2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
68 #include "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/hash_table.h"
76 #include "link_varyings.h"
77 #include "ir_optimization.h"
78 #include "ir_rvalue_visitor.h"
79 #include "ir_uniform.h"
81 #include "main/shaderobj.h"
82 #include "main/enums.h"
88 * Visitor that determines whether or not a variable is ever written.
90 class find_assignment_visitor
: public ir_hierarchical_visitor
{
92 find_assignment_visitor(const char *name
)
93 : name(name
), found(false)
98 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
100 ir_variable
*const var
= ir
->lhs
->variable_referenced();
102 if (strcmp(name
, var
->name
) == 0) {
107 return visit_continue_with_parent
;
110 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
112 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
113 actual_node
, &ir
->actual_parameters
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
115 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
117 if (sig_param
->data
.mode
== ir_var_function_out
||
118 sig_param
->data
.mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
127 if (ir
->return_deref
!= NULL
) {
128 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
130 if (strcmp(name
, var
->name
) == 0) {
136 return visit_continue_with_parent
;
139 bool variable_found()
145 const char *name
; /**< Find writes to a variable with this name. */
146 bool found
; /**< Was a write to the variable found? */
151 * Visitor that determines whether or not a variable is ever read.
153 class find_deref_visitor
: public ir_hierarchical_visitor
{
155 find_deref_visitor(const char *name
)
156 : name(name
), found(false)
161 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
163 if (strcmp(this->name
, ir
->var
->name
) == 0) {
168 return visit_continue
;
171 bool variable_found() const
177 const char *name
; /**< Find writes to a variable with this name. */
178 bool found
; /**< Was a write to the variable found? */
182 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
184 unsigned num_vertices
;
185 gl_shader_program
*prog
;
187 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
189 this->num_vertices
= num_vertices
;
193 virtual ~geom_array_resize_visitor()
198 virtual ir_visitor_status
visit(ir_variable
*var
)
200 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
201 return visit_continue
;
203 unsigned size
= var
->type
->length
;
205 /* Generate a link error if the shader has declared this array with an
208 if (size
&& size
!= this->num_vertices
) {
209 linker_error(this->prog
, "size of array %s declared as %u, "
210 "but number of input vertices is %u\n",
211 var
->name
, size
, this->num_vertices
);
212 return visit_continue
;
215 /* Generate a link error if the shader attempts to access an input
216 * array using an index too large for its actual size assigned at link
219 if (var
->data
.max_array_access
>= this->num_vertices
) {
220 linker_error(this->prog
, "geometry shader accesses element %i of "
221 "%s, but only %i input vertices\n",
222 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
223 return visit_continue
;
226 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
228 var
->data
.max_array_access
= this->num_vertices
- 1;
230 return visit_continue
;
233 /* Dereferences of input variables need to be updated so that their type
234 * matches the newly assigned type of the variable they are accessing. */
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 /* Dereferences of 2D input arrays need to be updated so that their type
242 * matches the newly assigned type of the array they are accessing. */
243 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
245 const glsl_type
*const vt
= ir
->array
->type
;
247 ir
->type
= vt
->fields
.array
;
248 return visit_continue
;
252 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
254 unsigned num_vertices
;
255 gl_shader_program
*prog
;
257 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
259 this->num_vertices
= num_vertices
;
263 virtual ~tess_eval_array_resize_visitor()
268 virtual ir_visitor_status
visit(ir_variable
*var
)
270 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
271 return visit_continue
;
273 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
275 var
->data
.max_array_access
= this->num_vertices
- 1;
277 return visit_continue
;
280 /* Dereferences of input variables need to be updated so that their type
281 * matches the newly assigned type of the variable they are accessing. */
282 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
284 ir
->type
= ir
->var
->type
;
285 return visit_continue
;
288 /* Dereferences of 2D input arrays need to be updated so that their type
289 * matches the newly assigned type of the array they are accessing. */
290 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
292 const glsl_type
*const vt
= ir
->array
->type
;
294 ir
->type
= vt
->fields
.array
;
295 return visit_continue
;
299 class barrier_use_visitor
: public ir_hierarchical_visitor
{
301 barrier_use_visitor(gl_shader_program
*prog
)
302 : prog(prog
), in_main(false), after_return(false), control_flow(0)
306 virtual ~barrier_use_visitor()
311 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
313 if (strcmp(ir
->name
, "main") == 0)
316 return visit_continue
;
319 virtual ir_visitor_status
visit_leave(ir_function
*)
322 after_return
= false;
323 return visit_continue
;
326 virtual ir_visitor_status
visit_leave(ir_return
*)
329 return visit_continue
;
332 virtual ir_visitor_status
visit_enter(ir_if
*)
335 return visit_continue
;
338 virtual ir_visitor_status
visit_leave(ir_if
*)
341 return visit_continue
;
344 virtual ir_visitor_status
visit_enter(ir_loop
*)
347 return visit_continue
;
350 virtual ir_visitor_status
visit_leave(ir_loop
*)
353 return visit_continue
;
356 /* FINISHME: `switch` is not expressed at the IR level -- it's already
357 * been lowered to a mess of `if`s. We'll correctly disallow any use of
358 * barrier() in a conditional path within the switch, but not in a path
359 * which is always hit.
362 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
364 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
365 /* Use of barrier(); determine if it is legal: */
367 linker_error(prog
, "Builtin barrier() may only be used in main");
372 linker_error(prog
, "Builtin barrier() may not be used after return");
376 if (control_flow
!= 0) {
377 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
381 return visit_continue
;
385 gl_shader_program
*prog
;
386 bool in_main
, after_return
;
391 * Visitor that determines the highest stream id to which a (geometry) shader
392 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
394 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
396 find_emit_vertex_visitor(int max_allowed
)
397 : max_stream_allowed(max_allowed
),
398 invalid_stream_id(0),
399 invalid_stream_id_from_emit_vertex(false),
400 end_primitive_found(false),
401 uses_non_zero_stream(false)
406 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
408 int stream_id
= ir
->stream_id();
411 invalid_stream_id
= stream_id
;
412 invalid_stream_id_from_emit_vertex
= true;
416 if (stream_id
> max_stream_allowed
) {
417 invalid_stream_id
= stream_id
;
418 invalid_stream_id_from_emit_vertex
= true;
423 uses_non_zero_stream
= true;
425 return visit_continue
;
428 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
430 end_primitive_found
= true;
432 int stream_id
= ir
->stream_id();
435 invalid_stream_id
= stream_id
;
436 invalid_stream_id_from_emit_vertex
= false;
440 if (stream_id
> max_stream_allowed
) {
441 invalid_stream_id
= stream_id
;
442 invalid_stream_id_from_emit_vertex
= false;
447 uses_non_zero_stream
= true;
449 return visit_continue
;
454 return invalid_stream_id
!= 0;
457 const char *error_func()
459 return invalid_stream_id_from_emit_vertex
?
460 "EmitStreamVertex" : "EndStreamPrimitive";
465 return invalid_stream_id
;
470 return uses_non_zero_stream
;
473 bool uses_end_primitive()
475 return end_primitive_found
;
479 int max_stream_allowed
;
480 int invalid_stream_id
;
481 bool invalid_stream_id_from_emit_vertex
;
482 bool end_primitive_found
;
483 bool uses_non_zero_stream
;
486 /* Class that finds array derefs and check if indexes are dynamic. */
487 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
490 dynamic_sampler_array_indexing_visitor() :
491 dynamic_sampler_array_indexing(false)
495 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
497 if (!ir
->variable_referenced())
498 return visit_continue
;
500 if (!ir
->variable_referenced()->type
->contains_sampler())
501 return visit_continue
;
503 if (!ir
->array_index
->constant_expression_value()) {
504 dynamic_sampler_array_indexing
= true;
507 return visit_continue
;
510 bool uses_dynamic_sampler_array_indexing()
512 return dynamic_sampler_array_indexing
;
516 bool dynamic_sampler_array_indexing
;
519 } /* anonymous namespace */
522 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
526 ralloc_strcat(&prog
->InfoLog
, "error: ");
528 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
531 prog
->LinkStatus
= false;
536 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
540 ralloc_strcat(&prog
->InfoLog
, "warning: ");
542 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
549 * Given a string identifying a program resource, break it into a base name
550 * and an optional array index in square brackets.
552 * If an array index is present, \c out_base_name_end is set to point to the
553 * "[" that precedes the array index, and the array index itself is returned
556 * If no array index is present (or if the array index is negative or
557 * mal-formed), \c out_base_name_end, is set to point to the null terminator
558 * at the end of the input string, and -1 is returned.
560 * Only the final array index is parsed; if the string contains other array
561 * indices (or structure field accesses), they are left in the base name.
563 * No attempt is made to check that the base name is properly formed;
564 * typically the caller will look up the base name in a hash table, so
565 * ill-formed base names simply turn into hash table lookup failures.
568 parse_program_resource_name(const GLchar
*name
,
569 const GLchar
**out_base_name_end
)
571 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
573 * "When an integer array element or block instance number is part of
574 * the name string, it will be specified in decimal form without a "+"
575 * or "-" sign or any extra leading zeroes. Additionally, the name
576 * string will not include white space anywhere in the string."
579 const size_t len
= strlen(name
);
580 *out_base_name_end
= name
+ len
;
582 if (len
== 0 || name
[len
-1] != ']')
585 /* Walk backwards over the string looking for a non-digit character. This
586 * had better be the opening bracket for an array index.
588 * Initially, i specifies the location of the ']'. Since the string may
589 * contain only the ']' charcater, walk backwards very carefully.
592 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
595 if ((i
== 0) || name
[i
-1] != '[')
598 long array_index
= strtol(&name
[i
], NULL
, 10);
602 /* Check for leading zero */
603 if (name
[i
] == '0' && name
[i
+1] != ']')
606 *out_base_name_end
= name
+ (i
- 1);
612 link_invalidate_variable_locations(exec_list
*ir
)
614 foreach_in_list(ir_instruction
, node
, ir
) {
615 ir_variable
*const var
= node
->as_variable();
620 /* Only assign locations for variables that lack an explicit location.
621 * Explicit locations are set for all built-in variables, generic vertex
622 * shader inputs (via layout(location=...)), and generic fragment shader
623 * outputs (also via layout(location=...)).
625 if (!var
->data
.explicit_location
) {
626 var
->data
.location
= -1;
627 var
->data
.location_frac
= 0;
630 /* ir_variable::is_unmatched_generic_inout is used by the linker while
631 * connecting outputs from one stage to inputs of the next stage.
633 if (var
->data
.explicit_location
&&
634 var
->data
.location
< VARYING_SLOT_VAR0
) {
635 var
->data
.is_unmatched_generic_inout
= 0;
637 var
->data
.is_unmatched_generic_inout
= 1;
644 * Set clip_distance_array_size based on the given shader.
646 * Also check for errors based on incorrect usage of gl_ClipVertex and
649 * Return false if an error was reported.
652 analyze_clip_usage(struct gl_shader_program
*prog
,
653 struct gl_shader
*shader
,
654 GLuint
*clip_distance_array_size
)
656 *clip_distance_array_size
= 0;
658 if (!prog
->IsES
&& prog
->Version
>= 130) {
659 /* From section 7.1 (Vertex Shader Special Variables) of the
662 * "It is an error for a shader to statically write both
663 * gl_ClipVertex and gl_ClipDistance."
665 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
666 * gl_ClipVertex nor gl_ClipDistance.
668 find_assignment_visitor
clip_vertex("gl_ClipVertex");
669 find_assignment_visitor
clip_distance("gl_ClipDistance");
671 clip_vertex
.run(shader
->ir
);
672 clip_distance
.run(shader
->ir
);
673 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
674 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
675 "and `gl_ClipDistance'\n",
676 _mesa_shader_stage_to_string(shader
->Stage
));
680 if (clip_distance
.variable_found()) {
681 ir_variable
*clip_distance_var
=
682 shader
->symbols
->get_variable("gl_ClipDistance");
684 assert(clip_distance_var
);
685 *clip_distance_array_size
= clip_distance_var
->type
->length
;
692 * Verify that a vertex shader executable meets all semantic requirements.
694 * Also sets prog->Vert.ClipDistanceArraySize as a side effect.
696 * \param shader Vertex shader executable to be verified
699 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
700 struct gl_shader
*shader
)
705 /* From the GLSL 1.10 spec, page 48:
707 * "The variable gl_Position is available only in the vertex
708 * language and is intended for writing the homogeneous vertex
709 * position. All executions of a well-formed vertex shader
710 * executable must write a value into this variable. [...] The
711 * variable gl_Position is available only in the vertex
712 * language and is intended for writing the homogeneous vertex
713 * position. All executions of a well-formed vertex shader
714 * executable must write a value into this variable."
716 * while in GLSL 1.40 this text is changed to:
718 * "The variable gl_Position is available only in the vertex
719 * language and is intended for writing the homogeneous vertex
720 * position. It can be written at any time during shader
721 * execution. It may also be read back by a vertex shader
722 * after being written. This value will be used by primitive
723 * assembly, clipping, culling, and other fixed functionality
724 * operations, if present, that operate on primitives after
725 * vertex processing has occurred. Its value is undefined if
726 * the vertex shader executable does not write gl_Position."
728 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
729 * gl_Position is not an error.
731 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
732 find_assignment_visitor
find("gl_Position");
733 find
.run(shader
->ir
);
734 if (!find
.variable_found()) {
737 "vertex shader does not write to `gl_Position'."
