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 buffer blocks and checks that all definitions of
1169 * blocks agree on their contents.
1172 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1175 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1176 struct gl_uniform_block
*blks
= NULL
;
1177 unsigned *num_blks
= validate_ssbo
? &prog
->NumShaderStorageBlocks
:
1178 &prog
->NumUniformBlocks
;
1180 unsigned max_num_buffer_blocks
= 0;
1181 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1182 if (prog
->_LinkedShaders
[i
]) {
1183 if (validate_ssbo
) {
1184 max_num_buffer_blocks
+=
1185 prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1187 max_num_buffer_blocks
+=
1188 prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1193 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1194 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1196 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1197 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1198 InterfaceBlockStageIndex
[i
][j
] = -1;
1203 unsigned sh_num_blocks
;
1204 struct gl_uniform_block
**sh_blks
;
1205 if (validate_ssbo
) {
1206 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1207 sh_blks
= sh
->ShaderStorageBlocks
;
1209 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1210 sh_blks
= sh
->UniformBlocks
;
1213 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1214 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1218 linker_error(prog
, "buffer block `%s' has mismatching "
1219 "definitions\n", sh_blks
[j
]->Name
);
1221 for (unsigned k
= 0; k
<= i
; k
++) {
1222 delete[] InterfaceBlockStageIndex
[k
];
1227 InterfaceBlockStageIndex
[i
][index
] = j
;
1231 /* Update per stage block pointers to point to the program list.
1232 * FIXME: We should be able to free the per stage blocks here.
1234 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1235 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1236 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1238 if (stage_index
!= -1) {
1239 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1241 blks
[j
].stageref
|= (1 << i
);
1243 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1244 sh
->ShaderStorageBlocks
: sh
->UniformBlocks
;
1246 sh_blks
[stage_index
] = &blks
[j
];
1251 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1252 delete[] InterfaceBlockStageIndex
[i
];
1256 prog
->ShaderStorageBlocks
= blks
;
1258 prog
->UniformBlocks
= blks
;
1265 * Populates a shaders symbol table with all global declarations
1268 populate_symbol_table(gl_shader
*sh
)
1270 sh
->symbols
= new(sh
) glsl_symbol_table
;
1272 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1276 if ((func
= inst
->as_function()) != NULL
) {
1277 sh
->symbols
->add_function(func
);
1278 } else if ((var
= inst
->as_variable()) != NULL
) {
1279 if (var
->data
.mode
!= ir_var_temporary
)
1280 sh
->symbols
->add_variable(var
);
1287 * Remap variables referenced in an instruction tree
1289 * This is used when instruction trees are cloned from one shader and placed in
1290 * another. These trees will contain references to \c ir_variable nodes that
1291 * do not exist in the target shader. This function finds these \c ir_variable
1292 * references and replaces the references with matching variables in the target
1295 * If there is no matching variable in the target shader, a clone of the
1296 * \c ir_variable is made and added to the target shader. The new variable is
1297 * added to \b both the instruction stream and the symbol table.
1299 * \param inst IR tree that is to be processed.
1300 * \param symbols Symbol table containing global scope symbols in the
1302 * \param instructions Instruction stream where new variable declarations
1306 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1309 class remap_visitor
: public ir_hierarchical_visitor
{
1311 remap_visitor(struct gl_shader
*target
,
1314 this->target
= target
;
1315 this->symbols
= target
->symbols
;
1316 this->instructions
= target
->ir
;
1317 this->temps
= temps
;
1320 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1322 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1323 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1325 assert(var
!= NULL
);
1327 return visit_continue
;
1330 ir_variable
*const existing
=
1331 this->symbols
->get_variable(ir
->var
->name
);
1332 if (existing
!= NULL
)
1335 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1337 this->symbols
->add_variable(copy
);
1338 this->instructions
->push_head(copy
);
1342 return visit_continue
;
1346 struct gl_shader
*target
;
1347 glsl_symbol_table
*symbols
;
1348 exec_list
*instructions
;
1352 remap_visitor
v(target
, temps
);
1359 * Move non-declarations from one instruction stream to another
1361 * The intended usage pattern of this function is to pass the pointer to the
1362 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1363 * pointer) for \c last and \c false for \c make_copies on the first
1364 * call. Successive calls pass the return value of the previous call for
1365 * \c last and \c true for \c make_copies.
1367 * \param instructions Source instruction stream
1368 * \param last Instruction after which new instructions should be
1369 * inserted in the target instruction stream
1370 * \param make_copies Flag selecting whether instructions in \c instructions
1371 * should be copied (via \c ir_instruction::clone) into the
1372 * target list or moved.
1375 * The new "last" instruction in the target instruction stream. This pointer
1376 * is suitable for use as the \c last parameter of a later call to this
1380 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1381 bool make_copies
, gl_shader
*target
)
1383 hash_table
*temps
= NULL
;
1386 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1387 hash_table_pointer_compare
);
1389 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1390 if (inst
->as_function())
1393 ir_variable
*var
= inst
->as_variable();
1394 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1397 assert(inst
->as_assignment()
1399 || inst
->as_if() /* for initializers with the ?: operator */
1400 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1403 inst
= inst
->clone(target
, NULL
);
1406 hash_table_insert(temps
, inst
, var
);
1408 remap_variables(inst
, target
, temps
);
1413 last
->insert_after(inst
);
1418 hash_table_dtor(temps
);
1425 * This class is only used in link_intrastage_shaders() below but declaring
1426 * it inside that function leads to compiler warnings with some versions of
1429 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1431 array_sizing_visitor()
1432 : mem_ctx(ralloc_context(NULL
)),
1433 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1434 hash_table_pointer_compare
))
1438 ~array_sizing_visitor()
1440 hash_table_dtor(this->unnamed_interfaces
);
1441 ralloc_free(this->mem_ctx
);
1444 virtual ir_visitor_status
visit(ir_variable
*var
)
1446 const glsl_type
*type_without_array
;
1447 fixup_type(&var
->type
, var
->data
.max_array_access
,
1448 var
->data
.from_ssbo_unsized_array
);
1449 type_without_array
= var
->type
->without_array();
1450 if (var
->type
->is_interface()) {
1451 if (interface_contains_unsized_arrays(var
->type
)) {
1452 const glsl_type
*new_type
=
1453 resize_interface_members(var
->type
,
1454 var
->get_max_ifc_array_access(),
1455 var
->is_in_shader_storage_block());
1456 var
->type
= new_type
;
1457 var
->change_interface_type(new_type
);
1459 } else if (type_without_array
->is_interface()) {
1460 if (interface_contains_unsized_arrays(type_without_array
)) {
1461 const glsl_type
*new_type
=
1462 resize_interface_members(type_without_array
,
1463 var
->get_max_ifc_array_access(),
1464 var
->is_in_shader_storage_block());
1465 var
->change_interface_type(new_type
);
1466 var
->type
= update_interface_members_array(var
->type
, new_type
);
1468 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1469 /* Store a pointer to the variable in the unnamed_interfaces
1472 ir_variable
**interface_vars
= (ir_variable
**)
1473 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1474 if (interface_vars
== NULL
) {
1475 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1477 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1480 unsigned index
= ifc_type
->field_index(var
->name
);
1481 assert(index
< ifc_type
->length
);
1482 assert(interface_vars
[index
] == NULL
);
1483 interface_vars
[index
] = var
;
1485 return visit_continue
;
1489 * For each unnamed interface block that was discovered while running the
1490 * visitor, adjust the interface type to reflect the newly assigned array
1491 * sizes, and fix up the ir_variable nodes to point to the new interface
1494 void fixup_unnamed_interface_types()
1496 hash_table_call_foreach(this->unnamed_interfaces
,
1497 fixup_unnamed_interface_type
, NULL
);
1502 * If the type pointed to by \c type represents an unsized array, replace
1503 * it with a sized array whose size is determined by max_array_access.
1505 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1506 bool from_ssbo_unsized_array
)
1508 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1509 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1510 max_array_access
+ 1);
1511 assert(*type
!= NULL
);
1515 static const glsl_type
*
1516 update_interface_members_array(const glsl_type
*type
,
1517 const glsl_type
*new_interface_type
)
1519 const glsl_type
*element_type
= type
->fields
.array
;
1520 if (element_type
->is_array()) {
1521 const glsl_type
*new_array_type
=
1522 update_interface_members_array(element_type
, new_interface_type
);
1523 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1525 return glsl_type::get_array_instance(new_interface_type
,
1531 * Determine whether the given interface type contains unsized arrays (if
1532 * it doesn't, array_sizing_visitor doesn't need to process it).
1534 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1536 for (unsigned i
= 0; i
< type
->length
; i
++) {
1537 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1538 if (elem_type
->is_unsized_array())
1545 * Create a new interface type based on the given type, with unsized arrays
1546 * replaced by sized arrays whose size is determined by
1547 * max_ifc_array_access.
1549 static const glsl_type
*
1550 resize_interface_members(const glsl_type
*type
,
1551 const unsigned *max_ifc_array_access
,
1554 unsigned num_fields
= type
->length
;
1555 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1556 memcpy(fields
, type
->fields
.structure
,
1557 num_fields
* sizeof(*fields
));
1558 for (unsigned i
= 0; i
< num_fields
; i
++) {
1559 /* If SSBO last member is unsized array, we don't replace it by a sized
1562 if (is_ssbo
&& i
== (num_fields
- 1))
1563 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1566 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1569 glsl_interface_packing packing
=
1570 (glsl_interface_packing
) type
->interface_packing
;
1571 const glsl_type
*new_ifc_type
=
1572 glsl_type::get_interface_instance(fields
, num_fields
,
1573 packing
, type
->name
);
1575 return new_ifc_type
;
1578 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1581 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1582 ir_variable
**interface_vars
= (ir_variable
**) data
;
1583 unsigned num_fields
= ifc_type
->length
;
1584 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1585 memcpy(fields
, ifc_type
->fields
.structure
,
1586 num_fields
* sizeof(*fields
));
1587 bool interface_type_changed
= false;
1588 for (unsigned i
= 0; i
< num_fields
; i
++) {
1589 if (interface_vars
[i
] != NULL
&&
1590 fields
[i
].type
!= interface_vars
[i
]->type
) {
1591 fields
[i
].type
= interface_vars
[i
]->type
;
1592 interface_type_changed
= true;
1595 if (!interface_type_changed
) {
1599 glsl_interface_packing packing
=
1600 (glsl_interface_packing
) ifc_type
->interface_packing
;
1601 const glsl_type
*new_ifc_type
=
1602 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1605 for (unsigned i
= 0; i
< num_fields
; i
++) {
1606 if (interface_vars
[i
] != NULL
)
1607 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1612 * Memory context used to allocate the data in \c unnamed_interfaces.
