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 "main/core.h"
69 #include "glsl_symbol_table.h"
70 #include "glsl_parser_extras.h"
73 #include "program/hash_table.h"
75 #include "link_varyings.h"
76 #include "ir_optimization.h"
77 #include "ir_rvalue_visitor.h"
78 #include "ir_uniform.h"
80 #include "main/shaderobj.h"
81 #include "main/enums.h"
84 void linker_error(gl_shader_program
*, const char *, ...);
89 * Visitor that determines whether or not a variable is ever written.
91 class find_assignment_visitor
: public ir_hierarchical_visitor
{
93 find_assignment_visitor(const char *name
)
94 : name(name
), found(false)
99 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
101 ir_variable
*const var
= ir
->lhs
->variable_referenced();
103 if (strcmp(name
, var
->name
) == 0) {
108 return visit_continue_with_parent
;
111 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
113 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
114 actual_node
, &ir
->actual_parameters
) {
115 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
116 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
118 if (sig_param
->data
.mode
== ir_var_function_out
||
119 sig_param
->data
.mode
== ir_var_function_inout
) {
120 ir_variable
*var
= param_rval
->variable_referenced();
121 if (var
&& strcmp(name
, var
->name
) == 0) {
128 if (ir
->return_deref
!= NULL
) {
129 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
131 if (strcmp(name
, var
->name
) == 0) {
137 return visit_continue_with_parent
;
140 bool variable_found()
146 const char *name
; /**< Find writes to a variable with this name. */
147 bool found
; /**< Was a write to the variable found? */
152 * Visitor that determines whether or not a variable is ever read.
154 class find_deref_visitor
: public ir_hierarchical_visitor
{
156 find_deref_visitor(const char *name
)
157 : name(name
), found(false)
162 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
164 if (strcmp(this->name
, ir
->var
->name
) == 0) {
169 return visit_continue
;
172 bool variable_found() const
178 const char *name
; /**< Find writes to a variable with this name. */
179 bool found
; /**< Was a write to the variable found? */
183 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
185 unsigned num_vertices
;
186 gl_shader_program
*prog
;
188 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
190 this->num_vertices
= num_vertices
;
194 virtual ~geom_array_resize_visitor()
199 virtual ir_visitor_status
visit(ir_variable
*var
)
201 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
202 return visit_continue
;
204 unsigned size
= var
->type
->length
;
206 /* Generate a link error if the shader has declared this array with an
209 if (size
&& size
!= this->num_vertices
) {
210 linker_error(this->prog
, "size of array %s declared as %u, "
211 "but number of input vertices is %u\n",
212 var
->name
, size
, this->num_vertices
);
213 return visit_continue
;
216 /* Generate a link error if the shader attempts to access an input
217 * array using an index too large for its actual size assigned at link
220 if (var
->data
.max_array_access
>= this->num_vertices
) {
221 linker_error(this->prog
, "geometry shader accesses element %i of "
222 "%s, but only %i input vertices\n",
223 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
224 return visit_continue
;
227 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
229 var
->data
.max_array_access
= this->num_vertices
- 1;
231 return visit_continue
;
234 /* Dereferences of input variables need to be updated so that their type
235 * matches the newly assigned type of the variable they are accessing. */
236 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
238 ir
->type
= ir
->var
->type
;
239 return visit_continue
;
242 /* Dereferences of 2D input arrays need to be updated so that their type
243 * matches the newly assigned type of the array they are accessing. */
244 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
246 const glsl_type
*const vt
= ir
->array
->type
;
248 ir
->type
= vt
->fields
.array
;
249 return visit_continue
;
253 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
255 unsigned num_vertices
;
256 gl_shader_program
*prog
;
258 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
260 this->num_vertices
= num_vertices
;
264 virtual ~tess_eval_array_resize_visitor()
269 virtual ir_visitor_status
visit(ir_variable
*var
)
271 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
272 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
;
281 /* Dereferences of input variables need to be updated so that their type
282 * matches the newly assigned type of the variable they are accessing. */
283 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
285 ir
->type
= ir
->var
->type
;
286 return visit_continue
;
289 /* Dereferences of 2D input arrays need to be updated so that their type
290 * matches the newly assigned type of the array they are accessing. */
291 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
293 const glsl_type
*const vt
= ir
->array
->type
;
295 ir
->type
= vt
->fields
.array
;
296 return visit_continue
;
301 * Visitor that determines the highest stream id to which a (geometry) shader
302 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
304 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
306 find_emit_vertex_visitor(int max_allowed
)
307 : max_stream_allowed(max_allowed
),
308 invalid_stream_id(0),
309 invalid_stream_id_from_emit_vertex(false),
310 end_primitive_found(false),
311 uses_non_zero_stream(false)
316 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
318 int stream_id
= ir
->stream_id();
321 invalid_stream_id
= stream_id
;
322 invalid_stream_id_from_emit_vertex
= true;
326 if (stream_id
> max_stream_allowed
) {
327 invalid_stream_id
= stream_id
;
328 invalid_stream_id_from_emit_vertex
= true;
333 uses_non_zero_stream
= true;
335 return visit_continue
;
338 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
340 end_primitive_found
= true;
342 int stream_id
= ir
->stream_id();
345 invalid_stream_id
= stream_id
;
346 invalid_stream_id_from_emit_vertex
= false;
350 if (stream_id
> max_stream_allowed
) {
351 invalid_stream_id
= stream_id
;
352 invalid_stream_id_from_emit_vertex
= false;
357 uses_non_zero_stream
= true;
359 return visit_continue
;
364 return invalid_stream_id
!= 0;
367 const char *error_func()
369 return invalid_stream_id_from_emit_vertex
?
370 "EmitStreamVertex" : "EndStreamPrimitive";
375 return invalid_stream_id
;
380 return uses_non_zero_stream
;
383 bool uses_end_primitive()
385 return end_primitive_found
;
389 int max_stream_allowed
;
390 int invalid_stream_id
;
391 bool invalid_stream_id_from_emit_vertex
;
392 bool end_primitive_found
;
393 bool uses_non_zero_stream
;
396 /* Class that finds array derefs and check if indexes are dynamic. */
397 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
400 dynamic_sampler_array_indexing_visitor() :
401 dynamic_sampler_array_indexing(false)
405 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
407 if (!ir
->variable_referenced())
408 return visit_continue
;
410 if (!ir
->variable_referenced()->type
->contains_sampler())
411 return visit_continue
;
413 if (!ir
->array_index
->constant_expression_value()) {
414 dynamic_sampler_array_indexing
= true;
417 return visit_continue
;
420 bool uses_dynamic_sampler_array_indexing()
422 return dynamic_sampler_array_indexing
;
426 bool dynamic_sampler_array_indexing
;
429 } /* anonymous namespace */
432 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
436 ralloc_strcat(&prog
->InfoLog
, "error: ");
438 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
441 prog
->LinkStatus
= false;
446 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
450 ralloc_strcat(&prog
->InfoLog
, "warning: ");
452 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
459 * Given a string identifying a program resource, break it into a base name
460 * and an optional array index in square brackets.
462 * If an array index is present, \c out_base_name_end is set to point to the
463 * "[" that precedes the array index, and the array index itself is returned
466 * If no array index is present (or if the array index is negative or
467 * mal-formed), \c out_base_name_end, is set to point to the null terminator
468 * at the end of the input string, and -1 is returned.
470 * Only the final array index is parsed; if the string contains other array
471 * indices (or structure field accesses), they are left in the base name.
473 * No attempt is made to check that the base name is properly formed;
474 * typically the caller will look up the base name in a hash table, so
475 * ill-formed base names simply turn into hash table lookup failures.
478 parse_program_resource_name(const GLchar
*name
,
479 const GLchar
**out_base_name_end
)
481 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
483 * "When an integer array element or block instance number is part of
484 * the name string, it will be specified in decimal form without a "+"
485 * or "-" sign or any extra leading zeroes. Additionally, the name
486 * string will not include white space anywhere in the string."
489 const size_t len
= strlen(name
);
490 *out_base_name_end
= name
+ len
;
492 if (len
== 0 || name
[len
-1] != ']')
495 /* Walk backwards over the string looking for a non-digit character. This
496 * had better be the opening bracket for an array index.
498 * Initially, i specifies the location of the ']'. Since the string may
499 * contain only the ']' charcater, walk backwards very carefully.
502 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
505 if ((i
== 0) || name
[i
-1] != '[')
508 long array_index
= strtol(&name
[i
], NULL
, 10);
512 /* Check for leading zero */
513 if (name
[i
] == '0' && name
[i
+1] != ']')
516 *out_base_name_end
= name
+ (i
- 1);
522 link_invalidate_variable_locations(exec_list
*ir
)
524 foreach_in_list(ir_instruction
, node
, ir
) {
525 ir_variable
*const var
= node
->as_variable();
530 /* Only assign locations for variables that lack an explicit location.
531 * Explicit locations are set for all built-in variables, generic vertex
532 * shader inputs (via layout(location=...)), and generic fragment shader
533 * outputs (also via layout(location=...)).
535 if (!var
->data
.explicit_location
) {
536 var
->data
.location
= -1;
537 var
->data
.location_frac
= 0;
540 /* ir_variable::is_unmatched_generic_inout is used by the linker while
541 * connecting outputs from one stage to inputs of the next stage.
543 * There are two implicit assumptions here. First, we assume that any
544 * built-in variable (i.e., non-generic in or out) will have
545 * explicit_location set. Second, we assume that any generic in or out
546 * will not have explicit_location set.
548 * This second assumption will only be valid until
549 * GL_ARB_separate_shader_objects is supported. When that extension is
550 * implemented, this function will need some modifications.
552 if (!var
->data
.explicit_location
) {
553 var
->data
.is_unmatched_generic_inout
= 1;
555 var
->data
.is_unmatched_generic_inout
= 0;
562 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
564 * Also check for errors based on incorrect usage of gl_ClipVertex and
567 * Return false if an error was reported.
570 analyze_clip_usage(struct gl_shader_program
*prog
,
571 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
572 GLuint
*ClipDistanceArraySize
)
574 *ClipDistanceArraySize
= 0;
576 if (!prog
->IsES
&& prog
->Version
>= 130) {
577 /* From section 7.1 (Vertex Shader Special Variables) of the
580 * "It is an error for a shader to statically write both
581 * gl_ClipVertex and gl_ClipDistance."
583 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
584 * gl_ClipVertex nor gl_ClipDistance.