738 "It's value is undefined. \n");
741 "vertex shader does not write to `gl_Position'. \n");
747 analyze_clip_usage(prog
, shader
, &prog
->Vert
.ClipDistanceArraySize
);
751 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
752 struct gl_shader
*shader
)
757 analyze_clip_usage(prog
, shader
, &prog
->TessEval
.ClipDistanceArraySize
);
762 * Verify that a fragment shader executable meets all semantic requirements
764 * \param shader Fragment shader executable to be verified
767 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
768 struct gl_shader
*shader
)
773 find_assignment_visitor
frag_color("gl_FragColor");
774 find_assignment_visitor
frag_data("gl_FragData");
776 frag_color
.run(shader
->ir
);
777 frag_data
.run(shader
->ir
);
779 if (frag_color
.variable_found() && frag_data
.variable_found()) {
780 linker_error(prog
, "fragment shader writes to both "
781 "`gl_FragColor' and `gl_FragData'\n");
786 * Verify that a geometry shader executable meets all semantic requirements
788 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize as
791 * \param shader Geometry shader executable to be verified
794 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
795 struct gl_shader
*shader
)
800 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
801 prog
->Geom
.VerticesIn
= num_vertices
;
803 analyze_clip_usage(prog
, shader
, &prog
->Geom
.ClipDistanceArraySize
);
807 * Check if geometry shaders emit to non-zero streams and do corresponding
811 validate_geometry_shader_emissions(struct gl_context
*ctx
,
812 struct gl_shader_program
*prog
)
814 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
815 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
816 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
817 if (emit_vertex
.error()) {
818 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
819 "stream parameter are in the range [0, %d].\n",
820 emit_vertex
.error_func(),
821 emit_vertex
.error_stream(),
822 ctx
->Const
.MaxVertexStreams
- 1);
824 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
825 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
827 /* From the ARB_gpu_shader5 spec:
829 * "Multiple vertex streams are supported only if the output primitive
830 * type is declared to be "points". A program will fail to link if it
831 * contains a geometry shader calling EmitStreamVertex() or
832 * EndStreamPrimitive() if its output primitive type is not "points".
834 * However, in the same spec:
836 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
837 * with <stream> set to zero."
841 * "The function EndPrimitive() is equivalent to calling
842 * EndStreamPrimitive() with <stream> set to zero."
844 * Since we can call EmitVertex() and EndPrimitive() when we output
845 * primitives other than points, calling EmitStreamVertex(0) or
846 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
847 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
848 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
851 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
852 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
853 "with n>0 requires point output\n");
859 validate_intrastage_arrays(struct gl_shader_program
*prog
,
860 ir_variable
*const var
,
861 ir_variable
*const existing
)
863 /* Consider the types to be "the same" if both types are arrays
864 * of the same type and one of the arrays is implicitly sized.
865 * In addition, set the type of the linked variable to the
866 * explicitly sized array.
868 if (var
->type
->is_array() && existing
->type
->is_array()) {
869 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
870 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
871 if (var
->type
->length
!= 0) {
872 if (var
->type
->length
<= existing
->data
.max_array_access
) {
873 linker_error(prog
, "%s `%s' declared as type "
874 "`%s' but outermost dimension has an index"
877 var
->name
, var
->type
->name
,
878 existing
->data
.max_array_access
);
880 existing
->type
= var
->type
;
882 } else if (existing
->type
->length
!= 0) {
883 if(existing
->type
->length
<= var
->data
.max_array_access
&&
884 !existing
->data
.from_ssbo_unsized_array
) {
885 linker_error(prog
, "%s `%s' declared as type "
886 "`%s' but outermost dimension has an index"
889 var
->name
, existing
->type
->name
,
890 var
->data
.max_array_access
);
895 /* The arrays of structs could have different glsl_type pointers but
896 * they are actually the same type. Use record_compare() to check that.
898 if (existing
->type
->fields
.array
->is_record() &&
899 var
->type
->fields
.array
->is_record() &&
900 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
909 * Perform validation of global variables used across multiple shaders
912 cross_validate_globals(struct gl_shader_program
*prog
,
913 struct gl_shader
**shader_list
,
914 unsigned num_shaders
,
917 /* Examine all of the uniforms in all of the shaders and cross validate
920 glsl_symbol_table variables
;
921 for (unsigned i
= 0; i
< num_shaders
; i
++) {
922 if (shader_list
[i
] == NULL
)
925 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
926 ir_variable
*const var
= node
->as_variable();
931 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
934 /* don't cross validate subroutine uniforms */
935 if (var
->type
->contains_subroutine())
938 /* Don't cross validate temporaries that are at global scope. These
939 * will eventually get pulled into the shaders 'main'.
941 if (var
->data
.mode
== ir_var_temporary
)
944 /* If a global with this name has already been seen, verify that the
945 * new instance has the same type. In addition, if the globals have
946 * initializers, the values of the initializers must be the same.
948 ir_variable
*const existing
= variables
.get_variable(var
->name
);
949 if (existing
!= NULL
) {
950 /* Check if types match. Interface blocks have some special
951 * rules so we handle those elsewhere.
953 if (var
->type
!= existing
->type
&&
954 !var
->is_interface_instance()) {
955 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
956 if (var
->type
->is_record() && existing
->type
->is_record()
957 && existing
->type
->record_compare(var
->type
)) {
958 existing
->type
= var
->type
;
960 /* If it is an unsized array in a Shader Storage Block,
961 * two different shaders can access to different elements.
962 * Because of that, they might be converted to different
963 * sized arrays, then check that they are compatible but
964 * ignore the array size.
966 if (!(var
->data
.mode
== ir_var_shader_storage
&&
967 var
->data
.from_ssbo_unsized_array
&&
968 existing
->data
.mode
== ir_var_shader_storage
&&
969 existing
->data
.from_ssbo_unsized_array
&&
970 var
->type
->gl_type
== existing
->type
->gl_type
)) {
971 linker_error(prog
, "%s `%s' declared as type "
972 "`%s' and type `%s'\n",
974 var
->name
, var
->type
->name
,
975 existing
->type
->name
);
982 if (var
->data
.explicit_location
) {
983 if (existing
->data
.explicit_location
984 && (var
->data
.location
!= existing
->data
.location
)) {
985 linker_error(prog
, "explicit locations for %s "
986 "`%s' have differing values\n",
987 mode_string(var
), var
->name
);
991 existing
->data
.location
= var
->data
.location
;
992 existing
->data
.explicit_location
= true;
994 /* Check if uniform with implicit location was marked explicit
995 * by earlier shader stage. If so, mark it explicit in this stage
996 * too to make sure later processing does not treat it as
999 if (existing
->data
.explicit_location
) {
1000 var
->data
.location
= existing
->data
.location
;
1001 var
->data
.explicit_location
= true;
1005 /* From the GLSL 4.20 specification:
1006 * "A link error will result if two compilation units in a program
1007 * specify different integer-constant bindings for the same
1008 * opaque-uniform name. However, it is not an error to specify a
1009 * binding on some but not all declarations for the same name"
1011 if (var
->data
.explicit_binding
) {
1012 if (existing
->data
.explicit_binding
&&
1013 var
->data
.binding
!= existing
->data
.binding
) {
1014 linker_error(prog
, "explicit bindings for %s "
1015 "`%s' have differing values\n",
1016 mode_string(var
), var
->name
);
1020 existing
->data
.binding
= var
->data
.binding
;
1021 existing
->data
.explicit_binding
= true;
1024 if (var
->type
->contains_atomic() &&
1025 var
->data
.offset
!= existing
->data
.offset
) {
1026 linker_error(prog
, "offset specifications for %s "
1027 "`%s' have differing values\n",
1028 mode_string(var
), var
->name
);
1032 /* Validate layout qualifiers for gl_FragDepth.
1034 * From the AMD/ARB_conservative_depth specs:
1036 * "If gl_FragDepth is redeclared in any fragment shader in a
1037 * program, it must be redeclared in all fragment shaders in
1038 * that program that have static assignments to
1039 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1040 * fragment shaders in a single program must have the same set
1043 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1044 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1045 bool layout_differs
=
1046 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1048 if (layout_declared
&& layout_differs
) {
1050 "All redeclarations of gl_FragDepth in all "
1051 "fragment shaders in a single program must have "
1052 "the same set of qualifiers.\n");
1055 if (var
->data
.used
&& layout_differs
) {
1057 "If gl_FragDepth is redeclared with a layout "
1058 "qualifier in any fragment shader, it must be "
1059 "redeclared with the same layout qualifier in "
1060 "all fragment shaders that have assignments to "
1065 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1067 * "If a shared global has multiple initializers, the
1068 * initializers must all be constant expressions, and they
1069 * must all have the same value. Otherwise, a link error will
1070 * result. (A shared global having only one initializer does
1071 * not require that initializer to be a constant expression.)"
1073 * Previous to 4.20 the GLSL spec simply said that initializers
1074 * must have the same value. In this case of non-constant
1075 * initializers, this was impossible to determine. As a result,
1076 * no vendor actually implemented that behavior. The 4.20
1077 * behavior matches the implemented behavior of at least one other
1078 * vendor, so we'll implement that for all GLSL versions.
1080 if (var
->constant_initializer
!= NULL
) {
1081 if (existing
->constant_initializer
!= NULL
) {
1082 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1083 linker_error(prog
, "initializers for %s "
1084 "`%s' have differing values\n",
1085 mode_string(var
), var
->name
);
1089 /* If the first-seen instance of a particular uniform did not
1090 * have an initializer but a later instance does, copy the
1091 * initializer to the version stored in the symbol table.
1093 /* FINISHME: This is wrong. The constant_value field should
1094 * FINISHME: not be modified! Imagine a case where a shader
1095 * FINISHME: without an initializer is linked in two different
1096 * FINISHME: programs with shaders that have differing
1097 * FINISHME: initializers. Linking with the first will
1098 * FINISHME: modify the shader, and linking with the second
1099 * FINISHME: will fail.
1101 existing
->constant_initializer
=
1102 var
->constant_initializer
->clone(ralloc_parent(existing
),
1107 if (var
->data
.has_initializer
) {
1108 if (existing
->data
.has_initializer
1109 && (var
->constant_initializer
== NULL
1110 || existing
->constant_initializer
== NULL
)) {
1112 "shared global variable `%s' has multiple "
1113 "non-constant initializers.\n",
1118 /* Some instance had an initializer, so keep track of that. In
1119 * this location, all sorts of initializers (constant or
1120 * otherwise) will propagate the existence to the variable
1121 * stored in the symbol table.
1123 existing
->data
.has_initializer
= true;
1126 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1127 linker_error(prog
, "declarations for %s `%s' have "
1128 "mismatching invariant qualifiers\n",
1129 mode_string(var
), var
->name
);
1132 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1133 linker_error(prog
, "declarations for %s `%s' have "
1134 "mismatching centroid qualifiers\n",
1135 mode_string(var
), var
->name
);
1138 if (existing
->data
.sample
!= var
->data
.sample
) {
1139 linker_error(prog
, "declarations for %s `%s` have "
1140 "mismatching sample qualifiers\n",
1141 mode_string(var
), var
->name
);
1144 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1145 linker_error(prog
, "declarations for %s `%s` have "
1146 "mismatching image format qualifiers\n",
1147 mode_string(var
), var
->name
);
1151 variables
.add_variable(var
);
1158 * Perform validation of uniforms used across multiple shader stages
1161 cross_validate_uniforms(struct gl_shader_program
*prog
)
1163 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1164 MESA_SHADER_STAGES
, true);
1168 * Accumulates the array of prog->BufferInterfaceBlocks and checks that all
1169 * definitons of blocks agree on their contents.
1172 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1174 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1176 unsigned max_num_uniform_blocks
= 0;
1177 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1178 if (prog
->_LinkedShaders
[i
])
1179 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumBufferInterfaceBlocks
;
1182 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1183 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1185 InterfaceBlockStageIndex
[i
] = new int[max_num_uniform_blocks
];
1186 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1187 InterfaceBlockStageIndex
[i
][j
] = -1;
1192 for (unsigned int j
= 0; j
< sh
->NumBufferInterfaceBlocks
; j
++) {
1193 int index
= link_cross_validate_uniform_block(prog
,
1194 &prog
->BufferInterfaceBlocks
,
1195 &prog
->NumBufferInterfaceBlocks
,
1196 sh
->BufferInterfaceBlocks
[j
]);
1199 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1200 sh
->BufferInterfaceBlocks
[j
]->Name
);
1202 for (unsigned k
= 0; k
<= i
; k
++) {
1203 delete[] InterfaceBlockStageIndex
[k
];
1208 InterfaceBlockStageIndex
[i
][index
] = j
;
1212 /* Update per stage block pointers to point to the program list.
1213 * FIXME: We should be able to free the per stage blocks here.
1215 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1216 for (unsigned j
= 0; j
< prog
->NumBufferInterfaceBlocks
; j
++) {
1217 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1219 if (stage_index
!= -1) {
1220 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1222 prog
->BufferInterfaceBlocks
[j
].stageref
|= (1 << i
);
1224 sh
->BufferInterfaceBlocks
[stage_index
] =
1225 &prog
->BufferInterfaceBlocks
[j
];
1230 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1231 delete[] InterfaceBlockStageIndex
[i
];
1239 * Populates a shaders symbol table with all global declarations
1242 populate_symbol_table(gl_shader
*sh
)
1244 sh
->symbols
= new(sh
) glsl_symbol_table
;
1246 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1250 if ((func
= inst
->as_function()) != NULL
) {
1251 sh
->symbols
->add_function(func
);
1252 } else if ((var
= inst
->as_variable()) != NULL
) {
1253 if (var
->data
.mode
!= ir_var_temporary
)
1254 sh
->symbols
->add_variable(var
);
1261 * Remap variables referenced in an instruction tree
1263 * This is used when instruction trees are cloned from one shader and placed in
1264 * another. These trees will contain references to \c ir_variable nodes that
1265 * do not exist in the target shader. This function finds these \c ir_variable
1266 * references and replaces the references with matching variables in the target
1269 * If there is no matching variable in the target shader, a clone of the
1270 * \c ir_variable is made and added to the target shader. The new variable is
1271 * added to \b both the instruction stream and the symbol table.
1273 * \param inst IR tree that is to be processed.
