2 * Copyright © 2010 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
26 * GLSL linker implementation
28 * Given a set of shaders that are to be linked to generate a final program,
29 * there are three distinct stages.
31 * In the first stage shaders are partitioned into groups based on the shader
32 * type. All shaders of a particular type (e.g., vertex shaders) are linked
35 * - Undefined references in each shader are resolve to definitions in
37 * - Types and qualifiers of uniforms, outputs, and global variables defined
38 * in multiple shaders with the same name are verified to be the same.
39 * - Initializers for uniforms and global variables defined
40 * in multiple shaders with the same name are verified to be the same.
42 * The result, in the terminology of the GLSL spec, is a set of shader
43 * executables for each processing unit.
45 * After the first stage is complete, a series of semantic checks are performed
46 * on each of the shader executables.
48 * - Each shader executable must define a \c main function.
49 * - Each vertex shader executable must write to \c gl_Position.
50 * - Each fragment shader executable must write to either \c gl_FragData or
53 * In the final stage individual shader executables are linked to create a
54 * complete exectuable.
56 * - Types of uniforms defined in multiple shader stages with the same name
57 * are verified to be the same.
58 * - Initializers for uniforms defined in multiple shader stages with the
59 * same name are verified to be the same.
60 * - Types and qualifiers of outputs defined in one stage are verified to
61 * be the same as the types and qualifiers of inputs defined with the same
62 * name in a later stage.
64 * \author Ian Romanick <ian.d.romanick@intel.com>
68 #include "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/hash_table.h"
76 #include "link_varyings.h"
77 #include "ir_optimization.h"
78 #include "ir_rvalue_visitor.h"
79 #include "ir_uniform.h"
81 #include "main/shaderobj.h"
82 #include "main/enums.h"
88 * Visitor that determines whether or not a variable is ever written.
90 class find_assignment_visitor
: public ir_hierarchical_visitor
{
92 find_assignment_visitor(const char *name
)
93 : name(name
), found(false)
98 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
100 ir_variable
*const var
= ir
->lhs
->variable_referenced();
102 if (strcmp(name
, var
->name
) == 0) {
107 return visit_continue_with_parent
;
110 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
112 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
113 actual_node
, &ir
->actual_parameters
) {
114 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
115 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
117 if (sig_param
->data
.mode
== ir_var_function_out
||
118 sig_param
->data
.mode
== ir_var_function_inout
) {
119 ir_variable
*var
= param_rval
->variable_referenced();
120 if (var
&& strcmp(name
, var
->name
) == 0) {
127 if (ir
->return_deref
!= NULL
) {
128 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
130 if (strcmp(name
, var
->name
) == 0) {
136 return visit_continue_with_parent
;
139 bool variable_found()
145 const char *name
; /**< Find writes to a variable with this name. */
146 bool found
; /**< Was a write to the variable found? */
151 * Visitor that determines whether or not a variable is ever read.
153 class find_deref_visitor
: public ir_hierarchical_visitor
{
155 find_deref_visitor(const char *name
)
156 : name(name
), found(false)
161 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
163 if (strcmp(this->name
, ir
->var
->name
) == 0) {
168 return visit_continue
;
171 bool variable_found() const
177 const char *name
; /**< Find writes to a variable with this name. */
178 bool found
; /**< Was a write to the variable found? */
182 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
184 unsigned num_vertices
;
185 gl_shader_program
*prog
;
187 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
189 this->num_vertices
= num_vertices
;
193 virtual ~geom_array_resize_visitor()
198 virtual ir_visitor_status
visit(ir_variable
*var
)
200 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
201 return visit_continue
;
203 unsigned size
= var
->type
->length
;
205 /* Generate a link error if the shader has declared this array with an
208 if (size
&& size
!= this->num_vertices
) {
209 linker_error(this->prog
, "size of array %s declared as %u, "
210 "but number of input vertices is %u\n",
211 var
->name
, size
, this->num_vertices
);
212 return visit_continue
;
215 /* Generate a link error if the shader attempts to access an input
216 * array using an index too large for its actual size assigned at link
219 if (var
->data
.max_array_access
>= this->num_vertices
) {
220 linker_error(this->prog
, "geometry shader accesses element %i of "
221 "%s, but only %i input vertices\n",
222 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
223 return visit_continue
;
226 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
228 var
->data
.max_array_access
= this->num_vertices
- 1;
230 return visit_continue
;
233 /* Dereferences of input variables need to be updated so that their type
234 * matches the newly assigned type of the variable they are accessing. */
235 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
237 ir
->type
= ir
->var
->type
;
238 return visit_continue
;
241 /* Dereferences of 2D input arrays need to be updated so that their type
242 * matches the newly assigned type of the array they are accessing. */
243 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
245 const glsl_type
*const vt
= ir
->array
->type
;
247 ir
->type
= vt
->fields
.array
;
248 return visit_continue
;
252 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
254 unsigned num_vertices
;
255 gl_shader_program
*prog
;
257 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
259 this->num_vertices
= num_vertices
;
263 virtual ~tess_eval_array_resize_visitor()
268 virtual ir_visitor_status
visit(ir_variable
*var
)
270 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
271 return visit_continue
;
273 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
275 var
->data
.max_array_access
= this->num_vertices
- 1;
277 return visit_continue
;
280 /* Dereferences of input variables need to be updated so that their type
281 * matches the newly assigned type of the variable they are accessing. */
282 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
284 ir
->type
= ir
->var
->type
;
285 return visit_continue
;
288 /* Dereferences of 2D input arrays need to be updated so that their type
289 * matches the newly assigned type of the array they are accessing. */
290 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
292 const glsl_type
*const vt
= ir
->array
->type
;
294 ir
->type
= vt
->fields
.array
;
295 return visit_continue
;
299 class barrier_use_visitor
: public ir_hierarchical_visitor
{
301 barrier_use_visitor(gl_shader_program
*prog
)
302 : prog(prog
), in_main(false), after_return(false), control_flow(0)
306 virtual ~barrier_use_visitor()
311 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
313 if (strcmp(ir
->name
, "main") == 0)
316 return visit_continue
;
319 virtual ir_visitor_status
visit_leave(ir_function
*)
322 after_return
= false;
323 return visit_continue
;
326 virtual ir_visitor_status
visit_leave(ir_return
*)
329 return visit_continue
;
332 virtual ir_visitor_status
visit_enter(ir_if
*)
335 return visit_continue
;
338 virtual ir_visitor_status
visit_leave(ir_if
*)
341 return visit_continue
;
344 virtual ir_visitor_status
visit_enter(ir_loop
*)
347 return visit_continue
;
350 virtual ir_visitor_status
visit_leave(ir_loop
*)
353 return visit_continue
;
356 /* FINISHME: `switch` is not expressed at the IR level -- it's already
357 * been lowered to a mess of `if`s. We'll correctly disallow any use of
358 * barrier() in a conditional path within the switch, but not in a path
359 * which is always hit.
