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_linked_shader
*linked_shader
,
1663 struct gl_shader
**shader_list
,
1664 unsigned num_shaders
)
1666 linked_shader
->info
.TessCtrl
.VerticesOut
= 0;
1668 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
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 (linked_shader
->info
.TessCtrl
.VerticesOut
!= 0 &&
1686 linked_shader
->info
.TessCtrl
.VerticesOut
!=
1687 shader
->info
.TessCtrl
.VerticesOut
) {
1688 linker_error(prog
, "tessellation control shader defined with "
1689 "conflicting output vertex count (%d and %d)\n",
1690 linked_shader
->info
.TessCtrl
.VerticesOut
,
1691 shader
->info
.TessCtrl
.VerticesOut
);
1694 linked_shader
->info
.TessCtrl
.VerticesOut
=
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 (linked_shader
->info
.TessCtrl
.VerticesOut
== 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
= 0;
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
== 0)
1808 linked_shader
->info
.TessEval
.Spacing
= GL_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 linked_shader
->info
.redeclares_gl_fragcoord
= false;
1830 linked_shader
->info
.uses_gl_fragcoord
= false;
1831 linked_shader
->info
.origin_upper_left
= false;
1832 linked_shader
->info
.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 ((linked_shader
->info
.redeclares_gl_fragcoord
1848 && !shader
->info
.redeclares_gl_fragcoord
1849 && shader
->info
.uses_gl_fragcoord
)
1850 || (shader
->info
.redeclares_gl_fragcoord
1851 && !linked_shader
->info
.redeclares_gl_fragcoord
1852 && linked_shader
->info
.uses_gl_fragcoord
)) {
1853 linker_error(prog
, "fragment shader defined with conflicting "
1854 "layout qualifiers for gl_FragCoord\n");
1857 /* From the GLSL 1.50 spec, page 39:
1859 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1860 * single program must have the same set of qualifiers."
1862 if (linked_shader
->info
.redeclares_gl_fragcoord
&&
1863 shader
->info
.redeclares_gl_fragcoord
&&
1864 (shader
->info
.origin_upper_left
!=
1865 linked_shader
->info
.origin_upper_left
||
1866 shader
->info
.pixel_center_integer
!=
1867 linked_shader
->info
.pixel_center_integer
)) {
1868 linker_error(prog
, "fragment shader defined with conflicting "
1869 "layout qualifiers for gl_FragCoord\n");
1872 /* Update the linked shader state. Note that uses_gl_fragcoord should
1873 * accumulate the results. The other values should replace. If there
1874 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1875 * are already known to be the same.
1877 if (shader
->info
.redeclares_gl_fragcoord
||
1878 shader
->info
.uses_gl_fragcoord
) {
1879 linked_shader
->info
.redeclares_gl_fragcoord
=
1880 shader
->info
.redeclares_gl_fragcoord
;
1881 linked_shader
->info
.uses_gl_fragcoord
=
1882 linked_shader
->info
.uses_gl_fragcoord
||
1883 shader
->info
.uses_gl_fragcoord
;
1884 linked_shader
->info
.origin_upper_left
=
1885 shader
->info
.origin_upper_left
;
1886 linked_shader
->info
.pixel_center_integer
=
1887 shader
->info
.pixel_center_integer
;
1890 linked_shader
->info
.EarlyFragmentTests
|=
1891 shader
->info
.EarlyFragmentTests
;
1892 linked_shader
->info
.InnerCoverage
|=
1893 shader
->info
.InnerCoverage
;
1894 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1895 shader
->info
.PostDepthCoverage
;
1897 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1902 * Performs the cross-validation of geometry shader max_vertices and
1903 * primitive type layout qualifiers for the attached geometry shaders,
1904 * and propagates them to the linked GS and linked shader program.
1907 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1908 struct gl_linked_shader
*linked_shader
,
1909 struct gl_shader
**shader_list
,
1910 unsigned num_shaders
)
1912 linked_shader
->info
.Geom
.VerticesOut
= -1;
1913 linked_shader
->info
.Geom
.Invocations
= 0;
1914 linked_shader
->info
.Geom
.InputType
= PRIM_UNKNOWN
;
1915 linked_shader
->info
.Geom
.OutputType
= PRIM_UNKNOWN
;
1917 /* No in/out qualifiers defined for anything but GLSL 1.50+
1918 * geometry shaders so far.
1920 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
||
1921 prog
->data
->Version
< 150)
1924 /* From the GLSL 1.50 spec, page 46:
1926 * "All geometry shader output layout declarations in a program
1927 * must declare the same layout and same value for
1928 * max_vertices. There must be at least one geometry output
1929 * layout declaration somewhere in a program, but not all
1930 * geometry shaders (compilation units) are required to
1934 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1935 struct gl_shader
*shader
= shader_list
[i
];
1937 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1938 if (linked_shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
&&
1939 linked_shader
->info
.Geom
.InputType
!=
1940 shader
->info
.Geom
.InputType
) {
1941 linker_error(prog
, "geometry shader defined with conflicting "
1945 linked_shader
->info
.Geom
.InputType
= shader
->info
.Geom
.InputType
;
1948 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1949 if (linked_shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
&&
1950 linked_shader
->info
.Geom
.OutputType
!=
1951 shader
->info
.Geom
.OutputType
) {
1952 linker_error(prog
, "geometry shader defined with conflicting "
1956 linked_shader
->info
.Geom
.OutputType
= shader
->info
.Geom
.OutputType
;
1959 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1960 if (linked_shader
->info
.Geom
.VerticesOut
!= -1 &&
1961 linked_shader
->info
.Geom
.VerticesOut
!=
1962 shader
->info
.Geom
.VerticesOut
) {
1963 linker_error(prog
, "geometry shader defined with conflicting "
1964 "output vertex count (%d and %d)\n",
1965 linked_shader
->info
.Geom
.VerticesOut
,
1966 shader
->info
.Geom
.VerticesOut
);
1969 linked_shader
->info
.Geom
.VerticesOut
= shader
->info
.Geom
.VerticesOut
;
1972 if (shader
->info
.Geom
.Invocations
!= 0) {
1973 if (linked_shader
->info
.Geom
.Invocations
!= 0 &&
1974 linked_shader
->info
.Geom
.Invocations
!=
1975 shader
->info
.Geom
.Invocations
) {
1976 linker_error(prog
, "geometry shader defined with conflicting "
1977 "invocation count (%d and %d)\n",
1978 linked_shader
->info
.Geom
.Invocations
,
1979 shader
->info
.Geom
.Invocations
);
1982 linked_shader
->info
.Geom
.Invocations
= shader
->info
.Geom
.Invocations
;
1986 /* Just do the intrastage -> interstage propagation right now,
1987 * since we already know we're in the right type of shader program
1990 if (linked_shader
->info
.Geom
.InputType
== PRIM_UNKNOWN
) {
1992 "geometry shader didn't declare primitive input type\n");
1996 if (linked_shader
->info
.Geom
.OutputType
== PRIM_UNKNOWN
) {
1998 "geometry shader didn't declare primitive output type\n");
2002 if (linked_shader
->info
.Geom
.VerticesOut
== -1) {
2004 "geometry shader didn't declare max_vertices\n");
2008 if (linked_shader
->info
.Geom
.Invocations
== 0)
2009 linked_shader
->info
.Geom
.Invocations
= 1;
2014 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2015 * qualifiers for the attached compute shaders, and propagate them to the
2016 * linked CS and linked shader program.
2019 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2020 struct gl_linked_shader
*linked_shader
,
2021 struct gl_shader
**shader_list
,
2022 unsigned num_shaders
)
2024 for (int i
= 0; i
< 3; i
++)
2025 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
2027 linked_shader
->info
.Comp
.LocalSizeVariable
= false;
2029 /* This function is called for all shader stages, but it only has an effect
2030 * for compute shaders.
2032 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2035 /* From the ARB_compute_shader spec, in the section describing local size
2038 * If multiple compute shaders attached to a single program object
2039 * declare local work-group size, the declarations must be identical;
2040 * otherwise a link-time error results. Furthermore, if a program
2041 * object contains any compute shaders, at least one must contain an
2042 * input layout qualifier specifying the local work sizes of the
2043 * program, or a link-time error will occur.
2045 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2046 struct gl_shader
*shader
= shader_list
[sh
];
2048 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2049 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2050 for (int i
= 0; i
< 3; i
++) {
2051 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2052 shader
->info
.Comp
.LocalSize
[i
]) {
2053 linker_error(prog
, "compute shader defined with conflicting "
2059 for (int i
= 0; i
< 3; i
++) {
2060 linked_shader
->info
.Comp
.LocalSize
[i
] =
2061 shader
->info
.Comp
.LocalSize
[i
];
2063 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2064 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2065 /* The ARB_compute_variable_group_size spec says:
2067 * If one compute shader attached to a program declares a
2068 * variable local group size and a second compute shader
2069 * attached to the same program declares a fixed local group
2070 * size, a link-time error results.
2072 linker_error(prog
, "compute shader defined with both fixed and "
2073 "variable local group size\n");
2076 linked_shader
->info
.Comp
.LocalSizeVariable
= true;
2080 /* Just do the intrastage -> interstage propagation right now,
2081 * since we already know we're in the right type of shader program
2084 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0 &&
2085 !linked_shader
->info
.Comp
.LocalSizeVariable
) {
2086 linker_error(prog
, "compute shader must contain a fixed or a variable "
2087 "local group size\n");
2090 for (int i
= 0; i
< 3; i
++)
2091 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2093 prog
->Comp
.LocalSizeVariable
=
2094 linked_shader
->info
.Comp
.LocalSizeVariable
;
2099 * Combine a group of shaders for a single stage to generate a linked shader
2102 * If this function is supplied a single shader, it is cloned, and the new
2103 * shader is returned.
2105 struct gl_linked_shader
*
2106 link_intrastage_shaders(void *mem_ctx
,
2107 struct gl_context
*ctx
,
2108 struct gl_shader_program
*prog
,
2109 struct gl_shader
**shader_list
,
2110 unsigned num_shaders
,
2111 bool allow_missing_main
)
2113 struct gl_uniform_block
*ubo_blocks
= NULL
;
2114 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2115 unsigned num_ubo_blocks
= 0;
2116 unsigned num_ssbo_blocks
= 0;
2118 /* Check that global variables defined in multiple shaders are consistent.
