;
2964 to_assign
[num_attr
].var
= var
;
2968 if (target_index
== MESA_SHADER_VERTEX
) {
2969 unsigned total_attribs_size
=
2970 _mesa_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
2971 _mesa_bitcount(double_storage_locations
);
2972 if (total_attribs_size
> max_index
) {
2974 "attempt to use %d vertex attribute slots only %d available ",
2975 total_attribs_size
, max_index
);
2980 /* If all of the attributes were assigned locations by the application (or
2981 * are built-in attributes with fixed locations), return early. This should
2982 * be the common case.
2987 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2989 if (target_index
== MESA_SHADER_VERTEX
) {
2990 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2991 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2992 * reserved to prevent it from being automatically allocated below.
2994 find_deref_visitor
find("gl_Vertex");
2996 if (find
.variable_found())
2997 used_locations
|= (1 << 0);
3000 for (unsigned i
= 0; i
< num_attr
; i
++) {
3001 /* Mask representing the contiguous slots that will be used by this
3004 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
3006 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
3009 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
3010 ? "vertex shader input" : "fragment shader output";
3013 "insufficient contiguous locations "
3014 "available for %s `%s'\n",
3015 string
, to_assign
[i
].var
->name
);
3019 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
3020 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
3021 used_locations
|= (use_mask
<< location
);
3023 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
3024 double_storage_locations
|= (use_mask
<< location
);
3027 /* Now that we have all the locations, from the GL 4.5 core spec, section
3028 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
3029 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
3030 * as equivalent single-precision types.
3032 if (target_index
== MESA_SHADER_VERTEX
) {
3033 unsigned total_attribs_size
=
3034 _mesa_bitcount(used_locations
& SAFE_MASK_FROM_INDEX(max_index
)) +
3035 _mesa_bitcount(double_storage_locations
);
3036 if (total_attribs_size
> max_index
) {
3038 "attempt to use %d vertex attribute slots only %d available ",
3039 total_attribs_size
, max_index
);
3048 * Match explicit locations of outputs to inputs and deactivate the
3049 * unmatch flag if found so we don't optimise them away.
3052 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
3053 gl_linked_shader
*consumer
)
3055 glsl_symbol_table parameters
;
3056 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
3059 /* Find all shader outputs in the "producer" stage.
3061 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
3062 ir_variable
*const var
= node
->as_variable();
3064 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
3067 if (var
->data
.explicit_location
&&
3068 var
->data
.location
>= VARYING_SLOT_VAR0
) {
3069 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
3070 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
3071 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
3075 /* Match inputs to outputs */
3076 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
3077 ir_variable
*const input
= node
->as_variable();
3079 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
3082 ir_variable
*output
= NULL
;
3083 if (input
->data
.explicit_location
3084 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
3085 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
3086 [input
->data
.location_frac
];
3088 if (output
!= NULL
){
3089 input
->data
.is_unmatched_generic_inout
= 0;
3090 output
->data
.is_unmatched_generic_inout
= 0;
3097 * Store the gl_FragDepth layout in the gl_shader_program struct.
3100 store_fragdepth_layout(struct gl_shader_program
*prog
)
3102 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3106 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3108 /* We don't look up the gl_FragDepth symbol directly because if
3109 * gl_FragDepth is not used in the shader, it's removed from the IR.
3110 * However, the symbol won't be removed from the symbol table.
3112 * We're only interested in the cases where the variable is NOT removed
3115 foreach_in_list(ir_instruction
, node
, ir
) {
3116 ir_variable
*const var
= node
->as_variable();
3118 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3122 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3123 switch (var
->data
.depth_layout
) {
3124 case ir_depth_layout_none
:
3125 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3127 case ir_depth_layout_any
:
3128 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3130 case ir_depth_layout_greater
:
3131 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3133 case ir_depth_layout_less
:
3134 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3136 case ir_depth_layout_unchanged
:
3137 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3148 * Validate the resources used by a program versus the implementation limits
3151 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3153 unsigned total_uniform_blocks
= 0;
3154 unsigned total_shader_storage_blocks
= 0;
3156 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3157 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3162 if (sh
->Program
->info
.num_textures
>
3163 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3164 linker_error(prog
, "Too many %s shader texture samplers\n",
3165 _mesa_shader_stage_to_string(i
));
3168 if (sh
->num_uniform_components
>
3169 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3170 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3171 linker_warning(prog
, "Too many %s shader default uniform block "
3172 "components, but the driver will try to optimize "
3173 "them out; this is non-portable out-of-spec "
3175 _mesa_shader_stage_to_string(i
));
3177 linker_error(prog
, "Too many %s shader default uniform block "
3179 _mesa_shader_stage_to_string(i
));
3183 if (sh
->num_combined_uniform_components
>
3184 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3185 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3186 linker_warning(prog
, "Too many %s shader uniform components, "
3187 "but the driver will try to optimize them out; "
3188 "this is non-portable out-of-spec behavior\n",
3189 _mesa_shader_stage_to_string(i
));
3191 linker_error(prog
, "Too many %s shader uniform components\n",
3192 _mesa_shader_stage_to_string(i
));
3196 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3197 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3199 const unsigned max_uniform_blocks
=
3200 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3201 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3202 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3203 _mesa_shader_stage_to_string(i
),
3204 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3207 const unsigned max_shader_storage_blocks
=
3208 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3209 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3210 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3211 _mesa_shader_stage_to_string(i
),
3212 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3216 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3217 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3218 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3221 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3222 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3223 total_shader_storage_blocks
,
3224 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3227 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3228 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3229 ctx
->Const
.MaxUniformBlockSize
) {
3230 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3231 prog
->data
->UniformBlocks
[i
].Name
,
3232 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3233 ctx
->Const
.MaxUniformBlockSize
);
3237 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3238 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3239 ctx
->Const
.MaxShaderStorageBlockSize
) {
3240 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3241 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3242 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3243 ctx
->Const
.MaxShaderStorageBlockSize
);
3249 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3251 unsigned mask
= prog
->data
->linked_stages
;
3253 const int i
= u_bit_scan(&mask
);
3254 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3256 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3257 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3260 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3266 if (p
->sh
.NumSubroutineFunctions
== 0) {
3267 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3270 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3271 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3272 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3273 if (fn
->types
[k
] == uni
->type
) {
3279 uni
->num_compatible_subroutines
= count
;
3285 check_subroutine_resources(struct gl_shader_program
*prog
)
3287 unsigned mask
= prog
->data
->linked_stages
;
3289 const int i
= u_bit_scan(&mask
);
3290 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3292 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3293 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3294 _mesa_shader_stage_to_string(i
));
3299 * Validate shader image resources.