362 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
364 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
365 /* Use of barrier(); determine if it is legal: */
367 linker_error(prog
, "Builtin barrier() may only be used in main");
372 linker_error(prog
, "Builtin barrier() may not be used after return");
376 if (control_flow
!= 0) {
377 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
381 return visit_continue
;
385 gl_shader_program
*prog
;
386 bool in_main
, after_return
;
391 * Visitor that determines the highest stream id to which a (geometry) shader
392 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
394 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
396 find_emit_vertex_visitor(int max_allowed
)
397 : max_stream_allowed(max_allowed
),
398 invalid_stream_id(0),
399 invalid_stream_id_from_emit_vertex(false),
400 end_primitive_found(false),
401 uses_non_zero_stream(false)
406 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
408 int stream_id
= ir
->stream_id();
411 invalid_stream_id
= stream_id
;
412 invalid_stream_id_from_emit_vertex
= true;
416 if (stream_id
> max_stream_allowed
) {
417 invalid_stream_id
= stream_id
;
418 invalid_stream_id_from_emit_vertex
= true;
423 uses_non_zero_stream
= true;
425 return visit_continue
;
428 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
430 end_primitive_found
= true;
432 int stream_id
= ir
->stream_id();
435 invalid_stream_id
= stream_id
;
436 invalid_stream_id_from_emit_vertex
= false;
440 if (stream_id
> max_stream_allowed
) {
441 invalid_stream_id
= stream_id
;
442 invalid_stream_id_from_emit_vertex
= false;
447 uses_non_zero_stream
= true;
449 return visit_continue
;
454 return invalid_stream_id
!= 0;
457 const char *error_func()
459 return invalid_stream_id_from_emit_vertex
?
460 "EmitStreamVertex" : "EndStreamPrimitive";
465 return invalid_stream_id
;
470 return uses_non_zero_stream
;
473 bool uses_end_primitive()
475 return end_primitive_found
;
479 int max_stream_allowed
;
480 int invalid_stream_id
;
481 bool invalid_stream_id_from_emit_vertex
;
482 bool end_primitive_found
;
483 bool uses_non_zero_stream
;
486 /* Class that finds array derefs and check if indexes are dynamic. */
487 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
490 dynamic_sampler_array_indexing_visitor() :
491 dynamic_sampler_array_indexing(false)
495 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
497 if (!ir
->variable_referenced())
498 return visit_continue
;
500 if (!ir
->variable_referenced()->type
->contains_sampler())
501 return visit_continue
;
503 if (!ir
->array_index
->constant_expression_value()) {
504 dynamic_sampler_array_indexing
= true;
507 return visit_continue
;
510 bool uses_dynamic_sampler_array_indexing()
512 return dynamic_sampler_array_indexing
;
516 bool dynamic_sampler_array_indexing
;
519 } /* anonymous namespace */
522 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
526 ralloc_strcat(&prog
->InfoLog
, "error: ");
528 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
531 prog
->LinkStatus
= false;
536 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
540 ralloc_strcat(&prog
->InfoLog
, "warning: ");
542 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
549 * Given a string identifying a program resource, break it into a base name
550 * and an optional array index in square brackets.
552 * If an array index is present, \c out_base_name_end is set to point to the
553 * "[" that precedes the array index, and the array index itself is returned
556 * If no array index is present (or if the array index is negative or
557 * mal-formed), \c out_base_name_end, is set to point to the null terminator
558 * at the end of the input string, and -1 is returned.
560 * Only the final array index is parsed; if the string contains other array
561 * indices (or structure field accesses), they are left in the base name.
563 * No attempt is made to check that the base name is properly formed;
564 * typically the caller will look up the base name in a hash table, so
565 * ill-formed base names simply turn into hash table lookup failures.
568 parse_program_resource_name(const GLchar
*name
,
569 const GLchar
**out_base_name_end
)
571 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
573 * "When an integer array element or block instance number is part of
574 * the name string, it will be specified in decimal form without a "+"
575 * or "-" sign or any extra leading zeroes. Additionally, the name
576 * string will not include white space anywhere in the string."
579 const size_t len
= strlen(name
);
580 *out_base_name_end
= name
+ len
;
582 if (len
== 0 || name
[len
-1] != ']')
585 /* Walk backwards over the string looking for a non-digit character. This
586 * had better be the opening bracket for an array index.
