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"
75 #include "program/prog_instruction.h"
77 #include "link_varyings.h"
78 #include "ir_optimization.h"
79 #include "ir_rvalue_visitor.h"
80 #include "ir_uniform.h"
82 #include "main/shaderobj.h"
83 #include "main/enums.h"
89 * Visitor that determines whether or not a variable is ever written.
91 class find_assignment_visitor
: public ir_hierarchical_visitor
{
93 find_assignment_visitor(const char *name
)
94 : name(name
), found(false)
99 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
101 ir_variable
*const var
= ir
->lhs
->variable_referenced();
103 if (strcmp(name
, var
->name
) == 0) {
108 return visit_continue_with_parent
;
111 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
113 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
114 actual_node
, &ir
->actual_parameters
) {
115 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
116 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
118 if (sig_param
->data
.mode
== ir_var_function_out
||
119 sig_param
->data
.mode
== ir_var_function_inout
) {
120 ir_variable
*var
= param_rval
->variable_referenced();
121 if (var
&& strcmp(name
, var
->name
) == 0) {
128 if (ir
->return_deref
!= NULL
) {
129 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
131 if (strcmp(name
, var
->name
) == 0) {
137 return visit_continue_with_parent
;
140 bool variable_found()
146 const char *name
; /**< Find writes to a variable with this name. */
147 bool found
; /**< Was a write to the variable found? */
152 * Visitor that determines whether or not a variable is ever read.
154 class find_deref_visitor
: public ir_hierarchical_visitor
{
156 find_deref_visitor(const char *name
)
157 : name(name
), found(false)
162 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
164 if (strcmp(this->name
, ir
->var
->name
) == 0) {
169 return visit_continue
;
172 bool variable_found() const
178 const char *name
; /**< Find writes to a variable with this name. */
179 bool found
; /**< Was a write to the variable found? */
183 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
185 unsigned num_vertices
;
186 gl_shader_program
*prog
;
188 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
190 this->num_vertices
= num_vertices
;
194 virtual ~geom_array_resize_visitor()
199 virtual ir_visitor_status
visit(ir_variable
*var
)
201 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
202 return visit_continue
;
204 unsigned size
= var
->type
->length
;
206 /* Generate a link error if the shader has declared this array with an
209 if (!var
->data
.implicit_sized_array
&&
210 size
&& size
!= this->num_vertices
) {
211 linker_error(this->prog
, "size of array %s declared as %u, "
212 "but number of input vertices is %u\n",
213 var
->name
, size
, this->num_vertices
);
214 return visit_continue
;
217 /* Generate a link error if the shader attempts to access an input
218 * array using an index too large for its actual size assigned at link
221 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
222 linker_error(this->prog
, "geometry shader accesses element %i of "
223 "%s, but only %i input vertices\n",
224 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
225 return visit_continue
;
228 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
230 var
->data
.max_array_access
= this->num_vertices
- 1;
232 return visit_continue
;
235 /* Dereferences of input variables need to be updated so that their type
236 * matches the newly assigned type of the variable they are accessing. */
237 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
239 ir
->type
= ir
->var
->type
;
240 return visit_continue
;
243 /* Dereferences of 2D input arrays need to be updated so that their type
244 * matches the newly assigned type of the array they are accessing. */
245 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
247 const glsl_type
*const vt
= ir
->array
->type
;
249 ir
->type
= vt
->fields
.array
;
250 return visit_continue
;
254 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
256 unsigned num_vertices
;
257 gl_shader_program
*prog
;
259 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
261 this->num_vertices
= num_vertices
;
265 virtual ~tess_eval_array_resize_visitor()
270 virtual ir_visitor_status
visit(ir_variable
*var
)
272 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
273 return visit_continue
;
275 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
277 var
->data
.max_array_access
= this->num_vertices
- 1;
279 return visit_continue
;
282 /* Dereferences of input variables need to be updated so that their type
283 * matches the newly assigned type of the variable they are accessing. */
284 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
286 ir
->type
= ir
->var
->type
;
287 return visit_continue
;
290 /* Dereferences of 2D input arrays need to be updated so that their type
291 * matches the newly assigned type of the array they are accessing. */
292 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
294 const glsl_type
*const vt
= ir
->array
->type
;
296 ir
->type
= vt
->fields
.array
;
297 return visit_continue
;
301 class barrier_use_visitor
: public ir_hierarchical_visitor
{
303 barrier_use_visitor(gl_shader_program
*prog
)
304 : prog(prog
), in_main(false), after_return(false), control_flow(0)
308 virtual ~barrier_use_visitor()
313 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
315 if (strcmp(ir
->name
, "main") == 0)
318 return visit_continue
;
321 virtual ir_visitor_status
visit_leave(ir_function
*)
324 after_return
= false;
325 return visit_continue
;
328 virtual ir_visitor_status
visit_leave(ir_return
*)
331 return visit_continue
;
334 virtual ir_visitor_status
visit_enter(ir_if
*)
337 return visit_continue
;
340 virtual ir_visitor_status
visit_leave(ir_if
*)
343 return visit_continue
;
346 virtual ir_visitor_status
visit_enter(ir_loop
*)
349 return visit_continue
;
352 virtual ir_visitor_status
visit_leave(ir_loop
*)
355 return visit_continue
;
358 /* FINISHME: `switch` is not expressed at the IR level -- it's already
359 * been lowered to a mess of `if`s. We'll correctly disallow any use of
360 * barrier() in a conditional path within the switch, but not in a path
361 * which is always hit.
364 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
366 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
367 /* Use of barrier(); determine if it is legal: */
369 linker_error(prog
, "Builtin barrier() may only be used in main");
374 linker_error(prog
, "Builtin barrier() may not be used after return");
378 if (control_flow
!= 0) {
379 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
383 return visit_continue
;
387 gl_shader_program
*prog
;
388 bool in_main
, after_return
;
393 * Visitor that determines the highest stream id to which a (geometry) shader
394 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
396 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
398 find_emit_vertex_visitor(int max_allowed
)
399 : max_stream_allowed(max_allowed
),
400 invalid_stream_id(0),
401 invalid_stream_id_from_emit_vertex(false),
402 end_primitive_found(false),
403 uses_non_zero_stream(false)
408 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
410 int stream_id
= ir
->stream_id();
413 invalid_stream_id
= stream_id
;
414 invalid_stream_id_from_emit_vertex
= true;
418 if (stream_id
> max_stream_allowed
) {
419 invalid_stream_id
= stream_id
;
420 invalid_stream_id_from_emit_vertex
= true;
425 uses_non_zero_stream
= true;
427 return visit_continue
;
430 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
432 end_primitive_found
= true;
434 int stream_id
= ir
->stream_id();
437 invalid_stream_id
= stream_id
;
438 invalid_stream_id_from_emit_vertex
= false;
442 if (stream_id
> max_stream_allowed
) {
443 invalid_stream_id
= stream_id
;
444 invalid_stream_id_from_emit_vertex
= false;
449 uses_non_zero_stream
= true;
451 return visit_continue
;
456 return invalid_stream_id
!= 0;
459 const char *error_func()
461 return invalid_stream_id_from_emit_vertex
?
462 "EmitStreamVertex" : "EndStreamPrimitive";
467 return invalid_stream_id
;
472 return uses_non_zero_stream
;
475 bool uses_end_primitive()
477 return end_primitive_found
;
481 int max_stream_allowed
;
482 int invalid_stream_id
;
483 bool invalid_stream_id_from_emit_vertex
;
484 bool end_primitive_found
;
485 bool uses_non_zero_stream
;
488 /* Class that finds array derefs and check if indexes are dynamic. */
489 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
492 dynamic_sampler_array_indexing_visitor() :
493 dynamic_sampler_array_indexing(false)
497 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
499 if (!ir
->variable_referenced())
500 return visit_continue
;
502 if (!ir
->variable_referenced()->type
->contains_sampler())
503 return visit_continue
;
505 if (!ir
->array_index
->constant_expression_value()) {
506 dynamic_sampler_array_indexing
= true;
509 return visit_continue
;
512 bool uses_dynamic_sampler_array_indexing()
514 return dynamic_sampler_array_indexing
;
518 bool dynamic_sampler_array_indexing
;
521 } /* anonymous namespace */
524 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
528 ralloc_strcat(&prog
->InfoLog
, "error: ");
530 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
533 prog
->LinkStatus
= false;
538 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
542 ralloc_strcat(&prog
->InfoLog
, "warning: ");
544 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
551 * Given a string identifying a program resource, break it into a base name
552 * and an optional array index in square brackets.
554 * If an array index is present, \c out_base_name_end is set to point to the
555 * "[" that precedes the array index, and the array index itself is returned
558 * If no array index is present (or if the array index is negative or
559 * mal-formed), \c out_base_name_end, is set to point to the null terminator
560 * at the end of the input string, and -1 is returned.
562 * Only the final array index is parsed; if the string contains other array
563 * indices (or structure field accesses), they are left in the base name.
565 * No attempt is made to check that the base name is properly formed;
566 * typically the caller will look up the base name in a hash table, so
567 * ill-formed base names simply turn into hash table lookup failures.
570 parse_program_resource_name(const GLchar
*name
,
571 const GLchar
**out_base_name_end
)
573 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
575 * "When an integer array element or block instance number is part of
576 * the name string, it will be specified in decimal form without a "+"
577 * or "-" sign or any extra leading zeroes. Additionally, the name
578 * string will not include white space anywhere in the string."
581 const size_t len
= strlen(name
);
582 *out_base_name_end
= name
+ len
;
584 if (len
== 0 || name
[len
-1] != ']')
587 /* Walk backwards over the string looking for a non-digit character. This
588 * had better be the opening bracket for an array index.
590 * Initially, i specifies the location of the ']'. Since the string may
591 * contain only the ']' charcater, walk backwards very carefully.
594 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
597 if ((i
== 0) || name
[i
-1] != '[')
600 long array_index
= strtol(&name
[i
], NULL
, 10);
604 /* Check for leading zero */
605 if (name
[i
] == '0' && name
[i
+1] != ']')
608 *out_base_name_end
= name
+ (i
- 1);
614 link_invalidate_variable_locations(exec_list
*ir
)
616 foreach_in_list(ir_instruction
, node
, ir
) {
617 ir_variable
*const var
= node
->as_variable();
622 /* Only assign locations for variables that lack an explicit location.
623 * Explicit locations are set for all built-in variables, generic vertex
624 * shader inputs (via layout(location=...)), and generic fragment shader
625 * outputs (also via layout(location=...)).
627 if (!var
->data
.explicit_location
) {
628 var
->data
.location
= -1;
629 var
->data
.location_frac
= 0;
632 /* ir_variable::is_unmatched_generic_inout is used by the linker while
633 * connecting outputs from one stage to inputs of the next stage.
635 if (var
->data
.explicit_location
&&
636 var
->data
.location
< VARYING_SLOT_VAR0
) {
637 var
->data
.is_unmatched_generic_inout
= 0;
639 var
->data
.is_unmatched_generic_inout
= 1;
646 * Set clip_distance_array_size based and cull_distance_array_size on the given
649 * Also check for errors based on incorrect usage of gl_ClipVertex and
650 * gl_ClipDistance and gl_CullDistance.
651 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
652 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
654 * Return false if an error was reported.
657 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
658 struct gl_shader
*shader
,
659 struct gl_context
*ctx
,
660 GLuint
*clip_distance_array_size
,
661 GLuint
*cull_distance_array_size
)
663 *clip_distance_array_size
= 0;
664 *cull_distance_array_size
= 0;
666 if (prog
->Version
>= (prog
->IsES
? 300 : 130)) {
667 /* From section 7.1 (Vertex Shader Special Variables) of the
670 * "It is an error for a shader to statically write both
671 * gl_ClipVertex and gl_ClipDistance."
673 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
674 * gl_ClipVertex nor gl_ClipDistance. However with
675 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
677 find_assignment_visitor
clip_distance("gl_ClipDistance");
678 find_assignment_visitor
cull_distance("gl_CullDistance");
680 clip_distance
.run(shader
->ir
);
681 cull_distance
.run(shader
->ir
);
683 /* From the ARB_cull_distance spec:
685 * It is a compile-time or link-time error for the set of shaders forming
686 * a program to statically read or write both gl_ClipVertex and either
687 * gl_ClipDistance or gl_CullDistance.
689 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
693 find_assignment_visitor
clip_vertex("gl_ClipVertex");
695 clip_vertex
.run(shader
->ir
);
697 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
698 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
699 "and `gl_ClipDistance'\n",
700 _mesa_shader_stage_to_string(shader
->Stage
));
703 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
704 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
705 "and `gl_CullDistance'\n",
706 _mesa_shader_stage_to_string(shader
->Stage
));
711 if (clip_distance
.variable_found()) {
712 ir_variable
*clip_distance_var
=
713 shader
->symbols
->get_variable("gl_ClipDistance");
714 assert(clip_distance_var
);
715 *clip_distance_array_size
= clip_distance_var
->type
->length
;
717 if (cull_distance
.variable_found()) {
718 ir_variable
*cull_distance_var
=
719 shader
->symbols
->get_variable("gl_CullDistance");
720 assert(cull_distance_var
);
721 *cull_distance_array_size
= cull_distance_var
->type
->length
;
723 /* From the ARB_cull_distance spec:
725 * It is a compile-time or link-time error for the set of shaders forming
726 * a program to have the sum of the sizes of the gl_ClipDistance and
727 * gl_CullDistance arrays to be larger than
728 * gl_MaxCombinedClipAndCullDistances.
730 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
731 ctx
->Const
.MaxClipPlanes
) {
732 linker_error(prog
, "%s shader: the combined size of "
733 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
735 "gl_MaxCombinedClipAndCullDistances (%u)",
736 _mesa_shader_stage_to_string(shader
->Stage
),
737 ctx
->Const
.MaxClipPlanes
);
744 * Verify that a vertex shader executable meets all semantic requirements.
746 * Also sets prog->Vert.ClipDistanceArraySize and
747 * prog->Vert.CullDistanceArraySize as a side effect.
749 * \param shader Vertex shader executable to be verified
752 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
753 struct gl_shader
*shader
,
754 struct gl_context
*ctx
)
759 /* From the GLSL 1.10 spec, page 48:
761 * "The variable gl_Position is available only in the vertex
762 * language and is intended for writing the homogeneous vertex
763 * position. All executions of a well-formed vertex shader
764 * executable must write a value into this variable. [...] The
765 * variable gl_Position is available only in the vertex
766 * language and is intended for writing the homogeneous vertex
767 * position. All executions of a well-formed vertex shader
768 * executable must write a value into this variable."