1617 * Hash table from const glsl_type * to an array of ir_variable *'s
1618 * pointing to the ir_variables constituting each unnamed interface block.
1620 hash_table
*unnamed_interfaces
;
1624 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1628 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1629 struct gl_shader_program
*prog
,
1630 struct gl_shader
*linked_shader
,
1631 struct gl_shader
**shader_list
,
1632 unsigned num_shaders
)
1634 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1635 linked_shader
->TransformFeedback
.BufferStride
[i
] = 0;
1638 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1639 struct gl_shader
*shader
= shader_list
[i
];
1641 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1642 if (shader
->TransformFeedback
.BufferStride
[j
]) {
1643 if (linked_shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1644 shader
->TransformFeedback
.BufferStride
[j
] != 0 &&
1645 linked_shader
->TransformFeedback
.BufferStride
[j
] !=
1646 shader
->TransformFeedback
.BufferStride
[j
]) {
1648 "intrastage shaders defined with conflicting "
1649 "xfb_stride for buffer %d (%d and %d)\n", j
,
1650 linked_shader
->TransformFeedback
.BufferStride
[j
],
1651 shader
->TransformFeedback
.BufferStride
[j
]);
1655 if (shader
->TransformFeedback
.BufferStride
[j
])
1656 linked_shader
->TransformFeedback
.BufferStride
[j
] =
1657 shader
->TransformFeedback
.BufferStride
[j
];
1662 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1663 if (linked_shader
->TransformFeedback
.BufferStride
[j
]) {
1664 prog
->TransformFeedback
.BufferStride
[j
] =
1665 linked_shader
->TransformFeedback
.BufferStride
[j
];
1667 /* We will validate doubles at a later stage */
1668 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1669 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1670 "multiple of 4 or if its applied to a type that is "
1671 "or contains a double a multiple of 8.",
1672 prog
->TransformFeedback
.BufferStride
[j
]);
1676 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1677 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1679 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1680 "limit has been exceeded.");
1688 * Performs the cross-validation of tessellation control shader vertices and
1689 * layout qualifiers for the attached tessellation control shaders,
1690 * and propagates them to the linked TCS and linked shader program.
1693 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1694 struct gl_shader
*linked_shader
,
1695 struct gl_shader
**shader_list
,
1696 unsigned num_shaders
)
1698 linked_shader
->TessCtrl
.VerticesOut
= 0;
1700 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1703 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1705 * "All tessellation control shader layout declarations in a program
1706 * must specify the same output patch vertex count. There must be at
1707 * least one layout qualifier specifying an output patch vertex count
1708 * in any program containing tessellation control shaders; however,
1709 * such a declaration is not required in all tessellation control
1713 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1714 struct gl_shader
*shader
= shader_list
[i
];
1716 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1717 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1718 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1719 linker_error(prog
, "tessellation control shader defined with "
1720 "conflicting output vertex count (%d and %d)\n",
1721 linked_shader
->TessCtrl
.VerticesOut
,
1722 shader
->TessCtrl
.VerticesOut
);
1725 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1729 /* Just do the intrastage -> interstage propagation right now,
1730 * since we already know we're in the right type of shader program
1733 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1734 linker_error(prog
, "tessellation control shader didn't declare "
1735 "vertices out layout qualifier\n");
1738 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1743 * Performs the cross-validation of tessellation evaluation shader
1744 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1745 * for the attached tessellation evaluation shaders, and propagates them
1746 * to the linked TES and linked shader program.
1749 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1750 struct gl_shader
*linked_shader
,
1751 struct gl_shader
**shader_list
,
1752 unsigned num_shaders
)
1754 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1755 linked_shader
->TessEval
.Spacing
= 0;
1756 linked_shader
->TessEval
.VertexOrder
= 0;
1757 linked_shader
->TessEval
.PointMode
= -1;
1759 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1762 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1764 * "At least one tessellation evaluation shader (compilation unit) in
1765 * a program must declare a primitive mode in its input layout.
1766 * Declaration vertex spacing, ordering, and point mode identifiers is
1767 * optional. It is not required that all tessellation evaluation
1768 * shaders in a program declare a primitive mode. If spacing or
1769 * vertex ordering declarations are omitted, the tessellation
1770 * primitive generator will use equal spacing or counter-clockwise
1771 * vertex ordering, respectively. If a point mode declaration is
1772 * omitted, the tessellation primitive generator will produce lines or
1773 * triangles according to the primitive mode."
1776 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1777 struct gl_shader
*shader
= shader_list
[i
];
1779 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1780 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1781 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1782 linker_error(prog
, "tessellation evaluation shader defined with "
1783 "conflicting input primitive modes.\n");
1786 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1789 if (shader
->TessEval
.Spacing
!= 0) {
1790 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1791 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1792 linker_error(prog
, "tessellation evaluation shader defined with "
1793 "conflicting vertex spacing.\n");
1796 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1799 if (shader
->TessEval
.VertexOrder
!= 0) {
1800 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1801 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1802 linker_error(prog
, "tessellation evaluation shader defined with "
1803 "conflicting ordering.\n");
1806 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1809 if (shader
->TessEval
.PointMode
!= -1) {
1810 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1811 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1812 linker_error(prog
, "tessellation evaluation shader defined with "
1813 "conflicting point modes.\n");
1816 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1821 /* Just do the intrastage -> interstage propagation right now,
1822 * since we already know we're in the right type of shader program
1825 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1827 "tessellation evaluation shader didn't declare input "
1828 "primitive modes.\n");
1831 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1833 if (linked_shader
->TessEval
.Spacing
== 0)
1834 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1835 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1837 if (linked_shader
->TessEval
.VertexOrder
== 0)
1838 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1839 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1841 if (linked_shader
->TessEval
.PointMode
== -1)
1842 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1843 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1848 * Performs the cross-validation of layout qualifiers specified in
1849 * redeclaration of gl_FragCoord for the attached fragment shaders,
1850 * and propagates them to the linked FS and linked shader program.
1853 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1854 struct gl_shader
*linked_shader
,
1855 struct gl_shader
**shader_list
,
1856 unsigned num_shaders
)
1858 linked_shader
->redeclares_gl_fragcoord
= false;
1859 linked_shader
->uses_gl_fragcoord
= false;
1860 linked_shader
->origin_upper_left
= false;
1861 linked_shader
->pixel_center_integer
= false;
1863 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1864 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1867 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1868 struct gl_shader
*shader
= shader_list
[i
];
1869 /* From the GLSL 1.50 spec, page 39:
1871 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1872 * it must be redeclared in all the fragment shaders in that program
1873 * that have a static use gl_FragCoord."
1875 if ((linked_shader
->redeclares_gl_fragcoord
1876 && !shader
->redeclares_gl_fragcoord
1877 && shader
->uses_gl_fragcoord
)
1878 || (shader
->redeclares_gl_fragcoord
1879 && !linked_shader
->redeclares_gl_fragcoord
1880 && linked_shader
->uses_gl_fragcoord
)) {
1881 linker_error(prog
, "fragment shader defined with conflicting "
1882 "layout qualifiers for gl_FragCoord\n");
1885 /* From the GLSL 1.50 spec, page 39:
1887 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1888 * single program must have the same set of qualifiers."
1890 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1891 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1892 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1893 linker_error(prog
, "fragment shader defined with conflicting "
1894 "layout qualifiers for gl_FragCoord\n");
1897 /* Update the linked shader state. Note that uses_gl_fragcoord should
1898 * accumulate the results. The other values should replace. If there
1899 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1900 * are already known to be the same.
1902 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1903 linked_shader
->redeclares_gl_fragcoord
=
1904 shader
->redeclares_gl_fragcoord
;
1905 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1906 || shader
->uses_gl_fragcoord
;
1907 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1908 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1911 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1916 * Performs the cross-validation of geometry shader max_vertices and
1917 * primitive type layout qualifiers for the attached geometry shaders,
1918 * and propagates them to the linked GS and linked shader program.
1921 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1922 struct gl_shader
*linked_shader
,
1923 struct gl_shader
**shader_list
,
1924 unsigned num_shaders
)
1926 linked_shader
->Geom
.VerticesOut
= 0;
1927 linked_shader
->Geom
.Invocations
= 0;
1928 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1929 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1931 /* No in/out qualifiers defined for anything but GLSL 1.50+
1932 * geometry shaders so far.
1934 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1937 /* From the GLSL 1.50 spec, page 46:
1939 * "All geometry shader output layout declarations in a program
1940 * must declare the same layout and same value for
1941 * max_vertices. There must be at least one geometry output
1942 * layout declaration somewhere in a program, but not all
1943 * geometry shaders (compilation units) are required to
1947 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1948 struct gl_shader
*shader
= shader_list
[i
];
1950 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1951 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1952 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1953 linker_error(prog
, "geometry shader defined with conflicting "
1957 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1960 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1961 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1962 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1963 linker_error(prog
, "geometry shader defined with conflicting "
1967 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1970 if (shader
->Geom
.VerticesOut
!= 0) {
1971 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1972 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1973 linker_error(prog
, "geometry shader defined with conflicting "
1974 "output vertex count (%d and %d)\n",
1975 linked_shader
->Geom
.VerticesOut
,
1976 shader
->Geom
.VerticesOut
);
1979 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1982 if (shader
->Geom
.Invocations
!= 0) {
1983 if (linked_shader
->Geom
.Invocations
!= 0 &&
1984 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1985 linker_error(prog
, "geometry shader defined with conflicting "
1986 "invocation count (%d and %d)\n",
1987 linked_shader
->Geom
.Invocations
,
1988 shader
->Geom
.Invocations
);
1991 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1995 /* Just do the intrastage -> interstage propagation right now,
1996 * since we already know we're in the right type of shader program
1999 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
2001 "geometry shader didn't declare primitive input type\n");
2004 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
2006 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
2008 "geometry shader didn't declare primitive output type\n");
2011 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
2013 if (linked_shader
->Geom
.VerticesOut
== 0) {
2015 "geometry shader didn't declare max_vertices\n");
2018 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
2020 if (linked_shader
->Geom
.Invocations
== 0)
2021 linked_shader
->Geom
.Invocations
= 1;
2023 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
2028 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2029 * qualifiers for the attached compute shaders, and propagate them to the
2030 * linked CS and linked shader program.
2033 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2034 struct gl_shader
*linked_shader
,
2035 struct gl_shader
**shader_list
,
2036 unsigned num_shaders
)
2038 for (int i
= 0; i
< 3; i
++)
2039 linked_shader
->Comp
.LocalSize
[i
] = 0;
2041 /* This function is called for all shader stages, but it only has an effect
2042 * for compute shaders.