586 find_assignment_visitor
clip_vertex("gl_ClipVertex");
587 find_assignment_visitor
clip_distance("gl_ClipDistance");
589 clip_vertex
.run(shader
->ir
);
590 clip_distance
.run(shader
->ir
);
591 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
592 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
593 "and `gl_ClipDistance'\n",
594 _mesa_shader_stage_to_string(shader
->Stage
));
597 *UsesClipDistance
= clip_distance
.variable_found();
598 ir_variable
*clip_distance_var
=
599 shader
->symbols
->get_variable("gl_ClipDistance");
600 if (clip_distance_var
)
601 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
603 *UsesClipDistance
= false;
609 * Verify that a vertex shader executable meets all semantic requirements.
611 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
614 * \param shader Vertex shader executable to be verified
617 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
618 struct gl_shader
*shader
)
623 /* From the GLSL 1.10 spec, page 48:
625 * "The variable gl_Position is available only in the vertex
626 * language and is intended for writing the homogeneous vertex
627 * position. All executions of a well-formed vertex shader
628 * executable must write a value into this variable. [...] The
629 * variable gl_Position is available only in the vertex
630 * language and is intended for writing the homogeneous vertex
631 * position. All executions of a well-formed vertex shader
632 * executable must write a value into this variable."
634 * while in GLSL 1.40 this text is changed to:
636 * "The variable gl_Position is available only in the vertex
637 * language and is intended for writing the homogeneous vertex
638 * position. It can be written at any time during shader
639 * execution. It may also be read back by a vertex shader
640 * after being written. This value will be used by primitive
641 * assembly, clipping, culling, and other fixed functionality
642 * operations, if present, that operate on primitives after
643 * vertex processing has occurred. Its value is undefined if
644 * the vertex shader executable does not write gl_Position."
646 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
647 * gl_Position is not an error.
649 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
650 find_assignment_visitor
find("gl_Position");
651 find
.run(shader
->ir
);
652 if (!find
.variable_found()) {
655 "vertex shader does not write to `gl_Position'."
656 "It's value is undefined. \n");
659 "vertex shader does not write to `gl_Position'. \n");
665 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
666 &prog
->Vert
.ClipDistanceArraySize
);
670 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
671 struct gl_shader
*shader
)
676 analyze_clip_usage(prog
, shader
, &prog
->TessEval
.UsesClipDistance
,
677 &prog
->TessEval
.ClipDistanceArraySize
);
682 * Verify that a fragment shader executable meets all semantic requirements
684 * \param shader Fragment shader executable to be verified
687 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
688 struct gl_shader
*shader
)
693 find_assignment_visitor
frag_color("gl_FragColor");
694 find_assignment_visitor
frag_data("gl_FragData");
696 frag_color
.run(shader
->ir
);
697 frag_data
.run(shader
->ir
);
699 if (frag_color
.variable_found() && frag_data
.variable_found()) {
700 linker_error(prog
, "fragment shader writes to both "
701 "`gl_FragColor' and `gl_FragData'\n");
706 * Verify that a geometry shader executable meets all semantic requirements
708 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
709 * prog->Geom.ClipDistanceArraySize as a side effect.
711 * \param shader Geometry shader executable to be verified
714 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
715 struct gl_shader
*shader
)
720 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
721 prog
->Geom
.VerticesIn
= num_vertices
;
723 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
724 &prog
->Geom
.ClipDistanceArraySize
);
728 * Check if geometry shaders emit to non-zero streams and do corresponding
732 validate_geometry_shader_emissions(struct gl_context
*ctx
,
733 struct gl_shader_program
*prog
)
735 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
736 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
737 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
738 if (emit_vertex
.error()) {
739 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
740 "stream parameter are in the range [0, %d].\n",
741 emit_vertex
.error_func(),
742 emit_vertex
.error_stream(),
743 ctx
->Const
.MaxVertexStreams
- 1);
745 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
746 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
748 /* From the ARB_gpu_shader5 spec:
750 * "Multiple vertex streams are supported only if the output primitive
751 * type is declared to be "points". A program will fail to link if it
752 * contains a geometry shader calling EmitStreamVertex() or
753 * EndStreamPrimitive() if its output primitive type is not "points".
755 * However, in the same spec:
757 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
758 * with <stream> set to zero."
762 * "The function EndPrimitive() is equivalent to calling
763 * EndStreamPrimitive() with <stream> set to zero."
765 * Since we can call EmitVertex() and EndPrimitive() when we output
766 * primitives other than points, calling EmitStreamVertex(0) or
767 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
768 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
769 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
772 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
773 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
774 "with n>0 requires point output\n");
780 validate_intrastage_arrays(struct gl_shader_program
*prog
,
781 ir_variable
*const var
,
782 ir_variable
*const existing
)
784 /* Consider the types to be "the same" if both types are arrays
785 * of the same type and one of the arrays is implicitly sized.
786 * In addition, set the type of the linked variable to the
787 * explicitly sized array.
789 if (var
->type
->is_array() && existing
->type
->is_array() &&
790 (var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
791 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
792 if (var
->type
->length
!= 0) {
793 if (var
->type
->length
<= existing
->data
.max_array_access
) {
794 linker_error(prog
, "%s `%s' declared as type "
795 "`%s' but outermost dimension has an index"
798 var
->name
, var
->type
->name
,
799 existing
->data
.max_array_access
);
801 existing
->type
= var
->type
;
803 } else if (existing
->type
->length
!= 0) {
804 if(existing
->type
->length
<= var
->data
.max_array_access
) {
805 linker_error(prog
, "%s `%s' declared as type "
806 "`%s' but outermost dimension has an index"
809 var
->name
, existing
->type
->name
,
810 var
->data
.max_array_access
);
820 * Perform validation of global variables used across multiple shaders
823 cross_validate_globals(struct gl_shader_program
*prog
,
824 struct gl_shader
**shader_list
,
825 unsigned num_shaders
,
828 /* Examine all of the uniforms in all of the shaders and cross validate
831 glsl_symbol_table variables
;
832 for (unsigned i
= 0; i
< num_shaders
; i
++) {
833 if (shader_list
[i
] == NULL
)
836 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
837 ir_variable
*const var
= node
->as_variable();
842 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
845 /* Don't cross validate temporaries that are at global scope. These
846 * will eventually get pulled into the shaders 'main'.
848 if (var
->data
.mode
== ir_var_temporary
)
851 /* If a global with this name has already been seen, verify that the
852 * new instance has the same type. In addition, if the globals have
853 * initializers, the values of the initializers must be the same.
855 ir_variable
*const existing
= variables
.get_variable(var
->name
);
856 if (existing
!= NULL
) {
857 /* Check if types match. Interface blocks have some special
858 * rules so we handle those elsewhere.
860 if (var
->type
!= existing
->type
&&
861 !var
->is_interface_instance()) {
862 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
863 if (var
->type
->is_record() && existing
->type
->is_record()
864 && existing
->type
->record_compare(var
->type
)) {
865 existing
->type
= var
->type
;
867 linker_error(prog
, "%s `%s' declared as type "
868 "`%s' and type `%s'\n",
870 var
->name
, var
->type
->name
,
871 existing
->type
->name
);
877 if (var
->data
.explicit_location
) {
878 if (existing
->data
.explicit_location
879 && (var
->data
.location
!= existing
->data
.location
)) {
880 linker_error(prog
, "explicit locations for %s "
881 "`%s' have differing values\n",
882 mode_string(var
), var
->name
);
886 existing
->data
.location
= var
->data
.location
;
887 existing
->data
.explicit_location
= true;
890 /* From the GLSL 4.20 specification:
891 * "A link error will result if two compilation units in a program
892 * specify different integer-constant bindings for the same
893 * opaque-uniform name. However, it is not an error to specify a
894 * binding on some but not all declarations for the same name"
896 if (var
->data
.explicit_binding
) {
897 if (existing
->data
.explicit_binding
&&
898 var
->data
.binding
!= existing
->data
.binding
) {
899 linker_error(prog
, "explicit bindings for %s "
900 "`%s' have differing values\n",
901 mode_string(var
), var
->name
);
905 existing
->data
.binding
= var
->data
.binding
;
906 existing
->data
.explicit_binding
= true;
909 if (var
->type
->contains_atomic() &&
910 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
911 linker_error(prog
, "offset specifications for %s "
912 "`%s' have differing values\n",
913 mode_string(var
), var
->name
);
917 /* Validate layout qualifiers for gl_FragDepth.
919 * From the AMD/ARB_conservative_depth specs:
921 * "If gl_FragDepth is redeclared in any fragment shader in a
922 * program, it must be redeclared in all fragment shaders in
923 * that program that have static assignments to
924 * gl_FragDepth. All redeclarations of gl_FragDepth in all
925 * fragment shaders in a single program must have the same set
928 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
929 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
930 bool layout_differs
=
931 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
933 if (layout_declared
&& layout_differs
) {
935 "All redeclarations of gl_FragDepth in all "
936 "fragment shaders in a single program must have "
937 "the same set of qualifiers.\n");
940 if (var
->data
.used
&& layout_differs
) {
942 "If gl_FragDepth is redeclared with a layout "
943 "qualifier in any fragment shader, it must be "
944 "redeclared with the same layout qualifier in "
945 "all fragment shaders that have assignments to "
950 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
952 * "If a shared global has multiple initializers, the
953 * initializers must all be constant expressions, and they
954 * must all have the same value. Otherwise, a link error will
955 * result. (A shared global having only one initializer does
956 * not require that initializer to be a constant expression.)"
958 * Previous to 4.20 the GLSL spec simply said that initializers
959 * must have the same value. In this case of non-constant
960 * initializers, this was impossible to determine. As a result,
961 * no vendor actually implemented that behavior. The 4.20
962 * behavior matches the implemented behavior of at least one other
963 * vendor, so we'll implement that for all GLSL versions.
965 if (var
->constant_initializer
!= NULL
) {
966 if (existing
->constant_initializer
!= NULL
) {
967 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
968 linker_error(prog
, "initializers for %s "
969 "`%s' have differing values\n",
970 mode_string(var
), var
->name
);
974 /* If the first-seen instance of a particular uniform did not
975 * have an initializer but a later instance does, copy the
976 * initializer to the version stored in the symbol table.