770 * while in GLSL 1.40 this text is changed to:
772 * "The variable gl_Position is available only in the vertex
773 * language and is intended for writing the homogeneous vertex
774 * position. It can be written at any time during shader
775 * execution. It may also be read back by a vertex shader
776 * after being written. This value will be used by primitive
777 * assembly, clipping, culling, and other fixed functionality
778 * operations, if present, that operate on primitives after
779 * vertex processing has occurred. Its value is undefined if
780 * the vertex shader executable does not write gl_Position."
782 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
783 * gl_Position is not an error.
785 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
786 find_assignment_visitor
find("gl_Position");
787 find
.run(shader
->ir
);
788 if (!find
.variable_found()) {
791 "vertex shader does not write to `gl_Position'."
792 "It's value is undefined. \n");
795 "vertex shader does not write to `gl_Position'. \n");
801 analyze_clip_cull_usage(prog
, shader
, ctx
,
802 &prog
->Vert
.ClipDistanceArraySize
,
803 &prog
->Vert
.CullDistanceArraySize
);
807 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
808 struct gl_shader
*shader
,
809 struct gl_context
*ctx
)
814 analyze_clip_cull_usage(prog
, shader
, ctx
,
815 &prog
->TessEval
.ClipDistanceArraySize
,
816 &prog
->TessEval
.CullDistanceArraySize
);
821 * Verify that a fragment shader executable meets all semantic requirements
823 * \param shader Fragment shader executable to be verified
826 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
827 struct gl_shader
*shader
)
832 find_assignment_visitor
frag_color("gl_FragColor");
833 find_assignment_visitor
frag_data("gl_FragData");
835 frag_color
.run(shader
->ir
);
836 frag_data
.run(shader
->ir
);
838 if (frag_color
.variable_found() && frag_data
.variable_found()) {
839 linker_error(prog
, "fragment shader writes to both "
840 "`gl_FragColor' and `gl_FragData'\n");
845 * Verify that a geometry shader executable meets all semantic requirements
847 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize and
848 * prog->Geom.CullDistanceArraySize as a side effect.
850 * \param shader Geometry shader executable to be verified
853 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
854 struct gl_shader
*shader
,
855 struct gl_context
*ctx
)
860 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
861 prog
->Geom
.VerticesIn
= num_vertices
;
863 analyze_clip_cull_usage(prog
, shader
, ctx
,
864 &prog
->Geom
.ClipDistanceArraySize
,
865 &prog
->Geom
.CullDistanceArraySize
);
869 * Check if geometry shaders emit to non-zero streams and do corresponding
873 validate_geometry_shader_emissions(struct gl_context
*ctx
,
874 struct gl_shader_program
*prog
)
876 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
877 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
878 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
879 if (emit_vertex
.error()) {
880 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
881 "stream parameter are in the range [0, %d].\n",
882 emit_vertex
.error_func(),
883 emit_vertex
.error_stream(),
884 ctx
->Const
.MaxVertexStreams
- 1);
886 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
887 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
889 /* From the ARB_gpu_shader5 spec:
891 * "Multiple vertex streams are supported only if the output primitive
892 * type is declared to be "points". A program will fail to link if it
893 * contains a geometry shader calling EmitStreamVertex() or
894 * EndStreamPrimitive() if its output primitive type is not "points".
896 * However, in the same spec:
898 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
899 * with <stream> set to zero."
903 * "The function EndPrimitive() is equivalent to calling
904 * EndStreamPrimitive() with <stream> set to zero."
906 * Since we can call EmitVertex() and EndPrimitive() when we output
907 * primitives other than points, calling EmitStreamVertex(0) or
908 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
909 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
910 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
913 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
914 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
915 "with n>0 requires point output\n");
921 validate_intrastage_arrays(struct gl_shader_program
*prog
,
922 ir_variable
*const var
,
923 ir_variable
*const existing
)
925 /* Consider the types to be "the same" if both types are arrays
926 * of the same type and one of the arrays is implicitly sized.
927 * In addition, set the type of the linked variable to the
928 * explicitly sized array.
930 if (var
->type
->is_array() && existing
->type
->is_array()) {
931 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
932 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
933 if (var
->type
->length
!= 0) {
934 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
935 linker_error(prog
, "%s `%s' declared as type "
936 "`%s' but outermost dimension has an index"
939 var
->name
, var
->type
->name
,
940 existing
->data
.max_array_access
);
942 existing
->type
= var
->type
;
944 } else if (existing
->type
->length
!= 0) {
945 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
946 !existing
->data
.from_ssbo_unsized_array
) {
947 linker_error(prog
, "%s `%s' declared as type "
948 "`%s' but outermost dimension has an index"
951 var
->name
, existing
->type
->name
,
952 var
->data
.max_array_access
);
957 /* The arrays of structs could have different glsl_type pointers but
958 * they are actually the same type. Use record_compare() to check that.
960 if (existing
->type
->fields
.array
->is_record() &&
961 var
->type
->fields
.array
->is_record() &&
962 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
971 * Perform validation of global variables used across multiple shaders
974 cross_validate_globals(struct gl_shader_program
*prog
,
975 struct gl_shader
**shader_list
,
976 unsigned num_shaders
,
979 /* Examine all of the uniforms in all of the shaders and cross validate
982 glsl_symbol_table variables
;
983 for (unsigned i
= 0; i
< num_shaders
; i
++) {
984 if (shader_list
[i
] == NULL
)
987 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
988 ir_variable
*const var
= node
->as_variable();
993 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
996 /* don't cross validate subroutine uniforms */
997 if (var
->type
->contains_subroutine())
1000 /* Don't cross validate temporaries that are at global scope. These
1001 * will eventually get pulled into the shaders 'main'.
1003 if (var
->data
.mode
== ir_var_temporary
)
1006 /* If a global with this name has already been seen, verify that the
1007 * new instance has the same type. In addition, if the globals have
1008 * initializers, the values of the initializers must be the same.
1010 ir_variable
*const existing
= variables
.get_variable(var
->name
);
1011 if (existing
!= NULL
) {
1012 /* Check if types match. Interface blocks have some special
1013 * rules so we handle those elsewhere.
1015 if (var
->type
!= existing
->type
&&
1016 !var
->is_interface_instance()) {
1017 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
1018 if (var
->type
->is_record() && existing
->type
->is_record()
1019 && existing
->type
->record_compare(var
->type
)) {
1020 existing
->type
= var
->type
;
1022 /* If it is an unsized array in a Shader Storage Block,
1023 * two different shaders can access to different elements.
1024 * Because of that, they might be converted to different
1025 * sized arrays, then check that they are compatible but
1026 * ignore the array size.
1028 if (!(var
->data
.mode
== ir_var_shader_storage
&&
1029 var
->data
.from_ssbo_unsized_array
&&
1030 existing
->data
.mode
== ir_var_shader_storage
&&
1031 existing
->data
.from_ssbo_unsized_array
&&
1032 var
->type
->gl_type
== existing
->type
->gl_type
)) {
1033 linker_error(prog
, "%s `%s' declared as type "
1034 "`%s' and type `%s'\n",
1036 var
->name
, var
->type
->name
,
1037 existing
->type
->name
);
1044 if (var
->data
.explicit_location
) {
1045 if (existing
->data
.explicit_location
1046 && (var
->data
.location
!= existing
->data
.location
)) {
1047 linker_error(prog
, "explicit locations for %s "
1048 "`%s' have differing values\n",
1049 mode_string(var
), var
->name
);
1053 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
1054 linker_error(prog
, "explicit components for %s "
1055 "`%s' have differing values\n",
1056 mode_string(var
), var
->name
);
1060 existing
->data
.location
= var
->data
.location
;
1061 existing
->data
.explicit_location
= true;
1063 /* Check if uniform with implicit location was marked explicit
1064 * by earlier shader stage. If so, mark it explicit in this stage
1065 * too to make sure later processing does not treat it as
1068 if (existing
->data
.explicit_location
) {
1069 var
->data
.location
= existing
->data
.location
;
1070 var
->data
.explicit_location
= true;
1074 /* From the GLSL 4.20 specification:
1075 * "A link error will result if two compilation units in a program
1076 * specify different integer-constant bindings for the same
1077 * opaque-uniform name. However, it is not an error to specify a
1078 * binding on some but not all declarations for the same name"
1080 if (var
->data
.explicit_binding
) {
1081 if (existing
->data
.explicit_binding
&&
1082 var
->data
.binding
!= existing
->data
.binding
) {
1083 linker_error(prog
, "explicit bindings for %s "
1084 "`%s' have differing values\n",
1085 mode_string(var
), var
->name
);
1089 existing
->data
.binding
= var
->data
.binding
;
1090 existing
->data
.explicit_binding
= true;
1093 if (var
->type
->contains_atomic() &&
1094 var
->data
.offset
!= existing
->data
.offset
) {
1095 linker_error(prog
, "offset specifications for %s "
1096 "`%s' have differing values\n",
1097 mode_string(var
), var
->name
);
1101 /* Validate layout qualifiers for gl_FragDepth.
1103 * From the AMD/ARB_conservative_depth specs:
1105 * "If gl_FragDepth is redeclared in any fragment shader in a
1106 * program, it must be redeclared in all fragment shaders in
1107 * that program that have static assignments to
1108 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1109 * fragment shaders in a single program must have the same set
1112 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1113 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1114 bool layout_differs
=
1115 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1117 if (layout_declared
&& layout_differs
) {
1119 "All redeclarations of gl_FragDepth in all "
1120 "fragment shaders in a single program must have "
1121 "the same set of qualifiers.\n");
1124 if (var
->data
.used
&& layout_differs
) {
1126 "If gl_FragDepth is redeclared with a layout "
1127 "qualifier in any fragment shader, it must be "
1128 "redeclared with the same layout qualifier in "
1129 "all fragment shaders that have assignments to "
1134 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1136 * "If a shared global has multiple initializers, the
1137 * initializers must all be constant expressions, and they
1138 * must all have the same value. Otherwise, a link error will
1139 * result. (A shared global having only one initializer does
1140 * not require that initializer to be a constant expression.)"
1142 * Previous to 4.20 the GLSL spec simply said that initializers
1143 * must have the same value. In this case of non-constant
1144 * initializers, this was impossible to determine. As a result,
1145 * no vendor actually implemented that behavior. The 4.20
1146 * behavior matches the implemented behavior of at least one other
1147 * vendor, so we'll implement that for all GLSL versions.
1149 if (var
->constant_initializer
!= NULL
) {
1150 if (existing
->constant_initializer
!= NULL
) {
1151 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1152 linker_error(prog
, "initializers for %s "
1153 "`%s' have differing values\n",
1154 mode_string(var
), var
->name
);
1158 /* If the first-seen instance of a particular uniform did
1159 * not have an initializer but a later instance does,
1160 * replace the former with the later.
1162 variables
.replace_variable(existing
->name
, var
);
1166 if (var
->data
.has_initializer
) {
1167 if (existing
->data
.has_initializer
1168 && (var
->constant_initializer
== NULL
1169 || existing
->constant_initializer
== NULL
)) {
1171 "shared global variable `%s' has multiple "
1172 "non-constant initializers.\n",
1178 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1179 linker_error(prog
, "declarations for %s `%s' have "
1180 "mismatching invariant qualifiers\n",
1181 mode_string(var
), var
->name
);
1184 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1185 linker_error(prog
, "declarations for %s `%s' have "
1186 "mismatching centroid qualifiers\n",
1187 mode_string(var
), var
->name
);
1190 if (existing
->data
.sample
!= var
->data
.sample
) {
1191 linker_error(prog
, "declarations for %s `%s` have "
1192 "mismatching sample qualifiers\n",
1193 mode_string(var
), var
->name
);
1196 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1197 linker_error(prog
, "declarations for %s `%s` have "
1198 "mismatching image format qualifiers\n",
1199 mode_string(var
), var
->name
);
1203 variables
.add_variable(var
);
1210 * Perform validation of uniforms used across multiple shader stages
1213 cross_validate_uniforms(struct gl_shader_program
*prog
)
1215 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1216 MESA_SHADER_STAGES
, true);
1220 * Accumulates the array of buffer blocks and checks that all definitions of
1221 * blocks agree on their contents.
1224 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1227 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1228 struct gl_uniform_block
*blks
= NULL
;
1229 unsigned *num_blks
= validate_ssbo
? &prog
->NumShaderStorageBlocks
:
1230 &prog
->NumUniformBlocks
;
1232 unsigned max_num_buffer_blocks
= 0;
1233 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1234 if (prog
->_LinkedShaders
[i
]) {
1235 if (validate_ssbo
) {
1236 max_num_buffer_blocks
+=
1237 prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1239 max_num_buffer_blocks
+=
1240 prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1245 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1246 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1248 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1249 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1250 InterfaceBlockStageIndex
[i
][j
] = -1;
1255 unsigned sh_num_blocks
;
1256 struct gl_uniform_block
**sh_blks
;
1257 if (validate_ssbo
) {
1258 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1259 sh_blks
= sh
->ShaderStorageBlocks
;
1261 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1262 sh_blks
= sh
->UniformBlocks
;
1265 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1266 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1270 linker_error(prog
, "buffer block `%s' has mismatching "
1271 "definitions\n", sh_blks
[j
]->Name
);
1273 for (unsigned k
= 0; k
<= i
; k
++) {
1274 delete[] InterfaceBlockStageIndex
[k
];
1279 InterfaceBlockStageIndex
[i
][index
] = j
;
1283 /* Update per stage block pointers to point to the program list.
1284 * FIXME: We should be able to free the per stage blocks here.
1286 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1287 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1288 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1290 if (stage_index
!= -1) {
1291 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1293 blks
[j
].stageref
|= (1 << i
);
1295 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1296 sh
->ShaderStorageBlocks
: sh
->UniformBlocks
;
1298 sh_blks
[stage_index
] = &blks
[j
];
1303 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1304 delete[] InterfaceBlockStageIndex
[i
];
1308 prog
->ShaderStorageBlocks
= blks
;
1310 prog
->UniformBlocks
= blks
;
1317 * Populates a shaders symbol table with all global declarations
1320 populate_symbol_table(gl_shader
*sh
)
1322 sh
->symbols
= new(sh
) glsl_symbol_table
;
1324 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1328 if ((func
= inst
->as_function()) != NULL
) {
1329 sh
->symbols
->add_function(func
);
1330 } else if ((var
= inst
->as_variable()) != NULL
) {
1331 if (var
->data
.mode
!= ir_var_temporary
)
1332 sh
->symbols
->add_variable(var
);
1339 * Remap variables referenced in an instruction tree
1341 * This is used when instruction trees are cloned from one shader and placed in
1342 * another. These trees will contain references to \c ir_variable nodes that
1343 * do not exist in the target shader. This function finds these \c ir_variable
1344 * references and replaces the references with matching variables in the target
1347 * If there is no matching variable in the target shader, a clone of the
1348 * \c ir_variable is made and added to the target shader. The new variable is
1349 * added to \b both the instruction stream and the symbol table.