1274 * \param symbols Symbol table containing global scope symbols in the
1276 * \param instructions Instruction stream where new variable declarations
1280 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1283 class remap_visitor
: public ir_hierarchical_visitor
{
1285 remap_visitor(struct gl_shader
*target
,
1288 this->target
= target
;
1289 this->symbols
= target
->symbols
;
1290 this->instructions
= target
->ir
;
1291 this->temps
= temps
;
1294 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1296 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1297 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1299 assert(var
!= NULL
);
1301 return visit_continue
;
1304 ir_variable
*const existing
=
1305 this->symbols
->get_variable(ir
->var
->name
);
1306 if (existing
!= NULL
)
1309 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1311 this->symbols
->add_variable(copy
);
1312 this->instructions
->push_head(copy
);
1316 return visit_continue
;
1320 struct gl_shader
*target
;
1321 glsl_symbol_table
*symbols
;
1322 exec_list
*instructions
;
1326 remap_visitor
v(target
, temps
);
1333 * Move non-declarations from one instruction stream to another
1335 * The intended usage pattern of this function is to pass the pointer to the
1336 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1337 * pointer) for \c last and \c false for \c make_copies on the first
1338 * call. Successive calls pass the return value of the previous call for
1339 * \c last and \c true for \c make_copies.
1341 * \param instructions Source instruction stream
1342 * \param last Instruction after which new instructions should be
1343 * inserted in the target instruction stream
1344 * \param make_copies Flag selecting whether instructions in \c instructions
1345 * should be copied (via \c ir_instruction::clone) into the
1346 * target list or moved.
1349 * The new "last" instruction in the target instruction stream. This pointer
1350 * is suitable for use as the \c last parameter of a later call to this
1354 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1355 bool make_copies
, gl_shader
*target
)
1357 hash_table
*temps
= NULL
;
1360 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1361 hash_table_pointer_compare
);
1363 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1364 if (inst
->as_function())
1367 ir_variable
*var
= inst
->as_variable();
1368 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1371 assert(inst
->as_assignment()
1373 || inst
->as_if() /* for initializers with the ?: operator */
1374 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1377 inst
= inst
->clone(target
, NULL
);
1380 hash_table_insert(temps
, inst
, var
);
1382 remap_variables(inst
, target
, temps
);
1387 last
->insert_after(inst
);
1392 hash_table_dtor(temps
);
1399 * This class is only used in link_intrastage_shaders() below but declaring
1400 * it inside that function leads to compiler warnings with some versions of
1403 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1405 array_sizing_visitor()
1406 : mem_ctx(ralloc_context(NULL
)),
1407 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1408 hash_table_pointer_compare
))
1412 ~array_sizing_visitor()
1414 hash_table_dtor(this->unnamed_interfaces
);
1415 ralloc_free(this->mem_ctx
);
1418 virtual ir_visitor_status
visit(ir_variable
*var
)
1420 const glsl_type
*type_without_array
;
1421 fixup_type(&var
->type
, var
->data
.max_array_access
,
1422 var
->data
.from_ssbo_unsized_array
);
1423 type_without_array
= var
->type
->without_array();
1424 if (var
->type
->is_interface()) {
1425 if (interface_contains_unsized_arrays(var
->type
)) {
1426 const glsl_type
*new_type
=
1427 resize_interface_members(var
->type
,
1428 var
->get_max_ifc_array_access(),
1429 var
->is_in_shader_storage_block());
1430 var
->type
= new_type
;
1431 var
->change_interface_type(new_type
);
1433 } else if (type_without_array
->is_interface()) {
1434 if (interface_contains_unsized_arrays(type_without_array
)) {
1435 const glsl_type
*new_type
=
1436 resize_interface_members(type_without_array
,
1437 var
->get_max_ifc_array_access(),
1438 var
->is_in_shader_storage_block());
1439 var
->change_interface_type(new_type
);
1440 var
->type
= update_interface_members_array(var
->type
, new_type
);
1442 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1443 /* Store a pointer to the variable in the unnamed_interfaces
1446 ir_variable
**interface_vars
= (ir_variable
**)
1447 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1448 if (interface_vars
== NULL
) {
1449 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1451 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1454 unsigned index
= ifc_type
->field_index(var
->name
);
1455 assert(index
< ifc_type
->length
);
1456 assert(interface_vars
[index
] == NULL
);
1457 interface_vars
[index
] = var
;
1459 return visit_continue
;
1463 * For each unnamed interface block that was discovered while running the
1464 * visitor, adjust the interface type to reflect the newly assigned array
1465 * sizes, and fix up the ir_variable nodes to point to the new interface
1468 void fixup_unnamed_interface_types()
1470 hash_table_call_foreach(this->unnamed_interfaces
,
1471 fixup_unnamed_interface_type
, NULL
);
1476 * If the type pointed to by \c type represents an unsized array, replace
1477 * it with a sized array whose size is determined by max_array_access.
1479 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1480 bool from_ssbo_unsized_array
)
1482 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1483 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1484 max_array_access
+ 1);
1485 assert(*type
!= NULL
);
1489 static const glsl_type
*
1490 update_interface_members_array(const glsl_type
*type
,
1491 const glsl_type
*new_interface_type
)
1493 const glsl_type
*element_type
= type
->fields
.array
;
1494 if (element_type
->is_array()) {
1495 const glsl_type
*new_array_type
=
1496 update_interface_members_array(element_type
, new_interface_type
);
1497 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1499 return glsl_type::get_array_instance(new_interface_type
,
1505 * Determine whether the given interface type contains unsized arrays (if
1506 * it doesn't, array_sizing_visitor doesn't need to process it).
1508 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1510 for (unsigned i
= 0; i
< type
->length
; i
++) {
1511 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1512 if (elem_type
->is_unsized_array())
1519 * Create a new interface type based on the given type, with unsized arrays
1520 * replaced by sized arrays whose size is determined by
1521 * max_ifc_array_access.
1523 static const glsl_type
*
1524 resize_interface_members(const glsl_type
*type
,
1525 const unsigned *max_ifc_array_access
,
1528 unsigned num_fields
= type
->length
;
1529 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1530 memcpy(fields
, type
->fields
.structure
,
1531 num_fields
* sizeof(*fields
));
1532 for (unsigned i
= 0; i
< num_fields
; i
++) {
1533 /* If SSBO last member is unsized array, we don't replace it by a sized
1536 if (is_ssbo
&& i
== (num_fields
- 1))
1537 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1540 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1543 glsl_interface_packing packing
=
1544 (glsl_interface_packing
) type
->interface_packing
;
1545 const glsl_type
*new_ifc_type
=
1546 glsl_type::get_interface_instance(fields
, num_fields
,
1547 packing
, type
->name
);
1549 return new_ifc_type
;
1552 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1555 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1556 ir_variable
**interface_vars
= (ir_variable
**) data
;
1557 unsigned num_fields
= ifc_type
->length
;
1558 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1559 memcpy(fields
, ifc_type
->fields
.structure
,
1560 num_fields
* sizeof(*fields
));
1561 bool interface_type_changed
= false;
1562 for (unsigned i
= 0; i
< num_fields
; i
++) {
1563 if (interface_vars
[i
] != NULL
&&
1564 fields
[i
].type
!= interface_vars
[i
]->type
) {
1565 fields
[i
].type
= interface_vars
[i
]->type
;
1566 interface_type_changed
= true;
1569 if (!interface_type_changed
) {
1573 glsl_interface_packing packing
=
1574 (glsl_interface_packing
) ifc_type
->interface_packing
;
1575 const glsl_type
*new_ifc_type
=
1576 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1579 for (unsigned i
= 0; i
< num_fields
; i
++) {
1580 if (interface_vars
[i
] != NULL
)
1581 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1586 * Memory context used to allocate the data in \c unnamed_interfaces.
1591 * Hash table from const glsl_type * to an array of ir_variable *'s
1592 * pointing to the ir_variables constituting each unnamed interface block.
1594 hash_table
*unnamed_interfaces
;
1598 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1602 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1603 struct gl_shader_program
*prog
,
1604 struct gl_shader
*linked_shader
,
1605 struct gl_shader
**shader_list
,
1606 unsigned num_shaders
)
1608 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1609 linked_shader
->TransformFeedback
.BufferStride
[i
] = 0;
1612 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1613 struct gl_shader
*shader
= shader_list
[i
];
1615 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1616 if (shader
->TransformFeedback
.BufferStride
[j
]) {
1617 if (linked_shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1618 shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1619 linked_shader
->TransformFeedback
.BufferStride
[j
] !=
1620 shader
->TransformFeedback
.BufferStride
[j
]) {
1622 "intrastage shaders defined with conflicting "
1623 "xfb_stride for buffer %d (%d and %d)\n", j
,
1624 linked_shader
->TransformFeedback
.BufferStride
[j
],
1625 shader
->TransformFeedback
.BufferStride
[j
]);
1629 if (shader
->TransformFeedback
.BufferStride
[j
])
1630 linked_shader
->TransformFeedback
.BufferStride
[j
] =
1631 shader
->TransformFeedback
.BufferStride
[j
];
1636 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1637 if (linked_shader
->TransformFeedback
.BufferStride
[j
]) {
1638 prog
->TransformFeedback
.BufferStride
[j
] =
1639 linked_shader
->TransformFeedback
.BufferStride
[j
];
1641 /* We will validate doubles at a later stage */
1642 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1643 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1644 "multiple of 4 or if its applied to a type that is "
1645 "or contains a double a multiple of 8.",
1646 prog
->TransformFeedback
.BufferStride
[j
]);
1650 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1651 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1653 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1654 "limit has been exceeded.");
1662 * Performs the cross-validation of tessellation control shader vertices and
1663 * layout qualifiers for the attached tessellation control shaders,
1664 * and propagates them to the linked TCS and linked shader program.
1667 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1668 struct gl_shader
*linked_shader
,
1669 struct gl_shader
**shader_list
,
1670 unsigned num_shaders
)
1672 linked_shader
->TessCtrl
.VerticesOut
= 0;
1674 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1677 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1679 * "All tessellation control shader layout declarations in a program
1680 * must specify the same output patch vertex count. There must be at
1681 * least one layout qualifier specifying an output patch vertex count
1682 * in any program containing tessellation control shaders; however,
1683 * such a declaration is not required in all tessellation control
1687 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1688 struct gl_shader
*shader
= shader_list
[i
];
1690 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1691 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1692 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1693 linker_error(prog
, "tessellation control shader defined with "
1694 "conflicting output vertex count (%d and %d)\n",
1695 linked_shader
->TessCtrl
.VerticesOut
,
1696 shader
->TessCtrl
.VerticesOut
);
1699 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1703 /* Just do the intrastage -> interstage propagation right now,
1704 * since we already know we're in the right type of shader program
1707 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1708 linker_error(prog
, "tessellation control shader didn't declare "
1709 "vertices out layout qualifier\n");
1712 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1717 * Performs the cross-validation of tessellation evaluation shader
1718 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1719 * for the attached tessellation evaluation shaders, and propagates them
1720 * to the linked TES and linked shader program.
1723 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1724 struct gl_shader
*linked_shader
,
1725 struct gl_shader
**shader_list
,
1726 unsigned num_shaders
)
1728 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1729 linked_shader
->TessEval
.Spacing
= 0;
1730 linked_shader
->TessEval
.VertexOrder
= 0;
1731 linked_shader
->TessEval
.PointMode
= -1;
1733 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1736 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1738 * "At least one tessellation evaluation shader (compilation unit) in
1739 * a program must declare a primitive mode in its input layout.
1740 * Declaration vertex spacing, ordering, and point mode identifiers is
1741 * optional. It is not required that all tessellation evaluation
1742 * shaders in a program declare a primitive mode. If spacing or
1743 * vertex ordering declarations are omitted, the tessellation
1744 * primitive generator will use equal spacing or counter-clockwise
1745 * vertex ordering, respectively. If a point mode declaration is
1746 * omitted, the tessellation primitive generator will produce lines or
1747 * triangles according to the primitive mode."
1750 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1751 struct gl_shader
*shader
= shader_list
[i
];
1753 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1754 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1755 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1756 linker_error(prog
, "tessellation evaluation shader defined with "
1757 "conflicting input primitive modes.\n");
1760 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1763 if (shader
->TessEval
.Spacing
!= 0) {
1764 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1765 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1766 linker_error(prog
, "tessellation evaluation shader defined with "
1767 "conflicting vertex spacing.\n");
1770 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1773 if (shader
->TessEval
.VertexOrder
!= 0) {
1774 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1775 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1776 linker_error(prog
, "tessellation evaluation shader defined with "
1777 "conflicting ordering.\n");
1780 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1783 if (shader
->TessEval
.PointMode
!= -1) {
1784 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1785 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1786 linker_error(prog
, "tessellation evaluation shader defined with "
1787 "conflicting point modes.\n");
1790 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1795 /* Just do the intrastage -> interstage propagation right now,
1796 * since we already know we're in the right type of shader program
1799 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1801 "tessellation evaluation shader didn't declare input "
1802 "primitive modes.\n");
1805 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1807 if (linked_shader
->TessEval
.Spacing
== 0)
1808 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1809 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1811 if (linked_shader
->TessEval
.VertexOrder
== 0)
1812 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1813 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1815 if (linked_shader
->TessEval
.PointMode
== -1)
1816 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1817 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1822 * Performs the cross-validation of layout qualifiers specified in
1823 * redeclaration of gl_FragCoord for the attached fragment shaders,
1824 * and propagates them to the linked FS and linked shader program.