2044 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2047 /* From the ARB_compute_shader spec, in the section describing local size
2050 * If multiple compute shaders attached to a single program object
2051 * declare local work-group size, the declarations must be identical;
2052 * otherwise a link-time error results. Furthermore, if a program
2053 * object contains any compute shaders, at least one must contain an
2054 * input layout qualifier specifying the local work sizes of the
2055 * program, or a link-time error will occur.
2057 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2058 struct gl_shader
*shader
= shader_list
[sh
];
2060 if (shader
->Comp
.LocalSize
[0] != 0) {
2061 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
2062 for (int i
= 0; i
< 3; i
++) {
2063 if (linked_shader
->Comp
.LocalSize
[i
] !=
2064 shader
->Comp
.LocalSize
[i
]) {
2065 linker_error(prog
, "compute shader defined with conflicting "
2071 for (int i
= 0; i
< 3; i
++)
2072 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
2076 /* Just do the intrastage -> interstage propagation right now,
2077 * since we already know we're in the right type of shader program
2080 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
2081 linker_error(prog
, "compute shader didn't declare local size\n");
2084 for (int i
= 0; i
< 3; i
++)
2085 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
2090 * Combine a group of shaders for a single stage to generate a linked shader
2093 * If this function is supplied a single shader, it is cloned, and the new
2094 * shader is returned.
2096 static struct gl_shader
*
2097 link_intrastage_shaders(void *mem_ctx
,
2098 struct gl_context
*ctx
,
2099 struct gl_shader_program
*prog
,
2100 struct gl_shader
**shader_list
,
2101 unsigned num_shaders
)
2103 struct gl_uniform_block
*ubo_blocks
= NULL
;
2104 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2105 unsigned num_ubo_blocks
= 0;
2106 unsigned num_ssbo_blocks
= 0;
2108 /* Check that global variables defined in multiple shaders are consistent.
2110 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
2111 if (!prog
->LinkStatus
)
2114 /* Check that interface blocks defined in multiple shaders are consistent.
2116 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2118 if (!prog
->LinkStatus
)
2121 /* Link up uniform blocks defined within this stage. */
2122 link_uniform_blocks(mem_ctx
, ctx
, prog
, shader_list
, num_shaders
,
2123 &ubo_blocks
, &num_ubo_blocks
, &ssbo_blocks
,
2126 if (!prog
->LinkStatus
)
2129 /* Check that there is only a single definition of each function signature
2130 * across all shaders.
2132 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2133 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2134 ir_function
*const f
= node
->as_function();
2139 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2140 ir_function
*const other
=
2141 shader_list
[j
]->symbols
->get_function(f
->name
);
2143 /* If the other shader has no function (and therefore no function
2144 * signatures) with the same name, skip to the next shader.
2149 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2150 if (!sig
->is_defined
|| sig
->is_builtin())
2153 ir_function_signature
*other_sig
=
2154 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2156 if ((other_sig
!= NULL
) && other_sig
->is_defined
2157 && !other_sig
->is_builtin()) {
2158 linker_error(prog
, "function `%s' is multiply defined\n",
2167 /* Find the shader that defines main, and make a clone of it.
2169 * Starting with the clone, search for undefined references. If one is
2170 * found, find the shader that defines it. Clone the reference and add
2171 * it to the shader. Repeat until there are no undefined references or
2172 * until a reference cannot be resolved.
2174 gl_shader
*main
= NULL
;
2175 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2176 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2177 main
= shader_list
[i
];
2183 linker_error(prog
, "%s shader lacks `main'\n",
2184 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2188 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2189 linked
->ir
= new(linked
) exec_list
;
2190 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2192 /* Copy ubo blocks to linked shader list */
2193 linked
->UniformBlocks
=
2194 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2195 ralloc_steal(linked
, ubo_blocks
);
2196 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2197 linked
->UniformBlocks
[i
] = &ubo_blocks
[i
];
2199 linked
->NumUniformBlocks
= num_ubo_blocks
;
2201 /* Copy ssbo blocks to linked shader list */
2202 linked
->ShaderStorageBlocks
=
2203 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2204 ralloc_steal(linked
, ssbo_blocks
);
2205 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2206 linked
->ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2208 linked
->NumShaderStorageBlocks
= num_ssbo_blocks
;
2210 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2211 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2212 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2213 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2214 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2215 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2218 populate_symbol_table(linked
);
2220 /* The pointer to the main function in the final linked shader (i.e., the
2221 * copy of the original shader that contained the main function).
2223 ir_function_signature
*const main_sig
=
2224 _mesa_get_main_function_signature(linked
);
2226 /* Move any instructions other than variable declarations or function
2227 * declarations into main.
2229 exec_node
*insertion_point
=
2230 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2233 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2234 if (shader_list
[i
] == main
)
2237 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2238 insertion_point
, true, linked
);
2241 /* Check if any shader needs built-in functions. */
2242 bool need_builtins
= false;
2243 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2244 if (shader_list
[i
]->uses_builtin_functions
) {
2245 need_builtins
= true;
2251 if (need_builtins
) {
2252 /* Make a temporary array one larger than shader_list, which will hold
2253 * the built-in function shader as well.
2255 gl_shader
**linking_shaders
= (gl_shader
**)
2256 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2258 ok
= linking_shaders
!= NULL
;
2261 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2262 _mesa_glsl_initialize_builtin_functions();
2263 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2265 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2267 free(linking_shaders
);
2269 _mesa_error_no_memory(__func__
);
2272 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2277 _mesa_delete_shader(ctx
, linked
);
2281 /* At this point linked should contain all of the linked IR, so
2282 * validate it to make sure nothing went wrong.
2284 validate_ir_tree(linked
->ir
);
2286 /* Set the size of geometry shader input arrays */
2287 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2288 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2289 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2290 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2291 ir
->accept(&input_resize_visitor
);
2295 if (ctx
->Const
.VertexID_is_zero_based
)
2296 lower_vertex_id(linked
);
2298 /* Validate correct usage of barrier() in the tess control shader */
2299 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2300 barrier_use_visitor
visitor(prog
);
2301 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2302 ir
->accept(&visitor
);
2306 /* Make a pass over all variable declarations to ensure that arrays with
2307 * unspecified sizes have a size specified. The size is inferred from the
2308 * max_array_access field.
2310 array_sizing_visitor v
;
2312 v
.fixup_unnamed_interface_types();
2318 * Update the sizes of linked shader uniform arrays to the maximum
2321 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2323 * If one or more elements of an array are active,
2324 * GetActiveUniform will return the name of the array in name,
2325 * subject to the restrictions listed above. The type of the array
2326 * is returned in type. The size parameter contains the highest
2327 * array element index used, plus one. The compiler or linker
2328 * determines the highest index used. There will be only one
2329 * active uniform reported by the GL per uniform array.
2333 update_array_sizes(struct gl_shader_program
*prog
)
2335 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2336 if (prog
->_LinkedShaders
[i
] == NULL
)
2339 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2340 ir_variable
*const var
= node
->as_variable();
2342 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2343 !var
->type
->is_array())
2346 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2347 * will not be eliminated. Since we always do std140, just
2348 * don't resize arrays in UBOs.
2350 * Atomic counters are supposed to get deterministic
2351 * locations assigned based on the declaration ordering and
2352 * sizes, array compaction would mess that up.
2354 * Subroutine uniforms are not removed.
2356 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2357 var
->type
->contains_subroutine())
2360 unsigned int size
= var
->data
.max_array_access
;
2361 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2362 if (prog
->_LinkedShaders
[j
] == NULL
)
2365 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2366 ir_variable
*other_var
= node2
->as_variable();
2370 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2371 other_var
->data
.max_array_access
> size
) {
2372 size
= other_var
->data
.max_array_access
;
2377 if (size
+ 1 != var
->type
->length
) {
2378 /* If this is a built-in uniform (i.e., it's backed by some
2379 * fixed-function state), adjust the number of state slots to
2380 * match the new array size. The number of slots per array entry
2381 * is not known. It seems safe to assume that the total number of
2382 * slots is an integer multiple of the number of array elements.
2383 * Determine the number of slots per array element by dividing by
2384 * the old (total) size.
2386 const unsigned num_slots
= var
->get_num_state_slots();
2387 if (num_slots
> 0) {
2388 var
->set_num_state_slots((size
+ 1)
2389 * (num_slots
/ var
->type
->length
));
2392 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2394 /* FINISHME: We should update the types of array
2395 * dereferences of this variable now.
2403 * Resize tessellation evaluation per-vertex inputs to the size of
2404 * tessellation control per-vertex outputs.
2407 resize_tes_inputs(struct gl_context
*ctx
,
2408 struct gl_shader_program
*prog
)
2410 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2413 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2414 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2416 /* If no control shader is present, then the TES inputs are statically
2417 * sized to MaxPatchVertices; the actual size of the arrays won't be
2418 * known until draw time.
2420 const int num_vertices
= tcs
2421 ? tcs
->TessCtrl
.VerticesOut
2422 : ctx
->Const
.MaxPatchVertices
;
2424 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2425 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2426 ir
->accept(&input_resize_visitor
);
2430 /* Convert the gl_PatchVerticesIn system value into a constant, since
2431 * the value is known at this point.
2433 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2434 ir_variable
*var
= ir
->as_variable();
2435 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2436 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2437 void *mem_ctx
= ralloc_parent(var
);
2438 var
->data
.mode
= ir_var_auto
;
2439 var
->data
.location
= 0;
2440 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2447 * Find a contiguous set of available bits in a bitmask.
2449 * \param used_mask Bits representing used (1) and unused (0) locations
2450 * \param needed_count Number of contiguous bits needed.
2453 * Base location of the available bits on success or -1 on failure.
2456 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2458 unsigned needed_mask
= (1 << needed_count
) - 1;
2459 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2461 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2462 * cannot optimize possibly infinite loops" for the loop below.
2464 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2467 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2468 if ((needed_mask
& ~used_mask
) == needed_mask
)
2479 * Assign locations for either VS inputs or FS outputs
2481 * \param prog Shader program whose variables need locations assigned
2482 * \param constants Driver specific constant values for the program.
2483 * \param target_index Selector for the program target to receive location
2484 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2485 * \c MESA_SHADER_FRAGMENT.
2488 * If locations are successfully assigned, true is returned. Otherwise an
2489 * error is emitted to the shader link log and false is returned.
2492 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2493 struct gl_constants
*constants
,
2494 unsigned target_index
)
2496 /* Maximum number of generic locations. This corresponds to either the
2497 * maximum number of draw buffers or the maximum number of generic
2500 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2501 constants
->Program
[target_index
].MaxAttribs
:
2502 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2504 /* Mark invalid locations as being used.