1351 * \param inst IR tree that is to be processed.
1352 * \param symbols Symbol table containing global scope symbols in the
1354 * \param instructions Instruction stream where new variable declarations
1358 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1361 class remap_visitor
: public ir_hierarchical_visitor
{
1363 remap_visitor(struct gl_shader
*target
,
1366 this->target
= target
;
1367 this->symbols
= target
->symbols
;
1368 this->instructions
= target
->ir
;
1369 this->temps
= temps
;
1372 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1374 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1375 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1377 assert(var
!= NULL
);
1379 return visit_continue
;
1382 ir_variable
*const existing
=
1383 this->symbols
->get_variable(ir
->var
->name
);
1384 if (existing
!= NULL
)
1387 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1389 this->symbols
->add_variable(copy
);
1390 this->instructions
->push_head(copy
);
1394 return visit_continue
;
1398 struct gl_shader
*target
;
1399 glsl_symbol_table
*symbols
;
1400 exec_list
*instructions
;
1404 remap_visitor
v(target
, temps
);
1411 * Move non-declarations from one instruction stream to another
1413 * The intended usage pattern of this function is to pass the pointer to the
1414 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1415 * pointer) for \c last and \c false for \c make_copies on the first
1416 * call. Successive calls pass the return value of the previous call for
1417 * \c last and \c true for \c make_copies.
1419 * \param instructions Source instruction stream
1420 * \param last Instruction after which new instructions should be
1421 * inserted in the target instruction stream
1422 * \param make_copies Flag selecting whether instructions in \c instructions
1423 * should be copied (via \c ir_instruction::clone) into the
1424 * target list or moved.
1427 * The new "last" instruction in the target instruction stream. This pointer
1428 * is suitable for use as the \c last parameter of a later call to this
1432 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1433 bool make_copies
, gl_shader
*target
)
1435 hash_table
*temps
= NULL
;
1438 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1439 hash_table_pointer_compare
);
1441 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1442 if (inst
->as_function())
1445 ir_variable
*var
= inst
->as_variable();
1446 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1449 assert(inst
->as_assignment()
1451 || inst
->as_if() /* for initializers with the ?: operator */
1452 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1455 inst
= inst
->clone(target
, NULL
);
1458 hash_table_insert(temps
, inst
, var
);
1460 remap_variables(inst
, target
, temps
);
1465 last
->insert_after(inst
);
1470 hash_table_dtor(temps
);
1477 * This class is only used in link_intrastage_shaders() below but declaring
1478 * it inside that function leads to compiler warnings with some versions of
1481 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1483 array_sizing_visitor()
1484 : mem_ctx(ralloc_context(NULL
)),
1485 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1486 hash_table_pointer_compare
))
1490 ~array_sizing_visitor()
1492 hash_table_dtor(this->unnamed_interfaces
);
1493 ralloc_free(this->mem_ctx
);
1496 virtual ir_visitor_status
visit(ir_variable
*var
)
1498 const glsl_type
*type_without_array
;
1499 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1500 fixup_type(&var
->type
, var
->data
.max_array_access
,
1501 var
->data
.from_ssbo_unsized_array
,
1502 &implicit_sized_array
);
1503 var
->data
.implicit_sized_array
= implicit_sized_array
;
1504 type_without_array
= var
->type
->without_array();
1505 if (var
->type
->is_interface()) {
1506 if (interface_contains_unsized_arrays(var
->type
)) {
1507 const glsl_type
*new_type
=
1508 resize_interface_members(var
->type
,
1509 var
->get_max_ifc_array_access(),
1510 var
->is_in_shader_storage_block());
1511 var
->type
= new_type
;
1512 var
->change_interface_type(new_type
);
1514 } else if (type_without_array
->is_interface()) {
1515 if (interface_contains_unsized_arrays(type_without_array
)) {
1516 const glsl_type
*new_type
=
1517 resize_interface_members(type_without_array
,
1518 var
->get_max_ifc_array_access(),
1519 var
->is_in_shader_storage_block());
1520 var
->change_interface_type(new_type
);
1521 var
->type
= update_interface_members_array(var
->type
, new_type
);
1523 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1524 /* Store a pointer to the variable in the unnamed_interfaces
1527 ir_variable
**interface_vars
= (ir_variable
**)
1528 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1529 if (interface_vars
== NULL
) {
1530 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1532 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1535 unsigned index
= ifc_type
->field_index(var
->name
);
1536 assert(index
< ifc_type
->length
);
1537 assert(interface_vars
[index
] == NULL
);
1538 interface_vars
[index
] = var
;
1540 return visit_continue
;
1544 * For each unnamed interface block that was discovered while running the
1545 * visitor, adjust the interface type to reflect the newly assigned array
1546 * sizes, and fix up the ir_variable nodes to point to the new interface
1549 void fixup_unnamed_interface_types()
1551 hash_table_call_foreach(this->unnamed_interfaces
,
1552 fixup_unnamed_interface_type
, NULL
);
1557 * If the type pointed to by \c type represents an unsized array, replace
1558 * it with a sized array whose size is determined by max_array_access.
1560 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1561 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1563 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1564 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1565 max_array_access
+ 1);
1566 *implicit_sized
= true;
1567 assert(*type
!= NULL
);
1571 static const glsl_type
*
1572 update_interface_members_array(const glsl_type
*type
,
1573 const glsl_type
*new_interface_type
)
1575 const glsl_type
*element_type
= type
->fields
.array
;
1576 if (element_type
->is_array()) {
1577 const glsl_type
*new_array_type
=
1578 update_interface_members_array(element_type
, new_interface_type
);
1579 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1581 return glsl_type::get_array_instance(new_interface_type
,
1587 * Determine whether the given interface type contains unsized arrays (if
1588 * it doesn't, array_sizing_visitor doesn't need to process it).
1590 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1592 for (unsigned i
= 0; i
< type
->length
; i
++) {
1593 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1594 if (elem_type
->is_unsized_array())
1601 * Create a new interface type based on the given type, with unsized arrays
1602 * replaced by sized arrays whose size is determined by
1603 * max_ifc_array_access.
1605 static const glsl_type
*
1606 resize_interface_members(const glsl_type
*type
,
1607 const int *max_ifc_array_access
,
1610 unsigned num_fields
= type
->length
;
1611 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1612 memcpy(fields
, type
->fields
.structure
,
1613 num_fields
* sizeof(*fields
));
1614 for (unsigned i
= 0; i
< num_fields
; i
++) {
1615 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1616 /* If SSBO last member is unsized array, we don't replace it by a sized
1619 if (is_ssbo
&& i
== (num_fields
- 1))
1620 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1621 true, &implicit_sized_array
);
1623 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1624 false, &implicit_sized_array
);
1625 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1627 glsl_interface_packing packing
=
1628 (glsl_interface_packing
) type
->interface_packing
;
1629 const glsl_type
*new_ifc_type
=
1630 glsl_type::get_interface_instance(fields
, num_fields
,
1631 packing
, type
->name
);
1633 return new_ifc_type
;
1636 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1639 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1640 ir_variable
**interface_vars
= (ir_variable
**) data
;
1641 unsigned num_fields
= ifc_type
->length
;
1642 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1643 memcpy(fields
, ifc_type
->fields
.structure
,
1644 num_fields
* sizeof(*fields
));
1645 bool interface_type_changed
= false;
1646 for (unsigned i
= 0; i
< num_fields
; i
++) {
1647 if (interface_vars
[i
] != NULL
&&
1648 fields
[i
].type
!= interface_vars
[i
]->type
) {
1649 fields
[i
].type
= interface_vars
[i
]->type
;
1650 interface_type_changed
= true;
1653 if (!interface_type_changed
) {
1657 glsl_interface_packing packing
=
1658 (glsl_interface_packing
) ifc_type
->interface_packing
;
1659 const glsl_type
*new_ifc_type
=
1660 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1663 for (unsigned i
= 0; i
< num_fields
; i
++) {
1664 if (interface_vars
[i
] != NULL
)
1665 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1670 * Memory context used to allocate the data in \c unnamed_interfaces.
1675 * Hash table from const glsl_type * to an array of ir_variable *'s
1676 * pointing to the ir_variables constituting each unnamed interface block.
1678 hash_table
*unnamed_interfaces
;
1682 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1686 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1687 struct gl_shader_program
*prog
,
1688 struct gl_shader
*linked_shader
,
1689 struct gl_shader
**shader_list
,
1690 unsigned num_shaders
)
1692 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1693 linked_shader
->TransformFeedback
.BufferStride
[i
] = 0;
1696 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1697 struct gl_shader
*shader
= shader_list
[i
];
1699 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1700 if (shader
->TransformFeedback
.BufferStride
[j
]) {
1701 if (linked_shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1702 shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1703 linked_shader
->TransformFeedback
.BufferStride
[j
] !=
1704 shader
->TransformFeedback
.BufferStride
[j
]) {
1706 "intrastage shaders defined with conflicting "
1707 "xfb_stride for buffer %d (%d and %d)\n", j
,
1708 linked_shader
->TransformFeedback
.BufferStride
[j
],
1709 shader
->TransformFeedback
.BufferStride
[j
]);
1713 if (shader
->TransformFeedback
.BufferStride
[j
])
1714 linked_shader
->TransformFeedback
.BufferStride
[j
] =
1715 shader
->TransformFeedback
.BufferStride
[j
];
1720 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1721 if (linked_shader
->TransformFeedback
.BufferStride
[j
]) {
1722 prog
->TransformFeedback
.BufferStride
[j
] =
1723 linked_shader
->TransformFeedback
.BufferStride
[j
];
1725 /* We will validate doubles at a later stage */
1726 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1727 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1728 "multiple of 4 or if its applied to a type that is "
1729 "or contains a double a multiple of 8.",
1730 prog
->TransformFeedback
.BufferStride
[j
]);
1734 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1735 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1737 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1738 "limit has been exceeded.");
1746 * Performs the cross-validation of tessellation control shader vertices and
1747 * layout qualifiers for the attached tessellation control shaders,
1748 * and propagates them to the linked TCS and linked shader program.
1751 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1752 struct gl_shader
*linked_shader
,
1753 struct gl_shader
**shader_list
,
1754 unsigned num_shaders
)
1756 linked_shader
->TessCtrl
.VerticesOut
= 0;
1758 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1761 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1763 * "All tessellation control shader layout declarations in a program
1764 * must specify the same output patch vertex count. There must be at
1765 * least one layout qualifier specifying an output patch vertex count
1766 * in any program containing tessellation control shaders; however,
1767 * such a declaration is not required in all tessellation control
1771 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1772 struct gl_shader
*shader
= shader_list
[i
];
1774 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1775 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1776 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1777 linker_error(prog
, "tessellation control shader defined with "
1778 "conflicting output vertex count (%d and %d)\n",
1779 linked_shader
->TessCtrl
.VerticesOut
,
1780 shader
->TessCtrl
.VerticesOut
);
1783 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1787 /* Just do the intrastage -> interstage propagation right now,
1788 * since we already know we're in the right type of shader program
1791 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1792 linker_error(prog
, "tessellation control shader didn't declare "
1793 "vertices out layout qualifier\n");
1796 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1801 * Performs the cross-validation of tessellation evaluation shader
1802 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1803 * for the attached tessellation evaluation shaders, and propagates them
1804 * to the linked TES and linked shader program.
1807 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1808 struct gl_shader
*linked_shader
,
1809 struct gl_shader
**shader_list
,
1810 unsigned num_shaders
)
1812 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1813 linked_shader
->TessEval
.Spacing
= 0;
1814 linked_shader
->TessEval
.VertexOrder
= 0;
1815 linked_shader
->TessEval
.PointMode
= -1;
1817 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1820 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1822 * "At least one tessellation evaluation shader (compilation unit) in
1823 * a program must declare a primitive mode in its input layout.
1824 * Declaration vertex spacing, ordering, and point mode identifiers is
1825 * optional. It is not required that all tessellation evaluation
1826 * shaders in a program declare a primitive mode. If spacing or
1827 * vertex ordering declarations are omitted, the tessellation
1828 * primitive generator will use equal spacing or counter-clockwise
1829 * vertex ordering, respectively. If a point mode declaration is
1830 * omitted, the tessellation primitive generator will produce lines or
1831 * triangles according to the primitive mode."
1834 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1835 struct gl_shader
*shader
= shader_list
[i
];
1837 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1838 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1839 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1840 linker_error(prog
, "tessellation evaluation shader defined with "
1841 "conflicting input primitive modes.\n");
1844 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1847 if (shader
->TessEval
.Spacing
!= 0) {
1848 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1849 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1850 linker_error(prog
, "tessellation evaluation shader defined with "
1851 "conflicting vertex spacing.\n");
1854 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1857 if (shader
->TessEval
.VertexOrder
!= 0) {
1858 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1859 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1860 linker_error(prog
, "tessellation evaluation shader defined with "
1861 "conflicting ordering.\n");
1864 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1867 if (shader
->TessEval
.PointMode
!= -1) {
1868 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1869 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1870 linker_error(prog
, "tessellation evaluation shader defined with "
1871 "conflicting point modes.\n");
1874 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1879 /* Just do the intrastage -> interstage propagation right now,
1880 * since we already know we're in the right type of shader program
1883 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1885 "tessellation evaluation shader didn't declare input "
1886 "primitive modes.\n");
1890 if (linked_shader
->TessEval
.Spacing
== 0)
1891 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1893 if (linked_shader
->TessEval
.VertexOrder
== 0)
1894 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1896 if (linked_shader
->TessEval
.PointMode
== -1)
1897 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1902 * Performs the cross-validation of layout qualifiers specified in
1903 * redeclaration of gl_FragCoord for the attached fragment shaders,
1904 * and propagates them to the linked FS and linked shader program.
1907 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1908 struct gl_shader
*linked_shader
,
1909 struct gl_shader
**shader_list
,
1910 unsigned num_shaders
)
1912 linked_shader
->redeclares_gl_fragcoord
= false;
1913 linked_shader
->uses_gl_fragcoord
= false;
1914 linked_shader
->origin_upper_left
= false;
1915 linked_shader
->pixel_center_integer
= false;
1917 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1918 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1921 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1922 struct gl_shader
*shader
= shader_list
[i
];
1923 /* From the GLSL 1.50 spec, page 39:
1925 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1926 * it must be redeclared in all the fragment shaders in that program
1927 * that have a static use gl_FragCoord."