1827 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1828 struct gl_shader
*linked_shader
,
1829 struct gl_shader
**shader_list
,
1830 unsigned num_shaders
)
1832 linked_shader
->redeclares_gl_fragcoord
= false;
1833 linked_shader
->uses_gl_fragcoord
= false;
1834 linked_shader
->origin_upper_left
= false;
1835 linked_shader
->pixel_center_integer
= false;
1837 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1838 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1841 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1842 struct gl_shader
*shader
= shader_list
[i
];
1843 /* From the GLSL 1.50 spec, page 39:
1845 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1846 * it must be redeclared in all the fragment shaders in that program
1847 * that have a static use gl_FragCoord."
1849 if ((linked_shader
->redeclares_gl_fragcoord
1850 && !shader
->redeclares_gl_fragcoord
1851 && shader
->uses_gl_fragcoord
)
1852 || (shader
->redeclares_gl_fragcoord
1853 && !linked_shader
->redeclares_gl_fragcoord
1854 && linked_shader
->uses_gl_fragcoord
)) {
1855 linker_error(prog
, "fragment shader defined with conflicting "
1856 "layout qualifiers for gl_FragCoord\n");
1859 /* From the GLSL 1.50 spec, page 39:
1861 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1862 * single program must have the same set of qualifiers."
1864 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1865 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1866 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1867 linker_error(prog
, "fragment shader defined with conflicting "
1868 "layout qualifiers for gl_FragCoord\n");
1871 /* Update the linked shader state. Note that uses_gl_fragcoord should
1872 * accumulate the results. The other values should replace. If there
1873 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1874 * are already known to be the same.
1876 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1877 linked_shader
->redeclares_gl_fragcoord
=
1878 shader
->redeclares_gl_fragcoord
;
1879 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1880 || shader
->uses_gl_fragcoord
;
1881 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1882 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1885 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1890 * Performs the cross-validation of geometry shader max_vertices and
1891 * primitive type layout qualifiers for the attached geometry shaders,
1892 * and propagates them to the linked GS and linked shader program.
1895 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1896 struct gl_shader
*linked_shader
,
1897 struct gl_shader
**shader_list
,
1898 unsigned num_shaders
)
1900 linked_shader
->Geom
.VerticesOut
= 0;
1901 linked_shader
->Geom
.Invocations
= 0;
1902 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1903 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1905 /* No in/out qualifiers defined for anything but GLSL 1.50+
1906 * geometry shaders so far.
1908 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1911 /* From the GLSL 1.50 spec, page 46:
1913 * "All geometry shader output layout declarations in a program
1914 * must declare the same layout and same value for
1915 * max_vertices. There must be at least one geometry output
1916 * layout declaration somewhere in a program, but not all
1917 * geometry shaders (compilation units) are required to
1921 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1922 struct gl_shader
*shader
= shader_list
[i
];
1924 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1925 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1926 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1927 linker_error(prog
, "geometry shader defined with conflicting "
1931 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1934 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1935 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1936 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1937 linker_error(prog
, "geometry shader defined with conflicting "
1941 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1944 if (shader
->Geom
.VerticesOut
!= 0) {
1945 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1946 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1947 linker_error(prog
, "geometry shader defined with conflicting "
1948 "output vertex count (%d and %d)\n",
1949 linked_shader
->Geom
.VerticesOut
,
1950 shader
->Geom
.VerticesOut
);
1953 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1956 if (shader
->Geom
.Invocations
!= 0) {
1957 if (linked_shader
->Geom
.Invocations
!= 0 &&
1958 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1959 linker_error(prog
, "geometry shader defined with conflicting "
1960 "invocation count (%d and %d)\n",
1961 linked_shader
->Geom
.Invocations
,
1962 shader
->Geom
.Invocations
);
1965 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1969 /* Just do the intrastage -> interstage propagation right now,
1970 * since we already know we're in the right type of shader program
1973 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1975 "geometry shader didn't declare primitive input type\n");
1978 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1980 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1982 "geometry shader didn't declare primitive output type\n");
1985 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1987 if (linked_shader
->Geom
.VerticesOut
== 0) {
1989 "geometry shader didn't declare max_vertices\n");
1992 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1994 if (linked_shader
->Geom
.Invocations
== 0)
1995 linked_shader
->Geom
.Invocations
= 1;
1997 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
2002 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2003 * qualifiers for the attached compute shaders, and propagate them to the
2004 * linked CS and linked shader program.
2007 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2008 struct gl_shader
*linked_shader
,
2009 struct gl_shader
**shader_list
,
2010 unsigned num_shaders
)
2012 for (int i
= 0; i
< 3; i
++)
2013 linked_shader
->Comp
.LocalSize
[i
] = 0;
2015 /* This function is called for all shader stages, but it only has an effect
2016 * for compute shaders.
2018 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2021 /* From the ARB_compute_shader spec, in the section describing local size
2024 * If multiple compute shaders attached to a single program object
2025 * declare local work-group size, the declarations must be identical;
2026 * otherwise a link-time error results. Furthermore, if a program
2027 * object contains any compute shaders, at least one must contain an
2028 * input layout qualifier specifying the local work sizes of the
2029 * program, or a link-time error will occur.
2031 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2032 struct gl_shader
*shader
= shader_list
[sh
];
2034 if (shader
->Comp
.LocalSize
[0] != 0) {
2035 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
2036 for (int i
= 0; i
< 3; i
++) {
2037 if (linked_shader
->Comp
.LocalSize
[i
] !=
2038 shader
->Comp
.LocalSize
[i
]) {
2039 linker_error(prog
, "compute shader defined with conflicting "
2045 for (int i
= 0; i
< 3; i
++)
2046 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
2050 /* Just do the intrastage -> interstage propagation right now,
2051 * since we already know we're in the right type of shader program
2054 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
2055 linker_error(prog
, "compute shader didn't declare local size\n");
2058 for (int i
= 0; i
< 3; i
++)
2059 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
2064 * Combine a group of shaders for a single stage to generate a linked shader
2067 * If this function is supplied a single shader, it is cloned, and the new
2068 * shader is returned.
2070 static struct gl_shader
*
2071 link_intrastage_shaders(void *mem_ctx
,
2072 struct gl_context
*ctx
,
2073 struct gl_shader_program
*prog
,
2074 struct gl_shader
**shader_list
,
2075 unsigned num_shaders
)
2077 struct gl_uniform_block
*uniform_blocks
= NULL
;
2079 /* Check that global variables defined in multiple shaders are consistent.
2081 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
2082 if (!prog
->LinkStatus
)
2085 /* Check that interface blocks defined in multiple shaders are consistent.
2087 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2089 if (!prog
->LinkStatus
)
2092 /* Link up uniform blocks defined within this stage. */
2093 const unsigned num_uniform_blocks
=
2094 link_uniform_blocks(mem_ctx
, ctx
, prog
, shader_list
, num_shaders
,
2096 if (!prog
->LinkStatus
)
2099 /* Check that there is only a single definition of each function signature
2100 * across all shaders.
2102 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2103 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2104 ir_function
*const f
= node
->as_function();
2109 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2110 ir_function
*const other
=
2111 shader_list
[j
]->symbols
->get_function(f
->name
);
2113 /* If the other shader has no function (and therefore no function
2114 * signatures) with the same name, skip to the next shader.
2119 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2120 if (!sig
->is_defined
|| sig
->is_builtin())
2123 ir_function_signature
*other_sig
=
2124 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2126 if ((other_sig
!= NULL
) && other_sig
->is_defined
2127 && !other_sig
->is_builtin()) {
2128 linker_error(prog
, "function `%s' is multiply defined\n",
2137 /* Find the shader that defines main, and make a clone of it.
2139 * Starting with the clone, search for undefined references. If one is
2140 * found, find the shader that defines it. Clone the reference and add
2141 * it to the shader. Repeat until there are no undefined references or
2142 * until a reference cannot be resolved.
2144 gl_shader
*main
= NULL
;
2145 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2146 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2147 main
= shader_list
[i
];
2153 linker_error(prog
, "%s shader lacks `main'\n",
2154 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2158 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2159 linked
->ir
= new(linked
) exec_list
;
2160 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2162 linked
->BufferInterfaceBlocks
=
2163 ralloc_array(linked
, gl_uniform_block
*, num_uniform_blocks
);
2165 ralloc_steal(linked
, uniform_blocks
);
2166 for (unsigned i
= 0; i
< num_uniform_blocks
; i
++) {
2167 linked
->BufferInterfaceBlocks
[i
] = &uniform_blocks
[i
];
2170 linked
->NumBufferInterfaceBlocks
= num_uniform_blocks
;
2172 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2173 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2174 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2175 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2176 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2177 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2180 populate_symbol_table(linked
);
2182 /* The pointer to the main function in the final linked shader (i.e., the
2183 * copy of the original shader that contained the main function).
2185 ir_function_signature
*const main_sig
=
2186 _mesa_get_main_function_signature(linked
);
2188 /* Move any instructions other than variable declarations or function
2189 * declarations into main.
2191 exec_node
*insertion_point
=
2192 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2195 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2196 if (shader_list
[i
] == main
)
2199 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2200 insertion_point
, true, linked
);
2203 /* Check if any shader needs built-in functions. */
2204 bool need_builtins
= false;
2205 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2206 if (shader_list
[i
]->uses_builtin_functions
) {
2207 need_builtins
= true;
2213 if (need_builtins
) {
2214 /* Make a temporary array one larger than shader_list, which will hold
2215 * the built-in function shader as well.
2217 gl_shader
**linking_shaders
= (gl_shader
**)
2218 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2220 ok
= linking_shaders
!= NULL
;
2223 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2224 _mesa_glsl_initialize_builtin_functions();
2225 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2227 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2229 free(linking_shaders
);
2231 _mesa_error_no_memory(__func__
);
2234 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2239 _mesa_delete_shader(ctx
, linked
);
2243 /* At this point linked should contain all of the linked IR, so
2244 * validate it to make sure nothing went wrong.
2246 validate_ir_tree(linked
->ir
);
2248 /* Set the size of geometry shader input arrays */
2249 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2250 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2251 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2252 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2253 ir
->accept(&input_resize_visitor
);
2257 if (ctx
->Const
.VertexID_is_zero_based
)
2258 lower_vertex_id(linked
);
2260 /* Validate correct usage of barrier() in the tess control shader */
2261 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2262 barrier_use_visitor
visitor(prog
);
2263 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2264 ir
->accept(&visitor
);
2268 /* Make a pass over all variable declarations to ensure that arrays with
2269 * unspecified sizes have a size specified. The size is inferred from the
2270 * max_array_access field.
2272 array_sizing_visitor v
;
2274 v
.fixup_unnamed_interface_types();
2280 * Update the sizes of linked shader uniform arrays to the maximum
2283 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2285 * If one or more elements of an array are active,
2286 * GetActiveUniform will return the name of the array in name,
2287 * subject to the restrictions listed above. The type of the array
2288 * is returned in type. The size parameter contains the highest
2289 * array element index used, plus one. The compiler or linker
2290 * determines the highest index used. There will be only one
2291 * active uniform reported by the GL per uniform array.
2295 update_array_sizes(struct gl_shader_program
*prog
)
2297 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2298 if (prog
->_LinkedShaders
[i
] == NULL
)
2301 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2302 ir_variable
*const var
= node
->as_variable();
2304 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2305 !var
->type
->is_array())
2308 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2309 * will not be eliminated. Since we always do std140, just
2310 * don't resize arrays in UBOs.
2312 * Atomic counters are supposed to get deterministic
2313 * locations assigned based on the declaration ordering and
2314 * sizes, array compaction would mess that up.
2316 * Subroutine uniforms are not removed.
2318 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2319 var
->type
->contains_subroutine())
2322 unsigned int size
= var
->data
.max_array_access
;
2323 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2324 if (prog
->_LinkedShaders
[j
] == NULL
)
2327 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2328 ir_variable
*other_var
= node2
->as_variable();
2332 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2333 other_var
->data
.max_array_access
> size
) {
2334 size
= other_var
->data
.max_array_access
;
2339 if (size
+ 1 != var
->type
->length
) {
2340 /* If this is a built-in uniform (i.e., it's backed by some
2341 * fixed-function state), adjust the number of state slots to
2342 * match the new array size. The number of slots per array entry
2343 * is not known. It seems safe to assume that the total number of
2344 * slots is an integer multiple of the number of array elements.
2345 * Determine the number of slots per array element by dividing by
2346 * the old (total) size.
2348 const unsigned num_slots
= var
->get_num_state_slots();
2349 if (num_slots
> 0) {
2350 var
->set_num_state_slots((size
+ 1)
2351 * (num_slots
/ var
->type
->length
));
2354 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2356 /* FINISHME: We should update the types of array
2357 * dereferences of this variable now.
2365 * Resize tessellation evaluation per-vertex inputs to the size of
2366 * tessellation control per-vertex outputs.
2369 resize_tes_inputs(struct gl_context
*ctx
,
2370 struct gl_shader_program
*prog
)
2372 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2375 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2376 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2378 /* If no control shader is present, then the TES inputs are statically
2379 * sized to MaxPatchVertices; the actual size of the arrays won't be
2380 * known until draw time.
2382 const int num_vertices
= tcs
2383 ? tcs
->TessCtrl
.VerticesOut
2384 : ctx
->Const
.MaxPatchVertices
;
2386 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2387 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2388 ir
->accept(&input_resize_visitor
);
2392 /* Convert the gl_PatchVerticesIn system value into a constant, since
2393 * the value is known at this point.
2395 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2396 ir_variable
*var
= ir
->as_variable();
2397 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2398 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2399 void *mem_ctx
= ralloc_parent(var
);
2400 var
->data
.mode
= ir_var_auto
;
2401 var
->data
.location
= 0;
2402 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2409 * Find a contiguous set of available bits in a bitmask.
2411 * \param used_mask Bits representing used (1) and unused (0) locations
2412 * \param needed_count Number of contiguous bits needed.
2415 * Base location of the available bits on success or -1 on failure.
2418 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2420 unsigned needed_mask
= (1 << needed_count
) - 1;
2421 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2423 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2424 * cannot optimize possibly infinite loops" for the loop below.
