2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
68 #include "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/hash_table.h"
76 #include "link_varyings.h"
77 #include "ir_optimization.h"
78 #include "ir_rvalue_visitor.h"
79 #include "ir_uniform.h"
81 #include "main/shaderobj.h"
82 #include "main/enums.h"
88 * Visitor that determines whether or not a variable is ever written.
90 class find_assignment_visitor
: public ir_hierarchical_visitor
{
92 find_assignment_visitor(const char *name
)
93 : name(name
), found(false)
98 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
100 ir_variable
*const var
= ir
->lhs
->variable_referenced();
102 if (strcmp(name
, var
->name
) == 0) {
107 return visit_continue_with_parent
;
110 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
112 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
113 actual_node
, &ir
->actual_parameters
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
115 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
117 if (sig_param
->data
.mode
== ir_var_function_out
||
118 sig_param
->data
.mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
127 if (ir
->return_deref
!= NULL
) {
128 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
130 if (strcmp(name
, var
->name
) == 0) {
136 return visit_continue_with_parent
;
139 bool variable_found()
145 const char *name
; /**< Find writes to a variable with this name. */
146 bool found
; /**< Was a write to the variable found? */
151 * Visitor that determines whether or not a variable is ever read.
153 class find_deref_visitor
: public ir_hierarchical_visitor
{
155 find_deref_visitor(const char *name
)
156 : name(name
), found(false)
161 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
163 if (strcmp(this->name
, ir
->var
->name
) == 0) {
168 return visit_continue
;
171 bool variable_found() const
177 const char *name
; /**< Find writes to a variable with this name. */
178 bool found
; /**< Was a write to the variable found? */
182 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
184 unsigned num_vertices
;
185 gl_shader_program
*prog
;
187 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
189 this->num_vertices
= num_vertices
;
193 virtual ~geom_array_resize_visitor()
198 virtual ir_visitor_status
visit(ir_variable
*var
)
200 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
201 return visit_continue
;
203 unsigned size
= var
->type
->length
;
205 /* Generate a link error if the shader has declared this array with an
208 if (size
&& size
!= this->num_vertices
) {
209 linker_error(this->prog
, "size of array %s declared as %u, "
210 "but number of input vertices is %u\n",
211 var
->name
, size
, this->num_vertices
);
212 return visit_continue
;
215 /* Generate a link error if the shader attempts to access an input
216 * array using an index too large for its actual size assigned at link
219 if (var
->data
.max_array_access
>= this->num_vertices
) {
220 linker_error(this->prog
, "geometry shader accesses element %i of "
221 "%s, but only %i input vertices\n",
222 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
223 return visit_continue
;
226 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
228 var
->data
.max_array_access
= this->num_vertices
- 1;
230 return visit_continue
;
233 /* Dereferences of input variables need to be updated so that their type
234 * matches the newly assigned type of the variable they are accessing. */
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 /* Dereferences of 2D input arrays need to be updated so that their type
242 * matches the newly assigned type of the array they are accessing. */
243 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
245 const glsl_type
*const vt
= ir
->array
->type
;
247 ir
->type
= vt
->fields
.array
;
248 return visit_continue
;
252 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
254 unsigned num_vertices
;
255 gl_shader_program
*prog
;
257 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
259 this->num_vertices
= num_vertices
;
263 virtual ~tess_eval_array_resize_visitor()
268 virtual ir_visitor_status
visit(ir_variable
*var
)
270 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
271 return visit_continue
;
273 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
275 var
->data
.max_array_access
= this->num_vertices
- 1;
277 return visit_continue
;
280 /* Dereferences of input variables need to be updated so that their type
281 * matches the newly assigned type of the variable they are accessing. */
282 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
284 ir
->type
= ir
->var
->type
;
285 return visit_continue
;
288 /* Dereferences of 2D input arrays need to be updated so that their type
289 * matches the newly assigned type of the array they are accessing. */
290 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
292 const glsl_type
*const vt
= ir
->array
->type
;
294 ir
->type
= vt
->fields
.array
;
295 return visit_continue
;
299 class barrier_use_visitor
: public ir_hierarchical_visitor
{
301 barrier_use_visitor(gl_shader_program
*prog
)
302 : prog(prog
), in_main(false), after_return(false), control_flow(0)
306 virtual ~barrier_use_visitor()
311 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
313 if (strcmp(ir
->name
, "main") == 0)
316 return visit_continue
;
319 virtual ir_visitor_status
visit_leave(ir_function
*)
322 after_return
= false;
323 return visit_continue
;
326 virtual ir_visitor_status
visit_leave(ir_return
*)
329 return visit_continue
;
332 virtual ir_visitor_status
visit_enter(ir_if
*)
335 return visit_continue
;
338 virtual ir_visitor_status
visit_leave(ir_if
*)
341 return visit_continue
;
344 virtual ir_visitor_status
visit_enter(ir_loop
*)
347 return visit_continue
;
350 virtual ir_visitor_status
visit_leave(ir_loop
*)
353 return visit_continue
;
356 /* FINISHME: `switch` is not expressed at the IR level -- it's already
357 * been lowered to a mess of `if`s. We'll correctly disallow any use of
358 * barrier() in a conditional path within the switch, but not in a path
359 * which is always hit.
362 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
364 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
365 /* Use of barrier(); determine if it is legal: */
367 linker_error(prog
, "Builtin barrier() may only be used in main");
372 linker_error(prog
, "Builtin barrier() may not be used after return");
376 if (control_flow
!= 0) {
377 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
381 return visit_continue
;
385 gl_shader_program
*prog
;
386 bool in_main
, after_return
;
391 * Visitor that determines the highest stream id to which a (geometry) shader
392 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
394 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
396 find_emit_vertex_visitor(int max_allowed
)
397 : max_stream_allowed(max_allowed
),
398 invalid_stream_id(0),
399 invalid_stream_id_from_emit_vertex(false),
400 end_primitive_found(false),
401 uses_non_zero_stream(false)
406 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
408 int stream_id
= ir
->stream_id();
411 invalid_stream_id
= stream_id
;
412 invalid_stream_id_from_emit_vertex
= true;
416 if (stream_id
> max_stream_allowed
) {
417 invalid_stream_id
= stream_id
;
418 invalid_stream_id_from_emit_vertex
= true;
423 uses_non_zero_stream
= true;
425 return visit_continue
;
428 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
430 end_primitive_found
= true;
432 int stream_id
= ir
->stream_id();
435 invalid_stream_id
= stream_id
;
436 invalid_stream_id_from_emit_vertex
= false;
440 if (stream_id
> max_stream_allowed
) {
441 invalid_stream_id
= stream_id
;
442 invalid_stream_id_from_emit_vertex
= false;
447 uses_non_zero_stream
= true;
449 return visit_continue
;
454 return invalid_stream_id
!= 0;
457 const char *error_func()
459 return invalid_stream_id_from_emit_vertex
?
460 "EmitStreamVertex" : "EndStreamPrimitive";
465 return invalid_stream_id
;
470 return uses_non_zero_stream
;
473 bool uses_end_primitive()
475 return end_primitive_found
;
479 int max_stream_allowed
;
480 int invalid_stream_id
;
481 bool invalid_stream_id_from_emit_vertex
;
482 bool end_primitive_found
;
483 bool uses_non_zero_stream
;
486 /* Class that finds array derefs and check if indexes are dynamic. */
487 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
490 dynamic_sampler_array_indexing_visitor() :
491 dynamic_sampler_array_indexing(false)
495 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
497 if (!ir
->variable_referenced())
498 return visit_continue
;
500 if (!ir
->variable_referenced()->type
->contains_sampler())
501 return visit_continue
;
503 if (!ir
->array_index
->constant_expression_value()) {
504 dynamic_sampler_array_indexing
= true;
507 return visit_continue
;
510 bool uses_dynamic_sampler_array_indexing()
512 return dynamic_sampler_array_indexing
;
516 bool dynamic_sampler_array_indexing
;
519 } /* anonymous namespace */
522 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
526 ralloc_strcat(&prog
->InfoLog
, "error: ");
528 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
531 prog
->LinkStatus
= false;
536 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
540 ralloc_strcat(&prog
->InfoLog
, "warning: ");
542 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
549 * Given a string identifying a program resource, break it into a base name
550 * and an optional array index in square brackets.
552 * If an array index is present, \c out_base_name_end is set to point to the
553 * "[" that precedes the array index, and the array index itself is returned
556 * If no array index is present (or if the array index is negative or
557 * mal-formed), \c out_base_name_end, is set to point to the null terminator
558 * at the end of the input string, and -1 is returned.
560 * Only the final array index is parsed; if the string contains other array
561 * indices (or structure field accesses), they are left in the base name.
563 * No attempt is made to check that the base name is properly formed;
564 * typically the caller will look up the base name in a hash table, so
565 * ill-formed base names simply turn into hash table lookup failures.
568 parse_program_resource_name(const GLchar
*name
,
569 const GLchar
**out_base_name_end
)
571 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
573 * "When an integer array element or block instance number is part of
574 * the name string, it will be specified in decimal form without a "+"
575 * or "-" sign or any extra leading zeroes. Additionally, the name
576 * string will not include white space anywhere in the string."
579 const size_t len
= strlen(name
);
580 *out_base_name_end
= name
+ len
;
582 if (len
== 0 || name
[len
-1] != ']')
585 /* Walk backwards over the string looking for a non-digit character. This
586 * had better be the opening bracket for an array index.
588 * Initially, i specifies the location of the ']'. Since the string may
589 * contain only the ']' charcater, walk backwards very carefully.
592 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
595 if ((i
== 0) || name
[i
-1] != '[')
598 long array_index
= strtol(&name
[i
], NULL
, 10);
602 /* Check for leading zero */
603 if (name
[i
] == '0' && name
[i
+1] != ']')
606 *out_base_name_end
= name
+ (i
- 1);
612 link_invalidate_variable_locations(exec_list
*ir
)
614 foreach_in_list(ir_instruction
, node
, ir
) {
615 ir_variable
*const var
= node
->as_variable();
620 /* Only assign locations for variables that lack an explicit location.
621 * Explicit locations are set for all built-in variables, generic vertex
622 * shader inputs (via layout(location=...)), and generic fragment shader
623 * outputs (also via layout(location=...)).
625 if (!var
->data
.explicit_location
) {
626 var
->data
.location
= -1;
627 var
->data
.location_frac
= 0;
630 /* ir_variable::is_unmatched_generic_inout is used by the linker while
631 * connecting outputs from one stage to inputs of the next stage.
633 if (var
->data
.explicit_location
&&
634 var
->data
.location
< VARYING_SLOT_VAR0
) {
635 var
->data
.is_unmatched_generic_inout
= 0;
637 var
->data
.is_unmatched_generic_inout
= 1;
644 * Set clip_distance_array_size based and cull_distance_array_size on the given
647 * Also check for errors based on incorrect usage of gl_ClipVertex and
648 * gl_ClipDistance and gl_CullDistance.
649 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
650 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
652 * Return false if an error was reported.
655 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
656 struct gl_shader
*shader
,
657 struct gl_context
*ctx
,
658 GLuint
*clip_distance_array_size
,
659 GLuint
*cull_distance_array_size
)
661 *clip_distance_array_size
= 0;
662 *cull_distance_array_size
= 0;
664 if (!prog
->IsES
&& prog
->Version
>= 130) {
665 /* From section 7.1 (Vertex Shader Special Variables) of the
668 * "It is an error for a shader to statically write both
669 * gl_ClipVertex and gl_ClipDistance."
671 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
672 * gl_ClipVertex nor gl_ClipDistance.
674 find_assignment_visitor
clip_vertex("gl_ClipVertex");
675 find_assignment_visitor
clip_distance("gl_ClipDistance");
676 find_assignment_visitor
cull_distance("gl_CullDistance");
678 clip_vertex
.run(shader
->ir
);
679 clip_distance
.run(shader
->ir
);
680 cull_distance
.run(shader
->ir
);
682 /* From the ARB_cull_distance spec:
684 * It is a compile-time or link-time error for the set of shaders forming
685 * a program to statically read or write both gl_ClipVertex and either
686 * gl_ClipDistance or gl_CullDistance.
688 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
689 * gl_ClipVertex, gl_ClipDistance or gl_CullDistance.
691 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
692 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
693 "and `gl_ClipDistance'\n",
694 _mesa_shader_stage_to_string(shader
->Stage
));
697 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
698 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
699 "and `gl_CullDistance'\n",
700 _mesa_shader_stage_to_string(shader
->Stage
));
704 if (clip_distance
.variable_found()) {
705 ir_variable
*clip_distance_var
=
706 shader
->symbols
->get_variable("gl_ClipDistance");
707 assert(clip_distance_var
);
708 *clip_distance_array_size
= clip_distance_var
->type
->length
;
710 if (cull_distance
.variable_found()) {
711 ir_variable
*cull_distance_var
=
712 shader
->symbols
->get_variable("gl_CullDistance");
713 assert(cull_distance_var
);
714 *cull_distance_array_size
= cull_distance_var
->type
->length
;
716 /* From the ARB_cull_distance spec:
718 * It is a compile-time or link-time error for the set of shaders forming
719 * a program to have the sum of the sizes of the gl_ClipDistance and
720 * gl_CullDistance arrays to be larger than
721 * gl_MaxCombinedClipAndCullDistances.
723 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
724 ctx
->Const
.MaxClipPlanes
) {
725 linker_error(prog
, "%s shader: the combined size of "
726 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
728 "gl_MaxCombinedClipAndCullDistances (%u)",
729 _mesa_shader_stage_to_string(shader
->Stage
),
730 ctx
->Const
.MaxClipPlanes
);
737 * Verify that a vertex shader executable meets all semantic requirements.
739 * Also sets prog->Vert.ClipDistanceArraySize and
740 * prog->Vert.CullDistanceArraySize as a side effect.