2120 glsl_symbol_table variables
;
2121 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2122 if (shader_list
[i
] == NULL
)
2124 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2127 if (!prog
->data
->LinkStatus
)
2130 /* Check that interface blocks defined in multiple shaders are consistent.
2132 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2134 if (!prog
->data
->LinkStatus
)
2137 /* Check that there is only a single definition of each function signature
2138 * across all shaders.
2140 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2141 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2142 ir_function
*const f
= node
->as_function();
2147 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2148 ir_function
*const other
=
2149 shader_list
[j
]->symbols
->get_function(f
->name
);
2151 /* If the other shader has no function (and therefore no function
2152 * signatures) with the same name, skip to the next shader.
2157 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2158 if (!sig
->is_defined
)
2161 ir_function_signature
*other_sig
=
2162 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2164 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2165 linker_error(prog
, "function `%s' is multiply defined\n",
2174 /* Find the shader that defines main, and make a clone of it.
2176 * Starting with the clone, search for undefined references. If one is
2177 * found, find the shader that defines it. Clone the reference and add
2178 * it to the shader. Repeat until there are no undefined references or
2179 * until a reference cannot be resolved.
2181 gl_shader
*main
= NULL
;
2182 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2183 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2184 main
= shader_list
[i
];
2189 if (main
== NULL
&& allow_missing_main
)
2190 main
= shader_list
[0];
2193 linker_error(prog
, "%s shader lacks `main'\n",
2194 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2198 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2199 linked
->Stage
= shader_list
[0]->Stage
;
2201 /* Create program and attach it to the linked shader */
2202 struct gl_program
*gl_prog
=
2203 ctx
->Driver
.NewProgram(ctx
,
2204 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2207 prog
->data
->LinkStatus
= false;
2208 _mesa_delete_linked_shader(ctx
, linked
);
2212 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2214 /* Don't use _mesa_reference_program() just take ownership */
2215 linked
->Program
= gl_prog
;
2217 linked
->ir
= new(linked
) exec_list
;
2218 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2220 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2221 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2222 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2223 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2224 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2225 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2228 populate_symbol_table(linked
);
2230 /* The pointer to the main function in the final linked shader (i.e., the
2231 * copy of the original shader that contained the main function).
2233 ir_function_signature
*const main_sig
=
2234 _mesa_get_main_function_signature(linked
->symbols
);
2236 /* Move any instructions other than variable declarations or function
2237 * declarations into main.
2239 if (main_sig
!= NULL
) {
2240 exec_node
*insertion_point
=
2241 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2244 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2245 if (shader_list
[i
] == main
)
2248 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2249 insertion_point
, true, linked
);
2253 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2254 _mesa_delete_linked_shader(ctx
, linked
);
2258 /* Make a pass over all variable declarations to ensure that arrays with
2259 * unspecified sizes have a size specified. The size is inferred from the
2260 * max_array_access field.
2262 array_sizing_visitor v
;
2264 v
.fixup_unnamed_interface_types();
2266 /* Link up uniform blocks defined within this stage. */
2267 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2268 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2270 if (!prog
->data
->LinkStatus
) {
2271 _mesa_delete_linked_shader(ctx
, linked
);
2275 /* Copy ubo blocks to linked shader list */
2276 linked
->Program
->sh
.UniformBlocks
=
2277 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2278 ralloc_steal(linked
, ubo_blocks
);
2279 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2280 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2282 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2284 /* Copy ssbo blocks to linked shader list */
2285 linked
->Program
->sh
.ShaderStorageBlocks
=
2286 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2287 ralloc_steal(linked
, ssbo_blocks
);
2288 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2289 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2291 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2293 /* At this point linked should contain all of the linked IR, so
2294 * validate it to make sure nothing went wrong.
2296 validate_ir_tree(linked
->ir
);
2298 /* Set the size of geometry shader input arrays */
2299 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2300 unsigned num_vertices
= vertices_per_prim(linked
->info
.Geom
.InputType
);
2301 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2302 MESA_SHADER_GEOMETRY
);
2303 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2304 ir
->accept(&input_resize_visitor
);
2308 if (ctx
->Const
.VertexID_is_zero_based
)
2309 lower_vertex_id(linked
);
2312 /* Compute the source checksum. */
2313 linked
->SourceChecksum
= 0;
2314 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2315 if (shader_list
[i
] == NULL
)
2317 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2325 * Update the sizes of linked shader uniform arrays to the maximum
2328 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2330 * If one or more elements of an array are active,
2331 * GetActiveUniform will return the name of the array in name,
2332 * subject to the restrictions listed above. The type of the array
2333 * is returned in type. The size parameter contains the highest
2334 * array element index used, plus one. The compiler or linker
2335 * determines the highest index used. There will be only one
2336 * active uniform reported by the GL per uniform array.
2340 update_array_sizes(struct gl_shader_program
*prog
)
2342 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2343 if (prog
->_LinkedShaders
[i
] == NULL
)
2346 bool types_were_updated
= false;
2348 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2349 ir_variable
*const var
= node
->as_variable();
2351 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2352 !var
->type
->is_array())
2355 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2356 * will not be eliminated. Since we always do std140, just
2357 * don't resize arrays in UBOs.
2359 * Atomic counters are supposed to get deterministic
2360 * locations assigned based on the declaration ordering and
2361 * sizes, array compaction would mess that up.
2363 * Subroutine uniforms are not removed.
2365 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2366 var
->type
->contains_subroutine() || var
->constant_initializer
)
2369 int size
= var
->data
.max_array_access
;
2370 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2371 if (prog
->_LinkedShaders
[j
] == NULL
)
2374 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2375 ir_variable
*other_var
= node2
->as_variable();
2379 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2380 other_var
->data
.max_array_access
> size
) {
2381 size
= other_var
->data
.max_array_access
;
2386 if (size
+ 1 != (int)var
->type
->length
) {
2387 /* If this is a built-in uniform (i.e., it's backed by some
2388 * fixed-function state), adjust the number of state slots to
2389 * match the new array size. The number of slots per array entry
2390 * is not known. It seems safe to assume that the total number of
2391 * slots is an integer multiple of the number of array elements.
2392 * Determine the number of slots per array element by dividing by
2393 * the old (total) size.
2395 const unsigned num_slots
= var
->get_num_state_slots();
2396 if (num_slots
> 0) {
2397 var
->set_num_state_slots((size
+ 1)
2398 * (num_slots
/ var
->type
->length
));
2401 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2403 types_were_updated
= true;
2407 /* Update the types of dereferences in case we changed any. */
2408 if (types_were_updated
) {
2409 deref_type_updater v
;
2410 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2416 * Resize tessellation evaluation per-vertex inputs to the size of
2417 * tessellation control per-vertex outputs.
2420 resize_tes_inputs(struct gl_context
*ctx
,
2421 struct gl_shader_program
*prog
)
2423 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2426 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2427 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2429 /* If no control shader is present, then the TES inputs are statically
2430 * sized to MaxPatchVertices; the actual size of the arrays won't be
2431 * known until draw time.
2433 const int num_vertices
= tcs
2434 ? tcs
->info
.TessCtrl
.VerticesOut
2435 : ctx
->Const
.MaxPatchVertices
;
2437 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2438 MESA_SHADER_TESS_EVAL
);
2439 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2440 ir
->accept(&input_resize_visitor
);
2443 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2444 /* Convert the gl_PatchVerticesIn system value into a constant, since
2445 * the value is known at this point.
2447 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2448 ir_variable
*var
= ir
->as_variable();
2449 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2450 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2451 void *mem_ctx
= ralloc_parent(var
);
2452 var
->data
.location
= 0;
2453 var
->data
.explicit_location
= false;
2455 var
->data
.mode
= ir_var_auto
;
2456 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2458 var
->data
.mode
= ir_var_uniform
;
2459 var
->data
.how_declared
= ir_var_hidden
;
2460 var
->allocate_state_slots(1);
2461 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2462 slot0
->swizzle
= SWIZZLE_XXXX
;
2463 slot0
->tokens
[0] = STATE_INTERNAL
;
2464 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2465 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2466 slot0
->tokens
[i
] = 0;
2474 * Find a contiguous set of available bits in a bitmask.
2476 * \param used_mask Bits representing used (1) and unused (0) locations
2477 * \param needed_count Number of contiguous bits needed.
2480 * Base location of the available bits on success or -1 on failure.
2483 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2485 unsigned needed_mask
= (1 << needed_count
) - 1;
2486 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2488 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2489 * cannot optimize possibly infinite loops" for the loop below.
2491 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2494 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2495 if ((needed_mask
& ~used_mask
) == needed_mask
)
2506 * Assign locations for either VS inputs or FS outputs
2508 * \param mem_ctx Temporary ralloc context used for linking
2509 * \param prog Shader program whose variables need locations assigned
2510 * \param constants Driver specific constant values for the program.
2511 * \param target_index Selector for the program target to receive location
2512 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2513 * \c MESA_SHADER_FRAGMENT.
2516 * If locations are successfully assigned, true is returned. Otherwise an
2517 * error is emitted to the shader link log and false is returned.
2520 assign_attribute_or_color_locations(void *mem_ctx
,
2521 gl_shader_program
*prog
,
2522 struct gl_constants
*constants
,
2523 unsigned target_index
)
2525 /* Maximum number of generic locations. This corresponds to either the
2526 * maximum number of draw buffers or the maximum number of generic
2529 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2530 constants
->Program
[target_index
].MaxAttribs
:
2531 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2533 /* Mark invalid locations as being used.
2535 unsigned used_locations
= (max_index
>= 32)
2536 ? ~0 : ~((1 << max_index
) - 1);
2537 unsigned double_storage_locations
= 0;
2539 assert((target_index
== MESA_SHADER_VERTEX
)
2540 || (target_index
== MESA_SHADER_FRAGMENT
));
2542 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2546 /* Operate in a total of four passes.
2548 * 1. Invalidate the location assignments for all vertex shader inputs.
2550 * 2. Assign locations for inputs that have user-defined (via
2551 * glBindVertexAttribLocation) locations and outputs that have
2552 * user-defined locations (via glBindFragDataLocation).
2554 * 3. Sort the attributes without assigned locations by number of slots
2555 * required in decreasing order. Fragmentation caused by attribute
2556 * locations assigned by the application may prevent large attributes
2557 * from having enough contiguous space.