978 /* FINISHME: This is wrong. The constant_value field should
979 * FINISHME: not be modified! Imagine a case where a shader
980 * FINISHME: without an initializer is linked in two different
981 * FINISHME: programs with shaders that have differing
982 * FINISHME: initializers. Linking with the first will
983 * FINISHME: modify the shader, and linking with the second
984 * FINISHME: will fail.
986 existing
->constant_initializer
=
987 var
->constant_initializer
->clone(ralloc_parent(existing
),
992 if (var
->data
.has_initializer
) {
993 if (existing
->data
.has_initializer
994 && (var
->constant_initializer
== NULL
995 || existing
->constant_initializer
== NULL
)) {
997 "shared global variable `%s' has multiple "
998 "non-constant initializers.\n",
1003 /* Some instance had an initializer, so keep track of that. In
1004 * this location, all sorts of initializers (constant or
1005 * otherwise) will propagate the existence to the variable
1006 * stored in the symbol table.
1008 existing
->data
.has_initializer
= true;
1011 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1012 linker_error(prog
, "declarations for %s `%s' have "
1013 "mismatching invariant qualifiers\n",
1014 mode_string(var
), var
->name
);
1017 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1018 linker_error(prog
, "declarations for %s `%s' have "
1019 "mismatching centroid qualifiers\n",
1020 mode_string(var
), var
->name
);
1023 if (existing
->data
.sample
!= var
->data
.sample
) {
1024 linker_error(prog
, "declarations for %s `%s` have "
1025 "mismatching sample qualifiers\n",
1026 mode_string(var
), var
->name
);
1030 variables
.add_variable(var
);
1037 * Perform validation of uniforms used across multiple shader stages
1040 cross_validate_uniforms(struct gl_shader_program
*prog
)
1042 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1043 MESA_SHADER_STAGES
, true);
1047 * Accumulates the array of prog->UniformBlocks and checks that all
1048 * definitons of blocks agree on their contents.
1051 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1053 unsigned max_num_uniform_blocks
= 0;
1054 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1055 if (prog
->_LinkedShaders
[i
])
1056 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1059 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1060 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1062 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
1063 max_num_uniform_blocks
);
1064 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1065 prog
->UniformBlockStageIndex
[i
][j
] = -1;
1070 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
1071 int index
= link_cross_validate_uniform_block(prog
,
1072 &prog
->UniformBlocks
,
1073 &prog
->NumUniformBlocks
,
1074 &sh
->UniformBlocks
[j
]);
1077 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1078 sh
->UniformBlocks
[j
].Name
);
1082 prog
->UniformBlockStageIndex
[i
][index
] = j
;
1091 * Populates a shaders symbol table with all global declarations
1094 populate_symbol_table(gl_shader
*sh
)
1096 sh
->symbols
= new(sh
) glsl_symbol_table
;
1098 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1102 if ((func
= inst
->as_function()) != NULL
) {
1103 sh
->symbols
->add_function(func
);
1104 } else if ((var
= inst
->as_variable()) != NULL
) {
1105 if (var
->data
.mode
!= ir_var_temporary
)
1106 sh
->symbols
->add_variable(var
);
1113 * Remap variables referenced in an instruction tree
1115 * This is used when instruction trees are cloned from one shader and placed in
1116 * another. These trees will contain references to \c ir_variable nodes that
1117 * do not exist in the target shader. This function finds these \c ir_variable
1118 * references and replaces the references with matching variables in the target
1121 * If there is no matching variable in the target shader, a clone of the
1122 * \c ir_variable is made and added to the target shader. The new variable is
1123 * added to \b both the instruction stream and the symbol table.
1125 * \param inst IR tree that is to be processed.
1126 * \param symbols Symbol table containing global scope symbols in the
1128 * \param instructions Instruction stream where new variable declarations
1132 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1135 class remap_visitor
: public ir_hierarchical_visitor
{
1137 remap_visitor(struct gl_shader
*target
,
1140 this->target
= target
;
1141 this->symbols
= target
->symbols
;
1142 this->instructions
= target
->ir
;
1143 this->temps
= temps
;
1146 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1148 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1149 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1151 assert(var
!= NULL
);
1153 return visit_continue
;
1156 ir_variable
*const existing
=
1157 this->symbols
->get_variable(ir
->var
->name
);
1158 if (existing
!= NULL
)
1161 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1163 this->symbols
->add_variable(copy
);
1164 this->instructions
->push_head(copy
);
1168 return visit_continue
;
1172 struct gl_shader
*target
;
1173 glsl_symbol_table
*symbols
;
1174 exec_list
*instructions
;
1178 remap_visitor
v(target
, temps
);
1185 * Move non-declarations from one instruction stream to another
1187 * The intended usage pattern of this function is to pass the pointer to the
1188 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1189 * pointer) for \c last and \c false for \c make_copies on the first
1190 * call. Successive calls pass the return value of the previous call for
1191 * \c last and \c true for \c make_copies.
1193 * \param instructions Source instruction stream
1194 * \param last Instruction after which new instructions should be
1195 * inserted in the target instruction stream
1196 * \param make_copies Flag selecting whether instructions in \c instructions
1197 * should be copied (via \c ir_instruction::clone) into the
1198 * target list or moved.
1201 * The new "last" instruction in the target instruction stream. This pointer
1202 * is suitable for use as the \c last parameter of a later call to this
1206 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1207 bool make_copies
, gl_shader
*target
)
1209 hash_table
*temps
= NULL
;
1212 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1213 hash_table_pointer_compare
);
1215 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1216 if (inst
->as_function())
1219 ir_variable
*var
= inst
->as_variable();
1220 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1223 assert(inst
->as_assignment()
1225 || inst
->as_if() /* for initializers with the ?: operator */
1226 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1229 inst
= inst
->clone(target
, NULL
);
1232 hash_table_insert(temps
, inst
, var
);
1234 remap_variables(inst
, target
, temps
);
1239 last
->insert_after(inst
);
1244 hash_table_dtor(temps
);
1250 * Get the function signature for main from a shader
1252 ir_function_signature
*
1253 link_get_main_function_signature(gl_shader
*sh
)
1255 ir_function
*const f
= sh
->symbols
->get_function("main");
1257 exec_list void_parameters
;
1259 /* Look for the 'void main()' signature and ensure that it's defined.
1260 * This keeps the linker from accidentally pick a shader that just
1261 * contains a prototype for main.
1263 * We don't have to check for multiple definitions of main (in multiple
1264 * shaders) because that would have already been caught above.
1266 ir_function_signature
*sig
=
1267 f
->matching_signature(NULL
, &void_parameters
, false);
1268 if ((sig
!= NULL
) && sig
->is_defined
) {
1278 * This class is only used in link_intrastage_shaders() below but declaring
1279 * it inside that function leads to compiler warnings with some versions of
1282 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1284 array_sizing_visitor()
1285 : mem_ctx(ralloc_context(NULL
)),
1286 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1287 hash_table_pointer_compare
))
1291 ~array_sizing_visitor()
1293 hash_table_dtor(this->unnamed_interfaces
);
1294 ralloc_free(this->mem_ctx
);
1297 virtual ir_visitor_status
visit(ir_variable
*var
)
1299 fixup_type(&var
->type
, var
->data
.max_array_access
);
1300 if (var
->type
->is_interface()) {
1301 if (interface_contains_unsized_arrays(var
->type
)) {
1302 const glsl_type
*new_type
=
1303 resize_interface_members(var
->type
,
1304 var
->get_max_ifc_array_access());
1305 var
->type
= new_type
;
1306 var
->change_interface_type(new_type
);
1308 } else if (var
->type
->is_array() &&
1309 var
->type
->fields
.array
->is_interface()) {
1310 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1311 const glsl_type
*new_type
=
1312 resize_interface_members(var
->type
->fields
.array
,
1313 var
->get_max_ifc_array_access());
1314 var
->change_interface_type(new_type
);
1315 var
->type
= update_interface_members_array(var
->type
, new_type
);
1317 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1318 /* Store a pointer to the variable in the unnamed_interfaces
1321 ir_variable
**interface_vars
= (ir_variable
**)
1322 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1323 if (interface_vars
== NULL
) {
1324 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1326 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1329 unsigned index
= ifc_type
->field_index(var
->name
);
1330 assert(index
< ifc_type
->length
);
1331 assert(interface_vars
[index
] == NULL
);
1332 interface_vars
[index
] = var
;
1334 return visit_continue
;
1338 * For each unnamed interface block that was discovered while running the
1339 * visitor, adjust the interface type to reflect the newly assigned array
1340 * sizes, and fix up the ir_variable nodes to point to the new interface
1343 void fixup_unnamed_interface_types()
1345 hash_table_call_foreach(this->unnamed_interfaces
,
1346 fixup_unnamed_interface_type
, NULL
);
1351 * If the type pointed to by \c type represents an unsized array, replace
1352 * it with a sized array whose size is determined by max_array_access.
1354 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1356 if ((*type
)->is_unsized_array()) {
1357 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1358 max_array_access
+ 1);
1359 assert(*type
!= NULL
);
1363 static const glsl_type
*
1364 update_interface_members_array(const glsl_type
*type
,
1365 const glsl_type
*new_interface_type
)
1367 const glsl_type
*element_type
= type
->fields
.array
;
1368 if (element_type
->is_array()) {
1369 const glsl_type
*new_array_type
=
1370 update_interface_members_array(element_type
, new_interface_type
);
1371 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1373 return glsl_type::get_array_instance(new_interface_type
,
1379 * Determine whether the given interface type contains unsized arrays (if
1380 * it doesn't, array_sizing_visitor doesn't need to process it).
1382 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1384 for (unsigned i
= 0; i
< type
->length
; i
++) {
1385 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1386 if (elem_type
->is_unsized_array())
1393 * Create a new interface type based on the given type, with unsized arrays
1394 * replaced by sized arrays whose size is determined by
1395 * max_ifc_array_access.