742 * \param shader Vertex shader executable to be verified
745 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
746 struct gl_shader
*shader
,
747 struct gl_context
*ctx
)
752 /* From the GLSL 1.10 spec, page 48:
754 * "The variable gl_Position is available only in the vertex
755 * language and is intended for writing the homogeneous vertex
756 * position. All executions of a well-formed vertex shader
757 * executable must write a value into this variable. [...] The
758 * variable gl_Position is available only in the vertex
759 * language and is intended for writing the homogeneous vertex
760 * position. All executions of a well-formed vertex shader
761 * executable must write a value into this variable."
763 * while in GLSL 1.40 this text is changed to:
765 * "The variable gl_Position is available only in the vertex
766 * language and is intended for writing the homogeneous vertex
767 * position. It can be written at any time during shader
768 * execution. It may also be read back by a vertex shader
769 * after being written. This value will be used by primitive
770 * assembly, clipping, culling, and other fixed functionality
771 * operations, if present, that operate on primitives after
772 * vertex processing has occurred. Its value is undefined if
773 * the vertex shader executable does not write gl_Position."
775 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
776 * gl_Position is not an error.
778 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
779 find_assignment_visitor
find("gl_Position");
780 find
.run(shader
->ir
);
781 if (!find
.variable_found()) {
784 "vertex shader does not write to `gl_Position'."
785 "It's value is undefined. \n");
788 "vertex shader does not write to `gl_Position'. \n");
794 analyze_clip_cull_usage(prog
, shader
, ctx
,
795 &prog
->Vert
.ClipDistanceArraySize
,
796 &prog
->Vert
.CullDistanceArraySize
);
800 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
801 struct gl_shader
*shader
,
802 struct gl_context
*ctx
)
807 analyze_clip_cull_usage(prog
, shader
, ctx
,
808 &prog
->TessEval
.ClipDistanceArraySize
,
809 &prog
->TessEval
.CullDistanceArraySize
);
814 * Verify that a fragment shader executable meets all semantic requirements
816 * \param shader Fragment shader executable to be verified
819 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
820 struct gl_shader
*shader
)
825 find_assignment_visitor
frag_color("gl_FragColor");
826 find_assignment_visitor
frag_data("gl_FragData");
828 frag_color
.run(shader
->ir
);
829 frag_data
.run(shader
->ir
);
831 if (frag_color
.variable_found() && frag_data
.variable_found()) {
832 linker_error(prog
, "fragment shader writes to both "
833 "`gl_FragColor' and `gl_FragData'\n");
838 * Verify that a geometry shader executable meets all semantic requirements
840 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize and
841 * prog->Geom.CullDistanceArraySize as a side effect.
843 * \param shader Geometry shader executable to be verified
846 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
847 struct gl_shader
*shader
,
848 struct gl_context
*ctx
)
853 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
854 prog
->Geom
.VerticesIn
= num_vertices
;
856 analyze_clip_cull_usage(prog
, shader
, ctx
,
857 &prog
->Geom
.ClipDistanceArraySize
,
858 &prog
->Geom
.CullDistanceArraySize
);
862 * Check if geometry shaders emit to non-zero streams and do corresponding
866 validate_geometry_shader_emissions(struct gl_context
*ctx
,
867 struct gl_shader_program
*prog
)
869 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
870 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
871 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
872 if (emit_vertex
.error()) {
873 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
874 "stream parameter are in the range [0, %d].\n",
875 emit_vertex
.error_func(),
876 emit_vertex
.error_stream(),
877 ctx
->Const
.MaxVertexStreams
- 1);
879 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
880 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
882 /* From the ARB_gpu_shader5 spec:
884 * "Multiple vertex streams are supported only if the output primitive
885 * type is declared to be "points". A program will fail to link if it
886 * contains a geometry shader calling EmitStreamVertex() or
887 * EndStreamPrimitive() if its output primitive type is not "points".
889 * However, in the same spec:
891 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
892 * with <stream> set to zero."
896 * "The function EndPrimitive() is equivalent to calling
897 * EndStreamPrimitive() with <stream> set to zero."
899 * Since we can call EmitVertex() and EndPrimitive() when we output
900 * primitives other than points, calling EmitStreamVertex(0) or
901 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
902 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
903 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
906 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
907 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
908 "with n>0 requires point output\n");
914 validate_intrastage_arrays(struct gl_shader_program
*prog
,
915 ir_variable
*const var
,
916 ir_variable
*const existing
)
918 /* Consider the types to be "the same" if both types are arrays
919 * of the same type and one of the arrays is implicitly sized.
920 * In addition, set the type of the linked variable to the
921 * explicitly sized array.
923 if (var
->type
->is_array() && existing
->type
->is_array()) {
924 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
925 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
926 if (var
->type
->length
!= 0) {
927 if (var
->type
->length
<= existing
->data
.max_array_access
) {
928 linker_error(prog
, "%s `%s' declared as type "
929 "`%s' but outermost dimension has an index"
932 var
->name
, var
->type
->name
,
933 existing
->data
.max_array_access
);
935 existing
->type
= var
->type
;
937 } else if (existing
->type
->length
!= 0) {
938 if(existing
->type
->length
<= var
->data
.max_array_access
&&
939 !existing
->data
.from_ssbo_unsized_array
) {
940 linker_error(prog
, "%s `%s' declared as type "
941 "`%s' but outermost dimension has an index"
944 var
->name
, existing
->type
->name
,
945 var
->data
.max_array_access
);
950 /* The arrays of structs could have different glsl_type pointers but
951 * they are actually the same type. Use record_compare() to check that.
953 if (existing
->type
->fields
.array
->is_record() &&
954 var
->type
->fields
.array
->is_record() &&
955 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
964 * Perform validation of global variables used across multiple shaders
967 cross_validate_globals(struct gl_shader_program
*prog
,
968 struct gl_shader
**shader_list
,
969 unsigned num_shaders
,
972 /* Examine all of the uniforms in all of the shaders and cross validate
975 glsl_symbol_table variables
;
976 for (unsigned i
= 0; i
< num_shaders
; i
++) {
977 if (shader_list
[i
] == NULL
)
980 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
981 ir_variable
*const var
= node
->as_variable();
986 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
989 /* don't cross validate subroutine uniforms */
990 if (var
->type
->contains_subroutine())
993 /* Don't cross validate temporaries that are at global scope. These
994 * will eventually get pulled into the shaders 'main'.
996 if (var
->data
.mode
== ir_var_temporary
)
999 /* If a global with this name has already been seen, verify that the
1000 * new instance has the same type. In addition, if the globals have
1001 * initializers, the values of the initializers must be the same.
1003 ir_variable
*const existing
= variables
.get_variable(var
->name
);
1004 if (existing
!= NULL
) {
1005 /* Check if types match. Interface blocks have some special
1006 * rules so we handle those elsewhere.
1008 if (var
->type
!= existing
->type
&&
1009 !var
->is_interface_instance()) {
1010 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
1011 if (var
->type
->is_record() && existing
->type
->is_record()
1012 && existing
->type
->record_compare(var
->type
)) {
1013 existing
->type
= var
->type
;
1015 /* If it is an unsized array in a Shader Storage Block,
1016 * two different shaders can access to different elements.
1017 * Because of that, they might be converted to different
1018 * sized arrays, then check that they are compatible but
1019 * ignore the array size.
1021 if (!(var
->data
.mode
== ir_var_shader_storage
&&
1022 var
->data
.from_ssbo_unsized_array
&&
1023 existing
->data
.mode
== ir_var_shader_storage
&&
1024 existing
->data
.from_ssbo_unsized_array
&&
1025 var
->type
->gl_type
== existing
->type
->gl_type
)) {
1026 linker_error(prog
, "%s `%s' declared as type "
1027 "`%s' and type `%s'\n",
1029 var
->name
, var
->type
->name
,
1030 existing
->type
->name
);
1037 if (var
->data
.explicit_location
) {
1038 if (existing
->data
.explicit_location
1039 && (var
->data
.location
!= existing
->data
.location
)) {
1040 linker_error(prog
, "explicit locations for %s "
1041 "`%s' have differing values\n",
1042 mode_string(var
), var
->name
);
1046 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
1047 linker_error(prog
, "explicit components for %s "
1048 "`%s' have differing values\n",
1049 mode_string(var
), var
->name
);
1053 existing
->data
.location
= var
->data
.location
;
1054 existing
->data
.explicit_location
= true;
1056 /* Check if uniform with implicit location was marked explicit
1057 * by earlier shader stage. If so, mark it explicit in this stage
1058 * too to make sure later processing does not treat it as
1061 if (existing
->data
.explicit_location
) {
1062 var
->data
.location
= existing
->data
.location
;
1063 var
->data
.explicit_location
= true;
1067 /* From the GLSL 4.20 specification:
1068 * "A link error will result if two compilation units in a program
1069 * specify different integer-constant bindings for the same
1070 * opaque-uniform name. However, it is not an error to specify a
1071 * binding on some but not all declarations for the same name"
1073 if (var
->data
.explicit_binding
) {
1074 if (existing
->data
.explicit_binding
&&
1075 var
->data
.binding
!= existing
->data
.binding
) {
1076 linker_error(prog
, "explicit bindings for %s "
1077 "`%s' have differing values\n",
1078 mode_string(var
), var
->name
);
1082 existing
->data
.binding
= var
->data
.binding
;
1083 existing
->data
.explicit_binding
= true;
1086 if (var
->type
->contains_atomic() &&
1087 var
->data
.offset
!= existing
->data
.offset
) {
1088 linker_error(prog
, "offset specifications for %s "
1089 "`%s' have differing values\n",
1090 mode_string(var
), var
->name
);
1094 /* Validate layout qualifiers for gl_FragDepth.
1096 * From the AMD/ARB_conservative_depth specs:
1098 * "If gl_FragDepth is redeclared in any fragment shader in a
1099 * program, it must be redeclared in all fragment shaders in
1100 * that program that have static assignments to
1101 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1102 * fragment shaders in a single program must have the same set
1105 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1106 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1107 bool layout_differs
=
1108 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1110 if (layout_declared
&& layout_differs
) {
1112 "All redeclarations of gl_FragDepth in all "
1113 "fragment shaders in a single program must have "
1114 "the same set of qualifiers.\n");
1117 if (var
->data
.used
&& layout_differs
) {
1119 "If gl_FragDepth is redeclared with a layout "
1120 "qualifier in any fragment shader, it must be "
1121 "redeclared with the same layout qualifier in "
1122 "all fragment shaders that have assignments to "
1127 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1129 * "If a shared global has multiple initializers, the
1130 * initializers must all be constant expressions, and they
1131 * must all have the same value. Otherwise, a link error will
1132 * result. (A shared global having only one initializer does
1133 * not require that initializer to be a constant expression.)"
1135 * Previous to 4.20 the GLSL spec simply said that initializers
1136 * must have the same value. In this case of non-constant
1137 * initializers, this was impossible to determine. As a result,
1138 * no vendor actually implemented that behavior. The 4.20
1139 * behavior matches the implemented behavior of at least one other
1140 * vendor, so we'll implement that for all GLSL versions.
1142 if (var
->constant_initializer
!= NULL
) {
1143 if (existing
->constant_initializer
!= NULL
) {
1144 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1145 linker_error(prog
, "initializers for %s "
1146 "`%s' have differing values\n",
1147 mode_string(var
), var
->name
);
1151 /* If the first-seen instance of a particular uniform did
1152 * not have an initializer but a later instance does,
1153 * replace the former with the later.
1155 variables
.replace_variable(existing
->name
, var
);
1159 if (var
->data
.has_initializer
) {
1160 if (existing
->data
.has_initializer
1161 && (var
->constant_initializer
== NULL
1162 || existing
->constant_initializer
== NULL
)) {
1164 "shared global variable `%s' has multiple "
1165 "non-constant initializers.\n",
1171 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1172 linker_error(prog
, "declarations for %s `%s' have "
1173 "mismatching invariant qualifiers\n",
1174 mode_string(var
), var
->name
);
1177 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1178 linker_error(prog
, "declarations for %s `%s' have "
1179 "mismatching centroid qualifiers\n",
1180 mode_string(var
), var
->name
);
1183 if (existing
->data
.sample
!= var
->data
.sample
) {
1184 linker_error(prog
, "declarations for %s `%s` have "
1185 "mismatching sample qualifiers\n",
1186 mode_string(var
), var
->name
);
1189 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1190 linker_error(prog
, "declarations for %s `%s` have "
1191 "mismatching image format qualifiers\n",
1192 mode_string(var
), var
->name
);
1196 variables
.add_variable(var
);
1203 * Perform validation of uniforms used across multiple shader stages
1206 cross_validate_uniforms(struct gl_shader_program
*prog
)
1208 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1209 MESA_SHADER_STAGES
, true);
1213 * Accumulates the array of buffer blocks and checks that all definitions of
1214 * blocks agree on their contents.
1217 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1220 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1221 struct gl_uniform_block
*blks
= NULL
;
1222 unsigned *num_blks
= validate_ssbo
? &prog
->NumShaderStorageBlocks
:
1223 &prog
->NumUniformBlocks
;
1225 unsigned max_num_buffer_blocks
= 0;
1226 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1227 if (prog
->_LinkedShaders
[i
]) {
1228 if (validate_ssbo
) {
1229 max_num_buffer_blocks
+=
1230 prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1232 max_num_buffer_blocks
+=
1233 prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1238 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1239 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1241 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1242 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1243 InterfaceBlockStageIndex
[i
][j
] = -1;
1248 unsigned sh_num_blocks
;
1249 struct gl_uniform_block
**sh_blks
;
1250 if (validate_ssbo
) {
1251 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1252 sh_blks
= sh
->ShaderStorageBlocks
;
1254 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1255 sh_blks
= sh
->UniformBlocks
;
1258 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1259 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1263 linker_error(prog
, "buffer block `%s' has mismatching "
1264 "definitions\n", sh_blks
[j
]->Name
);
1266 for (unsigned k
= 0; k
<= i
; k
++) {
1267 delete[] InterfaceBlockStageIndex
[k
];
1272 InterfaceBlockStageIndex
[i
][index
] = j
;
1276 /* Update per stage block pointers to point to the program list.
1277 * FIXME: We should be able to free the per stage blocks here.