2559 * 4. Assign locations to any inputs without assigned locations.
2562 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2563 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2565 const enum ir_variable_mode direction
=
2566 (target_index
== MESA_SHADER_VERTEX
)
2567 ? ir_var_shader_in
: ir_var_shader_out
;
2570 /* Temporary storage for the set of attributes that need locations assigned.
2576 /* Used below in the call to qsort. */
2577 static int compare(const void *a
, const void *b
)
2579 const temp_attr
*const l
= (const temp_attr
*) a
;
2580 const temp_attr
*const r
= (const temp_attr
*) b
;
2582 /* Reversed because we want a descending order sort below. */
2583 return r
->slots
- l
->slots
;
2586 assert(max_index
<= 32);
2588 /* Temporary array for the set of attributes that have locations assigned.
2590 ir_variable
*assigned
[16];
2592 unsigned num_attr
= 0;
2593 unsigned assigned_attr
= 0;
2595 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2596 ir_variable
*const var
= node
->as_variable();
2598 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2601 if (var
->data
.explicit_location
) {
2602 var
->data
.is_unmatched_generic_inout
= 0;
2603 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2604 || (var
->data
.location
< 0)) {
2606 "invalid explicit location %d specified for `%s'\n",
2607 (var
->data
.location
< 0)
2608 ? var
->data
.location
2609 : var
->data
.location
- generic_base
,
2613 } else if (target_index
== MESA_SHADER_VERTEX
) {
2616 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2617 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2618 var
->data
.location
= binding
;
2619 var
->data
.is_unmatched_generic_inout
= 0;
2621 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2624 const char *name
= var
->name
;
2625 const glsl_type
*type
= var
->type
;
2628 /* Check if there's a binding for the variable name */
2629 if (prog
->FragDataBindings
->get(binding
, name
)) {
2630 assert(binding
>= FRAG_RESULT_DATA0
);
2631 var
->data
.location
= binding
;
2632 var
->data
.is_unmatched_generic_inout
= 0;
2634 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2635 var
->data
.index
= index
;
2640 /* If not, but it's an array type, look for name[0] */
2641 if (type
->is_array()) {
2642 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2643 type
= type
->fields
.array
;
2651 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2654 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2656 * "Output binding assignments will cause LinkProgram to fail:
2658 * If the program has an active output assigned to a location greater
2659 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2660 * an active output assigned an index greater than or equal to one;"
2662 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2663 var
->data
.location
- generic_base
>=
2664 (int) constants
->MaxDualSourceDrawBuffers
) {
2666 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2667 "with index %u for %s\n",
2668 var
->data
.location
- generic_base
, var
->data
.index
,
2673 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2675 /* If the variable is not a built-in and has a location statically
2676 * assigned in the shader (presumably via a layout qualifier), make sure
2677 * that it doesn't collide with other assigned locations. Otherwise,
2678 * add it to the list of variables that need linker-assigned locations.
2680 if (var
->data
.location
!= -1) {
2681 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2682 /* From page 61 of the OpenGL 4.0 spec:
2684 * "LinkProgram will fail if the attribute bindings assigned
2685 * by BindAttribLocation do not leave not enough space to
2686 * assign a location for an active matrix attribute or an
2687 * active attribute array, both of which require multiple
2688 * contiguous generic attributes."
2690 * I think above text prohibits the aliasing of explicit and
2691 * automatic assignments. But, aliasing is allowed in manual
2692 * assignments of attribute locations. See below comments for
2695 * From OpenGL 4.0 spec, page 61:
2697 * "It is possible for an application to bind more than one
2698 * attribute name to the same location. This is referred to as
2699 * aliasing. This will only work if only one of the aliased
2700 * attributes is active in the executable program, or if no
2701 * path through the shader consumes more than one attribute of
2702 * a set of attributes aliased to the same location. A link
2703 * error can occur if the linker determines that every path
2704 * through the shader consumes multiple aliased attributes,
2705 * but implementations are not required to generate an error
2708 * From GLSL 4.30 spec, page 54:
2710 * "A program will fail to link if any two non-vertex shader
2711 * input variables are assigned to the same location. For
2712 * vertex shaders, multiple input variables may be assigned
2713 * to the same location using either layout qualifiers or via
2714 * the OpenGL API. However, such aliasing is intended only to
2715 * support vertex shaders where each execution path accesses
2716 * at most one input per each location. Implementations are
2717 * permitted, but not required, to generate link-time errors
2718 * if they detect that every path through the vertex shader
2719 * executable accesses multiple inputs assigned to any single
2720 * location. For all shader types, a program will fail to link
2721 * if explicit location assignments leave the linker unable
2722 * to find space for other variables without explicit
2725 * From OpenGL ES 3.0 spec, page 56:
2727 * "Binding more than one attribute name to the same location
2728 * is referred to as aliasing, and is not permitted in OpenGL
2729 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2730 * fail when this condition exists. However, aliasing is
2731 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2732 * This will only work if only one of the aliased attributes
2733 * is active in the executable program, or if no path through
2734 * the shader consumes more than one attribute of a set of
2735 * attributes aliased to the same location. A link error can
2736 * occur if the linker determines that every path through the
2737 * shader consumes multiple aliased attributes, but implemen-
2738 * tations are not required to generate an error in this case."
2740 * After looking at above references from OpenGL, OpenGL ES and
2741 * GLSL specifications, we allow aliasing of vertex input variables
2742 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2744 * NOTE: This is not required by the spec but its worth mentioning
2745 * here that we're not doing anything to make sure that no path
2746 * through the vertex shader executable accesses multiple inputs
2747 * assigned to any single location.
2750 /* Mask representing the contiguous slots that will be used by
2753 const unsigned attr
= var
->data
.location
- generic_base
;
2754 const unsigned use_mask
= (1 << slots
) - 1;
2755 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2756 ? "vertex shader input" : "fragment shader output";
2758 /* Generate a link error if the requested locations for this
2759 * attribute exceed the maximum allowed attribute location.
2761 if (attr
+ slots
> max_index
) {
2763 "insufficient contiguous locations "
2764 "available for %s `%s' %d %d %d\n", string
,
2765 var
->name
, used_locations
, use_mask
, attr
);
2769 /* Generate a link error if the set of bits requested for this
2770 * attribute overlaps any previously allocated bits.
2772 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2773 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2774 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2777 * "Additionally, for fragment shader outputs, if two
2778 * variables are placed within the same location, they
2779 * must have the same underlying type (floating-point or
2780 * integer). No component aliasing of output variables or
2781 * members is allowed.
2783 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2784 unsigned assigned_slots
=
2785 assigned
[i
]->type
->count_attribute_slots(false);
2786 unsigned assig_attr
=
2787 assigned
[i
]->data
.location
- generic_base
;
2788 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2790 if ((assigned_use_mask
<< assig_attr
) &
2791 (use_mask
<< attr
)) {
2793 const glsl_type
*assigned_type
=
2794 assigned
[i
]->type
->without_array();
2795 const glsl_type
*type
= var
->type
->without_array();
2796 if (assigned_type
->base_type
!= type
->base_type
) {
2797 linker_error(prog
, "types do not match for aliased"
2798 " %ss %s and %s\n", string
,
2799 assigned
[i
]->name
, var
->name
);
2803 unsigned assigned_component_mask
=
2804 ((1 << assigned_type
->vector_elements
) - 1) <<
2805 assigned
[i
]->data
.location_frac
;
2806 unsigned component_mask
=
2807 ((1 << type
->vector_elements
) - 1) <<
2808 var
->data
.location_frac
;
2809 if (assigned_component_mask
& component_mask
) {
2810 linker_error(prog
, "overlapping component is "
2811 "assigned to %ss %s and %s "
2813 string
, assigned
[i
]->name
, var
->name
,
2814 var
->data
.location_frac
);
2819 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2820 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2821 linker_error(prog
, "overlapping location is assigned "
2822 "to %s `%s' %d %d %d\n", string
, var
->name
,
2823 used_locations
, use_mask
, attr
);
2826 linker_warning(prog
, "overlapping location is assigned "
2827 "to %s `%s' %d %d %d\n", string
, var
->name
,
2828 used_locations
, use_mask
, attr
);
2832 used_locations
|= (use_mask
<< attr
);
2834 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2836 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2837 * active attribute variables may fail to link, unless
2838 * device-dependent optimizations are able to make the program
2839 * fit within available hardware resources. For the purposes
2840 * of this test, attribute variables of the type dvec3, dvec4,
2841 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2842 * count as consuming twice as many attributes as equivalent
2843 * single-precision types. While these types use the same number
2844 * of generic attributes as their single-precision equivalents,
2845 * implementations are permitted to consume two single-precision
2846 * vectors of internal storage for each three- or four-component
2847 * double-precision vector."
2849 * Mark this attribute slot as taking up twice as much space
2850 * so we can count it properly against limits. According to
2851 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2852 * is optional behavior, but it seems preferable.
2854 if (var
->type
->without_array()->is_dual_slot())
2855 double_storage_locations
|= (use_mask
<< attr
);
2858 assigned
[assigned_attr
] = var
;
2864 if (num_attr
>= max_index
) {
2865 linker_error(prog
, "too many %s (max %u)",
2866 target_index
== MESA_SHADER_VERTEX
?
2867 "vertex shader inputs" : "fragment shader outputs",
2871 to_assign
[num_attr
].slots
= slots
;
2872 to_assign
[num_attr
].var
= var
;
2876 if (target_index
== MESA_SHADER_VERTEX
) {
2877 unsigned total_attribs_size
=
2878 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2879 _mesa_bitcount(double_storage_locations
);
2880 if (total_attribs_size
> max_index
) {
2882 "attempt to use %d vertex attribute slots only %d available ",
2883 total_attribs_size
, max_index
);
2888 /* If all of the attributes were assigned locations by the application (or
2889 * are built-in attributes with fixed locations), return early. This should
2890 * be the common case.
2895 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2897 if (target_index
== MESA_SHADER_VERTEX
) {
2898 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2899 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2900 * reserved to prevent it from being automatically allocated below.