3302 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3304 unsigned total_image_units
= 0;
3305 unsigned fragment_outputs
= 0;
3306 unsigned total_shader_storage_blocks
= 0;
3308 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3311 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3312 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3315 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3316 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3317 _mesa_shader_stage_to_string(i
),
3318 sh
->Program
->info
.num_images
,
3319 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3321 total_image_units
+= sh
->Program
->info
.num_images
;
3322 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3324 if (i
== MESA_SHADER_FRAGMENT
) {
3325 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3326 ir_variable
*var
= node
->as_variable();
3327 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3328 /* since there are no double fs outputs - pass false */
3329 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3335 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3336 linker_error(prog
, "Too many combined image uniforms\n");
3338 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3339 ctx
->Const
.MaxCombinedShaderOutputResources
)
3340 linker_error(prog
, "Too many combined image uniforms, shader storage "
3341 " buffers and fragment outputs\n");
3346 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3347 * for a variable, checks for overlaps between other uniforms using explicit
3351 reserve_explicit_locations(struct gl_shader_program
*prog
,
3352 string_to_uint_map
*map
, ir_variable
*var
)
3354 unsigned slots
= var
->type
->uniform_locations();
3355 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3356 unsigned return_value
= slots
;
3358 /* Resize remap table if locations do not fit in the current one. */
3359 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3360 prog
->UniformRemapTable
=
3361 reralloc(prog
, prog
->UniformRemapTable
,
3362 gl_uniform_storage
*,
3365 if (!prog
->UniformRemapTable
) {
3366 linker_error(prog
, "Out of memory during linking.\n");
3370 /* Initialize allocated space. */
3371 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3372 prog
->UniformRemapTable
[i
] = NULL
;
3374 prog
->NumUniformRemapTable
= max_loc
+ 1;
3377 for (unsigned i
= 0; i
< slots
; i
++) {
3378 unsigned loc
= var
->data
.location
+ i
;
3380 /* Check if location is already used. */
3381 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3383 /* Possibly same uniform from a different stage, this is ok. */
3385 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3390 /* ARB_explicit_uniform_location specification states:
3392 * "No two default-block uniform variables in the program can have
3393 * the same location, even if they are unused, otherwise a compiler
3394 * or linker error will be generated."
3397 "location qualifier for uniform %s overlaps "
3398 "previously used location\n",
3403 /* Initialize location as inactive before optimization
3404 * rounds and location assignment.
3406 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3409 /* Note, base location used for arrays. */
3410 map
->put(var
->data
.location
, var
->name
);
3412 return return_value
;
3416 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3417 struct gl_program
*p
,
3420 unsigned slots
= var
->type
->uniform_locations();
3421 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3423 /* Resize remap table if locations do not fit in the current one. */
3424 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3425 p
->sh
.SubroutineUniformRemapTable
=
3426 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3427 gl_uniform_storage
*,
3430 if (!p
->sh
.SubroutineUniformRemapTable
) {
3431 linker_error(prog
, "Out of memory during linking.\n");
3435 /* Initialize allocated space. */
3436 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3437 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3439 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3442 for (unsigned i
= 0; i
< slots
; i
++) {
3443 unsigned loc
= var
->data
.location
+ i
;
3445 /* Check if location is already used. */
3446 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3448 /* ARB_explicit_uniform_location specification states:
3449 * "No two subroutine uniform variables can have the same location
3450 * in the same shader stage, otherwise a compiler or linker error
3451 * will be generated."
3454 "location qualifier for uniform %s overlaps "
3455 "previously used location\n",
3460 /* Initialize location as inactive before optimization
3461 * rounds and location assignment.
3463 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3469 * Check and reserve all explicit uniform locations, called before
3470 * any optimizations happen to handle also inactive uniforms and
3471 * inactive array elements that may get trimmed away.
3474 check_explicit_uniform_locations(struct gl_context
*ctx
,
3475 struct gl_shader_program
*prog
)
3477 prog
->NumExplicitUniformLocations
= 0;
3479 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3482 /* This map is used to detect if overlapping explicit locations
3483 * occur with the same uniform (from different stage) or a different one.
3485 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3488 linker_error(prog
, "Out of memory during linking.\n");
3492 unsigned entries_total
= 0;
3493 unsigned mask
= prog
->data
->linked_stages
;
3495 const int i
= u_bit_scan(&mask
);
3496 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3498 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3499 ir_variable
*var
= node
->as_variable();
3500 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3503 if (var
->data
.explicit_location
) {
3505 if (var
->type
->without_array()->is_subroutine())
3506 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3508 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3512 entries_total
+= slots
;
3523 struct empty_uniform_block
*current_block
= NULL
;
3525 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3526 /* We found empty space in UniformRemapTable. */
3527 if (prog
->UniformRemapTable
[i
] == NULL
) {
3528 /* We've found the beginning of a new continous block of empty slots */
3529 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3530 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3531 current_block
->start
= i
;
3532 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3533 ¤t_block
->link
);
3536 /* The current block continues, so we simply increment its slots */
3537 current_block
->slots
++;
3542 prog
->NumExplicitUniformLocations
= entries_total
;
3546 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3547 GLenum type
, const char *name
)
3549 bool found_interface
= false;
3550 unsigned block_name_len
= 0;
3551 const char *block_name_dot
= strchr(name
, '.');
3553 /* These rules only apply to buffer variables. So we return
3554 * true for the rest of types.
3556 if (type
!= GL_BUFFER_VARIABLE
)
3559 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3560 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3561 block_name_len
= strlen(block_name
);
3563 const char *block_square_bracket
= strchr(block_name
, '[');
3564 if (block_square_bracket
) {
3565 /* The block is part of an array of named interfaces,
3566 * for the name comparison we ignore the "[x]" part.
3568 block_name_len
-= strlen(block_square_bracket
);
3571 if (block_name_dot
) {
3572 /* Check if the variable name starts with the interface
3573 * name. The interface name (if present) should have the
3574 * length than the interface block name we are comparing to.
3576 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3577 if (len
!= block_name_len
)
3581 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3582 found_interface
= true;
3587 /* We remove the interface name from the buffer variable name,
3588 * including the dot that follows it.
3590 if (found_interface
)
3591 name
= name
+ block_name_len
+ 1;
3593 /* The ARB_program_interface_query spec says:
3595 * "For an active shader storage block member declared as an array, an
3596 * entry will be generated only for the first array element, regardless
3597 * of its type. For arrays of aggregate types, the enumeration rules
3598 * are applied recursively for the single enumerated array element."