1279 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1280 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1281 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1283 if (stage_index
!= -1) {
1284 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1286 blks
[j
].stageref
|= (1 << i
);
1288 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1289 sh
->ShaderStorageBlocks
: sh
->UniformBlocks
;
1291 sh_blks
[stage_index
] = &blks
[j
];
1296 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1297 delete[] InterfaceBlockStageIndex
[i
];
1301 prog
->ShaderStorageBlocks
= blks
;
1303 prog
->UniformBlocks
= blks
;
1310 * Populates a shaders symbol table with all global declarations
1313 populate_symbol_table(gl_shader
*sh
)
1315 sh
->symbols
= new(sh
) glsl_symbol_table
;
1317 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1321 if ((func
= inst
->as_function()) != NULL
) {
1322 sh
->symbols
->add_function(func
);
1323 } else if ((var
= inst
->as_variable()) != NULL
) {
1324 if (var
->data
.mode
!= ir_var_temporary
)
1325 sh
->symbols
->add_variable(var
);
1332 * Remap variables referenced in an instruction tree
1334 * This is used when instruction trees are cloned from one shader and placed in
1335 * another. These trees will contain references to \c ir_variable nodes that
1336 * do not exist in the target shader. This function finds these \c ir_variable
1337 * references and replaces the references with matching variables in the target
1340 * If there is no matching variable in the target shader, a clone of the
1341 * \c ir_variable is made and added to the target shader. The new variable is
1342 * added to \b both the instruction stream and the symbol table.
1344 * \param inst IR tree that is to be processed.
1345 * \param symbols Symbol table containing global scope symbols in the
1347 * \param instructions Instruction stream where new variable declarations
1351 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1354 class remap_visitor
: public ir_hierarchical_visitor
{
1356 remap_visitor(struct gl_shader
*target
,
1359 this->target
= target
;
1360 this->symbols
= target
->symbols
;
1361 this->instructions
= target
->ir
;
1362 this->temps
= temps
;
1365 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1367 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1368 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1370 assert(var
!= NULL
);
1372 return visit_continue
;
1375 ir_variable
*const existing
=
1376 this->symbols
->get_variable(ir
->var
->name
);
1377 if (existing
!= NULL
)
1380 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1382 this->symbols
->add_variable(copy
);
1383 this->instructions
->push_head(copy
);
1387 return visit_continue
;
1391 struct gl_shader
*target
;
1392 glsl_symbol_table
*symbols
;
1393 exec_list
*instructions
;
1397 remap_visitor
v(target
, temps
);
1404 * Move non-declarations from one instruction stream to another
1406 * The intended usage pattern of this function is to pass the pointer to the
1407 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1408 * pointer) for \c last and \c false for \c make_copies on the first
1409 * call. Successive calls pass the return value of the previous call for
1410 * \c last and \c true for \c make_copies.
1412 * \param instructions Source instruction stream
1413 * \param last Instruction after which new instructions should be
1414 * inserted in the target instruction stream
1415 * \param make_copies Flag selecting whether instructions in \c instructions
1416 * should be copied (via \c ir_instruction::clone) into the
1417 * target list or moved.
1420 * The new "last" instruction in the target instruction stream. This pointer
1421 * is suitable for use as the \c last parameter of a later call to this
1425 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1426 bool make_copies
, gl_shader
*target
)
1428 hash_table
*temps
= NULL
;
1431 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1432 hash_table_pointer_compare
);
1434 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1435 if (inst
->as_function())
1438 ir_variable
*var
= inst
->as_variable();
1439 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1442 assert(inst
->as_assignment()
1444 || inst
->as_if() /* for initializers with the ?: operator */
1445 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1448 inst
= inst
->clone(target
, NULL
);
1451 hash_table_insert(temps
, inst
, var
);
1453 remap_variables(inst
, target
, temps
);
1458 last
->insert_after(inst
);
1463 hash_table_dtor(temps
);
1470 * This class is only used in link_intrastage_shaders() below but declaring
1471 * it inside that function leads to compiler warnings with some versions of
1474 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1476 array_sizing_visitor()
1477 : mem_ctx(ralloc_context(NULL
)),
1478 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1479 hash_table_pointer_compare
))
1483 ~array_sizing_visitor()
1485 hash_table_dtor(this->unnamed_interfaces
);
1486 ralloc_free(this->mem_ctx
);
1489 virtual ir_visitor_status
visit(ir_variable
*var
)
1491 const glsl_type
*type_without_array
;
1492 fixup_type(&var
->type
, var
->data
.max_array_access
,
1493 var
->data
.from_ssbo_unsized_array
);
1494 type_without_array
= var
->type
->without_array();
1495 if (var
->type
->is_interface()) {
1496 if (interface_contains_unsized_arrays(var
->type
)) {
1497 const glsl_type
*new_type
=
1498 resize_interface_members(var
->type
,
1499 var
->get_max_ifc_array_access(),
1500 var
->is_in_shader_storage_block());
1501 var
->type
= new_type
;
1502 var
->change_interface_type(new_type
);
1504 } else if (type_without_array
->is_interface()) {
1505 if (interface_contains_unsized_arrays(type_without_array
)) {
1506 const glsl_type
*new_type
=
1507 resize_interface_members(type_without_array
,
1508 var
->get_max_ifc_array_access(),
1509 var
->is_in_shader_storage_block());
1510 var
->change_interface_type(new_type
);
1511 var
->type
= update_interface_members_array(var
->type
, new_type
);
1513 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1514 /* Store a pointer to the variable in the unnamed_interfaces
1517 ir_variable
**interface_vars
= (ir_variable
**)
1518 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1519 if (interface_vars
== NULL
) {
1520 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1522 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1525 unsigned index
= ifc_type
->field_index(var
->name
);
1526 assert(index
< ifc_type
->length
);
1527 assert(interface_vars
[index
] == NULL
);
1528 interface_vars
[index
] = var
;
1530 return visit_continue
;
1534 * For each unnamed interface block that was discovered while running the
1535 * visitor, adjust the interface type to reflect the newly assigned array
1536 * sizes, and fix up the ir_variable nodes to point to the new interface
1539 void fixup_unnamed_interface_types()
1541 hash_table_call_foreach(this->unnamed_interfaces
,
1542 fixup_unnamed_interface_type
, NULL
);
1547 * If the type pointed to by \c type represents an unsized array, replace
1548 * it with a sized array whose size is determined by max_array_access.
1550 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1551 bool from_ssbo_unsized_array
)
1553 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1554 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1555 max_array_access
+ 1);
1556 assert(*type
!= NULL
);
1560 static const glsl_type
*
1561 update_interface_members_array(const glsl_type
*type
,
1562 const glsl_type
*new_interface_type
)
1564 const glsl_type
*element_type
= type
->fields
.array
;
1565 if (element_type
->is_array()) {
1566 const glsl_type
*new_array_type
=
1567 update_interface_members_array(element_type
, new_interface_type
);
1568 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1570 return glsl_type::get_array_instance(new_interface_type
,
1576 * Determine whether the given interface type contains unsized arrays (if
1577 * it doesn't, array_sizing_visitor doesn't need to process it).
1579 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1581 for (unsigned i
= 0; i
< type
->length
; i
++) {
1582 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1583 if (elem_type
->is_unsized_array())
1590 * Create a new interface type based on the given type, with unsized arrays
1591 * replaced by sized arrays whose size is determined by
1592 * max_ifc_array_access.
1594 static const glsl_type
*
1595 resize_interface_members(const glsl_type
*type
,
1596 const unsigned *max_ifc_array_access
,
1599 unsigned num_fields
= type
->length
;
1600 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1601 memcpy(fields
, type
->fields
.structure
,
1602 num_fields
* sizeof(*fields
));
1603 for (unsigned i
= 0; i
< num_fields
; i
++) {
1604 /* If SSBO last member is unsized array, we don't replace it by a sized
1607 if (is_ssbo
&& i
== (num_fields
- 1))
1608 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1611 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1614 glsl_interface_packing packing
=
1615 (glsl_interface_packing
) type
->interface_packing
;
1616 const glsl_type
*new_ifc_type
=
1617 glsl_type::get_interface_instance(fields
, num_fields
,
1618 packing
, type
->name
);
1620 return new_ifc_type
;
1623 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1626 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1627 ir_variable
**interface_vars
= (ir_variable
**) data
;
1628 unsigned num_fields
= ifc_type
->length
;
1629 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1630 memcpy(fields
, ifc_type
->fields
.structure
,
1631 num_fields
* sizeof(*fields
));
1632 bool interface_type_changed
= false;
1633 for (unsigned i
= 0; i
< num_fields
; i
++) {
1634 if (interface_vars
[i
] != NULL
&&
1635 fields
[i
].type
!= interface_vars
[i
]->type
) {
1636 fields
[i
].type
= interface_vars
[i
]->type
;
1637 interface_type_changed
= true;
1640 if (!interface_type_changed
) {
1644 glsl_interface_packing packing
=
1645 (glsl_interface_packing
) ifc_type
->interface_packing
;
1646 const glsl_type
*new_ifc_type
=
1647 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1650 for (unsigned i
= 0; i
< num_fields
; i
++) {
1651 if (interface_vars
[i
] != NULL
)
1652 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1657 * Memory context used to allocate the data in \c unnamed_interfaces.
1662 * Hash table from const glsl_type * to an array of ir_variable *'s
1663 * pointing to the ir_variables constituting each unnamed interface block.
1665 hash_table
*unnamed_interfaces
;
1669 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1673 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1674 struct gl_shader_program
*prog
,
1675 struct gl_shader
*linked_shader
,
1676 struct gl_shader
**shader_list
,
1677 unsigned num_shaders
)
1679 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1680 linked_shader
->TransformFeedback
.BufferStride
[i
] = 0;
1683 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1684 struct gl_shader
*shader
= shader_list
[i
];
1686 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1687 if (shader
->TransformFeedback
.BufferStride
[j
]) {
1688 if (linked_shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1689 shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1690 linked_shader
->TransformFeedback
.BufferStride
[j
] !=
1691 shader
->TransformFeedback
.BufferStride
[j
]) {
1693 "intrastage shaders defined with conflicting "
1694 "xfb_stride for buffer %d (%d and %d)\n", j
,
1695 linked_shader
->TransformFeedback
.BufferStride
[j
],
1696 shader
->TransformFeedback
.BufferStride
[j
]);
1700 if (shader
->TransformFeedback
.BufferStride
[j
])
1701 linked_shader
->TransformFeedback
.BufferStride
[j
] =
1702 shader
->TransformFeedback
.BufferStride
[j
];
1707 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1708 if (linked_shader
->TransformFeedback
.BufferStride
[j
]) {
1709 prog
->TransformFeedback
.BufferStride
[j
] =
1710 linked_shader
->TransformFeedback
.BufferStride
[j
];
1712 /* We will validate doubles at a later stage */
1713 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1714 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1715 "multiple of 4 or if its applied to a type that is "
1716 "or contains a double a multiple of 8.",
1717 prog
->TransformFeedback
.BufferStride
[j
]);
1721 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1722 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1724 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1725 "limit has been exceeded.");
1733 * Performs the cross-validation of tessellation control shader vertices and
1734 * layout qualifiers for the attached tessellation control shaders,
1735 * and propagates them to the linked TCS and linked shader program.
1738 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1739 struct gl_shader
*linked_shader
,
1740 struct gl_shader
**shader_list
,
1741 unsigned num_shaders
)
1743 linked_shader
->TessCtrl
.VerticesOut
= 0;
1745 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1748 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1750 * "All tessellation control shader layout declarations in a program
1751 * must specify the same output patch vertex count. There must be at
1752 * least one layout qualifier specifying an output patch vertex count
1753 * in any program containing tessellation control shaders; however,
1754 * such a declaration is not required in all tessellation control
1758 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1759 struct gl_shader
*shader
= shader_list
[i
];
1761 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1762 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1763 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1764 linker_error(prog
, "tessellation control shader defined with "
1765 "conflicting output vertex count (%d and %d)\n",
1766 linked_shader
->TessCtrl
.VerticesOut
,
1767 shader
->TessCtrl
.VerticesOut
);
1770 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1774 /* Just do the intrastage -> interstage propagation right now,
1775 * since we already know we're in the right type of shader program
1778 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1779 linker_error(prog
, "tessellation control shader didn't declare "
1780 "vertices out layout qualifier\n");
1783 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1788 * Performs the cross-validation of tessellation evaluation shader
1789 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1790 * for the attached tessellation evaluation shaders, and propagates them
1791 * to the linked TES and linked shader program.
1794 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1795 struct gl_shader
*linked_shader
,
1796 struct gl_shader
**shader_list
,
1797 unsigned num_shaders
)
1799 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1800 linked_shader
->TessEval
.Spacing
= 0;
1801 linked_shader
->TessEval
.VertexOrder
= 0;
1802 linked_shader
->TessEval
.PointMode
= -1;
1804 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1807 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1809 * "At least one tessellation evaluation shader (compilation unit) in
1810 * a program must declare a primitive mode in its input layout.
1811 * Declaration vertex spacing, ordering, and point mode identifiers is
1812 * optional. It is not required that all tessellation evaluation
1813 * shaders in a program declare a primitive mode. If spacing or
1814 * vertex ordering declarations are omitted, the tessellation
1815 * primitive generator will use equal spacing or counter-clockwise
1816 * vertex ordering, respectively. If a point mode declaration is
1817 * omitted, the tessellation primitive generator will produce lines or
1818 * triangles according to the primitive mode."