2506 unsigned used_locations
= (max_index
>= 32)
2507 ? ~0 : ~((1 << max_index
) - 1);
2508 unsigned double_storage_locations
= 0;
2510 assert((target_index
== MESA_SHADER_VERTEX
)
2511 || (target_index
== MESA_SHADER_FRAGMENT
));
2513 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2517 /* Operate in a total of four passes.
2519 * 1. Invalidate the location assignments for all vertex shader inputs.
2521 * 2. Assign locations for inputs that have user-defined (via
2522 * glBindVertexAttribLocation) locations and outputs that have
2523 * user-defined locations (via glBindFragDataLocation).
2525 * 3. Sort the attributes without assigned locations by number of slots
2526 * required in decreasing order. Fragmentation caused by attribute
2527 * locations assigned by the application may prevent large attributes
2528 * from having enough contiguous space.
2530 * 4. Assign locations to any inputs without assigned locations.
2533 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2534 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2536 const enum ir_variable_mode direction
=
2537 (target_index
== MESA_SHADER_VERTEX
)
2538 ? ir_var_shader_in
: ir_var_shader_out
;
2541 /* Temporary storage for the set of attributes that need locations assigned.
2547 /* Used below in the call to qsort. */
2548 static int compare(const void *a
, const void *b
)
2550 const temp_attr
*const l
= (const temp_attr
*) a
;
2551 const temp_attr
*const r
= (const temp_attr
*) b
;
2553 /* Reversed because we want a descending order sort below. */
2554 return r
->slots
- l
->slots
;
2557 assert(max_index
<= 32);
2559 unsigned num_attr
= 0;
2561 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2562 ir_variable
*const var
= node
->as_variable();
2564 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2567 if (var
->data
.explicit_location
) {
2568 var
->data
.is_unmatched_generic_inout
= 0;
2569 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2570 || (var
->data
.location
< 0)) {
2572 "invalid explicit location %d specified for `%s'\n",
2573 (var
->data
.location
< 0)
2574 ? var
->data
.location
2575 : var
->data
.location
- generic_base
,
2579 } else if (target_index
== MESA_SHADER_VERTEX
) {
2582 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2583 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2584 var
->data
.location
= binding
;
2585 var
->data
.is_unmatched_generic_inout
= 0;
2587 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2591 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2592 assert(binding
>= FRAG_RESULT_DATA0
);
2593 var
->data
.location
= binding
;
2594 var
->data
.is_unmatched_generic_inout
= 0;
2596 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2597 var
->data
.index
= index
;
2602 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2604 * "Output binding assignments will cause LinkProgram to fail:
2606 * If the program has an active output assigned to a location greater
2607 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2608 * an active output assigned an index greater than or equal to one;"
2610 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2611 var
->data
.location
- generic_base
>=
2612 (int) constants
->MaxDualSourceDrawBuffers
) {
2614 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2615 "with index %u for %s\n",
2616 var
->data
.location
- generic_base
, var
->data
.index
,
2621 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2623 /* If the variable is not a built-in and has a location statically
2624 * assigned in the shader (presumably via a layout qualifier), make sure
2625 * that it doesn't collide with other assigned locations. Otherwise,
2626 * add it to the list of variables that need linker-assigned locations.
2628 if (var
->data
.location
!= -1) {
2629 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2630 /* From page 61 of the OpenGL 4.0 spec:
2632 * "LinkProgram will fail if the attribute bindings assigned
2633 * by BindAttribLocation do not leave not enough space to
2634 * assign a location for an active matrix attribute or an
2635 * active attribute array, both of which require multiple
2636 * contiguous generic attributes."
2638 * I think above text prohibits the aliasing of explicit and
2639 * automatic assignments. But, aliasing is allowed in manual
2640 * assignments of attribute locations. See below comments for
2643 * From OpenGL 4.0 spec, page 61:
2645 * "It is possible for an application to bind more than one
2646 * attribute name to the same location. This is referred to as
2647 * aliasing. This will only work if only one of the aliased
2648 * attributes is active in the executable program, or if no
2649 * path through the shader consumes more than one attribute of
2650 * a set of attributes aliased to the same location. A link
2651 * error can occur if the linker determines that every path
2652 * through the shader consumes multiple aliased attributes,
2653 * but implementations are not required to generate an error
2656 * From GLSL 4.30 spec, page 54:
2658 * "A program will fail to link if any two non-vertex shader
2659 * input variables are assigned to the same location. For
2660 * vertex shaders, multiple input variables may be assigned
2661 * to the same location using either layout qualifiers or via
2662 * the OpenGL API. However, such aliasing is intended only to
2663 * support vertex shaders where each execution path accesses
2664 * at most one input per each location. Implementations are
2665 * permitted, but not required, to generate link-time errors
2666 * if they detect that every path through the vertex shader
2667 * executable accesses multiple inputs assigned to any single
2668 * location. For all shader types, a program will fail to link
2669 * if explicit location assignments leave the linker unable
2670 * to find space for other variables without explicit
2673 * From OpenGL ES 3.0 spec, page 56:
2675 * "Binding more than one attribute name to the same location
2676 * is referred to as aliasing, and is not permitted in OpenGL
2677 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2678 * fail when this condition exists. However, aliasing is
2679 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2680 * This will only work if only one of the aliased attributes
2681 * is active in the executable program, or if no path through
2682 * the shader consumes more than one attribute of a set of
2683 * attributes aliased to the same location. A link error can
2684 * occur if the linker determines that every path through the
2685 * shader consumes multiple aliased attributes, but implemen-
2686 * tations are not required to generate an error in this case."
2688 * After looking at above references from OpenGL, OpenGL ES and
2689 * GLSL specifications, we allow aliasing of vertex input variables
2690 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2692 * NOTE: This is not required by the spec but its worth mentioning
2693 * here that we're not doing anything to make sure that no path
2694 * through the vertex shader executable accesses multiple inputs
2695 * assigned to any single location.
2698 /* Mask representing the contiguous slots that will be used by
2701 const unsigned attr
= var
->data
.location
- generic_base
;
2702 const unsigned use_mask
= (1 << slots
) - 1;
2703 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2704 ? "vertex shader input" : "fragment shader output";
2706 /* Generate a link error if the requested locations for this
2707 * attribute exceed the maximum allowed attribute location.
2709 if (attr
+ slots
> max_index
) {
2711 "insufficient contiguous locations "
2712 "available for %s `%s' %d %d %d\n", string
,
2713 var
->name
, used_locations
, use_mask
, attr
);
2717 /* Generate a link error if the set of bits requested for this
2718 * attribute overlaps any previously allocated bits.
2720 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2721 if (target_index
== MESA_SHADER_FRAGMENT
||
2722 (prog
->IsES
&& prog
->Version
>= 300)) {
2724 "overlapping location is assigned "
2725 "to %s `%s' %d %d %d\n", string
,
2726 var
->name
, used_locations
, use_mask
, attr
);
2729 linker_warning(prog
,
2730 "overlapping location is assigned "
2731 "to %s `%s' %d %d %d\n", string
,
2732 var
->name
, used_locations
, use_mask
, attr
);
2736 used_locations
|= (use_mask
<< attr
);
2738 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2740 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2741 * active attribute variables may fail to link, unless
2742 * device-dependent optimizations are able to make the program
2743 * fit within available hardware resources. For the purposes
2744 * of this test, attribute variables of the type dvec3, dvec4,
2745 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2746 * count as consuming twice as many attributes as equivalent
2747 * single-precision types. While these types use the same number
2748 * of generic attributes as their single-precision equivalents,
2749 * implementations are permitted to consume two single-precision
2750 * vectors of internal storage for each three- or four-component
2751 * double-precision vector."
2753 * Mark this attribute slot as taking up twice as much space
2754 * so we can count it properly against limits. According to
2755 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2756 * is optional behavior, but it seems preferable.
2758 if (var
->type
->without_array()->is_dual_slot_double())
2759 double_storage_locations
|= (use_mask
<< attr
);
2765 if (num_attr
>= max_index
) {
2766 linker_error(prog
, "too many %s (max %u)",
2767 target_index
== MESA_SHADER_VERTEX
?
2768 "vertex shader inputs" : "fragment shader outputs",
2772 to_assign
[num_attr
].slots
= slots
;
2773 to_assign
[num_attr
].var
= var
;
2777 if (target_index
== MESA_SHADER_VERTEX
) {
2778 unsigned total_attribs_size
=
2779 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2780 _mesa_bitcount(double_storage_locations
);
2781 if (total_attribs_size
> max_index
) {
2783 "attempt to use %d vertex attribute slots only %d available ",
2784 total_attribs_size
, max_index
);
2789 /* If all of the attributes were assigned locations by the application (or
2790 * are built-in attributes with fixed locations), return early. This should
2791 * be the common case.
2796 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2798 if (target_index
== MESA_SHADER_VERTEX
) {
2799 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2800 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2801 * reserved to prevent it from being automatically allocated below.
2803 find_deref_visitor
find("gl_Vertex");
2805 if (find
.variable_found())
2806 used_locations
|= (1 << 0);
2809 for (unsigned i
= 0; i
< num_attr
; i
++) {
2810 /* Mask representing the contiguous slots that will be used by this
2813 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2815 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2818 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2819 ? "vertex shader input" : "fragment shader output";
2822 "insufficient contiguous locations "
2823 "available for %s `%s'\n",
2824 string
, to_assign
[i
].var
->name
);
2828 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2829 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2830 used_locations
|= (use_mask
<< location
);
2837 * Match explicit locations of outputs to inputs and deactivate the
2838 * unmatch flag if found so we don't optimise them away.
2841 match_explicit_outputs_to_inputs(struct gl_shader_program
*prog
,
2842 gl_shader
*producer
,
2843 gl_shader
*consumer
)
2845 glsl_symbol_table parameters
;
2846 ir_variable
*explicit_locations
[MAX_VARYING
] = { NULL
};
2848 /* Find all shader outputs in the "producer" stage.
2850 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2851 ir_variable
*const var
= node
->as_variable();
2853 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2856 if (var
->data
.explicit_location
&&
2857 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2858 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2859 if (explicit_locations
[idx
] == NULL
)
2860 explicit_locations
[idx
] = var
;
2864 /* Match inputs to outputs */
2865 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2866 ir_variable
*const input
= node
->as_variable();
2868 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2871 ir_variable
*output
= NULL
;
2872 if (input
->data
.explicit_location
2873 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2874 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
];
2876 if (output
!= NULL
){
2877 input
->data
.is_unmatched_generic_inout
= 0;
2878 output
->data
.is_unmatched_generic_inout
= 0;
2885 * Store the gl_FragDepth layout in the gl_shader_program struct.
2888 store_fragdepth_layout(struct gl_shader_program
*prog
)
2890 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2894 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2896 /* We don't look up the gl_FragDepth symbol directly because if
2897 * gl_FragDepth is not used in the shader, it's removed from the IR.