3600 const char *struct_first_dot
= strchr(name
, '.');
3601 const char *first_square_bracket
= strchr(name
, '[');
3603 /* The buffer variable is on top level and it is not an array */
3604 if (!first_square_bracket
) {
3606 /* The shader storage block member is a struct, then generate the entry */
3607 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3610 /* Shader storage block member is an array, only generate an entry for the
3611 * first array element.
3613 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3621 add_program_resource(struct gl_shader_program
*prog
,
3622 struct set
*resource_set
,
3623 GLenum type
, const void *data
, uint8_t stages
)
3627 /* If resource already exists, do not add it again. */
3628 if (_mesa_set_search(resource_set
, data
))
3631 prog
->data
->ProgramResourceList
=
3632 reralloc(prog
->data
,
3633 prog
->data
->ProgramResourceList
,
3634 gl_program_resource
,
3635 prog
->data
->NumProgramResourceList
+ 1);
3637 if (!prog
->data
->ProgramResourceList
) {
3638 linker_error(prog
, "Out of memory during linking.\n");
3642 struct gl_program_resource
*res
=
3643 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3647 res
->StageReferences
= stages
;
3649 prog
->data
->NumProgramResourceList
++;
3651 _mesa_set_add(resource_set
, data
);
3656 /* Function checks if a variable var is a packed varying and
3657 * if given name is part of packed varying's list.
3659 * If a variable is a packed varying, it has a name like
3660 * 'packed:a,b,c' where a, b and c are separate variables.
3663 included_in_packed_varying(ir_variable
*var
, const char *name
)
3665 if (strncmp(var
->name
, "packed:", 7) != 0)
3668 char *list
= strdup(var
->name
+ 7);
3673 char *token
= strtok_r(list
, ",", &saveptr
);
3675 if (strcmp(token
, name
) == 0) {
3679 token
= strtok_r(NULL
, ",", &saveptr
);
3686 * Function builds a stage reference bitmask from variable name.
3689 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3694 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3695 * used for reference mask in gl_program_resource will need to be changed.
3697 assert(MESA_SHADER_STAGES
< 8);
3699 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3700 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3704 /* Shader symbol table may contain variables that have
3705 * been optimized away. Search IR for the variable instead.
3707 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3708 ir_variable
*var
= node
->as_variable();
3710 unsigned baselen
= strlen(var
->name
);
3712 if (included_in_packed_varying(var
, name
)) {
3717 /* Type needs to match if specified, otherwise we might
3718 * pick a variable with same name but different interface.
3720 if (var
->data
.mode
!= mode
)
3723 if (strncmp(var
->name
, name
, baselen
) == 0) {
3724 /* Check for exact name matches but also check for arrays and
3727 if (name
[baselen
] == '\0' ||
3728 name
[baselen
] == '[' ||
3729 name
[baselen
] == '.') {
3741 * Create gl_shader_variable from ir_variable class.
3743 static gl_shader_variable
*
3744 create_shader_variable(struct gl_shader_program
*shProg
,
3745 const ir_variable
*in
,
3746 const char *name
, const glsl_type
*type
,
3747 const glsl_type
*interface_type
,
3748 bool use_implicit_location
, int location
,
3749 const glsl_type
*outermost_struct_type
)
3751 /* Allocate zero-initialized memory to ensure that bitfield padding
3754 gl_shader_variable
*out
= rzalloc(shProg
, struct gl_shader_variable
);
3758 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3759 * expect to see gl_VertexID in the program resource list. Pretend.
3761 if (in
->data
.mode
== ir_var_system_value
&&
3762 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3763 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3764 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3765 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3766 (in
->data
.mode
== ir_var_system_value
&&
3767 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3768 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3769 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3770 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3771 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3772 (in
->data
.mode
== ir_var_system_value
&&
3773 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3774 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3775 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3777 out
->name
= ralloc_strdup(shProg
, name
);
3783 /* The ARB_program_interface_query spec says:
3785 * "Not all active variables are assigned valid locations; the
3786 * following variables will have an effective location of -1:
3788 * * uniforms declared as atomic counters;
3790 * * members of a uniform block;
3792 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3794 * * inputs or outputs not declared with a "location" layout
3795 * qualifier, except for vertex shader inputs and fragment shader
3798 if (in
->type
->is_atomic_uint() || is_gl_identifier(in
->name
) ||
3799 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3802 out
->location
= location
;
3806 out
->outermost_struct_type
= outermost_struct_type
;
3807 out
->interface_type
= interface_type
;
3808 out
->component
= in
->data
.location_frac
;
3809 out
->index
= in
->data
.index
;
3810 out
->patch
= in
->data
.patch
;
3811 out
->mode
= in
->data
.mode
;
3812 out
->interpolation
= in
->data
.interpolation
;
3813 out
->explicit_location
= in
->data
.explicit_location
;
3814 out
->precision
= in
->data
.precision
;
3820 add_shader_variable(const struct gl_context
*ctx
,
3821 struct gl_shader_program
*shProg
,
3822 struct set
*resource_set
,
3823 unsigned stage_mask
,
3824 GLenum programInterface
, ir_variable
*var
,
3825 const char *name
, const glsl_type
*type
,
3826 bool use_implicit_location
, int location
,
3827 bool inouts_share_location
,
3828 const glsl_type
*outermost_struct_type
= NULL
)
3830 const glsl_type
*interface_type
= var
->get_interface_type();
3832 if (outermost_struct_type
== NULL
) {
3833 if (var
->data
.from_named_ifc_block
) {
3834 const char *interface_name
= interface_type
->name
;
3836 if (interface_type
->is_array()) {
3837 /* Issue #16 of the ARB_program_interface_query spec says:
3839 * "* If a variable is a member of an interface block without an
3840 * instance name, it is enumerated using just the variable name.
3842 * * If a variable is a member of an interface block with an
3843 * instance name, it is enumerated as "BlockName.Member", where
3844 * "BlockName" is the name of the interface block (not the
3845 * instance name) and "Member" is the name of the variable."
3847 * In particular, it indicates that it should be "BlockName",
3848 * not "BlockName[array length]". The conformance suite and
3849 * dEQP both require this behavior.
3851 * Here, we unwrap the extra array level added by named interface
3852 * block array lowering so we have the correct variable type. We
3853 * also unwrap the interface type when constructing the name.
3855 * We leave interface_type the same so that ES 3.x SSO pipeline
3856 * validation can enforce the rules requiring array length to
3857 * match on interface blocks.
