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"
77 #include "util/string_to_uint_map.h"
79 #include "link_varyings.h"
80 #include "ir_optimization.h"
81 #include "ir_rvalue_visitor.h"
82 #include "ir_uniform.h"
84 #include "main/shaderobj.h"
85 #include "main/enums.h"
91 * Visitor that determines whether or not a variable is ever written.
93 class find_assignment_visitor
: public ir_hierarchical_visitor
{
95 find_assignment_visitor(const char *name
)
96 : name(name
), found(false)
101 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
103 ir_variable
*const var
= ir
->lhs
->variable_referenced();
105 if (strcmp(name
, var
->name
) == 0) {
110 return visit_continue_with_parent
;
113 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
115 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
116 actual_node
, &ir
->actual_parameters
) {
117 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
118 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
120 if (sig_param
->data
.mode
== ir_var_function_out
||
121 sig_param
->data
.mode
== ir_var_function_inout
) {
122 ir_variable
*var
= param_rval
->variable_referenced();
123 if (var
&& strcmp(name
, var
->name
) == 0) {
130 if (ir
->return_deref
!= NULL
) {
131 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
133 if (strcmp(name
, var
->name
) == 0) {
139 return visit_continue_with_parent
;
142 bool variable_found()
148 const char *name
; /**< Find writes to a variable with this name. */
149 bool found
; /**< Was a write to the variable found? */
154 * Visitor that determines whether or not a variable is ever read.
156 class find_deref_visitor
: public ir_hierarchical_visitor
{
158 find_deref_visitor(const char *name
)
159 : name(name
), found(false)
164 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
166 if (strcmp(this->name
, ir
->var
->name
) == 0) {
171 return visit_continue
;
174 bool variable_found() const
180 const char *name
; /**< Find writes to a variable with this name. */
181 bool found
; /**< Was a write to the variable found? */
186 * A visitor helper that provides methods for updating the types of
187 * ir_dereferences. Classes that update variable types (say, updating
188 * array sizes) will want to use this so that dereference types stay in sync.
190 class deref_type_updater
: public ir_hierarchical_visitor
{
192 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
194 ir
->type
= ir
->var
->type
;
195 return visit_continue
;
198 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
200 const glsl_type
*const vt
= ir
->array
->type
;
202 ir
->type
= vt
->fields
.array
;
203 return visit_continue
;
206 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
208 for (unsigned i
= 0; i
< ir
->record
->type
->length
; i
++) {
209 const struct glsl_struct_field
*field
=
210 &ir
->record
->type
->fields
.structure
[i
];
211 if (strcmp(field
->name
, ir
->field
) == 0) {
212 ir
->type
= field
->type
;
216 return visit_continue
;
221 class array_resize_visitor
: public deref_type_updater
{
223 unsigned num_vertices
;
224 gl_shader_program
*prog
;
225 gl_shader_stage stage
;
227 array_resize_visitor(unsigned num_vertices
,
228 gl_shader_program
*prog
,
229 gl_shader_stage stage
)
231 this->num_vertices
= num_vertices
;
236 virtual ~array_resize_visitor()
241 virtual ir_visitor_status
visit(ir_variable
*var
)
243 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
245 return visit_continue
;
247 unsigned size
= var
->type
->length
;
249 if (stage
== MESA_SHADER_GEOMETRY
) {
250 /* Generate a link error if the shader has declared this array with
253 if (!var
->data
.implicit_sized_array
&&
254 size
&& size
!= this->num_vertices
) {
255 linker_error(this->prog
, "size of array %s declared as %u, "
256 "but number of input vertices is %u\n",
257 var
->name
, size
, this->num_vertices
);
258 return visit_continue
;
261 /* Generate a link error if the shader attempts to access an input
262 * array using an index too large for its actual size assigned at
265 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
266 linker_error(this->prog
, "%s shader accesses element %i of "
267 "%s, but only %i input vertices\n",
268 _mesa_shader_stage_to_string(this->stage
),
269 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
270 return visit_continue
;
274 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
276 var
->data
.max_array_access
= this->num_vertices
- 1;
278 return visit_continue
;
283 * Visitor that determines the highest stream id to which a (geometry) shader
284 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
286 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
288 find_emit_vertex_visitor(int max_allowed
)
289 : max_stream_allowed(max_allowed
),
290 invalid_stream_id(0),
291 invalid_stream_id_from_emit_vertex(false),
292 end_primitive_found(false),
293 uses_non_zero_stream(false)
298 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
300 int stream_id
= ir
->stream_id();
303 invalid_stream_id
= stream_id
;
304 invalid_stream_id_from_emit_vertex
= true;
308 if (stream_id
> max_stream_allowed
) {
309 invalid_stream_id
= stream_id
;
310 invalid_stream_id_from_emit_vertex
= true;
315 uses_non_zero_stream
= true;
317 return visit_continue
;
320 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
322 end_primitive_found
= true;
324 int stream_id
= ir
->stream_id();
327 invalid_stream_id
= stream_id
;
328 invalid_stream_id_from_emit_vertex
= false;
332 if (stream_id
> max_stream_allowed
) {
333 invalid_stream_id
= stream_id
;
334 invalid_stream_id_from_emit_vertex
= false;
339 uses_non_zero_stream
= true;
341 return visit_continue
;
346 return invalid_stream_id
!= 0;
349 const char *error_func()
351 return invalid_stream_id_from_emit_vertex
?
352 "EmitStreamVertex" : "EndStreamPrimitive";
357 return invalid_stream_id
;
362 return uses_non_zero_stream
;
365 bool uses_end_primitive()
367 return end_primitive_found
;
371 int max_stream_allowed
;
372 int invalid_stream_id
;
373 bool invalid_stream_id_from_emit_vertex
;
374 bool end_primitive_found
;
375 bool uses_non_zero_stream
;
378 /* Class that finds array derefs and check if indexes are dynamic. */
379 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
382 dynamic_sampler_array_indexing_visitor() :
383 dynamic_sampler_array_indexing(false)
387 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
389 if (!ir
->variable_referenced())
390 return visit_continue
;
392 if (!ir
->variable_referenced()->type
->contains_sampler())
393 return visit_continue
;
395 if (!ir
->array_index
->constant_expression_value()) {
396 dynamic_sampler_array_indexing
= true;
399 return visit_continue
;
402 bool uses_dynamic_sampler_array_indexing()
404 return dynamic_sampler_array_indexing
;
408 bool dynamic_sampler_array_indexing
;
411 } /* anonymous namespace */
414 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
418 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
420 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
423 prog
->data
->LinkStatus
= false;
428 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
432 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
434 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
441 * Given a string identifying a program resource, break it into a base name
442 * and an optional array index in square brackets.
444 * If an array index is present, \c out_base_name_end is set to point to the
445 * "[" that precedes the array index, and the array index itself is returned
448 * If no array index is present (or if the array index is negative or
449 * mal-formed), \c out_base_name_end, is set to point to the null terminator
450 * at the end of the input string, and -1 is returned.
452 * Only the final array index is parsed; if the string contains other array
453 * indices (or structure field accesses), they are left in the base name.
455 * No attempt is made to check that the base name is properly formed;
456 * typically the caller will look up the base name in a hash table, so
457 * ill-formed base names simply turn into hash table lookup failures.
460 parse_program_resource_name(const GLchar
*name
,
461 const GLchar
**out_base_name_end
)
463 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
465 * "When an integer array element or block instance number is part of
466 * the name string, it will be specified in decimal form without a "+"
467 * or "-" sign or any extra leading zeroes. Additionally, the name
468 * string will not include white space anywhere in the string."
471 const size_t len
= strlen(name
);
472 *out_base_name_end
= name
+ len
;
474 if (len
== 0 || name
[len
-1] != ']')
477 /* Walk backwards over the string looking for a non-digit character. This
478 * had better be the opening bracket for an array index.
480 * Initially, i specifies the location of the ']'. Since the string may
481 * contain only the ']' charcater, walk backwards very carefully.
484 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
487 if ((i
== 0) || name
[i
-1] != '[')
490 long array_index
= strtol(&name
[i
], NULL
, 10);
494 /* Check for leading zero */
495 if (name
[i
] == '0' && name
[i
+1] != ']')
498 *out_base_name_end
= name
+ (i
- 1);
504 link_invalidate_variable_locations(exec_list
*ir
)
506 foreach_in_list(ir_instruction
, node
, ir
) {
507 ir_variable
*const var
= node
->as_variable();
512 /* Only assign locations for variables that lack an explicit location.
513 * Explicit locations are set for all built-in variables, generic vertex
514 * shader inputs (via layout(location=...)), and generic fragment shader
515 * outputs (also via layout(location=...)).
517 if (!var
->data
.explicit_location
) {
518 var
->data
.location
= -1;
519 var
->data
.location_frac
= 0;
522 /* ir_variable::is_unmatched_generic_inout is used by the linker while
523 * connecting outputs from one stage to inputs of the next stage.
525 if (var
->data
.explicit_location
&&
526 var
->data
.location
< VARYING_SLOT_VAR0
) {
527 var
->data
.is_unmatched_generic_inout
= 0;
529 var
->data
.is_unmatched_generic_inout
= 1;
536 * Set clip_distance_array_size based and cull_distance_array_size on the given
539 * Also check for errors based on incorrect usage of gl_ClipVertex and
540 * gl_ClipDistance and gl_CullDistance.
541 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
542 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
544 * Return false if an error was reported.
547 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
548 struct gl_linked_shader
*shader
,
549 struct gl_context
*ctx
,
550 GLuint
*clip_distance_array_size
,
551 GLuint
*cull_distance_array_size
)
553 *clip_distance_array_size
= 0;
554 *cull_distance_array_size
= 0;
556 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
557 /* From section 7.1 (Vertex Shader Special Variables) of the
560 * "It is an error for a shader to statically write both
561 * gl_ClipVertex and gl_ClipDistance."
563 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
564 * gl_ClipVertex nor gl_ClipDistance. However with
565 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
567 find_assignment_visitor
clip_distance("gl_ClipDistance");
568 find_assignment_visitor
cull_distance("gl_CullDistance");
570 clip_distance
.run(shader
->ir
);
571 cull_distance
.run(shader
->ir
);
573 /* From the ARB_cull_distance spec:
575 * It is a compile-time or link-time error for the set of shaders forming
576 * a program to statically read or write both gl_ClipVertex and either
577 * gl_ClipDistance or gl_CullDistance.
579 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
583 find_assignment_visitor
clip_vertex("gl_ClipVertex");
585 clip_vertex
.run(shader
->ir
);
587 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
588 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
589 "and `gl_ClipDistance'\n",
590 _mesa_shader_stage_to_string(shader
->Stage
));
593 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
594 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
595 "and `gl_CullDistance'\n",
596 _mesa_shader_stage_to_string(shader
->Stage
));
601 if (clip_distance
.variable_found()) {
602 ir_variable
*clip_distance_var
=
603 shader
->symbols
->get_variable("gl_ClipDistance");
604 assert(clip_distance_var
);
605 *clip_distance_array_size
= clip_distance_var
->type
->length
;
607 if (cull_distance
.variable_found()) {
608 ir_variable
*cull_distance_var
=
609 shader
->symbols
->get_variable("gl_CullDistance");
610 assert(cull_distance_var
);
611 *cull_distance_array_size
= cull_distance_var
->type
->length
;
613 /* From the ARB_cull_distance spec:
615 * It is a compile-time or link-time error for the set of shaders forming
616 * a program to have the sum of the sizes of the gl_ClipDistance and
617 * gl_CullDistance arrays to be larger than
618 * gl_MaxCombinedClipAndCullDistances.
620 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
621 ctx
->Const
.MaxClipPlanes
) {
622 linker_error(prog
, "%s shader: the combined size of "
623 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
625 "gl_MaxCombinedClipAndCullDistances (%u)",
626 _mesa_shader_stage_to_string(shader
->Stage
),
627 ctx
->Const
.MaxClipPlanes
);
634 * Verify that a vertex shader executable meets all semantic requirements.
636 * Also sets prog->Vert.ClipDistanceArraySize and
637 * prog->Vert.CullDistanceArraySize as a side effect.
639 * \param shader Vertex shader executable to be verified
642 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
643 struct gl_linked_shader
*shader
,
644 struct gl_context
*ctx
)
649 /* From the GLSL 1.10 spec, page 48:
651 * "The variable gl_Position is available only in the vertex
652 * language and is intended for writing the homogeneous vertex
653 * position. All executions of a well-formed vertex shader
654 * executable must write a value into this variable. [...] The
655 * variable gl_Position is available only in the vertex
656 * language and is intended for writing the homogeneous vertex
657 * position. All executions of a well-formed vertex shader
658 * executable must write a value into this variable."
660 * while in GLSL 1.40 this text is changed to:
662 * "The variable gl_Position is available only in the vertex
663 * language and is intended for writing the homogeneous vertex
664 * position. It can be written at any time during shader
665 * execution. It may also be read back by a vertex shader
666 * after being written. This value will be used by primitive
667 * assembly, clipping, culling, and other fixed functionality
668 * operations, if present, that operate on primitives after
669 * vertex processing has occurred. Its value is undefined if
670 * the vertex shader executable does not write gl_Position."
672 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
673 * gl_Position is not an error.
675 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
676 find_assignment_visitor
find("gl_Position");
677 find
.run(shader
->ir
);
678 if (!find
.variable_found()) {
681 "vertex shader does not write to `gl_Position'. "
682 "Its value is undefined. \n");
685 "vertex shader does not write to `gl_Position'. \n");
691 analyze_clip_cull_usage(prog
, shader
, ctx
,
692 &prog
->Vert
.ClipDistanceArraySize
,
693 &prog
->Vert
.CullDistanceArraySize
);
697 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
698 struct gl_linked_shader
*shader
,
699 struct gl_context
*ctx
)
704 analyze_clip_cull_usage(prog
, shader
, ctx
,
705 &prog
->TessEval
.ClipDistanceArraySize
,
706 &prog
->TessEval
.CullDistanceArraySize
);
711 * Verify that a fragment shader executable meets all semantic requirements
713 * \param shader Fragment shader executable to be verified
716 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
717 struct gl_linked_shader
*shader
)
722 find_assignment_visitor
frag_color("gl_FragColor");
723 find_assignment_visitor
frag_data("gl_FragData");
725 frag_color
.run(shader
->ir
);
726 frag_data
.run(shader
->ir
);
728 if (frag_color
.variable_found() && frag_data
.variable_found()) {
729 linker_error(prog
, "fragment shader writes to both "
730 "`gl_FragColor' and `gl_FragData'\n");
735 * Verify that a geometry shader executable meets all semantic requirements
737 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize and
738 * prog->Geom.CullDistanceArraySize as a side effect.