1929 if ((linked_shader
->redeclares_gl_fragcoord
1930 && !shader
->redeclares_gl_fragcoord
1931 && shader
->uses_gl_fragcoord
)
1932 || (shader
->redeclares_gl_fragcoord
1933 && !linked_shader
->redeclares_gl_fragcoord
1934 && linked_shader
->uses_gl_fragcoord
)) {
1935 linker_error(prog
, "fragment shader defined with conflicting "
1936 "layout qualifiers for gl_FragCoord\n");
1939 /* From the GLSL 1.50 spec, page 39:
1941 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1942 * single program must have the same set of qualifiers."
1944 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1945 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1946 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1947 linker_error(prog
, "fragment shader defined with conflicting "
1948 "layout qualifiers for gl_FragCoord\n");
1951 /* Update the linked shader state. Note that uses_gl_fragcoord should
1952 * accumulate the results. The other values should replace. If there
1953 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1954 * are already known to be the same.
1956 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1957 linked_shader
->redeclares_gl_fragcoord
=
1958 shader
->redeclares_gl_fragcoord
;
1959 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1960 || shader
->uses_gl_fragcoord
;
1961 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1962 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1965 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1970 * Performs the cross-validation of geometry shader max_vertices and
1971 * primitive type layout qualifiers for the attached geometry shaders,
1972 * and propagates them to the linked GS and linked shader program.
1975 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1976 struct gl_shader
*linked_shader
,
1977 struct gl_shader
**shader_list
,
1978 unsigned num_shaders
)
1980 linked_shader
->Geom
.VerticesOut
= -1;
1981 linked_shader
->Geom
.Invocations
= 0;
1982 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1983 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1985 /* No in/out qualifiers defined for anything but GLSL 1.50+
1986 * geometry shaders so far.
1988 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1991 /* From the GLSL 1.50 spec, page 46:
1993 * "All geometry shader output layout declarations in a program
1994 * must declare the same layout and same value for
1995 * max_vertices. There must be at least one geometry output
1996 * layout declaration somewhere in a program, but not all
1997 * geometry shaders (compilation units) are required to
2001 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2002 struct gl_shader
*shader
= shader_list
[i
];
2004 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
2005 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
2006 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
2007 linker_error(prog
, "geometry shader defined with conflicting "
2011 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
2014 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
2015 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
2016 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
2017 linker_error(prog
, "geometry shader defined with conflicting "
2021 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
2024 if (shader
->Geom
.VerticesOut
!= -1) {
2025 if (linked_shader
->Geom
.VerticesOut
!= -1 &&
2026 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
2027 linker_error(prog
, "geometry shader defined with conflicting "
2028 "output vertex count (%d and %d)\n",
2029 linked_shader
->Geom
.VerticesOut
,
2030 shader
->Geom
.VerticesOut
);
2033 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
2036 if (shader
->Geom
.Invocations
!= 0) {
2037 if (linked_shader
->Geom
.Invocations
!= 0 &&
2038 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
2039 linker_error(prog
, "geometry shader defined with conflicting "
2040 "invocation count (%d and %d)\n",
2041 linked_shader
->Geom
.Invocations
,
2042 shader
->Geom
.Invocations
);
2045 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
2049 /* Just do the intrastage -> interstage propagation right now,
2050 * since we already know we're in the right type of shader program
2053 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
2055 "geometry shader didn't declare primitive input type\n");
2058 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
2060 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
2062 "geometry shader didn't declare primitive output type\n");
2065 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
2067 if (linked_shader
->Geom
.VerticesOut
== -1) {
2069 "geometry shader didn't declare max_vertices\n");
2072 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
2074 if (linked_shader
->Geom
.Invocations
== 0)
2075 linked_shader
->Geom
.Invocations
= 1;
2077 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
2082 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2083 * qualifiers for the attached compute shaders, and propagate them to the
2084 * linked CS and linked shader program.
2087 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2088 struct gl_shader
*linked_shader
,
2089 struct gl_shader
**shader_list
,
2090 unsigned num_shaders
)
2092 for (int i
= 0; i
< 3; i
++)
2093 linked_shader
->Comp
.LocalSize
[i
] = 0;
2095 /* This function is called for all shader stages, but it only has an effect
2096 * for compute shaders.
2098 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2101 /* From the ARB_compute_shader spec, in the section describing local size
2104 * If multiple compute shaders attached to a single program object
2105 * declare local work-group size, the declarations must be identical;
2106 * otherwise a link-time error results. Furthermore, if a program
2107 * object contains any compute shaders, at least one must contain an
2108 * input layout qualifier specifying the local work sizes of the
2109 * program, or a link-time error will occur.
2111 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2112 struct gl_shader
*shader
= shader_list
[sh
];
2114 if (shader
->Comp
.LocalSize
[0] != 0) {
2115 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
2116 for (int i
= 0; i
< 3; i
++) {
2117 if (linked_shader
->Comp
.LocalSize
[i
] !=
2118 shader
->Comp
.LocalSize
[i
]) {
2119 linker_error(prog
, "compute shader defined with conflicting "
2125 for (int i
= 0; i
< 3; i
++)
2126 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
2130 /* Just do the intrastage -> interstage propagation right now,
2131 * since we already know we're in the right type of shader program
2134 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
2135 linker_error(prog
, "compute shader didn't declare local size\n");
2138 for (int i
= 0; i
< 3; i
++)
2139 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
2144 * Combine a group of shaders for a single stage to generate a linked shader
2147 * If this function is supplied a single shader, it is cloned, and the new
2148 * shader is returned.
2150 static struct gl_shader
*
2151 link_intrastage_shaders(void *mem_ctx
,
2152 struct gl_context
*ctx
,
2153 struct gl_shader_program
*prog
,
2154 struct gl_shader
**shader_list
,
2155 unsigned num_shaders
)
2157 struct gl_uniform_block
*ubo_blocks
= NULL
;
2158 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2159 unsigned num_ubo_blocks
= 0;
2160 unsigned num_ssbo_blocks
= 0;
2162 /* Check that global variables defined in multiple shaders are consistent.
2164 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
2165 if (!prog
->LinkStatus
)
2168 /* Check that interface blocks defined in multiple shaders are consistent.
2170 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2172 if (!prog
->LinkStatus
)
2175 /* Check that there is only a single definition of each function signature
2176 * across all shaders.
2178 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2179 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2180 ir_function
*const f
= node
->as_function();
2185 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2186 ir_function
*const other
=
2187 shader_list
[j
]->symbols
->get_function(f
->name
);
2189 /* If the other shader has no function (and therefore no function
2190 * signatures) with the same name, skip to the next shader.
2195 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2196 if (!sig
->is_defined
|| sig
->is_builtin())
2199 ir_function_signature
*other_sig
=
2200 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2202 if ((other_sig
!= NULL
) && other_sig
->is_defined
2203 && !other_sig
->is_builtin()) {
2204 linker_error(prog
, "function `%s' is multiply defined\n",
2213 /* Find the shader that defines main, and make a clone of it.
2215 * Starting with the clone, search for undefined references. If one is
2216 * found, find the shader that defines it. Clone the reference and add
2217 * it to the shader. Repeat until there are no undefined references or
2218 * until a reference cannot be resolved.
2220 gl_shader
*main
= NULL
;
2221 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2222 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2223 main
= shader_list
[i
];
2229 linker_error(prog
, "%s shader lacks `main'\n",
2230 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2234 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, shader_list
[0]->Stage
);
2235 linked
->ir
= new(linked
) exec_list
;
2236 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2238 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2239 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2240 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2241 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2242 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2243 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2246 populate_symbol_table(linked
);
2248 /* The pointer to the main function in the final linked shader (i.e., the
2249 * copy of the original shader that contained the main function).
2251 ir_function_signature
*const main_sig
=
2252 _mesa_get_main_function_signature(linked
);
2254 /* Move any instructions other than variable declarations or function
2255 * declarations into main.
2257 exec_node
*insertion_point
=
2258 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2261 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2262 if (shader_list
[i
] == main
)
2265 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2266 insertion_point
, true, linked
);
2269 /* Check if any shader needs built-in functions. */
2270 bool need_builtins
= false;
2271 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2272 if (shader_list
[i
]->uses_builtin_functions
) {
2273 need_builtins
= true;
2279 if (need_builtins
) {
2280 /* Make a temporary array one larger than shader_list, which will hold
2281 * the built-in function shader as well.
2283 gl_shader
**linking_shaders
= (gl_shader
**)
2284 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2286 ok
= linking_shaders
!= NULL
;
2289 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2290 _mesa_glsl_initialize_builtin_functions();
2291 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2293 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2295 free(linking_shaders
);
2297 _mesa_error_no_memory(__func__
);
2300 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2305 _mesa_delete_shader(ctx
, linked
);
2309 /* Make a pass over all variable declarations to ensure that arrays with
2310 * unspecified sizes have a size specified. The size is inferred from the
2311 * max_array_access field.
2313 array_sizing_visitor v
;
2315 v
.fixup_unnamed_interface_types();
2317 /* Link up uniform blocks defined within this stage. */
2318 link_uniform_blocks(mem_ctx
, ctx
, prog
, &linked
, 1,
2319 &ubo_blocks
, &num_ubo_blocks
, &ssbo_blocks
,
2322 if (!prog
->LinkStatus
) {
2323 _mesa_delete_shader(ctx
, linked
);
2327 /* Copy ubo blocks to linked shader list */
2328 linked
->UniformBlocks
=
2329 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2330 ralloc_steal(linked
, ubo_blocks
);
2331 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2332 linked
->UniformBlocks
[i
] = &ubo_blocks
[i
];
2334 linked
->NumUniformBlocks
= num_ubo_blocks
;
2336 /* Copy ssbo blocks to linked shader list */
2337 linked
->ShaderStorageBlocks
=
2338 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2339 ralloc_steal(linked
, ssbo_blocks
);
2340 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2341 linked
->ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2343 linked
->NumShaderStorageBlocks
= num_ssbo_blocks
;
2345 /* At this point linked should contain all of the linked IR, so
2346 * validate it to make sure nothing went wrong.
2348 validate_ir_tree(linked
->ir
);
2350 /* Set the size of geometry shader input arrays */
2351 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2352 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2353 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2354 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2355 ir
->accept(&input_resize_visitor
);
2359 if (ctx
->Const
.VertexID_is_zero_based
)
2360 lower_vertex_id(linked
);
2362 /* Validate correct usage of barrier() in the tess control shader */
2363 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2364 barrier_use_visitor
visitor(prog
);
2365 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2366 ir
->accept(&visitor
);
2374 * Update the sizes of linked shader uniform arrays to the maximum
2377 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2379 * If one or more elements of an array are active,
2380 * GetActiveUniform will return the name of the array in name,
2381 * subject to the restrictions listed above. The type of the array
2382 * is returned in type. The size parameter contains the highest
2383 * array element index used, plus one. The compiler or linker
2384 * determines the highest index used. There will be only one
2385 * active uniform reported by the GL per uniform array.
2389 update_array_sizes(struct gl_shader_program
*prog
)
2391 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2392 if (prog
->_LinkedShaders
[i
] == NULL
)
2395 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2396 ir_variable
*const var
= node
->as_variable();
2398 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2399 !var
->type
->is_array())
2402 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2403 * will not be eliminated. Since we always do std140, just
2404 * don't resize arrays in UBOs.
2406 * Atomic counters are supposed to get deterministic
2407 * locations assigned based on the declaration ordering and
2408 * sizes, array compaction would mess that up.
2410 * Subroutine uniforms are not removed.
2412 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2413 var
->type
->contains_subroutine() || var
->constant_initializer
)
2416 int size
= var
->data
.max_array_access
;
2417 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2418 if (prog
->_LinkedShaders
[j
] == NULL
)
2421 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2422 ir_variable
*other_var
= node2
->as_variable();
2426 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2427 other_var
->data
.max_array_access
> size
) {
2428 size
= other_var
->data
.max_array_access
;
2433 if (size
+ 1 != (int)var
->type
->length
) {
2434 /* If this is a built-in uniform (i.e., it's backed by some
2435 * fixed-function state), adjust the number of state slots to
2436 * match the new array size. The number of slots per array entry
2437 * is not known. It seems safe to assume that the total number of
2438 * slots is an integer multiple of the number of array elements.
2439 * Determine the number of slots per array element by dividing by
2440 * the old (total) size.
2442 const unsigned num_slots
= var
->get_num_state_slots();
2443 if (num_slots
> 0) {
2444 var
->set_num_state_slots((size
+ 1)
2445 * (num_slots
/ var
->type
->length
));
2448 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2450 /* FINISHME: We should update the types of array
2451 * dereferences of this variable now.
2459 * Resize tessellation evaluation per-vertex inputs to the size of
2460 * tessellation control per-vertex outputs.
2463 resize_tes_inputs(struct gl_context
*ctx
,
2464 struct gl_shader_program
*prog
)
2466 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2469 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2470 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2472 /* If no control shader is present, then the TES inputs are statically
2473 * sized to MaxPatchVertices; the actual size of the arrays won't be
2474 * known until draw time.
2476 const int num_vertices
= tcs
2477 ? tcs
->TessCtrl
.VerticesOut
2478 : ctx
->Const
.MaxPatchVertices
;
2480 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2481 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2482 ir
->accept(&input_resize_visitor
);
2485 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2486 /* Convert the gl_PatchVerticesIn system value into a constant, since
2487 * the value is known at this point.
2489 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2490 ir_variable
*var
= ir
->as_variable();
2491 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2492 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2493 void *mem_ctx
= ralloc_parent(var
);
2494 var
->data
.location
= 0;
2495 var
->data
.explicit_location
= false;
2497 var
->data
.mode
= ir_var_auto
;
2498 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2500 var
->data
.mode
= ir_var_uniform
;
2501 var
->data
.how_declared
= ir_var_hidden
;
2502 var
->allocate_state_slots(1);
2503 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2504 slot0
->swizzle
= SWIZZLE_XXXX
;
2505 slot0
->tokens
[0] = STATE_INTERNAL
;
2506 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2507 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2508 slot0
->tokens
[i
] = 0;
2516 * Find a contiguous set of available bits in a bitmask.
2518 * \param used_mask Bits representing used (1) and unused (0) locations
2519 * \param needed_count Number of contiguous bits needed.
2522 * Base location of the available bits on success or -1 on failure.
2525 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2527 unsigned needed_mask
= (1 << needed_count
) - 1;
2528 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2530 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2531 * cannot optimize possibly infinite loops" for the loop below.