3859 type
= type
->fields
.array
;
3861 interface_name
= interface_type
->fields
.array
->name
;
3864 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3868 switch (type
->base_type
) {
3869 case GLSL_TYPE_STRUCT
: {
3870 /* The ARB_program_interface_query spec says:
3872 * "For an active variable declared as a structure, a separate entry
3873 * will be generated for each active structure member. The name of
3874 * each entry is formed by concatenating the name of the structure,
3875 * the "." character, and the name of the structure member. If a
3876 * structure member to enumerate is itself a structure or array,
3877 * these enumeration rules are applied recursively."
3879 if (outermost_struct_type
== NULL
)
3880 outermost_struct_type
= type
;
3882 unsigned field_location
= location
;
3883 for (unsigned i
= 0; i
< type
->length
; i
++) {
3884 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3885 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3886 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3887 stage_mask
, programInterface
,
3888 var
, field_name
, field
->type
,
3889 use_implicit_location
, field_location
,
3890 false, outermost_struct_type
))
3893 field_location
+= field
->type
->count_attribute_slots(false);
3898 case GLSL_TYPE_ARRAY
: {
3899 /* The ARB_program_interface_query spec says:
3901 * "For an active variable declared as an array of basic types, a
3902 * single entry will be generated, with its name string formed by
3903 * concatenating the name of the array and the string "[0]"."
3905 * "For an active variable declared as an array of an aggregate data
3906 * type (structures or arrays), a separate entry will be generated
3907 * for each active array element, unless noted immediately below.
3908 * The name of each entry is formed by concatenating the name of
3909 * the array, the "[" character, an integer identifying the element
3910 * number, and the "]" character. These enumeration rules are
3911 * applied recursively, treating each enumerated array element as a
3912 * separate active variable."
3914 const struct glsl_type
*array_type
= type
->fields
.array
;
3915 if (array_type
->base_type
== GLSL_TYPE_STRUCT
||
3916 array_type
->base_type
== GLSL_TYPE_ARRAY
) {
3917 unsigned elem_location
= location
;
3918 unsigned stride
= inouts_share_location
? 0 :
3919 array_type
->count_attribute_slots(false);
3920 for (unsigned i
= 0; i
< type
->length
; i
++) {
3921 char *elem
= ralloc_asprintf(shProg
, "%s[%d]", name
, i
);
3922 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3923 stage_mask
, programInterface
,
3924 var
, elem
, array_type
,
3925 use_implicit_location
, elem_location
,
3926 false, outermost_struct_type
))
3928 elem_location
+= stride
;
3936 /* The ARB_program_interface_query spec says:
3938 * "For an active variable declared as a single instance of a basic
3939 * type, a single entry will be generated, using the variable name
3940 * from the shader source."
3942 gl_shader_variable
*sha_v
=
3943 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3944 use_implicit_location
, location
,
3945 outermost_struct_type
);
3949 return add_program_resource(shProg
, resource_set
,
3950 programInterface
, sha_v
, stage_mask
);
3956 inout_has_same_location(const ir_variable
*var
, unsigned stage
)
3958 if (!var
->data
.patch
&&
3959 ((var
->data
.mode
== ir_var_shader_out
&&
3960 stage
== MESA_SHADER_TESS_CTRL
) ||
3961 (var
->data
.mode
== ir_var_shader_in
&&
3962 (stage
== MESA_SHADER_TESS_CTRL
|| stage
== MESA_SHADER_TESS_EVAL
||
3963 stage
== MESA_SHADER_GEOMETRY
))))
3970 add_interface_variables(const struct gl_context
*ctx
,
3971 struct gl_shader_program
*shProg
,
3972 struct set
*resource_set
,
3973 unsigned stage
, GLenum programInterface
)
3975 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3977 foreach_in_list(ir_instruction
, node
, ir
) {
3978 ir_variable
*var
= node
->as_variable();
3980 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3985 switch (var
->data
.mode
) {
3986 case ir_var_system_value
:
3987 case ir_var_shader_in
:
3988 if (programInterface
!= GL_PROGRAM_INPUT
)
3990 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3991 : int(VARYING_SLOT_VAR0
);
3993 case ir_var_shader_out
:
3994 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3996 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3997 : int(VARYING_SLOT_VAR0
);
4003 if (var
->data
.patch
)
4004 loc_bias
= int(VARYING_SLOT_PATCH0
);
4006 /* Skip packed varyings, packed varyings are handled separately
4007 * by add_packed_varyings.
4009 if (strncmp(var
->name
, "packed:", 7) == 0)
4012 /* Skip fragdata arrays, these are handled separately
4013 * by add_fragdata_arrays.
4015 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
4018 const bool vs_input_or_fs_output
=
4019 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
4020 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
4022 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4023 1 << stage
, programInterface
,
4024 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
4025 var
->data
.location
- loc_bias
,
4026 inout_has_same_location(var
, stage
)))
4033 add_packed_varyings(const struct gl_context
*ctx
,
4034 struct gl_shader_program
*shProg
,
4035 struct set
*resource_set
,
4036 int stage
, GLenum type
)
4038 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
4041 if (!sh
|| !sh
->packed_varyings
)
4044 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
4045 ir_variable
*var
= node
->as_variable();
4047 switch (var
->data
.mode
) {
4048 case ir_var_shader_in
:
4049 iface
= GL_PROGRAM_INPUT
;
4051 case ir_var_shader_out
:
4052 iface
= GL_PROGRAM_OUTPUT
;
4055 unreachable("unexpected type");
4058 if (type
== iface
) {
4059 const int stage_mask
=
4060 build_stageref(shProg
, var
->name
, var
->data
.mode
);
4061 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4063 iface
, var
, var
->name
, var
->type
, false,
4064 var
->data
.location
- VARYING_SLOT_VAR0
,
4065 inout_has_same_location(var
, stage
)))
4074 add_fragdata_arrays(const struct gl_context
*ctx
,
4075 struct gl_shader_program
*shProg
,
4076 struct set
*resource_set
)
4078 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4080 if (!sh
|| !sh
->fragdata_arrays
)
4083 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
4084 ir_variable
*var
= node
->as_variable();
4086 assert(var
->data
.mode
== ir_var_shader_out
);
4088 if (!add_shader_variable(ctx
, shProg
, resource_set
,
4089 1 << MESA_SHADER_FRAGMENT
,
4090 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
4091 true, var
->data
.location
- FRAG_RESULT_DATA0
,
4100 get_top_level_name(const char *name
)
4102 const char *first_dot
= strchr(name
, '.');
4103 const char *first_square_bracket
= strchr(name
, '[');
4106 /* The ARB_program_interface_query spec says:
4108 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4109 * the number of active array elements of the top-level shader storage
4110 * block member containing to the active variable is written to
4111 * <params>. If the top-level block member is not declared as an
4112 * array, the value one is written to <params>. If the top-level block
4113 * member is an array with no declared size, the value zero is written
4117 /* The buffer variable is on top level.*/
4118 if (!first_square_bracket
&& !first_dot
)
4119 name_size
= strlen(name
);
4120 else if ((!first_square_bracket
||
4121 (first_dot
&& first_dot
< first_square_bracket
)))
4122 name_size
= first_dot
- name
;
4124 name_size
= first_square_bracket
- name
;
4126 return strndup(name
, name_size
);
4130 get_var_name(const char *name
)
4132 const char *first_dot
= strchr(name
, '.');
4135 return strdup(name
);
4137 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4141 is_top_level_shader_storage_block_member(const char* name
,
4142 const char* interface_name
,
4143 const char* field_name
)
4145 bool result
= false;
4147 /* If the given variable is already a top-level shader storage
4148 * block member, then return array_size = 1.