1821 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1822 struct gl_shader
*shader
= shader_list
[i
];
1824 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1825 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1826 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1827 linker_error(prog
, "tessellation evaluation shader defined with "
1828 "conflicting input primitive modes.\n");
1831 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1834 if (shader
->TessEval
.Spacing
!= 0) {
1835 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1836 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1837 linker_error(prog
, "tessellation evaluation shader defined with "
1838 "conflicting vertex spacing.\n");
1841 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1844 if (shader
->TessEval
.VertexOrder
!= 0) {
1845 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1846 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1847 linker_error(prog
, "tessellation evaluation shader defined with "
1848 "conflicting ordering.\n");
1851 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1854 if (shader
->TessEval
.PointMode
!= -1) {
1855 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1856 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1857 linker_error(prog
, "tessellation evaluation shader defined with "
1858 "conflicting point modes.\n");
1861 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1866 /* Just do the intrastage -> interstage propagation right now,
1867 * since we already know we're in the right type of shader program
1870 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1872 "tessellation evaluation shader didn't declare input "
1873 "primitive modes.\n");
1876 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1878 if (linked_shader
->TessEval
.Spacing
== 0)
1879 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1880 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1882 if (linked_shader
->TessEval
.VertexOrder
== 0)
1883 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1884 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1886 if (linked_shader
->TessEval
.PointMode
== -1)
1887 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1888 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1893 * Performs the cross-validation of layout qualifiers specified in
1894 * redeclaration of gl_FragCoord for the attached fragment shaders,
1895 * and propagates them to the linked FS and linked shader program.
1898 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1899 struct gl_shader
*linked_shader
,
1900 struct gl_shader
**shader_list
,
1901 unsigned num_shaders
)
1903 linked_shader
->redeclares_gl_fragcoord
= false;
1904 linked_shader
->uses_gl_fragcoord
= false;
1905 linked_shader
->origin_upper_left
= false;
1906 linked_shader
->pixel_center_integer
= false;
1908 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1909 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1912 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1913 struct gl_shader
*shader
= shader_list
[i
];
1914 /* From the GLSL 1.50 spec, page 39:
1916 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1917 * it must be redeclared in all the fragment shaders in that program
1918 * that have a static use gl_FragCoord."
1920 if ((linked_shader
->redeclares_gl_fragcoord
1921 && !shader
->redeclares_gl_fragcoord
1922 && shader
->uses_gl_fragcoord
)
1923 || (shader
->redeclares_gl_fragcoord
1924 && !linked_shader
->redeclares_gl_fragcoord
1925 && linked_shader
->uses_gl_fragcoord
)) {
1926 linker_error(prog
, "fragment shader defined with conflicting "
1927 "layout qualifiers for gl_FragCoord\n");
1930 /* From the GLSL 1.50 spec, page 39:
1932 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1933 * single program must have the same set of qualifiers."
1935 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1936 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1937 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1938 linker_error(prog
, "fragment shader defined with conflicting "
1939 "layout qualifiers for gl_FragCoord\n");
1942 /* Update the linked shader state. Note that uses_gl_fragcoord should
1943 * accumulate the results. The other values should replace. If there
1944 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1945 * are already known to be the same.
1947 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1948 linked_shader
->redeclares_gl_fragcoord
=
1949 shader
->redeclares_gl_fragcoord
;
1950 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1951 || shader
->uses_gl_fragcoord
;
1952 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1953 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1956 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1961 * Performs the cross-validation of geometry shader max_vertices and
1962 * primitive type layout qualifiers for the attached geometry shaders,
1963 * and propagates them to the linked GS and linked shader program.
1966 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1967 struct gl_shader
*linked_shader
,
1968 struct gl_shader
**shader_list
,
1969 unsigned num_shaders
)
1971 linked_shader
->Geom
.VerticesOut
= 0;
1972 linked_shader
->Geom
.Invocations
= 0;
1973 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1974 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1976 /* No in/out qualifiers defined for anything but GLSL 1.50+
1977 * geometry shaders so far.
1979 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1982 /* From the GLSL 1.50 spec, page 46:
1984 * "All geometry shader output layout declarations in a program
1985 * must declare the same layout and same value for
1986 * max_vertices. There must be at least one geometry output
1987 * layout declaration somewhere in a program, but not all
1988 * geometry shaders (compilation units) are required to
1992 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1993 struct gl_shader
*shader
= shader_list
[i
];
1995 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1996 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1997 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1998 linker_error(prog
, "geometry shader defined with conflicting "
2002 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
2005 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
2006 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
2007 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
2008 linker_error(prog
, "geometry shader defined with conflicting "
2012 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
2015 if (shader
->Geom
.VerticesOut
!= 0) {
2016 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
2017 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
2018 linker_error(prog
, "geometry shader defined with conflicting "
2019 "output vertex count (%d and %d)\n",
2020 linked_shader
->Geom
.VerticesOut
,
2021 shader
->Geom
.VerticesOut
);
2024 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
2027 if (shader
->Geom
.Invocations
!= 0) {
2028 if (linked_shader
->Geom
.Invocations
!= 0 &&
2029 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
2030 linker_error(prog
, "geometry shader defined with conflicting "
2031 "invocation count (%d and %d)\n",
2032 linked_shader
->Geom
.Invocations
,
2033 shader
->Geom
.Invocations
);
2036 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
2040 /* Just do the intrastage -> interstage propagation right now,
2041 * since we already know we're in the right type of shader program
2044 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
2046 "geometry shader didn't declare primitive input type\n");
2049 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
2051 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
2053 "geometry shader didn't declare primitive output type\n");
2056 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
2058 if (linked_shader
->Geom
.VerticesOut
== 0) {
2060 "geometry shader didn't declare max_vertices\n");
2063 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
2065 if (linked_shader
->Geom
.Invocations
== 0)
2066 linked_shader
->Geom
.Invocations
= 1;
2068 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
2073 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2074 * qualifiers for the attached compute shaders, and propagate them to the
2075 * linked CS and linked shader program.
2078 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2079 struct gl_shader
*linked_shader
,
2080 struct gl_shader
**shader_list
,
2081 unsigned num_shaders
)
2083 for (int i
= 0; i
< 3; i
++)
2084 linked_shader
->Comp
.LocalSize
[i
] = 0;
2086 /* This function is called for all shader stages, but it only has an effect
2087 * for compute shaders.
2089 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2092 /* From the ARB_compute_shader spec, in the section describing local size
2095 * If multiple compute shaders attached to a single program object
2096 * declare local work-group size, the declarations must be identical;
2097 * otherwise a link-time error results. Furthermore, if a program
2098 * object contains any compute shaders, at least one must contain an
2099 * input layout qualifier specifying the local work sizes of the
2100 * program, or a link-time error will occur.
2102 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2103 struct gl_shader
*shader
= shader_list
[sh
];
2105 if (shader
->Comp
.LocalSize
[0] != 0) {
2106 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
2107 for (int i
= 0; i
< 3; i
++) {
2108 if (linked_shader
->Comp
.LocalSize
[i
] !=
2109 shader
->Comp
.LocalSize
[i
]) {
2110 linker_error(prog
, "compute shader defined with conflicting "
2116 for (int i
= 0; i
< 3; i
++)
2117 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
2121 /* Just do the intrastage -> interstage propagation right now,
2122 * since we already know we're in the right type of shader program
2125 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
2126 linker_error(prog
, "compute shader didn't declare local size\n");
2129 for (int i
= 0; i
< 3; i
++)
2130 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
2135 * Combine a group of shaders for a single stage to generate a linked shader
2138 * If this function is supplied a single shader, it is cloned, and the new
2139 * shader is returned.
2141 static struct gl_shader
*
2142 link_intrastage_shaders(void *mem_ctx
,
2143 struct gl_context
*ctx
,
2144 struct gl_shader_program
*prog
,
2145 struct gl_shader
**shader_list
,
2146 unsigned num_shaders
)
2148 struct gl_uniform_block
*ubo_blocks
= NULL
;
2149 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2150 unsigned num_ubo_blocks
= 0;
2151 unsigned num_ssbo_blocks
= 0;
2153 /* Check that global variables defined in multiple shaders are consistent.
2155 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
2156 if (!prog
->LinkStatus
)
2159 /* Check that interface blocks defined in multiple shaders are consistent.
2161 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2163 if (!prog
->LinkStatus
)
2166 /* Link up uniform blocks defined within this stage. */
2167 link_uniform_blocks(mem_ctx
, ctx
, prog
, shader_list
, num_shaders
,
2168 &ubo_blocks
, &num_ubo_blocks
, &ssbo_blocks
,
2171 if (!prog
->LinkStatus
)
2174 /* Check that there is only a single definition of each function signature
2175 * across all shaders.
2177 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2178 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2179 ir_function
*const f
= node
->as_function();
2184 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2185 ir_function
*const other
=
2186 shader_list
[j
]->symbols
->get_function(f
->name
);
2188 /* If the other shader has no function (and therefore no function
2189 * signatures) with the same name, skip to the next shader.
2194 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2195 if (!sig
->is_defined
|| sig
->is_builtin())
2198 ir_function_signature
*other_sig
=
2199 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2201 if ((other_sig
!= NULL
) && other_sig
->is_defined
2202 && !other_sig
->is_builtin()) {
2203 linker_error(prog
, "function `%s' is multiply defined\n",
2212 /* Find the shader that defines main, and make a clone of it.
2214 * Starting with the clone, search for undefined references. If one is
2215 * found, find the shader that defines it. Clone the reference and add
2216 * it to the shader. Repeat until there are no undefined references or
2217 * until a reference cannot be resolved.
2219 gl_shader
*main
= NULL
;
2220 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2221 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2222 main
= shader_list
[i
];
2228 linker_error(prog
, "%s shader lacks `main'\n",
2229 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2233 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2234 linked
->ir
= new(linked
) exec_list
;
2235 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2237 /* Copy ubo blocks to linked shader list */
2238 linked
->UniformBlocks
=
2239 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2240 ralloc_steal(linked
, ubo_blocks
);
2241 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2242 linked
->UniformBlocks
[i
] = &ubo_blocks
[i
];
2244 linked
->NumUniformBlocks
= num_ubo_blocks
;
2246 /* Copy ssbo blocks to linked shader list */
2247 linked
->ShaderStorageBlocks
=
2248 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2249 ralloc_steal(linked
, ssbo_blocks
);
2250 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2251 linked
->ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2253 linked
->NumShaderStorageBlocks
= num_ssbo_blocks
;
2255 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2256 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2257 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2258 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2259 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2260 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2263 populate_symbol_table(linked
);
2265 /* The pointer to the main function in the final linked shader (i.e., the
2266 * copy of the original shader that contained the main function).
2268 ir_function_signature
*const main_sig
=
2269 _mesa_get_main_function_signature(linked
);
2271 /* Move any instructions other than variable declarations or function
2272 * declarations into main.
2274 exec_node
*insertion_point
=
2275 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2278 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2279 if (shader_list
[i
] == main
)
2282 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2283 insertion_point
, true, linked
);
2286 /* Check if any shader needs built-in functions. */
2287 bool need_builtins
= false;
2288 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2289 if (shader_list
[i
]->uses_builtin_functions
) {
2290 need_builtins
= true;
2296 if (need_builtins
) {
2297 /* Make a temporary array one larger than shader_list, which will hold
2298 * the built-in function shader as well.
2300 gl_shader
**linking_shaders
= (gl_shader
**)
2301 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2303 ok
= linking_shaders
!= NULL
;
2306 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2307 _mesa_glsl_initialize_builtin_functions();
2308 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2310 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2312 free(linking_shaders
);
2314 _mesa_error_no_memory(__func__
);
2317 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2322 _mesa_delete_shader(ctx
, linked
);
2326 /* At this point linked should contain all of the linked IR, so
2327 * validate it to make sure nothing went wrong.
2329 validate_ir_tree(linked
->ir
);
2331 /* Set the size of geometry shader input arrays */
2332 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2333 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2334 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2335 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2336 ir
->accept(&input_resize_visitor
);
2340 if (ctx
->Const
.VertexID_is_zero_based
)
2341 lower_vertex_id(linked
);
2343 /* Validate correct usage of barrier() in the tess control shader */
2344 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2345 barrier_use_visitor
visitor(prog
);
2346 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2347 ir
->accept(&visitor
);
2351 /* Make a pass over all variable declarations to ensure that arrays with
2352 * unspecified sizes have a size specified. The size is inferred from the
2353 * max_array_access field.
2355 array_sizing_visitor v
;
2357 v
.fixup_unnamed_interface_types();
2363 * Update the sizes of linked shader uniform arrays to the maximum
2366 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2368 * If one or more elements of an array are active,
2369 * GetActiveUniform will return the name of the array in name,
2370 * subject to the restrictions listed above. The type of the array
2371 * is returned in type. The size parameter contains the highest
2372 * array element index used, plus one. The compiler or linker
2373 * determines the highest index used. There will be only one
2374 * active uniform reported by the GL per uniform array.
2378 update_array_sizes(struct gl_shader_program
*prog
)
2380 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2381 if (prog
->_LinkedShaders
[i
] == NULL
)
2384 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2385 ir_variable
*const var
= node
->as_variable();
2387 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2388 !var
->type
->is_array())
2391 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2392 * will not be eliminated. Since we always do std140, just
2393 * don't resize arrays in UBOs.
2395 * Atomic counters are supposed to get deterministic
2396 * locations assigned based on the declaration ordering and
2397 * sizes, array compaction would mess that up.
2399 * Subroutine uniforms are not removed.
2401 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2402 var
->type
->contains_subroutine())
2405 unsigned int size
= var
->data
.max_array_access
;
2406 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2407 if (prog
->_LinkedShaders
[j
] == NULL
)
2410 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2411 ir_variable
*other_var
= node2
->as_variable();
2415 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2416 other_var
->data
.max_array_access
> size
) {
2417 size
= other_var
->data
.max_array_access
;
2422 if (size
+ 1 != var
->type
->length
) {
2423 /* If this is a built-in uniform (i.e., it's backed by some
2424 * fixed-function state), adjust the number of state slots to
2425 * match the new array size. The number of slots per array entry
2426 * is not known. It seems safe to assume that the total number of
2427 * slots is an integer multiple of the number of array elements.
2428 * Determine the number of slots per array element by dividing by
2429 * the old (total) size.
2431 const unsigned num_slots
= var
->get_num_state_slots();
2432 if (num_slots
> 0) {
2433 var
->set_num_state_slots((size
+ 1)
2434 * (num_slots
/ var
->type
->length
));
2437 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2439 /* FINISHME: We should update the types of array
2440 * dereferences of this variable now.