2533 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2536 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2537 if ((needed_mask
& ~used_mask
) == needed_mask
)
2548 * Assign locations for either VS inputs or FS outputs
2550 * \param prog Shader program whose variables need locations assigned
2551 * \param constants Driver specific constant values for the program.
2552 * \param target_index Selector for the program target to receive location
2553 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2554 * \c MESA_SHADER_FRAGMENT.
2557 * If locations are successfully assigned, true is returned. Otherwise an
2558 * error is emitted to the shader link log and false is returned.
2561 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2562 struct gl_constants
*constants
,
2563 unsigned target_index
)
2565 /* Maximum number of generic locations. This corresponds to either the
2566 * maximum number of draw buffers or the maximum number of generic
2569 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2570 constants
->Program
[target_index
].MaxAttribs
:
2571 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2573 /* Mark invalid locations as being used.
2575 unsigned used_locations
= (max_index
>= 32)
2576 ? ~0 : ~((1 << max_index
) - 1);
2577 unsigned double_storage_locations
= 0;
2579 assert((target_index
== MESA_SHADER_VERTEX
)
2580 || (target_index
== MESA_SHADER_FRAGMENT
));
2582 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2586 /* Operate in a total of four passes.
2588 * 1. Invalidate the location assignments for all vertex shader inputs.
2590 * 2. Assign locations for inputs that have user-defined (via
2591 * glBindVertexAttribLocation) locations and outputs that have
2592 * user-defined locations (via glBindFragDataLocation).
2594 * 3. Sort the attributes without assigned locations by number of slots
2595 * required in decreasing order. Fragmentation caused by attribute
2596 * locations assigned by the application may prevent large attributes
2597 * from having enough contiguous space.
2599 * 4. Assign locations to any inputs without assigned locations.
2602 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2603 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2605 const enum ir_variable_mode direction
=
2606 (target_index
== MESA_SHADER_VERTEX
)
2607 ? ir_var_shader_in
: ir_var_shader_out
;
2610 /* Temporary storage for the set of attributes that need locations assigned.
2616 /* Used below in the call to qsort. */
2617 static int compare(const void *a
, const void *b
)
2619 const temp_attr
*const l
= (const temp_attr
*) a
;
2620 const temp_attr
*const r
= (const temp_attr
*) b
;
2622 /* Reversed because we want a descending order sort below. */
2623 return r
->slots
- l
->slots
;
2626 assert(max_index
<= 32);
2628 /* Temporary array for the set of attributes that have locations assigned.
2630 ir_variable
*assigned
[16];
2632 unsigned num_attr
= 0;
2633 unsigned assigned_attr
= 0;
2635 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2636 ir_variable
*const var
= node
->as_variable();
2638 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2641 if (var
->data
.explicit_location
) {
2642 var
->data
.is_unmatched_generic_inout
= 0;
2643 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2644 || (var
->data
.location
< 0)) {
2646 "invalid explicit location %d specified for `%s'\n",
2647 (var
->data
.location
< 0)
2648 ? var
->data
.location
2649 : var
->data
.location
- generic_base
,
2653 } else if (target_index
== MESA_SHADER_VERTEX
) {
2656 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2657 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2658 var
->data
.location
= binding
;
2659 var
->data
.is_unmatched_generic_inout
= 0;
2661 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2665 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2666 assert(binding
>= FRAG_RESULT_DATA0
);
2667 var
->data
.location
= binding
;
2668 var
->data
.is_unmatched_generic_inout
= 0;
2670 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2671 var
->data
.index
= index
;
2676 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2678 * "Output binding assignments will cause LinkProgram to fail:
2680 * If the program has an active output assigned to a location greater
2681 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2682 * an active output assigned an index greater than or equal to one;"
2684 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2685 var
->data
.location
- generic_base
>=
2686 (int) constants
->MaxDualSourceDrawBuffers
) {
2688 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2689 "with index %u for %s\n",
2690 var
->data
.location
- generic_base
, var
->data
.index
,
2695 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2697 /* If the variable is not a built-in and has a location statically
2698 * assigned in the shader (presumably via a layout qualifier), make sure
2699 * that it doesn't collide with other assigned locations. Otherwise,
2700 * add it to the list of variables that need linker-assigned locations.
2702 if (var
->data
.location
!= -1) {
2703 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2704 /* From page 61 of the OpenGL 4.0 spec:
2706 * "LinkProgram will fail if the attribute bindings assigned
2707 * by BindAttribLocation do not leave not enough space to
2708 * assign a location for an active matrix attribute or an
2709 * active attribute array, both of which require multiple
2710 * contiguous generic attributes."
2712 * I think above text prohibits the aliasing of explicit and
2713 * automatic assignments. But, aliasing is allowed in manual
2714 * assignments of attribute locations. See below comments for
2717 * From OpenGL 4.0 spec, page 61:
2719 * "It is possible for an application to bind more than one
2720 * attribute name to the same location. This is referred to as
2721 * aliasing. This will only work if only one of the aliased
2722 * attributes is active in the executable program, or if no
2723 * path through the shader consumes more than one attribute of
2724 * a set of attributes aliased to the same location. A link
2725 * error can occur if the linker determines that every path
2726 * through the shader consumes multiple aliased attributes,
2727 * but implementations are not required to generate an error
2730 * From GLSL 4.30 spec, page 54:
2732 * "A program will fail to link if any two non-vertex shader
2733 * input variables are assigned to the same location. For
2734 * vertex shaders, multiple input variables may be assigned
2735 * to the same location using either layout qualifiers or via
2736 * the OpenGL API. However, such aliasing is intended only to
2737 * support vertex shaders where each execution path accesses
2738 * at most one input per each location. Implementations are
2739 * permitted, but not required, to generate link-time errors
2740 * if they detect that every path through the vertex shader
2741 * executable accesses multiple inputs assigned to any single
2742 * location. For all shader types, a program will fail to link
2743 * if explicit location assignments leave the linker unable
2744 * to find space for other variables without explicit
2747 * From OpenGL ES 3.0 spec, page 56:
2749 * "Binding more than one attribute name to the same location
2750 * is referred to as aliasing, and is not permitted in OpenGL
2751 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2752 * fail when this condition exists. However, aliasing is
2753 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2754 * This will only work if only one of the aliased attributes
2755 * is active in the executable program, or if no path through
2756 * the shader consumes more than one attribute of a set of
2757 * attributes aliased to the same location. A link error can
2758 * occur if the linker determines that every path through the
2759 * shader consumes multiple aliased attributes, but implemen-
2760 * tations are not required to generate an error in this case."
2762 * After looking at above references from OpenGL, OpenGL ES and
2763 * GLSL specifications, we allow aliasing of vertex input variables
2764 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2766 * NOTE: This is not required by the spec but its worth mentioning
2767 * here that we're not doing anything to make sure that no path
2768 * through the vertex shader executable accesses multiple inputs
2769 * assigned to any single location.
2772 /* Mask representing the contiguous slots that will be used by
2775 const unsigned attr
= var
->data
.location
- generic_base
;
2776 const unsigned use_mask
= (1 << slots
) - 1;
2777 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2778 ? "vertex shader input" : "fragment shader output";
2780 /* Generate a link error if the requested locations for this
2781 * attribute exceed the maximum allowed attribute location.
2783 if (attr
+ slots
> max_index
) {
2785 "insufficient contiguous locations "
2786 "available for %s `%s' %d %d %d\n", string
,
2787 var
->name
, used_locations
, use_mask
, attr
);
2791 /* Generate a link error if the set of bits requested for this
2792 * attribute overlaps any previously allocated bits.
2794 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2795 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2796 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2799 * "Additionally, for fragment shader outputs, if two
2800 * variables are placed within the same location, they
2801 * must have the same underlying type (floating-point or
2802 * integer). No component aliasing of output variables or
2803 * members is allowed.
2805 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2806 unsigned assigned_slots
=
2807 assigned
[i
]->type
->count_attribute_slots(false);
2808 unsigned assig_attr
=
2809 assigned
[i
]->data
.location
- generic_base
;
2810 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2812 if ((assigned_use_mask
<< assig_attr
) &
2813 (use_mask
<< attr
)) {
2815 const glsl_type
*assigned_type
=
2816 assigned
[i
]->type
->without_array();
2817 const glsl_type
*type
= var
->type
->without_array();
2818 if (assigned_type
->base_type
!= type
->base_type
) {
2819 linker_error(prog
, "types do not match for aliased"
2820 " %ss %s and %s\n", string
,
2821 assigned
[i
]->name
, var
->name
);
2825 unsigned assigned_component_mask
=
2826 ((1 << assigned_type
->vector_elements
) - 1) <<
2827 assigned
[i
]->data
.location_frac
;
2828 unsigned component_mask
=
2829 ((1 << type
->vector_elements
) - 1) <<
2830 var
->data
.location_frac
;
2831 if (assigned_component_mask
& component_mask
) {
2832 linker_error(prog
, "overlapping component is "
2833 "assigned to %ss %s and %s "
2835 string
, assigned
[i
]->name
, var
->name
,
2836 var
->data
.location_frac
);
2841 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2842 (prog
->IsES
&& prog
->Version
>= 300)) {
2843 linker_error(prog
, "overlapping location is assigned "
2844 "to %s `%s' %d %d %d\n", string
, var
->name
,
2845 used_locations
, use_mask
, attr
);
2848 linker_warning(prog
, "overlapping location is assigned "
2849 "to %s `%s' %d %d %d\n", string
, var
->name
,
2850 used_locations
, use_mask
, attr
);
2854 used_locations
|= (use_mask
<< attr
);
2856 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2858 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2859 * active attribute variables may fail to link, unless
2860 * device-dependent optimizations are able to make the program
2861 * fit within available hardware resources. For the purposes
2862 * of this test, attribute variables of the type dvec3, dvec4,
2863 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2864 * count as consuming twice as many attributes as equivalent
2865 * single-precision types. While these types use the same number
2866 * of generic attributes as their single-precision equivalents,
2867 * implementations are permitted to consume two single-precision
2868 * vectors of internal storage for each three- or four-component
2869 * double-precision vector."
2871 * Mark this attribute slot as taking up twice as much space
2872 * so we can count it properly against limits. According to
2873 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2874 * is optional behavior, but it seems preferable.
2876 if (var
->type
->without_array()->is_dual_slot())
2877 double_storage_locations
|= (use_mask
<< attr
);
2880 assigned
[assigned_attr
] = var
;
2886 if (num_attr
>= max_index
) {
2887 linker_error(prog
, "too many %s (max %u)",
2888 target_index
== MESA_SHADER_VERTEX
?
2889 "vertex shader inputs" : "fragment shader outputs",
2893 to_assign
[num_attr
].slots
= slots
;
2894 to_assign
[num_attr
].var
= var
;
2898 if (target_index
== MESA_SHADER_VERTEX
) {
2899 unsigned total_attribs_size
=
2900 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2901 _mesa_bitcount(double_storage_locations
);
2902 if (total_attribs_size
> max_index
) {
2904 "attempt to use %d vertex attribute slots only %d available ",
2905 total_attribs_size
, max_index
);
2910 /* If all of the attributes were assigned locations by the application (or
2911 * are built-in attributes with fixed locations), return early. This should
2912 * be the common case.
2917 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2919 if (target_index
== MESA_SHADER_VERTEX
) {
2920 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2921 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2922 * reserved to prevent it from being automatically allocated below.
2924 find_deref_visitor
find("gl_Vertex");
2926 if (find
.variable_found())
2927 used_locations
|= (1 << 0);
2930 for (unsigned i
= 0; i
< num_attr
; i
++) {
2931 /* Mask representing the contiguous slots that will be used by this
2934 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2936 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2939 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2940 ? "vertex shader input" : "fragment shader output";
2943 "insufficient contiguous locations "
2944 "available for %s `%s'\n",
2945 string
, to_assign
[i
].var
->name
);
2949 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2950 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2951 used_locations
|= (use_mask
<< location
);
2953 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2954 double_storage_locations
|= (use_mask
<< location
);
2957 /* Now that we have all the locations, from the GL 4.5 core spec, section
2958 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2959 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2960 * as equivalent single-precision types.
2962 if (target_index
== MESA_SHADER_VERTEX
) {
2963 unsigned total_attribs_size
=
2964 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2965 _mesa_bitcount(double_storage_locations
);
2966 if (total_attribs_size
> max_index
) {
2968 "attempt to use %d vertex attribute slots only %d available ",
2969 total_attribs_size
, max_index
);
2978 * Match explicit locations of outputs to inputs and deactivate the
2979 * unmatch flag if found so we don't optimise them away.
2982 match_explicit_outputs_to_inputs(gl_shader
*producer
,
2983 gl_shader
*consumer
)
2985 glsl_symbol_table parameters
;
2986 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2989 /* Find all shader outputs in the "producer" stage.
2991 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2992 ir_variable
*const var
= node
->as_variable();
2994 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2997 if (var
->data
.explicit_location
&&
2998 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2999 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
3000 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
3001 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
3005 /* Match inputs to outputs */
3006 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
3007 ir_variable
*const input
= node
->as_variable();
3009 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
3012 ir_variable
*output
= NULL
;
3013 if (input
->data
.explicit_location
3014 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
3015 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
3016 [input
->data
.location_frac
];
3018 if (output
!= NULL
){
3019 input
->data
.is_unmatched_generic_inout
= 0;
3020 output
->data
.is_unmatched_generic_inout
= 0;
3027 * Store the gl_FragDepth layout in the gl_shader_program struct.
3030 store_fragdepth_layout(struct gl_shader_program
*prog
)
3032 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3036 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3038 /* We don't look up the gl_FragDepth symbol directly because if
3039 * gl_FragDepth is not used in the shader, it's removed from the IR.
3040 * However, the symbol won't be removed from the symbol table.