588 * Initially, i specifies the location of the ']'. Since the string may
589 * contain only the ']' charcater, walk backwards very carefully.
592 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
595 if ((i
== 0) || name
[i
-1] != '[')
598 long array_index
= strtol(&name
[i
], NULL
, 10);
602 /* Check for leading zero */
603 if (name
[i
] == '0' && name
[i
+1] != ']')
606 *out_base_name_end
= name
+ (i
- 1);
612 link_invalidate_variable_locations(exec_list
*ir
)
614 foreach_in_list(ir_instruction
, node
, ir
) {
615 ir_variable
*const var
= node
->as_variable();
620 /* Only assign locations for variables that lack an explicit location.
621 * Explicit locations are set for all built-in variables, generic vertex
622 * shader inputs (via layout(location=...)), and generic fragment shader
623 * outputs (also via layout(location=...)).
625 if (!var
->data
.explicit_location
) {
626 var
->data
.location
= -1;
627 var
->data
.location_frac
= 0;
630 /* ir_variable::is_unmatched_generic_inout is used by the linker while
631 * connecting outputs from one stage to inputs of the next stage.
633 if (var
->data
.explicit_location
&&
634 var
->data
.location
< VARYING_SLOT_VAR0
) {
635 var
->data
.is_unmatched_generic_inout
= 0;
637 var
->data
.is_unmatched_generic_inout
= 1;
644 * Set clip_distance_array_size based on the given shader.
646 * Also check for errors based on incorrect usage of gl_ClipVertex and
649 * Return false if an error was reported.
652 analyze_clip_usage(struct gl_shader_program
*prog
,
653 struct gl_shader
*shader
,
654 GLuint
*clip_distance_array_size
)
656 *clip_distance_array_size
= 0;
658 if (!prog
->IsES
&& prog
->Version
>= 130) {
659 /* From section 7.1 (Vertex Shader Special Variables) of the
662 * "It is an error for a shader to statically write both
663 * gl_ClipVertex and gl_ClipDistance."
665 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
666 * gl_ClipVertex nor gl_ClipDistance.
668 find_assignment_visitor
clip_vertex("gl_ClipVertex");
669 find_assignment_visitor
clip_distance("gl_ClipDistance");
671 clip_vertex
.run(shader
->ir
);
672 clip_distance
.run(shader
->ir
);
673 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
674 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
675 "and `gl_ClipDistance'\n",
676 _mesa_shader_stage_to_string(shader
->Stage
));
680 if (clip_distance
.variable_found()) {
681 ir_variable
*clip_distance_var
=
682 shader
->symbols
->get_variable("gl_ClipDistance");
684 assert(clip_distance_var
);
685 *clip_distance_array_size
= clip_distance_var
->type
->length
;
692 * Verify that a vertex shader executable meets all semantic requirements.
694 * Also sets prog->Vert.ClipDistanceArraySize as a side effect.
696 * \param shader Vertex shader executable to be verified
699 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
700 struct gl_shader
*shader
)
705 /* From the GLSL 1.10 spec, page 48:
707 * "The variable gl_Position is available only in the vertex
708 * language and is intended for writing the homogeneous vertex
709 * position. All executions of a well-formed vertex shader
710 * executable must write a value into this variable. [...] The
711 * variable gl_Position is available only in the vertex
712 * language and is intended for writing the homogeneous vertex
713 * position. All executions of a well-formed vertex shader
714 * executable must write a value into this variable."
716 * while in GLSL 1.40 this text is changed to:
718 * "The variable gl_Position is available only in the vertex
719 * language and is intended for writing the homogeneous vertex
720 * position. It can be written at any time during shader
721 * execution. It may also be read back by a vertex shader
722 * after being written. This value will be used by primitive
723 * assembly, clipping, culling, and other fixed functionality
724 * operations, if present, that operate on primitives after
725 * vertex processing has occurred. Its value is undefined if
726 * the vertex shader executable does not write gl_Position."
728 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
729 * gl_Position is not an error.
731 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
732 find_assignment_visitor
find("gl_Position");
733 find
.run(shader
->ir
);
734 if (!find
.variable_found()) {
737 "vertex shader does not write to `gl_Position'."
738 "It's value is undefined. \n");
741 "vertex shader does not write to `gl_Position'. \n");
747 analyze_clip_usage(prog
, shader
, &prog
->Vert
.ClipDistanceArraySize
);
751 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
752 struct gl_shader
*shader
)
757 analyze_clip_usage(prog
, shader
, &prog
->TessEval
.ClipDistanceArraySize
);
762 * Verify that a fragment shader executable meets all semantic requirements
764 * \param shader Fragment shader executable to be verified
767 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
768 struct gl_shader
*shader
)
773 find_assignment_visitor
frag_color("gl_FragColor");
774 find_assignment_visitor
frag_data("gl_FragData");
776 frag_color
.run(shader
->ir
);
777 frag_data
.run(shader
->ir
);
779 if (frag_color
.variable_found() && frag_data
.variable_found()) {
780 linker_error(prog
, "fragment shader writes to both "
781 "`gl_FragColor' and `gl_FragData'\n");
786 * Verify that a geometry shader executable meets all semantic requirements
788 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize as
791 * \param shader Geometry shader executable to be verified
794 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
795 struct gl_shader
*shader
)
800 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
801 prog
->Geom
.VerticesIn
= num_vertices
;
803 analyze_clip_usage(prog
, shader
, &prog
->Geom
.ClipDistanceArraySize
);
807 * Check if geometry shaders emit to non-zero streams and do corresponding
811 validate_geometry_shader_emissions(struct gl_context
*ctx
,
812 struct gl_shader_program
*prog
)
814 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
815 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
816 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
817 if (emit_vertex
.error()) {
818 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
819 "stream parameter are in the range [0, %d].\n",
820 emit_vertex
.error_func(),
821 emit_vertex
.error_stream(),
822 ctx
->Const
.MaxVertexStreams
- 1);
824 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
825 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
827 /* From the ARB_gpu_shader5 spec:
829 * "Multiple vertex streams are supported only if the output primitive
830 * type is declared to be "points". A program will fail to link if it
831 * contains a geometry shader calling EmitStreamVertex() or
832 * EndStreamPrimitive() if its output primitive type is not "points".