2448 * Resize tessellation evaluation per-vertex inputs to the size of
2449 * tessellation control per-vertex outputs.
2452 resize_tes_inputs(struct gl_context
*ctx
,
2453 struct gl_shader_program
*prog
)
2455 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2458 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2459 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2461 /* If no control shader is present, then the TES inputs are statically
2462 * sized to MaxPatchVertices; the actual size of the arrays won't be
2463 * known until draw time.
2465 const int num_vertices
= tcs
2466 ? tcs
->TessCtrl
.VerticesOut
2467 : ctx
->Const
.MaxPatchVertices
;
2469 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2470 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2471 ir
->accept(&input_resize_visitor
);
2475 /* Convert the gl_PatchVerticesIn system value into a constant, since
2476 * the value is known at this point.
2478 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2479 ir_variable
*var
= ir
->as_variable();
2480 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2481 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2482 void *mem_ctx
= ralloc_parent(var
);
2483 var
->data
.mode
= ir_var_auto
;
2484 var
->data
.location
= 0;
2485 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2492 * Find a contiguous set of available bits in a bitmask.
2494 * \param used_mask Bits representing used (1) and unused (0) locations
2495 * \param needed_count Number of contiguous bits needed.
2498 * Base location of the available bits on success or -1 on failure.
2501 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2503 unsigned needed_mask
= (1 << needed_count
) - 1;
2504 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2506 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2507 * cannot optimize possibly infinite loops" for the loop below.
2509 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2512 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2513 if ((needed_mask
& ~used_mask
) == needed_mask
)
2524 * Assign locations for either VS inputs or FS outputs
2526 * \param prog Shader program whose variables need locations assigned
2527 * \param constants Driver specific constant values for the program.
2528 * \param target_index Selector for the program target to receive location
2529 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2530 * \c MESA_SHADER_FRAGMENT.
2533 * If locations are successfully assigned, true is returned. Otherwise an
2534 * error is emitted to the shader link log and false is returned.
2537 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2538 struct gl_constants
*constants
,
2539 unsigned target_index
)
2541 /* Maximum number of generic locations. This corresponds to either the
2542 * maximum number of draw buffers or the maximum number of generic
2545 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2546 constants
->Program
[target_index
].MaxAttribs
:
2547 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2549 /* Mark invalid locations as being used.
2551 unsigned used_locations
= (max_index
>= 32)
2552 ? ~0 : ~((1 << max_index
) - 1);
2553 unsigned double_storage_locations
= 0;
2555 assert((target_index
== MESA_SHADER_VERTEX
)
2556 || (target_index
== MESA_SHADER_FRAGMENT
));
2558 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2562 /* Operate in a total of four passes.
2564 * 1. Invalidate the location assignments for all vertex shader inputs.
2566 * 2. Assign locations for inputs that have user-defined (via
2567 * glBindVertexAttribLocation) locations and outputs that have
2568 * user-defined locations (via glBindFragDataLocation).
2570 * 3. Sort the attributes without assigned locations by number of slots
2571 * required in decreasing order. Fragmentation caused by attribute
2572 * locations assigned by the application may prevent large attributes
2573 * from having enough contiguous space.
2575 * 4. Assign locations to any inputs without assigned locations.
2578 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2579 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2581 const enum ir_variable_mode direction
=
2582 (target_index
== MESA_SHADER_VERTEX
)
2583 ? ir_var_shader_in
: ir_var_shader_out
;
2586 /* Temporary storage for the set of attributes that need locations assigned.
2592 /* Used below in the call to qsort. */
2593 static int compare(const void *a
, const void *b
)
2595 const temp_attr
*const l
= (const temp_attr
*) a
;
2596 const temp_attr
*const r
= (const temp_attr
*) b
;
2598 /* Reversed because we want a descending order sort below. */
2599 return r
->slots
- l
->slots
;
2602 assert(max_index
<= 32);
2604 unsigned num_attr
= 0;
2606 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2607 ir_variable
*const var
= node
->as_variable();
2609 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2612 if (var
->data
.explicit_location
) {
2613 var
->data
.is_unmatched_generic_inout
= 0;
2614 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2615 || (var
->data
.location
< 0)) {
2617 "invalid explicit location %d specified for `%s'\n",
2618 (var
->data
.location
< 0)
2619 ? var
->data
.location
2620 : var
->data
.location
- generic_base
,
2624 } else if (target_index
== MESA_SHADER_VERTEX
) {
2627 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2628 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2629 var
->data
.location
= binding
;
2630 var
->data
.is_unmatched_generic_inout
= 0;
2632 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2636 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2637 assert(binding
>= FRAG_RESULT_DATA0
);
2638 var
->data
.location
= binding
;
2639 var
->data
.is_unmatched_generic_inout
= 0;
2641 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2642 var
->data
.index
= index
;
2647 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2649 * "Output binding assignments will cause LinkProgram to fail:
2651 * If the program has an active output assigned to a location greater
2652 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2653 * an active output assigned an index greater than or equal to one;"
2655 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2656 var
->data
.location
- generic_base
>=
2657 (int) constants
->MaxDualSourceDrawBuffers
) {
2659 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2660 "with index %u for %s\n",
2661 var
->data
.location
- generic_base
, var
->data
.index
,
2666 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2668 /* If the variable is not a built-in and has a location statically
2669 * assigned in the shader (presumably via a layout qualifier), make sure
2670 * that it doesn't collide with other assigned locations. Otherwise,
2671 * add it to the list of variables that need linker-assigned locations.
2673 if (var
->data
.location
!= -1) {
2674 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2675 /* From page 61 of the OpenGL 4.0 spec:
2677 * "LinkProgram will fail if the attribute bindings assigned
2678 * by BindAttribLocation do not leave not enough space to
2679 * assign a location for an active matrix attribute or an
2680 * active attribute array, both of which require multiple
2681 * contiguous generic attributes."
2683 * I think above text prohibits the aliasing of explicit and
2684 * automatic assignments. But, aliasing is allowed in manual
2685 * assignments of attribute locations. See below comments for
2688 * From OpenGL 4.0 spec, page 61:
2690 * "It is possible for an application to bind more than one
2691 * attribute name to the same location. This is referred to as
2692 * aliasing. This will only work if only one of the aliased
2693 * attributes is active in the executable program, or if no
2694 * path through the shader consumes more than one attribute of
2695 * a set of attributes aliased to the same location. A link
2696 * error can occur if the linker determines that every path
2697 * through the shader consumes multiple aliased attributes,
2698 * but implementations are not required to generate an error
2701 * From GLSL 4.30 spec, page 54:
2703 * "A program will fail to link if any two non-vertex shader
2704 * input variables are assigned to the same location. For
2705 * vertex shaders, multiple input variables may be assigned
2706 * to the same location using either layout qualifiers or via
2707 * the OpenGL API. However, such aliasing is intended only to
2708 * support vertex shaders where each execution path accesses
2709 * at most one input per each location. Implementations are
2710 * permitted, but not required, to generate link-time errors
2711 * if they detect that every path through the vertex shader
2712 * executable accesses multiple inputs assigned to any single
2713 * location. For all shader types, a program will fail to link
2714 * if explicit location assignments leave the linker unable
2715 * to find space for other variables without explicit
2718 * From OpenGL ES 3.0 spec, page 56:
2720 * "Binding more than one attribute name to the same location
2721 * is referred to as aliasing, and is not permitted in OpenGL
2722 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2723 * fail when this condition exists. However, aliasing is
2724 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2725 * This will only work if only one of the aliased attributes
2726 * is active in the executable program, or if no path through
2727 * the shader consumes more than one attribute of a set of
2728 * attributes aliased to the same location. A link error can
2729 * occur if the linker determines that every path through the
2730 * shader consumes multiple aliased attributes, but implemen-
2731 * tations are not required to generate an error in this case."
2733 * After looking at above references from OpenGL, OpenGL ES and
2734 * GLSL specifications, we allow aliasing of vertex input variables
2735 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2737 * NOTE: This is not required by the spec but its worth mentioning
2738 * here that we're not doing anything to make sure that no path
2739 * through the vertex shader executable accesses multiple inputs
2740 * assigned to any single location.
2743 /* Mask representing the contiguous slots that will be used by
2746 const unsigned attr
= var
->data
.location
- generic_base
;
2747 const unsigned use_mask
= (1 << slots
) - 1;
2748 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2749 ? "vertex shader input" : "fragment shader output";
2751 /* Generate a link error if the requested locations for this
2752 * attribute exceed the maximum allowed attribute location.
2754 if (attr
+ slots
> max_index
) {
2756 "insufficient contiguous locations "
2757 "available for %s `%s' %d %d %d\n", string
,
2758 var
->name
, used_locations
, use_mask
, attr
);
2762 /* Generate a link error if the set of bits requested for this
2763 * attribute overlaps any previously allocated bits.
2765 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2766 if (target_index
== MESA_SHADER_FRAGMENT
||
2767 (prog
->IsES
&& prog
->Version
>= 300)) {
2769 "overlapping location is assigned "
2770 "to %s `%s' %d %d %d\n", string
,
2771 var
->name
, used_locations
, use_mask
, attr
);
2774 linker_warning(prog
,
2775 "overlapping location is assigned "
2776 "to %s `%s' %d %d %d\n", string
,
2777 var
->name
, used_locations
, use_mask
, attr
);
2781 used_locations
|= (use_mask
<< attr
);
2783 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2785 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2786 * active attribute variables may fail to link, unless
2787 * device-dependent optimizations are able to make the program
2788 * fit within available hardware resources. For the purposes
2789 * of this test, attribute variables of the type dvec3, dvec4,
2790 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2791 * count as consuming twice as many attributes as equivalent
2792 * single-precision types. While these types use the same number
2793 * of generic attributes as their single-precision equivalents,
2794 * implementations are permitted to consume two single-precision
2795 * vectors of internal storage for each three- or four-component
2796 * double-precision vector."
2798 * Mark this attribute slot as taking up twice as much space
2799 * so we can count it properly against limits. According to
2800 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2801 * is optional behavior, but it seems preferable.
2803 if (var
->type
->without_array()->is_dual_slot_double())
2804 double_storage_locations
|= (use_mask
<< attr
);
2810 if (num_attr
>= max_index
) {
2811 linker_error(prog
, "too many %s (max %u)",
2812 target_index
== MESA_SHADER_VERTEX
?
2813 "vertex shader inputs" : "fragment shader outputs",
2817 to_assign
[num_attr
].slots
= slots
;
2818 to_assign
[num_attr
].var
= var
;
2822 if (target_index
== MESA_SHADER_VERTEX
) {
2823 unsigned total_attribs_size
=
2824 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2825 _mesa_bitcount(double_storage_locations
);
2826 if (total_attribs_size
> max_index
) {
2828 "attempt to use %d vertex attribute slots only %d available ",
2829 total_attribs_size
, max_index
);
2834 /* If all of the attributes were assigned locations by the application (or
2835 * are built-in attributes with fixed locations), return early. This should
2836 * be the common case.
2841 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2843 if (target_index
== MESA_SHADER_VERTEX
) {
2844 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2845 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2846 * reserved to prevent it from being automatically allocated below.
2848 find_deref_visitor
find("gl_Vertex");
2850 if (find
.variable_found())
2851 used_locations
|= (1 << 0);
2854 for (unsigned i
= 0; i
< num_attr
; i
++) {
2855 /* Mask representing the contiguous slots that will be used by this
2858 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2860 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2863 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2864 ? "vertex shader input" : "fragment shader output";
2867 "insufficient contiguous locations "
2868 "available for %s `%s'\n",
2869 string
, to_assign
[i
].var
->name
);
2873 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2874 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2875 used_locations
|= (use_mask
<< location
);
2882 * Match explicit locations of outputs to inputs and deactivate the
2883 * unmatch flag if found so we don't optimise them away.
2886 match_explicit_outputs_to_inputs(gl_shader
*producer
,
2887 gl_shader
*consumer
)
2889 glsl_symbol_table parameters
;
2890 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2893 /* Find all shader outputs in the "producer" stage.
2895 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2896 ir_variable
*const var
= node
->as_variable();
2898 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2901 if (var
->data
.explicit_location
&&
2902 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2903 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2904 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2905 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2909 /* Match inputs to outputs */
2910 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2911 ir_variable
*const input
= node
->as_variable();
2913 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2916 ir_variable
*output
= NULL
;
2917 if (input
->data
.explicit_location
2918 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2919 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2920 [input
->data
.location_frac
];
2922 if (output
!= NULL
){
2923 input
->data
.is_unmatched_generic_inout
= 0;
2924 output
->data
.is_unmatched_generic_inout
= 0;
2931 * Store the gl_FragDepth layout in the gl_shader_program struct.
2934 store_fragdepth_layout(struct gl_shader_program
*prog
)
2936 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2940 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2942 /* We don't look up the gl_FragDepth symbol directly because if
2943 * gl_FragDepth is not used in the shader, it's removed from the IR.
2944 * However, the symbol won't be removed from the symbol table.