3042 * We're only interested in the cases where the variable is NOT removed
3045 foreach_in_list(ir_instruction
, node
, ir
) {
3046 ir_variable
*const var
= node
->as_variable();
3048 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3052 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3053 switch (var
->data
.depth_layout
) {
3054 case ir_depth_layout_none
:
3055 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3057 case ir_depth_layout_any
:
3058 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3060 case ir_depth_layout_greater
:
3061 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3063 case ir_depth_layout_less
:
3064 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3066 case ir_depth_layout_unchanged
:
3067 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3078 * Validate the resources used by a program versus the implementation limits
3081 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3083 unsigned total_uniform_blocks
= 0;
3084 unsigned total_shader_storage_blocks
= 0;
3086 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3087 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3092 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3093 linker_error(prog
, "Too many %s shader texture samplers\n",
3094 _mesa_shader_stage_to_string(i
));
3097 if (sh
->num_uniform_components
>
3098 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3099 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3100 linker_warning(prog
, "Too many %s shader default uniform block "
3101 "components, but the driver will try to optimize "
3102 "them out; this is non-portable out-of-spec "
3104 _mesa_shader_stage_to_string(i
));
3106 linker_error(prog
, "Too many %s shader default uniform block "
3108 _mesa_shader_stage_to_string(i
));
3112 if (sh
->num_combined_uniform_components
>
3113 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3114 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3115 linker_warning(prog
, "Too many %s shader uniform components, "
3116 "but the driver will try to optimize them out; "
3117 "this is non-portable out-of-spec behavior\n",
3118 _mesa_shader_stage_to_string(i
));
3120 linker_error(prog
, "Too many %s shader uniform components\n",
3121 _mesa_shader_stage_to_string(i
));
3125 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3126 total_uniform_blocks
+= sh
->NumUniformBlocks
;
3128 const unsigned max_uniform_blocks
=
3129 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3130 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
3131 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3132 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
3133 max_uniform_blocks
);
3136 const unsigned max_shader_storage_blocks
=
3137 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3138 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
3139 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3140 _mesa_shader_stage_to_string(i
),
3141 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
3145 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3146 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3147 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3150 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3151 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3152 total_shader_storage_blocks
,
3153 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3156 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
3157 if (prog
->UniformBlocks
[i
].UniformBufferSize
>
3158 ctx
->Const
.MaxUniformBlockSize
) {
3159 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3160 prog
->UniformBlocks
[i
].Name
,
3161 prog
->UniformBlocks
[i
].UniformBufferSize
,
3162 ctx
->Const
.MaxUniformBlockSize
);
3166 for (unsigned i
= 0; i
< prog
->NumShaderStorageBlocks
; i
++) {
3167 if (prog
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3168 ctx
->Const
.MaxShaderStorageBlockSize
) {
3169 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3170 prog
->ShaderStorageBlocks
[i
].Name
,
3171 prog
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3172 ctx
->Const
.MaxShaderStorageBlockSize
);
3178 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3180 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3181 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3186 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3187 if (sh
->SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3190 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3195 sh
->NumSubroutineUniforms
++;
3197 if (sh
->NumSubroutineFunctions
== 0) {
3198 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3201 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3202 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3203 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3204 if (fn
->types
[k
] == uni
->type
) {
3210 uni
->num_compatible_subroutines
= count
;
3216 check_subroutine_resources(struct gl_shader_program
*prog
)
3218 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3219 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3222 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3223 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3224 _mesa_shader_stage_to_string(i
));
3229 * Validate shader image resources.
3232 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3234 unsigned total_image_units
= 0;
3235 unsigned fragment_outputs
= 0;
3236 unsigned total_shader_storage_blocks
= 0;
3238 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3241 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3242 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3245 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3246 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3247 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3248 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3250 total_image_units
+= sh
->NumImages
;
3251 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3253 if (i
== MESA_SHADER_FRAGMENT
) {
3254 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3255 ir_variable
*var
= node
->as_variable();
3256 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3257 /* since there are no double fs outputs - pass false */
3258 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3264 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3265 linker_error(prog
, "Too many combined image uniforms\n");
3267 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3268 ctx
->Const
.MaxCombinedShaderOutputResources
)
3269 linker_error(prog
, "Too many combined image uniforms, shader storage "
3270 " buffers and fragment outputs\n");
3275 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3276 * for a variable, checks for overlaps between other uniforms using explicit
3280 reserve_explicit_locations(struct gl_shader_program
*prog
,
3281 string_to_uint_map
*map
, ir_variable
*var
)
3283 unsigned slots
= var
->type
->uniform_locations();
3284 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3285 unsigned return_value
= slots
;
3287 /* Resize remap table if locations do not fit in the current one. */
3288 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3289 prog
->UniformRemapTable
=
3290 reralloc(prog
, prog
->UniformRemapTable
,
3291 gl_uniform_storage
*,
3294 if (!prog
->UniformRemapTable
) {
3295 linker_error(prog
, "Out of memory during linking.\n");
3299 /* Initialize allocated space. */
3300 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3301 prog
->UniformRemapTable
[i
] = NULL
;
3303 prog
->NumUniformRemapTable
= max_loc
+ 1;
3306 for (unsigned i
= 0; i
< slots
; i
++) {
3307 unsigned loc
= var
->data
.location
+ i
;
3309 /* Check if location is already used. */
3310 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3312 /* Possibly same uniform from a different stage, this is ok. */
3314 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3319 /* ARB_explicit_uniform_location specification states:
3321 * "No two default-block uniform variables in the program can have
3322 * the same location, even if they are unused, otherwise a compiler
3323 * or linker error will be generated."
3326 "location qualifier for uniform %s overlaps "
3327 "previously used location\n",
3332 /* Initialize location as inactive before optimization
3333 * rounds and location assignment.
3335 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3338 /* Note, base location used for arrays. */
3339 map
->put(var
->data
.location
, var
->name
);
3341 return return_value
;
3345 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3346 struct gl_shader
*sh
,
3349 unsigned slots
= var
->type
->uniform_locations();
3350 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3352 /* Resize remap table if locations do not fit in the current one. */
3353 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3354 sh
->SubroutineUniformRemapTable
=
3355 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3356 gl_uniform_storage
*,
3359 if (!sh
->SubroutineUniformRemapTable
) {
3360 linker_error(prog
, "Out of memory during linking.\n");
3364 /* Initialize allocated space. */
3365 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3366 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3368 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3371 for (unsigned i
= 0; i
< slots
; i
++) {
3372 unsigned loc
= var
->data
.location
+ i
;
3374 /* Check if location is already used. */
3375 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3377 /* ARB_explicit_uniform_location specification states:
3378 * "No two subroutine uniform variables can have the same location
3379 * in the same shader stage, otherwise a compiler or linker error
3380 * will be generated."
3383 "location qualifier for uniform %s overlaps "
3384 "previously used location\n",
3389 /* Initialize location as inactive before optimization
3390 * rounds and location assignment.
3392 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3398 * Check and reserve all explicit uniform locations, called before
3399 * any optimizations happen to handle also inactive uniforms and
3400 * inactive array elements that may get trimmed away.
3403 check_explicit_uniform_locations(struct gl_context
*ctx
,
3404 struct gl_shader_program
*prog
)
3406 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3409 /* This map is used to detect if overlapping explicit locations
3410 * occur with the same uniform (from different stage) or a different one.
3412 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3415 linker_error(prog
, "Out of memory during linking.\n");
3419 unsigned entries_total
= 0;
3420 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3421 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3426 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3427 ir_variable
*var
= node
->as_variable();
3428 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3431 if (var
->data
.explicit_location
) {
3433 if (var
->type
->without_array()->is_subroutine())
3434 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3436 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3440 entries_total
+= slots
;
3451 struct empty_uniform_block
*current_block
= NULL
;
3453 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3454 /* We found empty space in UniformRemapTable. */
3455 if (prog
->UniformRemapTable
[i
] == NULL
) {
3456 /* We've found the beginning of a new continous block of empty slots */
3457 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3458 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3459 current_block
->start
= i
;
3460 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3461 ¤t_block
->link
);
3464 /* The current block continues, so we simply increment its slots */
3465 current_block
->slots
++;
3470 return entries_total
;
3474 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3475 GLenum type
, const char *name
)
3477 bool found_interface
= false;
3478 unsigned block_name_len
= 0;
3479 const char *block_name_dot
= strchr(name
, '.');
3481 /* These rules only apply to buffer variables. So we return
3482 * true for the rest of types.
3484 if (type
!= GL_BUFFER_VARIABLE
)
3487 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
3488 const char *block_name
= shProg
->ShaderStorageBlocks
[i
].Name
;
3489 block_name_len
= strlen(block_name
);
3491 const char *block_square_bracket
= strchr(block_name
, '[');
3492 if (block_square_bracket
) {
3493 /* The block is part of an array of named interfaces,
3494 * for the name comparison we ignore the "[x]" part.
3496 block_name_len
-= strlen(block_square_bracket
);
3499 if (block_name_dot
) {
3500 /* Check if the variable name starts with the interface
3501 * name. The interface name (if present) should have the
3502 * length than the interface block name we are comparing to.
3504 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3505 if (len
!= block_name_len
)
3509 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3510 found_interface
= true;
3515 /* We remove the interface name from the buffer variable name,
3516 * including the dot that follows it.
3518 if (found_interface
)
3519 name
= name
+ block_name_len
+ 1;
3521 /* The ARB_program_interface_query spec says:
3523 * "For an active shader storage block member declared as an array, an
3524 * entry will be generated only for the first array element, regardless
3525 * of its type. For arrays of aggregate types, the enumeration rules
3526 * are applied recursively for the single enumerated array element."
3528 const char *struct_first_dot
= strchr(name
, '.');
3529 const char *first_square_bracket
= strchr(name
, '[');
3531 /* The buffer variable is on top level and it is not an array */
3532 if (!first_square_bracket
) {
3534 /* The shader storage block member is a struct, then generate the entry */
3535 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3538 /* Shader storage block member is an array, only generate an entry for the
3539 * first array element.
3541 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3549 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3550 const void *data
, uint8_t stages
)
3554 /* If resource already exists, do not add it again. */
3555 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3556 if (prog
->ProgramResourceList
[i
].Data
== data
)
3559 prog
->ProgramResourceList
=
3561 prog
->ProgramResourceList
,
3562 gl_program_resource
,
3563 prog
->NumProgramResourceList
+ 1);
3565 if (!prog
->ProgramResourceList
) {
3566 linker_error(prog
, "Out of memory during linking.\n");
3570 struct gl_program_resource
*res
=
3571 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3575 res
->StageReferences
= stages
;
3577 prog
->NumProgramResourceList
++;
3582 /* Function checks if a variable var is a packed varying and
3583 * if given name is part of packed varying's list.
3585 * If a variable is a packed varying, it has a name like
3586 * 'packed:a,b,c' where a, b and c are separate variables.
3589 included_in_packed_varying(ir_variable
*var
, const char *name
)
3591 if (strncmp(var
->name
, "packed:", 7) != 0)
3594 char *list
= strdup(var
->name
+ 7);
3599 char *token
= strtok_r(list
, ",", &saveptr
);
3601 if (strcmp(token
, name
) == 0) {
3605 token
= strtok_r(NULL
, ",", &saveptr
);
3612 * Function builds a stage reference bitmask from variable name.
3615 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3620 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3621 * used for reference mask in gl_program_resource will need to be changed.
3623 assert(MESA_SHADER_STAGES
< 8);
3625 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3626 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3630 /* Shader symbol table may contain variables that have
3631 * been optimized away. Search IR for the variable instead.
3633 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3634 ir_variable
*var
= node
->as_variable();
3636 unsigned baselen
= strlen(var
->name
);
3638 if (included_in_packed_varying(var
, name
)) {
3643 /* Type needs to match if specified, otherwise we might
3644 * pick a variable with same name but different interface.
3646 if (var
->data
.mode
!= mode
)
3649 if (strncmp(var
->name
, name
, baselen
) == 0) {
3650 /* Check for exact name matches but also check for arrays and
3653 if (name
[baselen
] == '\0' ||
3654 name
[baselen
] == '[' ||
3655 name
[baselen
] == '.') {
3667 * Create gl_shader_variable from ir_variable class.
3669 static gl_shader_variable
*
3670 create_shader_variable(struct gl_shader_program
*shProg
,
3671 const ir_variable
*in
,
3672 const char *name
, const glsl_type
*type
,
3673 bool use_implicit_location
, int location
,
3674 const glsl_type
*outermost_struct_type
)
3676 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3680 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3681 * expect to see gl_VertexID in the program resource list. Pretend.
3683 if (in
->data
.mode
== ir_var_system_value
&&
3684 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3685 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3687 out
->name
= ralloc_strdup(shProg
, name
);
3693 /* The ARB_program_interface_query spec says:
3695 * "Not all active variables are assigned valid locations; the
3696 * following variables will have an effective location of -1:
3698 * * uniforms declared as atomic counters;
3700 * * members of a uniform block;
3702 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3704 * * inputs or outputs not declared with a "location" layout
3705 * qualifier, except for vertex shader inputs and fragment shader
3708 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3709 is_gl_identifier(in
->name
) ||
3710 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3713 out
->location
= location
;
3717 out
->outermost_struct_type
= outermost_struct_type
;
3718 out
->interface_type
= in
->get_interface_type();
3719 out
->component
= in
->data
.location_frac
;
3720 out
->index
= in
->data
.index
;
3721 out
->patch
= in
->data
.patch
;
3722 out
->mode
= in
->data
.mode
;
3723 out
->interpolation
= in
->data
.interpolation
;
3724 out
->explicit_location
= in
->data
.explicit_location
;
3725 out
->precision
= in
->data
.precision
;
3731 add_shader_variable(struct gl_shader_program
*shProg
, unsigned stage_mask
,
3732 GLenum programInterface
, ir_variable
*var
,
3733 const char *name
, const glsl_type
*type
,
3734 bool use_implicit_location
, int location
,
3735 const glsl_type
*outermost_struct_type
= NULL
)
3737 const bool is_vertex_input
=
3738 programInterface
== GL_PROGRAM_INPUT
&&
3739 stage_mask
== MESA_SHADER_VERTEX
;
3741 switch (type
->base_type
) {
3742 case GLSL_TYPE_STRUCT
: {
3743 /* The ARB_program_interface_query spec says:
3745 * "For an active variable declared as a structure, a separate entry
3746 * will be generated for each active structure member. The name of
3747 * each entry is formed by concatenating the name of the structure,
3748 * the "." character, and the name of the structure member. If a
3749 * structure member to enumerate is itself a structure or array,
3750 * these enumeration rules are applied recursively."
3752 if (outermost_struct_type
== NULL
)
3753 outermost_struct_type
= type
;
3755 unsigned field_location
= location
;
3756 for (unsigned i
= 0; i
< type
->length
; i
++) {
3757 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3758 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3759 if (!add_shader_variable(shProg
, stage_mask
, programInterface
,
3760 var
, field_name
, field
->type
,
3761 use_implicit_location
, field_location
,
3762 outermost_struct_type
))
3766 field
->type
->count_attribute_slots(is_vertex_input
);
3772 /* Issue #16 of the ARB_program_interface_query spec says:
3774 * "* If a variable is a member of an interface block without an
3775 * instance name, it is enumerated using just the variable name.
3777 * * If a variable is a member of an interface block with an instance
3778 * name, it is enumerated as "BlockName.Member", where "BlockName" is
3779 * the name of the interface block (not the instance name) and
3780 * "Member" is the name of the variable."