834 * However, in the same spec:
836 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
837 * with <stream> set to zero."
841 * "The function EndPrimitive() is equivalent to calling
842 * EndStreamPrimitive() with <stream> set to zero."
844 * Since we can call EmitVertex() and EndPrimitive() when we output
845 * primitives other than points, calling EmitStreamVertex(0) or
846 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
847 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
848 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
851 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
852 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
853 "with n>0 requires point output\n");
859 validate_intrastage_arrays(struct gl_shader_program
*prog
,
860 ir_variable
*const var
,
861 ir_variable
*const existing
)
863 /* Consider the types to be "the same" if both types are arrays
864 * of the same type and one of the arrays is implicitly sized.
865 * In addition, set the type of the linked variable to the
866 * explicitly sized array.
868 if (var
->type
->is_array() && existing
->type
->is_array()) {
869 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
870 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
871 if (var
->type
->length
!= 0) {
872 if (var
->type
->length
<= existing
->data
.max_array_access
) {
873 linker_error(prog
, "%s `%s' declared as type "
874 "`%s' but outermost dimension has an index"
877 var
->name
, var
->type
->name
,
878 existing
->data
.max_array_access
);
880 existing
->type
= var
->type
;
882 } else if (existing
->type
->length
!= 0) {
883 if(existing
->type
->length
<= var
->data
.max_array_access
&&
884 !existing
->data
.from_ssbo_unsized_array
) {
885 linker_error(prog
, "%s `%s' declared as type "
886 "`%s' but outermost dimension has an index"
889 var
->name
, existing
->type
->name
,
890 var
->data
.max_array_access
);
895 /* The arrays of structs could have different glsl_type pointers but
896 * they are actually the same type. Use record_compare() to check that.
898 if (existing
->type
->fields
.array
->is_record() &&
899 var
->type
->fields
.array
->is_record() &&
900 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
909 * Perform validation of global variables used across multiple shaders
912 cross_validate_globals(struct gl_shader_program
*prog
,
913 struct gl_shader
**shader_list
,
914 unsigned num_shaders
,
917 /* Examine all of the uniforms in all of the shaders and cross validate
920 glsl_symbol_table variables
;
921 for (unsigned i
= 0; i
< num_shaders
; i
++) {
922 if (shader_list
[i
] == NULL
)
925 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
926 ir_variable
*const var
= node
->as_variable();
931 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
934 /* don't cross validate subroutine uniforms */
935 if (var
->type
->contains_subroutine())
938 /* Don't cross validate temporaries that are at global scope. These
939 * will eventually get pulled into the shaders 'main'.
941 if (var
->data
.mode
== ir_var_temporary
)
944 /* If a global with this name has already been seen, verify that the
945 * new instance has the same type. In addition, if the globals have
946 * initializers, the values of the initializers must be the same.
948 ir_variable
*const existing
= variables
.get_variable(var
->name
);
949 if (existing
!= NULL
) {
950 /* Check if types match. Interface blocks have some special
951 * rules so we handle those elsewhere.
953 if (var
->type
!= existing
->type
&&
954 !var
->is_interface_instance()) {
955 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
956 if (var
->type
->is_record() && existing
->type
->is_record()
957 && existing
->type
->record_compare(var
->type
)) {
958 existing
->type
= var
->type
;
960 /* If it is an unsized array in a Shader Storage Block,
961 * two different shaders can access to different elements.
962 * Because of that, they might be converted to different
963 * sized arrays, then check that they are compatible but
964 * ignore the array size.
966 if (!(var
->data
.mode
== ir_var_shader_storage
&&
967 var
->data
.from_ssbo_unsized_array
&&
968 existing
->data
.mode
== ir_var_shader_storage
&&
969 existing
->data
.from_ssbo_unsized_array
&&
970 var
->type
->gl_type
== existing
->type
->gl_type
)) {
971 linker_error(prog
, "%s `%s' declared as type "
972 "`%s' and type `%s'\n",
974 var
->name
, var
->type
->name
,
975 existing
->type
->name
);
982 if (var
->data
.explicit_location
) {
983 if (existing
->data
.explicit_location
984 && (var
->data
.location
!= existing
->data
.location
)) {
985 linker_error(prog
, "explicit locations for %s "
986 "`%s' have differing values\n",
987 mode_string(var
), var
->name
);
991 existing
->data
.location
= var
->data
.location
;
992 existing
->data
.explicit_location
= true;
994 /* Check if uniform with implicit location was marked explicit
995 * by earlier shader stage. If so, mark it explicit in this stage
996 * too to make sure later processing does not treat it as
999 if (existing
->data
.explicit_location
) {
1000 var
->data
.location
= existing
->data
.location
;
1001 var
->data
.explicit_location
= true;
1005 /* From the GLSL 4.20 specification:
1006 * "A link error will result if two compilation units in a program
1007 * specify different integer-constant bindings for the same
1008 * opaque-uniform name. However, it is not an error to specify a
1009 * binding on some but not all declarations for the same name"
1011 if (var
->data
.explicit_binding
) {
1012 if (existing
->data
.explicit_binding
&&
1013 var
->data
.binding
!= existing
->data
.binding
) {
1014 linker_error(prog
, "explicit bindings for %s "
1015 "`%s' have differing values\n",
1016 mode_string(var
), var
->name
);
1020 existing
->data
.binding
= var
->data
.binding
;
1021 existing
->data
.explicit_binding
= true;
1024 if (var
->type
->contains_atomic() &&
1025 var
->data
.offset
!= existing
->data
.offset
) {
1026 linker_error(prog
, "offset specifications for %s "
1027 "`%s' have differing values\n",
1028 mode_string(var
), var
->name
);
1032 /* Validate layout qualifiers for gl_FragDepth.