1397 static const glsl_type
*
1398 resize_interface_members(const glsl_type
*type
,
1399 const unsigned *max_ifc_array_access
)
1401 unsigned num_fields
= type
->length
;
1402 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1403 memcpy(fields
, type
->fields
.structure
,
1404 num_fields
* sizeof(*fields
));
1405 for (unsigned i
= 0; i
< num_fields
; i
++) {
1406 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1408 glsl_interface_packing packing
=
1409 (glsl_interface_packing
) type
->interface_packing
;
1410 const glsl_type
*new_ifc_type
=
1411 glsl_type::get_interface_instance(fields
, num_fields
,
1412 packing
, type
->name
);
1414 return new_ifc_type
;
1417 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1420 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1421 ir_variable
**interface_vars
= (ir_variable
**) data
;
1422 unsigned num_fields
= ifc_type
->length
;
1423 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1424 memcpy(fields
, ifc_type
->fields
.structure
,
1425 num_fields
* sizeof(*fields
));
1426 bool interface_type_changed
= false;
1427 for (unsigned i
= 0; i
< num_fields
; i
++) {
1428 if (interface_vars
[i
] != NULL
&&
1429 fields
[i
].type
!= interface_vars
[i
]->type
) {
1430 fields
[i
].type
= interface_vars
[i
]->type
;
1431 interface_type_changed
= true;
1434 if (!interface_type_changed
) {
1438 glsl_interface_packing packing
=
1439 (glsl_interface_packing
) ifc_type
->interface_packing
;
1440 const glsl_type
*new_ifc_type
=
1441 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1444 for (unsigned i
= 0; i
< num_fields
; i
++) {
1445 if (interface_vars
[i
] != NULL
)
1446 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1451 * Memory context used to allocate the data in \c unnamed_interfaces.
1456 * Hash table from const glsl_type * to an array of ir_variable *'s
1457 * pointing to the ir_variables constituting each unnamed interface block.
1459 hash_table
*unnamed_interfaces
;
1464 * Performs the cross-validation of tessellation control shader vertices and
1465 * layout qualifiers for the attached tessellation control shaders,
1466 * and propagates them to the linked TCS and linked shader program.
1469 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1470 struct gl_shader
*linked_shader
,
1471 struct gl_shader
**shader_list
,
1472 unsigned num_shaders
)
1474 linked_shader
->TessCtrl
.VerticesOut
= 0;
1476 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1479 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1481 * "All tessellation control shader layout declarations in a program
1482 * must specify the same output patch vertex count. There must be at
1483 * least one layout qualifier specifying an output patch vertex count
1484 * in any program containing tessellation control shaders; however,
1485 * such a declaration is not required in all tessellation control
1489 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1490 struct gl_shader
*shader
= shader_list
[i
];
1492 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1493 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1494 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1495 linker_error(prog
, "tessellation control shader defined with "
1496 "conflicting output vertex count (%d and %d)\n",
1497 linked_shader
->TessCtrl
.VerticesOut
,
1498 shader
->TessCtrl
.VerticesOut
);
1501 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1505 /* Just do the intrastage -> interstage propagation right now,
1506 * since we already know we're in the right type of shader program
1509 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1510 linker_error(prog
, "tessellation control shader didn't declare "
1511 "vertices out layout qualifier\n");
1514 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1519 * Performs the cross-validation of tessellation evaluation shader
1520 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1521 * for the attached tessellation evaluation shaders, and propagates them
1522 * to the linked TES and linked shader program.
1525 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1526 struct gl_shader
*linked_shader
,
1527 struct gl_shader
**shader_list
,
1528 unsigned num_shaders
)
1530 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1531 linked_shader
->TessEval
.Spacing
= 0;
1532 linked_shader
->TessEval
.VertexOrder
= 0;
1533 linked_shader
->TessEval
.PointMode
= -1;
1535 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1538 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1540 * "At least one tessellation evaluation shader (compilation unit) in
1541 * a program must declare a primitive mode in its input layout.
1542 * Declaration vertex spacing, ordering, and point mode identifiers is
1543 * optional. It is not required that all tessellation evaluation
1544 * shaders in a program declare a primitive mode. If spacing or
1545 * vertex ordering declarations are omitted, the tessellation
1546 * primitive generator will use equal spacing or counter-clockwise
1547 * vertex ordering, respectively. If a point mode declaration is
1548 * omitted, the tessellation primitive generator will produce lines or
1549 * triangles according to the primitive mode."
1552 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1553 struct gl_shader
*shader
= shader_list
[i
];
1555 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1556 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1557 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1558 linker_error(prog
, "tessellation evaluation shader defined with "
1559 "conflicting input primitive modes.\n");
1562 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1565 if (shader
->TessEval
.Spacing
!= 0) {
1566 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1567 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1568 linker_error(prog
, "tessellation evaluation shader defined with "
1569 "conflicting vertex spacing.\n");
1572 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1575 if (shader
->TessEval
.VertexOrder
!= 0) {
1576 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1577 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1578 linker_error(prog
, "tessellation evaluation shader defined with "
1579 "conflicting ordering.\n");
1582 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1585 if (shader
->TessEval
.PointMode
!= -1) {
1586 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1587 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1588 linker_error(prog
, "tessellation evaluation shader defined with "
1589 "conflicting point modes.\n");
1592 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1597 /* Just do the intrastage -> interstage propagation right now,
1598 * since we already know we're in the right type of shader program
1601 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1603 "tessellation evaluation shader didn't declare input "
1604 "primitive modes.\n");
1607 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1609 if (linked_shader
->TessEval
.Spacing
== 0)
1610 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1611 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1613 if (linked_shader
->TessEval
.VertexOrder
== 0)
1614 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1615 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1617 if (linked_shader
->TessEval
.PointMode
== -1)
1618 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1619 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1624 * Performs the cross-validation of layout qualifiers specified in
1625 * redeclaration of gl_FragCoord for the attached fragment shaders,
1626 * and propagates them to the linked FS and linked shader program.
1629 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1630 struct gl_shader
*linked_shader
,
1631 struct gl_shader
**shader_list
,
1632 unsigned num_shaders
)
1634 linked_shader
->redeclares_gl_fragcoord
= false;
1635 linked_shader
->uses_gl_fragcoord
= false;
1636 linked_shader
->origin_upper_left
= false;
1637 linked_shader
->pixel_center_integer
= false;
1639 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1640 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1643 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1644 struct gl_shader
*shader
= shader_list
[i
];
1645 /* From the GLSL 1.50 spec, page 39:
1647 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1648 * it must be redeclared in all the fragment shaders in that program
1649 * that have a static use gl_FragCoord."
1651 if ((linked_shader
->redeclares_gl_fragcoord
1652 && !shader
->redeclares_gl_fragcoord
1653 && shader
->uses_gl_fragcoord
)
1654 || (shader
->redeclares_gl_fragcoord
1655 && !linked_shader
->redeclares_gl_fragcoord
1656 && linked_shader
->uses_gl_fragcoord
)) {
1657 linker_error(prog
, "fragment shader defined with conflicting "
1658 "layout qualifiers for gl_FragCoord\n");
1661 /* From the GLSL 1.50 spec, page 39:
1663 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1664 * single program must have the same set of qualifiers."
1666 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1667 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1668 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1669 linker_error(prog
, "fragment shader defined with conflicting "
1670 "layout qualifiers for gl_FragCoord\n");
1673 /* Update the linked shader state. Note that uses_gl_fragcoord should
1674 * accumulate the results. The other values should replace. If there
1675 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1676 * are already known to be the same.
1678 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1679 linked_shader
->redeclares_gl_fragcoord
=
1680 shader
->redeclares_gl_fragcoord
;
1681 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1682 || shader
->uses_gl_fragcoord
;
1683 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1684 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1687 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1692 * Performs the cross-validation of geometry shader max_vertices and
1693 * primitive type layout qualifiers for the attached geometry shaders,
1694 * and propagates them to the linked GS and linked shader program.
1697 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1698 struct gl_shader
*linked_shader
,
1699 struct gl_shader
**shader_list
,
1700 unsigned num_shaders
)
1702 linked_shader
->Geom
.VerticesOut
= 0;
1703 linked_shader
->Geom
.Invocations
= 0;
1704 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1705 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1707 /* No in/out qualifiers defined for anything but GLSL 1.50+
1708 * geometry shaders so far.
1710 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1713 /* From the GLSL 1.50 spec, page 46:
1715 * "All geometry shader output layout declarations in a program
1716 * must declare the same layout and same value for
1717 * max_vertices. There must be at least one geometry output
1718 * layout declaration somewhere in a program, but not all
1719 * geometry shaders (compilation units) are required to
1723 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1724 struct gl_shader
*shader
= shader_list
[i
];
1726 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1727 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1728 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1729 linker_error(prog
, "geometry shader defined with conflicting "
1733 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1736 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1737 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1738 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1739 linker_error(prog
, "geometry shader defined with conflicting "
1743 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1746 if (shader
->Geom
.VerticesOut
!= 0) {
1747 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1748 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1749 linker_error(prog
, "geometry shader defined with conflicting "
1750 "output vertex count (%d and %d)\n",
1751 linked_shader
->Geom
.VerticesOut
,
1752 shader
->Geom
.VerticesOut
);
1755 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1758 if (shader
->Geom
.Invocations
!= 0) {
1759 if (linked_shader
->Geom
.Invocations
!= 0 &&
1760 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1761 linker_error(prog
, "geometry shader defined with conflicting "
1762 "invocation count (%d and %d)\n",
1763 linked_shader
->Geom
.Invocations
,
1764 shader
->Geom
.Invocations
);
1767 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1771 /* Just do the intrastage -> interstage propagation right now,
1772 * since we already know we're in the right type of shader program
1775 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1777 "geometry shader didn't declare primitive input type\n");
1780 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1782 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1784 "geometry shader didn't declare primitive output type\n");
1787 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1789 if (linked_shader
->Geom
.VerticesOut
== 0) {
1791 "geometry shader didn't declare max_vertices\n");
1794 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1796 if (linked_shader
->Geom
.Invocations
== 0)
1797 linked_shader
->Geom
.Invocations
= 1;
1799 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1804 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1805 * qualifiers for the attached compute shaders, and propagate them to the
1806 * linked CS and linked shader program.
1809 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1810 struct gl_shader
*linked_shader
,
1811 struct gl_shader
**shader_list
,
1812 unsigned num_shaders
)
1814 for (int i
= 0; i
< 3; i
++)
1815 linked_shader
->Comp
.LocalSize
[i
] = 0;
1817 /* This function is called for all shader stages, but it only has an effect
1818 * for compute shaders.
1820 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1823 /* From the ARB_compute_shader spec, in the section describing local size
1826 * If multiple compute shaders attached to a single program object
1827 * declare local work-group size, the declarations must be identical;
1828 * otherwise a link-time error results. Furthermore, if a program
1829 * object contains any compute shaders, at least one must contain an
1830 * input layout qualifier specifying the local work sizes of the
1831 * program, or a link-time error will occur.