2946 * We're only interested in the cases where the variable is NOT removed
2949 foreach_in_list(ir_instruction
, node
, ir
) {
2950 ir_variable
*const var
= node
->as_variable();
2952 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2956 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2957 switch (var
->data
.depth_layout
) {
2958 case ir_depth_layout_none
:
2959 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2961 case ir_depth_layout_any
:
2962 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2964 case ir_depth_layout_greater
:
2965 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2967 case ir_depth_layout_less
:
2968 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2970 case ir_depth_layout_unchanged
:
2971 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2982 * Validate the resources used by a program versus the implementation limits
2985 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2987 unsigned total_uniform_blocks
= 0;
2988 unsigned total_shader_storage_blocks
= 0;
2990 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2991 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2996 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2997 linker_error(prog
, "Too many %s shader texture samplers\n",
2998 _mesa_shader_stage_to_string(i
));
3001 if (sh
->num_uniform_components
>
3002 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3003 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3004 linker_warning(prog
, "Too many %s shader default uniform block "
3005 "components, but the driver will try to optimize "
3006 "them out; this is non-portable out-of-spec "
3008 _mesa_shader_stage_to_string(i
));
3010 linker_error(prog
, "Too many %s shader default uniform block "
3012 _mesa_shader_stage_to_string(i
));
3016 if (sh
->num_combined_uniform_components
>
3017 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3018 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3019 linker_warning(prog
, "Too many %s shader uniform components, "
3020 "but the driver will try to optimize them out; "
3021 "this is non-portable out-of-spec behavior\n",
3022 _mesa_shader_stage_to_string(i
));
3024 linker_error(prog
, "Too many %s shader uniform components\n",
3025 _mesa_shader_stage_to_string(i
));
3029 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3030 total_uniform_blocks
+= sh
->NumUniformBlocks
;
3032 const unsigned max_uniform_blocks
=
3033 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3034 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
3035 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3036 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
3037 max_uniform_blocks
);
3040 const unsigned max_shader_storage_blocks
=
3041 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3042 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
3043 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3044 _mesa_shader_stage_to_string(i
),
3045 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
3049 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3050 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3051 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3054 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3055 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3056 total_shader_storage_blocks
,
3057 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3060 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
3061 if (prog
->UniformBlocks
[i
].UniformBufferSize
>
3062 ctx
->Const
.MaxUniformBlockSize
) {
3063 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3064 prog
->UniformBlocks
[i
].Name
,
3065 prog
->UniformBlocks
[i
].UniformBufferSize
,
3066 ctx
->Const
.MaxUniformBlockSize
);
3070 for (unsigned i
= 0; i
< prog
->NumShaderStorageBlocks
; i
++) {
3071 if (prog
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3072 ctx
->Const
.MaxShaderStorageBlockSize
) {
3073 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3074 prog
->ShaderStorageBlocks
[i
].Name
,
3075 prog
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3076 ctx
->Const
.MaxShaderStorageBlockSize
);
3082 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3084 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3085 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3090 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3091 if (sh
->SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3094 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3099 sh
->NumSubroutineUniforms
++;
3101 if (sh
->NumSubroutineFunctions
== 0) {
3102 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3105 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3106 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3107 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3108 if (fn
->types
[k
] == uni
->type
) {
3114 uni
->num_compatible_subroutines
= count
;
3120 check_subroutine_resources(struct gl_shader_program
*prog
)
3122 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3123 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3126 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3127 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3128 _mesa_shader_stage_to_string(i
));
3133 * Validate shader image resources.
3136 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3138 unsigned total_image_units
= 0;
3139 unsigned fragment_outputs
= 0;
3140 unsigned total_shader_storage_blocks
= 0;
3142 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3145 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3146 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3149 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3150 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3151 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3152 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3154 total_image_units
+= sh
->NumImages
;
3155 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3157 if (i
== MESA_SHADER_FRAGMENT
) {
3158 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3159 ir_variable
*var
= node
->as_variable();
3160 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3161 /* since there are no double fs outputs - pass false */
3162 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3168 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3169 linker_error(prog
, "Too many combined image uniforms\n");
3171 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3172 ctx
->Const
.MaxCombinedShaderOutputResources
)
3173 linker_error(prog
, "Too many combined image uniforms, shader storage "
3174 " buffers and fragment outputs\n");
3179 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3180 * for a variable, checks for overlaps between other uniforms using explicit
3184 reserve_explicit_locations(struct gl_shader_program
*prog
,
3185 string_to_uint_map
*map
, ir_variable
*var
)
3187 unsigned slots
= var
->type
->uniform_locations();
3188 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3189 unsigned return_value
= slots
;
3191 /* Resize remap table if locations do not fit in the current one. */
3192 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3193 prog
->UniformRemapTable
=
3194 reralloc(prog
, prog
->UniformRemapTable
,
3195 gl_uniform_storage
*,
3198 if (!prog
->UniformRemapTable
) {
3199 linker_error(prog
, "Out of memory during linking.\n");
3203 /* Initialize allocated space. */
3204 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3205 prog
->UniformRemapTable
[i
] = NULL
;
3207 prog
->NumUniformRemapTable
= max_loc
+ 1;
3210 for (unsigned i
= 0; i
< slots
; i
++) {
3211 unsigned loc
= var
->data
.location
+ i
;
3213 /* Check if location is already used. */
3214 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3216 /* Possibly same uniform from a different stage, this is ok. */
3218 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3223 /* ARB_explicit_uniform_location specification states:
3225 * "No two default-block uniform variables in the program can have
3226 * the same location, even if they are unused, otherwise a compiler
3227 * or linker error will be generated."
3230 "location qualifier for uniform %s overlaps "
3231 "previously used location\n",
3236 /* Initialize location as inactive before optimization
3237 * rounds and location assignment.
3239 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3242 /* Note, base location used for arrays. */
3243 map
->put(var
->data
.location
, var
->name
);
3245 return return_value
;
3249 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3250 struct gl_shader
*sh
,
3253 unsigned slots
= var
->type
->uniform_locations();
3254 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3256 /* Resize remap table if locations do not fit in the current one. */
3257 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3258 sh
->SubroutineUniformRemapTable
=
3259 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3260 gl_uniform_storage
*,
3263 if (!sh
->SubroutineUniformRemapTable
) {
3264 linker_error(prog
, "Out of memory during linking.\n");
3268 /* Initialize allocated space. */
3269 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3270 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3272 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3275 for (unsigned i
= 0; i
< slots
; i
++) {
3276 unsigned loc
= var
->data
.location
+ i
;
3278 /* Check if location is already used. */
3279 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3281 /* ARB_explicit_uniform_location specification states:
3282 * "No two subroutine uniform variables can have the same location
3283 * in the same shader stage, otherwise a compiler or linker error
3284 * will be generated."
3287 "location qualifier for uniform %s overlaps "
3288 "previously used location\n",
3293 /* Initialize location as inactive before optimization
3294 * rounds and location assignment.
3296 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3302 * Check and reserve all explicit uniform locations, called before
3303 * any optimizations happen to handle also inactive uniforms and
3304 * inactive array elements that may get trimmed away.
3307 check_explicit_uniform_locations(struct gl_context
*ctx
,
3308 struct gl_shader_program
*prog
)
3310 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3313 /* This map is used to detect if overlapping explicit locations
3314 * occur with the same uniform (from different stage) or a different one.
3316 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3319 linker_error(prog
, "Out of memory during linking.\n");
3323 unsigned entries_total
= 0;
3324 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3325 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3330 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3331 ir_variable
*var
= node
->as_variable();
3332 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3335 if (var
->data
.explicit_location
) {
3337 if (var
->type
->without_array()->is_subroutine())
3338 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3340 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3344 entries_total
+= slots
;
3355 struct empty_uniform_block
*current_block
= NULL
;
3357 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3358 /* We found empty space in UniformRemapTable. */
3359 if (prog
->UniformRemapTable
[i
] == NULL
) {
3360 /* We've found the beginning of a new continous block of empty slots */
3361 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3362 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3363 current_block
->start
= i
;
3364 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3365 ¤t_block
->link
);
3368 /* The current block continues, so we simply increment its slots */
3369 current_block
->slots
++;
3374 return entries_total
;
3378 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3379 GLenum type
, const char *name
)
3381 bool found_interface
= false;
3382 unsigned block_name_len
= 0;
3383 const char *block_name_dot
= strchr(name
, '.');
3385 /* These rules only apply to buffer variables. So we return
3386 * true for the rest of types.
3388 if (type
!= GL_BUFFER_VARIABLE
)
3391 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
3392 const char *block_name
= shProg
->ShaderStorageBlocks
[i
].Name
;
3393 block_name_len
= strlen(block_name
);
3395 const char *block_square_bracket
= strchr(block_name
, '[');
3396 if (block_square_bracket
) {
3397 /* The block is part of an array of named interfaces,
3398 * for the name comparison we ignore the "[x]" part.
3400 block_name_len
-= strlen(block_square_bracket
);
3403 if (block_name_dot
) {
3404 /* Check if the variable name starts with the interface
3405 * name. The interface name (if present) should have the
3406 * length than the interface block name we are comparing to.
3408 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3409 if (len
!= block_name_len
)
3413 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3414 found_interface
= true;
3419 /* We remove the interface name from the buffer variable name,
3420 * including the dot that follows it.
3422 if (found_interface
)
3423 name
= name
+ block_name_len
+ 1;
3425 /* The ARB_program_interface_query spec says:
3427 * "For an active shader storage block member declared as an array, an
3428 * entry will be generated only for the first array element, regardless
3429 * of its type. For arrays of aggregate types, the enumeration rules
3430 * are applied recursively for the single enumerated array element."
3432 const char *struct_first_dot
= strchr(name
, '.');
3433 const char *first_square_bracket
= strchr(name
, '[');
3435 /* The buffer variable is on top level and it is not an array */
3436 if (!first_square_bracket
) {
3438 /* The shader storage block member is a struct, then generate the entry */
3439 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3442 /* Shader storage block member is an array, only generate an entry for the
3443 * first array element.
3445 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3453 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3454 const void *data
, uint8_t stages
)
3458 /* If resource already exists, do not add it again. */
3459 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3460 if (prog
->ProgramResourceList
[i
].Data
== data
)
3463 prog
->ProgramResourceList
=
3465 prog
->ProgramResourceList
,
3466 gl_program_resource
,
3467 prog
->NumProgramResourceList
+ 1);
3469 if (!prog
->ProgramResourceList
) {
3470 linker_error(prog
, "Out of memory during linking.\n");
3474 struct gl_program_resource
*res
=
3475 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3479 res
->StageReferences
= stages
;
3481 prog
->NumProgramResourceList
++;
3486 /* Function checks if a variable var is a packed varying and
3487 * if given name is part of packed varying's list.
3489 * If a variable is a packed varying, it has a name like
3490 * 'packed:a,b,c' where a, b and c are separate variables.
3493 included_in_packed_varying(ir_variable
*var
, const char *name
)
3495 if (strncmp(var
->name
, "packed:", 7) != 0)
3498 char *list
= strdup(var
->name
+ 7);
3503 char *token
= strtok_r(list
, ",", &saveptr
);
3505 if (strcmp(token
, name
) == 0) {
3509 token
= strtok_r(NULL
, ",", &saveptr
);
3516 * Function builds a stage reference bitmask from variable name.
3519 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3524 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3525 * used for reference mask in gl_program_resource will need to be changed.
3527 assert(MESA_SHADER_STAGES
< 8);
3529 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3530 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3534 /* Shader symbol table may contain variables that have
3535 * been optimized away. Search IR for the variable instead.
3537 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3538 ir_variable
*var
= node
->as_variable();
3540 unsigned baselen
= strlen(var
->name
);
3542 if (included_in_packed_varying(var
, name
)) {
3547 /* Type needs to match if specified, otherwise we might
3548 * pick a variable with same name but different interface.
3550 if (var
->data
.mode
!= mode
)
3553 if (strncmp(var
->name
, name
, baselen
) == 0) {
3554 /* Check for exact name matches but also check for arrays and
3557 if (name
[baselen
] == '\0' ||
3558 name
[baselen
] == '[' ||
3559 name
[baselen
] == '.') {
3571 * Create gl_shader_variable from ir_variable class.
3573 static gl_shader_variable
*
3574 create_shader_variable(struct gl_shader_program
*shProg
,
3575 const ir_variable
*in
,
3576 const char *name
, const glsl_type
*type
,
3577 bool use_implicit_location
, int location
)
3579 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3583 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3584 * expect to see gl_VertexID in the program resource list. Pretend.
3586 if (in
->data
.mode
== ir_var_system_value
&&
3587 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3588 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3590 out
->name
= ralloc_strdup(shProg
, name
);
3596 /* The ARB_program_interface_query spec says:
3598 * "Not all active variables are assigned valid locations; the
3599 * following variables will have an effective location of -1:
3601 * * uniforms declared as atomic counters;
3603 * * members of a uniform block;
3605 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3607 * * inputs or outputs not declared with a "location" layout
3608 * qualifier, except for vertex shader inputs and fragment shader
3611 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3612 is_gl_identifier(in
->name
) ||
3613 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3616 out
->location
= location
;
3620 out
->component
= in
->data
.location_frac
;
3621 out
->index
= in
->data
.index
;
3622 out
->patch
= in
->data
.patch
;
3623 out
->mode
= in
->data
.mode
;
3629 add_shader_variable(struct gl_shader_program
*shProg
, unsigned stage_mask
,
3630 GLenum programInterface
, ir_variable
*var
,
3631 const char *name
, const glsl_type
*type
,
3632 bool use_implicit_location
, int location
)
3634 const bool is_vertex_input
=
3635 programInterface
== GL_PROGRAM_INPUT
&&
3636 stage_mask
== MESA_SHADER_VERTEX
;
3638 switch (type
->base_type
) {
3639 case GLSL_TYPE_STRUCT
: {
3640 /* The ARB_program_interface_query spec says:
3642 * "For an active variable declared as a structure, a separate entry
3643 * will be generated for each active structure member. The name of
3644 * each entry is formed by concatenating the name of the structure,
3645 * the "." character, and the name of the structure member. If a
3646 * structure member to enumerate is itself a structure or array,
3647 * these enumeration rules are applied recursively."
3649 unsigned field_location
= location
;
3650 for (unsigned i
= 0; i
< type
->length
; i
++) {
3651 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3652 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3653 if (!add_shader_variable(shProg
, stage_mask
, programInterface
,
3654 var
, field_name
, field
->type
,
3655 use_implicit_location
, field_location
))
3659 field
->type
->count_attribute_slots(is_vertex_input
);
3665 /* Issue #16 of the ARB_program_interface_query spec says:
3667 * "* If a variable is a member of an interface block without an
3668 * instance name, it is enumerated using just the variable name.
3670 * * If a variable is a member of an interface block with an instance
3671 * name, it is enumerated as "BlockName.Member", where "BlockName" is
3672 * the name of the interface block (not the instance name) and
3673 * "Member" is the name of the variable."