1034 * From the AMD/ARB_conservative_depth specs:
1036 * "If gl_FragDepth is redeclared in any fragment shader in a
1037 * program, it must be redeclared in all fragment shaders in
1038 * that program that have static assignments to
1039 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1040 * fragment shaders in a single program must have the same set
1043 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1044 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1045 bool layout_differs
=
1046 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1048 if (layout_declared
&& layout_differs
) {
1050 "All redeclarations of gl_FragDepth in all "
1051 "fragment shaders in a single program must have "
1052 "the same set of qualifiers.\n");
1055 if (var
->data
.used
&& layout_differs
) {
1057 "If gl_FragDepth is redeclared with a layout "
1058 "qualifier in any fragment shader, it must be "
1059 "redeclared with the same layout qualifier in "
1060 "all fragment shaders that have assignments to "
1065 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1067 * "If a shared global has multiple initializers, the
1068 * initializers must all be constant expressions, and they
1069 * must all have the same value. Otherwise, a link error will
1070 * result. (A shared global having only one initializer does
1071 * not require that initializer to be a constant expression.)"
1073 * Previous to 4.20 the GLSL spec simply said that initializers
1074 * must have the same value. In this case of non-constant
1075 * initializers, this was impossible to determine. As a result,
1076 * no vendor actually implemented that behavior. The 4.20
1077 * behavior matches the implemented behavior of at least one other
1078 * vendor, so we'll implement that for all GLSL versions.
1080 if (var
->constant_initializer
!= NULL
) {
1081 if (existing
->constant_initializer
!= NULL
) {
1082 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1083 linker_error(prog
, "initializers for %s "
1084 "`%s' have differing values\n",
1085 mode_string(var
), var
->name
);
1089 /* If the first-seen instance of a particular uniform did not
1090 * have an initializer but a later instance does, copy the
1091 * initializer to the version stored in the symbol table.
1093 /* FINISHME: This is wrong. The constant_value field should
1094 * FINISHME: not be modified! Imagine a case where a shader
1095 * FINISHME: without an initializer is linked in two different
1096 * FINISHME: programs with shaders that have differing
1097 * FINISHME: initializers. Linking with the first will
1098 * FINISHME: modify the shader, and linking with the second
1099 * FINISHME: will fail.
1101 existing
->constant_initializer
=
1102 var
->constant_initializer
->clone(ralloc_parent(existing
),
1107 if (var
->data
.has_initializer
) {
1108 if (existing
->data
.has_initializer
1109 && (var
->constant_initializer
== NULL
1110 || existing
->constant_initializer
== NULL
)) {
1112 "shared global variable `%s' has multiple "
1113 "non-constant initializers.\n",
1118 /* Some instance had an initializer, so keep track of that. In
1119 * this location, all sorts of initializers (constant or
1120 * otherwise) will propagate the existence to the variable
1121 * stored in the symbol table.
1123 existing
->data
.has_initializer
= true;
1126 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1127 linker_error(prog
, "declarations for %s `%s' have "
1128 "mismatching invariant qualifiers\n",
1129 mode_string(var
), var
->name
);
1132 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1133 linker_error(prog
, "declarations for %s `%s' have "
1134 "mismatching centroid qualifiers\n",
1135 mode_string(var
), var
->name
);
1138 if (existing
->data
.sample
!= var
->data
.sample
) {
1139 linker_error(prog
, "declarations for %s `%s` have "
1140 "mismatching sample qualifiers\n",
1141 mode_string(var
), var
->name
);
1144 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1145 linker_error(prog
, "declarations for %s `%s` have "
1146 "mismatching image format qualifiers\n",
1147 mode_string(var
), var
->name
);
1151 variables
.add_variable(var
);
1158 * Perform validation of uniforms used across multiple shader stages
1161 cross_validate_uniforms(struct gl_shader_program
*prog
)
1163 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1164 MESA_SHADER_STAGES
, true);
1168 * Accumulates the array of prog->BufferInterfaceBlocks and checks that all
1169 * definitons of blocks agree on their contents.