740 * \param shader Geometry shader executable to be verified
743 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
744 struct gl_linked_shader
*shader
,
745 struct gl_context
*ctx
)
750 unsigned num_vertices
= vertices_per_prim(shader
->info
.Geom
.InputType
);
751 prog
->Geom
.VerticesIn
= num_vertices
;
753 analyze_clip_cull_usage(prog
, shader
, ctx
,
754 &prog
->Geom
.ClipDistanceArraySize
,
755 &prog
->Geom
.CullDistanceArraySize
);
759 * Check if geometry shaders emit to non-zero streams and do corresponding
763 validate_geometry_shader_emissions(struct gl_context
*ctx
,
764 struct gl_shader_program
*prog
)
766 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
769 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
770 emit_vertex
.run(sh
->ir
);
771 if (emit_vertex
.error()) {
772 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
773 "stream parameter are in the range [0, %d].\n",
774 emit_vertex
.error_func(),
775 emit_vertex
.error_stream(),
776 ctx
->Const
.MaxVertexStreams
- 1);
778 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
779 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
781 /* From the ARB_gpu_shader5 spec:
783 * "Multiple vertex streams are supported only if the output primitive
784 * type is declared to be "points". A program will fail to link if it
785 * contains a geometry shader calling EmitStreamVertex() or
786 * EndStreamPrimitive() if its output primitive type is not "points".
788 * However, in the same spec:
790 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
791 * with <stream> set to zero."
795 * "The function EndPrimitive() is equivalent to calling
796 * EndStreamPrimitive() with <stream> set to zero."
798 * Since we can call EmitVertex() and EndPrimitive() when we output
799 * primitives other than points, calling EmitStreamVertex(0) or
800 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
801 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
802 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
805 if (prog
->Geom
.UsesStreams
&& sh
->info
.Geom
.OutputType
!= GL_POINTS
) {
806 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
807 "with n>0 requires point output\n");
813 validate_intrastage_arrays(struct gl_shader_program
*prog
,
814 ir_variable
*const var
,
815 ir_variable
*const existing
)
817 /* Consider the types to be "the same" if both types are arrays
818 * of the same type and one of the arrays is implicitly sized.
819 * In addition, set the type of the linked variable to the
820 * explicitly sized array.
822 if (var
->type
->is_array() && existing
->type
->is_array()) {
823 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
824 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
825 if (var
->type
->length
!= 0) {
826 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
827 linker_error(prog
, "%s `%s' declared as type "
828 "`%s' but outermost dimension has an index"
831 var
->name
, var
->type
->name
,
832 existing
->data
.max_array_access
);
834 existing
->type
= var
->type
;
836 } else if (existing
->type
->length
!= 0) {
837 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
838 !existing
->data
.from_ssbo_unsized_array
) {
839 linker_error(prog
, "%s `%s' declared as type "
840 "`%s' but outermost dimension has an index"
843 var
->name
, existing
->type
->name
,
844 var
->data
.max_array_access
);
849 /* The arrays of structs could have different glsl_type pointers but
850 * they are actually the same type. Use record_compare() to check that.
852 if (existing
->type
->fields
.array
->is_record() &&
853 var
->type
->fields
.array
->is_record() &&
854 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
863 * Perform validation of global variables used across multiple shaders
866 cross_validate_globals(struct gl_shader_program
*prog
,
867 struct exec_list
*ir
, glsl_symbol_table
*variables
,
870 foreach_in_list(ir_instruction
, node
, ir
) {
871 ir_variable
*const var
= node
->as_variable();
876 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
879 /* don't cross validate subroutine uniforms */
880 if (var
->type
->contains_subroutine())
883 /* Don't cross validate temporaries that are at global scope. These
884 * will eventually get pulled into the shaders 'main'.
886 if (var
->data
.mode
== ir_var_temporary
)
889 /* If a global with this name has already been seen, verify that the
890 * new instance has the same type. In addition, if the globals have
891 * initializers, the values of the initializers must be the same.
893 ir_variable
*const existing
= variables
->get_variable(var
->name
);
894 if (existing
!= NULL
) {
895 /* Check if types match. Interface blocks have some special
896 * rules so we handle those elsewhere.
898 if (var
->type
!= existing
->type
&&
899 !var
->is_interface_instance()) {
900 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
901 if (var
->type
->is_record() && existing
->type
->is_record()
902 && existing
->type
->record_compare(var
->type
)) {
903 existing
->type
= var
->type
;
905 /* If it is an unsized array in a Shader Storage Block,
906 * two different shaders can access to different elements.
907 * Because of that, they might be converted to different
908 * sized arrays, then check that they are compatible but
909 * ignore the array size.
911 if (!(var
->data
.mode
== ir_var_shader_storage
&&
912 var
->data
.from_ssbo_unsized_array
&&
913 existing
->data
.mode
== ir_var_shader_storage
&&
914 existing
->data
.from_ssbo_unsized_array
&&
915 var
->type
->gl_type
== existing
->type
->gl_type
)) {
916 linker_error(prog
, "%s `%s' declared as type "
917 "`%s' and type `%s'\n",
919 var
->name
, var
->type
->name
,
920 existing
->type
->name
);
927 if (var
->data
.explicit_location
) {
928 if (existing
->data
.explicit_location
929 && (var
->data
.location
!= existing
->data
.location
)) {
930 linker_error(prog
, "explicit locations for %s "
931 "`%s' have differing values\n",
932 mode_string(var
), var
->name
);
936 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
937 linker_error(prog
, "explicit components for %s `%s' have "
938 "differing values\n", mode_string(var
), var
->name
);
942 existing
->data
.location
= var
->data
.location
;
943 existing
->data
.explicit_location
= true;
945 /* Check if uniform with implicit location was marked explicit
946 * by earlier shader stage. If so, mark it explicit in this stage
947 * too to make sure later processing does not treat it as
950 if (existing
->data
.explicit_location
) {
951 var
->data
.location
= existing
->data
.location
;
952 var
->data
.explicit_location
= true;
956 /* From the GLSL 4.20 specification:
957 * "A link error will result if two compilation units in a program
958 * specify different integer-constant bindings for the same
959 * opaque-uniform name. However, it is not an error to specify a
960 * binding on some but not all declarations for the same name"
962 if (var
->data
.explicit_binding
) {
963 if (existing
->data
.explicit_binding
&&
964 var
->data
.binding
!= existing
->data
.binding
) {
965 linker_error(prog
, "explicit bindings for %s "
966 "`%s' have differing values\n",
967 mode_string(var
), var
->name
);
971 existing
->data
.binding
= var
->data
.binding
;
972 existing
->data
.explicit_binding
= true;
975 if (var
->type
->contains_atomic() &&
976 var
->data
.offset
!= existing
->data
.offset
) {
977 linker_error(prog
, "offset specifications for %s "
978 "`%s' have differing values\n",
979 mode_string(var
), var
->name
);
983 /* Validate layout qualifiers for gl_FragDepth.
985 * From the AMD/ARB_conservative_depth specs:
987 * "If gl_FragDepth is redeclared in any fragment shader in a
988 * program, it must be redeclared in all fragment shaders in
989 * that program that have static assignments to
990 * gl_FragDepth. All redeclarations of gl_FragDepth in all
991 * fragment shaders in a single program must have the same set
994 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
995 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
996 bool layout_differs
=
997 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
999 if (layout_declared
&& layout_differs
) {
1001 "All redeclarations of gl_FragDepth in all "
1002 "fragment shaders in a single program must have "
1003 "the same set of qualifiers.\n");
1006 if (var
->data
.used
&& layout_differs
) {
1008 "If gl_FragDepth is redeclared with a layout "
1009 "qualifier in any fragment shader, it must be "
1010 "redeclared with the same layout qualifier in "
1011 "all fragment shaders that have assignments to "
1016 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1018 * "If a shared global has multiple initializers, the
1019 * initializers must all be constant expressions, and they
1020 * must all have the same value. Otherwise, a link error will
1021 * result. (A shared global having only one initializer does
1022 * not require that initializer to be a constant expression.)"
1024 * Previous to 4.20 the GLSL spec simply said that initializers
1025 * must have the same value. In this case of non-constant
1026 * initializers, this was impossible to determine. As a result,
1027 * no vendor actually implemented that behavior. The 4.20
1028 * behavior matches the implemented behavior of at least one other
1029 * vendor, so we'll implement that for all GLSL versions.
1031 if (var
->constant_initializer
!= NULL
) {
1032 if (existing
->constant_initializer
!= NULL
) {
1033 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1034 linker_error(prog
, "initializers for %s "
1035 "`%s' have differing values\n",
1036 mode_string(var
), var
->name
);
1040 /* If the first-seen instance of a particular uniform did
1041 * not have an initializer but a later instance does,
1042 * replace the former with the later.
1044 variables
->replace_variable(existing
->name
, var
);
1048 if (var
->data
.has_initializer
) {
1049 if (existing
->data
.has_initializer
1050 && (var
->constant_initializer
== NULL
1051 || existing
->constant_initializer
== NULL
)) {
1053 "shared global variable `%s' has multiple "
1054 "non-constant initializers.\n",
1060 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1061 linker_error(prog
, "declarations for %s `%s' have "
1062 "mismatching invariant qualifiers\n",
1063 mode_string(var
), var
->name
);
1066 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1067 linker_error(prog
, "declarations for %s `%s' have "
1068 "mismatching centroid qualifiers\n",
1069 mode_string(var
), var
->name
);
1072 if (existing
->data
.sample
!= var
->data
.sample
) {
1073 linker_error(prog
, "declarations for %s `%s` have "
1074 "mismatching sample qualifiers\n",
1075 mode_string(var
), var
->name
);
1078 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1079 linker_error(prog
, "declarations for %s `%s` have "
1080 "mismatching image format qualifiers\n",
1081 mode_string(var
), var
->name
);
1085 /* Only in GLSL ES 3.10, the precision qualifier should not match
1086 * between block members defined in matched block names within a
1089 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1090 * member should match.
1092 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1093 !var
->get_interface_type()) &&
1094 existing
->data
.precision
!= var
->data
.precision
) {
1095 linker_error(prog
, "declarations for %s `%s` have "
1096 "mismatching precision qualifiers\n",
1097 mode_string(var
), var
->name
);
1101 variables
->add_variable(var
);
1107 * Perform validation of uniforms used across multiple shader stages
1110 cross_validate_uniforms(struct gl_shader_program
*prog
)
1112 glsl_symbol_table variables
;
1113 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1114 if (prog
->_LinkedShaders
[i
] == NULL
)
1117 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1123 * Accumulates the array of buffer blocks and checks that all definitions of
1124 * blocks agree on their contents.
1127 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1130 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1131 struct gl_uniform_block
*blks
= NULL
;
1132 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1133 &prog
->data
->NumUniformBlocks
;
1135 unsigned max_num_buffer_blocks
= 0;
1136 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1137 if (prog
->_LinkedShaders
[i
]) {
1138 if (validate_ssbo
) {
1139 max_num_buffer_blocks
+=
1140 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1142 max_num_buffer_blocks
+=
1143 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1148 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1149 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1151 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1152 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1153 InterfaceBlockStageIndex
[i
][j
] = -1;
1158 unsigned sh_num_blocks
;
1159 struct gl_uniform_block
**sh_blks
;
1160 if (validate_ssbo
) {
1161 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1162 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1164 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1165 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1168 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1169 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1173 linker_error(prog
, "buffer block `%s' has mismatching "
1174 "definitions\n", sh_blks
[j
]->Name
);
1176 for (unsigned k
= 0; k
<= i
; k
++) {
1177 delete[] InterfaceBlockStageIndex
[k
];
1182 InterfaceBlockStageIndex
[i
][index
] = j
;
1186 /* Update per stage block pointers to point to the program list.
1187 * FIXME: We should be able to free the per stage blocks here.
1189 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1190 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1191 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1193 if (stage_index
!= -1) {
1194 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1196 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1197 sh
->Program
->sh
.ShaderStorageBlocks
:
1198 sh
->Program
->sh
.UniformBlocks
;
1200 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1201 sh_blks
[stage_index
] = &blks
[j
];
1206 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1207 delete[] InterfaceBlockStageIndex
[i
];
1211 prog
->data
->ShaderStorageBlocks
= blks
;
1213 prog
->data
->UniformBlocks
= blks
;
1220 * Populates a shaders symbol table with all global declarations
1223 populate_symbol_table(gl_linked_shader
*sh
)
1225 sh
->symbols
= new(sh
) glsl_symbol_table
;
1227 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1231 if ((func
= inst
->as_function()) != NULL
) {
1232 sh
->symbols
->add_function(func
);
1233 } else if ((var
= inst
->as_variable()) != NULL
) {
1234 if (var
->data
.mode
!= ir_var_temporary
)
1235 sh
->symbols
->add_variable(var
);
1242 * Remap variables referenced in an instruction tree
1244 * This is used when instruction trees are cloned from one shader and placed in
1245 * another. These trees will contain references to \c ir_variable nodes that
1246 * do not exist in the target shader. This function finds these \c ir_variable
1247 * references and replaces the references with matching variables in the target
1250 * If there is no matching variable in the target shader, a clone of the
1251 * \c ir_variable is made and added to the target shader. The new variable is
1252 * added to \b both the instruction stream and the symbol table.
1254 * \param inst IR tree that is to be processed.
1255 * \param symbols Symbol table containing global scope symbols in the
1257 * \param instructions Instruction stream where new variable declarations
1261 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1264 class remap_visitor
: public ir_hierarchical_visitor
{
1266 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1268 this->target
= target
;
1269 this->symbols
= target
->symbols
;
1270 this->instructions
= target
->ir
;
1271 this->temps
= temps
;
1274 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1276 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1277 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1278 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1280 assert(var
!= NULL
);
1282 return visit_continue
;
1285 ir_variable
*const existing
=
1286 this->symbols
->get_variable(ir
->var
->name
);
1287 if (existing
!= NULL
)
1290 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1292 this->symbols
->add_variable(copy
);
1293 this->instructions
->push_head(copy
);
1297 return visit_continue
;
1301 struct gl_linked_shader
*target
;
1302 glsl_symbol_table
*symbols
;
1303 exec_list
*instructions
;
1307 remap_visitor
v(target
, temps
);
1314 * Move non-declarations from one instruction stream to another
1316 * The intended usage pattern of this function is to pass the pointer to the
1317 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1318 * pointer) for \c last and \c false for \c make_copies on the first
1319 * call. Successive calls pass the return value of the previous call for
1320 * \c last and \c true for \c make_copies.
1322 * \param instructions Source instruction stream
1323 * \param last Instruction after which new instructions should be
1324 * inserted in the target instruction stream
1325 * \param make_copies Flag selecting whether instructions in \c instructions
1326 * should be copied (via \c ir_instruction::clone) into the
1327 * target list or moved.