1833 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1834 struct gl_shader
*shader
= shader_list
[sh
];
1836 if (shader
->Comp
.LocalSize
[0] != 0) {
1837 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1838 for (int i
= 0; i
< 3; i
++) {
1839 if (linked_shader
->Comp
.LocalSize
[i
] !=
1840 shader
->Comp
.LocalSize
[i
]) {
1841 linker_error(prog
, "compute shader defined with conflicting "
1847 for (int i
= 0; i
< 3; i
++)
1848 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1852 /* Just do the intrastage -> interstage propagation right now,
1853 * since we already know we're in the right type of shader program
1856 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1857 linker_error(prog
, "compute shader didn't declare local size\n");
1860 for (int i
= 0; i
< 3; i
++)
1861 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1866 * Combine a group of shaders for a single stage to generate a linked shader
1869 * If this function is supplied a single shader, it is cloned, and the new
1870 * shader is returned.
1872 static struct gl_shader
*
1873 link_intrastage_shaders(void *mem_ctx
,
1874 struct gl_context
*ctx
,
1875 struct gl_shader_program
*prog
,
1876 struct gl_shader
**shader_list
,
1877 unsigned num_shaders
)
1879 struct gl_uniform_block
*uniform_blocks
= NULL
;
1881 /* Check that global variables defined in multiple shaders are consistent.
1883 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1884 if (!prog
->LinkStatus
)
1887 /* Check that interface blocks defined in multiple shaders are consistent.
1889 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1891 if (!prog
->LinkStatus
)
1894 /* Link up uniform blocks defined within this stage. */
1895 const unsigned num_uniform_blocks
=
1896 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1898 if (!prog
->LinkStatus
)
1901 /* Check that there is only a single definition of each function signature
1902 * across all shaders.
1904 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1905 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1906 ir_function
*const f
= node
->as_function();
1911 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1912 ir_function
*const other
=
1913 shader_list
[j
]->symbols
->get_function(f
->name
);
1915 /* If the other shader has no function (and therefore no function
1916 * signatures) with the same name, skip to the next shader.
1921 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1922 if (!sig
->is_defined
|| sig
->is_builtin())
1925 ir_function_signature
*other_sig
=
1926 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1928 if ((other_sig
!= NULL
) && other_sig
->is_defined
1929 && !other_sig
->is_builtin()) {
1930 linker_error(prog
, "function `%s' is multiply defined\n",
1939 /* Find the shader that defines main, and make a clone of it.
1941 * Starting with the clone, search for undefined references. If one is
1942 * found, find the shader that defines it. Clone the reference and add
1943 * it to the shader. Repeat until there are no undefined references or
1944 * until a reference cannot be resolved.
1946 gl_shader
*main
= NULL
;
1947 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1948 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1949 main
= shader_list
[i
];
1955 linker_error(prog
, "%s shader lacks `main'\n",
1956 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1960 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1961 linked
->ir
= new(linked
) exec_list
;
1962 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1964 linked
->UniformBlocks
= uniform_blocks
;
1965 linked
->NumUniformBlocks
= num_uniform_blocks
;
1966 ralloc_steal(linked
, linked
->UniformBlocks
);
1968 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1969 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1970 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1971 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1972 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1974 populate_symbol_table(linked
);
1976 /* The pointer to the main function in the final linked shader (i.e., the
1977 * copy of the original shader that contained the main function).
1979 ir_function_signature
*const main_sig
=
1980 link_get_main_function_signature(linked
);
1982 /* Move any instructions other than variable declarations or function
1983 * declarations into main.
1985 exec_node
*insertion_point
=
1986 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1989 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1990 if (shader_list
[i
] == main
)
1993 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1994 insertion_point
, true, linked
);
1997 /* Check if any shader needs built-in functions. */
1998 bool need_builtins
= false;
1999 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2000 if (shader_list
[i
]->uses_builtin_functions
) {
2001 need_builtins
= true;
2007 if (need_builtins
) {
2008 /* Make a temporary array one larger than shader_list, which will hold
2009 * the built-in function shader as well.
2011 gl_shader
**linking_shaders
= (gl_shader
**)
2012 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2014 ok
= linking_shaders
!= NULL
;
2017 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2018 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2020 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2022 free(linking_shaders
);
2024 _mesa_error_no_memory(__func__
);
2027 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2032 ctx
->Driver
.DeleteShader(ctx
, linked
);
2036 /* At this point linked should contain all of the linked IR, so
2037 * validate it to make sure nothing went wrong.
2039 validate_ir_tree(linked
->ir
);
2041 /* Set the size of geometry shader input arrays */
2042 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2043 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2044 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2045 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2046 ir
->accept(&input_resize_visitor
);
2050 if (ctx
->Const
.VertexID_is_zero_based
)
2051 lower_vertex_id(linked
);
2053 /* Make a pass over all variable declarations to ensure that arrays with
2054 * unspecified sizes have a size specified. The size is inferred from the
2055 * max_array_access field.
2057 array_sizing_visitor v
;
2059 v
.fixup_unnamed_interface_types();
2065 * Update the sizes of linked shader uniform arrays to the maximum
2068 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2070 * If one or more elements of an array are active,
2071 * GetActiveUniform will return the name of the array in name,
2072 * subject to the restrictions listed above. The type of the array
2073 * is returned in type. The size parameter contains the highest
2074 * array element index used, plus one. The compiler or linker
2075 * determines the highest index used. There will be only one
2076 * active uniform reported by the GL per uniform array.
2080 update_array_sizes(struct gl_shader_program
*prog
)
2082 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2083 if (prog
->_LinkedShaders
[i
] == NULL
)
2086 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2087 ir_variable
*const var
= node
->as_variable();
2089 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2090 !var
->type
->is_array())
2093 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2094 * will not be eliminated. Since we always do std140, just
2095 * don't resize arrays in UBOs.
2097 * Atomic counters are supposed to get deterministic
2098 * locations assigned based on the declaration ordering and
2099 * sizes, array compaction would mess that up.
2101 if (var
->is_in_buffer_block() || var
->type
->contains_atomic())
2104 unsigned int size
= var
->data
.max_array_access
;
2105 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2106 if (prog
->_LinkedShaders
[j
] == NULL
)
2109 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2110 ir_variable
*other_var
= node2
->as_variable();
2114 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2115 other_var
->data
.max_array_access
> size
) {
2116 size
= other_var
->data
.max_array_access
;
2121 if (size
+ 1 != var
->type
->length
) {
2122 /* If this is a built-in uniform (i.e., it's backed by some
2123 * fixed-function state), adjust the number of state slots to
2124 * match the new array size. The number of slots per array entry
2125 * is not known. It seems safe to assume that the total number of
2126 * slots is an integer multiple of the number of array elements.
2127 * Determine the number of slots per array element by dividing by
2128 * the old (total) size.
2130 const unsigned num_slots
= var
->get_num_state_slots();
2131 if (num_slots
> 0) {
2132 var
->set_num_state_slots((size
+ 1)
2133 * (num_slots
/ var
->type
->length
));
2136 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2138 /* FINISHME: We should update the types of array
2139 * dereferences of this variable now.
2147 * Resize tessellation evaluation per-vertex inputs to the size of
2148 * tessellation control per-vertex outputs.
2151 resize_tes_inputs(struct gl_context
*ctx
,
2152 struct gl_shader_program
*prog
)
2154 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2157 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2158 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2160 /* If no control shader is present, then the TES inputs are statically
2161 * sized to MaxPatchVertices; the actual size of the arrays won't be
2162 * known until draw time.
2164 const int num_vertices
= tcs
2165 ? tcs
->TessCtrl
.VerticesOut
2166 : ctx
->Const
.MaxPatchVertices
;
2168 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2169 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2170 ir
->accept(&input_resize_visitor
);
2175 * Find a contiguous set of available bits in a bitmask.
2177 * \param used_mask Bits representing used (1) and unused (0) locations
2178 * \param needed_count Number of contiguous bits needed.
2181 * Base location of the available bits on success or -1 on failure.
2184 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2186 unsigned needed_mask
= (1 << needed_count
) - 1;
2187 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2189 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2190 * cannot optimize possibly infinite loops" for the loop below.
2192 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2195 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2196 if ((needed_mask
& ~used_mask
) == needed_mask
)
2207 * Assign locations for either VS inputs or FS outputs
2209 * \param prog Shader program whose variables need locations assigned
2210 * \param target_index Selector for the program target to receive location
2211 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2212 * \c MESA_SHADER_FRAGMENT.
2213 * \param max_index Maximum number of generic locations. This corresponds
2214 * to either the maximum number of draw buffers or the
2215 * maximum number of generic attributes.
2218 * If locations are successfully assigned, true is returned. Otherwise an
2219 * error is emitted to the shader link log and false is returned.
2222 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2223 unsigned target_index
,
2226 /* Mark invalid locations as being used.
2228 unsigned used_locations
= (max_index
>= 32)
2229 ? ~0 : ~((1 << max_index
) - 1);
2231 assert((target_index
== MESA_SHADER_VERTEX
)
2232 || (target_index
== MESA_SHADER_FRAGMENT
));
2234 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2238 /* Operate in a total of four passes.
2240 * 1. Invalidate the location assignments for all vertex shader inputs.
2242 * 2. Assign locations for inputs that have user-defined (via
2243 * glBindVertexAttribLocation) locations and outputs that have
2244 * user-defined locations (via glBindFragDataLocation).
2246 * 3. Sort the attributes without assigned locations by number of slots
2247 * required in decreasing order. Fragmentation caused by attribute
2248 * locations assigned by the application may prevent large attributes
2249 * from having enough contiguous space.
2251 * 4. Assign locations to any inputs without assigned locations.
2254 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2255 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2257 const enum ir_variable_mode direction
=
2258 (target_index
== MESA_SHADER_VERTEX
)
2259 ? ir_var_shader_in
: ir_var_shader_out
;
2262 /* Temporary storage for the set of attributes that need locations assigned.