3782 const char *prefixed_name
= (var
->data
.from_named_ifc_block
&&
3783 !is_gl_identifier(var
->name
))
3784 ? ralloc_asprintf(shProg
, "%s.%s", var
->get_interface_type()->name
,
3788 /* The ARB_program_interface_query spec says:
3790 * "For an active variable declared as a single instance of a basic
3791 * type, a single entry will be generated, using the variable name
3792 * from the shader source."
3794 gl_shader_variable
*sha_v
=
3795 create_shader_variable(shProg
, var
, prefixed_name
, type
,
3796 use_implicit_location
, location
,
3797 outermost_struct_type
);
3801 return add_program_resource(shProg
, programInterface
, sha_v
, stage_mask
);
3807 add_interface_variables(struct gl_shader_program
*shProg
,
3808 unsigned stage
, GLenum programInterface
)
3810 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3812 foreach_in_list(ir_instruction
, node
, ir
) {
3813 ir_variable
*var
= node
->as_variable();
3815 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3820 switch (var
->data
.mode
) {
3821 case ir_var_system_value
:
3822 case ir_var_shader_in
:
3823 if (programInterface
!= GL_PROGRAM_INPUT
)
3825 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3826 : int(VARYING_SLOT_VAR0
);
3828 case ir_var_shader_out
:
3829 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3831 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3832 : int(VARYING_SLOT_VAR0
);
3838 /* Skip packed varyings, packed varyings are handled separately
3839 * by add_packed_varyings.
3841 if (strncmp(var
->name
, "packed:", 7) == 0)
3844 /* Skip fragdata arrays, these are handled separately
3845 * by add_fragdata_arrays.
3847 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3850 const bool vs_input_or_fs_output
=
3851 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3852 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3854 if (!add_shader_variable(shProg
, 1 << stage
, programInterface
,
3855 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3856 var
->data
.location
- loc_bias
))
3863 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
, GLenum type
)
3865 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3868 if (!sh
|| !sh
->packed_varyings
)
3871 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3872 ir_variable
*var
= node
->as_variable();
3874 switch (var
->data
.mode
) {
3875 case ir_var_shader_in
:
3876 iface
= GL_PROGRAM_INPUT
;
3878 case ir_var_shader_out
:
3879 iface
= GL_PROGRAM_OUTPUT
;
3882 unreachable("unexpected type");
3885 if (type
== iface
) {
3886 const int stage_mask
=
3887 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3888 if (!add_shader_variable(shProg
, stage_mask
,
3889 iface
, var
, var
->name
, var
->type
, false,
3890 var
->data
.location
- VARYING_SLOT_VAR0
))
3899 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3901 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3903 if (!sh
|| !sh
->fragdata_arrays
)
3906 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3907 ir_variable
*var
= node
->as_variable();
3909 assert(var
->data
.mode
== ir_var_shader_out
);
3911 if (!add_shader_variable(shProg
,
3912 1 << MESA_SHADER_FRAGMENT
,
3913 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3914 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3922 get_top_level_name(const char *name
)
3924 const char *first_dot
= strchr(name
, '.');
3925 const char *first_square_bracket
= strchr(name
, '[');
3928 /* The ARB_program_interface_query spec says:
3930 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3931 * the number of active array elements of the top-level shader storage
3932 * block member containing to the active variable is written to
3933 * <params>. If the top-level block member is not declared as an
3934 * array, the value one is written to <params>. If the top-level block
3935 * member is an array with no declared size, the value zero is written
3939 /* The buffer variable is on top level.*/
3940 if (!first_square_bracket
&& !first_dot
)
3941 name_size
= strlen(name
);
3942 else if ((!first_square_bracket
||
3943 (first_dot
&& first_dot
< first_square_bracket
)))
3944 name_size
= first_dot
- name
;
3946 name_size
= first_square_bracket
- name
;
3948 return strndup(name
, name_size
);
3952 get_var_name(const char *name
)
3954 const char *first_dot
= strchr(name
, '.');
3957 return strdup(name
);
3959 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3963 is_top_level_shader_storage_block_member(const char* name
,
3964 const char* interface_name
,
3965 const char* field_name
)
3967 bool result
= false;
3969 /* If the given variable is already a top-level shader storage
3970 * block member, then return array_size = 1.
3971 * We could have two possibilities: if we have an instanced
3972 * shader storage block or not instanced.
3974 * For the first, we check create a name as it was in top level and
3975 * compare it with the real name. If they are the same, then
3976 * the variable is already at top-level.
3978 * Full instanced name is: interface name + '.' + var name +
3981 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3982 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3983 if (!full_instanced_name
) {
3984 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3988 snprintf(full_instanced_name
, name_length
, "%s.%s",
3989 interface_name
, field_name
);
3991 /* Check if its top-level shader storage block member of an
3992 * instanced interface block, or of a unnamed interface block.
3994 if (strcmp(name
, full_instanced_name
) == 0 ||
3995 strcmp(name
, field_name
) == 0)
3998 free(full_instanced_name
);
4003 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4004 char *interface_name
, char *var_name
)
4006 /* The ARB_program_interface_query spec says:
4008 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4009 * the number of active array elements of the top-level shader storage
4010 * block member containing to the active variable is written to
4011 * <params>. If the top-level block member is not declared as an
4012 * array, the value one is written to <params>. If the top-level block
4013 * member is an array with no declared size, the value zero is written
4016 if (is_top_level_shader_storage_block_member(uni
->name
,
4020 else if (field
->type
->is_unsized_array())
4022 else if (field
->type
->is_array())
4023 return field
->type
->length
;
4029 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4030 const glsl_struct_field
*field
, char *interface_name
,
4033 /* The ARB_program_interface_query spec says:
4035 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4036 * identifying the stride between array elements of the top-level
4037 * shader storage block member containing the active variable is
4038 * written to <params>. For top-level block members declared as
4039 * arrays, the value written is the difference, in basic machine units,
4040 * between the offsets of the active variable for consecutive elements
4041 * in the top-level array. For top-level block members not declared as
4042 * an array, zero is written to <params>."
4044 if (field
->type
->is_array()) {
4045 const enum glsl_matrix_layout matrix_layout
=
4046 glsl_matrix_layout(field
->matrix_layout
);
4047 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4048 const glsl_type
*array_type
= field
->type
->fields
.array
;
4050 if (is_top_level_shader_storage_block_member(uni
->name
,
4055 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4056 if (array_type
->is_record() || array_type
->is_array())
4057 return glsl_align(array_type
->std140_size(row_major
), 16);
4059 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4061 return array_type
->std430_array_stride(row_major
);
4068 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4069 struct gl_uniform_storage
*uni
)
4071 int block_index
= uni
->block_index
;
4072 int array_size
= -1;
4073 int array_stride
= -1;
4074 char *var_name
= get_top_level_name(uni
->name
);
4075 char *interface_name
=
4076 get_top_level_name(uni
->is_shader_storage
?
4077 shProg
->ShaderStorageBlocks
[block_index
].Name
:
4078 shProg
->UniformBlocks
[block_index
].Name
);
4080 if (strcmp(var_name
, interface_name
) == 0) {
4081 /* Deal with instanced array of SSBOs */
4082 char *temp_name
= get_var_name(uni
->name
);
4084 linker_error(shProg
, "Out of memory during linking.\n");
4085 goto write_top_level_array_size_and_stride
;
4088 var_name
= get_top_level_name(temp_name
);
4091 linker_error(shProg
, "Out of memory during linking.\n");
4092 goto write_top_level_array_size_and_stride
;
4096 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
4097 if (shProg
->Shaders
[i
] == NULL
)
4100 const gl_shader
*stage
= shProg
->Shaders
[i
];
4101 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
4102 ir_variable
*var
= node
->as_variable();
4103 if (!var
|| !var
->get_interface_type() ||
4104 var
->data
.mode
!= ir_var_shader_storage
)
4107 const glsl_type
*interface
= var
->get_interface_type();
4109 if (strcmp(interface_name
, interface
->name
) != 0)
4112 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4113 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4114 if (strcmp(field
->name
, var_name
) != 0)
4117 array_stride
= get_array_stride(uni
, interface
, field
,
4118 interface_name
, var_name
);
4119 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4120 goto write_top_level_array_size_and_stride
;
4124 write_top_level_array_size_and_stride
:
4125 free(interface_name
);
4127 uni
->top_level_array_stride
= array_stride
;
4128 uni
->top_level_array_size
= array_size
;
4132 * Builds up a list of program resources that point to existing
4136 build_program_resource_list(struct gl_context
*ctx
,
4137 struct gl_shader_program
*shProg
)
4139 /* Rebuild resource list. */
4140 if (shProg
->ProgramResourceList
) {
4141 ralloc_free(shProg
->ProgramResourceList
);
4142 shProg
->ProgramResourceList
= NULL
;
4143 shProg
->NumProgramResourceList
= 0;
4146 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4148 /* Determine first input and final output stage. These are used to
4149 * detect which variables should be enumerated in the resource list
4150 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4152 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4153 if (!shProg
->_LinkedShaders
[i
])
4155 if (input_stage
== MESA_SHADER_STAGES
)
4160 /* Empty shader, no resources. */
4161 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4164 /* Program interface needs to expose varyings in case of SSO. */
4165 if (shProg
->SeparateShader
) {
4166 if (!add_packed_varyings(shProg
, input_stage
, GL_PROGRAM_INPUT
))
4169 if (!add_packed_varyings(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
4173 if (!add_fragdata_arrays(shProg
))
4176 /* Add inputs and outputs to the resource list. */
4177 if (!add_interface_variables(shProg
, input_stage
, GL_PROGRAM_INPUT
))
4180 if (!add_interface_variables(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
4183 /* Add transform feedback varyings. */
4184 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
4185 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
4186 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
4187 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
4193 /* Add transform feedback buffers. */
4194 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4195 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
4196 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
4197 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_BUFFER
,
4198 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
4204 /* Add uniforms from uniform storage. */
4205 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4206 /* Do not add uniforms internally used by Mesa. */
4207 if (shProg
->UniformStorage
[i
].hidden
)
4211 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
4214 /* Add stagereferences for uniforms in a uniform block. */
4215 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
4216 int block_index
= shProg
->UniformStorage
[i
].block_index
;
4217 if (block_index
!= -1) {
4218 stageref
|= is_shader_storage
?
4219 shProg
->ShaderStorageBlocks
[block_index
].stageref
:
4220 shProg
->UniformBlocks
[block_index
].stageref
;
4223 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4224 if (!should_add_buffer_variable(shProg
, type
,
4225 shProg
->UniformStorage
[i
].name
))
4228 if (is_shader_storage
) {
4229 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
4232 if (!add_program_resource(shProg
, type
,
4233 &shProg
->UniformStorage
[i
], stageref
))
4237 /* Add program uniform blocks. */
4238 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
4239 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
4240 &shProg
->UniformBlocks
[i
], 0))
4244 /* Add program shader storage blocks. */
4245 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
4246 if (!add_program_resource(shProg
, GL_SHADER_STORAGE_BLOCK
,
4247 &shProg
->ShaderStorageBlocks
[i
], 0))
4251 /* Add atomic counter buffers. */
4252 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
4253 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
4254 &shProg
->AtomicBuffers
[i
], 0))
4258 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4260 if (!shProg
->UniformStorage
[i
].hidden
)
4263 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4264 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
4265 !shProg
->UniformStorage
[i
].type
->is_subroutine())
4268 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4269 /* add shader subroutines */
4270 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
4275 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4276 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
4282 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4283 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4284 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
4291 * This check is done to make sure we allow only constant expression
4292 * indexing and "constant-index-expression" (indexing with an expression
4293 * that includes loop induction variable).
4296 validate_sampler_array_indexing(struct gl_context
*ctx
,
4297 struct gl_shader_program
*prog
)
4299 dynamic_sampler_array_indexing_visitor v
;
4300 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4301 if (prog
->_LinkedShaders
[i
] == NULL
)
4304 bool no_dynamic_indexing
=
4305 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4307 /* Search for array derefs in shader. */
4308 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4309 if (v
.uses_dynamic_sampler_array_indexing()) {
4310 const char *msg
= "sampler arrays indexed with non-constant "
4311 "expressions is forbidden in GLSL %s %u";
4312 /* Backend has indicated that it has no dynamic indexing support. */
4313 if (no_dynamic_indexing
) {
4314 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4317 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4325 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4327 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4328 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4333 sh
->MaxSubroutineFunctionIndex
= 0;
4334 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4335 ir_function
*fn
= node
->as_function();
4339 if (fn
->is_subroutine
)
4340 sh
->NumSubroutineUniformTypes
++;
4342 if (!fn
->num_subroutine_types
)
4345 /* these should have been calculated earlier. */
4346 assert(fn
->subroutine_index
!= -1);
4347 if (sh
->NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4348 linker_error(prog
, "Too many subroutine functions declared.\n");
4351 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4352 struct gl_subroutine_function
,
4353 sh
->NumSubroutineFunctions
+ 1);
4354 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4355 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4356 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4357 ralloc_array(sh
, const struct glsl_type
*,
4358 fn
->num_subroutine_types
);
4360 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4363 * "Each subroutine with an index qualifier in the shader must be
4364 * given a unique index, otherwise a compile or link error will be
4367 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4368 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4369 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4370 linker_error(prog
, "each subroutine index qualifier in the "
4371 "shader must be unique\n");
4375 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4376 fn
->subroutine_index
;
4378 if (fn
->subroutine_index
> (int)sh
->MaxSubroutineFunctionIndex
)
4379 sh
->MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4381 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4382 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4383 sh
->NumSubroutineFunctions
++;
4389 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4391 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4393 foreach_in_list(ir_instruction
, node
, ir
) {
4394 ir_variable
*const var
= node
->as_variable();
4396 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4399 /* Don't set always active on builtins that haven't been redeclared */
4400 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4403 var
->data
.always_active_io
= true;
4408 * When separate shader programs are enabled, only input/outputs between
4409 * the stages of a multi-stage separate program can be safely removed
4410 * from the shader interface. Other inputs/outputs must remain active.
4413 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4415 unsigned first
, last
;
4416 assert(prog
->SeparateShader
);
4418 first
= MESA_SHADER_STAGES
;
4421 /* Determine first and last stage. Excluding the compute stage */
4422 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4423 if (!prog
->_LinkedShaders
[i
])
4425 if (first
== MESA_SHADER_STAGES
)
4430 if (first
== MESA_SHADER_STAGES
)
4433 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4434 gl_shader
*sh
= prog
->_LinkedShaders
[stage
];
4438 if (first
== last
) {
4439 /* For a single shader program only allow inputs to the vertex shader
4440 * and outputs from the fragment shader to be removed.