1330 * The new "last" instruction in the target instruction stream. This pointer
1331 * is suitable for use as the \c last parameter of a later call to this
1335 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1336 bool make_copies
, gl_linked_shader
*target
)
1338 hash_table
*temps
= NULL
;
1341 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1342 _mesa_key_pointer_equal
);
1344 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1345 if (inst
->as_function())
1348 ir_variable
*var
= inst
->as_variable();
1349 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1352 assert(inst
->as_assignment()
1354 || inst
->as_if() /* for initializers with the ?: operator */
1355 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1358 inst
= inst
->clone(target
, NULL
);
1361 _mesa_hash_table_insert(temps
, var
, inst
);
1363 remap_variables(inst
, target
, temps
);
1368 last
->insert_after(inst
);
1373 _mesa_hash_table_destroy(temps
, NULL
);
1380 * This class is only used in link_intrastage_shaders() below but declaring
1381 * it inside that function leads to compiler warnings with some versions of
1384 class array_sizing_visitor
: public deref_type_updater
{
1386 array_sizing_visitor()
1387 : mem_ctx(ralloc_context(NULL
)),
1388 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1389 _mesa_key_pointer_equal
))
1393 ~array_sizing_visitor()
1395 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1396 ralloc_free(this->mem_ctx
);
1399 virtual ir_visitor_status
visit(ir_variable
*var
)
1401 const glsl_type
*type_without_array
;
1402 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1403 fixup_type(&var
->type
, var
->data
.max_array_access
,
1404 var
->data
.from_ssbo_unsized_array
,
1405 &implicit_sized_array
);
1406 var
->data
.implicit_sized_array
= implicit_sized_array
;
1407 type_without_array
= var
->type
->without_array();
1408 if (var
->type
->is_interface()) {
1409 if (interface_contains_unsized_arrays(var
->type
)) {
1410 const glsl_type
*new_type
=
1411 resize_interface_members(var
->type
,
1412 var
->get_max_ifc_array_access(),
1413 var
->is_in_shader_storage_block());
1414 var
->type
= new_type
;
1415 var
->change_interface_type(new_type
);
1417 } else if (type_without_array
->is_interface()) {
1418 if (interface_contains_unsized_arrays(type_without_array
)) {
1419 const glsl_type
*new_type
=
1420 resize_interface_members(type_without_array
,
1421 var
->get_max_ifc_array_access(),
1422 var
->is_in_shader_storage_block());
1423 var
->change_interface_type(new_type
);
1424 var
->type
= update_interface_members_array(var
->type
, new_type
);
1426 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1427 /* Store a pointer to the variable in the unnamed_interfaces
1431 _mesa_hash_table_search(this->unnamed_interfaces
,
1434 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1436 if (interface_vars
== NULL
) {
1437 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1439 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1442 unsigned index
= ifc_type
->field_index(var
->name
);
1443 assert(index
< ifc_type
->length
);
1444 assert(interface_vars
[index
] == NULL
);
1445 interface_vars
[index
] = var
;
1447 return visit_continue
;
1451 * For each unnamed interface block that was discovered while running the
1452 * visitor, adjust the interface type to reflect the newly assigned array
1453 * sizes, and fix up the ir_variable nodes to point to the new interface
1456 void fixup_unnamed_interface_types()
1458 hash_table_call_foreach(this->unnamed_interfaces
,
1459 fixup_unnamed_interface_type
, NULL
);
1464 * If the type pointed to by \c type represents an unsized array, replace
1465 * it with a sized array whose size is determined by max_array_access.
1467 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1468 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1470 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1471 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1472 max_array_access
+ 1);
1473 *implicit_sized
= true;
1474 assert(*type
!= NULL
);
1478 static const glsl_type
*
1479 update_interface_members_array(const glsl_type
*type
,
1480 const glsl_type
*new_interface_type
)
1482 const glsl_type
*element_type
= type
->fields
.array
;
1483 if (element_type
->is_array()) {
1484 const glsl_type
*new_array_type
=
1485 update_interface_members_array(element_type
, new_interface_type
);
1486 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1488 return glsl_type::get_array_instance(new_interface_type
,
1494 * Determine whether the given interface type contains unsized arrays (if
1495 * it doesn't, array_sizing_visitor doesn't need to process it).
1497 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1499 for (unsigned i
= 0; i
< type
->length
; i
++) {
1500 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1501 if (elem_type
->is_unsized_array())
1508 * Create a new interface type based on the given type, with unsized arrays
1509 * replaced by sized arrays whose size is determined by
1510 * max_ifc_array_access.
1512 static const glsl_type
*
1513 resize_interface_members(const glsl_type
*type
,
1514 const int *max_ifc_array_access
,
1517 unsigned num_fields
= type
->length
;
1518 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1519 memcpy(fields
, type
->fields
.structure
,
1520 num_fields
* sizeof(*fields
));
1521 for (unsigned i
= 0; i
< num_fields
; i
++) {
1522 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1523 /* If SSBO last member is unsized array, we don't replace it by a sized
1526 if (is_ssbo
&& i
== (num_fields
- 1))
1527 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1528 true, &implicit_sized_array
);
1530 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1531 false, &implicit_sized_array
);
1532 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1534 glsl_interface_packing packing
=
1535 (glsl_interface_packing
) type
->interface_packing
;
1536 bool row_major
= (bool) type
->interface_row_major
;
1537 const glsl_type
*new_ifc_type
=
1538 glsl_type::get_interface_instance(fields
, num_fields
,
1539 packing
, row_major
, type
->name
);
1541 return new_ifc_type
;
1544 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1547 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1548 ir_variable
**interface_vars
= (ir_variable
**) data
;
1549 unsigned num_fields
= ifc_type
->length
;
1550 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1551 memcpy(fields
, ifc_type
->fields
.structure
,
1552 num_fields
* sizeof(*fields
));
1553 bool interface_type_changed
= false;
1554 for (unsigned i
= 0; i
< num_fields
; i
++) {
1555 if (interface_vars
[i
] != NULL
&&
1556 fields
[i
].type
!= interface_vars
[i
]->type
) {
1557 fields
[i
].type
= interface_vars
[i
]->type
;
1558 interface_type_changed
= true;
1561 if (!interface_type_changed
) {
1565 glsl_interface_packing packing
=
1566 (glsl_interface_packing
) ifc_type
->interface_packing
;
1567 bool row_major
= (bool) ifc_type
->interface_row_major
;
1568 const glsl_type
*new_ifc_type
=
1569 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1570 row_major
, ifc_type
->name
);
1572 for (unsigned i
= 0; i
< num_fields
; i
++) {
1573 if (interface_vars
[i
] != NULL
)
1574 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1579 * Memory context used to allocate the data in \c unnamed_interfaces.
1584 * Hash table from const glsl_type * to an array of ir_variable *'s
1585 * pointing to the ir_variables constituting each unnamed interface block.
1587 hash_table
*unnamed_interfaces
;
1591 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1595 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1596 struct gl_shader_program
*prog
,
1597 struct gl_linked_shader
*linked_shader
,
1598 struct gl_shader
**shader_list
,
1599 unsigned num_shaders
)
1601 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1602 linked_shader
->info
.TransformFeedback
.BufferStride
[i
] = 0;
1605 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1606 struct gl_shader
*shader
= shader_list
[i
];
1608 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1609 if (shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1610 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1611 shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1612 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] !=
1613 shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1615 "intrastage shaders defined with conflicting "
1616 "xfb_stride for buffer %d (%d and %d)\n", j
,
1618 info
.TransformFeedback
.BufferStride
[j
],
1619 shader
->info
.TransformFeedback
.BufferStride
[j
]);
1623 if (shader
->info
.TransformFeedback
.BufferStride
[j
])
1624 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] =
1625 shader
->info
.TransformFeedback
.BufferStride
[j
];
1630 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1631 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1632 prog
->TransformFeedback
.BufferStride
[j
] =
1633 linked_shader
->info
.TransformFeedback
.BufferStride
[j
];
1635 /* We will validate doubles at a later stage */
1636 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1637 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1638 "multiple of 4 or if its applied to a type that is "
1639 "or contains a double a multiple of 8.",
1640 prog
->TransformFeedback
.BufferStride
[j
]);
1644 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1645 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1647 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1648 "limit has been exceeded.");
1656 * Performs the cross-validation of tessellation control shader vertices and
1657 * layout qualifiers for the attached tessellation control shaders,
1658 * and propagates them to the linked TCS and linked shader program.
1661 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1662 struct gl_program
*gl_prog
,
1663 struct gl_shader
**shader_list
,
1664 unsigned num_shaders
)
1666 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1669 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1671 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1673 * "All tessellation control shader layout declarations in a program
1674 * must specify the same output patch vertex count. There must be at
1675 * least one layout qualifier specifying an output patch vertex count
1676 * in any program containing tessellation control shaders; however,
1677 * such a declaration is not required in all tessellation control
1681 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1682 struct gl_shader
*shader
= shader_list
[i
];
1684 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1685 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1686 gl_prog
->info
.tess
.tcs_vertices_out
!=
1687 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1688 linker_error(prog
, "tessellation control shader defined with "
1689 "conflicting output vertex count (%d and %d)\n",
1690 gl_prog
->info
.tess
.tcs_vertices_out
,
1691 shader
->info
.TessCtrl
.VerticesOut
);
1694 gl_prog
->info
.tess
.tcs_vertices_out
=
1695 shader
->info
.TessCtrl
.VerticesOut
;
1699 /* Just do the intrastage -> interstage propagation right now,
1700 * since we already know we're in the right type of shader program
1703 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1704 linker_error(prog
, "tessellation control shader didn't declare "
1705 "vertices out layout qualifier\n");
1712 * Performs the cross-validation of tessellation evaluation shader
1713 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1714 * for the attached tessellation evaluation shaders, and propagates them
1715 * to the linked TES and linked shader program.
1718 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1719 struct gl_linked_shader
*linked_shader
,
1720 struct gl_shader
**shader_list
,
1721 unsigned num_shaders
)
1723 linked_shader
->info
.TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1724 linked_shader
->info
.TessEval
.Spacing
= TESS_SPACING_UNSPECIFIED
;
1725 linked_shader
->info
.TessEval
.VertexOrder
= 0;
1726 linked_shader
->info
.TessEval
.PointMode
= -1;
1728 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1731 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1733 * "At least one tessellation evaluation shader (compilation unit) in
1734 * a program must declare a primitive mode in its input layout.
1735 * Declaration vertex spacing, ordering, and point mode identifiers is
1736 * optional. It is not required that all tessellation evaluation
1737 * shaders in a program declare a primitive mode. If spacing or
1738 * vertex ordering declarations are omitted, the tessellation
1739 * primitive generator will use equal spacing or counter-clockwise
1740 * vertex ordering, respectively. If a point mode declaration is
1741 * omitted, the tessellation primitive generator will produce lines or
1742 * triangles according to the primitive mode."
1745 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1746 struct gl_shader
*shader
= shader_list
[i
];
1748 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1749 if (linked_shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1750 linked_shader
->info
.TessEval
.PrimitiveMode
!=
1751 shader
->info
.TessEval
.PrimitiveMode
) {
1752 linker_error(prog
, "tessellation evaluation shader defined with "
1753 "conflicting input primitive modes.\n");
1756 linked_shader
->info
.TessEval
.PrimitiveMode
= shader
->info
.TessEval
.PrimitiveMode
;
1759 if (shader
->info
.TessEval
.Spacing
!= 0) {
1760 if (linked_shader
->info
.TessEval
.Spacing
!= 0 &&
1761 linked_shader
->info
.TessEval
.Spacing
!=
1762 shader
->info
.TessEval
.Spacing
) {
1763 linker_error(prog
, "tessellation evaluation shader defined with "
1764 "conflicting vertex spacing.\n");
1767 linked_shader
->info
.TessEval
.Spacing
= shader
->info
.TessEval
.Spacing
;
1770 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1771 if (linked_shader
->info
.TessEval
.VertexOrder
!= 0 &&
1772 linked_shader
->info
.TessEval
.VertexOrder
!=
1773 shader
->info
.TessEval
.VertexOrder
) {
1774 linker_error(prog
, "tessellation evaluation shader defined with "
1775 "conflicting ordering.\n");
1778 linked_shader
->info
.TessEval
.VertexOrder
=
1779 shader
->info
.TessEval
.VertexOrder
;
1782 if (shader
->info
.TessEval
.PointMode
!= -1) {
1783 if (linked_shader
->info
.TessEval
.PointMode
!= -1 &&
1784 linked_shader
->info
.TessEval
.PointMode
!=
1785 shader
->info
.TessEval
.PointMode
) {
1786 linker_error(prog
, "tessellation evaluation shader defined with "
1787 "conflicting point modes.\n");
1790 linked_shader
->info
.TessEval
.PointMode
=
1791 shader
->info
.TessEval
.PointMode
;
1796 /* Just do the intrastage -> interstage propagation right now,
1797 * since we already know we're in the right type of shader program
1800 if (linked_shader
->info
.TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1802 "tessellation evaluation shader didn't declare input "
1803 "primitive modes.\n");
1807 if (linked_shader
->info
.TessEval
.Spacing
== TESS_SPACING_UNSPECIFIED
)
1808 linked_shader
->info
.TessEval
.Spacing
= TESS_SPACING_EQUAL
;
1810 if (linked_shader
->info
.TessEval
.VertexOrder
== 0)
1811 linked_shader
->info
.TessEval
.VertexOrder
= GL_CCW
;
1813 if (linked_shader
->info
.TessEval
.PointMode
== -1)
1814 linked_shader
->info
.TessEval
.PointMode
= GL_FALSE
;
1819 * Performs the cross-validation of layout qualifiers specified in
1820 * redeclaration of gl_FragCoord for the attached fragment shaders,
1821 * and propagates them to the linked FS and linked shader program.
1824 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1825 struct gl_linked_shader
*linked_shader
,
1826 struct gl_shader
**shader_list
,
1827 unsigned num_shaders
)
1829 bool redeclares_gl_fragcoord
= false;
1830 bool uses_gl_fragcoord
= false;
1831 bool origin_upper_left
= false;
1832 bool pixel_center_integer
= false;
1834 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1835 (prog
->data
->Version
< 150 &&
1836 !prog
->ARB_fragment_coord_conventions_enable
))
1839 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1840 struct gl_shader
*shader
= shader_list
[i
];
1841 /* From the GLSL 1.50 spec, page 39:
1843 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1844 * it must be redeclared in all the fragment shaders in that program
1845 * that have a static use gl_FragCoord."
1847 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1848 shader
->uses_gl_fragcoord
)
1849 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1850 uses_gl_fragcoord
)) {
1851 linker_error(prog
, "fragment shader defined with conflicting "
1852 "layout qualifiers for gl_FragCoord\n");
1855 /* From the GLSL 1.50 spec, page 39:
1857 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1858 * single program must have the same set of qualifiers."
1860 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1861 (shader
->origin_upper_left
!= origin_upper_left
||
1862 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1863 linker_error(prog
, "fragment shader defined with conflicting "
1864 "layout qualifiers for gl_FragCoord\n");
1867 /* Update the linked shader state. Note that uses_gl_fragcoord should
1868 * accumulate the results. The other values should replace. If there
1869 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1870 * are already known to be the same.