2426 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2429 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2430 if ((needed_mask
& ~used_mask
) == needed_mask
)
2441 * Assign locations for either VS inputs or FS outputs
2443 * \param prog Shader program whose variables need locations assigned
2444 * \param constants Driver specific constant values for the program.
2445 * \param target_index Selector for the program target to receive location
2446 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2447 * \c MESA_SHADER_FRAGMENT.
2450 * If locations are successfully assigned, true is returned. Otherwise an
2451 * error is emitted to the shader link log and false is returned.
2454 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2455 struct gl_constants
*constants
,
2456 unsigned target_index
)
2458 /* Maximum number of generic locations. This corresponds to either the
2459 * maximum number of draw buffers or the maximum number of generic
2462 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2463 constants
->Program
[target_index
].MaxAttribs
:
2464 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2466 /* Mark invalid locations as being used.
2468 unsigned used_locations
= (max_index
>= 32)
2469 ? ~0 : ~((1 << max_index
) - 1);
2470 unsigned double_storage_locations
= 0;
2472 assert((target_index
== MESA_SHADER_VERTEX
)
2473 || (target_index
== MESA_SHADER_FRAGMENT
));
2475 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2479 /* Operate in a total of four passes.
2481 * 1. Invalidate the location assignments for all vertex shader inputs.
2483 * 2. Assign locations for inputs that have user-defined (via
2484 * glBindVertexAttribLocation) locations and outputs that have
2485 * user-defined locations (via glBindFragDataLocation).
2487 * 3. Sort the attributes without assigned locations by number of slots
2488 * required in decreasing order. Fragmentation caused by attribute
2489 * locations assigned by the application may prevent large attributes
2490 * from having enough contiguous space.
2492 * 4. Assign locations to any inputs without assigned locations.
2495 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2496 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2498 const enum ir_variable_mode direction
=
2499 (target_index
== MESA_SHADER_VERTEX
)
2500 ? ir_var_shader_in
: ir_var_shader_out
;
2503 /* Temporary storage for the set of attributes that need locations assigned.
2509 /* Used below in the call to qsort. */
2510 static int compare(const void *a
, const void *b
)
2512 const temp_attr
*const l
= (const temp_attr
*) a
;
2513 const temp_attr
*const r
= (const temp_attr
*) b
;
2515 /* Reversed because we want a descending order sort below. */
2516 return r
->slots
- l
->slots
;
2519 assert(max_index
<= 32);
2521 unsigned num_attr
= 0;
2523 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2524 ir_variable
*const var
= node
->as_variable();
2526 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2529 if (var
->data
.explicit_location
) {
2530 var
->data
.is_unmatched_generic_inout
= 0;
2531 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2532 || (var
->data
.location
< 0)) {
2534 "invalid explicit location %d specified for `%s'\n",
2535 (var
->data
.location
< 0)
2536 ? var
->data
.location
2537 : var
->data
.location
- generic_base
,
2541 } else if (target_index
== MESA_SHADER_VERTEX
) {
2544 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2545 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2546 var
->data
.location
= binding
;
2547 var
->data
.is_unmatched_generic_inout
= 0;
2549 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2553 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2554 assert(binding
>= FRAG_RESULT_DATA0
);
2555 var
->data
.location
= binding
;
2556 var
->data
.is_unmatched_generic_inout
= 0;
2558 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2559 var
->data
.index
= index
;
2564 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2566 * "Output binding assignments will cause LinkProgram to fail:
2568 * If the program has an active output assigned to a location greater
2569 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2570 * an active output assigned an index greater than or equal to one;"
2572 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2573 var
->data
.location
- generic_base
>=
2574 (int) constants
->MaxDualSourceDrawBuffers
) {
2576 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2577 "with index %u for %s\n",
2578 var
->data
.location
- generic_base
, var
->data
.index
,
2583 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
? true : false);
2585 /* If the variable is not a built-in and has a location statically
2586 * assigned in the shader (presumably via a layout qualifier), make sure
2587 * that it doesn't collide with other assigned locations. Otherwise,
2588 * add it to the list of variables that need linker-assigned locations.
2590 if (var
->data
.location
!= -1) {
2591 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2592 /* From page 61 of the OpenGL 4.0 spec:
2594 * "LinkProgram will fail if the attribute bindings assigned
2595 * by BindAttribLocation do not leave not enough space to
2596 * assign a location for an active matrix attribute or an
2597 * active attribute array, both of which require multiple
2598 * contiguous generic attributes."
2600 * I think above text prohibits the aliasing of explicit and
2601 * automatic assignments. But, aliasing is allowed in manual
2602 * assignments of attribute locations. See below comments for
2605 * From OpenGL 4.0 spec, page 61:
2607 * "It is possible for an application to bind more than one
2608 * attribute name to the same location. This is referred to as
2609 * aliasing. This will only work if only one of the aliased
2610 * attributes is active in the executable program, or if no
2611 * path through the shader consumes more than one attribute of
2612 * a set of attributes aliased to the same location. A link
2613 * error can occur if the linker determines that every path
2614 * through the shader consumes multiple aliased attributes,
2615 * but implementations are not required to generate an error
2618 * From GLSL 4.30 spec, page 54:
2620 * "A program will fail to link if any two non-vertex shader
2621 * input variables are assigned to the same location. For
2622 * vertex shaders, multiple input variables may be assigned
2623 * to the same location using either layout qualifiers or via
2624 * the OpenGL API. However, such aliasing is intended only to
2625 * support vertex shaders where each execution path accesses
2626 * at most one input per each location. Implementations are
2627 * permitted, but not required, to generate link-time errors
2628 * if they detect that every path through the vertex shader
2629 * executable accesses multiple inputs assigned to any single
2630 * location. For all shader types, a program will fail to link
2631 * if explicit location assignments leave the linker unable
2632 * to find space for other variables without explicit
2635 * From OpenGL ES 3.0 spec, page 56:
2637 * "Binding more than one attribute name to the same location
2638 * is referred to as aliasing, and is not permitted in OpenGL
2639 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2640 * fail when this condition exists. However, aliasing is
2641 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2642 * This will only work if only one of the aliased attributes
2643 * is active in the executable program, or if no path through
2644 * the shader consumes more than one attribute of a set of
2645 * attributes aliased to the same location. A link error can
2646 * occur if the linker determines that every path through the
2647 * shader consumes multiple aliased attributes, but implemen-
2648 * tations are not required to generate an error in this case."
2650 * After looking at above references from OpenGL, OpenGL ES and
2651 * GLSL specifications, we allow aliasing of vertex input variables
2652 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2654 * NOTE: This is not required by the spec but its worth mentioning
2655 * here that we're not doing anything to make sure that no path
2656 * through the vertex shader executable accesses multiple inputs
2657 * assigned to any single location.
2660 /* Mask representing the contiguous slots that will be used by
2663 const unsigned attr
= var
->data
.location
- generic_base
;
2664 const unsigned use_mask
= (1 << slots
) - 1;
2665 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2666 ? "vertex shader input" : "fragment shader output";
2668 /* Generate a link error if the requested locations for this
2669 * attribute exceed the maximum allowed attribute location.
2671 if (attr
+ slots
> max_index
) {
2673 "insufficient contiguous locations "
2674 "available for %s `%s' %d %d %d\n", string
,
2675 var
->name
, used_locations
, use_mask
, attr
);
2679 /* Generate a link error if the set of bits requested for this
2680 * attribute overlaps any previously allocated bits.
2682 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2683 if (target_index
== MESA_SHADER_FRAGMENT
||
2684 (prog
->IsES
&& prog
->Version
>= 300)) {
2686 "overlapping location is assigned "
2687 "to %s `%s' %d %d %d\n", string
,
2688 var
->name
, used_locations
, use_mask
, attr
);
2691 linker_warning(prog
,
2692 "overlapping location is assigned "
2693 "to %s `%s' %d %d %d\n", string
,
2694 var
->name
, used_locations
, use_mask
, attr
);
2698 used_locations
|= (use_mask
<< attr
);
2700 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2702 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2703 * active attribute variables may fail to link, unless
2704 * device-dependent optimizations are able to make the program
2705 * fit within available hardware resources. For the purposes
2706 * of this test, attribute variables of the type dvec3, dvec4,
2707 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2708 * count as consuming twice as many attributes as equivalent
2709 * single-precision types. While these types use the same number
2710 * of generic attributes as their single-precision equivalents,
2711 * implementations are permitted to consume two single-precision
2712 * vectors of internal storage for each three- or four-component
2713 * double-precision vector."
2715 * Mark this attribute slot as taking up twice as much space
2716 * so we can count it properly against limits. According to
2717 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2718 * is optional behavior, but it seems preferable.
2720 if (var
->type
->without_array()->is_dual_slot_double())
2721 double_storage_locations
|= (use_mask
<< attr
);
2727 if (num_attr
>= max_index
) {
2728 linker_error(prog
, "too many %s (max %u)",
2729 target_index
== MESA_SHADER_VERTEX
?
2730 "vertex shader inputs" : "fragment shader outputs",
2734 to_assign
[num_attr
].slots
= slots
;
2735 to_assign
[num_attr
].var
= var
;
2739 if (target_index
== MESA_SHADER_VERTEX
) {
2740 unsigned total_attribs_size
=
2741 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2742 _mesa_bitcount(double_storage_locations
);
2743 if (total_attribs_size
> max_index
) {
2745 "attempt to use %d vertex attribute slots only %d available ",
2746 total_attribs_size
, max_index
);
2751 /* If all of the attributes were assigned locations by the application (or
2752 * are built-in attributes with fixed locations), return early. This should
2753 * be the common case.
2758 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2760 if (target_index
== MESA_SHADER_VERTEX
) {
2761 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2762 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2763 * reserved to prevent it from being automatically allocated below.
2765 find_deref_visitor
find("gl_Vertex");
2767 if (find
.variable_found())
2768 used_locations
|= (1 << 0);
2771 for (unsigned i
= 0; i
< num_attr
; i
++) {
2772 /* Mask representing the contiguous slots that will be used by this
2775 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2777 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2780 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2781 ? "vertex shader input" : "fragment shader output";
2784 "insufficient contiguous locations "
2785 "available for %s `%s'\n",
2786 string
, to_assign
[i
].var
->name
);
2790 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2791 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2792 used_locations
|= (use_mask
<< location
);
2799 * Match explicit locations of outputs to inputs and deactivate the
2800 * unmatch flag if found so we don't optimise them away.
2803 match_explicit_outputs_to_inputs(struct gl_shader_program
*prog
,
2804 gl_shader
*producer
,
2805 gl_shader
*consumer
)
2807 glsl_symbol_table parameters
;
2808 ir_variable
*explicit_locations
[MAX_VARYING
] = { NULL
};
2810 /* Find all shader outputs in the "producer" stage.
2812 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2813 ir_variable
*const var
= node
->as_variable();
2815 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2818 if (var
->data
.explicit_location
&&
2819 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2820 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2821 if (explicit_locations
[idx
] == NULL
)
2822 explicit_locations
[idx
] = var
;
2826 /* Match inputs to outputs */
2827 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2828 ir_variable
*const input
= node
->as_variable();
2830 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2833 ir_variable
*output
= NULL
;
2834 if (input
->data
.explicit_location
2835 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2836 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
];
2838 if (output
!= NULL
){
2839 input
->data
.is_unmatched_generic_inout
= 0;
2840 output
->data
.is_unmatched_generic_inout
= 0;
2847 * Store the gl_FragDepth layout in the gl_shader_program struct.
2850 store_fragdepth_layout(struct gl_shader_program
*prog
)
2852 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2856 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2858 /* We don't look up the gl_FragDepth symbol directly because if
2859 * gl_FragDepth is not used in the shader, it's removed from the IR.
2860 * However, the symbol won't be removed from the symbol table.
2862 * We're only interested in the cases where the variable is NOT removed
2865 foreach_in_list(ir_instruction
, node
, ir
) {
2866 ir_variable
*const var
= node
->as_variable();
2868 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2872 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2873 switch (var
->data
.depth_layout
) {
2874 case ir_depth_layout_none
:
2875 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2877 case ir_depth_layout_any
:
2878 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2880 case ir_depth_layout_greater
:
2881 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2883 case ir_depth_layout_less
:
2884 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2886 case ir_depth_layout_unchanged
:
2887 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2898 * Validate the resources used by a program versus the implementation limits
2901 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2903 unsigned total_uniform_blocks
= 0;
2904 unsigned total_shader_storage_blocks
= 0;
2906 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2907 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2912 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2913 linker_error(prog
, "Too many %s shader texture samplers\n",
2914 _mesa_shader_stage_to_string(i
));
2917 if (sh
->num_uniform_components
>
2918 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2919 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2920 linker_warning(prog
, "Too many %s shader default uniform block "
2921 "components, but the driver will try to optimize "
2922 "them out; this is non-portable out-of-spec "
2924 _mesa_shader_stage_to_string(i
));
2926 linker_error(prog
, "Too many %s shader default uniform block "
2928 _mesa_shader_stage_to_string(i
));
2932 if (sh
->num_combined_uniform_components
>
2933 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2934 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2935 linker_warning(prog
, "Too many %s shader uniform components, "
2936 "but the driver will try to optimize them out; "
2937 "this is non-portable out-of-spec behavior\n",
2938 _mesa_shader_stage_to_string(i
));
2940 linker_error(prog
, "Too many %s shader uniform components\n",
2941 _mesa_shader_stage_to_string(i
));
2945 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
2946 total_uniform_blocks
+= sh
->NumUniformBlocks
;
2948 const unsigned max_uniform_blocks
=
2949 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2950 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
2951 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2952 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
2953 max_uniform_blocks
);
2956 const unsigned max_shader_storage_blocks
=
2957 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
2958 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
2959 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
2960 _mesa_shader_stage_to_string(i
),
2961 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
2965 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2966 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2967 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
2970 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
2971 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
2972 total_shader_storage_blocks
,
2973 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
2976 for (unsigned i
= 0; i
< prog
->NumBufferInterfaceBlocks
; i
++) {
2977 /* Don't check SSBOs for Uniform Block Size */
2978 if (!prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2979 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2980 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2981 prog
->BufferInterfaceBlocks
[i
].Name
,
2982 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2983 ctx
->Const
.MaxUniformBlockSize
);
2986 if (prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2987 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxShaderStorageBlockSize
) {
2988 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
2989 prog
->BufferInterfaceBlocks
[i
].Name
,
2990 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2991 ctx
->Const
.MaxShaderStorageBlockSize
);
2997 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
2999 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3000 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3005 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3006 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3012 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3013 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3014 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3015 if (fn
->types
[k
] == uni
->type
) {
3021 uni
->num_compatible_subroutines
= count
;
3027 check_subroutine_resources(struct gl_shader_program
*prog
)
3029 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3030 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3033 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3034 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3035 _mesa_shader_stage_to_string(i
));
3040 * Validate shader image resources.