2898 * However, the symbol won't be removed from the symbol table.
2900 * We're only interested in the cases where the variable is NOT removed
2903 foreach_in_list(ir_instruction
, node
, ir
) {
2904 ir_variable
*const var
= node
->as_variable();
2906 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2910 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2911 switch (var
->data
.depth_layout
) {
2912 case ir_depth_layout_none
:
2913 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2915 case ir_depth_layout_any
:
2916 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2918 case ir_depth_layout_greater
:
2919 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2921 case ir_depth_layout_less
:
2922 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2924 case ir_depth_layout_unchanged
:
2925 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2936 * Validate the resources used by a program versus the implementation limits
2939 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2941 unsigned total_uniform_blocks
= 0;
2942 unsigned total_shader_storage_blocks
= 0;
2944 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2945 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2950 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2951 linker_error(prog
, "Too many %s shader texture samplers\n",
2952 _mesa_shader_stage_to_string(i
));
2955 if (sh
->num_uniform_components
>
2956 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2957 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2958 linker_warning(prog
, "Too many %s shader default uniform block "
2959 "components, but the driver will try to optimize "
2960 "them out; this is non-portable out-of-spec "
2962 _mesa_shader_stage_to_string(i
));
2964 linker_error(prog
, "Too many %s shader default uniform block "
2966 _mesa_shader_stage_to_string(i
));
2970 if (sh
->num_combined_uniform_components
>
2971 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2972 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2973 linker_warning(prog
, "Too many %s shader uniform components, "
2974 "but the driver will try to optimize them out; "
2975 "this is non-portable out-of-spec behavior\n",
2976 _mesa_shader_stage_to_string(i
));
2978 linker_error(prog
, "Too many %s shader uniform components\n",
2979 _mesa_shader_stage_to_string(i
));
2983 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
2984 total_uniform_blocks
+= sh
->NumUniformBlocks
;
2986 const unsigned max_uniform_blocks
=
2987 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2988 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
2989 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2990 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
2991 max_uniform_blocks
);
2994 const unsigned max_shader_storage_blocks
=
2995 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
2996 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
2997 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
2998 _mesa_shader_stage_to_string(i
),
2999 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
3003 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3004 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3005 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3008 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3009 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3010 total_shader_storage_blocks
,
3011 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3014 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
3015 if (prog
->UniformBlocks
[i
].UniformBufferSize
>
3016 ctx
->Const
.MaxUniformBlockSize
) {
3017 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3018 prog
->UniformBlocks
[i
].Name
,
3019 prog
->UniformBlocks
[i
].UniformBufferSize
,
3020 ctx
->Const
.MaxUniformBlockSize
);
3024 for (unsigned i
= 0; i
< prog
->NumShaderStorageBlocks
; i
++) {
3025 if (prog
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3026 ctx
->Const
.MaxShaderStorageBlockSize
) {
3027 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3028 prog
->ShaderStorageBlocks
[i
].Name
,
3029 prog
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3030 ctx
->Const
.MaxShaderStorageBlockSize
);
3036 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3038 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3039 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3044 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3045 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3051 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3052 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3053 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3054 if (fn
->types
[k
] == uni
->type
) {
3060 uni
->num_compatible_subroutines
= count
;
3066 check_subroutine_resources(struct gl_shader_program
*prog
)
3068 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3069 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3072 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3073 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3074 _mesa_shader_stage_to_string(i
));
3079 * Validate shader image resources.
3082 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3084 unsigned total_image_units
= 0;
3085 unsigned fragment_outputs
= 0;
3086 unsigned total_shader_storage_blocks
= 0;
3088 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3091 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3092 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3095 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3096 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3097 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3098 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3100 total_image_units
+= sh
->NumImages
;
3101 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3103 if (i
== MESA_SHADER_FRAGMENT
) {
3104 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3105 ir_variable
*var
= node
->as_variable();
3106 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3107 /* since there are no double fs outputs - pass false */
3108 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3114 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3115 linker_error(prog
, "Too many combined image uniforms\n");
3117 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3118 ctx
->Const
.MaxCombinedShaderOutputResources
)
3119 linker_error(prog
, "Too many combined image uniforms, shader storage "
3120 " buffers and fragment outputs\n");
3125 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3126 * for a variable, checks for overlaps between other uniforms using explicit
3130 reserve_explicit_locations(struct gl_shader_program
*prog
,
3131 string_to_uint_map
*map
, ir_variable
*var
)
3133 unsigned slots
= var
->type
->uniform_locations();
3134 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3135 unsigned return_value
= slots
;
3137 /* Resize remap table if locations do not fit in the current one. */
3138 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3139 prog
->UniformRemapTable
=
3140 reralloc(prog
, prog
->UniformRemapTable
,
3141 gl_uniform_storage
*,
3144 if (!prog
->UniformRemapTable
) {
3145 linker_error(prog
, "Out of memory during linking.\n");
3149 /* Initialize allocated space. */
3150 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3151 prog
->UniformRemapTable
[i
] = NULL
;
3153 prog
->NumUniformRemapTable
= max_loc
+ 1;
3156 for (unsigned i
= 0; i
< slots
; i
++) {
3157 unsigned loc
= var
->data
.location
+ i
;
3159 /* Check if location is already used. */
3160 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3162 /* Possibly same uniform from a different stage, this is ok. */
3164 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3169 /* ARB_explicit_uniform_location specification states:
3171 * "No two default-block uniform variables in the program can have
3172 * the same location, even if they are unused, otherwise a compiler
3173 * or linker error will be generated."
3176 "location qualifier for uniform %s overlaps "
3177 "previously used location\n",
3182 /* Initialize location as inactive before optimization
3183 * rounds and location assignment.
3185 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3188 /* Note, base location used for arrays. */
3189 map
->put(var
->data
.location
, var
->name
);
3191 return return_value
;
3195 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3196 struct gl_shader
*sh
,
3199 unsigned slots
= var
->type
->uniform_locations();
3200 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3202 /* Resize remap table if locations do not fit in the current one. */
3203 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3204 sh
->SubroutineUniformRemapTable
=
3205 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3206 gl_uniform_storage
*,
3209 if (!sh
->SubroutineUniformRemapTable
) {
3210 linker_error(prog
, "Out of memory during linking.\n");
3214 /* Initialize allocated space. */
3215 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3216 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3218 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3221 for (unsigned i
= 0; i
< slots
; i
++) {
3222 unsigned loc
= var
->data
.location
+ i
;
3224 /* Check if location is already used. */
3225 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3227 /* ARB_explicit_uniform_location specification states:
3228 * "No two subroutine uniform variables can have the same location
3229 * in the same shader stage, otherwise a compiler or linker error
3230 * will be generated."
3233 "location qualifier for uniform %s overlaps "
3234 "previously used location\n",
3239 /* Initialize location as inactive before optimization
3240 * rounds and location assignment.
3242 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3248 * Check and reserve all explicit uniform locations, called before
3249 * any optimizations happen to handle also inactive uniforms and
3250 * inactive array elements that may get trimmed away.
3253 check_explicit_uniform_locations(struct gl_context
*ctx
,
3254 struct gl_shader_program
*prog
)
3256 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3259 /* This map is used to detect if overlapping explicit locations
3260 * occur with the same uniform (from different stage) or a different one.
3262 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3265 linker_error(prog
, "Out of memory during linking.\n");
3269 unsigned entries_total
= 0;
3270 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3271 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3276 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3277 ir_variable
*var
= node
->as_variable();
3278 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3281 if (var
->data
.explicit_location
) {
3283 if (var
->type
->without_array()->is_subroutine())
3284 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3286 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3290 entries_total
+= slots
;
3301 struct empty_uniform_block
*current_block
= NULL
;
3303 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3304 /* We found empty space in UniformRemapTable. */
3305 if (prog
->UniformRemapTable
[i
] == NULL
) {
3306 /* We've found the beginning of a new continous block of empty slots */
3307 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3308 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3309 current_block
->start
= i
;
3310 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3311 ¤t_block
->link
);
3314 /* The current block continues, so we simply increment its slots */
3315 current_block
->slots
++;
3320 return entries_total
;
3324 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3325 GLenum type
, const char *name
)
3327 bool found_interface
= false;
3328 unsigned block_name_len
= 0;
3329 const char *block_name_dot
= strchr(name
, '.');
3331 /* These rules only apply to buffer variables. So we return
3332 * true for the rest of types.
3334 if (type
!= GL_BUFFER_VARIABLE
)
3337 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
3338 const char *block_name
= shProg
->ShaderStorageBlocks
[i
].Name
;
3339 block_name_len
= strlen(block_name
);
3341 const char *block_square_bracket
= strchr(block_name
, '[');
3342 if (block_square_bracket
) {
3343 /* The block is part of an array of named interfaces,
3344 * for the name comparison we ignore the "[x]" part.
3346 block_name_len
-= strlen(block_square_bracket
);
3349 if (block_name_dot
) {
3350 /* Check if the variable name starts with the interface
3351 * name. The interface name (if present) should have the
3352 * length than the interface block name we are comparing to.
3354 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3355 if (len
!= block_name_len
)
3359 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3360 found_interface
= true;
3365 /* We remove the interface name from the buffer variable name,
3366 * including the dot that follows it.
3368 if (found_interface
)
3369 name
= name
+ block_name_len
+ 1;
3371 /* From: ARB_program_interface_query extension:
3373 * "For an active shader storage block member declared as an array, an
3374 * entry will be generated only for the first array element, regardless
3375 * of its type. For arrays of aggregate types, the enumeration rules are
3376 * applied recursively for the single enumerated array element.
3378 const char *struct_first_dot
= strchr(name
, '.');
3379 const char *first_square_bracket
= strchr(name
, '[');
3381 /* The buffer variable is on top level and it is not an array */
3382 if (!first_square_bracket
) {
3384 /* The shader storage block member is a struct, then generate the entry */
3385 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3388 /* Shader storage block member is an array, only generate an entry for the
3389 * first array element.
3391 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3399 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3400 const void *data
, uint8_t stages
)
3404 /* If resource already exists, do not add it again. */
3405 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3406 if (prog
->ProgramResourceList
[i
].Data
== data
)
3409 prog
->ProgramResourceList
=
3411 prog
->ProgramResourceList
,
3412 gl_program_resource
,
3413 prog
->NumProgramResourceList
+ 1);
3415 if (!prog
->ProgramResourceList
) {
3416 linker_error(prog
, "Out of memory during linking.\n");
3420 struct gl_program_resource
*res
=
3421 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3425 res
->StageReferences
= stages
;
3427 prog
->NumProgramResourceList
++;
3432 /* Function checks if a variable var is a packed varying and
3433 * if given name is part of packed varying's list.