2902 find_deref_visitor
find("gl_Vertex");
2904 if (find
.variable_found())
2905 used_locations
|= (1 << 0);
2908 for (unsigned i
= 0; i
< num_attr
; i
++) {
2909 /* Mask representing the contiguous slots that will be used by this
2912 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2914 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2917 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2918 ? "vertex shader input" : "fragment shader output";
2921 "insufficient contiguous locations "
2922 "available for %s `%s'\n",
2923 string
, to_assign
[i
].var
->name
);
2927 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2928 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2929 used_locations
|= (use_mask
<< location
);
2931 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2932 double_storage_locations
|= (use_mask
<< location
);
2935 /* Now that we have all the locations, from the GL 4.5 core spec, section
2936 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2937 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2938 * as equivalent single-precision types.
2940 if (target_index
== MESA_SHADER_VERTEX
) {
2941 unsigned total_attribs_size
=
2942 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2943 _mesa_bitcount(double_storage_locations
);
2944 if (total_attribs_size
> max_index
) {
2946 "attempt to use %d vertex attribute slots only %d available ",
2947 total_attribs_size
, max_index
);
2956 * Match explicit locations of outputs to inputs and deactivate the
2957 * unmatch flag if found so we don't optimise them away.
2960 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2961 gl_linked_shader
*consumer
)
2963 glsl_symbol_table parameters
;
2964 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2967 /* Find all shader outputs in the "producer" stage.
2969 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2970 ir_variable
*const var
= node
->as_variable();
2972 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2975 if (var
->data
.explicit_location
&&
2976 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2977 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2978 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2979 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2983 /* Match inputs to outputs */
2984 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2985 ir_variable
*const input
= node
->as_variable();
2987 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2990 ir_variable
*output
= NULL
;
2991 if (input
->data
.explicit_location
2992 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2993 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2994 [input
->data
.location_frac
];
2996 if (output
!= NULL
){
2997 input
->data
.is_unmatched_generic_inout
= 0;
2998 output
->data
.is_unmatched_generic_inout
= 0;
3005 * Store the gl_FragDepth layout in the gl_shader_program struct.
3008 store_fragdepth_layout(struct gl_shader_program
*prog
)
3010 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3014 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3016 /* We don't look up the gl_FragDepth symbol directly because if
3017 * gl_FragDepth is not used in the shader, it's removed from the IR.
3018 * However, the symbol won't be removed from the symbol table.
3020 * We're only interested in the cases where the variable is NOT removed
3023 foreach_in_list(ir_instruction
, node
, ir
) {
3024 ir_variable
*const var
= node
->as_variable();
3026 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3030 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3031 switch (var
->data
.depth_layout
) {
3032 case ir_depth_layout_none
:
3033 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3035 case ir_depth_layout_any
:
3036 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3038 case ir_depth_layout_greater
:
3039 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3041 case ir_depth_layout_less
:
3042 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3044 case ir_depth_layout_unchanged
:
3045 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3056 * Validate the resources used by a program versus the implementation limits
3059 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3061 unsigned total_uniform_blocks
= 0;
3062 unsigned total_shader_storage_blocks
= 0;
3064 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3065 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3070 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3071 linker_error(prog
, "Too many %s shader texture samplers\n",
3072 _mesa_shader_stage_to_string(i
));
3075 if (sh
->num_uniform_components
>
3076 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3077 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3078 linker_warning(prog
, "Too many %s shader default uniform block "
3079 "components, but the driver will try to optimize "
3080 "them out; this is non-portable out-of-spec "
3082 _mesa_shader_stage_to_string(i
));
3084 linker_error(prog
, "Too many %s shader default uniform block "
3086 _mesa_shader_stage_to_string(i
));
3090 if (sh
->num_combined_uniform_components
>
3091 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3092 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3093 linker_warning(prog
, "Too many %s shader uniform components, "
3094 "but the driver will try to optimize them out; "
3095 "this is non-portable out-of-spec behavior\n",
3096 _mesa_shader_stage_to_string(i
));
3098 linker_error(prog
, "Too many %s shader uniform components\n",
3099 _mesa_shader_stage_to_string(i
));
3103 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3104 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3106 const unsigned max_uniform_blocks
=
3107 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3108 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3109 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3110 _mesa_shader_stage_to_string(i
),
3111 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3114 const unsigned max_shader_storage_blocks
=
3115 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3116 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3117 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3118 _mesa_shader_stage_to_string(i
),
3119 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3123 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3124 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3125 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3128 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3129 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3130 total_shader_storage_blocks
,
3131 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3134 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3135 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3136 ctx
->Const
.MaxUniformBlockSize
) {
3137 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3138 prog
->data
->UniformBlocks
[i
].Name
,
3139 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3140 ctx
->Const
.MaxUniformBlockSize
);
3144 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3145 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3146 ctx
->Const
.MaxShaderStorageBlockSize
) {
3147 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3148 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3149 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3150 ctx
->Const
.MaxShaderStorageBlockSize
);
3156 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3158 unsigned mask
= prog
->data
->linked_stages
;
3160 const int i
= u_bit_scan(&mask
);
3161 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3163 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3164 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3167 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3173 if (p
->sh
.NumSubroutineFunctions
== 0) {
3174 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3177 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3178 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3179 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3180 if (fn
->types
[k
] == uni
->type
) {
3186 uni
->num_compatible_subroutines
= count
;
3192 check_subroutine_resources(struct gl_shader_program
*prog
)
3194 unsigned mask
= prog
->data
->linked_stages
;
3196 const int i
= u_bit_scan(&mask
);
3197 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3199 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3200 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3201 _mesa_shader_stage_to_string(i
));
3206 * Validate shader image resources.
3209 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3211 unsigned total_image_units
= 0;
3212 unsigned fragment_outputs
= 0;
3213 unsigned total_shader_storage_blocks
= 0;
3215 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3218 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3219 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3222 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3223 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3224 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3225 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3227 total_image_units
+= sh
->NumImages
;
3228 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3230 if (i
== MESA_SHADER_FRAGMENT
) {
3231 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3232 ir_variable
*var
= node
->as_variable();
3233 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3234 /* since there are no double fs outputs - pass false */
3235 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3241 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3242 linker_error(prog
, "Too many combined image uniforms\n");
3244 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3245 ctx
->Const
.MaxCombinedShaderOutputResources
)
3246 linker_error(prog
, "Too many combined image uniforms, shader storage "
3247 " buffers and fragment outputs\n");
3252 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3253 * for a variable, checks for overlaps between other uniforms using explicit
3257 reserve_explicit_locations(struct gl_shader_program
*prog
,
3258 string_to_uint_map
*map
, ir_variable
*var
)
3260 unsigned slots
= var
->type
->uniform_locations();
3261 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3262 unsigned return_value
= slots
;
3264 /* Resize remap table if locations do not fit in the current one. */
3265 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3266 prog
->UniformRemapTable
=
3267 reralloc(prog
, prog
->UniformRemapTable
,
3268 gl_uniform_storage
*,
3271 if (!prog
->UniformRemapTable
) {
3272 linker_error(prog
, "Out of memory during linking.\n");
3276 /* Initialize allocated space. */
3277 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3278 prog
->UniformRemapTable
[i
] = NULL
;
3280 prog
->NumUniformRemapTable
= max_loc
+ 1;
3283 for (unsigned i
= 0; i
< slots
; i
++) {
3284 unsigned loc
= var
->data
.location
+ i
;
3286 /* Check if location is already used. */
3287 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3289 /* Possibly same uniform from a different stage, this is ok. */
3291 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3296 /* ARB_explicit_uniform_location specification states:
3298 * "No two default-block uniform variables in the program can have
3299 * the same location, even if they are unused, otherwise a compiler
3300 * or linker error will be generated."
3303 "location qualifier for uniform %s overlaps "
3304 "previously used location\n",
3309 /* Initialize location as inactive before optimization
3310 * rounds and location assignment.
3312 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3315 /* Note, base location used for arrays. */
3316 map
->put(var
->data
.location
, var
->name
);
3318 return return_value
;
3322 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3323 struct gl_program
*p
,
3326 unsigned slots
= var
->type
->uniform_locations();
3327 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3329 /* Resize remap table if locations do not fit in the current one. */
3330 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3331 p
->sh
.SubroutineUniformRemapTable
=
3332 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3333 gl_uniform_storage
*,
3336 if (!p
->sh
.SubroutineUniformRemapTable
) {
3337 linker_error(prog
, "Out of memory during linking.\n");
3341 /* Initialize allocated space. */
3342 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3343 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3345 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3348 for (unsigned i
= 0; i
< slots
; i
++) {
3349 unsigned loc
= var
->data
.location
+ i
;
3351 /* Check if location is already used. */
3352 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3354 /* ARB_explicit_uniform_location specification states:
3355 * "No two subroutine uniform variables can have the same location
3356 * in the same shader stage, otherwise a compiler or linker error
3357 * will be generated."
3360 "location qualifier for uniform %s overlaps "
3361 "previously used location\n",
3366 /* Initialize location as inactive before optimization
3367 * rounds and location assignment.
3369 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3375 * Check and reserve all explicit uniform locations, called before
3376 * any optimizations happen to handle also inactive uniforms and
3377 * inactive array elements that may get trimmed away.
3380 check_explicit_uniform_locations(struct gl_context
*ctx
,
3381 struct gl_shader_program
*prog
)
3383 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3386 /* This map is used to detect if overlapping explicit locations
3387 * occur with the same uniform (from different stage) or a different one.
3389 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3392 linker_error(prog
, "Out of memory during linking.\n");
3396 unsigned entries_total
= 0;
3397 unsigned mask
= prog
->data
->linked_stages
;
3399 const int i
= u_bit_scan(&mask
);
3400 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3402 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3403 ir_variable
*var
= node
->as_variable();
3404 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3407 if (var
->data
.explicit_location
) {
3409 if (var
->type
->without_array()->is_subroutine())
3410 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3412 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3416 entries_total
+= slots
;
3427 struct empty_uniform_block
*current_block
= NULL
;
3429 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3430 /* We found empty space in UniformRemapTable. */
3431 if (prog
->UniformRemapTable
[i
] == NULL
) {
3432 /* We've found the beginning of a new continous block of empty slots */
3433 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3434 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3435 current_block
->start
= i
;
3436 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3437 ¤t_block
->link
);
3440 /* The current block continues, so we simply increment its slots */
3441 current_block
->slots
++;
3446 return entries_total
;
3450 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3451 GLenum type
, const char *name
)
3453 bool found_interface
= false;
3454 unsigned block_name_len
= 0;
3455 const char *block_name_dot
= strchr(name
, '.');
3457 /* These rules only apply to buffer variables. So we return
3458 * true for the rest of types.