4149 * We could have two possibilities: if we have an instanced
4150 * shader storage block or not instanced.
4152 * For the first, we check create a name as it was in top level and
4153 * compare it with the real name. If they are the same, then
4154 * the variable is already at top-level.
4156 * Full instanced name is: interface name + '.' + var name +
4159 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4160 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4161 if (!full_instanced_name
) {
4162 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4166 snprintf(full_instanced_name
, name_length
, "%s.%s",
4167 interface_name
, field_name
);
4169 /* Check if its top-level shader storage block member of an
4170 * instanced interface block, or of a unnamed interface block.
4172 if (strcmp(name
, full_instanced_name
) == 0 ||
4173 strcmp(name
, field_name
) == 0)
4176 free(full_instanced_name
);
4181 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4182 char *interface_name
, char *var_name
)
4184 /* The ARB_program_interface_query spec says:
4186 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4187 * the number of active array elements of the top-level shader storage
4188 * block member containing to the active variable is written to
4189 * <params>. If the top-level block member is not declared as an
4190 * array, the value one is written to <params>. If the top-level block
4191 * member is an array with no declared size, the value zero is written
4194 if (is_top_level_shader_storage_block_member(uni
->name
,
4198 else if (field
->type
->is_unsized_array())
4200 else if (field
->type
->is_array())
4201 return field
->type
->length
;
4207 get_array_stride(struct gl_context
*ctx
, struct gl_uniform_storage
*uni
,
4208 const glsl_type
*interface
, const glsl_struct_field
*field
,
4209 char *interface_name
, char *var_name
)
4211 /* The ARB_program_interface_query spec says:
4213 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4214 * identifying the stride between array elements of the top-level
4215 * shader storage block member containing the active variable is
4216 * written to <params>. For top-level block members declared as
4217 * arrays, the value written is the difference, in basic machine units,
4218 * between the offsets of the active variable for consecutive elements
4219 * in the top-level array. For top-level block members not declared as
4220 * an array, zero is written to <params>."
4222 if (field
->type
->is_array()) {
4223 const enum glsl_matrix_layout matrix_layout
=
4224 glsl_matrix_layout(field
->matrix_layout
);
4225 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4226 const glsl_type
*array_type
= field
->type
->fields
.array
;
4228 if (is_top_level_shader_storage_block_member(uni
->name
,
4233 if (GLSL_INTERFACE_PACKING_STD140
==
4235 get_internal_ifc_packing(ctx
->Const
.UseSTD430AsDefaultPacking
)) {
4236 if (array_type
->is_record() || array_type
->is_array())
4237 return glsl_align(array_type
->std140_size(row_major
), 16);
4239 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4241 return array_type
->std430_array_stride(row_major
);
4248 calculate_array_size_and_stride(struct gl_context
*ctx
,
4249 struct gl_shader_program
*shProg
,
4250 struct gl_uniform_storage
*uni
)
4252 int block_index
= uni
->block_index
;
4253 int array_size
= -1;
4254 int array_stride
= -1;
4255 char *var_name
= get_top_level_name(uni
->name
);
4256 char *interface_name
=
4257 get_top_level_name(uni
->is_shader_storage
?
4258 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4259 shProg
->data
->UniformBlocks
[block_index
].Name
);
4261 if (strcmp(var_name
, interface_name
) == 0) {
4262 /* Deal with instanced array of SSBOs */
4263 char *temp_name
= get_var_name(uni
->name
);
4265 linker_error(shProg
, "Out of memory during linking.\n");
4266 goto write_top_level_array_size_and_stride
;
4269 var_name
= get_top_level_name(temp_name
);
4272 linker_error(shProg
, "Out of memory during linking.\n");
4273 goto write_top_level_array_size_and_stride
;
4277 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4278 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4282 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4283 ir_variable
*var
= node
->as_variable();
4284 if (!var
|| !var
->get_interface_type() ||
4285 var
->data
.mode
!= ir_var_shader_storage
)
4288 const glsl_type
*interface
= var
->get_interface_type();
4290 if (strcmp(interface_name
, interface
->name
) != 0)
4293 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4294 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4295 if (strcmp(field
->name
, var_name
) != 0)
4298 array_stride
= get_array_stride(ctx
, uni
, interface
, field
,
4299 interface_name
, var_name
);
4300 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4301 goto write_top_level_array_size_and_stride
;
4305 write_top_level_array_size_and_stride
:
4306 free(interface_name
);
4308 uni
->top_level_array_stride
= array_stride
;
4309 uni
->top_level_array_size
= array_size
;
4313 * Builds up a list of program resources that point to existing
4317 build_program_resource_list(struct gl_context
*ctx
,
4318 struct gl_shader_program
*shProg
)
4320 /* Rebuild resource list. */
4321 if (shProg
->data
->ProgramResourceList
) {
4322 ralloc_free(shProg
->data
->ProgramResourceList
);
4323 shProg
->data
->ProgramResourceList
= NULL
;
4324 shProg
->data
->NumProgramResourceList
= 0;
4327 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4329 /* Determine first input and final output stage. These are used to
4330 * detect which variables should be enumerated in the resource list
4331 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4333 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4334 if (!shProg
->_LinkedShaders
[i
])
4336 if (input_stage
== MESA_SHADER_STAGES
)
4341 /* Empty shader, no resources. */
4342 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4345 struct set
*resource_set
= _mesa_set_create(NULL
,
4347 _mesa_key_pointer_equal
);
4349 /* Program interface needs to expose varyings in case of SSO. */
4350 if (shProg
->SeparateShader
) {
4351 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4352 input_stage
, GL_PROGRAM_INPUT
))
4355 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4356 output_stage
, GL_PROGRAM_OUTPUT
))
4360 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4363 /* Add inputs and outputs to the resource list. */
4364 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4365 input_stage
, GL_PROGRAM_INPUT
))
4368 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4369 output_stage
, GL_PROGRAM_OUTPUT
))
4372 if (shProg
->last_vert_prog
) {
4373 struct gl_transform_feedback_info
*linked_xfb
=
4374 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4376 /* Add transform feedback varyings. */
4377 if (linked_xfb
->NumVarying
> 0) {
4378 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4379 if (!add_program_resource(shProg
, resource_set
,
4380 GL_TRANSFORM_FEEDBACK_VARYING
,
4381 &linked_xfb
->Varyings
[i
], 0))
4386 /* Add transform feedback buffers. */
4387 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4388 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4389 linked_xfb
->Buffers
[i
].Binding
= i
;
4390 if (!add_program_resource(shProg
, resource_set
,
4391 GL_TRANSFORM_FEEDBACK_BUFFER
,
4392 &linked_xfb
->Buffers
[i
], 0))
4398 /* Add uniforms from uniform storage. */
4399 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4400 /* Do not add uniforms internally used by Mesa. */
4401 if (shProg
->data
->UniformStorage
[i
].hidden
)
4405 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4408 /* Add stagereferences for uniforms in a uniform block. */
4409 bool is_shader_storage
=
4410 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4411 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4412 if (block_index
!= -1) {
4413 stageref
|= is_shader_storage
?