1172 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1174 unsigned max_num_uniform_blocks
= 0;
1175 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1176 if (prog
->_LinkedShaders
[i
])
1177 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumBufferInterfaceBlocks
;
1180 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1181 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1183 prog
->InterfaceBlockStageIndex
[i
] = ralloc_array(prog
, int,
1184 max_num_uniform_blocks
);
1185 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1186 prog
->InterfaceBlockStageIndex
[i
][j
] = -1;
1191 for (unsigned int j
= 0; j
< sh
->NumBufferInterfaceBlocks
; j
++) {
1192 int index
= link_cross_validate_uniform_block(prog
,
1193 &prog
->BufferInterfaceBlocks
,
1194 &prog
->NumBufferInterfaceBlocks
,
1195 &sh
->BufferInterfaceBlocks
[j
]);
1198 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1199 sh
->BufferInterfaceBlocks
[j
].Name
);
1203 prog
->InterfaceBlockStageIndex
[i
][index
] = j
;
1212 * Populates a shaders symbol table with all global declarations
1215 populate_symbol_table(gl_shader
*sh
)
1217 sh
->symbols
= new(sh
) glsl_symbol_table
;
1219 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1223 if ((func
= inst
->as_function()) != NULL
) {
1224 sh
->symbols
->add_function(func
);
1225 } else if ((var
= inst
->as_variable()) != NULL
) {
1226 if (var
->data
.mode
!= ir_var_temporary
)
1227 sh
->symbols
->add_variable(var
);
1234 * Remap variables referenced in an instruction tree
1236 * This is used when instruction trees are cloned from one shader and placed in
1237 * another. These trees will contain references to \c ir_variable nodes that
1238 * do not exist in the target shader. This function finds these \c ir_variable
1239 * references and replaces the references with matching variables in the target
1242 * If there is no matching variable in the target shader, a clone of the
1243 * \c ir_variable is made and added to the target shader. The new variable is
1244 * added to \b both the instruction stream and the symbol table.
1246 * \param inst IR tree that is to be processed.
1247 * \param symbols Symbol table containing global scope symbols in the
1249 * \param instructions Instruction stream where new variable declarations
1253 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1256 class remap_visitor
: public ir_hierarchical_visitor
{
1258 remap_visitor(struct gl_shader
*target
,
1261 this->target
= target
;
1262 this->symbols
= target
->symbols
;
1263 this->instructions
= target
->ir
;
1264 this->temps
= temps
;
1267 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1269 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1270 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1272 assert(var
!= NULL
);
1274 return visit_continue
;
1277 ir_variable
*const existing
=
1278 this->symbols
->get_variable(ir
->var
->name
);
1279 if (existing
!= NULL
)
1282 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1284 this->symbols
->add_variable(copy
);
1285 this->instructions
->push_head(copy
);
1289 return visit_continue
;
1293 struct gl_shader
*target
;
1294 glsl_symbol_table
*symbols
;
1295 exec_list
*instructions
;
1299 remap_visitor
v(target
, temps
);
1306 * Move non-declarations from one instruction stream to another
1308 * The intended usage pattern of this function is to pass the pointer to the
1309 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1310 * pointer) for \c last and \c false for \c make_copies on the first
1311 * call. Successive calls pass the return value of the previous call for
1312 * \c last and \c true for \c make_copies.
1314 * \param instructions Source instruction stream
1315 * \param last Instruction after which new instructions should be
1316 * inserted in the target instruction stream
1317 * \param make_copies Flag selecting whether instructions in \c instructions
1318 * should be copied (via \c ir_instruction::clone) into the
1319 * target list or moved.
1322 * The new "last" instruction in the target instruction stream. This pointer
1323 * is suitable for use as the \c last parameter of a later call to this
1327 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1328 bool make_copies
, gl_shader
*target
)
1330 hash_table
*temps
= NULL
;
1333 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1334 hash_table_pointer_compare
);
1336 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1337 if (inst
->as_function())
1340 ir_variable
*var
= inst
->as_variable();
1341 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1344 assert(inst
->as_assignment()
1346 || inst
->as_if() /* for initializers with the ?: operator */
1347 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1350 inst
= inst
->clone(target
, NULL
);
1353 hash_table_insert(temps
, inst
, var
);
1355 remap_variables(inst
, target
, temps
);
1360 last
->insert_after(inst
);
1365 hash_table_dtor(temps
);
1372 * This class is only used in link_intrastage_shaders() below but declaring
1373 * it inside that function leads to compiler warnings with some versions of
1376 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1378 array_sizing_visitor()
1379 : mem_ctx(ralloc_context(NULL
)),
1380 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1381 hash_table_pointer_compare
))
1385 ~array_sizing_visitor()
1387 hash_table_dtor(this->unnamed_interfaces
);
1388 ralloc_free(this->mem_ctx
);
1391 virtual ir_visitor_status
visit(ir_variable
*var
)
1393 const glsl_type
*type_without_array
;
1394 fixup_type(&var
->type
, var
->data
.max_array_access
,
1395 var
->data
.from_ssbo_unsized_array
);
1396 type_without_array
= var
->type
->without_array();
1397 if (var
->type
->is_interface()) {
1398 if (interface_contains_unsized_arrays(var
->type
)) {
1399 const glsl_type
*new_type
=
1400 resize_interface_members(var
->type
,
1401 var
->get_max_ifc_array_access(),
1402 var
->is_in_shader_storage_block());
1403 var
->type
= new_type
;
1404 var
->change_interface_type(new_type
);
1406 } else if (type_without_array
->is_interface()) {
1407 if (interface_contains_unsized_arrays(type_without_array
)) {
1408 const glsl_type
*new_type
=
1409 resize_interface_members(type_without_array
,
1410 var
->get_max_ifc_array_access(),
1411 var
->is_in_shader_storage_block());
1412 var
->change_interface_type(new_type
);
1413 var
->type
= update_interface_members_array(var
->type
, new_type
);
1415 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1416 /* Store a pointer to the variable in the unnamed_interfaces
1419 ir_variable
**interface_vars
= (ir_variable
**)
1420 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1421 if (interface_vars
== NULL
) {
1422 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1424 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1427 unsigned index
= ifc_type
->field_index(var
->name
);
1428 assert(index
< ifc_type
->length
);
1429 assert(interface_vars
[index
] == NULL
);
1430 interface_vars
[index
] = var
;
1432 return visit_continue
;
1436 * For each unnamed interface block that was discovered while running the
1437 * visitor, adjust the interface type to reflect the newly assigned array
1438 * sizes, and fix up the ir_variable nodes to point to the new interface
1441 void fixup_unnamed_interface_types()
1443 hash_table_call_foreach(this->unnamed_interfaces
,
1444 fixup_unnamed_interface_type
, NULL
);
1449 * If the type pointed to by \c type represents an unsized array, replace
1450 * it with a sized array whose size is determined by max_array_access.