3043 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3045 unsigned total_image_units
= 0;
3046 unsigned fragment_outputs
= 0;
3047 unsigned total_shader_storage_blocks
= 0;
3049 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3052 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3053 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3056 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3057 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3058 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3059 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3061 total_image_units
+= sh
->NumImages
;
3062 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3064 if (i
== MESA_SHADER_FRAGMENT
) {
3065 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3066 ir_variable
*var
= node
->as_variable();
3067 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3068 /* since there are no double fs outputs - pass false */
3069 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3075 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3076 linker_error(prog
, "Too many combined image uniforms\n");
3078 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3079 ctx
->Const
.MaxCombinedShaderOutputResources
)
3080 linker_error(prog
, "Too many combined image uniforms, shader storage "
3081 " buffers and fragment outputs\n");
3086 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3087 * for a variable, checks for overlaps between other uniforms using explicit
3091 reserve_explicit_locations(struct gl_shader_program
*prog
,
3092 string_to_uint_map
*map
, ir_variable
*var
)
3094 unsigned slots
= var
->type
->uniform_locations();
3095 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3096 unsigned return_value
= slots
;
3098 /* Resize remap table if locations do not fit in the current one. */
3099 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3100 prog
->UniformRemapTable
=
3101 reralloc(prog
, prog
->UniformRemapTable
,
3102 gl_uniform_storage
*,
3105 if (!prog
->UniformRemapTable
) {
3106 linker_error(prog
, "Out of memory during linking.\n");
3110 /* Initialize allocated space. */
3111 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3112 prog
->UniformRemapTable
[i
] = NULL
;
3114 prog
->NumUniformRemapTable
= max_loc
+ 1;
3117 for (unsigned i
= 0; i
< slots
; i
++) {
3118 unsigned loc
= var
->data
.location
+ i
;
3120 /* Check if location is already used. */
3121 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3123 /* Possibly same uniform from a different stage, this is ok. */
3125 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3130 /* ARB_explicit_uniform_location specification states:
3132 * "No two default-block uniform variables in the program can have
3133 * the same location, even if they are unused, otherwise a compiler
3134 * or linker error will be generated."
3137 "location qualifier for uniform %s overlaps "
3138 "previously used location\n",
3143 /* Initialize location as inactive before optimization
3144 * rounds and location assignment.
3146 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3149 /* Note, base location used for arrays. */
3150 map
->put(var
->data
.location
, var
->name
);
3152 return return_value
;
3156 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3157 struct gl_shader
*sh
,
3160 unsigned slots
= var
->type
->uniform_locations();
3161 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3163 /* Resize remap table if locations do not fit in the current one. */
3164 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3165 sh
->SubroutineUniformRemapTable
=
3166 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3167 gl_uniform_storage
*,
3170 if (!sh
->SubroutineUniformRemapTable
) {
3171 linker_error(prog
, "Out of memory during linking.\n");
3175 /* Initialize allocated space. */
3176 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3177 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3179 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3182 for (unsigned i
= 0; i
< slots
; i
++) {
3183 unsigned loc
= var
->data
.location
+ i
;
3185 /* Check if location is already used. */
3186 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3188 /* ARB_explicit_uniform_location specification states:
3189 * "No two subroutine uniform variables can have the same location
3190 * in the same shader stage, otherwise a compiler or linker error
3191 * will be generated."
3194 "location qualifier for uniform %s overlaps "
3195 "previously used location\n",
3200 /* Initialize location as inactive before optimization
3201 * rounds and location assignment.
3203 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3209 * Check and reserve all explicit uniform locations, called before
3210 * any optimizations happen to handle also inactive uniforms and
3211 * inactive array elements that may get trimmed away.
3214 check_explicit_uniform_locations(struct gl_context
*ctx
,
3215 struct gl_shader_program
*prog
)
3217 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3220 /* This map is used to detect if overlapping explicit locations
3221 * occur with the same uniform (from different stage) or a different one.
3223 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3226 linker_error(prog
, "Out of memory during linking.\n");
3230 unsigned entries_total
= 0;
3231 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3232 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3237 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3238 ir_variable
*var
= node
->as_variable();
3239 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3242 if (var
->data
.explicit_location
) {
3244 if (var
->type
->without_array()->is_subroutine())
3245 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3247 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3251 entries_total
+= slots
;
3262 struct empty_uniform_block
*current_block
= NULL
;
3264 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3265 /* We found empty space in UniformRemapTable. */
3266 if (prog
->UniformRemapTable
[i
] == NULL
) {
3267 /* We've found the beginning of a new continous block of empty slots */
3268 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3269 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3270 current_block
->start
= i
;
3271 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3272 ¤t_block
->link
);
3275 /* The current block continues, so we simply increment its slots */
3276 current_block
->slots
++;
3281 return entries_total
;
3285 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3286 GLenum type
, const char *name
)
3288 bool found_interface
= false;
3289 unsigned block_name_len
= 0;
3290 const char *block_name_dot
= strchr(name
, '.');
3292 /* These rules only apply to buffer variables. So we return
3293 * true for the rest of types.
3295 if (type
!= GL_BUFFER_VARIABLE
)
3298 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3299 const char *block_name
= shProg
->BufferInterfaceBlocks
[i
].Name
;
3300 block_name_len
= strlen(block_name
);
3302 const char *block_square_bracket
= strchr(block_name
, '[');
3303 if (block_square_bracket
) {
3304 /* The block is part of an array of named interfaces,
3305 * for the name comparison we ignore the "[x]" part.
3307 block_name_len
-= strlen(block_square_bracket
);
3310 if (block_name_dot
) {
3311 /* Check if the variable name starts with the interface
3312 * name. The interface name (if present) should have the
3313 * length than the interface block name we are comparing to.
3315 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3316 if (len
!= block_name_len
)
3320 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3321 found_interface
= true;
3326 /* We remove the interface name from the buffer variable name,
3327 * including the dot that follows it.
3329 if (found_interface
)
3330 name
= name
+ block_name_len
+ 1;
3332 /* From: ARB_program_interface_query extension:
3334 * "For an active shader storage block member declared as an array, an
3335 * entry will be generated only for the first array element, regardless
3336 * of its type. For arrays of aggregate types, the enumeration rules are
3337 * applied recursively for the single enumerated array element.
3339 const char *struct_first_dot
= strchr(name
, '.');
3340 const char *first_square_bracket
= strchr(name
, '[');
3342 /* The buffer variable is on top level and it is not an array */
3343 if (!first_square_bracket
) {
3345 /* The shader storage block member is a struct, then generate the entry */
3346 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3349 /* Shader storage block member is an array, only generate an entry for the
3350 * first array element.
3352 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3360 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3361 const void *data
, uint8_t stages
)
3365 /* If resource already exists, do not add it again. */
3366 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3367 if (prog
->ProgramResourceList
[i
].Data
== data
)
3370 prog
->ProgramResourceList
=
3372 prog
->ProgramResourceList
,
3373 gl_program_resource
,
3374 prog
->NumProgramResourceList
+ 1);
3376 if (!prog
->ProgramResourceList
) {
3377 linker_error(prog
, "Out of memory during linking.\n");
3381 struct gl_program_resource
*res
=
3382 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3386 res
->StageReferences
= stages
;
3388 prog
->NumProgramResourceList
++;
3393 /* Function checks if a variable var is a packed varying and
3394 * if given name is part of packed varying's list.
3396 * If a variable is a packed varying, it has a name like
3397 * 'packed:a,b,c' where a, b and c are separate variables.
3400 included_in_packed_varying(ir_variable
*var
, const char *name
)
3402 if (strncmp(var
->name
, "packed:", 7) != 0)
3405 char *list
= strdup(var
->name
+ 7);
3410 char *token
= strtok_r(list
, ",", &saveptr
);
3412 if (strcmp(token
, name
) == 0) {
3416 token
= strtok_r(NULL
, ",", &saveptr
);
3423 * Function builds a stage reference bitmask from variable name.
3426 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3431 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3432 * used for reference mask in gl_program_resource will need to be changed.
3434 assert(MESA_SHADER_STAGES
< 8);
3436 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3437 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3441 /* Shader symbol table may contain variables that have
3442 * been optimized away. Search IR for the variable instead.
3444 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3445 ir_variable
*var
= node
->as_variable();
3447 unsigned baselen
= strlen(var
->name
);
3449 if (included_in_packed_varying(var
, name
)) {
3454 /* Type needs to match if specified, otherwise we might
3455 * pick a variable with same name but different interface.
3457 if (var
->data
.mode
!= mode
)
3460 if (strncmp(var
->name
, name
, baselen
) == 0) {
3461 /* Check for exact name matches but also check for arrays and
3464 if (name
[baselen
] == '\0' ||
3465 name
[baselen
] == '[' ||
3466 name
[baselen
] == '.') {
3478 * Create gl_shader_variable from ir_variable class.
3480 static gl_shader_variable
*
3481 create_shader_variable(struct gl_shader_program
*shProg
,
3482 const ir_variable
*in
, bool use_implicit_location
,
3485 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3489 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3490 * expect to see gl_VertexID in the program resource list. Pretend.
3492 if (in
->data
.mode
== ir_var_system_value
&&
3493 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3494 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3496 out
->name
= ralloc_strdup(shProg
, in
->name
);
3502 /* From the ARB_program_interface_query specification:
3504 * "Not all active variables are assigned valid locations; the
3505 * following variables will have an effective location of -1:
3507 * * uniforms declared as atomic counters;
3509 * * members of a uniform block;
3511 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3513 * * inputs or outputs not declared with a "location" layout qualifier,
3514 * except for vertex shader inputs and fragment shader outputs."
3516 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3517 is_gl_identifier(in
->name
) ||
3518 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3521 out
->location
= in
->data
.location
- location_bias
;
3524 out
->type
= in
->type
;
3525 out
->index
= in
->data
.index
;
3526 out
->patch
= in
->data
.patch
;
3527 out
->mode
= in
->data
.mode
;
3533 add_interface_variables(struct gl_shader_program
*shProg
,
3534 unsigned stage
, GLenum programInterface
)
3536 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3538 foreach_in_list(ir_instruction
, node
, ir
) {
3539 ir_variable
*var
= node
->as_variable();
3541 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3546 switch (var
->data
.mode
) {
3547 case ir_var_system_value
:
3548 case ir_var_shader_in
:
3549 if (programInterface
!= GL_PROGRAM_INPUT
)
3551 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3552 : int(VARYING_SLOT_VAR0
);
3554 case ir_var_shader_out
:
3555 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3557 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3558 : int(VARYING_SLOT_VAR0
);
3564 /* Skip packed varyings, packed varyings are handled separately
3565 * by add_packed_varyings.
3567 if (strncmp(var
->name
, "packed:", 7) == 0)
3570 /* Skip fragdata arrays, these are handled separately
3571 * by add_fragdata_arrays.
3573 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3576 const bool vs_input_or_fs_output
=
3577 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3578 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3580 gl_shader_variable
*sha_v
=
3581 create_shader_variable(shProg
, var
, vs_input_or_fs_output
, loc_bias
);
3585 if (!add_program_resource(shProg
, programInterface
, sha_v
, 1 << stage
))
3592 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
, GLenum type
)
3594 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3597 if (!sh
|| !sh
->packed_varyings
)
3600 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3601 ir_variable
*var
= node
->as_variable();
3603 switch (var
->data
.mode
) {
3604 case ir_var_shader_in
:
3605 iface
= GL_PROGRAM_INPUT
;
3607 case ir_var_shader_out
:
3608 iface
= GL_PROGRAM_OUTPUT
;
3611 unreachable("unexpected type");
3614 if (type
== iface
) {
3615 gl_shader_variable
*sha_v
=
3616 create_shader_variable(shProg
, var
, false, VARYING_SLOT_VAR0
);
3619 if (!add_program_resource(shProg
, iface
, sha_v
,
3620 build_stageref(shProg
, sha_v
->name
,
3630 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3632 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3634 if (!sh
|| !sh
->fragdata_arrays
)
3637 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3638 ir_variable
*var
= node
->as_variable();
3640 assert(var
->data
.mode
== ir_var_shader_out
);
3641 gl_shader_variable
*sha_v
=
3642 create_shader_variable(shProg
, var
, true, FRAG_RESULT_DATA0
);
3645 if (!add_program_resource(shProg
, GL_PROGRAM_OUTPUT
, sha_v
,
3646 1 << MESA_SHADER_FRAGMENT
))
3654 get_top_level_name(const char *name
)
3656 const char *first_dot
= strchr(name
, '.');
3657 const char *first_square_bracket
= strchr(name
, '[');
3659 /* From ARB_program_interface_query spec:
3661 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying the
3662 * number of active array elements of the top-level shader storage block
3663 * member containing to the active variable is written to <params>. If the
3664 * top-level block member is not declared as an array, the value one is
3665 * written to <params>. If the top-level block member is an array with no
3666 * declared size, the value zero is written to <params>.