4414 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4415 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4418 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4419 if (!should_add_buffer_variable(shProg
, type
,
4420 shProg
->data
->UniformStorage
[i
].name
))
4423 if (is_shader_storage
) {
4424 calculate_array_size_and_stride(ctx
, shProg
,
4425 &shProg
->data
->UniformStorage
[i
]);
4428 if (!add_program_resource(shProg
, resource_set
, type
,
4429 &shProg
->data
->UniformStorage
[i
], stageref
))
4433 /* Add program uniform blocks. */
4434 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4435 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4436 &shProg
->data
->UniformBlocks
[i
], 0))
4440 /* Add program shader storage blocks. */
4441 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4442 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4443 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4447 /* Add atomic counter buffers. */
4448 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4449 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4450 &shProg
->data
->AtomicBuffers
[i
], 0))
4454 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4456 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4459 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4460 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4461 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4464 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4465 /* add shader subroutines */
4466 if (!add_program_resource(shProg
, resource_set
,
4467 type
, &shProg
->data
->UniformStorage
[i
], 0))
4472 unsigned mask
= shProg
->data
->linked_stages
;
4474 const int i
= u_bit_scan(&mask
);
4475 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4477 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4478 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4479 if (!add_program_resource(shProg
, resource_set
,
4480 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4485 _mesa_set_destroy(resource_set
, NULL
);
4489 * This check is done to make sure we allow only constant expression
4490 * indexing and "constant-index-expression" (indexing with an expression
4491 * that includes loop induction variable).
4494 validate_sampler_array_indexing(struct gl_context
*ctx
,
4495 struct gl_shader_program
*prog
)
4497 dynamic_sampler_array_indexing_visitor v
;
4498 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4499 if (prog
->_LinkedShaders
[i
] == NULL
)
4502 bool no_dynamic_indexing
=
4503 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4505 /* Search for array derefs in shader. */
4506 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4507 if (v
.uses_dynamic_sampler_array_indexing()) {
4508 const char *msg
= "sampler arrays indexed with non-constant "
4509 "expressions is forbidden in GLSL %s %u";
4510 /* Backend has indicated that it has no dynamic indexing support. */
4511 if (no_dynamic_indexing
) {
4512 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4513 prog
->data
->Version
);
4516 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4517 prog
->data
->Version
);
4525 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4527 unsigned mask
= prog
->data
->linked_stages
;
4529 const int i
= u_bit_scan(&mask
);
4530 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4532 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4533 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4534 ir_function
*fn
= node
->as_function();
4538 if (fn
->is_subroutine
)
4539 p
->sh
.NumSubroutineUniformTypes
++;
4541 if (!fn
->num_subroutine_types
)
4544 /* these should have been calculated earlier. */
4545 assert(fn
->subroutine_index
!= -1);
4546 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4547 linker_error(prog
, "Too many subroutine functions declared.\n");
4550 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4551 struct gl_subroutine_function
,
4552 p
->sh
.NumSubroutineFunctions
+ 1);
4553 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4554 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4555 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4556 ralloc_array(p
, const struct glsl_type
*,
4557 fn
->num_subroutine_types
);
4559 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4562 * "Each subroutine with an index qualifier in the shader must be
4563 * given a unique index, otherwise a compile or link error will be
4566 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4567 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4568 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4569 linker_error(prog
, "each subroutine index qualifier in the "
4570 "shader must be unique\n");
4574 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4575 fn
->subroutine_index
;
4577 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4578 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4580 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4581 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4582 p
->sh
.NumSubroutineFunctions
++;
4588 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4590 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4592 foreach_in_list(ir_instruction
, node
, ir
) {
4593 ir_variable
*const var
= node
->as_variable();
4595 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4598 /* Don't set always active on builtins that haven't been redeclared */
4599 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4602 var
->data
.always_active_io
= true;
4607 * When separate shader programs are enabled, only input/outputs between
4608 * the stages of a multi-stage separate program can be safely removed
4609 * from the shader interface. Other inputs/outputs must remain active.
4612 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4614 unsigned first
, last
;
4615 assert(prog
->SeparateShader
);
4617 first
= MESA_SHADER_STAGES
;
4620 /* Determine first and last stage. Excluding the compute stage */
4621 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4622 if (!prog
->_LinkedShaders
[i
])
4624 if (first
== MESA_SHADER_STAGES
)
4629 if (first
== MESA_SHADER_STAGES
)
4632 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4633 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4637 /* Prevent the removal of inputs to the first and outputs from the last
4638 * stage, unless they are the initial pipeline inputs or final pipeline
4639 * outputs, respectively.