3460 if (type
!= GL_BUFFER_VARIABLE
)
3463 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3464 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3465 block_name_len
= strlen(block_name
);
3467 const char *block_square_bracket
= strchr(block_name
, '[');
3468 if (block_square_bracket
) {
3469 /* The block is part of an array of named interfaces,
3470 * for the name comparison we ignore the "[x]" part.
3472 block_name_len
-= strlen(block_square_bracket
);
3475 if (block_name_dot
) {
3476 /* Check if the variable name starts with the interface
3477 * name. The interface name (if present) should have the
3478 * length than the interface block name we are comparing to.
3480 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3481 if (len
!= block_name_len
)
3485 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3486 found_interface
= true;
3491 /* We remove the interface name from the buffer variable name,
3492 * including the dot that follows it.
3494 if (found_interface
)
3495 name
= name
+ block_name_len
+ 1;
3497 /* The ARB_program_interface_query spec says:
3499 * "For an active shader storage block member declared as an array, an
3500 * entry will be generated only for the first array element, regardless
3501 * of its type. For arrays of aggregate types, the enumeration rules
3502 * are applied recursively for the single enumerated array element."
3504 const char *struct_first_dot
= strchr(name
, '.');
3505 const char *first_square_bracket
= strchr(name
, '[');
3507 /* The buffer variable is on top level and it is not an array */
3508 if (!first_square_bracket
) {
3510 /* The shader storage block member is a struct, then generate the entry */
3511 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3514 /* Shader storage block member is an array, only generate an entry for the
3515 * first array element.
3517 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3525 add_program_resource(struct gl_shader_program
*prog
,
3526 struct set
*resource_set
,
3527 GLenum type
, const void *data
, uint8_t stages
)
3531 /* If resource already exists, do not add it again. */
3532 if (_mesa_set_search(resource_set
, data
))
3535 prog
->ProgramResourceList
=
3537 prog
->ProgramResourceList
,
3538 gl_program_resource
,
3539 prog
->NumProgramResourceList
+ 1);
3541 if (!prog
->ProgramResourceList
) {
3542 linker_error(prog
, "Out of memory during linking.\n");
3546 struct gl_program_resource
*res
=
3547 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3551 res
->StageReferences
= stages
;
3553 prog
->NumProgramResourceList
++;
3555 _mesa_set_add(resource_set
, data
);
3560 /* Function checks if a variable var is a packed varying and
3561 * if given name is part of packed varying's list.
3563 * If a variable is a packed varying, it has a name like
3564 * 'packed:a,b,c' where a, b and c are separate variables.
3567 included_in_packed_varying(ir_variable
*var
, const char *name
)
3569 if (strncmp(var
->name
, "packed:", 7) != 0)
3572 char *list
= strdup(var
->name
+ 7);
3577 char *token
= strtok_r(list
, ",", &saveptr
);
3579 if (strcmp(token
, name
) == 0) {
3583 token
= strtok_r(NULL
, ",", &saveptr
);
3590 * Function builds a stage reference bitmask from variable name.
3593 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3598 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3599 * used for reference mask in gl_program_resource will need to be changed.
3601 assert(MESA_SHADER_STAGES
< 8);
3603 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3604 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3608 /* Shader symbol table may contain variables that have
3609 * been optimized away. Search IR for the variable instead.
3611 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3612 ir_variable
*var
= node
->as_variable();
3614 unsigned baselen
= strlen(var
->name
);
3616 if (included_in_packed_varying(var
, name
)) {
3621 /* Type needs to match if specified, otherwise we might
3622 * pick a variable with same name but different interface.
3624 if (var
->data
.mode
!= mode
)
3627 if (strncmp(var
->name
, name
, baselen
) == 0) {
3628 /* Check for exact name matches but also check for arrays and
3631 if (name
[baselen
] == '\0' ||
3632 name
[baselen
] == '[' ||
3633 name
[baselen
] == '.') {
3645 * Create gl_shader_variable from ir_variable class.
3647 static gl_shader_variable
*
3648 create_shader_variable(struct gl_shader_program
*shProg
,
3649 const ir_variable
*in
,
3650 const char *name
, const glsl_type
*type
,
3651 const glsl_type
*interface_type
,
3652 bool use_implicit_location
, int location
,
3653 const glsl_type
*outermost_struct_type
)
3655 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3659 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3660 * expect to see gl_VertexID in the program resource list. Pretend.
3662 if (in
->data
.mode
== ir_var_system_value
&&
3663 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3664 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3665 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3666 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3667 (in
->data
.mode
== ir_var_system_value
&&
3668 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3669 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3670 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3671 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3672 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3673 (in
->data
.mode
== ir_var_system_value
&&
3674 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3675 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3676 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3678 out
->name
= ralloc_strdup(shProg
, name
);
3684 /* The ARB_program_interface_query spec says:
3686 * "Not all active variables are assigned valid locations; the
3687 * following variables will have an effective location of -1:
3689 * * uniforms declared as atomic counters;
3691 * * members of a uniform block;
3693 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3695 * * inputs or outputs not declared with a "location" layout
3696 * qualifier, except for vertex shader inputs and fragment shader
3699 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3700 is_gl_identifier(in
->name
) ||
3701 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3704 out
->location
= location
;
3708 out
->outermost_struct_type
= outermost_struct_type
;
3709 out
->interface_type
= interface_type
;
3710 out
->component
= in
->data
.location_frac
;
3711 out
->index
= in
->data
.index
;
3712 out
->patch
= in
->data
.patch
;
3713 out
->mode
= in
->data
.mode
;
3714 out
->interpolation
= in
->data
.interpolation
;
3715 out
->explicit_location
= in
->data
.explicit_location
;
3716 out
->precision
= in
->data
.precision
;
3721 static const glsl_type
*
3722 resize_to_max_patch_vertices(const struct gl_context
*ctx
,
3723 const glsl_type
*type
)
3728 return glsl_type::get_array_instance(type
->fields
.array
,
3729 ctx
->Const
.MaxPatchVertices
);
3733 add_shader_variable(const struct gl_context
*ctx
,
3734 struct gl_shader_program
*shProg
,
3735 struct set
*resource_set
,
3736 unsigned stage_mask
,
3737 GLenum programInterface
, ir_variable
*var
,
3738 const char *name
, const glsl_type
*type
,
3739 bool use_implicit_location
, int location
,
3740 const glsl_type
*outermost_struct_type
= NULL
)
3742 const glsl_type
*interface_type
= var
->get_interface_type();
3744 if (outermost_struct_type
== NULL
) {
3745 /* Unsized (non-patch) TCS output/TES input arrays are implicitly
3746 * sized to gl_MaxPatchVertices. Internally, we shrink them to a
3749 * This can cause trouble with SSO programs. Since the TCS declares
3750 * the number of output vertices, we can always shrink TCS output
3751 * arrays. However, the TES might not be linked with a TCS, in
3752 * which case it won't know the size of the patch. In other words,
3753 * the TCS and TES may disagree on the (smaller) array sizes. This
3754 * can result in the resource names differing across stages, causing
3755 * SSO validation failures and other cascading issues.
3757 * Expanding the array size to the full gl_MaxPatchVertices fixes
3758 * these issues. It's also what program interface queries expect,
3759 * as that is the official size of the array.
3761 if (var
->data
.tess_varying_implicit_sized_array
) {
3762 type
= resize_to_max_patch_vertices(ctx
, type
);
3763 interface_type
= resize_to_max_patch_vertices(ctx
, interface_type
);
3766 if (var
->data
.from_named_ifc_block
) {
3767 const char *interface_name
= interface_type
->name
;
3769 if (interface_type
->is_array()) {
3770 /* Issue #16 of the ARB_program_interface_query spec says:
3772 * "* If a variable is a member of an interface block without an
3773 * instance name, it is enumerated using just the variable name.
3775 * * If a variable is a member of an interface block with an
3776 * instance name, it is enumerated as "BlockName.Member", where
3777 * "BlockName" is the name of the interface block (not the
3778 * instance name) and "Member" is the name of the variable."
3780 * In particular, it indicates that it should be "BlockName",
3781 * not "BlockName[array length]". The conformance suite and
3782 * dEQP both require this behavior.
3784 * Here, we unwrap the extra array level added by named interface
3785 * block array lowering so we have the correct variable type. We
3786 * also unwrap the interface type when constructing the name.
3788 * We leave interface_type the same so that ES 3.x SSO pipeline
3789 * validation can enforce the rules requiring array length to
3790 * match on interface blocks.
3792 type
= type
->fields
.array
;
3794 interface_name
= interface_type
->fields
.array
->name
;
3797 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3801 switch (type
->base_type
) {
3802 case GLSL_TYPE_STRUCT
: {
3803 /* The ARB_program_interface_query spec says:
3805 * "For an active variable declared as a structure, a separate entry
3806 * will be generated for each active structure member. The name of
3807 * each entry is formed by concatenating the name of the structure,
3808 * the "." character, and the name of the structure member. If a
3809 * structure member to enumerate is itself a structure or array,
3810 * these enumeration rules are applied recursively."
3812 if (outermost_struct_type
== NULL
)
3813 outermost_struct_type
= type
;
3815 unsigned field_location
= location
;
3816 for (unsigned i
= 0; i
< type
->length
; i
++) {
3817 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3818 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3819 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3820 stage_mask
, programInterface
,
3821 var
, field_name
, field
->type
,
3822 use_implicit_location
, field_location
,
3823 outermost_struct_type
))
3826 field_location
+= field
->type
->count_attribute_slots(false);
3832 /* The ARB_program_interface_query spec says:
3834 * "For an active variable declared as a single instance of a basic
3835 * type, a single entry will be generated, using the variable name
3836 * from the shader source."