3675 const char *prefixed_name
= (var
->data
.from_named_ifc_block
&&
3676 !is_gl_identifier(var
->name
))
3677 ? ralloc_asprintf(shProg
, "%s.%s", var
->get_interface_type()->name
,
3681 /* The ARB_program_interface_query spec says:
3683 * "For an active variable declared as a single instance of a basic
3684 * type, a single entry will be generated, using the variable name
3685 * from the shader source."
3687 gl_shader_variable
*sha_v
=
3688 create_shader_variable(shProg
, var
, prefixed_name
, type
,
3689 use_implicit_location
, location
);
3693 return add_program_resource(shProg
, programInterface
, sha_v
, stage_mask
);
3699 add_interface_variables(struct gl_shader_program
*shProg
,
3700 unsigned stage
, GLenum programInterface
)
3702 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3704 foreach_in_list(ir_instruction
, node
, ir
) {
3705 ir_variable
*var
= node
->as_variable();
3707 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3712 switch (var
->data
.mode
) {
3713 case ir_var_system_value
:
3714 case ir_var_shader_in
:
3715 if (programInterface
!= GL_PROGRAM_INPUT
)
3717 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3718 : int(VARYING_SLOT_VAR0
);
3720 case ir_var_shader_out
:
3721 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3723 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3724 : int(VARYING_SLOT_VAR0
);
3730 /* Skip packed varyings, packed varyings are handled separately
3731 * by add_packed_varyings.
3733 if (strncmp(var
->name
, "packed:", 7) == 0)
3736 /* Skip fragdata arrays, these are handled separately
3737 * by add_fragdata_arrays.
3739 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3742 const bool vs_input_or_fs_output
=
3743 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3744 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3746 if (!add_shader_variable(shProg
, 1 << stage
, programInterface
,
3747 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3748 var
->data
.location
- loc_bias
))
3755 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
, GLenum type
)
3757 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3760 if (!sh
|| !sh
->packed_varyings
)
3763 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3764 ir_variable
*var
= node
->as_variable();
3766 switch (var
->data
.mode
) {
3767 case ir_var_shader_in
:
3768 iface
= GL_PROGRAM_INPUT
;
3770 case ir_var_shader_out
:
3771 iface
= GL_PROGRAM_OUTPUT
;
3774 unreachable("unexpected type");
3777 if (type
== iface
) {
3778 const int stage_mask
=
3779 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3780 if (!add_shader_variable(shProg
, stage_mask
,
3781 iface
, var
, var
->name
, var
->type
, false,
3782 var
->data
.location
- VARYING_SLOT_VAR0
))
3791 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3793 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3795 if (!sh
|| !sh
->fragdata_arrays
)
3798 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3799 ir_variable
*var
= node
->as_variable();
3801 assert(var
->data
.mode
== ir_var_shader_out
);
3803 if (!add_shader_variable(shProg
,
3804 1 << MESA_SHADER_FRAGMENT
,
3805 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3806 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3814 get_top_level_name(const char *name
)
3816 const char *first_dot
= strchr(name
, '.');
3817 const char *first_square_bracket
= strchr(name
, '[');
3820 /* The ARB_program_interface_query spec says:
3822 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3823 * the number of active array elements of the top-level shader storage
3824 * block member containing to the active variable is written to
3825 * <params>. If the top-level block member is not declared as an
3826 * array, the value one is written to <params>. If the top-level block
3827 * member is an array with no declared size, the value zero is written
3831 /* The buffer variable is on top level.*/
3832 if (!first_square_bracket
&& !first_dot
)
3833 name_size
= strlen(name
);
3834 else if ((!first_square_bracket
||
3835 (first_dot
&& first_dot
< first_square_bracket
)))
3836 name_size
= first_dot
- name
;
3838 name_size
= first_square_bracket
- name
;
3840 return strndup(name
, name_size
);
3844 get_var_name(const char *name
)
3846 const char *first_dot
= strchr(name
, '.');
3849 return strdup(name
);
3851 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3855 is_top_level_shader_storage_block_member(const char* name
,
3856 const char* interface_name
,
3857 const char* field_name
)
3859 bool result
= false;
3861 /* If the given variable is already a top-level shader storage
3862 * block member, then return array_size = 1.
3863 * We could have two possibilities: if we have an instanced
3864 * shader storage block or not instanced.
3866 * For the first, we check create a name as it was in top level and
3867 * compare it with the real name. If they are the same, then
3868 * the variable is already at top-level.
3870 * Full instanced name is: interface name + '.' + var name +
3873 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3874 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3875 if (!full_instanced_name
) {
3876 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3880 snprintf(full_instanced_name
, name_length
, "%s.%s",
3881 interface_name
, field_name
);
3883 /* Check if its top-level shader storage block member of an
3884 * instanced interface block, or of a unnamed interface block.
3886 if (strcmp(name
, full_instanced_name
) == 0 ||
3887 strcmp(name
, field_name
) == 0)
3890 free(full_instanced_name
);
3895 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3896 char *interface_name
, char *var_name
)
3898 /* The ARB_program_interface_query spec says:
3900 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3901 * the number of active array elements of the top-level shader storage
3902 * block member containing to the active variable is written to
3903 * <params>. If the top-level block member is not declared as an
3904 * array, the value one is written to <params>. If the top-level block
3905 * member is an array with no declared size, the value zero is written
3908 if (is_top_level_shader_storage_block_member(uni
->name
,
3912 else if (field
->type
->is_unsized_array())
3914 else if (field
->type
->is_array())
3915 return field
->type
->length
;
3921 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3922 const glsl_struct_field
*field
, char *interface_name
,
3925 /* The ARB_program_interface_query spec says:
3927 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3928 * identifying the stride between array elements of the top-level
3929 * shader storage block member containing the active variable is
3930 * written to <params>. For top-level block members declared as
3931 * arrays, the value written is the difference, in basic machine units,
3932 * between the offsets of the active variable for consecutive elements
3933 * in the top-level array. For top-level block members not declared as
3934 * an array, zero is written to <params>."
3936 if (field
->type
->is_array()) {
3937 const enum glsl_matrix_layout matrix_layout
=
3938 glsl_matrix_layout(field
->matrix_layout
);
3939 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3940 const glsl_type
*array_type
= field
->type
->fields
.array
;
3942 if (is_top_level_shader_storage_block_member(uni
->name
,
3947 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3948 if (array_type
->is_record() || array_type
->is_array())
3949 return glsl_align(array_type
->std140_size(row_major
), 16);
3951 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3953 return array_type
->std430_array_stride(row_major
);
3960 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3961 struct gl_uniform_storage
*uni
)
3963 int block_index
= uni
->block_index
;
3964 int array_size
= -1;
3965 int array_stride
= -1;
3966 char *var_name
= get_top_level_name(uni
->name
);
3967 char *interface_name
=
3968 get_top_level_name(uni
->is_shader_storage
?
3969 shProg
->ShaderStorageBlocks
[block_index
].Name
:
3970 shProg
->UniformBlocks
[block_index
].Name
);
3972 if (strcmp(var_name
, interface_name
) == 0) {
3973 /* Deal with instanced array of SSBOs */
3974 char *temp_name
= get_var_name(uni
->name
);
3976 linker_error(shProg
, "Out of memory during linking.\n");
3977 goto write_top_level_array_size_and_stride
;
3980 var_name
= get_top_level_name(temp_name
);
3983 linker_error(shProg
, "Out of memory during linking.\n");
3984 goto write_top_level_array_size_and_stride
;
3988 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
3989 if (shProg
->Shaders
[i
] == NULL
)
3992 const gl_shader
*stage
= shProg
->Shaders
[i
];
3993 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
3994 ir_variable
*var
= node
->as_variable();
3995 if (!var
|| !var
->get_interface_type() ||
3996 var
->data
.mode
!= ir_var_shader_storage
)
3999 const glsl_type
*interface
= var
->get_interface_type();
4001 if (strcmp(interface_name
, interface
->name
) != 0)
4004 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4005 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4006 if (strcmp(field
->name
, var_name
) != 0)
4009 array_stride
= get_array_stride(uni
, interface
, field
,
4010 interface_name
, var_name
);
4011 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4012 goto write_top_level_array_size_and_stride
;
4016 write_top_level_array_size_and_stride
:
4017 free(interface_name
);
4019 uni
->top_level_array_stride
= array_stride
;
4020 uni
->top_level_array_size
= array_size
;
4024 * Builds up a list of program resources that point to existing
4028 build_program_resource_list(struct gl_context
*ctx
,
4029 struct gl_shader_program
*shProg
)
4031 /* Rebuild resource list. */
4032 if (shProg
->ProgramResourceList
) {
4033 ralloc_free(shProg
->ProgramResourceList
);
4034 shProg
->ProgramResourceList
= NULL
;
4035 shProg
->NumProgramResourceList
= 0;
4038 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4040 /* Determine first input and final output stage. These are used to
4041 * detect which variables should be enumerated in the resource list
4042 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4044 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4045 if (!shProg
->_LinkedShaders
[i
])
4047 if (input_stage
== MESA_SHADER_STAGES
)
4052 /* Empty shader, no resources. */
4053 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4056 /* Program interface needs to expose varyings in case of SSO. */
4057 if (shProg
->SeparateShader
) {
4058 if (!add_packed_varyings(shProg
, input_stage
, GL_PROGRAM_INPUT
))
4061 if (!add_packed_varyings(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
4065 if (!add_fragdata_arrays(shProg
))
4068 /* Add inputs and outputs to the resource list. */
4069 if (!add_interface_variables(shProg
, input_stage
, GL_PROGRAM_INPUT
))
4072 if (!add_interface_variables(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
4075 /* Add transform feedback varyings. */
4076 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
4077 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
4078 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
4079 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
4085 /* Add transform feedback buffers. */
4086 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4087 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
4088 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
4089 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_BUFFER
,
4090 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
4096 /* Add uniforms from uniform storage. */
4097 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4098 /* Do not add uniforms internally used by Mesa. */
4099 if (shProg
->UniformStorage
[i
].hidden
)
4103 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
4106 /* Add stagereferences for uniforms in a uniform block. */
4107 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
4108 int block_index
= shProg
->UniformStorage
[i
].block_index
;
4109 if (block_index
!= -1) {
4110 stageref
|= is_shader_storage
?
4111 shProg
->ShaderStorageBlocks
[block_index
].stageref
:
4112 shProg
->UniformBlocks
[block_index
].stageref
;
4115 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4116 if (!should_add_buffer_variable(shProg
, type
,
4117 shProg
->UniformStorage
[i
].name
))
4120 if (is_shader_storage
) {
4121 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
4124 if (!add_program_resource(shProg
, type
,
4125 &shProg
->UniformStorage
[i
], stageref
))
4129 /* Add program uniform blocks. */
4130 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
4131 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
4132 &shProg
->UniformBlocks
[i
], 0))
4136 /* Add program shader storage blocks. */
4137 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
4138 if (!add_program_resource(shProg
, GL_SHADER_STORAGE_BLOCK
,
4139 &shProg
->ShaderStorageBlocks
[i
], 0))
4143 /* Add atomic counter buffers. */
4144 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
4145 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
4146 &shProg
->AtomicBuffers
[i
], 0))
4150 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4152 if (!shProg
->UniformStorage
[i
].hidden
)
4155 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4156 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
4157 !shProg
->UniformStorage
[i
].type
->is_subroutine())
4160 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4161 /* add shader subroutines */
4162 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
4167 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4168 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
4174 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4175 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4176 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
4183 * This check is done to make sure we allow only constant expression
4184 * indexing and "constant-index-expression" (indexing with an expression
4185 * that includes loop induction variable).
4188 validate_sampler_array_indexing(struct gl_context
*ctx
,
4189 struct gl_shader_program
*prog
)
4191 dynamic_sampler_array_indexing_visitor v
;
4192 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4193 if (prog
->_LinkedShaders
[i
] == NULL
)
4196 bool no_dynamic_indexing
=
4197 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4199 /* Search for array derefs in shader. */
4200 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4201 if (v
.uses_dynamic_sampler_array_indexing()) {
4202 const char *msg
= "sampler arrays indexed with non-constant "
4203 "expressions is forbidden in GLSL %s %u";
4204 /* Backend has indicated that it has no dynamic indexing support. */
4205 if (no_dynamic_indexing
) {
4206 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4209 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4217 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4219 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4220 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4225 sh
->MaxSubroutineFunctionIndex
= 0;
4226 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4227 ir_function
*fn
= node
->as_function();
4231 if (fn
->is_subroutine
)
4232 sh
->NumSubroutineUniformTypes
++;
4234 if (!fn
->num_subroutine_types
)
4237 /* these should have been calculated earlier. */
4238 assert(fn
->subroutine_index
!= -1);
4239 if (sh
->NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4240 linker_error(prog
, "Too many subroutine functions declared.\n");
4243 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4244 struct gl_subroutine_function
,
4245 sh
->NumSubroutineFunctions
+ 1);
4246 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4247 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4248 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4249 ralloc_array(sh
, const struct glsl_type
*,
4250 fn
->num_subroutine_types
);
4252 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4255 * "Each subroutine with an index qualifier in the shader must be
4256 * given a unique index, otherwise a compile or link error will be
4259 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4260 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4261 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4262 linker_error(prog
, "each subroutine index qualifier in the "
4263 "shader must be unique\n");
4267 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4268 fn
->subroutine_index
;
4270 if (fn
->subroutine_index
> (int)sh
->MaxSubroutineFunctionIndex
)
4271 sh
->MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4273 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4274 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4275 sh
->NumSubroutineFunctions
++;
4281 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4283 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4285 foreach_in_list(ir_instruction
, node
, ir
) {
4286 ir_variable
*const var
= node
->as_variable();
4288 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4291 /* Don't set always active on builtins that haven't been redeclared */
4292 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4295 var
->data
.always_active_io
= true;
4300 * When separate shader programs are enabled, only input/outputs between
4301 * the stages of a multi-stage separate program can be safely removed
4302 * from the shader interface. Other inputs/outputs must remain active.