4641 * The removal of IO between shaders in the same program is always
4644 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4645 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4646 if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4647 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4652 link_and_validate_uniforms(struct gl_context
*ctx
,
4653 struct gl_shader_program
*prog
)
4655 update_array_sizes(prog
);
4656 link_assign_uniform_locations(prog
, ctx
);
4658 link_assign_atomic_counter_resources(ctx
, prog
);
4659 link_calculate_subroutine_compat(prog
);
4660 check_resources(ctx
, prog
);
4661 check_subroutine_resources(prog
);
4662 check_image_resources(ctx
, prog
);
4663 link_check_atomic_counter_resources(ctx
, prog
);
4667 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4668 struct gl_context
*ctx
,
4669 struct gl_shader_program
*prog
, void *mem_ctx
)
4671 /* Mark all generic shader inputs and outputs as unpaired. */
4672 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4673 if (prog
->_LinkedShaders
[i
] != NULL
) {
4674 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4678 unsigned prev
= first
;
4679 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4680 if (prog
->_LinkedShaders
[i
] == NULL
)
4683 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4684 prog
->_LinkedShaders
[i
]);
4688 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4689 MESA_SHADER_VERTEX
)) {
4693 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4694 MESA_SHADER_FRAGMENT
)) {
4698 prog
->last_vert_prog
= NULL
;
4699 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4700 if (prog
->_LinkedShaders
[i
] == NULL
)
4703 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4707 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4710 link_and_validate_uniforms(ctx
, prog
);
4712 if (!prog
->data
->LinkStatus
)
4715 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4716 if (prog
->_LinkedShaders
[i
] == NULL
)
4719 const struct gl_shader_compiler_options
*options
=
4720 &ctx
->Const
.ShaderCompilerOptions
[i
];
4722 if (options
->LowerBufferInterfaceBlocks
)
4723 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4724 options
->ClampBlockIndicesToArrayBounds
,
4725 ctx
->Const
.UseSTD430AsDefaultPacking
);
4727 if (i
== MESA_SHADER_COMPUTE
)
4728 lower_shared_reference(ctx
, prog
, prog
->_LinkedShaders
[i
]);
4730 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4731 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4738 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4741 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4742 /* Run it just once. */
4743 do_common_optimization(ir
, true, false,
4744 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4745 ctx
->Const
.NativeIntegers
);
4747 /* Repeat it until it stops making changes. */
4748 while (do_common_optimization(ir
, true, false,
4749 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4750 ctx
->Const
.NativeIntegers
))
4756 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4758 prog
->data
->LinkStatus
= LINKING_SUCCESS
; /* All error paths will set this to false */
4759 prog
->data
->Validated
= false;
4761 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4763 * "Linking can fail for a variety of reasons as specified in the
4764 * OpenGL Shading Language Specification, as well as any of the
4765 * following reasons:
4767 * - No shader objects are attached to program."
4769 * The Compatibility Profile specification does not list the error. In
4770 * Compatibility Profile missing shader stages are replaced by
4771 * fixed-function. This applies to the case where all stages are
4774 if (prog
->NumShaders
== 0) {
4775 if (ctx
->API
!= API_OPENGL_COMPAT
)
4776 linker_error(prog
, "no shaders attached to the program\n");
4780 #ifdef ENABLE_SHADER_CACHE
4781 /* If transform feedback used on the program then compile all shaders. */
4782 bool skip_cache
= false;
4783 if (prog
->TransformFeedback
.NumVarying
> 0) {
4784 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4785 _mesa_glsl_compile_shader(ctx
, prog
->Shaders
[i
], false, false, true);
4790 if (!skip_cache
&& shader_cache_read_program_metadata(ctx
, prog
))
4794 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4796 prog
->ARB_fragment_coord_conventions_enable
= false;
4798 /* Separate the shaders into groups based on their type.
4800 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4801 unsigned num_shaders
[MESA_SHADER_STAGES
];
4803 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4804 shader_list
[i
] = (struct gl_shader
**)
4805 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4809 unsigned min_version
= UINT_MAX
;
4810 unsigned max_version
= 0;
4811 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4812 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4813 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4815 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4816 linker_error(prog
, "all shaders must use same shading "
4817 "language version\n");
4821 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4822 prog
->ARB_fragment_coord_conventions_enable
= true;
4825 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4826 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4827 num_shaders
[shader_type
]++;
4830 /* In desktop GLSL, different shader versions may be linked together. In
4831 * GLSL ES, all shader versions must be the same.
4833 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4834 linker_error(prog
, "all shaders must use same shading "
4835 "language version\n");
4839 prog
->data
->Version
= max_version
;
4840 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4842 /* Some shaders have to be linked with some other shaders present.
4844 if (!prog
->SeparateShader
) {
4845 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4846 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4847 linker_error(prog
, "Geometry shader must be linked with "
4851 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4852 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4853 linker_error(prog
, "Tessellation evaluation shader must be linked "
4854 "with vertex shader\n");
4857 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4858 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4859 linker_error(prog
, "Tessellation control shader must be linked with "
4864 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4866 * "Linking can fail for [...] any of the following reasons:
4868 * * program contains an object to form a tessellation control
4869 * shader [...] and [...] the program is not separable and
4870 * contains no object to form a tessellation evaluation shader"
4872 * The OpenGL spec is contradictory. It allows linking without a tess
4873 * eval shader, but that can only be used with transform feedback and
4874 * rasterization disabled. However, transform feedback isn't allowed
4875 * with GL_PATCHES, so it can't be used.
4877 * More investigation showed that the idea of transform feedback after
4878 * a tess control shader was dropped, because some hw vendors couldn't
4879 * support tessellation without a tess eval shader, but the linker
4880 * section wasn't updated to reflect that.
4882 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4885 * Do what's reasonable and always require a tess eval shader if a tess
4886 * control shader is present.
4888 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4889 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4890 linker_error(prog
, "Tessellation control shader must be linked with "
4891 "tessellation evaluation shader\n");
4896 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4897 num_shaders
[MESA_SHADER_TESS_CTRL
] == 0) {
4898 linker_error(prog
, "GLSL ES requires non-separable programs "
4899 "containing a tessellation evaluation shader to also "
4900 "be linked with a tessellation control shader\n");
4906 /* Compute shaders have additional restrictions. */
4907 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4908 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4909 linker_error(prog
, "Compute shaders may not be linked with any other "
4910 "type of shader\n");
4913 /* Link all shaders for a particular stage and validate the result.