3435 * If a variable is a packed varying, it has a name like
3436 * 'packed:a,b,c' where a, b and c are separate variables.
3439 included_in_packed_varying(ir_variable
*var
, const char *name
)
3441 if (strncmp(var
->name
, "packed:", 7) != 0)
3444 char *list
= strdup(var
->name
+ 7);
3449 char *token
= strtok_r(list
, ",", &saveptr
);
3451 if (strcmp(token
, name
) == 0) {
3455 token
= strtok_r(NULL
, ",", &saveptr
);
3462 * Function builds a stage reference bitmask from variable name.
3465 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3470 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3471 * used for reference mask in gl_program_resource will need to be changed.
3473 assert(MESA_SHADER_STAGES
< 8);
3475 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3476 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3480 /* Shader symbol table may contain variables that have
3481 * been optimized away. Search IR for the variable instead.
3483 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3484 ir_variable
*var
= node
->as_variable();
3486 unsigned baselen
= strlen(var
->name
);
3488 if (included_in_packed_varying(var
, name
)) {
3493 /* Type needs to match if specified, otherwise we might
3494 * pick a variable with same name but different interface.
3496 if (var
->data
.mode
!= mode
)
3499 if (strncmp(var
->name
, name
, baselen
) == 0) {
3500 /* Check for exact name matches but also check for arrays and
3503 if (name
[baselen
] == '\0' ||
3504 name
[baselen
] == '[' ||
3505 name
[baselen
] == '.') {
3517 * Create gl_shader_variable from ir_variable class.
3519 static gl_shader_variable
*
3520 create_shader_variable(struct gl_shader_program
*shProg
,
3521 const ir_variable
*in
,
3522 const char *name
, const glsl_type
*type
,
3523 bool use_implicit_location
, int location
)
3525 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3529 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3530 * expect to see gl_VertexID in the program resource list. Pretend.
3532 if (in
->data
.mode
== ir_var_system_value
&&
3533 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3534 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3536 out
->name
= ralloc_strdup(shProg
, name
);
3542 /* From the ARB_program_interface_query specification:
3544 * "Not all active variables are assigned valid locations; the
3545 * following variables will have an effective location of -1:
3547 * * uniforms declared as atomic counters;
3549 * * members of a uniform block;
3551 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3553 * * inputs or outputs not declared with a "location" layout qualifier,
3554 * except for vertex shader inputs and fragment shader outputs."
3556 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3557 is_gl_identifier(in
->name
) ||
3558 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3561 out
->location
= location
;
3565 out
->index
= in
->data
.index
;
3566 out
->patch
= in
->data
.patch
;
3567 out
->mode
= in
->data
.mode
;
3573 add_shader_variable(struct gl_shader_program
*shProg
, unsigned stage_mask
,
3574 GLenum programInterface
, ir_variable
*var
,
3575 const char *name
, const glsl_type
*type
,
3576 bool use_implicit_location
, int location
)
3578 const bool is_vertex_input
=
3579 programInterface
== GL_PROGRAM_INPUT
&&
3580 stage_mask
== MESA_SHADER_VERTEX
;
3582 switch (type
->base_type
) {
3583 case GLSL_TYPE_STRUCT
: {
3584 /* From the ARB_program_interface_query specification:
3586 * "For an active variable declared as a structure, a separate entry
3587 * will be generated for each active structure member. The name of
3588 * each entry is formed by concatenating the name of the structure,
3589 * the "." character, and the name of the structure member. If a
3590 * structure member to enumerate is itself a structure or array, these
3591 * enumeration rules are applied recursively."
3593 unsigned field_location
= location
;
3594 for (unsigned i
= 0; i
< type
->length
; i
++) {
3595 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3596 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3597 if (!add_shader_variable(shProg
, stage_mask
, programInterface
,
3598 var
, field_name
, field
->type
,
3599 use_implicit_location
, field_location
))
3603 field
->type
->count_attribute_slots(is_vertex_input
);
3609 /* From the ARB_program_interface_query specification:
3611 * "For an active variable declared as a single instance of a basic
3612 * type, a single entry will be generated, using the variable name
3613 * from the shader source."
3615 gl_shader_variable
*sha_v
=
3616 create_shader_variable(shProg
, var
, name
, type
,
3617 use_implicit_location
, location
);
3621 return add_program_resource(shProg
, programInterface
, sha_v
, stage_mask
);
3627 add_interface_variables(struct gl_shader_program
*shProg
,
3628 unsigned stage
, GLenum programInterface
)
3630 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3632 foreach_in_list(ir_instruction
, node
, ir
) {
3633 ir_variable
*var
= node
->as_variable();
3635 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3640 switch (var
->data
.mode
) {
3641 case ir_var_system_value
:
3642 case ir_var_shader_in
:
3643 if (programInterface
!= GL_PROGRAM_INPUT
)
3645 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3646 : int(VARYING_SLOT_VAR0
);
3648 case ir_var_shader_out
:
3649 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3651 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3652 : int(VARYING_SLOT_VAR0
);
3658 /* Skip packed varyings, packed varyings are handled separately
3659 * by add_packed_varyings.
3661 if (strncmp(var
->name
, "packed:", 7) == 0)
3664 /* Skip fragdata arrays, these are handled separately
3665 * by add_fragdata_arrays.
3667 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3670 const bool vs_input_or_fs_output
=
3671 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3672 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3674 if (!add_shader_variable(shProg
, 1 << stage
, programInterface
,
3675 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3676 var
->data
.location
- loc_bias
))
3683 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
, GLenum type
)
3685 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3688 if (!sh
|| !sh
->packed_varyings
)
3691 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3692 ir_variable
*var
= node
->as_variable();
3694 switch (var
->data
.mode
) {
3695 case ir_var_shader_in
:
3696 iface
= GL_PROGRAM_INPUT
;
3698 case ir_var_shader_out
:
3699 iface
= GL_PROGRAM_OUTPUT
;
3702 unreachable("unexpected type");
3705 if (type
== iface
) {
3706 const int stage_mask
=
3707 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3708 if (!add_shader_variable(shProg
, stage_mask
,
3709 iface
, var
, var
->name
, var
->type
, false,
3710 var
->data
.location
- VARYING_SLOT_VAR0
))
3719 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3721 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3723 if (!sh
|| !sh
->fragdata_arrays
)
3726 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3727 ir_variable
*var
= node
->as_variable();
3729 assert(var
->data
.mode
== ir_var_shader_out
);
3731 if (!add_shader_variable(shProg
,
3732 1 << MESA_SHADER_FRAGMENT
,
3733 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3734 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3742 get_top_level_name(const char *name
)
3744 const char *first_dot
= strchr(name
, '.');
3745 const char *first_square_bracket
= strchr(name
, '[');
3747 /* From ARB_program_interface_query spec:
3749 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying the
3750 * number of active array elements of the top-level shader storage block
3751 * member containing to the active variable is written to <params>. If the
3752 * top-level block member is not declared as an array, the value one is
3753 * written to <params>. If the top-level block member is an array with no
3754 * declared size, the value zero is written to <params>.
3757 /* The buffer variable is on top level.*/
3758 if (!first_square_bracket
&& !first_dot
)
3759 name_size
= strlen(name
);
3760 else if ((!first_square_bracket
||
3761 (first_dot
&& first_dot
< first_square_bracket
)))
3762 name_size
= first_dot
- name
;
3764 name_size
= first_square_bracket
- name
;
3766 return strndup(name
, name_size
);
3770 get_var_name(const char *name
)
3772 const char *first_dot
= strchr(name
, '.');
3775 return strdup(name
);
3777 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3781 is_top_level_shader_storage_block_member(const char* name
,
3782 const char* interface_name
,
3783 const char* field_name
)
3785 bool result
= false;
3787 /* If the given variable is already a top-level shader storage
3788 * block member, then return array_size = 1.
3789 * We could have two possibilities: if we have an instanced
3790 * shader storage block or not instanced.
3792 * For the first, we check create a name as it was in top level and
3793 * compare it with the real name. If they are the same, then
3794 * the variable is already at top-level.
3796 * Full instanced name is: interface name + '.' + var name +
3799 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3800 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3801 if (!full_instanced_name
) {
3802 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3806 snprintf(full_instanced_name
, name_length
, "%s.%s",
3807 interface_name
, field_name
);
3809 /* Check if its top-level shader storage block member of an
3810 * instanced interface block, or of a unnamed interface block.
3812 if (strcmp(name
, full_instanced_name
) == 0 ||
3813 strcmp(name
, field_name
) == 0)
3816 free(full_instanced_name
);
3821 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3822 char *interface_name
, char *var_name
)
3824 /* From GL_ARB_program_interface_query spec:
3826 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer
3827 * identifying the number of active array elements of the top-level
3828 * shader storage block member containing to the active variable is
3829 * written to <params>. If the top-level block member is not
3830 * declared as an array, the value one is written to <params>. If
3831 * the top-level block member is an array with no declared size,
3832 * the value zero is written to <params>.
3834 if (is_top_level_shader_storage_block_member(uni
->name
,
3838 else if (field
->type
->is_unsized_array())
3840 else if (field
->type
->is_array())
3841 return field
->type
->length
;
3847 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3848 const glsl_struct_field
*field
, char *interface_name
,
3851 /* From GL_ARB_program_interface_query:
3853 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3854 * identifying the stride between array elements of the top-level
3855 * shader storage block member containing the active variable is
3856 * written to <params>. For top-level block members declared as
3857 * arrays, the value written is the difference, in basic machine
3858 * units, between the offsets of the active variable for
3859 * consecutive elements in the top-level array. For top-level
3860 * block members not declared as an array, zero is written to
3863 if (field
->type
->is_array()) {
3864 const enum glsl_matrix_layout matrix_layout
=
3865 glsl_matrix_layout(field
->matrix_layout
);
3866 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3867 const glsl_type
*array_type
= field
->type
->fields
.array
;
3869 if (is_top_level_shader_storage_block_member(uni
->name
,
3874 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3875 if (array_type
->is_record() || array_type
->is_array())
3876 return glsl_align(array_type
->std140_size(row_major
), 16);
3878 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3880 return array_type
->std430_array_stride(row_major
);
3887 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3888 struct gl_uniform_storage
*uni
)
3890 int block_index
= uni
->block_index
;
3891 int array_size
= -1;
3892 int array_stride
= -1;
3893 char *var_name
= get_top_level_name(uni
->name
);
3894 char *interface_name
=
3895 get_top_level_name(uni
->is_shader_storage
?