3838 gl_shader_variable
*sha_v
=
3839 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3840 use_implicit_location
, location
,
3841 outermost_struct_type
);
3845 return add_program_resource(shProg
, resource_set
,
3846 programInterface
, sha_v
, stage_mask
);
3852 add_interface_variables(const struct gl_context
*ctx
,
3853 struct gl_shader_program
*shProg
,
3854 struct set
*resource_set
,
3855 unsigned stage
, GLenum programInterface
)
3857 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3859 foreach_in_list(ir_instruction
, node
, ir
) {
3860 ir_variable
*var
= node
->as_variable();
3862 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3867 switch (var
->data
.mode
) {
3868 case ir_var_system_value
:
3869 case ir_var_shader_in
:
3870 if (programInterface
!= GL_PROGRAM_INPUT
)
3872 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3873 : int(VARYING_SLOT_VAR0
);
3875 case ir_var_shader_out
:
3876 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3878 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3879 : int(VARYING_SLOT_VAR0
);
3885 if (var
->data
.patch
)
3886 loc_bias
= int(VARYING_SLOT_PATCH0
);
3888 /* Skip packed varyings, packed varyings are handled separately
3889 * by add_packed_varyings.
3891 if (strncmp(var
->name
, "packed:", 7) == 0)
3894 /* Skip fragdata arrays, these are handled separately
3895 * by add_fragdata_arrays.
3897 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3900 const bool vs_input_or_fs_output
=
3901 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3902 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3904 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3905 1 << stage
, programInterface
,
3906 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3907 var
->data
.location
- loc_bias
))
3914 add_packed_varyings(const struct gl_context
*ctx
,
3915 struct gl_shader_program
*shProg
,
3916 struct set
*resource_set
,
3917 int stage
, GLenum type
)
3919 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3922 if (!sh
|| !sh
->packed_varyings
)
3925 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3926 ir_variable
*var
= node
->as_variable();
3928 switch (var
->data
.mode
) {
3929 case ir_var_shader_in
:
3930 iface
= GL_PROGRAM_INPUT
;
3932 case ir_var_shader_out
:
3933 iface
= GL_PROGRAM_OUTPUT
;
3936 unreachable("unexpected type");
3939 if (type
== iface
) {
3940 const int stage_mask
=
3941 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3942 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3944 iface
, var
, var
->name
, var
->type
, false,
3945 var
->data
.location
- VARYING_SLOT_VAR0
))
3954 add_fragdata_arrays(const struct gl_context
*ctx
,
3955 struct gl_shader_program
*shProg
,
3956 struct set
*resource_set
)
3958 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3960 if (!sh
|| !sh
->fragdata_arrays
)
3963 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3964 ir_variable
*var
= node
->as_variable();
3966 assert(var
->data
.mode
== ir_var_shader_out
);
3968 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3969 1 << MESA_SHADER_FRAGMENT
,
3970 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3971 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3979 get_top_level_name(const char *name
)
3981 const char *first_dot
= strchr(name
, '.');
3982 const char *first_square_bracket
= strchr(name
, '[');
3985 /* The ARB_program_interface_query spec says:
3987 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3988 * the number of active array elements of the top-level shader storage
3989 * block member containing to the active variable is written to
3990 * <params>. If the top-level block member is not declared as an
3991 * array, the value one is written to <params>. If the top-level block
3992 * member is an array with no declared size, the value zero is written
3996 /* The buffer variable is on top level.*/
3997 if (!first_square_bracket
&& !first_dot
)
3998 name_size
= strlen(name
);
3999 else if ((!first_square_bracket
||
4000 (first_dot
&& first_dot
< first_square_bracket
)))
4001 name_size
= first_dot
- name
;
4003 name_size
= first_square_bracket
- name
;
4005 return strndup(name
, name_size
);
4009 get_var_name(const char *name
)
4011 const char *first_dot
= strchr(name
, '.');
4014 return strdup(name
);
4016 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4020 is_top_level_shader_storage_block_member(const char* name
,
4021 const char* interface_name
,
4022 const char* field_name
)
4024 bool result
= false;
4026 /* If the given variable is already a top-level shader storage
4027 * block member, then return array_size = 1.
4028 * We could have two possibilities: if we have an instanced
4029 * shader storage block or not instanced.
4031 * For the first, we check create a name as it was in top level and
4032 * compare it with the real name. If they are the same, then
4033 * the variable is already at top-level.
4035 * Full instanced name is: interface name + '.' + var name +
4038 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4039 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4040 if (!full_instanced_name
) {
4041 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4045 snprintf(full_instanced_name
, name_length
, "%s.%s",
4046 interface_name
, field_name
);
4048 /* Check if its top-level shader storage block member of an
4049 * instanced interface block, or of a unnamed interface block.
4051 if (strcmp(name
, full_instanced_name
) == 0 ||
4052 strcmp(name
, field_name
) == 0)
4055 free(full_instanced_name
);
4060 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4061 char *interface_name
, char *var_name
)
4063 /* The ARB_program_interface_query spec says:
4065 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4066 * the number of active array elements of the top-level shader storage
4067 * block member containing to the active variable is written to
4068 * <params>. If the top-level block member is not declared as an
4069 * array, the value one is written to <params>. If the top-level block
4070 * member is an array with no declared size, the value zero is written
4073 if (is_top_level_shader_storage_block_member(uni
->name
,
4077 else if (field
->type
->is_unsized_array())
4079 else if (field
->type
->is_array())
4080 return field
->type
->length
;
4086 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4087 const glsl_struct_field
*field
, char *interface_name
,
4090 /* The ARB_program_interface_query spec says:
4092 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4093 * identifying the stride between array elements of the top-level
4094 * shader storage block member containing the active variable is
4095 * written to <params>. For top-level block members declared as
4096 * arrays, the value written is the difference, in basic machine units,
4097 * between the offsets of the active variable for consecutive elements
4098 * in the top-level array. For top-level block members not declared as
4099 * an array, zero is written to <params>."
4101 if (field
->type
->is_array()) {
4102 const enum glsl_matrix_layout matrix_layout
=
4103 glsl_matrix_layout(field
->matrix_layout
);
4104 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4105 const glsl_type
*array_type
= field
->type
->fields
.array
;
4107 if (is_top_level_shader_storage_block_member(uni
->name
,
4112 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4113 if (array_type
->is_record() || array_type
->is_array())
4114 return glsl_align(array_type
->std140_size(row_major
), 16);
4116 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4118 return array_type
->std430_array_stride(row_major
);
4125 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4126 struct gl_uniform_storage
*uni
)
4128 int block_index
= uni
->block_index
;
4129 int array_size
= -1;
4130 int array_stride
= -1;
4131 char *var_name
= get_top_level_name(uni
->name
);
4132 char *interface_name
=
4133 get_top_level_name(uni
->is_shader_storage
?
4134 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4135 shProg
->data
->UniformBlocks
[block_index
].Name
);
4137 if (strcmp(var_name
, interface_name
) == 0) {
4138 /* Deal with instanced array of SSBOs */
4139 char *temp_name
= get_var_name(uni
->name
);
4141 linker_error(shProg
, "Out of memory during linking.\n");
4142 goto write_top_level_array_size_and_stride
;
4145 var_name
= get_top_level_name(temp_name
);
4148 linker_error(shProg
, "Out of memory during linking.\n");
4149 goto write_top_level_array_size_and_stride
;
4153 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4154 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4158 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4159 ir_variable
*var
= node
->as_variable();
4160 if (!var
|| !var
->get_interface_type() ||
4161 var
->data
.mode
!= ir_var_shader_storage
)
4164 const glsl_type
*interface
= var
->get_interface_type();
4166 if (strcmp(interface_name
, interface
->name
) != 0)
4169 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4170 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4171 if (strcmp(field
->name
, var_name
) != 0)
4174 array_stride
= get_array_stride(uni
, interface
, field
,
4175 interface_name
, var_name
);
4176 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4177 goto write_top_level_array_size_and_stride
;
4181 write_top_level_array_size_and_stride
:
4182 free(interface_name
);
4184 uni
->top_level_array_stride
= array_stride
;
4185 uni
->top_level_array_size
= array_size
;
4189 * Builds up a list of program resources that point to existing
4193 build_program_resource_list(struct gl_context
*ctx
,
4194 struct gl_shader_program
*shProg
)
4196 /* Rebuild resource list. */
4197 if (shProg
->ProgramResourceList
) {
4198 ralloc_free(shProg
->ProgramResourceList
);
4199 shProg
->ProgramResourceList
= NULL
;
4200 shProg
->NumProgramResourceList
= 0;
4203 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4205 /* Determine first input and final output stage. These are used to
4206 * detect which variables should be enumerated in the resource list
4207 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4209 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4210 if (!shProg
->_LinkedShaders
[i
])
4212 if (input_stage
== MESA_SHADER_STAGES
)
4217 /* Empty shader, no resources. */
4218 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4221 struct set
*resource_set
= _mesa_set_create(NULL
,
4223 _mesa_key_pointer_equal
);
4225 /* Program interface needs to expose varyings in case of SSO. */
4226 if (shProg
->SeparateShader
) {
4227 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4228 input_stage
, GL_PROGRAM_INPUT
))
4231 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4232 output_stage
, GL_PROGRAM_OUTPUT
))
4236 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4239 /* Add inputs and outputs to the resource list. */
4240 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4241 input_stage
, GL_PROGRAM_INPUT
))
4244 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4245 output_stage
, GL_PROGRAM_OUTPUT
))
4248 struct gl_transform_feedback_info
*linked_xfb
=
4249 shProg
->xfb_program
->sh
.LinkedTransformFeedback
;
4251 /* Add transform feedback varyings. */
4252 if (linked_xfb
->NumVarying
> 0) {
4253 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4254 if (!add_program_resource(shProg
, resource_set
,
4255 GL_TRANSFORM_FEEDBACK_VARYING
,
4256 &linked_xfb
->Varyings
[i
], 0))
4261 /* Add transform feedback buffers. */
4262 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4263 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4264 linked_xfb
->Buffers
[i
].Binding
= i
;
4265 if (!add_program_resource(shProg
, resource_set
,
4266 GL_TRANSFORM_FEEDBACK_BUFFER
,
4267 &linked_xfb
->Buffers
[i
], 0))
4272 /* Add uniforms from uniform storage. */
4273 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4274 /* Do not add uniforms internally used by Mesa. */
4275 if (shProg
->data
->UniformStorage
[i
].hidden
)
4279 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4282 /* Add stagereferences for uniforms in a uniform block. */
4283 bool is_shader_storage
=
4284 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4285 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4286 if (block_index
!= -1) {
4287 stageref
|= is_shader_storage
?