3669 /* The buffer variable is on top level.*/
3670 if (!first_square_bracket
&& !first_dot
)
3671 name_size
= strlen(name
);
3672 else if ((!first_square_bracket
||
3673 (first_dot
&& first_dot
< first_square_bracket
)))
3674 name_size
= first_dot
- name
;
3676 name_size
= first_square_bracket
- name
;
3678 return strndup(name
, name_size
);
3682 get_var_name(const char *name
)
3684 const char *first_dot
= strchr(name
, '.');
3687 return strdup(name
);
3689 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3693 is_top_level_shader_storage_block_member(const char* name
,
3694 const char* interface_name
,
3695 const char* field_name
)
3697 bool result
= false;
3699 /* If the given variable is already a top-level shader storage
3700 * block member, then return array_size = 1.
3701 * We could have two possibilities: if we have an instanced
3702 * shader storage block or not instanced.
3704 * For the first, we check create a name as it was in top level and
3705 * compare it with the real name. If they are the same, then
3706 * the variable is already at top-level.
3708 * Full instanced name is: interface name + '.' + var name +
3711 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3712 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3713 if (!full_instanced_name
) {
3714 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3718 snprintf(full_instanced_name
, name_length
, "%s.%s",
3719 interface_name
, field_name
);
3721 /* Check if its top-level shader storage block member of an
3722 * instanced interface block, or of a unnamed interface block.
3724 if (strcmp(name
, full_instanced_name
) == 0 ||
3725 strcmp(name
, field_name
) == 0)
3728 free(full_instanced_name
);
3733 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3734 char *interface_name
, char *var_name
)
3736 /* From GL_ARB_program_interface_query spec:
3738 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer
3739 * identifying the number of active array elements of the top-level
3740 * shader storage block member containing to the active variable is
3741 * written to <params>. If the top-level block member is not
3742 * declared as an array, the value one is written to <params>. If
3743 * the top-level block member is an array with no declared size,
3744 * the value zero is written to <params>.
3746 if (is_top_level_shader_storage_block_member(uni
->name
,
3750 else if (field
->type
->is_unsized_array())
3752 else if (field
->type
->is_array())
3753 return field
->type
->length
;
3759 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3760 const glsl_struct_field
*field
, char *interface_name
,
3763 /* From GL_ARB_program_interface_query:
3765 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3766 * identifying the stride between array elements of the top-level
3767 * shader storage block member containing the active variable is
3768 * written to <params>. For top-level block members declared as
3769 * arrays, the value written is the difference, in basic machine
3770 * units, between the offsets of the active variable for
3771 * consecutive elements in the top-level array. For top-level
3772 * block members not declared as an array, zero is written to
3775 if (field
->type
->is_array()) {
3776 const enum glsl_matrix_layout matrix_layout
=
3777 glsl_matrix_layout(field
->matrix_layout
);
3778 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3779 const glsl_type
*array_type
= field
->type
->fields
.array
;
3781 if (is_top_level_shader_storage_block_member(uni
->name
,
3786 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3787 if (array_type
->is_record() || array_type
->is_array())
3788 return glsl_align(array_type
->std140_size(row_major
), 16);
3790 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3792 return array_type
->std430_array_stride(row_major
);
3799 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3800 struct gl_uniform_storage
*uni
)
3802 int block_index
= uni
->block_index
;
3803 int array_size
= -1;
3804 int array_stride
= -1;
3805 char *var_name
= get_top_level_name(uni
->name
);
3806 char *interface_name
=
3807 get_top_level_name(uni
->is_shader_storage
?
3808 shProg
->ShaderStorageBlocks
[block_index
]->Name
:
3809 shProg
->UniformBlocks
[block_index
]->Name
);
3811 if (strcmp(var_name
, interface_name
) == 0) {
3812 /* Deal with instanced array of SSBOs */
3813 char *temp_name
= get_var_name(uni
->name
);
3815 linker_error(shProg
, "Out of memory during linking.\n");
3816 goto write_top_level_array_size_and_stride
;
3819 var_name
= get_top_level_name(temp_name
);
3822 linker_error(shProg
, "Out of memory during linking.\n");
3823 goto write_top_level_array_size_and_stride
;
3827 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
3828 if (shProg
->Shaders
[i
] == NULL
)
3831 const gl_shader
*stage
= shProg
->Shaders
[i
];
3832 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
3833 ir_variable
*var
= node
->as_variable();
3834 if (!var
|| !var
->get_interface_type() ||
3835 var
->data
.mode
!= ir_var_shader_storage
)
3838 const glsl_type
*interface
= var
->get_interface_type();
3840 if (strcmp(interface_name
, interface
->name
) != 0)
3843 for (unsigned i
= 0; i
< interface
->length
; i
++) {
3844 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
3845 if (strcmp(field
->name
, var_name
) != 0)
3848 array_stride
= get_array_stride(uni
, interface
, field
,
3849 interface_name
, var_name
);
3850 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
3851 goto write_top_level_array_size_and_stride
;
3855 write_top_level_array_size_and_stride
:
3856 free(interface_name
);
3858 uni
->top_level_array_stride
= array_stride
;
3859 uni
->top_level_array_size
= array_size
;
3863 * Builds up a list of program resources that point to existing
3867 build_program_resource_list(struct gl_context
*ctx
,
3868 struct gl_shader_program
*shProg
)
3870 /* Rebuild resource list. */
3871 if (shProg
->ProgramResourceList
) {
3872 ralloc_free(shProg
->ProgramResourceList
);
3873 shProg
->ProgramResourceList
= NULL
;
3874 shProg
->NumProgramResourceList
= 0;
3877 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
3879 /* Determine first input and final output stage. These are used to
3880 * detect which variables should be enumerated in the resource list
3881 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
3883 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3884 if (!shProg
->_LinkedShaders
[i
])
3886 if (input_stage
== MESA_SHADER_STAGES
)
3891 /* Empty shader, no resources. */
3892 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
3895 /* Program interface needs to expose varyings in case of SSO. */
3896 if (shProg
->SeparateShader
) {
3897 if (!add_packed_varyings(shProg
, input_stage
, GL_PROGRAM_INPUT
))
3900 if (!add_packed_varyings(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
3904 if (!add_fragdata_arrays(shProg
))
3907 /* Add inputs and outputs to the resource list. */
3908 if (!add_interface_variables(shProg
, input_stage
, GL_PROGRAM_INPUT
))
3911 if (!add_interface_variables(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
3914 /* Add transform feedback varyings. */
3915 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
3916 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
3917 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
3918 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
3924 /* Add transform feedback buffers. */
3925 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
3926 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
3927 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
3928 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_BUFFER
,
3929 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
3935 /* Add uniforms from uniform storage. */
3936 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3937 /* Do not add uniforms internally used by Mesa. */
3938 if (shProg
->UniformStorage
[i
].hidden
)
3942 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
3945 /* Add stagereferences for uniforms in a uniform block. */
3946 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
3947 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3948 if (block_index
!= -1) {
3949 stageref
|= is_shader_storage
?
3950 shProg
->ShaderStorageBlocks
[block_index
]->stageref
:
3951 shProg
->UniformBlocks
[block_index
]->stageref
;
3954 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
3955 if (!should_add_buffer_variable(shProg
, type
,
3956 shProg
->UniformStorage
[i
].name
))
3959 if (is_shader_storage
) {
3960 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
3963 if (!add_program_resource(shProg
, type
,
3964 &shProg
->UniformStorage
[i
], stageref
))
3968 /* Add program uniform blocks and shader storage blocks. */
3969 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3970 bool is_shader_storage
= shProg
->BufferInterfaceBlocks
[i
].IsShaderStorage
;
3971 GLenum type
= is_shader_storage
? GL_SHADER_STORAGE_BLOCK
: GL_UNIFORM_BLOCK
;
3972 if (!add_program_resource(shProg
, type
,
3973 &shProg
->BufferInterfaceBlocks
[i
], 0))
3977 /* Add atomic counter buffers. */
3978 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3979 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3980 &shProg
->AtomicBuffers
[i
], 0))
3984 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3986 if (!shProg
->UniformStorage
[i
].hidden
)
3989 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
3990 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
3991 !shProg
->UniformStorage
[i
].type
->is_subroutine())
3994 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
3995 /* add shader subroutines */
3996 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
4001 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4002 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
4008 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4009 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4010 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
4017 * This check is done to make sure we allow only constant expression
4018 * indexing and "constant-index-expression" (indexing with an expression
4019 * that includes loop induction variable).
4022 validate_sampler_array_indexing(struct gl_context
*ctx
,
4023 struct gl_shader_program
*prog
)
4025 dynamic_sampler_array_indexing_visitor v
;
4026 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4027 if (prog
->_LinkedShaders
[i
] == NULL
)
4030 bool no_dynamic_indexing
=
4031 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4033 /* Search for array derefs in shader. */
4034 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4035 if (v
.uses_dynamic_sampler_array_indexing()) {
4036 const char *msg
= "sampler arrays indexed with non-constant "
4037 "expressions is forbidden in GLSL %s %u";
4038 /* Backend has indicated that it has no dynamic indexing support. */
4039 if (no_dynamic_indexing
) {
4040 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4043 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4051 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4053 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4054 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4059 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4060 ir_function
*fn
= node
->as_function();
4064 if (fn
->is_subroutine
)
4065 sh
->NumSubroutineUniformTypes
++;
4067 if (!fn
->num_subroutine_types
)
4070 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4071 struct gl_subroutine_function
,
4072 sh
->NumSubroutineFunctions
+ 1);
4073 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4074 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4075 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4076 ralloc_array(sh
, const struct glsl_type
*,
4077 fn
->num_subroutine_types
);
4079 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4082 * "Each subroutine with an index qualifier in the shader must be
4083 * given a unique index, otherwise a compile or link error will be
4086 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4087 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4088 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4089 linker_error(prog
, "each subroutine index qualifier in the "
4090 "shader must be unique\n");
4094 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4095 fn
->subroutine_index
;
4097 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4098 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4099 sh
->NumSubroutineFunctions
++;
4102 /* Assign index for subroutines without an explicit index*/
4104 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4105 while (sh
->SubroutineFunctions
[j
].index
== -1) {
4106 for (unsigned k
= 0; k
< sh
->NumSubroutineFunctions
; k
++) {
4107 if (sh
->SubroutineFunctions
[k
].index
== index
)
4109 else if (k
== sh
->NumSubroutineFunctions
- 1)
4110 sh
->SubroutineFunctions
[j
].index
= index
;
4119 split_ubos_and_ssbos(void *mem_ctx
,
4120 struct gl_uniform_block
**s_blks
,
4121 struct gl_uniform_block
*p_blks
,
4122 unsigned num_blocks
,
4123 struct gl_uniform_block
***ubos
,
4125 struct gl_uniform_block
***ssbos
,
4126 unsigned *num_ssbos
)
4128 unsigned num_ubo_blocks
= 0;
4129 unsigned num_ssbo_blocks
= 0;
4131 /* Are we spliting the list of blocks for the shader or the program */
4132 bool is_shader
= p_blks
== NULL
;
4134 for (unsigned i
= 0; i
< num_blocks
; i
++) {
4135 if (is_shader
? s_blks
[i
]->IsShaderStorage
: p_blks
[i
].IsShaderStorage
)
4141 *ubos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ubo_blocks
);
4144 *ssbos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ssbo_blocks
);
4147 for (unsigned i
= 0; i
< num_blocks
; i
++) {
4148 struct gl_uniform_block
*blk
= is_shader
? s_blks
[i
] : &p_blks
[i
];
4149 if (blk
->IsShaderStorage
) {
4150 (*ssbos
)[*num_ssbos
] = blk
;
4153 (*ubos
)[*num_ubos
] = blk
;
4158 assert(*num_ubos
+ *num_ssbos
== num_blocks
);
4162 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4164 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4166 foreach_in_list(ir_instruction
, node
, ir
) {
4167 ir_variable
*const var
= node
->as_variable();
4169 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4172 /* Don't set always active on builtins that haven't been redeclared */
4173 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4176 var
->data
.always_active_io
= true;
4181 * When separate shader programs are enabled, only input/outputs between
4182 * the stages of a multi-stage separate program can be safely removed
4183 * from the shader interface. Other inputs/outputs must remain active.
4186 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4188 unsigned first
, last
;
4189 assert(prog
->SeparateShader
);
4191 first
= MESA_SHADER_STAGES
;
4194 /* Determine first and last stage. Excluding the compute stage */
4195 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4196 if (!prog
->_LinkedShaders
[i
])
4198 if (first
== MESA_SHADER_STAGES
)
4203 if (first
== MESA_SHADER_STAGES
)
4206 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4207 gl_shader
*sh
= prog
->_LinkedShaders
[stage
];
4211 if (first
== last
) {
4212 /* For a single shader program only allow inputs to the vertex shader
4213 * and outputs from the fragment shader to be removed.
4215 if (stage
!= MESA_SHADER_VERTEX
)
4216 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4217 if (stage
!= MESA_SHADER_FRAGMENT
)
4218 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4220 /* For multi-stage separate shader programs only allow inputs and
4221 * outputs between the shader stages to be removed as well as inputs
4222 * to the vertex shader and outputs from the fragment shader.
4224 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4225 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4226 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4227 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4233 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4235 prog
->LinkStatus
= true; /* All error paths will set this to false */
4236 prog
->Validated
= false;
4237 prog
->_Used
= false;
4239 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4241 * "Linking can fail for a variety of reasons as specified in the
4242 * OpenGL Shading Language Specification, as well as any of the
4243 * following reasons:
4245 * - No shader objects are attached to program."