1872 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1873 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1874 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1875 origin_upper_left
= shader
->origin_upper_left
;
1876 pixel_center_integer
= shader
->pixel_center_integer
;
1879 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1880 shader
->EarlyFragmentTests
;
1881 linked_shader
->info
.InnerCoverage
|=
1882 shader
->info
.InnerCoverage
;
1883 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1884 shader
->info
.PostDepthCoverage
;
1886 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1891 * Performs the cross-validation of geometry shader max_vertices and
1892 * primitive type layout qualifiers for the attached geometry shaders,
1893 * and propagates them to the linked GS and linked shader program.
1896 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1897 struct gl_linked_shader
*linked_shader
,
1898 struct gl_shader
**shader_list
,
1899 unsigned num_shaders
)
1901 linked_shader
->info
.Geom
.VerticesOut
= -1;
1902 linked_shader
->info
.Geom
.Invocations
= 0;
1903 linked_shader
->info
.Geom
.InputType
= PRIM_UNKNOWN
;
1904 linked_shader
->info
.Geom
.OutputType
= PRIM_UNKNOWN
;
1906 /* No in/out qualifiers defined for anything but GLSL 1.50+
1907 * geometry shaders so far.
1909 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
||
1910 prog
->data
->Version
< 150)
1913 /* From the GLSL 1.50 spec, page 46:
1915 * "All geometry shader output layout declarations in a program
1916 * must declare the same layout and same value for
1917 * max_vertices. There must be at least one geometry output
1918 * layout declaration somewhere in a program, but not all
1919 * geometry shaders (compilation units) are required to
1923 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1924 struct gl_shader
*shader
= shader_list
[i
];
1926 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1927 if (linked_shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
&&
1928 linked_shader
->info
.Geom
.InputType
!=
1929 shader
->info
.Geom
.InputType
) {
1930 linker_error(prog
, "geometry shader defined with conflicting "
1934 linked_shader
->info
.Geom
.InputType
= shader
->info
.Geom
.InputType
;
1937 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1938 if (linked_shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
&&
1939 linked_shader
->info
.Geom
.OutputType
!=
1940 shader
->info
.Geom
.OutputType
) {
1941 linker_error(prog
, "geometry shader defined with conflicting "
1945 linked_shader
->info
.Geom
.OutputType
= shader
->info
.Geom
.OutputType
;
1948 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1949 if (linked_shader
->info
.Geom
.VerticesOut
!= -1 &&
1950 linked_shader
->info
.Geom
.VerticesOut
!=
1951 shader
->info
.Geom
.VerticesOut
) {
1952 linker_error(prog
, "geometry shader defined with conflicting "
1953 "output vertex count (%d and %d)\n",
1954 linked_shader
->info
.Geom
.VerticesOut
,
1955 shader
->info
.Geom
.VerticesOut
);
1958 linked_shader
->info
.Geom
.VerticesOut
= shader
->info
.Geom
.VerticesOut
;
1961 if (shader
->info
.Geom
.Invocations
!= 0) {
1962 if (linked_shader
->info
.Geom
.Invocations
!= 0 &&
1963 linked_shader
->info
.Geom
.Invocations
!=
1964 shader
->info
.Geom
.Invocations
) {
1965 linker_error(prog
, "geometry shader defined with conflicting "
1966 "invocation count (%d and %d)\n",
1967 linked_shader
->info
.Geom
.Invocations
,
1968 shader
->info
.Geom
.Invocations
);
1971 linked_shader
->info
.Geom
.Invocations
= shader
->info
.Geom
.Invocations
;
1975 /* Just do the intrastage -> interstage propagation right now,
1976 * since we already know we're in the right type of shader program
1979 if (linked_shader
->info
.Geom
.InputType
== PRIM_UNKNOWN
) {
1981 "geometry shader didn't declare primitive input type\n");
1985 if (linked_shader
->info
.Geom
.OutputType
== PRIM_UNKNOWN
) {
1987 "geometry shader didn't declare primitive output type\n");
1991 if (linked_shader
->info
.Geom
.VerticesOut
== -1) {
1993 "geometry shader didn't declare max_vertices\n");
1997 if (linked_shader
->info
.Geom
.Invocations
== 0)
1998 linked_shader
->info
.Geom
.Invocations
= 1;
2003 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2004 * qualifiers for the attached compute shaders, and propagate them to the
2005 * linked CS and linked shader program.
2008 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2009 struct gl_linked_shader
*linked_shader
,
2010 struct gl_shader
**shader_list
,
2011 unsigned num_shaders
)
2013 for (int i
= 0; i
< 3; i
++)
2014 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
2016 linked_shader
->info
.Comp
.LocalSizeVariable
= false;
2018 /* This function is called for all shader stages, but it only has an effect
2019 * for compute shaders.
2021 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2024 /* From the ARB_compute_shader spec, in the section describing local size
2027 * If multiple compute shaders attached to a single program object
2028 * declare local work-group size, the declarations must be identical;
2029 * otherwise a link-time error results. Furthermore, if a program
2030 * object contains any compute shaders, at least one must contain an
2031 * input layout qualifier specifying the local work sizes of the
2032 * program, or a link-time error will occur.
2034 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2035 struct gl_shader
*shader
= shader_list
[sh
];
2037 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2038 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2039 for (int i
= 0; i
< 3; i
++) {
2040 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2041 shader
->info
.Comp
.LocalSize
[i
]) {
2042 linker_error(prog
, "compute shader defined with conflicting "
2048 for (int i
= 0; i
< 3; i
++) {
2049 linked_shader
->info
.Comp
.LocalSize
[i
] =
2050 shader
->info
.Comp
.LocalSize
[i
];
2052 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2053 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2054 /* The ARB_compute_variable_group_size spec says:
2056 * If one compute shader attached to a program declares a
2057 * variable local group size and a second compute shader
2058 * attached to the same program declares a fixed local group
2059 * size, a link-time error results.
2061 linker_error(prog
, "compute shader defined with both fixed and "
2062 "variable local group size\n");
2065 linked_shader
->info
.Comp
.LocalSizeVariable
= true;
2069 /* Just do the intrastage -> interstage propagation right now,
2070 * since we already know we're in the right type of shader program
2073 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0 &&
2074 !linked_shader
->info
.Comp
.LocalSizeVariable
) {
2075 linker_error(prog
, "compute shader must contain a fixed or a variable "
2076 "local group size\n");
2079 for (int i
= 0; i
< 3; i
++)
2080 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2082 prog
->Comp
.LocalSizeVariable
=
2083 linked_shader
->info
.Comp
.LocalSizeVariable
;
2088 * Combine a group of shaders for a single stage to generate a linked shader
2091 * If this function is supplied a single shader, it is cloned, and the new
2092 * shader is returned.
2094 struct gl_linked_shader
*
2095 link_intrastage_shaders(void *mem_ctx
,
2096 struct gl_context
*ctx
,
2097 struct gl_shader_program
*prog
,
2098 struct gl_shader
**shader_list
,
2099 unsigned num_shaders
,
2100 bool allow_missing_main
)
2102 struct gl_uniform_block
*ubo_blocks
= NULL
;
2103 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2104 unsigned num_ubo_blocks
= 0;
2105 unsigned num_ssbo_blocks
= 0;
2107 /* Check that global variables defined in multiple shaders are consistent.
2109 glsl_symbol_table variables
;
2110 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2111 if (shader_list
[i
] == NULL
)
2113 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2116 if (!prog
->data
->LinkStatus
)
2119 /* Check that interface blocks defined in multiple shaders are consistent.
2121 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2123 if (!prog
->data
->LinkStatus
)
2126 /* Check that there is only a single definition of each function signature
2127 * across all shaders.
2129 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2130 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2131 ir_function
*const f
= node
->as_function();
2136 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2137 ir_function
*const other
=
2138 shader_list
[j
]->symbols
->get_function(f
->name
);
2140 /* If the other shader has no function (and therefore no function
2141 * signatures) with the same name, skip to the next shader.
2146 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2147 if (!sig
->is_defined
)
2150 ir_function_signature
*other_sig
=
2151 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2153 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2154 linker_error(prog
, "function `%s' is multiply defined\n",
2163 /* Find the shader that defines main, and make a clone of it.
2165 * Starting with the clone, search for undefined references. If one is
2166 * found, find the shader that defines it. Clone the reference and add
2167 * it to the shader. Repeat until there are no undefined references or
2168 * until a reference cannot be resolved.
2170 gl_shader
*main
= NULL
;
2171 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2172 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2173 main
= shader_list
[i
];
2178 if (main
== NULL
&& allow_missing_main
)
2179 main
= shader_list
[0];
2182 linker_error(prog
, "%s shader lacks `main'\n",
2183 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2187 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2188 linked
->Stage
= shader_list
[0]->Stage
;
2190 /* Create program and attach it to the linked shader */
2191 struct gl_program
*gl_prog
=
2192 ctx
->Driver
.NewProgram(ctx
,
2193 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2196 prog
->data
->LinkStatus
= false;
2197 _mesa_delete_linked_shader(ctx
, linked
);
2201 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2203 /* Don't use _mesa_reference_program() just take ownership */
2204 linked
->Program
= gl_prog
;
2206 linked
->ir
= new(linked
) exec_list
;
2207 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2209 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2210 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2211 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2212 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2213 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2214 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2217 populate_symbol_table(linked
);
2219 /* The pointer to the main function in the final linked shader (i.e., the
2220 * copy of the original shader that contained the main function).
2222 ir_function_signature
*const main_sig
=
2223 _mesa_get_main_function_signature(linked
->symbols
);
2225 /* Move any instructions other than variable declarations or function
2226 * declarations into main.
2228 if (main_sig
!= NULL
) {
2229 exec_node
*insertion_point
=
2230 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2233 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2234 if (shader_list
[i
] == main
)
2237 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2238 insertion_point
, true, linked
);
2242 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2243 _mesa_delete_linked_shader(ctx
, linked
);
2247 /* Make a pass over all variable declarations to ensure that arrays with
2248 * unspecified sizes have a size specified. The size is inferred from the
2249 * max_array_access field.
2251 array_sizing_visitor v
;
2253 v
.fixup_unnamed_interface_types();
2255 /* Link up uniform blocks defined within this stage. */
2256 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2257 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2259 if (!prog
->data
->LinkStatus
) {
2260 _mesa_delete_linked_shader(ctx
, linked
);
2264 /* Copy ubo blocks to linked shader list */
2265 linked
->Program
->sh
.UniformBlocks
=
2266 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2267 ralloc_steal(linked
, ubo_blocks
);
2268 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2269 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2271 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2273 /* Copy ssbo blocks to linked shader list */
2274 linked
->Program
->sh
.ShaderStorageBlocks
=
2275 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2276 ralloc_steal(linked
, ssbo_blocks
);
2277 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2278 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2280 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2282 /* At this point linked should contain all of the linked IR, so
2283 * validate it to make sure nothing went wrong.
2285 validate_ir_tree(linked
->ir
);
2287 /* Set the size of geometry shader input arrays */
2288 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2289 unsigned num_vertices
= vertices_per_prim(linked
->info
.Geom
.InputType
);
2290 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2291 MESA_SHADER_GEOMETRY
);
2292 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2293 ir
->accept(&input_resize_visitor
);
2297 if (ctx
->Const
.VertexID_is_zero_based
)
2298 lower_vertex_id(linked
);
2301 /* Compute the source checksum. */
2302 linked
->SourceChecksum
= 0;
2303 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2304 if (shader_list
[i
] == NULL
)
2306 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2314 * Update the sizes of linked shader uniform arrays to the maximum
2317 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2319 * If one or more elements of an array are active,
2320 * GetActiveUniform will return the name of the array in name,
2321 * subject to the restrictions listed above. The type of the array
2322 * is returned in type. The size parameter contains the highest
2323 * array element index used, plus one. The compiler or linker
2324 * determines the highest index used. There will be only one
2325 * active uniform reported by the GL per uniform array.
2329 update_array_sizes(struct gl_shader_program
*prog
)
2331 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2332 if (prog
->_LinkedShaders
[i
] == NULL
)
2335 bool types_were_updated
= false;
2337 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2338 ir_variable
*const var
= node
->as_variable();
2340 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2341 !var
->type
->is_array())
2344 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2345 * will not be eliminated. Since we always do std140, just
2346 * don't resize arrays in UBOs.
2348 * Atomic counters are supposed to get deterministic
2349 * locations assigned based on the declaration ordering and
2350 * sizes, array compaction would mess that up.
2352 * Subroutine uniforms are not removed.
2354 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2355 var
->type
->contains_subroutine() || var
->constant_initializer
)
2358 int size
= var
->data
.max_array_access
;
2359 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2360 if (prog
->_LinkedShaders
[j
] == NULL
)
2363 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2364 ir_variable
*other_var
= node2
->as_variable();
2368 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2369 other_var
->data
.max_array_access
> size
) {
2370 size
= other_var
->data
.max_array_access
;
2375 if (size
+ 1 != (int)var
->type
->length
) {
2376 /* If this is a built-in uniform (i.e., it's backed by some
2377 * fixed-function state), adjust the number of state slots to
2378 * match the new array size. The number of slots per array entry
2379 * is not known. It seems safe to assume that the total number of
2380 * slots is an integer multiple of the number of array elements.
2381 * Determine the number of slots per array element by dividing by
2382 * the old (total) size.
2384 const unsigned num_slots
= var
->get_num_state_slots();
2385 if (num_slots
> 0) {
2386 var
->set_num_state_slots((size
+ 1)
2387 * (num_slots
/ var
->type
->length
));
2390 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2392 types_were_updated
= true;
2396 /* Update the types of dereferences in case we changed any. */
2397 if (types_were_updated
) {
2398 deref_type_updater v
;
2399 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2405 * Resize tessellation evaluation per-vertex inputs to the size of
2406 * tessellation control per-vertex outputs.
2409 resize_tes_inputs(struct gl_context
*ctx
,
2410 struct gl_shader_program
*prog
)
2412 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2415 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2416 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2418 /* If no control shader is present, then the TES inputs are statically
2419 * sized to MaxPatchVertices; the actual size of the arrays won't be
2420 * known until draw time.
2422 const int num_vertices
= tcs
2423 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2424 : ctx
->Const
.MaxPatchVertices
;
2426 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2427 MESA_SHADER_TESS_EVAL
);
2428 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2429 ir
->accept(&input_resize_visitor
);
2432 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2433 /* Convert the gl_PatchVerticesIn system value into a constant, since
2434 * the value is known at this point.