2268 /* Used below in the call to qsort. */
2269 static int compare(const void *a
, const void *b
)
2271 const temp_attr
*const l
= (const temp_attr
*) a
;
2272 const temp_attr
*const r
= (const temp_attr
*) b
;
2274 /* Reversed because we want a descending order sort below. */
2275 return r
->slots
- l
->slots
;
2279 unsigned num_attr
= 0;
2280 unsigned total_attribs_size
= 0;
2282 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2283 ir_variable
*const var
= node
->as_variable();
2285 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2288 if (var
->data
.explicit_location
) {
2289 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2290 || (var
->data
.location
< 0)) {
2292 "invalid explicit location %d specified for `%s'\n",
2293 (var
->data
.location
< 0)
2294 ? var
->data
.location
2295 : var
->data
.location
- generic_base
,
2299 } else if (target_index
== MESA_SHADER_VERTEX
) {
2302 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2303 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2304 var
->data
.location
= binding
;
2305 var
->data
.is_unmatched_generic_inout
= 0;
2307 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2311 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2312 assert(binding
>= FRAG_RESULT_DATA0
);
2313 var
->data
.location
= binding
;
2314 var
->data
.is_unmatched_generic_inout
= 0;
2316 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2317 var
->data
.index
= index
;
2322 const unsigned slots
= var
->type
->count_attribute_slots();
2324 /* From GL4.5 core spec, section 11.1.1 (Vertex Attributes):
2326 * "A program with more than the value of MAX_VERTEX_ATTRIBS active
2327 * attribute variables may fail to link, unless device-dependent
2328 * optimizations are able to make the program fit within available
2329 * hardware resources. For the purposes of this test, attribute variables
2330 * of the type dvec3, dvec4, dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3,
2331 * and dmat4 may count as consuming twice as many attributes as equivalent
2332 * single-precision types. While these types use the same number of
2333 * generic attributes as their single-precision equivalents,
2334 * implementations are permitted to consume two single-precision vectors
2335 * of internal storage for each three- or four-component double-precision
2337 * Until someone has a good reason in Mesa, enforce that now.
2339 if (target_index
== MESA_SHADER_VERTEX
) {
2340 total_attribs_size
+= slots
;
2341 if (var
->type
->without_array() == glsl_type::dvec3_type
||
2342 var
->type
->without_array() == glsl_type::dvec4_type
||
2343 var
->type
->without_array() == glsl_type::dmat2x3_type
||
2344 var
->type
->without_array() == glsl_type::dmat2x4_type
||
2345 var
->type
->without_array() == glsl_type::dmat3_type
||
2346 var
->type
->without_array() == glsl_type::dmat3x4_type
||
2347 var
->type
->without_array() == glsl_type::dmat4x3_type
||
2348 var
->type
->without_array() == glsl_type::dmat4_type
)
2349 total_attribs_size
+= slots
;
2352 /* If the variable is not a built-in and has a location statically
2353 * assigned in the shader (presumably via a layout qualifier), make sure
2354 * that it doesn't collide with other assigned locations. Otherwise,
2355 * add it to the list of variables that need linker-assigned locations.
2357 if (var
->data
.location
!= -1) {
2358 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2359 /* From page 61 of the OpenGL 4.0 spec:
2361 * "LinkProgram will fail if the attribute bindings assigned
2362 * by BindAttribLocation do not leave not enough space to
2363 * assign a location for an active matrix attribute or an
2364 * active attribute array, both of which require multiple
2365 * contiguous generic attributes."
2367 * I think above text prohibits the aliasing of explicit and
2368 * automatic assignments. But, aliasing is allowed in manual
2369 * assignments of attribute locations. See below comments for
2372 * From OpenGL 4.0 spec, page 61:
2374 * "It is possible for an application to bind more than one
2375 * attribute name to the same location. This is referred to as
2376 * aliasing. This will only work if only one of the aliased
2377 * attributes is active in the executable program, or if no
2378 * path through the shader consumes more than one attribute of
2379 * a set of attributes aliased to the same location. A link
2380 * error can occur if the linker determines that every path
2381 * through the shader consumes multiple aliased attributes,
2382 * but implementations are not required to generate an error
2385 * From GLSL 4.30 spec, page 54:
2387 * "A program will fail to link if any two non-vertex shader
2388 * input variables are assigned to the same location. For
2389 * vertex shaders, multiple input variables may be assigned
2390 * to the same location using either layout qualifiers or via
2391 * the OpenGL API. However, such aliasing is intended only to
2392 * support vertex shaders where each execution path accesses
2393 * at most one input per each location. Implementations are
2394 * permitted, but not required, to generate link-time errors
2395 * if they detect that every path through the vertex shader
2396 * executable accesses multiple inputs assigned to any single
2397 * location. For all shader types, a program will fail to link
2398 * if explicit location assignments leave the linker unable
2399 * to find space for other variables without explicit
2402 * From OpenGL ES 3.0 spec, page 56:
2404 * "Binding more than one attribute name to the same location
2405 * is referred to as aliasing, and is not permitted in OpenGL
2406 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2407 * fail when this condition exists. However, aliasing is
2408 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2409 * This will only work if only one of the aliased attributes
2410 * is active in the executable program, or if no path through
2411 * the shader consumes more than one attribute of a set of
2412 * attributes aliased to the same location. A link error can
2413 * occur if the linker determines that every path through the
2414 * shader consumes multiple aliased attributes, but implemen-
2415 * tations are not required to generate an error in this case."
2417 * After looking at above references from OpenGL, OpenGL ES and
2418 * GLSL specifications, we allow aliasing of vertex input variables
2419 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2421 * NOTE: This is not required by the spec but its worth mentioning
2422 * here that we're not doing anything to make sure that no path
2423 * through the vertex shader executable accesses multiple inputs
2424 * assigned to any single location.
2427 /* Mask representing the contiguous slots that will be used by
2430 const unsigned attr
= var
->data
.location
- generic_base
;
2431 const unsigned use_mask
= (1 << slots
) - 1;
2432 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2433 ? "vertex shader input" : "fragment shader output";
2435 /* Generate a link error if the requested locations for this
2436 * attribute exceed the maximum allowed attribute location.
2438 if (attr
+ slots
> max_index
) {
2440 "insufficient contiguous locations "
2441 "available for %s `%s' %d %d %d\n", string
,
2442 var
->name
, used_locations
, use_mask
, attr
);
2446 /* Generate a link error if the set of bits requested for this
2447 * attribute overlaps any previously allocated bits.
2449 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2450 if (target_index
== MESA_SHADER_FRAGMENT
||
2451 (prog
->IsES
&& prog
->Version
>= 300)) {
2453 "overlapping location is assigned "
2454 "to %s `%s' %d %d %d\n", string
,
2455 var
->name
, used_locations
, use_mask
, attr
);
2458 linker_warning(prog
,
2459 "overlapping location is assigned "
2460 "to %s `%s' %d %d %d\n", string
,
2461 var
->name
, used_locations
, use_mask
, attr
);
2465 used_locations
|= (use_mask
<< attr
);
2471 to_assign
[num_attr
].slots
= slots
;
2472 to_assign
[num_attr
].var
= var
;
2476 if (target_index
== MESA_SHADER_VERTEX
) {
2477 if (total_attribs_size
> max_index
) {
2479 "attempt to use %d vertex attribute slots only %d available ",
2480 total_attribs_size
, max_index
);
2485 /* If all of the attributes were assigned locations by the application (or
2486 * are built-in attributes with fixed locations), return early. This should
2487 * be the common case.
2492 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2494 if (target_index
== MESA_SHADER_VERTEX
) {
2495 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2496 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2497 * reserved to prevent it from being automatically allocated below.
2499 find_deref_visitor
find("gl_Vertex");
2501 if (find
.variable_found())
2502 used_locations
|= (1 << 0);
2505 for (unsigned i
= 0; i
< num_attr
; i
++) {
2506 /* Mask representing the contiguous slots that will be used by this
2509 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2511 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2514 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2515 ? "vertex shader input" : "fragment shader output";
2518 "insufficient contiguous locations "
2519 "available for %s `%s'\n",
2520 string
, to_assign
[i
].var
->name
);
2524 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2525 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2526 used_locations
|= (use_mask
<< location
);
2534 * Demote shader inputs and outputs that are not used in other stages
2537 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2539 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2540 ir_variable
*const var
= node
->as_variable();
2542 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2545 /* A shader 'in' or 'out' variable is only really an input or output if
2546 * its value is used by other shader stages. This will cause the variable
2547 * to have a location assigned.
2549 if (var
->data
.is_unmatched_generic_inout
) {
2550 assert(var
->data
.mode
!= ir_var_temporary
);
2551 var
->data
.mode
= ir_var_auto
;
2558 * Store the gl_FragDepth layout in the gl_shader_program struct.
2561 store_fragdepth_layout(struct gl_shader_program
*prog
)
2563 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2567 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2569 /* We don't look up the gl_FragDepth symbol directly because if
2570 * gl_FragDepth is not used in the shader, it's removed from the IR.
2571 * However, the symbol won't be removed from the symbol table.
2573 * We're only interested in the cases where the variable is NOT removed
2576 foreach_in_list(ir_instruction
, node
, ir
) {
2577 ir_variable
*const var
= node
->as_variable();
2579 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2583 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2584 switch (var
->data
.depth_layout
) {
2585 case ir_depth_layout_none
:
2586 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2588 case ir_depth_layout_any
:
2589 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2591 case ir_depth_layout_greater
:
2592 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2594 case ir_depth_layout_less
:
2595 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2597 case ir_depth_layout_unchanged
:
2598 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2609 * Validate the resources used by a program versus the implementation limits
2612 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2614 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2615 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2620 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2621 linker_error(prog
, "Too many %s shader texture samplers\n",
2622 _mesa_shader_stage_to_string(i
));
2625 if (sh
->num_uniform_components
>
2626 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2627 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2628 linker_warning(prog
, "Too many %s shader default uniform block "
2629 "components, but the driver will try to optimize "
2630 "them out; this is non-portable out-of-spec "
2632 _mesa_shader_stage_to_string(i
));
2634 linker_error(prog
, "Too many %s shader default uniform block "
2636 _mesa_shader_stage_to_string(i
));
2640 if (sh
->num_combined_uniform_components
>
2641 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2642 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2643 linker_warning(prog
, "Too many %s shader uniform components, "
2644 "but the driver will try to optimize them out; "
2645 "this is non-portable out-of-spec behavior\n",
2646 _mesa_shader_stage_to_string(i
));
2648 linker_error(prog
, "Too many %s shader uniform components\n",
2649 _mesa_shader_stage_to_string(i
));
2654 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2655 unsigned total_uniform_blocks
= 0;
2657 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2658 if (prog
->UniformBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2659 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2660 prog
->UniformBlocks
[i
].Name
,
2661 prog
->UniformBlocks
[i
].UniformBufferSize
,
2662 ctx
->Const
.MaxUniformBlockSize
);
2665 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2666 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2668 total_uniform_blocks
++;
2672 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2673 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2674 prog
->NumUniformBlocks
,
2675 ctx
->Const
.MaxCombinedUniformBlocks
);
2677 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2678 const unsigned max_uniform_blocks
=
2679 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2680 if (blocks
[i
] > max_uniform_blocks
) {
2681 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2682 _mesa_shader_stage_to_string(i
),
2684 max_uniform_blocks
);
2693 * Validate shader image resources.