4915 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4916 if (num_shaders
[stage
] > 0) {
4917 gl_linked_shader
*const sh
=
4918 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4919 num_shaders
[stage
], false);
4921 if (!prog
->data
->LinkStatus
) {
4923 _mesa_delete_linked_shader(ctx
, sh
);
4928 case MESA_SHADER_VERTEX
:
4929 validate_vertex_shader_executable(prog
, sh
, ctx
);
4931 case MESA_SHADER_TESS_CTRL
:
4932 /* nothing to be done */
4934 case MESA_SHADER_TESS_EVAL
:
4935 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4937 case MESA_SHADER_GEOMETRY
:
4938 validate_geometry_shader_executable(prog
, sh
, ctx
);
4940 case MESA_SHADER_FRAGMENT
:
4941 validate_fragment_shader_executable(prog
, sh
);
4944 if (!prog
->data
->LinkStatus
) {
4946 _mesa_delete_linked_shader(ctx
, sh
);
4950 prog
->_LinkedShaders
[stage
] = sh
;
4951 prog
->data
->linked_stages
|= 1 << stage
;
4955 /* Here begins the inter-stage linking phase. Some initial validation is
4956 * performed, then locations are assigned for uniforms, attributes, and
4959 cross_validate_uniforms(prog
);
4960 if (!prog
->data
->LinkStatus
)
4963 unsigned first
, last
, prev
;
4965 first
= MESA_SHADER_STAGES
;
4968 /* Determine first and last stage. */
4969 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4970 if (!prog
->_LinkedShaders
[i
])
4972 if (first
== MESA_SHADER_STAGES
)
4977 check_explicit_uniform_locations(ctx
, prog
);
4978 link_assign_subroutine_types(prog
);
4980 if (!prog
->data
->LinkStatus
)
4983 resize_tes_inputs(ctx
, prog
);
4985 /* Validate the inputs of each stage with the output of the preceding
4989 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4990 if (prog
->_LinkedShaders
[i
] == NULL
)
4993 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4994 prog
->_LinkedShaders
[i
]);
4995 if (!prog
->data
->LinkStatus
)
4998 cross_validate_outputs_to_inputs(ctx
, prog
,
4999 prog
->_LinkedShaders
[prev
],
5000 prog
->_LinkedShaders
[i
]);
5001 if (!prog
->data
->LinkStatus
)
5007 /* The cross validation of outputs/inputs above validates explicit locations
5008 * but for SSO programs we need to do this also for the inputs in the
5009 * first stage and outputs of the last stage included in the program, since
5010 * there is no cross validation for these.
5012 if (prog
->SeparateShader
)
5013 validate_sso_explicit_locations(ctx
, prog
,
5014 (gl_shader_stage
) first
,
5015 (gl_shader_stage
) last
);
5017 /* Cross-validate uniform blocks between shader stages */
5018 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
5019 if (!prog
->data
->LinkStatus
)
5022 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5023 if (prog
->_LinkedShaders
[i
] != NULL
)
5024 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
5027 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
5028 * it before optimization because we want most of the checks to get
5029 * dropped thanks to constant propagation.
5031 * This rule also applies to GLSL ES 3.00.
5033 if (max_version
>= (prog
->IsES
? 300 : 130)) {
5034 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
5036 lower_discard_flow(sh
->ir
);
5040 if (prog
->SeparateShader
)
5041 disable_varying_optimizations_for_sso(prog
);
5044 if (!interstage_cross_validate_uniform_blocks(prog
, false))
5048 if (!interstage_cross_validate_uniform_blocks(prog
, true))
5051 /* Do common optimization before assigning storage for attributes,
5052 * uniforms, and varyings. Later optimization could possibly make
5053 * some of that unused.
5055 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5056 if (prog
->_LinkedShaders
[i
] == NULL
)
5059 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
5060 if (!prog
->data
->LinkStatus
)
5063 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
5064 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
5067 if (ctx
->Const
.LowerTessLevel
) {
5068 lower_tess_level(prog
->_LinkedShaders
[i
]);
5071 /* Call opts before lowering const arrays to uniforms so we can const
5072 * propagate any elements accessed directly.
5074 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5076 /* Call opts after lowering const arrays to copy propagate things. */
5077 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
5078 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
5080 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
5083 /* Validation for special cases where we allow sampler array indexing
5084 * with loop induction variable. This check emits a warning or error
5085 * depending if backend can handle dynamic indexing.
5087 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
5088 (prog
->IsES
&& prog
->data
->Version
< 300)) {
5089 if (!validate_sampler_array_indexing(ctx
, prog
))
5093 /* Check and validate stream emissions in geometry shaders */
5094 validate_geometry_shader_emissions(ctx
, prog
);
5096 store_fragdepth_layout(prog
);
5098 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
5101 /* Linking varyings can cause some extra, useless swizzles to be generated
5102 * due to packing and unpacking.
5104 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5105 if (prog
->_LinkedShaders
[i
] == NULL
)
5108 optimize_swizzles(prog
->_LinkedShaders
[i
]->ir
);
5111 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
5112 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
5113 * anything about shader linking when one of the shaders (vertex or
5114 * fragment shader) is absent. So, the extension shouldn't change the
5115 * behavior specified in GLSL specification.
5117 * From OpenGL ES 3.1 specification (7.3 Program Objects):
5118 * "Linking can fail for a variety of reasons as specified in the
5119 * OpenGL ES Shading Language Specification, as well as any of the
5120 * following reasons:
5124 * * program contains objects to form either a vertex shader or
5125 * fragment shader, and program is not separable, and does not
5126 * contain objects to form both a vertex shader and fragment
5129 * However, the only scenario in 3.1+ where we don't require them both is
5130 * when we have a compute shader. For example:
5132 * - No shaders is a link error.
5133 * - Geom or Tess without a Vertex shader is a link error which means we
5134 * always require a Vertex shader and hence a Fragment shader.
5135 * - Finally a Compute shader linked with any other stage is a link error.
5137 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
5138 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
5139 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
5140 linker_error(prog
, "program lacks a vertex shader\n");
5141 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5142 linker_error(prog
, "program lacks a fragment shader\n");
5147 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5148 free(shader_list
[i
]);
5149 if (prog
->_LinkedShaders
[i
] == NULL
)
5152 /* Do a final validation step to make sure that the IR wasn't
5153 * invalidated by any modifications performed after intrastage linking.
5155 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5157 /* Retain any live IR, but trash the rest. */
5158 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5160 /* The symbol table in the linked shaders may contain references to
5161 * variables that were removed (e.g., unused uniforms). Since it may
5162 * contain junk, there is no possible valid use. Delete it and set the
5165 delete prog
->_LinkedShaders
[i
]->symbols
;
5166 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5169 ralloc_free(mem_ctx
);