2436 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2437 ir_variable
*var
= ir
->as_variable();
2438 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2439 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2440 void *mem_ctx
= ralloc_parent(var
);
2441 var
->data
.location
= 0;
2442 var
->data
.explicit_location
= false;
2444 var
->data
.mode
= ir_var_auto
;
2445 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2447 var
->data
.mode
= ir_var_uniform
;
2448 var
->data
.how_declared
= ir_var_hidden
;
2449 var
->allocate_state_slots(1);
2450 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2451 slot0
->swizzle
= SWIZZLE_XXXX
;
2452 slot0
->tokens
[0] = STATE_INTERNAL
;
2453 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2454 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2455 slot0
->tokens
[i
] = 0;
2463 * Find a contiguous set of available bits in a bitmask.
2465 * \param used_mask Bits representing used (1) and unused (0) locations
2466 * \param needed_count Number of contiguous bits needed.
2469 * Base location of the available bits on success or -1 on failure.
2472 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2474 unsigned needed_mask
= (1 << needed_count
) - 1;
2475 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2477 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2478 * cannot optimize possibly infinite loops" for the loop below.
2480 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2483 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2484 if ((needed_mask
& ~used_mask
) == needed_mask
)
2495 * Assign locations for either VS inputs or FS outputs
2497 * \param mem_ctx Temporary ralloc context used for linking
2498 * \param prog Shader program whose variables need locations assigned
2499 * \param constants Driver specific constant values for the program.
2500 * \param target_index Selector for the program target to receive location
2501 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2502 * \c MESA_SHADER_FRAGMENT.
2505 * If locations are successfully assigned, true is returned. Otherwise an
2506 * error is emitted to the shader link log and false is returned.
2509 assign_attribute_or_color_locations(void *mem_ctx
,
2510 gl_shader_program
*prog
,
2511 struct gl_constants
*constants
,
2512 unsigned target_index
)
2514 /* Maximum number of generic locations. This corresponds to either the
2515 * maximum number of draw buffers or the maximum number of generic
2518 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2519 constants
->Program
[target_index
].MaxAttribs
:
2520 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2522 /* Mark invalid locations as being used.
2524 unsigned used_locations
= (max_index
>= 32)
2525 ? ~0 : ~((1 << max_index
) - 1);
2526 unsigned double_storage_locations
= 0;
2528 assert((target_index
== MESA_SHADER_VERTEX
)
2529 || (target_index
== MESA_SHADER_FRAGMENT
));
2531 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2535 /* Operate in a total of four passes.
2537 * 1. Invalidate the location assignments for all vertex shader inputs.
2539 * 2. Assign locations for inputs that have user-defined (via
2540 * glBindVertexAttribLocation) locations and outputs that have
2541 * user-defined locations (via glBindFragDataLocation).
2543 * 3. Sort the attributes without assigned locations by number of slots
2544 * required in decreasing order. Fragmentation caused by attribute
2545 * locations assigned by the application may prevent large attributes
2546 * from having enough contiguous space.
2548 * 4. Assign locations to any inputs without assigned locations.
2551 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2552 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2554 const enum ir_variable_mode direction
=
2555 (target_index
== MESA_SHADER_VERTEX
)
2556 ? ir_var_shader_in
: ir_var_shader_out
;
2559 /* Temporary storage for the set of attributes that need locations assigned.
2565 /* Used below in the call to qsort. */
2566 static int compare(const void *a
, const void *b
)
2568 const temp_attr
*const l
= (const temp_attr
*) a
;
2569 const temp_attr
*const r
= (const temp_attr
*) b
;
2571 /* Reversed because we want a descending order sort below. */
2572 return r
->slots
- l
->slots
;
2575 assert(max_index
<= 32);
2577 /* Temporary array for the set of attributes that have locations assigned.
2579 ir_variable
*assigned
[16];
2581 unsigned num_attr
= 0;
2582 unsigned assigned_attr
= 0;
2584 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2585 ir_variable
*const var
= node
->as_variable();
2587 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2590 if (var
->data
.explicit_location
) {
2591 var
->data
.is_unmatched_generic_inout
= 0;
2592 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2593 || (var
->data
.location
< 0)) {
2595 "invalid explicit location %d specified for `%s'\n",
2596 (var
->data
.location
< 0)
2597 ? var
->data
.location
2598 : var
->data
.location
- generic_base
,
2602 } else if (target_index
== MESA_SHADER_VERTEX
) {
2605 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2606 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2607 var
->data
.location
= binding
;
2608 var
->data
.is_unmatched_generic_inout
= 0;
2610 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2613 const char *name
= var
->name
;
2614 const glsl_type
*type
= var
->type
;
2617 /* Check if there's a binding for the variable name */
2618 if (prog
->FragDataBindings
->get(binding
, name
)) {
2619 assert(binding
>= FRAG_RESULT_DATA0
);
2620 var
->data
.location
= binding
;
2621 var
->data
.is_unmatched_generic_inout
= 0;
2623 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2624 var
->data
.index
= index
;
2629 /* If not, but it's an array type, look for name[0] */
2630 if (type
->is_array()) {
2631 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2632 type
= type
->fields
.array
;
2640 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2643 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2645 * "Output binding assignments will cause LinkProgram to fail:
2647 * If the program has an active output assigned to a location greater
2648 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2649 * an active output assigned an index greater than or equal to one;"
2651 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2652 var
->data
.location
- generic_base
>=
2653 (int) constants
->MaxDualSourceDrawBuffers
) {
2655 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2656 "with index %u for %s\n",
2657 var
->data
.location
- generic_base
, var
->data
.index
,
2662 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2664 /* If the variable is not a built-in and has a location statically
2665 * assigned in the shader (presumably via a layout qualifier), make sure
2666 * that it doesn't collide with other assigned locations. Otherwise,
2667 * add it to the list of variables that need linker-assigned locations.
2669 if (var
->data
.location
!= -1) {
2670 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2671 /* From page 61 of the OpenGL 4.0 spec:
2673 * "LinkProgram will fail if the attribute bindings assigned
2674 * by BindAttribLocation do not leave not enough space to
2675 * assign a location for an active matrix attribute or an
2676 * active attribute array, both of which require multiple
2677 * contiguous generic attributes."
2679 * I think above text prohibits the aliasing of explicit and
2680 * automatic assignments. But, aliasing is allowed in manual
2681 * assignments of attribute locations. See below comments for
2684 * From OpenGL 4.0 spec, page 61:
2686 * "It is possible for an application to bind more than one
2687 * attribute name to the same location. This is referred to as
2688 * aliasing. This will only work if only one of the aliased
2689 * attributes is active in the executable program, or if no
2690 * path through the shader consumes more than one attribute of
2691 * a set of attributes aliased to the same location. A link
2692 * error can occur if the linker determines that every path
2693 * through the shader consumes multiple aliased attributes,
2694 * but implementations are not required to generate an error
2697 * From GLSL 4.30 spec, page 54:
2699 * "A program will fail to link if any two non-vertex shader
2700 * input variables are assigned to the same location. For
2701 * vertex shaders, multiple input variables may be assigned
2702 * to the same location using either layout qualifiers or via
2703 * the OpenGL API. However, such aliasing is intended only to
2704 * support vertex shaders where each execution path accesses
2705 * at most one input per each location. Implementations are
2706 * permitted, but not required, to generate link-time errors
2707 * if they detect that every path through the vertex shader
2708 * executable accesses multiple inputs assigned to any single
2709 * location. For all shader types, a program will fail to link
2710 * if explicit location assignments leave the linker unable
2711 * to find space for other variables without explicit
2714 * From OpenGL ES 3.0 spec, page 56:
2716 * "Binding more than one attribute name to the same location
2717 * is referred to as aliasing, and is not permitted in OpenGL
2718 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2719 * fail when this condition exists. However, aliasing is
2720 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2721 * This will only work if only one of the aliased attributes
2722 * is active in the executable program, or if no path through
2723 * the shader consumes more than one attribute of a set of
2724 * attributes aliased to the same location. A link error can
2725 * occur if the linker determines that every path through the
2726 * shader consumes multiple aliased attributes, but implemen-
2727 * tations are not required to generate an error in this case."
2729 * After looking at above references from OpenGL, OpenGL ES and
2730 * GLSL specifications, we allow aliasing of vertex input variables
2731 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2733 * NOTE: This is not required by the spec but its worth mentioning
2734 * here that we're not doing anything to make sure that no path
2735 * through the vertex shader executable accesses multiple inputs
2736 * assigned to any single location.
2739 /* Mask representing the contiguous slots that will be used by
2742 const unsigned attr
= var
->data
.location
- generic_base
;
2743 const unsigned use_mask
= (1 << slots
) - 1;
2744 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2745 ? "vertex shader input" : "fragment shader output";
2747 /* Generate a link error if the requested locations for this
2748 * attribute exceed the maximum allowed attribute location.
2750 if (attr
+ slots
> max_index
) {
2752 "insufficient contiguous locations "
2753 "available for %s `%s' %d %d %d\n", string
,
2754 var
->name
, used_locations
, use_mask
, attr
);
2758 /* Generate a link error if the set of bits requested for this
2759 * attribute overlaps any previously allocated bits.
2761 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2762 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2763 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2766 * "Additionally, for fragment shader outputs, if two
2767 * variables are placed within the same location, they
2768 * must have the same underlying type (floating-point or
2769 * integer). No component aliasing of output variables or
2770 * members is allowed.
2772 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2773 unsigned assigned_slots
=
2774 assigned
[i
]->type
->count_attribute_slots(false);
2775 unsigned assig_attr
=
2776 assigned
[i
]->data
.location
- generic_base
;
2777 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2779 if ((assigned_use_mask
<< assig_attr
) &
2780 (use_mask
<< attr
)) {
2782 const glsl_type
*assigned_type
=
2783 assigned
[i
]->type
->without_array();
2784 const glsl_type
*type
= var
->type
->without_array();
2785 if (assigned_type
->base_type
!= type
->base_type
) {
2786 linker_error(prog
, "types do not match for aliased"
2787 " %ss %s and %s\n", string
,
2788 assigned
[i
]->name
, var
->name
);
2792 unsigned assigned_component_mask
=
2793 ((1 << assigned_type
->vector_elements
) - 1) <<
2794 assigned
[i
]->data
.location_frac
;
2795 unsigned component_mask
=
2796 ((1 << type
->vector_elements
) - 1) <<
2797 var
->data
.location_frac
;
2798 if (assigned_component_mask
& component_mask
) {
2799 linker_error(prog
, "overlapping component is "
2800 "assigned to %ss %s and %s "
2802 string
, assigned
[i
]->name
, var
->name
,
2803 var
->data
.location_frac
);
2808 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2809 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2810 linker_error(prog
, "overlapping location is assigned "
2811 "to %s `%s' %d %d %d\n", string
, var
->name
,
2812 used_locations
, use_mask
, attr
);
2815 linker_warning(prog
, "overlapping location is assigned "
2816 "to %s `%s' %d %d %d\n", string
, var
->name
,
2817 used_locations
, use_mask
, attr
);
2821 used_locations
|= (use_mask
<< attr
);
2823 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2825 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2826 * active attribute variables may fail to link, unless
2827 * device-dependent optimizations are able to make the program
2828 * fit within available hardware resources. For the purposes
2829 * of this test, attribute variables of the type dvec3, dvec4,
2830 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2831 * count as consuming twice as many attributes as equivalent
2832 * single-precision types. While these types use the same number
2833 * of generic attributes as their single-precision equivalents,
2834 * implementations are permitted to consume two single-precision
2835 * vectors of internal storage for each three- or four-component
2836 * double-precision vector."
2838 * Mark this attribute slot as taking up twice as much space
2839 * so we can count it properly against limits. According to
2840 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2841 * is optional behavior, but it seems preferable.
2843 if (var
->type
->without_array()->is_dual_slot())
2844 double_storage_locations
|= (use_mask
<< attr
);
2847 assigned
[assigned_attr
] = var
;
2853 if (num_attr
>= max_index
) {
2854 linker_error(prog
, "too many %s (max %u)",
2855 target_index
== MESA_SHADER_VERTEX
?
2856 "vertex shader inputs" : "fragment shader outputs",
2860 to_assign
[num_attr
].slots
= slots
;
2861 to_assign
[num_attr
].var
= var
;
2865 if (target_index
== MESA_SHADER_VERTEX
) {
2866 unsigned total_attribs_size
=
2867 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2868 _mesa_bitcount(double_storage_locations
);
2869 if (total_attribs_size
> max_index
) {
2871 "attempt to use %d vertex attribute slots only %d available ",
2872 total_attribs_size
, max_index
);
2877 /* If all of the attributes were assigned locations by the application (or
2878 * are built-in attributes with fixed locations), return early. This should
2879 * be the common case.
2884 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2886 if (target_index
== MESA_SHADER_VERTEX
) {
2887 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2888 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2889 * reserved to prevent it from being automatically allocated below.
2891 find_deref_visitor
find("gl_Vertex");
2893 if (find
.variable_found())
2894 used_locations
|= (1 << 0);
2897 for (unsigned i
= 0; i
< num_attr
; i
++) {
2898 /* Mask representing the contiguous slots that will be used by this
2901 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2903 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2906 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2907 ? "vertex shader input" : "fragment shader output";
2910 "insufficient contiguous locations "
2911 "available for %s `%s'\n",
2912 string
, to_assign
[i
].var
->name
);
2916 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2917 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2918 used_locations
|= (use_mask
<< location
);
2920 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2921 double_storage_locations
|= (use_mask
<< location
);
2924 /* Now that we have all the locations, from the GL 4.5 core spec, section
2925 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2926 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2927 * as equivalent single-precision types.
2929 if (target_index
== MESA_SHADER_VERTEX
) {
2930 unsigned total_attribs_size
=
2931 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2932 _mesa_bitcount(double_storage_locations
);
2933 if (total_attribs_size
> max_index
) {
2935 "attempt to use %d vertex attribute slots only %d available ",
2936 total_attribs_size
, max_index
);
2945 * Match explicit locations of outputs to inputs and deactivate the
2946 * unmatch flag if found so we don't optimise them away.
2949 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2950 gl_linked_shader
*consumer
)
2952 glsl_symbol_table parameters
;
2953 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2956 /* Find all shader outputs in the "producer" stage.
2958 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2959 ir_variable
*const var
= node
->as_variable();
2961 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2964 if (var
->data
.explicit_location
&&
2965 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2966 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2967 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2968 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2972 /* Match inputs to outputs */
2973 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2974 ir_variable
*const input
= node
->as_variable();
2976 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2979 ir_variable
*output
= NULL
;
2980 if (input
->data
.explicit_location
2981 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2982 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2983 [input
->data
.location_frac
];
2985 if (output
!= NULL
){
2986 input
->data
.is_unmatched_generic_inout
= 0;
2987 output
->data
.is_unmatched_generic_inout
= 0;
2994 * Store the gl_FragDepth layout in the gl_shader_program struct.
2997 store_fragdepth_layout(struct gl_shader_program
*prog
)
2999 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3003 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3005 /* We don't look up the gl_FragDepth symbol directly because if
3006 * gl_FragDepth is not used in the shader, it's removed from the IR.
3007 * However, the symbol won't be removed from the symbol table.