2696 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2698 unsigned total_image_units
= 0;
2699 unsigned fragment_outputs
= 0;
2701 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2704 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2705 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2708 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2709 linker_error(prog
, "Too many %s shader image uniforms\n",
2710 _mesa_shader_stage_to_string(i
));
2712 total_image_units
+= sh
->NumImages
;
2714 if (i
== MESA_SHADER_FRAGMENT
) {
2715 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2716 ir_variable
*var
= node
->as_variable();
2717 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2718 fragment_outputs
+= var
->type
->count_attribute_slots();
2724 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2725 linker_error(prog
, "Too many combined image uniforms\n");
2727 if (total_image_units
+ fragment_outputs
>
2728 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2729 linker_error(prog
, "Too many combined image uniforms and fragment outputs\n");
2734 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2735 * for a variable, checks for overlaps between other uniforms using explicit
2739 reserve_explicit_locations(struct gl_shader_program
*prog
,
2740 string_to_uint_map
*map
, ir_variable
*var
)
2742 unsigned slots
= var
->type
->uniform_locations();
2743 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2745 /* Resize remap table if locations do not fit in the current one. */
2746 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2747 prog
->UniformRemapTable
=
2748 reralloc(prog
, prog
->UniformRemapTable
,
2749 gl_uniform_storage
*,
2752 if (!prog
->UniformRemapTable
) {
2753 linker_error(prog
, "Out of memory during linking.\n");
2757 /* Initialize allocated space. */
2758 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2759 prog
->UniformRemapTable
[i
] = NULL
;
2761 prog
->NumUniformRemapTable
= max_loc
+ 1;
2764 for (unsigned i
= 0; i
< slots
; i
++) {
2765 unsigned loc
= var
->data
.location
+ i
;
2767 /* Check if location is already used. */
2768 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2770 /* Possibly same uniform from a different stage, this is ok. */
2772 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2775 /* ARB_explicit_uniform_location specification states:
2777 * "No two default-block uniform variables in the program can have
2778 * the same location, even if they are unused, otherwise a compiler
2779 * or linker error will be generated."
2782 "location qualifier for uniform %s overlaps "
2783 "previously used location\n",
2788 /* Initialize location as inactive before optimization
2789 * rounds and location assignment.
2791 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2794 /* Note, base location used for arrays. */
2795 map
->put(var
->data
.location
, var
->name
);
2801 * Check and reserve all explicit uniform locations, called before
2802 * any optimizations happen to handle also inactive uniforms and
2803 * inactive array elements that may get trimmed away.
2806 check_explicit_uniform_locations(struct gl_context
*ctx
,
2807 struct gl_shader_program
*prog
)
2809 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2812 /* This map is used to detect if overlapping explicit locations
2813 * occur with the same uniform (from different stage) or a different one.
2815 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2818 linker_error(prog
, "Out of memory during linking.\n");
2822 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2823 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2828 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2829 ir_variable
*var
= node
->as_variable();
2830 if (var
&& (var
->data
.mode
== ir_var_uniform
|| var
->data
.mode
== ir_var_shader_storage
) &&
2831 var
->data
.explicit_location
) {
2832 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2844 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
2845 const void *data
, uint8_t stages
)
2849 /* If resource already exists, do not add it again. */
2850 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
2851 if (prog
->ProgramResourceList
[i
].Data
== data
)
2854 prog
->ProgramResourceList
=
2856 prog
->ProgramResourceList
,
2857 gl_program_resource
,
2858 prog
->NumProgramResourceList
+ 1);
2860 if (!prog
->ProgramResourceList
) {
2861 linker_error(prog
, "Out of memory during linking.\n");
2865 struct gl_program_resource
*res
=
2866 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
2870 res
->StageReferences
= stages
;
2872 prog
->NumProgramResourceList
++;
2878 * Function builds a stage reference bitmask from variable name.
2881 build_stageref(struct gl_shader_program
*shProg
, const char *name
)
2885 /* Note, that we assume MAX 8 stages, if there will be more stages, type
2886 * used for reference mask in gl_program_resource will need to be changed.
2888 assert(MESA_SHADER_STAGES
< 8);
2890 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2891 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
2895 /* Shader symbol table may contain variables that have
2896 * been optimized away. Search IR for the variable instead.
2898 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2899 ir_variable
*var
= node
->as_variable();
2900 if (var
&& strcmp(var
->name
, name
) == 0) {
2910 add_interface_variables(struct gl_shader_program
*shProg
,
2911 struct gl_shader
*sh
, GLenum programInterface
)
2913 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2914 ir_variable
*var
= node
->as_variable();
2920 switch (var
->data
.mode
) {
2921 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
2922 * "For GetActiveAttrib, all active vertex shader input variables
2923 * are enumerated, including the special built-in inputs gl_VertexID
2924 * and gl_InstanceID."
2926 case ir_var_system_value
:
2927 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
2928 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
2929 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
2931 /* Mark special built-in inputs referenced by the vertex stage so
2932 * that they are considered active by the shader queries.
2934 mask
= (1 << (MESA_SHADER_VERTEX
));
2936 case ir_var_shader_in
:
2937 if (programInterface
!= GL_PROGRAM_INPUT
)
2940 case ir_var_shader_out
:
2941 if (programInterface
!= GL_PROGRAM_OUTPUT
)
2948 if (!add_program_resource(shProg
, programInterface
, var
,
2949 build_stageref(shProg
, var
->name
) | mask
))
2956 * Builds up a list of program resources that point to existing
2960 build_program_resource_list(struct gl_context
*ctx
,
2961 struct gl_shader_program
*shProg
)
2963 /* Rebuild resource list. */
2964 if (shProg
->ProgramResourceList
) {
2965 ralloc_free(shProg
->ProgramResourceList
);
2966 shProg
->ProgramResourceList
= NULL
;
2967 shProg
->NumProgramResourceList
= 0;
2970 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
2972 /* Determine first input and final output stage. These are used to
2973 * detect which variables should be enumerated in the resource list
2974 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
2976 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2977 if (!shProg
->_LinkedShaders
[i
])
2979 if (input_stage
== MESA_SHADER_STAGES
)
2984 /* Empty shader, no resources. */
2985 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
2988 /* Add inputs and outputs to the resource list. */
2989 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
],
2993 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
],
2997 /* Add transform feedback varyings. */
2998 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
2999 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
3001 build_stageref(shProg
,
3002 shProg
->LinkedTransformFeedback
.Varyings
[i
].Name
);
3003 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
3004 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
3010 /* Add uniforms from uniform storage. */
3011 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3012 /* Do not add uniforms internally used by Mesa. */
3013 if (shProg
->UniformStorage
[i
].hidden
)
3017 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
);
3019 /* Add stagereferences for uniforms in a uniform block. */
3020 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3021 if (block_index
!= -1) {
3022 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
3023 if (shProg
->UniformBlockStageIndex
[j
][block_index
] != -1)
3024 stageref
|= (1 << j
);
3028 if (!add_program_resource(shProg
, GL_UNIFORM
,
3029 &shProg
->UniformStorage
[i
], stageref
))
3033 /* Add program uniform blocks. */
3034 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
3035 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
3036 &shProg
->UniformBlocks
[i
], 0))
3040 /* Add atomic counter buffers. */
3041 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3042 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3043 &shProg
->AtomicBuffers
[i
], 0))
3047 /* TODO - following extensions will require more resource types:
3049 * GL_ARB_shader_storage_buffer_object
3050 * GL_ARB_shader_subroutine
3055 * This check is done to make sure we allow only constant expression
3056 * indexing and "constant-index-expression" (indexing with an expression
3057 * that includes loop induction variable).
3060 validate_sampler_array_indexing(struct gl_context
*ctx
,
3061 struct gl_shader_program
*prog
)
3063 dynamic_sampler_array_indexing_visitor v
;
3064 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3065 if (prog
->_LinkedShaders
[i
] == NULL
)
3068 bool no_dynamic_indexing
=
3069 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
3071 /* Search for array derefs in shader. */
3072 v
.run(prog
->_LinkedShaders
[i
]->ir
);
3073 if (v
.uses_dynamic_sampler_array_indexing()) {
3074 const char *msg
= "sampler arrays indexed with non-constant "
3075 "expressions is forbidden in GLSL %s %u";
3076 /* Backend has indicated that it has no dynamic indexing support. */
3077 if (no_dynamic_indexing
) {
3078 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3081 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3090 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3092 tfeedback_decl
*tfeedback_decls
= NULL
;
3093 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
3095 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
3097 prog
->LinkStatus
= true; /* All error paths will set this to false */
3098 prog
->Validated
= false;
3099 prog
->_Used
= false;
3101 prog
->ARB_fragment_coord_conventions_enable
= false;
3103 /* Separate the shaders into groups based on their type.
3105 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
3106 unsigned num_shaders
[MESA_SHADER_STAGES
];
3108 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3109 shader_list
[i
] = (struct gl_shader
**)
3110 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
3114 unsigned min_version
= UINT_MAX
;
3115 unsigned max_version
= 0;
3116 const bool is_es_prog
=
3117 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
3118 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
3119 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
3120 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
3122 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
3123 linker_error(prog
, "all shaders must use same shading "
3124 "language version\n");
3128 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
3129 prog
->ARB_fragment_coord_conventions_enable
= true;
3132 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
3133 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
3134 num_shaders
[shader_type
]++;
3137 /* In desktop GLSL, different shader versions may be linked together. In
3138 * GLSL ES, all shader versions must be the same.
3140 if (is_es_prog
&& min_version
!= max_version
) {
3141 linker_error(prog
, "all shaders must use same shading "
3142 "language version\n");
3146 prog
->Version
= max_version
;
3147 prog
->IsES
= is_es_prog
;
3149 /* Some shaders have to be linked with some other shaders present.