4288 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4289 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4292 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4293 if (!should_add_buffer_variable(shProg
, type
,
4294 shProg
->data
->UniformStorage
[i
].name
))
4297 if (is_shader_storage
) {
4298 calculate_array_size_and_stride(shProg
,
4299 &shProg
->data
->UniformStorage
[i
]);
4302 if (!add_program_resource(shProg
, resource_set
, type
,
4303 &shProg
->data
->UniformStorage
[i
], stageref
))
4307 /* Add program uniform blocks. */
4308 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4309 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4310 &shProg
->data
->UniformBlocks
[i
], 0))
4314 /* Add program shader storage blocks. */
4315 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4316 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4317 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4321 /* Add atomic counter buffers. */
4322 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4323 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4324 &shProg
->data
->AtomicBuffers
[i
], 0))
4328 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4330 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4333 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4334 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4335 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4338 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4339 /* add shader subroutines */
4340 if (!add_program_resource(shProg
, resource_set
,
4341 type
, &shProg
->data
->UniformStorage
[i
], 0))
4346 unsigned mask
= shProg
->data
->linked_stages
;
4348 const int i
= u_bit_scan(&mask
);
4349 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4351 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4352 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4353 if (!add_program_resource(shProg
, resource_set
,
4354 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4359 _mesa_set_destroy(resource_set
, NULL
);
4363 * This check is done to make sure we allow only constant expression
4364 * indexing and "constant-index-expression" (indexing with an expression
4365 * that includes loop induction variable).
4368 validate_sampler_array_indexing(struct gl_context
*ctx
,
4369 struct gl_shader_program
*prog
)
4371 dynamic_sampler_array_indexing_visitor v
;
4372 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4373 if (prog
->_LinkedShaders
[i
] == NULL
)
4376 bool no_dynamic_indexing
=
4377 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4379 /* Search for array derefs in shader. */
4380 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4381 if (v
.uses_dynamic_sampler_array_indexing()) {
4382 const char *msg
= "sampler arrays indexed with non-constant "
4383 "expressions is forbidden in GLSL %s %u";
4384 /* Backend has indicated that it has no dynamic indexing support. */
4385 if (no_dynamic_indexing
) {
4386 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4387 prog
->data
->Version
);
4390 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4391 prog
->data
->Version
);
4399 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4401 unsigned mask
= prog
->data
->linked_stages
;
4403 const int i
= u_bit_scan(&mask
);
4404 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4406 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4407 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4408 ir_function
*fn
= node
->as_function();
4412 if (fn
->is_subroutine
)
4413 p
->sh
.NumSubroutineUniformTypes
++;
4415 if (!fn
->num_subroutine_types
)
4418 /* these should have been calculated earlier. */
4419 assert(fn
->subroutine_index
!= -1);
4420 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4421 linker_error(prog
, "Too many subroutine functions declared.\n");
4424 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4425 struct gl_subroutine_function
,
4426 p
->sh
.NumSubroutineFunctions
+ 1);
4427 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4428 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4429 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4430 ralloc_array(p
, const struct glsl_type
*,
4431 fn
->num_subroutine_types
);
4433 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4436 * "Each subroutine with an index qualifier in the shader must be
4437 * given a unique index, otherwise a compile or link error will be
4440 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4441 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4442 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4443 linker_error(prog
, "each subroutine index qualifier in the "
4444 "shader must be unique\n");
4448 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4449 fn
->subroutine_index
;
4451 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4452 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4454 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4455 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4456 p
->sh
.NumSubroutineFunctions
++;
4462 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4464 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4466 foreach_in_list(ir_instruction
, node
, ir
) {
4467 ir_variable
*const var
= node
->as_variable();
4469 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4472 /* Don't set always active on builtins that haven't been redeclared */
4473 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4476 var
->data
.always_active_io
= true;
4481 * When separate shader programs are enabled, only input/outputs between
4482 * the stages of a multi-stage separate program can be safely removed
4483 * from the shader interface. Other inputs/outputs must remain active.
4486 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4488 unsigned first
, last
;
4489 assert(prog
->SeparateShader
);
4491 first
= MESA_SHADER_STAGES
;
4494 /* Determine first and last stage. Excluding the compute stage */
4495 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4496 if (!prog
->_LinkedShaders
[i
])
4498 if (first
== MESA_SHADER_STAGES
)
4503 if (first
== MESA_SHADER_STAGES
)
4506 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4507 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4511 if (first
== last
) {
4512 /* For a single shader program only allow inputs to the vertex shader
4513 * and outputs from the fragment shader to be removed.
4515 if (stage
!= MESA_SHADER_VERTEX
)
4516 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4517 if (stage
!= MESA_SHADER_FRAGMENT
)
4518 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4520 /* For multi-stage separate shader programs only allow inputs and
4521 * outputs between the shader stages to be removed as well as inputs
4522 * to the vertex shader and outputs from the fragment shader.
4524 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4525 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4526 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4527 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4533 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4534 unsigned num_explicit_uniform_locs
,
4535 struct gl_context
*ctx
,
4536 struct gl_shader_program
*prog
, void *mem_ctx
)
4538 bool has_xfb_qualifiers
= false;
4539 unsigned num_tfeedback_decls
= 0;
4540 char **varying_names
= NULL
;
4541 tfeedback_decl
*tfeedback_decls
= NULL
;
4543 /* Mark all generic shader inputs and outputs as unpaired. */
4544 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4545 if (prog
->_LinkedShaders
[i
] != NULL
) {
4546 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4550 unsigned prev
= first
;
4551 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4552 if (prog
->_LinkedShaders
[i
] == NULL
)
4555 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4556 prog
->_LinkedShaders
[i
]);
4560 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4561 MESA_SHADER_VERTEX
)) {
4565 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4566 MESA_SHADER_FRAGMENT
)) {
4570 /* From the ARB_enhanced_layouts spec:
4572 * "If the shader used to record output variables for transform feedback
4573 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4574 * qualifiers, the values specified by TransformFeedbackVaryings are
4575 * ignored, and the set of variables captured for transform feedback is
4576 * instead derived from the specified layout qualifiers."
4578 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4579 /* Find last stage before fragment shader */
4580 if (prog
->_LinkedShaders
[i
]) {
4581 has_xfb_qualifiers
=
4582 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4583 &num_tfeedback_decls
,
4589 if (!has_xfb_qualifiers
) {
4590 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4591 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4594 /* Find the program used for xfb. Even if we don't use xfb we still want to
4595 * set this so we can fill the default values for program interface query.
4597 prog
->xfb_program
= prog
->_LinkedShaders
[last
]->Program
;
4598 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4599 if (prog
->_LinkedShaders
[i
] == NULL
)
4602 prog
->xfb_program
= prog
->_LinkedShaders
[i
]->Program
;
4606 if (num_tfeedback_decls
!= 0) {
4607 /* From GL_EXT_transform_feedback:
4608 * A program will fail to link if:
4610 * * the <count> specified by TransformFeedbackVaryingsEXT is
4611 * non-zero, but the program object has no vertex or geometry
4614 if (first
>= MESA_SHADER_FRAGMENT
) {
4615 linker_error(prog
, "Transform feedback varyings specified, but "
4616 "no vertex, tessellation, or geometry shader is "
4621 tfeedback_decls
= rzalloc_array(mem_ctx
, tfeedback_decl
,
4622 num_tfeedback_decls
);
4623 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4624 varying_names
, tfeedback_decls
))
4628 /* If there is no fragment shader we need to set transform feedback.
4630 * For SSO we also need to assign output locations. We assign them here
4631 * because we need to do it for both single stage programs and multi stage
4634 if (last
< MESA_SHADER_FRAGMENT
&&
4635 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4636 const uint64_t reserved_out_slots
=
4637 reserved_varying_slot(prog
->_LinkedShaders
[last
], ir_var_shader_out
);
4638 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4639 prog
->_LinkedShaders
[last
], NULL
,
4640 num_tfeedback_decls
, tfeedback_decls
,
4641 reserved_out_slots
))
4645 if (last
<= MESA_SHADER_FRAGMENT
) {
4646 /* Remove unused varyings from the first/last stage unless SSO */
4647 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4648 prog
->_LinkedShaders
[first
],
4650 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4651 prog
->_LinkedShaders
[last
],
4654 /* If the program is made up of only a single stage */
4655 if (first
== last
) {
4656 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[last
];
4658 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4659 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4662 if (prog
->SeparateShader
) {
4663 const uint64_t reserved_slots
=
4664 reserved_varying_slot(sh
, ir_var_shader_in
);
4666 /* Assign input locations for SSO, output locations are already
4669 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4670 NULL
/* producer */,
4672 0 /* num_tfeedback_decls */,
4673 NULL
/* tfeedback_decls */,
4678 /* Linking the stages in the opposite order (from fragment to vertex)
4679 * ensures that inter-shader outputs written to in an earlier stage
4680 * are eliminated if they are (transitively) not used in a later
4684 for (int i
= next
- 1; i
>= 0; i
--) {
4685 if (prog
->_LinkedShaders
[i
] == NULL
&& i
!= 0)
4688 gl_linked_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4689 gl_linked_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4691 const uint64_t reserved_out_slots
=
4692 reserved_varying_slot(sh_i
, ir_var_shader_out
);
4693 const uint64_t reserved_in_slots
=
4694 reserved_varying_slot(sh_next
, ir_var_shader_in
);
4696 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4697 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4700 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4701 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4703 reserved_out_slots
| reserved_in_slots
))
4706 /* This must be done after all dead varyings are eliminated. */
4708 unsigned slots_used
= _mesa_bitcount_64(reserved_out_slots
);
4709 if (!check_against_output_limit(ctx
, prog
, sh_i
, slots_used
)) {
4714 unsigned slots_used
= _mesa_bitcount_64(reserved_in_slots
);
4715 if (!check_against_input_limit(ctx
, prog
, sh_next
, slots_used
))
4723 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4724 has_xfb_qualifiers
))
4727 update_array_sizes(prog
);
4728 link_assign_uniform_locations(prog
, ctx
, num_explicit_uniform_locs
);
4729 link_assign_atomic_counter_resources(ctx
, prog
);
4731 link_calculate_subroutine_compat(prog
);
4732 check_resources(ctx
, prog
);
4733 check_subroutine_resources(prog
);
4734 check_image_resources(ctx
, prog
);
4735 link_check_atomic_counter_resources(ctx
, prog
);
4737 if (!prog
->data
->LinkStatus
)
4740 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4741 if (prog
->_LinkedShaders
[i
] == NULL
)
4744 const struct gl_shader_compiler_options
*options
=
4745 &ctx
->Const
.ShaderCompilerOptions
[i
];
4747 if (options
->LowerBufferInterfaceBlocks
)
4748 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4749 options
->ClampBlockIndicesToArrayBounds
);
4751 if (i
== MESA_SHADER_COMPUTE
)
4752 lower_shared_reference(prog
->_LinkedShaders
[i
],
4753 &prog
->Comp
.SharedSize
);
4755 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4756 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4763 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4765 prog
->data
->LinkStatus
= true; /* All error paths will set this to false */
4766 prog
->data
->Validated
= false;
4768 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4770 * "Linking can fail for a variety of reasons as specified in the
4771 * OpenGL Shading Language Specification, as well as any of the
4772 * following reasons:
4774 * - No shader objects are attached to program."