3009 * We're only interested in the cases where the variable is NOT removed
3012 foreach_in_list(ir_instruction
, node
, ir
) {
3013 ir_variable
*const var
= node
->as_variable();
3015 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3019 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3020 switch (var
->data
.depth_layout
) {
3021 case ir_depth_layout_none
:
3022 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3024 case ir_depth_layout_any
:
3025 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3027 case ir_depth_layout_greater
:
3028 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3030 case ir_depth_layout_less
:
3031 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3033 case ir_depth_layout_unchanged
:
3034 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3045 * Validate the resources used by a program versus the implementation limits
3048 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3050 unsigned total_uniform_blocks
= 0;
3051 unsigned total_shader_storage_blocks
= 0;
3053 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3054 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3059 if (sh
->Program
->info
.num_textures
>
3060 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3061 linker_error(prog
, "Too many %s shader texture samplers\n",
3062 _mesa_shader_stage_to_string(i
));
3065 if (sh
->num_uniform_components
>
3066 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3067 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3068 linker_warning(prog
, "Too many %s shader default uniform block "
3069 "components, but the driver will try to optimize "
3070 "them out; this is non-portable out-of-spec "
3072 _mesa_shader_stage_to_string(i
));
3074 linker_error(prog
, "Too many %s shader default uniform block "
3076 _mesa_shader_stage_to_string(i
));
3080 if (sh
->num_combined_uniform_components
>
3081 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3082 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3083 linker_warning(prog
, "Too many %s shader uniform components, "
3084 "but the driver will try to optimize them out; "
3085 "this is non-portable out-of-spec behavior\n",
3086 _mesa_shader_stage_to_string(i
));
3088 linker_error(prog
, "Too many %s shader uniform components\n",
3089 _mesa_shader_stage_to_string(i
));
3093 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3094 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3096 const unsigned max_uniform_blocks
=
3097 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3098 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3099 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3100 _mesa_shader_stage_to_string(i
),
3101 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3104 const unsigned max_shader_storage_blocks
=
3105 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3106 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3107 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3108 _mesa_shader_stage_to_string(i
),
3109 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3113 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3114 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3115 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3118 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3119 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3120 total_shader_storage_blocks
,
3121 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3124 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3125 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3126 ctx
->Const
.MaxUniformBlockSize
) {
3127 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3128 prog
->data
->UniformBlocks
[i
].Name
,
3129 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3130 ctx
->Const
.MaxUniformBlockSize
);
3134 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3135 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3136 ctx
->Const
.MaxShaderStorageBlockSize
) {
3137 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3138 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3139 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3140 ctx
->Const
.MaxShaderStorageBlockSize
);
3146 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3148 unsigned mask
= prog
->data
->linked_stages
;
3150 const int i
= u_bit_scan(&mask
);
3151 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3153 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3154 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3157 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3163 if (p
->sh
.NumSubroutineFunctions
== 0) {
3164 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3167 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3168 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3169 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3170 if (fn
->types
[k
] == uni
->type
) {
3176 uni
->num_compatible_subroutines
= count
;
3182 check_subroutine_resources(struct gl_shader_program
*prog
)
3184 unsigned mask
= prog
->data
->linked_stages
;
3186 const int i
= u_bit_scan(&mask
);
3187 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3189 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3190 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3191 _mesa_shader_stage_to_string(i
));
3196 * Validate shader image resources.
3199 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3201 unsigned total_image_units
= 0;
3202 unsigned fragment_outputs
= 0;
3203 unsigned total_shader_storage_blocks
= 0;
3205 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3208 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3209 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3212 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3213 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3214 _mesa_shader_stage_to_string(i
),
3215 sh
->Program
->info
.num_images
,
3216 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3218 total_image_units
+= sh
->Program
->info
.num_images
;
3219 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3221 if (i
== MESA_SHADER_FRAGMENT
) {
3222 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3223 ir_variable
*var
= node
->as_variable();
3224 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3225 /* since there are no double fs outputs - pass false */
3226 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3232 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3233 linker_error(prog
, "Too many combined image uniforms\n");
3235 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3236 ctx
->Const
.MaxCombinedShaderOutputResources
)
3237 linker_error(prog
, "Too many combined image uniforms, shader storage "
3238 " buffers and fragment outputs\n");
3243 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3244 * for a variable, checks for overlaps between other uniforms using explicit
3248 reserve_explicit_locations(struct gl_shader_program
*prog
,
3249 string_to_uint_map
*map
, ir_variable
*var
)
3251 unsigned slots
= var
->type
->uniform_locations();
3252 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3253 unsigned return_value
= slots
;
3255 /* Resize remap table if locations do not fit in the current one. */
3256 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3257 prog
->UniformRemapTable
=
3258 reralloc(prog
, prog
->UniformRemapTable
,
3259 gl_uniform_storage
*,
3262 if (!prog
->UniformRemapTable
) {
3263 linker_error(prog
, "Out of memory during linking.\n");
3267 /* Initialize allocated space. */
3268 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3269 prog
->UniformRemapTable
[i
] = NULL
;
3271 prog
->NumUniformRemapTable
= max_loc
+ 1;
3274 for (unsigned i
= 0; i
< slots
; i
++) {
3275 unsigned loc
= var
->data
.location
+ i
;
3277 /* Check if location is already used. */
3278 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3280 /* Possibly same uniform from a different stage, this is ok. */
3282 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3287 /* ARB_explicit_uniform_location specification states:
3289 * "No two default-block uniform variables in the program can have
3290 * the same location, even if they are unused, otherwise a compiler
3291 * or linker error will be generated."
3294 "location qualifier for uniform %s overlaps "
3295 "previously used location\n",
3300 /* Initialize location as inactive before optimization
3301 * rounds and location assignment.
3303 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3306 /* Note, base location used for arrays. */
3307 map
->put(var
->data
.location
, var
->name
);
3309 return return_value
;
3313 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3314 struct gl_program
*p
,
3317 unsigned slots
= var
->type
->uniform_locations();
3318 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3320 /* Resize remap table if locations do not fit in the current one. */
3321 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3322 p
->sh
.SubroutineUniformRemapTable
=
3323 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3324 gl_uniform_storage
*,
3327 if (!p
->sh
.SubroutineUniformRemapTable
) {
3328 linker_error(prog
, "Out of memory during linking.\n");
3332 /* Initialize allocated space. */
3333 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3334 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3336 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3339 for (unsigned i
= 0; i
< slots
; i
++) {
3340 unsigned loc
= var
->data
.location
+ i
;
3342 /* Check if location is already used. */
3343 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3345 /* ARB_explicit_uniform_location specification states:
3346 * "No two subroutine uniform variables can have the same location
3347 * in the same shader stage, otherwise a compiler or linker error
3348 * will be generated."
3351 "location qualifier for uniform %s overlaps "
3352 "previously used location\n",
3357 /* Initialize location as inactive before optimization
3358 * rounds and location assignment.
3360 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3366 * Check and reserve all explicit uniform locations, called before
3367 * any optimizations happen to handle also inactive uniforms and
3368 * inactive array elements that may get trimmed away.
3371 check_explicit_uniform_locations(struct gl_context
*ctx
,
3372 struct gl_shader_program
*prog
)
3374 prog
->NumExplicitUniformLocations
= 0;
3376 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3379 /* This map is used to detect if overlapping explicit locations
3380 * occur with the same uniform (from different stage) or a different one.
3382 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3385 linker_error(prog
, "Out of memory during linking.\n");
3389 unsigned entries_total
= 0;
3390 unsigned mask
= prog
->data
->linked_stages
;
3392 const int i
= u_bit_scan(&mask
);
3393 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3395 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3396 ir_variable
*var
= node
->as_variable();
3397 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3400 if (var
->data
.explicit_location
) {
3402 if (var
->type
->without_array()->is_subroutine())
3403 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3405 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3409 entries_total
+= slots
;
3420 struct empty_uniform_block
*current_block
= NULL
;
3422 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3423 /* We found empty space in UniformRemapTable. */
3424 if (prog
->UniformRemapTable
[i
] == NULL
) {
3425 /* We've found the beginning of a new continous block of empty slots */
3426 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3427 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3428 current_block
->start
= i
;
3429 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3430 ¤t_block
->link
);
3433 /* The current block continues, so we simply increment its slots */
3434 current_block
->slots
++;
3439 prog
->NumExplicitUniformLocations
= entries_total
;
3443 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3444 GLenum type
, const char *name
)
3446 bool found_interface
= false;
3447 unsigned block_name_len
= 0;
3448 const char *block_name_dot
= strchr(name
, '.');
3450 /* These rules only apply to buffer variables. So we return
3451 * true for the rest of types.
3453 if (type
!= GL_BUFFER_VARIABLE
)
3456 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3457 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3458 block_name_len
= strlen(block_name
);
3460 const char *block_square_bracket
= strchr(block_name
, '[');
3461 if (block_square_bracket
) {
3462 /* The block is part of an array of named interfaces,
3463 * for the name comparison we ignore the "[x]" part.
3465 block_name_len
-= strlen(block_square_bracket
);
3468 if (block_name_dot
) {
3469 /* Check if the variable name starts with the interface
3470 * name. The interface name (if present) should have the
3471 * length than the interface block name we are comparing to.
3473 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3474 if (len
!= block_name_len
)
3478 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3479 found_interface
= true;
3484 /* We remove the interface name from the buffer variable name,
3485 * including the dot that follows it.
3487 if (found_interface
)
3488 name
= name
+ block_name_len
+ 1;
3490 /* The ARB_program_interface_query spec says:
3492 * "For an active shader storage block member declared as an array, an
3493 * entry will be generated only for the first array element, regardless
3494 * of its type. For arrays of aggregate types, the enumeration rules
3495 * are applied recursively for the single enumerated array element."
3497 const char *struct_first_dot
= strchr(name
, '.');
3498 const char *first_square_bracket
= strchr(name
, '[');
3500 /* The buffer variable is on top level and it is not an array */
3501 if (!first_square_bracket
) {
3503 /* The shader storage block member is a struct, then generate the entry */
3504 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3507 /* Shader storage block member is an array, only generate an entry for the
3508 * first array element.
3510 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3518 add_program_resource(struct gl_shader_program
*prog
,
3519 struct set
*resource_set
,
3520 GLenum type
, const void *data
, uint8_t stages
)
3524 /* If resource already exists, do not add it again. */
3525 if (_mesa_set_search(resource_set
, data
))
3528 prog
->data
->ProgramResourceList
=
3530 prog
->data
->ProgramResourceList
,
3531 gl_program_resource
,
3532 prog
->data
->NumProgramResourceList
+ 1);
3534 if (!prog
->data
->ProgramResourceList
) {
3535 linker_error(prog
, "Out of memory during linking.\n");
3539 struct gl_program_resource
*res
=
3540 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3544 res
->StageReferences
= stages
;
3546 prog
->data
->NumProgramResourceList
++;
3548 _mesa_set_add(resource_set
, data
);
3553 /* Function checks if a variable var is a packed varying and
3554 * if given name is part of packed varying's list.
3556 * If a variable is a packed varying, it has a name like
3557 * 'packed:a,b,c' where a, b and c are separate variables.
3560 included_in_packed_varying(ir_variable
*var
, const char *name
)
3562 if (strncmp(var
->name
, "packed:", 7) != 0)
3565 char *list
= strdup(var
->name
+ 7);
3570 char *token
= strtok_r(list
, ",", &saveptr
);
3572 if (strcmp(token
, name
) == 0) {
3576 token
= strtok_r(NULL
, ",", &saveptr
);
3583 * Function builds a stage reference bitmask from variable name.
3586 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3591 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3592 * used for reference mask in gl_program_resource will need to be changed.
3594 assert(MESA_SHADER_STAGES
< 8);
3596 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3597 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3601 /* Shader symbol table may contain variables that have
3602 * been optimized away. Search IR for the variable instead.
3604 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3605 ir_variable
*var
= node
->as_variable();
3607 unsigned baselen
= strlen(var
->name
);
3609 if (included_in_packed_varying(var
, name
)) {
3614 /* Type needs to match if specified, otherwise we might
3615 * pick a variable with same name but different interface.
3617 if (var
->data
.mode
!= mode
)
3620 if (strncmp(var
->name
, name
, baselen
) == 0) {
3621 /* Check for exact name matches but also check for arrays and
3624 if (name
[baselen
] == '\0' ||
3625 name
[baselen
] == '[' ||
3626 name
[baselen
] == '.') {
3638 * Create gl_shader_variable from ir_variable class.
3640 static gl_shader_variable
*
3641 create_shader_variable(struct gl_shader_program
*shProg
,
3642 const ir_variable
*in
,
3643 const char *name
, const glsl_type
*type
,
3644 const glsl_type
*interface_type
,
3645 bool use_implicit_location
, int location
,
3646 const glsl_type
*outermost_struct_type
)
3648 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3652 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3653 * expect to see gl_VertexID in the program resource list. Pretend.
3655 if (in
->data
.mode
== ir_var_system_value
&&
3656 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3657 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3658 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3659 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3660 (in
->data
.mode
== ir_var_system_value
&&
3661 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3662 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3663 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3664 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3665 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3666 (in
->data
.mode
== ir_var_system_value
&&
3667 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3668 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3669 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3671 out
->name
= ralloc_strdup(shProg
, name
);
3677 /* The ARB_program_interface_query spec says:
3679 * "Not all active variables are assigned valid locations; the
3680 * following variables will have an effective location of -1:
3682 * * uniforms declared as atomic counters;
3684 * * members of a uniform block;
3686 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3688 * * inputs or outputs not declared with a "location" layout
3689 * qualifier, except for vertex shader inputs and fragment shader
3692 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3693 is_gl_identifier(in
->name
) ||
3694 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3697 out
->location
= location
;
3701 out
->outermost_struct_type
= outermost_struct_type
;
3702 out
->interface_type
= interface_type
;
3703 out
->component
= in
->data
.location_frac
;
3704 out
->index
= in
->data
.index
;
3705 out
->patch
= in
->data
.patch
;
3706 out
->mode
= in
->data
.mode
;
3707 out
->interpolation
= in
->data
.interpolation
;
3708 out
->explicit_location
= in
->data
.explicit_location
;
3709 out
->precision
= in
->data
.precision
;
3714 static const glsl_type
*
3715 resize_to_max_patch_vertices(const struct gl_context
*ctx
,
3716 const glsl_type
*type
)
3721 return glsl_type::get_array_instance(type
->fields
.array
,
3722 ctx
->Const
.MaxPatchVertices
);
3726 add_shader_variable(const struct gl_context
*ctx
,
3727 struct gl_shader_program
*shProg
,
3728 struct set
*resource_set
,
3729 unsigned stage_mask
,
3730 GLenum programInterface
, ir_variable
*var
,
3731 const char *name
, const glsl_type
*type
,
3732 bool use_implicit_location
, int location
,
3733 const glsl_type
*outermost_struct_type
= NULL
)
3735 const glsl_type
*interface_type
= var
->get_interface_type();
3737 if (outermost_struct_type
== NULL
) {
3738 /* Unsized (non-patch) TCS output/TES input arrays are implicitly
3739 * sized to gl_MaxPatchVertices. Internally, we shrink them to a
3742 * This can cause trouble with SSO programs. Since the TCS declares
3743 * the number of output vertices, we can always shrink TCS output
3744 * arrays. However, the TES might not be linked with a TCS, in
3745 * which case it won't know the size of the patch. In other words,
3746 * the TCS and TES may disagree on the (smaller) array sizes. This
3747 * can result in the resource names differing across stages, causing
3748 * SSO validation failures and other cascading issues.