1452 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1453 bool from_ssbo_unsized_array
)
1455 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1456 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1457 max_array_access
+ 1);
1458 assert(*type
!= NULL
);
1462 static const glsl_type
*
1463 update_interface_members_array(const glsl_type
*type
,
1464 const glsl_type
*new_interface_type
)
1466 const glsl_type
*element_type
= type
->fields
.array
;
1467 if (element_type
->is_array()) {
1468 const glsl_type
*new_array_type
=
1469 update_interface_members_array(element_type
, new_interface_type
);
1470 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1472 return glsl_type::get_array_instance(new_interface_type
,
1478 * Determine whether the given interface type contains unsized arrays (if
1479 * it doesn't, array_sizing_visitor doesn't need to process it).
1481 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1483 for (unsigned i
= 0; i
< type
->length
; i
++) {
1484 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1485 if (elem_type
->is_unsized_array())
1492 * Create a new interface type based on the given type, with unsized arrays
1493 * replaced by sized arrays whose size is determined by
1494 * max_ifc_array_access.
1496 static const glsl_type
*
1497 resize_interface_members(const glsl_type
*type
,
1498 const unsigned *max_ifc_array_access
,
1501 unsigned num_fields
= type
->length
;
1502 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1503 memcpy(fields
, type
->fields
.structure
,
1504 num_fields
* sizeof(*fields
));
1505 for (unsigned i
= 0; i
< num_fields
; i
++) {
1506 /* If SSBO last member is unsized array, we don't replace it by a sized
1509 if (is_ssbo
&& i
== (num_fields
- 1))
1510 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1513 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1516 glsl_interface_packing packing
=
1517 (glsl_interface_packing
) type
->interface_packing
;
1518 const glsl_type
*new_ifc_type
=
1519 glsl_type::get_interface_instance(fields
, num_fields
,
1520 packing
, type
->name
);
1522 return new_ifc_type
;
1525 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1528 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1529 ir_variable
**interface_vars
= (ir_variable
**) data
;
1530 unsigned num_fields
= ifc_type
->length
;
1531 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1532 memcpy(fields
, ifc_type
->fields
.structure
,
1533 num_fields
* sizeof(*fields
));
1534 bool interface_type_changed
= false;
1535 for (unsigned i
= 0; i
< num_fields
; i
++) {
1536 if (interface_vars
[i
] != NULL
&&
1537 fields
[i
].type
!= interface_vars
[i
]->type
) {
1538 fields
[i
].type
= interface_vars
[i
]->type
;
1539 interface_type_changed
= true;
1542 if (!interface_type_changed
) {
1546 glsl_interface_packing packing
=
1547 (glsl_interface_packing
) ifc_type
->interface_packing
;
1548 const glsl_type
*new_ifc_type
=
1549 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1552 for (unsigned i
= 0; i
< num_fields
; i
++) {
1553 if (interface_vars
[i
] != NULL
)
1554 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1559 * Memory context used to allocate the data in \c unnamed_interfaces.
1564 * Hash table from const glsl_type * to an array of ir_variable *'s
1565 * pointing to the ir_variables constituting each unnamed interface block.
1567 hash_table
*unnamed_interfaces
;
1572 * Performs the cross-validation of tessellation control shader vertices and
1573 * layout qualifiers for the attached tessellation control shaders,
1574 * and propagates them to the linked TCS and linked shader program.
1577 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1578 struct gl_shader
*linked_shader
,
1579 struct gl_shader
**shader_list
,
1580 unsigned num_shaders
)
1582 linked_shader
->TessCtrl
.VerticesOut
= 0;
1584 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1587 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1589 * "All tessellation control shader layout declarations in a program
1590 * must specify the same output patch vertex count. There must be at
1591 * least one layout qualifier specifying an output patch vertex count
1592 * in any program containing tessellation control shaders; however,
1593 * such a declaration is not required in all tessellation control
1597 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1598 struct gl_shader
*shader
= shader_list
[i
];
1600 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1601 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1602 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1603 linker_error(prog
, "tessellation control shader defined with "
1604 "conflicting output vertex count (%d and %d)\n",
1605 linked_shader
->TessCtrl
.VerticesOut
,
1606 shader
->TessCtrl
.VerticesOut
);
1609 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1613 /* Just do the intrastage -> interstage propagation right now,
1614 * since we already know we're in the right type of shader program
1617 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1618 linker_error(prog
, "tessellation control shader didn't declare "
1619 "vertices out layout qualifier\n");
1622 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1627 * Performs the cross-validation of tessellation evaluation shader
1628 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1629 * for the attached tessellation evaluation shaders, and propagates them
1630 * to the linked TES and linked shader program.
1633 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1634 struct gl_shader
*linked_shader
,
1635 struct gl_shader
**shader_list
,
1636 unsigned num_shaders
)
1638 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1639 linked_shader
->TessEval
.Spacing
= 0;
1640 linked_shader
->TessEval
.VertexOrder
= 0;
1641 linked_shader
->TessEval
.PointMode
= -1;
1643 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1646 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1648 * "At least one tessellation evaluation shader (compilation unit) in
1649 * a program must declare a primitive mode in its input layout.