3896 shProg
->ShaderStorageBlocks
[block_index
].Name
:
3897 shProg
->UniformBlocks
[block_index
].Name
);
3899 if (strcmp(var_name
, interface_name
) == 0) {
3900 /* Deal with instanced array of SSBOs */
3901 char *temp_name
= get_var_name(uni
->name
);
3903 linker_error(shProg
, "Out of memory during linking.\n");
3904 goto write_top_level_array_size_and_stride
;
3907 var_name
= get_top_level_name(temp_name
);
3910 linker_error(shProg
, "Out of memory during linking.\n");
3911 goto write_top_level_array_size_and_stride
;
3915 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
3916 if (shProg
->Shaders
[i
] == NULL
)
3919 const gl_shader
*stage
= shProg
->Shaders
[i
];
3920 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
3921 ir_variable
*var
= node
->as_variable();
3922 if (!var
|| !var
->get_interface_type() ||
3923 var
->data
.mode
!= ir_var_shader_storage
)
3926 const glsl_type
*interface
= var
->get_interface_type();
3928 if (strcmp(interface_name
, interface
->name
) != 0)
3931 for (unsigned i
= 0; i
< interface
->length
; i
++) {
3932 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
3933 if (strcmp(field
->name
, var_name
) != 0)
3936 array_stride
= get_array_stride(uni
, interface
, field
,
3937 interface_name
, var_name
);
3938 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
3939 goto write_top_level_array_size_and_stride
;
3943 write_top_level_array_size_and_stride
:
3944 free(interface_name
);
3946 uni
->top_level_array_stride
= array_stride
;
3947 uni
->top_level_array_size
= array_size
;
3951 * Builds up a list of program resources that point to existing
3955 build_program_resource_list(struct gl_context
*ctx
,
3956 struct gl_shader_program
*shProg
)
3958 /* Rebuild resource list. */
3959 if (shProg
->ProgramResourceList
) {
3960 ralloc_free(shProg
->ProgramResourceList
);
3961 shProg
->ProgramResourceList
= NULL
;
3962 shProg
->NumProgramResourceList
= 0;
3965 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
3967 /* Determine first input and final output stage. These are used to
3968 * detect which variables should be enumerated in the resource list
3969 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
3971 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3972 if (!shProg
->_LinkedShaders
[i
])
3974 if (input_stage
== MESA_SHADER_STAGES
)
3979 /* Empty shader, no resources. */
3980 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
3983 /* Program interface needs to expose varyings in case of SSO. */
3984 if (shProg
->SeparateShader
) {
3985 if (!add_packed_varyings(shProg
, input_stage
, GL_PROGRAM_INPUT
))
3988 if (!add_packed_varyings(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
3992 if (!add_fragdata_arrays(shProg
))
3995 /* Add inputs and outputs to the resource list. */
3996 if (!add_interface_variables(shProg
, input_stage
, GL_PROGRAM_INPUT
))
3999 if (!add_interface_variables(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
4002 /* Add transform feedback varyings. */
4003 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
4004 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
4005 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
4006 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
4012 /* Add transform feedback buffers. */
4013 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4014 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
4015 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
4016 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_BUFFER
,
4017 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
4023 /* Add uniforms from uniform storage. */
4024 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4025 /* Do not add uniforms internally used by Mesa. */
4026 if (shProg
->UniformStorage
[i
].hidden
)
4030 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
4033 /* Add stagereferences for uniforms in a uniform block. */
4034 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
4035 int block_index
= shProg
->UniformStorage
[i
].block_index
;
4036 if (block_index
!= -1) {
4037 stageref
|= is_shader_storage
?
4038 shProg
->ShaderStorageBlocks
[block_index
].stageref
:
4039 shProg
->UniformBlocks
[block_index
].stageref
;
4042 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4043 if (!should_add_buffer_variable(shProg
, type
,
4044 shProg
->UniformStorage
[i
].name
))
4047 if (is_shader_storage
) {
4048 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
4051 if (!add_program_resource(shProg
, type
,
4052 &shProg
->UniformStorage
[i
], stageref
))
4056 /* Add program uniform blocks. */
4057 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
4058 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
4059 &shProg
->UniformBlocks
[i
], 0))
4063 /* Add program shader storage blocks. */
4064 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
4065 if (!add_program_resource(shProg
, GL_SHADER_STORAGE_BLOCK
,
4066 &shProg
->ShaderStorageBlocks
[i
], 0))
4070 /* Add atomic counter buffers. */
4071 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
4072 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
4073 &shProg
->AtomicBuffers
[i
], 0))
4077 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4079 if (!shProg
->UniformStorage
[i
].hidden
)
4082 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4083 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
4084 !shProg
->UniformStorage
[i
].type
->is_subroutine())
4087 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4088 /* add shader subroutines */
4089 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
4094 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4095 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
4101 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4102 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4103 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
4110 * This check is done to make sure we allow only constant expression
4111 * indexing and "constant-index-expression" (indexing with an expression
4112 * that includes loop induction variable).
4115 validate_sampler_array_indexing(struct gl_context
*ctx
,
4116 struct gl_shader_program
*prog
)
4118 dynamic_sampler_array_indexing_visitor v
;
4119 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4120 if (prog
->_LinkedShaders
[i
] == NULL
)
4123 bool no_dynamic_indexing
=
4124 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4126 /* Search for array derefs in shader. */
4127 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4128 if (v
.uses_dynamic_sampler_array_indexing()) {
4129 const char *msg
= "sampler arrays indexed with non-constant "
4130 "expressions is forbidden in GLSL %s %u";
4131 /* Backend has indicated that it has no dynamic indexing support. */
4132 if (no_dynamic_indexing
) {
4133 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4136 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4144 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4146 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4147 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4152 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4153 ir_function
*fn
= node
->as_function();
4157 if (fn
->is_subroutine
)
4158 sh
->NumSubroutineUniformTypes
++;
4160 if (!fn
->num_subroutine_types
)
4163 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4164 struct gl_subroutine_function
,
4165 sh
->NumSubroutineFunctions
+ 1);
4166 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4167 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4168 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4169 ralloc_array(sh
, const struct glsl_type
*,
4170 fn
->num_subroutine_types
);
4172 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4175 * "Each subroutine with an index qualifier in the shader must be
4176 * given a unique index, otherwise a compile or link error will be
4179 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4180 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4181 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4182 linker_error(prog
, "each subroutine index qualifier in the "
4183 "shader must be unique\n");
4187 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4188 fn
->subroutine_index
;
4190 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4191 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4192 sh
->NumSubroutineFunctions
++;
4195 /* Assign index for subroutines without an explicit index*/
4197 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4198 while (sh
->SubroutineFunctions
[j
].index
== -1) {
4199 for (unsigned k
= 0; k
< sh
->NumSubroutineFunctions
; k
++) {
4200 if (sh
->SubroutineFunctions
[k
].index
== index
)
4202 else if (k
== sh
->NumSubroutineFunctions
- 1)
4203 sh
->SubroutineFunctions
[j
].index
= index
;
4212 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4214 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4216 foreach_in_list(ir_instruction
, node
, ir
) {
4217 ir_variable
*const var
= node
->as_variable();
4219 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4222 /* Don't set always active on builtins that haven't been redeclared */
4223 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4226 var
->data
.always_active_io
= true;
4231 * When separate shader programs are enabled, only input/outputs between
4232 * the stages of a multi-stage separate program can be safely removed
4233 * from the shader interface. Other inputs/outputs must remain active.
4236 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4238 unsigned first
, last
;
4239 assert(prog
->SeparateShader
);
4241 first
= MESA_SHADER_STAGES
;
4244 /* Determine first and last stage. Excluding the compute stage */
4245 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4246 if (!prog
->_LinkedShaders
[i
])
4248 if (first
== MESA_SHADER_STAGES
)
4253 if (first
== MESA_SHADER_STAGES
)
4256 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4257 gl_shader
*sh
= prog
->_LinkedShaders
[stage
];
4261 if (first
== last
) {
4262 /* For a single shader program only allow inputs to the vertex shader
4263 * and outputs from the fragment shader to be removed.
4265 if (stage
!= MESA_SHADER_VERTEX
)
4266 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4267 if (stage
!= MESA_SHADER_FRAGMENT
)
4268 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4270 /* For multi-stage separate shader programs only allow inputs and
4271 * outputs between the shader stages to be removed as well as inputs
4272 * to the vertex shader and outputs from the fragment shader.
4274 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4275 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4276 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4277 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4283 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4285 prog
->LinkStatus
= true; /* All error paths will set this to false */
4286 prog
->Validated
= false;
4287 prog
->_Used
= false;
4289 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4291 * "Linking can fail for a variety of reasons as specified in the
4292 * OpenGL Shading Language Specification, as well as any of the
4293 * following reasons:
4295 * - No shader objects are attached to program."
4297 * The Compatibility Profile specification does not list the error. In
4298 * Compatibility Profile missing shader stages are replaced by
4299 * fixed-function. This applies to the case where all stages are
4302 if (prog
->NumShaders
== 0) {
4303 if (ctx
->API
!= API_OPENGL_COMPAT
)
4304 linker_error(prog
, "no shaders attached to the program\n");
4308 unsigned num_tfeedback_decls
= 0;
4309 unsigned int num_explicit_uniform_locs
= 0;
4310 bool has_xfb_qualifiers
= false;
4311 char **varying_names
= NULL
;
4312 tfeedback_decl
*tfeedback_decls
= NULL
;
4314 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4316 prog
->ARB_fragment_coord_conventions_enable
= false;
4318 /* Separate the shaders into groups based on their type.
4320 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4321 unsigned num_shaders
[MESA_SHADER_STAGES
];
4323 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4324 shader_list
[i
] = (struct gl_shader
**)
4325 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4329 unsigned min_version
= UINT_MAX
;
4330 unsigned max_version
= 0;
4331 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4332 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4333 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4335 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4336 linker_error(prog
, "all shaders must use same shading "
4337 "language version\n");
4341 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4342 prog
->ARB_fragment_coord_conventions_enable
= true;
4345 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4346 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4347 num_shaders
[shader_type
]++;
4350 /* In desktop GLSL, different shader versions may be linked together. In
4351 * GLSL ES, all shader versions must be the same.
4353 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4354 linker_error(prog
, "all shaders must use same shading "
4355 "language version\n");
4359 prog
->Version
= max_version
;
4360 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4362 /* Some shaders have to be linked with some other shaders present.
4364 if (!prog
->SeparateShader
) {
4365 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4366 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4367 linker_error(prog
, "Geometry shader must be linked with "
4371 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4372 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4373 linker_error(prog
, "Tessellation evaluation shader must be linked "
4374 "with vertex shader\n");
4377 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4378 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4379 linker_error(prog
, "Tessellation control shader must be linked with "
4384 /* The spec is self-contradictory here. It allows linking without a tess
4385 * eval shader, but that can only be used with transform feedback and
4386 * rasterization disabled. However, transform feedback isn't allowed
4387 * with GL_PATCHES, so it can't be used.