3151 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
3152 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3153 !prog
->SeparateShader
) {
3154 linker_error(prog
, "Geometry shader must be linked with "
3158 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
3159 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3160 !prog
->SeparateShader
) {
3161 linker_error(prog
, "Tessellation evaluation shader must be linked with "
3165 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3166 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3167 !prog
->SeparateShader
) {
3168 linker_error(prog
, "Tessellation control shader must be linked with "
3173 /* The spec is self-contradictory here. It allows linking without a tess
3174 * eval shader, but that can only be used with transform feedback and
3175 * rasterization disabled. However, transform feedback isn't allowed
3176 * with GL_PATCHES, so it can't be used.
3178 * More investigation showed that the idea of transform feedback after
3179 * a tess control shader was dropped, because some hw vendors couldn't
3180 * support tessellation without a tess eval shader, but the linker section
3181 * wasn't updated to reflect that.
3183 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
3186 * Do what's reasonable and always require a tess eval shader if a tess
3187 * control shader is present.
3189 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3190 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0 &&
3191 !prog
->SeparateShader
) {
3192 linker_error(prog
, "Tessellation control shader must be linked with "
3193 "tessellation evaluation shader\n");
3197 /* Compute shaders have additional restrictions. */
3198 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
3199 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
3200 linker_error(prog
, "Compute shaders may not be linked with any other "
3201 "type of shader\n");
3204 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3205 if (prog
->_LinkedShaders
[i
] != NULL
)
3206 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
3208 prog
->_LinkedShaders
[i
] = NULL
;
3211 /* Link all shaders for a particular stage and validate the result.
3213 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
3214 if (num_shaders
[stage
] > 0) {
3215 gl_shader
*const sh
=
3216 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
3217 num_shaders
[stage
]);
3219 if (!prog
->LinkStatus
) {
3221 ctx
->Driver
.DeleteShader(ctx
, sh
);
3226 case MESA_SHADER_VERTEX
:
3227 validate_vertex_shader_executable(prog
, sh
);
3229 case MESA_SHADER_TESS_CTRL
:
3230 /* nothing to be done */
3232 case MESA_SHADER_TESS_EVAL
:
3233 validate_tess_eval_shader_executable(prog
, sh
);
3235 case MESA_SHADER_GEOMETRY
:
3236 validate_geometry_shader_executable(prog
, sh
);
3238 case MESA_SHADER_FRAGMENT
:
3239 validate_fragment_shader_executable(prog
, sh
);
3242 if (!prog
->LinkStatus
) {
3244 ctx
->Driver
.DeleteShader(ctx
, sh
);
3248 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
3252 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
3253 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
3254 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
3255 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
3256 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
3257 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
3259 prog
->LastClipDistanceArraySize
= 0; /* Not used */
3261 /* Here begins the inter-stage linking phase. Some initial validation is
3262 * performed, then locations are assigned for uniforms, attributes, and
3265 cross_validate_uniforms(prog
);
3266 if (!prog
->LinkStatus
)
3271 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
3272 if (prog
->_LinkedShaders
[prev
] != NULL
)
3276 check_explicit_uniform_locations(ctx
, prog
);
3277 if (!prog
->LinkStatus
)
3280 resize_tes_inputs(ctx
, prog
);
3282 /* Validate the inputs of each stage with the output of the preceding
3285 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3286 if (prog
->_LinkedShaders
[i
] == NULL
)
3289 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
3290 prog
->_LinkedShaders
[i
]);
3291 if (!prog
->LinkStatus
)
3294 cross_validate_outputs_to_inputs(prog
,
3295 prog
->_LinkedShaders
[prev
],
3296 prog
->_LinkedShaders
[i
]);
3297 if (!prog
->LinkStatus
)
3303 /* Cross-validate uniform blocks between shader stages */
3304 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
3305 MESA_SHADER_STAGES
);
3306 if (!prog
->LinkStatus
)
3309 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3310 if (prog
->_LinkedShaders
[i
] != NULL
)
3311 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
3314 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
3315 * it before optimization because we want most of the checks to get
3316 * dropped thanks to constant propagation.
3318 * This rule also applies to GLSL ES 3.00.
3320 if (max_version
>= (is_es_prog
? 300 : 130)) {
3321 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3323 lower_discard_flow(sh
->ir
);
3327 if (!interstage_cross_validate_uniform_blocks(prog
))
3330 /* Do common optimization before assigning storage for attributes,
3331 * uniforms, and varyings. Later optimization could possibly make
3332 * some of that unused.
3334 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3335 if (prog
->_LinkedShaders
[i
] == NULL
)
3338 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
3339 if (!prog
->LinkStatus
)
3342 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
3343 lower_clip_distance(prog
->_LinkedShaders
[i
]);
3346 if (ctx
->Const
.LowerTessLevel
) {
3347 lower_tess_level(prog
->_LinkedShaders
[i
]);
3350 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
3351 &ctx
->Const
.ShaderCompilerOptions
[i
],
3352 ctx
->Const
.NativeIntegers
))
3355 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
3358 /* Validation for special cases where we allow sampler array indexing
3359 * with loop induction variable. This check emits a warning or error
3360 * depending if backend can handle dynamic indexing.
3362 if ((!prog
->IsES
&& prog
->Version
< 130) ||
3363 (prog
->IsES
&& prog
->Version
< 300)) {
3364 if (!validate_sampler_array_indexing(ctx
, prog
))
3368 /* Check and validate stream emissions in geometry shaders */
3369 validate_geometry_shader_emissions(ctx
, prog
);
3371 /* Mark all generic shader inputs and outputs as unpaired. */
3372 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3373 if (prog
->_LinkedShaders
[i
] != NULL
) {
3374 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
3378 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
,
3379 ctx
->Const
.Program
[MESA_SHADER_VERTEX
].MaxAttribs
)) {
3383 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
3387 unsigned first
, last
;
3389 first
= MESA_SHADER_STAGES
;
3392 /* Determine first and last stage. */
3393 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3394 if (!prog
->_LinkedShaders
[i
])
3396 if (first
== MESA_SHADER_STAGES
)
3401 if (num_tfeedback_decls
!= 0) {
3402 /* From GL_EXT_transform_feedback:
3403 * A program will fail to link if:
3405 * * the <count> specified by TransformFeedbackVaryingsEXT is
3406 * non-zero, but the program object has no vertex or geometry
3409 if (first
== MESA_SHADER_FRAGMENT
) {
3410 linker_error(prog
, "Transform feedback varyings specified, but "
3411 "no vertex or geometry shader is present.\n");
3415 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
3416 prog
->TransformFeedback
.NumVarying
);
3417 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
3418 prog
->TransformFeedback
.VaryingNames
,
3423 /* Linking the stages in the opposite order (from fragment to vertex)
3424 * ensures that inter-shader outputs written to in an earlier stage are
3425 * eliminated if they are (transitively) not used in a later stage.
3429 if (first
< MESA_SHADER_FRAGMENT
) {
3430 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
3432 if (first
== MESA_SHADER_GEOMETRY
) {
3433 /* There was no vertex shader, but we still have to assign varying
3434 * locations for use by geometry shader inputs in SSO.
3436 * If the shader is not separable (i.e., prog->SeparateShader is
3437 * false), linking will have already failed when first is
3438 * MESA_SHADER_GEOMETRY.
3440 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3441 NULL
, prog
->_LinkedShaders
[first
],
3442 num_tfeedback_decls
, tfeedback_decls
))
3446 if (last
!= MESA_SHADER_FRAGMENT
&&
3447 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
3448 /* There was no fragment shader, but we still have to assign varying
3449 * locations for use by transform feedback.
3451 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3453 num_tfeedback_decls
, tfeedback_decls
))
3457 do_dead_builtin_varyings(ctx
, sh
, NULL
,
3458 num_tfeedback_decls
, tfeedback_decls
);
3460 if (!prog
->SeparateShader
)
3461 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
3463 /* Eliminate code that is now dead due to unused outputs being demoted.
3465 while (do_dead_code(sh
->ir
, false))
3468 else if (first
== MESA_SHADER_FRAGMENT
) {
3469 /* If the program only contains a fragment shader...
3471 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
3473 do_dead_builtin_varyings(ctx
, NULL
, sh
,
3474 num_tfeedback_decls
, tfeedback_decls
);
3476 if (prog
->SeparateShader
) {
3477 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3478 NULL
/* producer */,
3480 0 /* num_tfeedback_decls */,
3481 NULL
/* tfeedback_decls */))
3484 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
3486 while (do_dead_code(sh
->ir
, false))
3491 for (int i
= next
- 1; i
>= 0; i
--) {
3492 if (prog
->_LinkedShaders
[i
] == NULL
)
3495 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
3496 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
3498 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
3499 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3503 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
3504 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3507 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
3508 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
3510 /* Eliminate code that is now dead due to unused outputs being demoted.
3512 while (do_dead_code(sh_i
->ir
, false))
3514 while (do_dead_code(sh_next
->ir
, false))
3517 /* This must be done after all dead varyings are eliminated. */
3518 if (!check_against_output_limit(ctx
, prog
, sh_i
))
3520 if (!check_against_input_limit(ctx
, prog
, sh_next
))
3526 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
3529 update_array_sizes(prog
);
3530 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
3531 link_assign_atomic_counter_resources(ctx
, prog
);
3532 store_fragdepth_layout(prog
);
3534 check_resources(ctx
, prog
);
3535 check_image_resources(ctx
, prog
);
3536 link_check_atomic_counter_resources(ctx
, prog
);
3538 if (!prog
->LinkStatus
)
3541 /* OpenGL ES requires that a vertex shader and a fragment shader both be
3542 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
3543 * anything about shader linking when one of the shaders (vertex or
3544 * fragment shader) is absent. So, the extension shouldn't change the
3545 * behavior specified in GLSL specification.
3547 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
3548 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
3549 linker_error(prog
, "program lacks a vertex shader\n");
3550 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3551 linker_error(prog
, "program lacks a fragment shader\n");
3555 /* FINISHME: Assign fragment shader output locations. */
3558 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3559 free(shader_list
[i
]);
3560 if (prog
->_LinkedShaders
[i
] == NULL
)
3563 /* Do a final validation step to make sure that the IR wasn't
3564 * invalidated by any modifications performed after intrastage linking.
3566 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
3568 /* Retain any live IR, but trash the rest. */
3569 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
3571 /* The symbol table in the linked shaders may contain references to
3572 * variables that were removed (e.g., unused uniforms). Since it may
3573 * contain junk, there is no possible valid use. Delete it and set the
3576 delete prog
->_LinkedShaders
[i
]->symbols
;
3577 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
3580 ralloc_free(mem_ctx
);