4776 * The Compatibility Profile specification does not list the error. In
4777 * Compatibility Profile missing shader stages are replaced by
4778 * fixed-function. This applies to the case where all stages are
4781 if (prog
->NumShaders
== 0) {
4782 if (ctx
->API
!= API_OPENGL_COMPAT
)
4783 linker_error(prog
, "no shaders attached to the program\n");
4787 unsigned int num_explicit_uniform_locs
= 0;
4789 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4791 prog
->ARB_fragment_coord_conventions_enable
= false;
4793 /* Separate the shaders into groups based on their type.
4795 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4796 unsigned num_shaders
[MESA_SHADER_STAGES
];
4798 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4799 shader_list
[i
] = (struct gl_shader
**)
4800 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4804 unsigned min_version
= UINT_MAX
;
4805 unsigned max_version
= 0;
4806 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4807 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4808 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4810 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4811 linker_error(prog
, "all shaders must use same shading "
4812 "language version\n");
4816 if (prog
->Shaders
[i
]->info
.ARB_fragment_coord_conventions_enable
) {
4817 prog
->ARB_fragment_coord_conventions_enable
= true;
4820 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4821 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4822 num_shaders
[shader_type
]++;
4825 /* In desktop GLSL, different shader versions may be linked together. In
4826 * GLSL ES, all shader versions must be the same.
4828 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4829 linker_error(prog
, "all shaders must use same shading "
4830 "language version\n");
4834 prog
->data
->Version
= max_version
;
4835 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4837 /* Some shaders have to be linked with some other shaders present.
4839 if (!prog
->SeparateShader
) {
4840 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4841 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4842 linker_error(prog
, "Geometry shader must be linked with "
4846 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4847 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4848 linker_error(prog
, "Tessellation evaluation shader must be linked "
4849 "with vertex shader\n");
4852 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4853 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4854 linker_error(prog
, "Tessellation control shader must be linked with "
4859 /* The spec is self-contradictory here. It allows linking without a tess
4860 * eval shader, but that can only be used with transform feedback and
4861 * rasterization disabled. However, transform feedback isn't allowed
4862 * with GL_PATCHES, so it can't be used.
4864 * More investigation showed that the idea of transform feedback after
4865 * a tess control shader was dropped, because some hw vendors couldn't
4866 * support tessellation without a tess eval shader, but the linker
4867 * section wasn't updated to reflect that.
4869 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4872 * Do what's reasonable and always require a tess eval shader if a tess
4873 * control shader is present.
4875 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4876 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4877 linker_error(prog
, "Tessellation control shader must be linked with "
4878 "tessellation evaluation shader\n");
4883 /* Compute shaders have additional restrictions. */
4884 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4885 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4886 linker_error(prog
, "Compute shaders may not be linked with any other "
4887 "type of shader\n");
4890 /* Link all shaders for a particular stage and validate the result.
4892 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4893 if (num_shaders
[stage
] > 0) {
4894 gl_linked_shader
*const sh
=
4895 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4896 num_shaders
[stage
], false);
4898 if (!prog
->data
->LinkStatus
) {
4900 _mesa_delete_linked_shader(ctx
, sh
);
4905 case MESA_SHADER_VERTEX
:
4906 validate_vertex_shader_executable(prog
, sh
, ctx
);
4908 case MESA_SHADER_TESS_CTRL
:
4909 /* nothing to be done */
4911 case MESA_SHADER_TESS_EVAL
:
4912 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4914 case MESA_SHADER_GEOMETRY
:
4915 validate_geometry_shader_executable(prog
, sh
, ctx
);
4917 case MESA_SHADER_FRAGMENT
:
4918 validate_fragment_shader_executable(prog
, sh
);
4921 if (!prog
->data
->LinkStatus
) {
4923 _mesa_delete_linked_shader(ctx
, sh
);
4927 prog
->_LinkedShaders
[stage
] = sh
;
4928 prog
->data
->linked_stages
|= 1 << stage
;
4932 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4933 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4934 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4935 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4936 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4937 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4938 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4939 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4940 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4942 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4943 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4946 /* Here begins the inter-stage linking phase. Some initial validation is
4947 * performed, then locations are assigned for uniforms, attributes, and
4950 cross_validate_uniforms(prog
);
4951 if (!prog
->data
->LinkStatus
)
4954 unsigned first
, last
, prev
;
4956 first
= MESA_SHADER_STAGES
;
4959 /* Determine first and last stage. */
4960 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4961 if (!prog
->_LinkedShaders
[i
])
4963 if (first
== MESA_SHADER_STAGES
)
4968 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4969 link_assign_subroutine_types(prog
);
4971 if (!prog
->data
->LinkStatus
)
4974 resize_tes_inputs(ctx
, prog
);
4976 /* Validate the inputs of each stage with the output of the preceding
4980 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4981 if (prog
->_LinkedShaders
[i
] == NULL
)
4984 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4985 prog
->_LinkedShaders
[i
]);
4986 if (!prog
->data
->LinkStatus
)
4989 cross_validate_outputs_to_inputs(prog
,
4990 prog
->_LinkedShaders
[prev
],
4991 prog
->_LinkedShaders
[i
]);
4992 if (!prog
->data
->LinkStatus
)
4998 /* Cross-validate uniform blocks between shader stages */
4999 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
5000 if (!prog
->data
->LinkStatus
)
5003 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5004 if (prog
->_LinkedShaders
[i
] != NULL
)
5005 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
5008 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
5009 * it before optimization because we want most of the checks to get
5010 * dropped thanks to constant propagation.
5012 * This rule also applies to GLSL ES 3.00.
5014 if (max_version
>= (prog
->IsES
? 300 : 130)) {
5015 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
5017 lower_discard_flow(sh
->ir
);
5021 if (prog
->SeparateShader
)
5022 disable_varying_optimizations_for_sso(prog
);
5025 if (!interstage_cross_validate_uniform_blocks(prog
, false))
5029 if (!interstage_cross_validate_uniform_blocks(prog
, true))
5032 /* Do common optimization before assigning storage for attributes,
5033 * uniforms, and varyings. Later optimization could possibly make
5034 * some of that unused.
5036 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5037 if (prog
->_LinkedShaders
[i
] == NULL
)
5040 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
5041 if (!prog
->data
->LinkStatus
)
5044 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
5045 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
5048 if (ctx
->Const
.LowerTessLevel
) {
5049 lower_tess_level(prog
->_LinkedShaders
[i
]);
5052 if (ctx
->Const
.GLSLOptimizeConservatively
) {
5053 /* Run it just once. */
5054 do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
5055 &ctx
->Const
.ShaderCompilerOptions
[i
],
5056 ctx
->Const
.NativeIntegers
);
5058 /* Repeat it until it stops making changes. */
5059 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
5060 &ctx
->Const
.ShaderCompilerOptions
[i
],
5061 ctx
->Const
.NativeIntegers
))
5065 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
5066 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
5069 /* Validation for special cases where we allow sampler array indexing
5070 * with loop induction variable. This check emits a warning or error
5071 * depending if backend can handle dynamic indexing.
5073 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
5074 (prog
->IsES
&& prog
->data
->Version
< 300)) {
5075 if (!validate_sampler_array_indexing(ctx
, prog
))
5079 /* Check and validate stream emissions in geometry shaders */
5080 validate_geometry_shader_emissions(ctx
, prog
);
5082 store_fragdepth_layout(prog
);
5084 if(!link_varyings_and_uniforms(first
, last
, num_explicit_uniform_locs
, ctx
,
5088 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
5089 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
5090 * anything about shader linking when one of the shaders (vertex or
5091 * fragment shader) is absent. So, the extension shouldn't change the
5092 * behavior specified in GLSL specification.
5094 * From OpenGL ES 3.1 specification (7.3 Program Objects):
5095 * "Linking can fail for a variety of reasons as specified in the
5096 * OpenGL ES Shading Language Specification, as well as any of the
5097 * following reasons:
5101 * * program contains objects to form either a vertex shader or
5102 * fragment shader, and program is not separable, and does not
5103 * contain objects to form both a vertex shader and fragment
5106 * However, the only scenario in 3.1+ where we don't require them both is
5107 * when we have a compute shader. For example:
5109 * - No shaders is a link error.
5110 * - Geom or Tess without a Vertex shader is a link error which means we
5111 * always require a Vertex shader and hence a Fragment shader.
5112 * - Finally a Compute shader linked with any other stage is a link error.
5114 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
5115 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
5116 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
5117 linker_error(prog
, "program lacks a vertex shader\n");
5118 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5119 linker_error(prog
, "program lacks a fragment shader\n");
5124 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5125 free(shader_list
[i
]);
5126 if (prog
->_LinkedShaders
[i
] == NULL
)
5129 /* Do a final validation step to make sure that the IR wasn't
5130 * invalidated by any modifications performed after intrastage linking.
5132 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5134 /* Retain any live IR, but trash the rest. */
5135 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5137 /* The symbol table in the linked shaders may contain references to
5138 * variables that were removed (e.g., unused uniforms). Since it may
5139 * contain junk, there is no possible valid use. Delete it and set the
5142 delete prog
->_LinkedShaders
[i
]->symbols
;
5143 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5146 ralloc_free(mem_ctx
);