4389 * More investigation showed that the idea of transform feedback after
4390 * a tess control shader was dropped, because some hw vendors couldn't
4391 * support tessellation without a tess eval shader, but the linker
4392 * section wasn't updated to reflect that.
4394 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4397 * Do what's reasonable and always require a tess eval shader if a tess
4398 * control shader is present.
4400 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4401 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4402 linker_error(prog
, "Tessellation control shader must be linked with "
4403 "tessellation evaluation shader\n");
4408 /* Compute shaders have additional restrictions. */
4409 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4410 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4411 linker_error(prog
, "Compute shaders may not be linked with any other "
4412 "type of shader\n");
4415 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4416 if (prog
->_LinkedShaders
[i
] != NULL
)
4417 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4419 prog
->_LinkedShaders
[i
] = NULL
;
4422 /* Link all shaders for a particular stage and validate the result.
4424 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4425 if (num_shaders
[stage
] > 0) {
4426 gl_shader
*const sh
=
4427 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4428 num_shaders
[stage
]);
4430 if (!prog
->LinkStatus
) {
4432 _mesa_delete_shader(ctx
, sh
);
4437 case MESA_SHADER_VERTEX
:
4438 validate_vertex_shader_executable(prog
, sh
);
4440 case MESA_SHADER_TESS_CTRL
:
4441 /* nothing to be done */
4443 case MESA_SHADER_TESS_EVAL
:
4444 validate_tess_eval_shader_executable(prog
, sh
);
4446 case MESA_SHADER_GEOMETRY
:
4447 validate_geometry_shader_executable(prog
, sh
);
4449 case MESA_SHADER_FRAGMENT
:
4450 validate_fragment_shader_executable(prog
, sh
);
4453 if (!prog
->LinkStatus
) {
4455 _mesa_delete_shader(ctx
, sh
);
4459 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4463 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
4464 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4465 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
4466 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4467 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
4468 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4470 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4472 /* Here begins the inter-stage linking phase. Some initial validation is
4473 * performed, then locations are assigned for uniforms, attributes, and
4476 cross_validate_uniforms(prog
);
4477 if (!prog
->LinkStatus
)
4480 unsigned first
, last
, prev
;
4482 first
= MESA_SHADER_STAGES
;
4485 /* Determine first and last stage. */
4486 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4487 if (!prog
->_LinkedShaders
[i
])
4489 if (first
== MESA_SHADER_STAGES
)
4494 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4495 link_assign_subroutine_types(prog
);
4497 if (!prog
->LinkStatus
)
4500 resize_tes_inputs(ctx
, prog
);
4502 /* Validate the inputs of each stage with the output of the preceding
4506 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4507 if (prog
->_LinkedShaders
[i
] == NULL
)
4510 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4511 prog
->_LinkedShaders
[i
]);
4512 if (!prog
->LinkStatus
)
4515 cross_validate_outputs_to_inputs(prog
,
4516 prog
->_LinkedShaders
[prev
],
4517 prog
->_LinkedShaders
[i
]);
4518 if (!prog
->LinkStatus
)
4524 /* Cross-validate uniform blocks between shader stages */
4525 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4526 MESA_SHADER_STAGES
);
4527 if (!prog
->LinkStatus
)
4530 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4531 if (prog
->_LinkedShaders
[i
] != NULL
)
4532 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4535 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4536 * it before optimization because we want most of the checks to get
4537 * dropped thanks to constant propagation.
4539 * This rule also applies to GLSL ES 3.00.
4541 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4542 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4544 lower_discard_flow(sh
->ir
);
4548 if (prog
->SeparateShader
)
4549 disable_varying_optimizations_for_sso(prog
);
4552 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4556 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4559 /* Do common optimization before assigning storage for attributes,
4560 * uniforms, and varyings. Later optimization could possibly make
4561 * some of that unused.
4563 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4564 if (prog
->_LinkedShaders
[i
] == NULL
)
4567 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4568 if (!prog
->LinkStatus
)
4571 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
4572 lower_clip_distance(prog
->_LinkedShaders
[i
]);
4575 if (ctx
->Const
.LowerTessLevel
) {
4576 lower_tess_level(prog
->_LinkedShaders
[i
]);
4579 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4580 &ctx
->Const
.ShaderCompilerOptions
[i
],
4581 ctx
->Const
.NativeIntegers
))
4584 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4587 /* Validation for special cases where we allow sampler array indexing
4588 * with loop induction variable. This check emits a warning or error
4589 * depending if backend can handle dynamic indexing.
4591 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4592 (prog
->IsES
&& prog
->Version
< 300)) {
4593 if (!validate_sampler_array_indexing(ctx
, prog
))
4597 /* Check and validate stream emissions in geometry shaders */
4598 validate_geometry_shader_emissions(ctx
, prog
);
4600 /* Mark all generic shader inputs and outputs as unpaired. */
4601 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4602 if (prog
->_LinkedShaders
[i
] != NULL
) {
4603 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4608 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4609 if (prog
->_LinkedShaders
[i
] == NULL
)
4612 match_explicit_outputs_to_inputs(prog
, prog
->_LinkedShaders
[prev
],
4613 prog
->_LinkedShaders
[i
]);
4617 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4618 MESA_SHADER_VERTEX
)) {
4622 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4623 MESA_SHADER_FRAGMENT
)) {
4627 /* From the ARB_enhanced_layouts spec:
4629 * "If the shader used to record output variables for transform feedback
4630 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4631 * qualifiers, the values specified by TransformFeedbackVaryings are
4632 * ignored, and the set of variables captured for transform feedback is
4633 * instead derived from the specified layout qualifiers."
4635 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4636 /* Find last stage before fragment shader */
4637 if (prog
->_LinkedShaders
[i
]) {
4638 has_xfb_qualifiers
=
4639 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4640 &num_tfeedback_decls
,
4646 if (!has_xfb_qualifiers
) {
4647 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4648 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4651 if (num_tfeedback_decls
!= 0) {
4652 /* From GL_EXT_transform_feedback:
4653 * A program will fail to link if:
4655 * * the <count> specified by TransformFeedbackVaryingsEXT is
4656 * non-zero, but the program object has no vertex or geometry
4659 if (first
>= MESA_SHADER_FRAGMENT
) {
4660 linker_error(prog
, "Transform feedback varyings specified, but "
4661 "no vertex, tessellation, or geometry shader is "
4666 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4667 num_tfeedback_decls
);
4668 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4669 varying_names
, tfeedback_decls
))
4673 /* If there is no fragment shader we need to set transform feedback.
4675 * For SSO we need also need to assign output locations, we assign them
4676 * here because we need to do it for both single stage programs and multi
4679 if (last
< MESA_SHADER_FRAGMENT
&&
4680 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4681 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4682 prog
->_LinkedShaders
[last
], NULL
,
4683 num_tfeedback_decls
, tfeedback_decls
))
4687 if (last
<= MESA_SHADER_FRAGMENT
) {
4688 /* Remove unused varyings from the first/last stage unless SSO */
4689 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4690 prog
->_LinkedShaders
[first
],
4692 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4693 prog
->_LinkedShaders
[last
],
4696 /* If the program is made up of only a single stage */
4697 if (first
== last
) {
4699 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4700 if (prog
->SeparateShader
) {
4701 /* Assign input locations for SSO, output locations are already
4704 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4705 NULL
/* producer */,
4707 0 /* num_tfeedback_decls */,
4708 NULL
/* tfeedback_decls */))
4712 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4713 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4716 /* Linking the stages in the opposite order (from fragment to vertex)
4717 * ensures that inter-shader outputs written to in an earlier stage
4718 * are eliminated if they are (transitively) not used in a later
4722 for (int i
= next
- 1; i
>= 0; i
--) {
4723 if (prog
->_LinkedShaders
[i
] == NULL
)
4726 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4727 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4729 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4730 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4734 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4735 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4738 /* This must be done after all dead varyings are eliminated. */
4739 if (!check_against_output_limit(ctx
, prog
, sh_i
))
4741 if (!check_against_input_limit(ctx
, prog
, sh_next
))
4749 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4750 has_xfb_qualifiers
))
4753 update_array_sizes(prog
);
4754 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
,
4755 num_explicit_uniform_locs
,
4756 ctx
->Const
.MaxUserAssignableUniformLocations
);
4757 link_assign_atomic_counter_resources(ctx
, prog
);
4758 store_fragdepth_layout(prog
);
4760 link_calculate_subroutine_compat(prog
);
4761 check_resources(ctx
, prog
);
4762 check_subroutine_resources(prog
);
4763 check_image_resources(ctx
, prog
);
4764 link_check_atomic_counter_resources(ctx
, prog
);
4766 if (!prog
->LinkStatus
)
4769 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4770 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4771 * anything about shader linking when one of the shaders (vertex or
4772 * fragment shader) is absent. So, the extension shouldn't change the
4773 * behavior specified in GLSL specification.
4775 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4776 * "Linking can fail for a variety of reasons as specified in the
4777 * OpenGL ES Shading Language Specification, as well as any of the
4778 * following reasons:
4782 * * program contains objects to form either a vertex shader or
4783 * fragment shader, and program is not separable, and does not
4784 * contain objects to form both a vertex shader and fragment
4787 * However, the only scenario in 3.1+ where we don't require them both is
4788 * when we have a compute shader. For example:
4790 * - No shaders is a link error.
4791 * - Geom or Tess without a Vertex shader is a link error which means we
4792 * always require a Vertex shader and hence a Fragment shader.
4793 * - Finally a Compute shader linked with any other stage is a link error.
4795 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4796 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4797 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4798 linker_error(prog
, "program lacks a vertex shader\n");
4799 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4800 linker_error(prog
, "program lacks a fragment shader\n");
4804 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4805 if (prog
->_LinkedShaders
[i
] == NULL
)
4808 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerBufferInterfaceBlocks
)
4809 lower_ubo_reference(prog
->_LinkedShaders
[i
]);
4811 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerShaderSharedVariables
)
4812 lower_shared_reference(prog
->_LinkedShaders
[i
],
4813 &prog
->Comp
.SharedSize
);
4815 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4819 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4820 free(shader_list
[i
]);
4821 if (prog
->_LinkedShaders
[i
] == NULL
)
4824 /* Do a final validation step to make sure that the IR wasn't
4825 * invalidated by any modifications performed after intrastage linking.
4827 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4829 /* Retain any live IR, but trash the rest. */
4830 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4832 /* The symbol table in the linked shaders may contain references to
4833 * variables that were removed (e.g., unused uniforms). Since it may
4834 * contain junk, there is no possible valid use. Delete it and set the
4837 delete prog
->_LinkedShaders
[i
]->symbols
;
4838 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4841 ralloc_free(mem_ctx
);