3750 * Expanding the array size to the full gl_MaxPatchVertices fixes
3751 * these issues. It's also what program interface queries expect,
3752 * as that is the official size of the array.
3754 if (var
->data
.tess_varying_implicit_sized_array
) {
3755 type
= resize_to_max_patch_vertices(ctx
, type
);
3756 interface_type
= resize_to_max_patch_vertices(ctx
, interface_type
);
3759 if (var
->data
.from_named_ifc_block
) {
3760 const char *interface_name
= interface_type
->name
;
3762 if (interface_type
->is_array()) {
3763 /* Issue #16 of the ARB_program_interface_query spec says:
3765 * "* If a variable is a member of an interface block without an
3766 * instance name, it is enumerated using just the variable name.
3768 * * If a variable is a member of an interface block with an
3769 * instance name, it is enumerated as "BlockName.Member", where
3770 * "BlockName" is the name of the interface block (not the
3771 * instance name) and "Member" is the name of the variable."
3773 * In particular, it indicates that it should be "BlockName",
3774 * not "BlockName[array length]". The conformance suite and
3775 * dEQP both require this behavior.
3777 * Here, we unwrap the extra array level added by named interface
3778 * block array lowering so we have the correct variable type. We
3779 * also unwrap the interface type when constructing the name.
3781 * We leave interface_type the same so that ES 3.x SSO pipeline
3782 * validation can enforce the rules requiring array length to
3783 * match on interface blocks.
3785 type
= type
->fields
.array
;
3787 interface_name
= interface_type
->fields
.array
->name
;
3790 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3794 switch (type
->base_type
) {
3795 case GLSL_TYPE_STRUCT
: {
3796 /* The ARB_program_interface_query spec says:
3798 * "For an active variable declared as a structure, a separate entry
3799 * will be generated for each active structure member. The name of
3800 * each entry is formed by concatenating the name of the structure,
3801 * the "." character, and the name of the structure member. If a
3802 * structure member to enumerate is itself a structure or array,
3803 * these enumeration rules are applied recursively."
3805 if (outermost_struct_type
== NULL
)
3806 outermost_struct_type
= type
;
3808 unsigned field_location
= location
;
3809 for (unsigned i
= 0; i
< type
->length
; i
++) {
3810 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3811 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3812 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3813 stage_mask
, programInterface
,
3814 var
, field_name
, field
->type
,
3815 use_implicit_location
, field_location
,
3816 outermost_struct_type
))
3819 field_location
+= field
->type
->count_attribute_slots(false);
3825 /* The ARB_program_interface_query spec says:
3827 * "For an active variable declared as a single instance of a basic
3828 * type, a single entry will be generated, using the variable name
3829 * from the shader source."
3831 gl_shader_variable
*sha_v
=
3832 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3833 use_implicit_location
, location
,
3834 outermost_struct_type
);
3838 return add_program_resource(shProg
, resource_set
,
3839 programInterface
, sha_v
, stage_mask
);
3845 add_interface_variables(const struct gl_context
*ctx
,
3846 struct gl_shader_program
*shProg
,
3847 struct set
*resource_set
,
3848 unsigned stage
, GLenum programInterface
)
3850 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3852 foreach_in_list(ir_instruction
, node
, ir
) {
3853 ir_variable
*var
= node
->as_variable();
3855 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3860 switch (var
->data
.mode
) {
3861 case ir_var_system_value
:
3862 case ir_var_shader_in
:
3863 if (programInterface
!= GL_PROGRAM_INPUT
)
3865 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3866 : int(VARYING_SLOT_VAR0
);
3868 case ir_var_shader_out
:
3869 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3871 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3872 : int(VARYING_SLOT_VAR0
);
3878 if (var
->data
.patch
)
3879 loc_bias
= int(VARYING_SLOT_PATCH0
);
3881 /* Skip packed varyings, packed varyings are handled separately
3882 * by add_packed_varyings.
3884 if (strncmp(var
->name
, "packed:", 7) == 0)
3887 /* Skip fragdata arrays, these are handled separately
3888 * by add_fragdata_arrays.
3890 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3893 const bool vs_input_or_fs_output
=
3894 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3895 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3897 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3898 1 << stage
, programInterface
,
3899 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3900 var
->data
.location
- loc_bias
))
3907 add_packed_varyings(const struct gl_context
*ctx
,
3908 struct gl_shader_program
*shProg
,
3909 struct set
*resource_set
,
3910 int stage
, GLenum type
)
3912 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3915 if (!sh
|| !sh
->packed_varyings
)
3918 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3919 ir_variable
*var
= node
->as_variable();
3921 switch (var
->data
.mode
) {
3922 case ir_var_shader_in
:
3923 iface
= GL_PROGRAM_INPUT
;
3925 case ir_var_shader_out
:
3926 iface
= GL_PROGRAM_OUTPUT
;
3929 unreachable("unexpected type");
3932 if (type
== iface
) {
3933 const int stage_mask
=
3934 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3935 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3937 iface
, var
, var
->name
, var
->type
, false,
3938 var
->data
.location
- VARYING_SLOT_VAR0
))
3947 add_fragdata_arrays(const struct gl_context
*ctx
,
3948 struct gl_shader_program
*shProg
,
3949 struct set
*resource_set
)
3951 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3953 if (!sh
|| !sh
->fragdata_arrays
)
3956 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3957 ir_variable
*var
= node
->as_variable();
3959 assert(var
->data
.mode
== ir_var_shader_out
);
3961 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3962 1 << MESA_SHADER_FRAGMENT
,
3963 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3964 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3972 get_top_level_name(const char *name
)
3974 const char *first_dot
= strchr(name
, '.');
3975 const char *first_square_bracket
= strchr(name
, '[');
3978 /* The ARB_program_interface_query spec says:
3980 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3981 * the number of active array elements of the top-level shader storage
3982 * block member containing to the active variable is written to
3983 * <params>. If the top-level block member is not declared as an
3984 * array, the value one is written to <params>. If the top-level block
3985 * member is an array with no declared size, the value zero is written
3989 /* The buffer variable is on top level.*/
3990 if (!first_square_bracket
&& !first_dot
)
3991 name_size
= strlen(name
);
3992 else if ((!first_square_bracket
||
3993 (first_dot
&& first_dot
< first_square_bracket
)))
3994 name_size
= first_dot
- name
;
3996 name_size
= first_square_bracket
- name
;
3998 return strndup(name
, name_size
);
4002 get_var_name(const char *name
)
4004 const char *first_dot
= strchr(name
, '.');
4007 return strdup(name
);
4009 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4013 is_top_level_shader_storage_block_member(const char* name
,
4014 const char* interface_name
,
4015 const char* field_name
)
4017 bool result
= false;
4019 /* If the given variable is already a top-level shader storage
4020 * block member, then return array_size = 1.
4021 * We could have two possibilities: if we have an instanced
4022 * shader storage block or not instanced.
4024 * For the first, we check create a name as it was in top level and
4025 * compare it with the real name. If they are the same, then
4026 * the variable is already at top-level.
4028 * Full instanced name is: interface name + '.' + var name +
4031 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4032 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4033 if (!full_instanced_name
) {
4034 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4038 snprintf(full_instanced_name
, name_length
, "%s.%s",
4039 interface_name
, field_name
);
4041 /* Check if its top-level shader storage block member of an
4042 * instanced interface block, or of a unnamed interface block.
4044 if (strcmp(name
, full_instanced_name
) == 0 ||
4045 strcmp(name
, field_name
) == 0)
4048 free(full_instanced_name
);
4053 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4054 char *interface_name
, char *var_name
)
4056 /* The ARB_program_interface_query spec says:
4058 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4059 * the number of active array elements of the top-level shader storage
4060 * block member containing to the active variable is written to
4061 * <params>. If the top-level block member is not declared as an
4062 * array, the value one is written to <params>. If the top-level block
4063 * member is an array with no declared size, the value zero is written
4066 if (is_top_level_shader_storage_block_member(uni
->name
,
4070 else if (field
->type
->is_unsized_array())
4072 else if (field
->type
->is_array())
4073 return field
->type
->length
;
4079 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4080 const glsl_struct_field
*field
, char *interface_name
,
4083 /* The ARB_program_interface_query spec says:
4085 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4086 * identifying the stride between array elements of the top-level
4087 * shader storage block member containing the active variable is
4088 * written to <params>. For top-level block members declared as
4089 * arrays, the value written is the difference, in basic machine units,
4090 * between the offsets of the active variable for consecutive elements
4091 * in the top-level array. For top-level block members not declared as
4092 * an array, zero is written to <params>."
4094 if (field
->type
->is_array()) {
4095 const enum glsl_matrix_layout matrix_layout
=
4096 glsl_matrix_layout(field
->matrix_layout
);
4097 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4098 const glsl_type
*array_type
= field
->type
->fields
.array
;
4100 if (is_top_level_shader_storage_block_member(uni
->name
,
4105 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4106 if (array_type
->is_record() || array_type
->is_array())
4107 return glsl_align(array_type
->std140_size(row_major
), 16);
4109 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4111 return array_type
->std430_array_stride(row_major
);
4118 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4119 struct gl_uniform_storage
*uni
)
4121 int block_index
= uni
->block_index
;
4122 int array_size
= -1;
4123 int array_stride
= -1;
4124 char *var_name
= get_top_level_name(uni
->name
);
4125 char *interface_name
=
4126 get_top_level_name(uni
->is_shader_storage
?
4127 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4128 shProg
->data
->UniformBlocks
[block_index
].Name
);
4130 if (strcmp(var_name
, interface_name
) == 0) {
4131 /* Deal with instanced array of SSBOs */
4132 char *temp_name
= get_var_name(uni
->name
);
4134 linker_error(shProg
, "Out of memory during linking.\n");
4135 goto write_top_level_array_size_and_stride
;
4138 var_name
= get_top_level_name(temp_name
);
4141 linker_error(shProg
, "Out of memory during linking.\n");
4142 goto write_top_level_array_size_and_stride
;
4146 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4147 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4151 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4152 ir_variable
*var
= node
->as_variable();
4153 if (!var
|| !var
->get_interface_type() ||
4154 var
->data
.mode
!= ir_var_shader_storage
)
4157 const glsl_type
*interface
= var
->get_interface_type();
4159 if (strcmp(interface_name
, interface
->name
) != 0)
4162 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4163 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4164 if (strcmp(field
->name
, var_name
) != 0)
4167 array_stride
= get_array_stride(uni
, interface
, field
,
4168 interface_name
, var_name
);
4169 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4170 goto write_top_level_array_size_and_stride
;
4174 write_top_level_array_size_and_stride
:
4175 free(interface_name
);
4177 uni
->top_level_array_stride
= array_stride
;
4178 uni
->top_level_array_size
= array_size
;
4182 * Builds up a list of program resources that point to existing
4186 build_program_resource_list(struct gl_context
*ctx
,
4187 struct gl_shader_program
*shProg
)
4189 /* Rebuild resource list. */
4190 if (shProg
->data
->ProgramResourceList
) {
4191 ralloc_free(shProg
->data
->ProgramResourceList
);
4192 shProg
->data
->ProgramResourceList
= NULL
;
4193 shProg
->data
->NumProgramResourceList
= 0;
4196 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4198 /* Determine first input and final output stage. These are used to
4199 * detect which variables should be enumerated in the resource list
4200 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4202 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4203 if (!shProg
->_LinkedShaders
[i
])
4205 if (input_stage
== MESA_SHADER_STAGES
)
4210 /* Empty shader, no resources. */
4211 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4214 struct set
*resource_set
= _mesa_set_create(NULL
,
4216 _mesa_key_pointer_equal
);
4218 /* Program interface needs to expose varyings in case of SSO. */
4219 if (shProg
->SeparateShader
) {
4220 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4221 input_stage
, GL_PROGRAM_INPUT
))
4224 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4225 output_stage
, GL_PROGRAM_OUTPUT
))
4229 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4232 /* Add inputs and outputs to the resource list. */
4233 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4234 input_stage
, GL_PROGRAM_INPUT
))
4237 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4238 output_stage
, GL_PROGRAM_OUTPUT
))
4241 struct gl_transform_feedback_info
*linked_xfb
=
4242 shProg
->xfb_program
->sh
.LinkedTransformFeedback
;
4244 /* Add transform feedback varyings. */
4245 if (linked_xfb
->NumVarying
> 0) {
4246 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4247 if (!add_program_resource(shProg
, resource_set
,
4248 GL_TRANSFORM_FEEDBACK_VARYING
,
4249 &linked_xfb
->Varyings
[i
], 0))
4254 /* Add transform feedback buffers. */
4255 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4256 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4257 linked_xfb
->Buffers
[i
].Binding
= i
;
4258 if (!add_program_resource(shProg
, resource_set
,
4259 GL_TRANSFORM_FEEDBACK_BUFFER
,
4260 &linked_xfb
->Buffers
[i
], 0))
4265 /* Add uniforms from uniform storage. */
4266 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4267 /* Do not add uniforms internally used by Mesa. */
4268 if (shProg
->data
->UniformStorage
[i
].hidden
)
4272 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4275 /* Add stagereferences for uniforms in a uniform block. */
4276 bool is_shader_storage
=
4277 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4278 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4279 if (block_index
!= -1) {
4280 stageref
|= is_shader_storage
?
4281 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4282 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4285 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4286 if (!should_add_buffer_variable(shProg
, type
,
4287 shProg
->data
->UniformStorage
[i
].name
))
4290 if (is_shader_storage
) {
4291 calculate_array_size_and_stride(shProg
,
4292 &shProg
->data
->UniformStorage
[i
]);
4295 if (!add_program_resource(shProg
, resource_set
, type
,
4296 &shProg
->data
->UniformStorage
[i
], stageref
))
4300 /* Add program uniform blocks. */
4301 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4302 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4303 &shProg
->data
->UniformBlocks
[i
], 0))
4307 /* Add program shader storage blocks. */
4308 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4309 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4310 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4314 /* Add atomic counter buffers. */
4315 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4316 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4317 &shProg
->data
->AtomicBuffers
[i
], 0))
4321 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4323 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4326 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4327 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4328 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4331 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4332 /* add shader subroutines */
4333 if (!add_program_resource(shProg
, resource_set
,
4334 type
, &shProg
->data
->UniformStorage
[i
], 0))
4339 unsigned mask
= shProg
->data
->linked_stages
;
4341 const int i
= u_bit_scan(&mask
);
4342 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4344 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4345 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4346 if (!add_program_resource(shProg
, resource_set
,
4347 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4352 _mesa_set_destroy(resource_set
, NULL
);
4356 * This check is done to make sure we allow only constant expression
4357 * indexing and "constant-index-expression" (indexing with an expression
4358 * that includes loop induction variable).