4247 * The Compatibility Profile specification does not list the error. In
4248 * Compatibility Profile missing shader stages are replaced by
4249 * fixed-function. This applies to the case where all stages are
4252 if (prog
->NumShaders
== 0) {
4253 if (ctx
->API
!= API_OPENGL_COMPAT
)
4254 linker_error(prog
, "no shaders attached to the program\n");
4258 unsigned num_tfeedback_decls
= 0;
4259 unsigned int num_explicit_uniform_locs
= 0;
4260 bool has_xfb_qualifiers
= false;
4261 char **varying_names
= NULL
;
4262 tfeedback_decl
*tfeedback_decls
= NULL
;
4264 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4266 prog
->ARB_fragment_coord_conventions_enable
= false;
4268 /* Separate the shaders into groups based on their type.
4270 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4271 unsigned num_shaders
[MESA_SHADER_STAGES
];
4273 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4274 shader_list
[i
] = (struct gl_shader
**)
4275 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4279 unsigned min_version
= UINT_MAX
;
4280 unsigned max_version
= 0;
4281 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4282 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4283 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4285 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4286 linker_error(prog
, "all shaders must use same shading "
4287 "language version\n");
4291 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4292 prog
->ARB_fragment_coord_conventions_enable
= true;
4295 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4296 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4297 num_shaders
[shader_type
]++;
4300 /* In desktop GLSL, different shader versions may be linked together. In
4301 * GLSL ES, all shader versions must be the same.
4303 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4304 linker_error(prog
, "all shaders must use same shading "
4305 "language version\n");
4309 prog
->Version
= max_version
;
4310 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4312 /* Some shaders have to be linked with some other shaders present.
4314 if (!prog
->SeparateShader
) {
4315 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4316 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4317 linker_error(prog
, "Geometry shader must be linked with "
4321 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4322 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4323 linker_error(prog
, "Tessellation evaluation shader must be linked "
4324 "with vertex shader\n");
4327 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4328 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4329 linker_error(prog
, "Tessellation control shader must be linked with "
4334 /* The spec is self-contradictory here. It allows linking without a tess
4335 * eval shader, but that can only be used with transform feedback and
4336 * rasterization disabled. However, transform feedback isn't allowed
4337 * with GL_PATCHES, so it can't be used.
4339 * More investigation showed that the idea of transform feedback after
4340 * a tess control shader was dropped, because some hw vendors couldn't
4341 * support tessellation without a tess eval shader, but the linker
4342 * section wasn't updated to reflect that.
4344 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4347 * Do what's reasonable and always require a tess eval shader if a tess
4348 * control shader is present.
4350 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4351 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4352 linker_error(prog
, "Tessellation control shader must be linked with "
4353 "tessellation evaluation shader\n");
4358 /* Compute shaders have additional restrictions. */
4359 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4360 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4361 linker_error(prog
, "Compute shaders may not be linked with any other "
4362 "type of shader\n");
4365 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4366 if (prog
->_LinkedShaders
[i
] != NULL
)
4367 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4369 prog
->_LinkedShaders
[i
] = NULL
;
4372 /* Link all shaders for a particular stage and validate the result.
4374 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4375 if (num_shaders
[stage
] > 0) {
4376 gl_shader
*const sh
=
4377 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4378 num_shaders
[stage
]);
4380 if (!prog
->LinkStatus
) {
4382 _mesa_delete_shader(ctx
, sh
);
4387 case MESA_SHADER_VERTEX
:
4388 validate_vertex_shader_executable(prog
, sh
);
4390 case MESA_SHADER_TESS_CTRL
:
4391 /* nothing to be done */
4393 case MESA_SHADER_TESS_EVAL
:
4394 validate_tess_eval_shader_executable(prog
, sh
);
4396 case MESA_SHADER_GEOMETRY
:
4397 validate_geometry_shader_executable(prog
, sh
);
4399 case MESA_SHADER_FRAGMENT
:
4400 validate_fragment_shader_executable(prog
, sh
);
4403 if (!prog
->LinkStatus
) {
4405 _mesa_delete_shader(ctx
, sh
);
4409 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4413 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
4414 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4415 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
4416 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4417 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
4418 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4420 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4422 /* Here begins the inter-stage linking phase. Some initial validation is
4423 * performed, then locations are assigned for uniforms, attributes, and
4426 cross_validate_uniforms(prog
);
4427 if (!prog
->LinkStatus
)
4430 unsigned first
, last
, prev
;
4432 first
= MESA_SHADER_STAGES
;
4435 /* Determine first and last stage. */
4436 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4437 if (!prog
->_LinkedShaders
[i
])
4439 if (first
== MESA_SHADER_STAGES
)
4444 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4445 link_assign_subroutine_types(prog
);
4447 if (!prog
->LinkStatus
)
4450 resize_tes_inputs(ctx
, prog
);
4452 /* Validate the inputs of each stage with the output of the preceding
4456 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4457 if (prog
->_LinkedShaders
[i
] == NULL
)
4460 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4461 prog
->_LinkedShaders
[i
]);
4462 if (!prog
->LinkStatus
)
4465 cross_validate_outputs_to_inputs(prog
,
4466 prog
->_LinkedShaders
[prev
],
4467 prog
->_LinkedShaders
[i
]);
4468 if (!prog
->LinkStatus
)
4474 /* Cross-validate uniform blocks between shader stages */
4475 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4476 MESA_SHADER_STAGES
);
4477 if (!prog
->LinkStatus
)
4480 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4481 if (prog
->_LinkedShaders
[i
] != NULL
)
4482 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4485 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4486 * it before optimization because we want most of the checks to get
4487 * dropped thanks to constant propagation.
4489 * This rule also applies to GLSL ES 3.00.
4491 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4492 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4494 lower_discard_flow(sh
->ir
);
4498 if (prog
->SeparateShader
)
4499 disable_varying_optimizations_for_sso(prog
);
4501 if (!interstage_cross_validate_uniform_blocks(prog
))
4504 /* Do common optimization before assigning storage for attributes,
4505 * uniforms, and varyings. Later optimization could possibly make
4506 * some of that unused.
4508 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4509 if (prog
->_LinkedShaders
[i
] == NULL
)
4512 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4513 if (!prog
->LinkStatus
)
4516 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
4517 lower_clip_distance(prog
->_LinkedShaders
[i
]);
4520 if (ctx
->Const
.LowerTessLevel
) {
4521 lower_tess_level(prog
->_LinkedShaders
[i
]);
4524 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4525 &ctx
->Const
.ShaderCompilerOptions
[i
],
4526 ctx
->Const
.NativeIntegers
))
4529 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4532 /* Validation for special cases where we allow sampler array indexing
4533 * with loop induction variable. This check emits a warning or error
4534 * depending if backend can handle dynamic indexing.
4536 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4537 (prog
->IsES
&& prog
->Version
< 300)) {
4538 if (!validate_sampler_array_indexing(ctx
, prog
))
4542 /* Check and validate stream emissions in geometry shaders */
4543 validate_geometry_shader_emissions(ctx
, prog
);
4545 /* Mark all generic shader inputs and outputs as unpaired. */
4546 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4547 if (prog
->_LinkedShaders
[i
] != NULL
) {
4548 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4553 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4554 if (prog
->_LinkedShaders
[i
] == NULL
)
4557 match_explicit_outputs_to_inputs(prog
, prog
->_LinkedShaders
[prev
],
4558 prog
->_LinkedShaders
[i
]);
4562 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4563 MESA_SHADER_VERTEX
)) {
4567 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4568 MESA_SHADER_FRAGMENT
)) {
4572 /* From the ARB_enhanced_layouts spec:
4574 * "If the shader used to record output variables for transform feedback
4575 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4576 * qualifiers, the values specified by TransformFeedbackVaryings are
4577 * ignored, and the set of variables captured for transform feedback is
4578 * instead derived from the specified layout qualifiers."
4580 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4581 /* Find last stage before fragment shader */
4582 if (prog
->_LinkedShaders
[i
]) {
4583 has_xfb_qualifiers
=
4584 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4585 &num_tfeedback_decls
,
4591 if (!has_xfb_qualifiers
) {
4592 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4593 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4596 if (num_tfeedback_decls
!= 0) {
4597 /* From GL_EXT_transform_feedback:
4598 * A program will fail to link if:
4600 * * the <count> specified by TransformFeedbackVaryingsEXT is
4601 * non-zero, but the program object has no vertex or geometry
4604 if (first
>= MESA_SHADER_FRAGMENT
) {
4605 linker_error(prog
, "Transform feedback varyings specified, but "
4606 "no vertex, tessellation, or geometry shader is "
4611 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4612 num_tfeedback_decls
);
4613 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4614 varying_names
, tfeedback_decls
))
4618 /* If there is no fragment shader we need to set transform feedback.
4620 * For SSO we need also need to assign output locations, we assign them
4621 * here because we need to do it for both single stage programs and multi
4624 if (last
< MESA_SHADER_FRAGMENT
&&
4625 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4626 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4627 prog
->_LinkedShaders
[last
], NULL
,
4628 num_tfeedback_decls
, tfeedback_decls
))
4632 if (last
<= MESA_SHADER_FRAGMENT
) {
4633 /* Remove unused varyings from the first/last stage unless SSO */
4634 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4635 prog
->_LinkedShaders
[first
],
4637 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4638 prog
->_LinkedShaders
[last
],
4641 /* If the program is made up of only a single stage */
4642 if (first
== last
) {
4644 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4645 if (prog
->SeparateShader
) {
4646 /* Assign input locations for SSO, output locations are already
4649 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4650 NULL
/* producer */,
4652 0 /* num_tfeedback_decls */,
4653 NULL
/* tfeedback_decls */))
4657 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4658 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4661 /* Linking the stages in the opposite order (from fragment to vertex)
4662 * ensures that inter-shader outputs written to in an earlier stage
4663 * are eliminated if they are (transitively) not used in a later
4667 for (int i
= next
- 1; i
>= 0; i
--) {
4668 if (prog
->_LinkedShaders
[i
] == NULL
)
4671 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4672 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4674 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4675 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4679 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4680 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4683 /* This must be done after all dead varyings are eliminated. */
4684 if (!check_against_output_limit(ctx
, prog
, sh_i
))
4686 if (!check_against_input_limit(ctx
, prog
, sh_next
))
4694 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4695 has_xfb_qualifiers
))
4698 /* Split BufferInterfaceBlocks into UniformBlocks and ShaderStorageBlocks
4699 * for gl_shader_program and gl_shader, so that drivers that need separate
4700 * index spaces for each set can have that.
4702 for (unsigned i
= MESA_SHADER_VERTEX
; i
< MESA_SHADER_STAGES
; i
++) {
4703 if (prog
->_LinkedShaders
[i
] != NULL
) {
4704 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4705 split_ubos_and_ssbos(sh
,
4706 sh
->BufferInterfaceBlocks
,
4708 sh
->NumBufferInterfaceBlocks
,
4710 &sh
->NumUniformBlocks
,
4711 &sh
->ShaderStorageBlocks
,
4712 &sh
->NumShaderStorageBlocks
);
4716 split_ubos_and_ssbos(prog
,
4718 prog
->BufferInterfaceBlocks
,
4719 prog
->NumBufferInterfaceBlocks
,
4720 &prog
->UniformBlocks
,
4721 &prog
->NumUniformBlocks
,
4722 &prog
->ShaderStorageBlocks
,
4723 &prog
->NumShaderStorageBlocks
);
4725 update_array_sizes(prog
);
4726 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
,
4727 num_explicit_uniform_locs
,
4728 ctx
->Const
.MaxUserAssignableUniformLocations
);
4729 link_assign_atomic_counter_resources(ctx
, prog
);
4730 store_fragdepth_layout(prog
);
4732 link_calculate_subroutine_compat(prog
);
4733 check_resources(ctx
, prog
);
4734 check_subroutine_resources(prog
);
4735 check_image_resources(ctx
, prog
);
4736 link_check_atomic_counter_resources(ctx
, prog
);
4738 if (!prog
->LinkStatus
)
4741 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4742 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4743 * anything about shader linking when one of the shaders (vertex or
4744 * fragment shader) is absent. So, the extension shouldn't change the
4745 * behavior specified in GLSL specification.
4747 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4748 * "Linking can fail for a variety of reasons as specified in the
4749 * OpenGL ES Shading Language Specification, as well as any of the
4750 * following reasons:
4754 * * program contains objects to form either a vertex shader or
4755 * fragment shader, and program is not separable, and does not
4756 * contain objects to form both a vertex shader and fragment
4759 * However, the only scenario in 3.1+ where we don't require them both is
4760 * when we have a compute shader. For example:
4762 * - No shaders is a link error.
4763 * - Geom or Tess without a Vertex shader is a link error which means we
4764 * always require a Vertex shader and hence a Fragment shader.
4765 * - Finally a Compute shader linked with any other stage is a link error.
4767 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4768 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4769 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4770 linker_error(prog
, "program lacks a vertex shader\n");
4771 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4772 linker_error(prog
, "program lacks a fragment shader\n");
4776 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4777 if (prog
->_LinkedShaders
[i
] == NULL
)
4780 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerBufferInterfaceBlocks
)
4781 lower_ubo_reference(prog
->_LinkedShaders
[i
]);
4783 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerShaderSharedVariables
)
4784 lower_shared_reference(prog
->_LinkedShaders
[i
],
4785 &prog
->Comp
.SharedSize
);
4787 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4791 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4792 free(shader_list
[i
]);
4793 if (prog
->_LinkedShaders
[i
] == NULL
)
4796 /* Do a final validation step to make sure that the IR wasn't
4797 * invalidated by any modifications performed after intrastage linking.
4799 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4801 /* Retain any live IR, but trash the rest. */
4802 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4804 /* The symbol table in the linked shaders may contain references to
4805 * variables that were removed (e.g., unused uniforms). Since it may
4806 * contain junk, there is no possible valid use. Delete it and set the
4809 delete prog
->_LinkedShaders
[i
]->symbols
;
4810 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4813 ralloc_free(mem_ctx
);