4305 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4307 unsigned first
, last
;
4308 assert(prog
->SeparateShader
);
4310 first
= MESA_SHADER_STAGES
;
4313 /* Determine first and last stage. Excluding the compute stage */
4314 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4315 if (!prog
->_LinkedShaders
[i
])
4317 if (first
== MESA_SHADER_STAGES
)
4322 if (first
== MESA_SHADER_STAGES
)
4325 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4326 gl_shader
*sh
= prog
->_LinkedShaders
[stage
];
4330 if (first
== last
) {
4331 /* For a single shader program only allow inputs to the vertex shader
4332 * and outputs from the fragment shader to be removed.
4334 if (stage
!= MESA_SHADER_VERTEX
)
4335 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4336 if (stage
!= MESA_SHADER_FRAGMENT
)
4337 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4339 /* For multi-stage separate shader programs only allow inputs and
4340 * outputs between the shader stages to be removed as well as inputs
4341 * to the vertex shader and outputs from the fragment shader.
4343 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4344 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4345 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4346 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4352 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4354 prog
->LinkStatus
= true; /* All error paths will set this to false */
4355 prog
->Validated
= false;
4356 prog
->_Used
= false;
4358 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4360 * "Linking can fail for a variety of reasons as specified in the
4361 * OpenGL Shading Language Specification, as well as any of the
4362 * following reasons:
4364 * - No shader objects are attached to program."
4366 * The Compatibility Profile specification does not list the error. In
4367 * Compatibility Profile missing shader stages are replaced by
4368 * fixed-function. This applies to the case where all stages are
4371 if (prog
->NumShaders
== 0) {
4372 if (ctx
->API
!= API_OPENGL_COMPAT
)
4373 linker_error(prog
, "no shaders attached to the program\n");
4377 unsigned num_tfeedback_decls
= 0;
4378 unsigned int num_explicit_uniform_locs
= 0;
4379 bool has_xfb_qualifiers
= false;
4380 char **varying_names
= NULL
;
4381 tfeedback_decl
*tfeedback_decls
= NULL
;
4383 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4385 prog
->ARB_fragment_coord_conventions_enable
= false;
4387 /* Separate the shaders into groups based on their type.
4389 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4390 unsigned num_shaders
[MESA_SHADER_STAGES
];
4392 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4393 shader_list
[i
] = (struct gl_shader
**)
4394 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4398 unsigned min_version
= UINT_MAX
;
4399 unsigned max_version
= 0;
4400 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4401 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4402 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4404 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4405 linker_error(prog
, "all shaders must use same shading "
4406 "language version\n");
4410 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4411 prog
->ARB_fragment_coord_conventions_enable
= true;
4414 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4415 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4416 num_shaders
[shader_type
]++;
4419 /* In desktop GLSL, different shader versions may be linked together. In
4420 * GLSL ES, all shader versions must be the same.
4422 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4423 linker_error(prog
, "all shaders must use same shading "
4424 "language version\n");
4428 prog
->Version
= max_version
;
4429 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4431 /* Some shaders have to be linked with some other shaders present.
4433 if (!prog
->SeparateShader
) {
4434 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4435 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4436 linker_error(prog
, "Geometry shader must be linked with "
4440 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4441 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4442 linker_error(prog
, "Tessellation evaluation shader must be linked "
4443 "with vertex shader\n");
4446 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4447 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4448 linker_error(prog
, "Tessellation control shader must be linked with "
4453 /* The spec is self-contradictory here. It allows linking without a tess
4454 * eval shader, but that can only be used with transform feedback and
4455 * rasterization disabled. However, transform feedback isn't allowed
4456 * with GL_PATCHES, so it can't be used.
4458 * More investigation showed that the idea of transform feedback after
4459 * a tess control shader was dropped, because some hw vendors couldn't
4460 * support tessellation without a tess eval shader, but the linker
4461 * section wasn't updated to reflect that.
4463 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4466 * Do what's reasonable and always require a tess eval shader if a tess
4467 * control shader is present.
4469 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4470 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4471 linker_error(prog
, "Tessellation control shader must be linked with "
4472 "tessellation evaluation shader\n");
4477 /* Compute shaders have additional restrictions. */
4478 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4479 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4480 linker_error(prog
, "Compute shaders may not be linked with any other "
4481 "type of shader\n");
4484 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4485 if (prog
->_LinkedShaders
[i
] != NULL
)
4486 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4488 prog
->_LinkedShaders
[i
] = NULL
;
4491 /* Link all shaders for a particular stage and validate the result.
4493 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4494 if (num_shaders
[stage
] > 0) {
4495 gl_shader
*const sh
=
4496 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4497 num_shaders
[stage
]);
4499 if (!prog
->LinkStatus
) {
4501 _mesa_delete_shader(ctx
, sh
);
4506 case MESA_SHADER_VERTEX
:
4507 validate_vertex_shader_executable(prog
, sh
, ctx
);
4509 case MESA_SHADER_TESS_CTRL
:
4510 /* nothing to be done */
4512 case MESA_SHADER_TESS_EVAL
:
4513 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4515 case MESA_SHADER_GEOMETRY
:
4516 validate_geometry_shader_executable(prog
, sh
, ctx
);
4518 case MESA_SHADER_FRAGMENT
:
4519 validate_fragment_shader_executable(prog
, sh
);
4522 if (!prog
->LinkStatus
) {
4524 _mesa_delete_shader(ctx
, sh
);
4528 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4532 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4533 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4534 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4535 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4536 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4537 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4538 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4539 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4540 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4542 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4543 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4546 /* Here begins the inter-stage linking phase. Some initial validation is
4547 * performed, then locations are assigned for uniforms, attributes, and
4550 cross_validate_uniforms(prog
);
4551 if (!prog
->LinkStatus
)
4554 unsigned first
, last
, prev
;
4556 first
= MESA_SHADER_STAGES
;
4559 /* Determine first and last stage. */
4560 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4561 if (!prog
->_LinkedShaders
[i
])
4563 if (first
== MESA_SHADER_STAGES
)
4568 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4569 link_assign_subroutine_types(prog
);
4571 if (!prog
->LinkStatus
)
4574 resize_tes_inputs(ctx
, prog
);
4576 /* Validate the inputs of each stage with the output of the preceding
4580 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4581 if (prog
->_LinkedShaders
[i
] == NULL
)
4584 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4585 prog
->_LinkedShaders
[i
]);
4586 if (!prog
->LinkStatus
)
4589 cross_validate_outputs_to_inputs(prog
,
4590 prog
->_LinkedShaders
[prev
],
4591 prog
->_LinkedShaders
[i
]);
4592 if (!prog
->LinkStatus
)
4598 /* Cross-validate uniform blocks between shader stages */
4599 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4600 MESA_SHADER_STAGES
);
4601 if (!prog
->LinkStatus
)
4604 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4605 if (prog
->_LinkedShaders
[i
] != NULL
)
4606 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4609 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4610 * it before optimization because we want most of the checks to get
4611 * dropped thanks to constant propagation.
4613 * This rule also applies to GLSL ES 3.00.
4615 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4616 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4618 lower_discard_flow(sh
->ir
);
4622 if (prog
->SeparateShader
)
4623 disable_varying_optimizations_for_sso(prog
);
4626 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4630 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4633 /* Do common optimization before assigning storage for attributes,
4634 * uniforms, and varyings. Later optimization could possibly make
4635 * some of that unused.
4637 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4638 if (prog
->_LinkedShaders
[i
] == NULL
)
4641 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4642 if (!prog
->LinkStatus
)
4645 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4646 lower_clip_distance(prog
->_LinkedShaders
[i
]);
4649 if (ctx
->Const
.LowerTessLevel
) {
4650 lower_tess_level(prog
->_LinkedShaders
[i
]);
4653 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4654 &ctx
->Const
.ShaderCompilerOptions
[i
],
4655 ctx
->Const
.NativeIntegers
))
4658 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4661 /* Validation for special cases where we allow sampler array indexing
4662 * with loop induction variable. This check emits a warning or error
4663 * depending if backend can handle dynamic indexing.
4665 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4666 (prog
->IsES
&& prog
->Version
< 300)) {
4667 if (!validate_sampler_array_indexing(ctx
, prog
))
4671 /* Check and validate stream emissions in geometry shaders */
4672 validate_geometry_shader_emissions(ctx
, prog
);
4674 /* Mark all generic shader inputs and outputs as unpaired. */
4675 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4676 if (prog
->_LinkedShaders
[i
] != NULL
) {
4677 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4682 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4683 if (prog
->_LinkedShaders
[i
] == NULL
)
4686 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4687 prog
->_LinkedShaders
[i
]);
4691 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4692 MESA_SHADER_VERTEX
)) {
4696 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4697 MESA_SHADER_FRAGMENT
)) {
4701 /* From the ARB_enhanced_layouts spec:
4703 * "If the shader used to record output variables for transform feedback
4704 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4705 * qualifiers, the values specified by TransformFeedbackVaryings are
4706 * ignored, and the set of variables captured for transform feedback is
4707 * instead derived from the specified layout qualifiers."
4709 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4710 /* Find last stage before fragment shader */
4711 if (prog
->_LinkedShaders
[i
]) {
4712 has_xfb_qualifiers
=
4713 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4714 &num_tfeedback_decls
,
4720 if (!has_xfb_qualifiers
) {
4721 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4722 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4725 if (num_tfeedback_decls
!= 0) {
4726 /* From GL_EXT_transform_feedback:
4727 * A program will fail to link if:
4729 * * the <count> specified by TransformFeedbackVaryingsEXT is
4730 * non-zero, but the program object has no vertex or geometry
4733 if (first
>= MESA_SHADER_FRAGMENT
) {
4734 linker_error(prog
, "Transform feedback varyings specified, but "
4735 "no vertex, tessellation, or geometry shader is "
4740 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4741 num_tfeedback_decls
);
4742 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4743 varying_names
, tfeedback_decls
))
4747 /* If there is no fragment shader we need to set transform feedback.
4749 * For SSO we also need to assign output locations. We assign them here
4750 * because we need to do it for both single stage programs and multi stage
4753 if (last
< MESA_SHADER_FRAGMENT
&&
4754 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4755 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4756 prog
->_LinkedShaders
[last
], NULL
,
4757 num_tfeedback_decls
, tfeedback_decls
))
4761 if (last
<= MESA_SHADER_FRAGMENT
) {
4762 /* Remove unused varyings from the first/last stage unless SSO */
4763 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4764 prog
->_LinkedShaders
[first
],
4766 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4767 prog
->_LinkedShaders
[last
],
4770 /* If the program is made up of only a single stage */
4771 if (first
== last
) {
4773 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4774 if (prog
->SeparateShader
) {
4775 /* Assign input locations for SSO, output locations are already
4778 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4779 NULL
/* producer */,
4781 0 /* num_tfeedback_decls */,
4782 NULL
/* tfeedback_decls */))
4786 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4787 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4790 /* Linking the stages in the opposite order (from fragment to vertex)
4791 * ensures that inter-shader outputs written to in an earlier stage
4792 * are eliminated if they are (transitively) not used in a later
4796 for (int i
= next
- 1; i
>= 0; i
--) {
4797 if (prog
->_LinkedShaders
[i
] == NULL
&& i
!= 0)
4800 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4801 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4803 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4804 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4808 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4809 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4812 /* This must be done after all dead varyings are eliminated. */
4814 if (!check_against_output_limit(ctx
, prog
, sh_i
)) {
4818 if (!check_against_input_limit(ctx
, prog
, sh_next
))
4826 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4827 has_xfb_qualifiers
))
4830 update_array_sizes(prog
);
4831 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
,
4832 num_explicit_uniform_locs
,
4833 ctx
->Const
.MaxUserAssignableUniformLocations
);
4834 link_assign_atomic_counter_resources(ctx
, prog
);
4835 store_fragdepth_layout(prog
);
4837 link_calculate_subroutine_compat(prog
);
4838 check_resources(ctx
, prog
);
4839 check_subroutine_resources(prog
);
4840 check_image_resources(ctx
, prog
);
4841 link_check_atomic_counter_resources(ctx
, prog
);
4843 if (!prog
->LinkStatus
)
4846 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4847 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4848 * anything about shader linking when one of the shaders (vertex or
4849 * fragment shader) is absent. So, the extension shouldn't change the
4850 * behavior specified in GLSL specification.
4852 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4853 * "Linking can fail for a variety of reasons as specified in the
4854 * OpenGL ES Shading Language Specification, as well as any of the
4855 * following reasons:
4859 * * program contains objects to form either a vertex shader or
4860 * fragment shader, and program is not separable, and does not
4861 * contain objects to form both a vertex shader and fragment
4864 * However, the only scenario in 3.1+ where we don't require them both is
4865 * when we have a compute shader. For example:
4867 * - No shaders is a link error.
4868 * - Geom or Tess without a Vertex shader is a link error which means we
4869 * always require a Vertex shader and hence a Fragment shader.
4870 * - Finally a Compute shader linked with any other stage is a link error.
4872 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4873 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4874 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4875 linker_error(prog
, "program lacks a vertex shader\n");
4876 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4877 linker_error(prog
, "program lacks a fragment shader\n");
4881 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4882 if (prog
->_LinkedShaders
[i
] == NULL
)
4885 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerBufferInterfaceBlocks
)
4886 lower_ubo_reference(prog
->_LinkedShaders
[i
]);
4888 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerShaderSharedVariables
)
4889 lower_shared_reference(prog
->_LinkedShaders
[i
],
4890 &prog
->Comp
.SharedSize
);
4892 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4893 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4897 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4898 free(shader_list
[i
]);
4899 if (prog
->_LinkedShaders
[i
] == NULL
)
4902 /* Do a final validation step to make sure that the IR wasn't
4903 * invalidated by any modifications performed after intrastage linking.
4905 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4907 /* Retain any live IR, but trash the rest. */
4908 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4910 /* The symbol table in the linked shaders may contain references to
4911 * variables that were removed (e.g., unused uniforms). Since it may
4912 * contain junk, there is no possible valid use. Delete it and set the
4915 delete prog
->_LinkedShaders
[i
]->symbols
;
4916 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4919 ralloc_free(mem_ctx
);