4361 validate_sampler_array_indexing(struct gl_context
*ctx
,
4362 struct gl_shader_program
*prog
)
4364 dynamic_sampler_array_indexing_visitor v
;
4365 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4366 if (prog
->_LinkedShaders
[i
] == NULL
)
4369 bool no_dynamic_indexing
=
4370 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4372 /* Search for array derefs in shader. */
4373 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4374 if (v
.uses_dynamic_sampler_array_indexing()) {
4375 const char *msg
= "sampler arrays indexed with non-constant "
4376 "expressions is forbidden in GLSL %s %u";
4377 /* Backend has indicated that it has no dynamic indexing support. */
4378 if (no_dynamic_indexing
) {
4379 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4380 prog
->data
->Version
);
4383 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4384 prog
->data
->Version
);
4392 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4394 unsigned mask
= prog
->data
->linked_stages
;
4396 const int i
= u_bit_scan(&mask
);
4397 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4399 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4400 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4401 ir_function
*fn
= node
->as_function();
4405 if (fn
->is_subroutine
)
4406 p
->sh
.NumSubroutineUniformTypes
++;
4408 if (!fn
->num_subroutine_types
)
4411 /* these should have been calculated earlier. */
4412 assert(fn
->subroutine_index
!= -1);
4413 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4414 linker_error(prog
, "Too many subroutine functions declared.\n");
4417 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4418 struct gl_subroutine_function
,
4419 p
->sh
.NumSubroutineFunctions
+ 1);
4420 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4421 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4422 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4423 ralloc_array(p
, const struct glsl_type
*,
4424 fn
->num_subroutine_types
);
4426 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4429 * "Each subroutine with an index qualifier in the shader must be
4430 * given a unique index, otherwise a compile or link error will be
4433 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4434 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4435 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4436 linker_error(prog
, "each subroutine index qualifier in the "
4437 "shader must be unique\n");
4441 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4442 fn
->subroutine_index
;
4444 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4445 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4447 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4448 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4449 p
->sh
.NumSubroutineFunctions
++;
4455 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4457 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4459 foreach_in_list(ir_instruction
, node
, ir
) {
4460 ir_variable
*const var
= node
->as_variable();
4462 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4465 /* Don't set always active on builtins that haven't been redeclared */
4466 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4469 var
->data
.always_active_io
= true;
4474 * When separate shader programs are enabled, only input/outputs between
4475 * the stages of a multi-stage separate program can be safely removed
4476 * from the shader interface. Other inputs/outputs must remain active.
4479 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4481 unsigned first
, last
;
4482 assert(prog
->SeparateShader
);
4484 first
= MESA_SHADER_STAGES
;
4487 /* Determine first and last stage. Excluding the compute stage */
4488 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4489 if (!prog
->_LinkedShaders
[i
])
4491 if (first
== MESA_SHADER_STAGES
)
4496 if (first
== MESA_SHADER_STAGES
)
4499 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4500 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4504 if (first
== last
) {
4505 /* For a single shader program only allow inputs to the vertex shader
4506 * and outputs from the fragment shader to be removed.
4508 if (stage
!= MESA_SHADER_VERTEX
)
4509 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4510 if (stage
!= MESA_SHADER_FRAGMENT
)
4511 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4513 /* For multi-stage separate shader programs only allow inputs and
4514 * outputs between the shader stages to be removed as well as inputs
4515 * to the vertex shader and outputs from the fragment shader.
4517 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4518 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4519 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4520 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4526 link_and_validate_uniforms(struct gl_context
*ctx
,
4527 struct gl_shader_program
*prog
)
4529 update_array_sizes(prog
);
4530 link_assign_uniform_locations(prog
, ctx
);
4532 link_assign_atomic_counter_resources(ctx
, prog
);
4533 link_calculate_subroutine_compat(prog
);
4534 check_resources(ctx
, prog
);
4535 check_subroutine_resources(prog
);
4536 check_image_resources(ctx
, prog
);
4537 link_check_atomic_counter_resources(ctx
, prog
);
4541 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4542 struct gl_context
*ctx
,
4543 struct gl_shader_program
*prog
, void *mem_ctx
)
4545 /* Mark all generic shader inputs and outputs as unpaired. */
4546 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4547 if (prog
->_LinkedShaders
[i
] != NULL
) {
4548 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4552 unsigned prev
= first
;
4553 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4554 if (prog
->_LinkedShaders
[i
] == NULL
)
4557 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4558 prog
->_LinkedShaders
[i
]);
4562 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4563 MESA_SHADER_VERTEX
)) {
4567 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4568 MESA_SHADER_FRAGMENT
)) {
4572 /* Find the program used for xfb. Even if we don't use xfb we still want to
4573 * set this so we can fill the default values for program interface query.
4575 prog
->xfb_program
= prog
->_LinkedShaders
[last
]->Program
;
4576 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4577 if (prog
->_LinkedShaders
[i
] == NULL
)
4580 prog
->xfb_program
= prog
->_LinkedShaders
[i
]->Program
;
4584 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4587 link_and_validate_uniforms(ctx
, prog
);
4589 if (!prog
->data
->LinkStatus
)
4592 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4593 if (prog
->_LinkedShaders
[i
] == NULL
)
4596 const struct gl_shader_compiler_options
*options
=
4597 &ctx
->Const
.ShaderCompilerOptions
[i
];
4599 if (options
->LowerBufferInterfaceBlocks
)
4600 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4601 options
->ClampBlockIndicesToArrayBounds
);
4603 if (i
== MESA_SHADER_COMPUTE
)
4604 lower_shared_reference(prog
->_LinkedShaders
[i
],
4605 &prog
->Comp
.SharedSize
);
4607 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4608 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4615 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4617 prog
->data
->LinkStatus
= true; /* All error paths will set this to false */
4618 prog
->data
->Validated
= false;
4620 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4622 * "Linking can fail for a variety of reasons as specified in the
4623 * OpenGL Shading Language Specification, as well as any of the
4624 * following reasons:
4626 * - No shader objects are attached to program."
4628 * The Compatibility Profile specification does not list the error. In
4629 * Compatibility Profile missing shader stages are replaced by
4630 * fixed-function. This applies to the case where all stages are
4633 if (prog
->NumShaders
== 0) {
4634 if (ctx
->API
!= API_OPENGL_COMPAT
)
4635 linker_error(prog
, "no shaders attached to the program\n");
4639 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4641 prog
->ARB_fragment_coord_conventions_enable
= false;
4643 /* Separate the shaders into groups based on their type.
4645 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4646 unsigned num_shaders
[MESA_SHADER_STAGES
];
4648 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4649 shader_list
[i
] = (struct gl_shader
**)
4650 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4654 unsigned min_version
= UINT_MAX
;
4655 unsigned max_version
= 0;
4656 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4657 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4658 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4660 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4661 linker_error(prog
, "all shaders must use same shading "
4662 "language version\n");
4666 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4667 prog
->ARB_fragment_coord_conventions_enable
= true;
4670 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4671 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4672 num_shaders
[shader_type
]++;
4675 /* In desktop GLSL, different shader versions may be linked together. In
4676 * GLSL ES, all shader versions must be the same.
4678 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4679 linker_error(prog
, "all shaders must use same shading "
4680 "language version\n");
4684 prog
->data
->Version
= max_version
;
4685 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4687 /* Some shaders have to be linked with some other shaders present.
4689 if (!prog
->SeparateShader
) {
4690 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4691 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4692 linker_error(prog
, "Geometry shader must be linked with "
4696 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4697 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4698 linker_error(prog
, "Tessellation evaluation shader must be linked "
4699 "with vertex shader\n");
4702 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4703 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4704 linker_error(prog
, "Tessellation control shader must be linked with "
4709 /* The spec is self-contradictory here. It allows linking without a tess
4710 * eval shader, but that can only be used with transform feedback and
4711 * rasterization disabled. However, transform feedback isn't allowed
4712 * with GL_PATCHES, so it can't be used.
4714 * More investigation showed that the idea of transform feedback after
4715 * a tess control shader was dropped, because some hw vendors couldn't
4716 * support tessellation without a tess eval shader, but the linker
4717 * section wasn't updated to reflect that.
4719 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4722 * Do what's reasonable and always require a tess eval shader if a tess
4723 * control shader is present.
4725 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4726 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4727 linker_error(prog
, "Tessellation control shader must be linked with "
4728 "tessellation evaluation shader\n");
4733 /* Compute shaders have additional restrictions. */
4734 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4735 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4736 linker_error(prog
, "Compute shaders may not be linked with any other "
4737 "type of shader\n");
4740 /* Link all shaders for a particular stage and validate the result.
4742 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4743 if (num_shaders
[stage
] > 0) {
4744 gl_linked_shader
*const sh
=
4745 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4746 num_shaders
[stage
], false);
4748 if (!prog
->data
->LinkStatus
) {
4750 _mesa_delete_linked_shader(ctx
, sh
);
4755 case MESA_SHADER_VERTEX
:
4756 validate_vertex_shader_executable(prog
, sh
, ctx
);
4758 case MESA_SHADER_TESS_CTRL
:
4759 /* nothing to be done */
4761 case MESA_SHADER_TESS_EVAL
:
4762 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4764 case MESA_SHADER_GEOMETRY
:
4765 validate_geometry_shader_executable(prog
, sh
, ctx
);
4767 case MESA_SHADER_FRAGMENT
:
4768 validate_fragment_shader_executable(prog
, sh
);
4771 if (!prog
->data
->LinkStatus
) {
4773 _mesa_delete_linked_shader(ctx
, sh
);
4777 prog
->_LinkedShaders
[stage
] = sh
;
4778 prog
->data
->linked_stages
|= 1 << stage
;
4782 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4783 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4784 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4785 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4786 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4787 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4788 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4789 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4790 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4792 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4793 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4796 /* Here begins the inter-stage linking phase. Some initial validation is
4797 * performed, then locations are assigned for uniforms, attributes, and
4800 cross_validate_uniforms(prog
);
4801 if (!prog
->data
->LinkStatus
)
4804 unsigned first
, last
, prev
;
4806 first
= MESA_SHADER_STAGES
;
4809 /* Determine first and last stage. */
4810 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4811 if (!prog
->_LinkedShaders
[i
])
4813 if (first
== MESA_SHADER_STAGES
)
4818 check_explicit_uniform_locations(ctx
, prog
);
4819 link_assign_subroutine_types(prog
);
4821 if (!prog
->data
->LinkStatus
)
4824 resize_tes_inputs(ctx
, prog
);
4826 /* Validate the inputs of each stage with the output of the preceding
4830 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4831 if (prog
->_LinkedShaders
[i
] == NULL
)
4834 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4835 prog
->_LinkedShaders
[i
]);
4836 if (!prog
->data
->LinkStatus
)
4839 cross_validate_outputs_to_inputs(prog
,
4840 prog
->_LinkedShaders
[prev
],
4841 prog
->_LinkedShaders
[i
]);
4842 if (!prog
->data
->LinkStatus
)
4848 /* Cross-validate uniform blocks between shader stages */
4849 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4850 if (!prog
->data
->LinkStatus
)
4853 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4854 if (prog
->_LinkedShaders
[i
] != NULL
)
4855 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4858 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4859 * it before optimization because we want most of the checks to get
4860 * dropped thanks to constant propagation.
4862 * This rule also applies to GLSL ES 3.00.
4864 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4865 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4867 lower_discard_flow(sh
->ir
);
4871 if (prog
->SeparateShader
)
4872 disable_varying_optimizations_for_sso(prog
);
4875 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4879 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4882 /* Do common optimization before assigning storage for attributes,
4883 * uniforms, and varyings. Later optimization could possibly make
4884 * some of that unused.
4886 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4887 if (prog
->_LinkedShaders
[i
] == NULL
)
4890 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4891 if (!prog
->data
->LinkStatus
)
4894 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4895 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4898 if (ctx
->Const
.LowerTessLevel
) {
4899 lower_tess_level(prog
->_LinkedShaders
[i
]);
4902 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4903 /* Run it just once. */
4904 do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4905 &ctx
->Const
.ShaderCompilerOptions
[i
],
4906 ctx
->Const
.NativeIntegers
);
4908 /* Repeat it until it stops making changes. */
4909 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4910 &ctx
->Const
.ShaderCompilerOptions
[i
],
4911 ctx
->Const
.NativeIntegers
))
4915 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
4916 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4919 /* Validation for special cases where we allow sampler array indexing
4920 * with loop induction variable. This check emits a warning or error
4921 * depending if backend can handle dynamic indexing.
4923 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
4924 (prog
->IsES
&& prog
->data
->Version
< 300)) {
4925 if (!validate_sampler_array_indexing(ctx
, prog
))
4929 /* Check and validate stream emissions in geometry shaders */
4930 validate_geometry_shader_emissions(ctx
, prog
);
4932 store_fragdepth_layout(prog
);
4934 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
4937 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4938 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4939 * anything about shader linking when one of the shaders (vertex or
4940 * fragment shader) is absent. So, the extension shouldn't change the
4941 * behavior specified in GLSL specification.
4943 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4944 * "Linking can fail for a variety of reasons as specified in the
4945 * OpenGL ES Shading Language Specification, as well as any of the
4946 * following reasons:
4950 * * program contains objects to form either a vertex shader or
4951 * fragment shader, and program is not separable, and does not
4952 * contain objects to form both a vertex shader and fragment
4955 * However, the only scenario in 3.1+ where we don't require them both is
4956 * when we have a compute shader. For example:
4958 * - No shaders is a link error.
4959 * - Geom or Tess without a Vertex shader is a link error which means we
4960 * always require a Vertex shader and hence a Fragment shader.
4961 * - Finally a Compute shader linked with any other stage is a link error.
4963 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4964 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4965 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4966 linker_error(prog
, "program lacks a vertex shader\n");
4967 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4968 linker_error(prog
, "program lacks a fragment shader\n");
4973 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4974 free(shader_list
[i
]);
4975 if (prog
->_LinkedShaders
[i
] == NULL
)
4978 /* Do a final validation step to make sure that the IR wasn't
4979 * invalidated by any modifications performed after intrastage linking.
4981 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4983 /* Retain any live IR, but trash the rest. */
4984 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4986 /* The symbol table in the linked shaders may contain references to
4987 * variables that were removed (e.g., unused uniforms). Since it may
4988 * contain junk, there is no possible valid use. Delete it and set the
4991 delete prog
->_LinkedShaders
[i
]->symbols
;
4992 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4995 ralloc_free(mem_ctx
);