4442 if (stage
!= MESA_SHADER_VERTEX
)
4443 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4444 if (stage
!= MESA_SHADER_FRAGMENT
)
4445 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4447 /* For multi-stage separate shader programs only allow inputs and
4448 * outputs between the shader stages to be removed as well as inputs
4449 * to the vertex shader and outputs from the fragment shader.
4451 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4452 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4453 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4454 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4460 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4462 prog
->LinkStatus
= true; /* All error paths will set this to false */
4463 prog
->Validated
= false;
4464 prog
->_Used
= false;
4466 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4468 * "Linking can fail for a variety of reasons as specified in the
4469 * OpenGL Shading Language Specification, as well as any of the
4470 * following reasons:
4472 * - No shader objects are attached to program."
4474 * The Compatibility Profile specification does not list the error. In
4475 * Compatibility Profile missing shader stages are replaced by
4476 * fixed-function. This applies to the case where all stages are
4479 if (prog
->NumShaders
== 0) {
4480 if (ctx
->API
!= API_OPENGL_COMPAT
)
4481 linker_error(prog
, "no shaders attached to the program\n");
4485 unsigned num_tfeedback_decls
= 0;
4486 unsigned int num_explicit_uniform_locs
= 0;
4487 bool has_xfb_qualifiers
= false;
4488 char **varying_names
= NULL
;
4489 tfeedback_decl
*tfeedback_decls
= NULL
;
4491 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4493 prog
->ARB_fragment_coord_conventions_enable
= false;
4495 /* Separate the shaders into groups based on their type.
4497 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4498 unsigned num_shaders
[MESA_SHADER_STAGES
];
4500 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4501 shader_list
[i
] = (struct gl_shader
**)
4502 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4506 unsigned min_version
= UINT_MAX
;
4507 unsigned max_version
= 0;
4508 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4509 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4510 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4512 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4513 linker_error(prog
, "all shaders must use same shading "
4514 "language version\n");
4518 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4519 prog
->ARB_fragment_coord_conventions_enable
= true;
4522 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4523 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4524 num_shaders
[shader_type
]++;
4527 /* In desktop GLSL, different shader versions may be linked together. In
4528 * GLSL ES, all shader versions must be the same.
4530 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4531 linker_error(prog
, "all shaders must use same shading "
4532 "language version\n");
4536 prog
->Version
= max_version
;
4537 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4539 /* Some shaders have to be linked with some other shaders present.
4541 if (!prog
->SeparateShader
) {
4542 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4543 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4544 linker_error(prog
, "Geometry shader must be linked with "
4548 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4549 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4550 linker_error(prog
, "Tessellation evaluation shader must be linked "
4551 "with vertex shader\n");
4554 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4555 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4556 linker_error(prog
, "Tessellation control shader must be linked with "
4561 /* The spec is self-contradictory here. It allows linking without a tess
4562 * eval shader, but that can only be used with transform feedback and
4563 * rasterization disabled. However, transform feedback isn't allowed
4564 * with GL_PATCHES, so it can't be used.
4566 * More investigation showed that the idea of transform feedback after
4567 * a tess control shader was dropped, because some hw vendors couldn't
4568 * support tessellation without a tess eval shader, but the linker
4569 * section wasn't updated to reflect that.
4571 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4574 * Do what's reasonable and always require a tess eval shader if a tess
4575 * control shader is present.
4577 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4578 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4579 linker_error(prog
, "Tessellation control shader must be linked with "
4580 "tessellation evaluation shader\n");
4585 /* Compute shaders have additional restrictions. */
4586 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4587 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4588 linker_error(prog
, "Compute shaders may not be linked with any other "
4589 "type of shader\n");
4592 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4593 if (prog
->_LinkedShaders
[i
] != NULL
)
4594 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4596 prog
->_LinkedShaders
[i
] = NULL
;
4599 /* Link all shaders for a particular stage and validate the result.
4601 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4602 if (num_shaders
[stage
] > 0) {
4603 gl_shader
*const sh
=
4604 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4605 num_shaders
[stage
]);
4607 if (!prog
->LinkStatus
) {
4609 _mesa_delete_shader(ctx
, sh
);
4614 case MESA_SHADER_VERTEX
:
4615 validate_vertex_shader_executable(prog
, sh
, ctx
);
4617 case MESA_SHADER_TESS_CTRL
:
4618 /* nothing to be done */
4620 case MESA_SHADER_TESS_EVAL
:
4621 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4623 case MESA_SHADER_GEOMETRY
:
4624 validate_geometry_shader_executable(prog
, sh
, ctx
);
4626 case MESA_SHADER_FRAGMENT
:
4627 validate_fragment_shader_executable(prog
, sh
);
4630 if (!prog
->LinkStatus
) {
4632 _mesa_delete_shader(ctx
, sh
);
4636 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4640 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4641 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4642 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4643 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4644 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4645 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4646 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4647 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4648 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4650 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4651 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4654 /* Here begins the inter-stage linking phase. Some initial validation is
4655 * performed, then locations are assigned for uniforms, attributes, and
4658 cross_validate_uniforms(prog
);
4659 if (!prog
->LinkStatus
)
4662 unsigned first
, last
, prev
;
4664 first
= MESA_SHADER_STAGES
;
4667 /* Determine first and last stage. */
4668 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4669 if (!prog
->_LinkedShaders
[i
])
4671 if (first
== MESA_SHADER_STAGES
)
4676 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4677 link_assign_subroutine_types(prog
);
4679 if (!prog
->LinkStatus
)
4682 resize_tes_inputs(ctx
, prog
);
4684 /* Validate the inputs of each stage with the output of the preceding
4688 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4689 if (prog
->_LinkedShaders
[i
] == NULL
)
4692 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4693 prog
->_LinkedShaders
[i
]);
4694 if (!prog
->LinkStatus
)
4697 cross_validate_outputs_to_inputs(prog
,
4698 prog
->_LinkedShaders
[prev
],
4699 prog
->_LinkedShaders
[i
]);
4700 if (!prog
->LinkStatus
)
4706 /* Cross-validate uniform blocks between shader stages */
4707 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4708 MESA_SHADER_STAGES
);
4709 if (!prog
->LinkStatus
)
4712 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4713 if (prog
->_LinkedShaders
[i
] != NULL
)
4714 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4717 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4718 * it before optimization because we want most of the checks to get
4719 * dropped thanks to constant propagation.
4721 * This rule also applies to GLSL ES 3.00.
4723 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4724 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4726 lower_discard_flow(sh
->ir
);
4730 if (prog
->SeparateShader
)
4731 disable_varying_optimizations_for_sso(prog
);
4734 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4738 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4741 /* Do common optimization before assigning storage for attributes,
4742 * uniforms, and varyings. Later optimization could possibly make
4743 * some of that unused.
4745 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4746 if (prog
->_LinkedShaders
[i
] == NULL
)
4749 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4750 if (!prog
->LinkStatus
)
4753 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4754 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4757 if (ctx
->Const
.LowerTessLevel
) {
4758 lower_tess_level(prog
->_LinkedShaders
[i
]);
4761 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4762 &ctx
->Const
.ShaderCompilerOptions
[i
],
4763 ctx
->Const
.NativeIntegers
))
4766 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4769 /* Validation for special cases where we allow sampler array indexing
4770 * with loop induction variable. This check emits a warning or error
4771 * depending if backend can handle dynamic indexing.
4773 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4774 (prog
->IsES
&& prog
->Version
< 300)) {
4775 if (!validate_sampler_array_indexing(ctx
, prog
))
4779 /* Check and validate stream emissions in geometry shaders */
4780 validate_geometry_shader_emissions(ctx
, prog
);
4782 /* Mark all generic shader inputs and outputs as unpaired. */
4783 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4784 if (prog
->_LinkedShaders
[i
] != NULL
) {
4785 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4790 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4791 if (prog
->_LinkedShaders
[i
] == NULL
)
4794 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4795 prog
->_LinkedShaders
[i
]);
4799 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4800 MESA_SHADER_VERTEX
)) {
4804 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4805 MESA_SHADER_FRAGMENT
)) {
4809 /* From the ARB_enhanced_layouts spec:
4811 * "If the shader used to record output variables for transform feedback
4812 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4813 * qualifiers, the values specified by TransformFeedbackVaryings are
4814 * ignored, and the set of variables captured for transform feedback is
4815 * instead derived from the specified layout qualifiers."
4817 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4818 /* Find last stage before fragment shader */
4819 if (prog
->_LinkedShaders
[i
]) {
4820 has_xfb_qualifiers
=
4821 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4822 &num_tfeedback_decls
,
4828 if (!has_xfb_qualifiers
) {
4829 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4830 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4833 if (num_tfeedback_decls
!= 0) {
4834 /* From GL_EXT_transform_feedback:
4835 * A program will fail to link if:
4837 * * the <count> specified by TransformFeedbackVaryingsEXT is
4838 * non-zero, but the program object has no vertex or geometry
4841 if (first
>= MESA_SHADER_FRAGMENT
) {
4842 linker_error(prog
, "Transform feedback varyings specified, but "
4843 "no vertex, tessellation, or geometry shader is "
4848 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4849 num_tfeedback_decls
);
4850 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4851 varying_names
, tfeedback_decls
))
4855 /* If there is no fragment shader we need to set transform feedback.
4857 * For SSO we also need to assign output locations. We assign them here
4858 * because we need to do it for both single stage programs and multi stage
4861 if (last
< MESA_SHADER_FRAGMENT
&&
4862 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4863 const uint64_t reserved_out_slots
=
4864 reserved_varying_slot(prog
->_LinkedShaders
[last
], ir_var_shader_out
);
4865 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4866 prog
->_LinkedShaders
[last
], NULL
,
4867 num_tfeedback_decls
, tfeedback_decls
,
4868 reserved_out_slots
))
4872 if (last
<= MESA_SHADER_FRAGMENT
) {
4873 /* Remove unused varyings from the first/last stage unless SSO */
4874 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4875 prog
->_LinkedShaders
[first
],
4877 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4878 prog
->_LinkedShaders
[last
],
4881 /* If the program is made up of only a single stage */
4882 if (first
== last
) {
4884 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4885 if (prog
->SeparateShader
) {
4886 const uint64_t reserved_slots
=
4887 reserved_varying_slot(sh
, ir_var_shader_in
);
4889 /* Assign input locations for SSO, output locations are already
4892 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4893 NULL
/* producer */,
4895 0 /* num_tfeedback_decls */,
4896 NULL
/* tfeedback_decls */,
4901 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4902 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4905 /* Linking the stages in the opposite order (from fragment to vertex)
4906 * ensures that inter-shader outputs written to in an earlier stage
4907 * are eliminated if they are (transitively) not used in a later
4911 for (int i
= next
- 1; i
>= 0; i
--) {
4912 if (prog
->_LinkedShaders
[i
] == NULL
&& i
!= 0)
4915 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4916 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4918 const uint64_t reserved_out_slots
=
4919 reserved_varying_slot(sh_i
, ir_var_shader_out
);
4920 const uint64_t reserved_in_slots
=
4921 reserved_varying_slot(sh_next
, ir_var_shader_in
);
4923 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4924 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4926 reserved_out_slots
| reserved_in_slots
))
4929 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4930 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4933 /* This must be done after all dead varyings are eliminated. */
4935 unsigned slots_used
= _mesa_bitcount_64(reserved_out_slots
);
4936 if (!check_against_output_limit(ctx
, prog
, sh_i
, slots_used
)) {
4941 unsigned slots_used
= _mesa_bitcount_64(reserved_in_slots
);
4942 if (!check_against_input_limit(ctx
, prog
, sh_next
, slots_used
))
4950 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4951 has_xfb_qualifiers
))
4954 update_array_sizes(prog
);
4955 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
,
4956 num_explicit_uniform_locs
,
4957 ctx
->Const
.MaxUserAssignableUniformLocations
);
4958 link_assign_atomic_counter_resources(ctx
, prog
);
4959 store_fragdepth_layout(prog
);
4961 link_calculate_subroutine_compat(prog
);
4962 check_resources(ctx
, prog
);
4963 check_subroutine_resources(prog
);
4964 check_image_resources(ctx
, prog
);
4965 link_check_atomic_counter_resources(ctx
, prog
);
4967 if (!prog
->LinkStatus
)
4970 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4971 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4972 * anything about shader linking when one of the shaders (vertex or
4973 * fragment shader) is absent. So, the extension shouldn't change the
4974 * behavior specified in GLSL specification.
4976 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4977 * "Linking can fail for a variety of reasons as specified in the
4978 * OpenGL ES Shading Language Specification, as well as any of the
4979 * following reasons:
4983 * * program contains objects to form either a vertex shader or
4984 * fragment shader, and program is not separable, and does not
4985 * contain objects to form both a vertex shader and fragment
4988 * However, the only scenario in 3.1+ where we don't require them both is
4989 * when we have a compute shader. For example:
4991 * - No shaders is a link error.
4992 * - Geom or Tess without a Vertex shader is a link error which means we
4993 * always require a Vertex shader and hence a Fragment shader.
4994 * - Finally a Compute shader linked with any other stage is a link error.
4996 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4997 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4998 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4999 linker_error(prog
, "program lacks a vertex shader\n");
5000 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
5001 linker_error(prog
, "program lacks a fragment shader\n");
5005 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5006 if (prog
->_LinkedShaders
[i
] == NULL
)
5009 const struct gl_shader_compiler_options
*options
=
5010 &ctx
->Const
.ShaderCompilerOptions
[i
];
5012 if (options
->LowerBufferInterfaceBlocks
)
5013 lower_ubo_reference(prog
->_LinkedShaders
[i
],
5014 options
->ClampBlockIndicesToArrayBounds
);
5016 if (options
->LowerShaderSharedVariables
)
5017 lower_shared_reference(prog
->_LinkedShaders
[i
],
5018 &prog
->Comp
.SharedSize
);
5020 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
5021 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
5025 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
5026 free(shader_list
[i
]);
5027 if (prog
->_LinkedShaders
[i
] == NULL
)
5030 /* Do a final validation step to make sure that the IR wasn't
5031 * invalidated by any modifications performed after intrastage linking.
5033 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
5035 /* Retain any live IR, but trash the rest. */
5036 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5038 /* The symbol table in the linked shaders may contain references to
5039 * variables that were removed (e.g., unused uniforms). Since it may
5040 * contain junk, there is no possible valid use. Delete it and set the
5043 delete prog
->_LinkedShaders
[i
]->symbols
;
5044 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5047 ralloc_free(mem_ctx
);