1650 * Declaration vertex spacing, ordering, and point mode identifiers is
1651 * optional. It is not required that all tessellation evaluation
1652 * shaders in a program declare a primitive mode. If spacing or
1653 * vertex ordering declarations are omitted, the tessellation
1654 * primitive generator will use equal spacing or counter-clockwise
1655 * vertex ordering, respectively. If a point mode declaration is
1656 * omitted, the tessellation primitive generator will produce lines or
1657 * triangles according to the primitive mode."
1660 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1661 struct gl_shader
*shader
= shader_list
[i
];
1663 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1664 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1665 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1666 linker_error(prog
, "tessellation evaluation shader defined with "
1667 "conflicting input primitive modes.\n");
1670 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1673 if (shader
->TessEval
.Spacing
!= 0) {
1674 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1675 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1676 linker_error(prog
, "tessellation evaluation shader defined with "
1677 "conflicting vertex spacing.\n");
1680 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1683 if (shader
->TessEval
.VertexOrder
!= 0) {
1684 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1685 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1686 linker_error(prog
, "tessellation evaluation shader defined with "
1687 "conflicting ordering.\n");
1690 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1693 if (shader
->TessEval
.PointMode
!= -1) {
1694 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1695 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1696 linker_error(prog
, "tessellation evaluation shader defined with "
1697 "conflicting point modes.\n");
1700 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1705 /* Just do the intrastage -> interstage propagation right now,
1706 * since we already know we're in the right type of shader program
1709 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1711 "tessellation evaluation shader didn't declare input "
1712 "primitive modes.\n");
1715 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1717 if (linked_shader
->TessEval
.Spacing
== 0)
1718 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1719 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1721 if (linked_shader
->TessEval
.VertexOrder
== 0)
1722 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1723 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1725 if (linked_shader
->TessEval
.PointMode
== -1)
1726 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1727 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1732 * Performs the cross-validation of layout qualifiers specified in
1733 * redeclaration of gl_FragCoord for the attached fragment shaders,
1734 * and propagates them to the linked FS and linked shader program.
1737 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1738 struct gl_shader
*linked_shader
,
1739 struct gl_shader
**shader_list
,
1740 unsigned num_shaders
)
1742 linked_shader
->redeclares_gl_fragcoord
= false;
1743 linked_shader
->uses_gl_fragcoord
= false;
1744 linked_shader
->origin_upper_left
= false;
1745 linked_shader
->pixel_center_integer
= false;
1747 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1748 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1751 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1752 struct gl_shader
*shader
= shader_list
[i
];
1753 /* From the GLSL 1.50 spec, page 39:
1755 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1756 * it must be redeclared in all the fragment shaders in that program
1757 * that have a static use gl_FragCoord."
1759 if ((linked_shader
->redeclares_gl_fragcoord
1760 && !shader
->redeclares_gl_fragcoord
1761 && shader
->uses_gl_fragcoord
)
1762 || (shader
->redeclares_gl_fragcoord
1763 && !linked_shader
->redeclares_gl_fragcoord
1764 && linked_shader
->uses_gl_fragcoord
)) {
1765 linker_error(prog
, "fragment shader defined with conflicting "
1766 "layout qualifiers for gl_FragCoord\n");
1769 /* From the GLSL 1.50 spec, page 39:
1771 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1772 * single program must have the same set of qualifiers."
1774 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1775 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1776 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1777 linker_error(prog
, "fragment shader defined with conflicting "
1778 "layout qualifiers for gl_FragCoord\n");
1781 /* Update the linked shader state. Note that uses_gl_fragcoord should
1782 * accumulate the results. The other values should replace. If there
1783 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1784 * are already known to be the same.
1786 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1787 linked_shader
->redeclares_gl_fragcoord
=
1788 shader
->redeclares_gl_fragcoord
;
1789 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1790 || shader
->uses_gl_fragcoord
;
1791 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1792 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1795 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1800 * Performs the cross-validation of geometry shader max_vertices and
1801 * primitive type layout qualifiers for the attached geometry shaders,
1802 * and propagates them to the linked GS and linked shader program.
1805 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1806 struct gl_shader
*linked_shader
,
1807 struct gl_shader
**shader_list
,
1808 unsigned num_shaders
)
1810 linked_shader
->Geom
.VerticesOut
= 0;
1811 linked_shader
->Geom
.Invocations
= 0;
1812 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1813 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1815 /* No in/out qualifiers defined for anything but GLSL 1.50+
1816 * geometry shaders so far.
1818 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1821 /* From the GLSL 1.50 spec, page 46:
1823 * "All geometry shader output layout declarations in a program
1824 * must declare the same layout and same value for
1825 * max_vertices. There must be at least one geometry output
1826 * layout declaration somewhere in a program, but not all
1827 * geometry shaders (compilation units) are required to
1831 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1832 struct gl_shader
*shader
= shader_list
[i
];
1834 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1835 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1836 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1837 linker_error(prog
, "geometry shader defined with conflicting "
1841 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1844 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1845 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1846 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1847 linker_error(prog
, "geometry shader defined with conflicting "
1851 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1854 if (shader
->Geom
.VerticesOut
!= 0) {
1855 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1856 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1857 linker_error(prog
, "geometry shader defined with conflicting "
1858 "output vertex count (%d and %d)\n",
1859 linked_shader
->Geom
.VerticesOut
,
1860 shader
->Geom
.VerticesOut
);
1863 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1866 if (shader
->Geom
.Invocations
!= 0) {
1867 if (linked_shader
->Geom
.Invocations
!= 0 &&
1868 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1869 linker_error(prog
, "geometry shader defined with conflicting "
1870 "invocation count (%d and %d)\n",
1871 linked_shader
->Geom
.Invocations
,
1872 shader
->Geom
.Invocations
);
1875 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1879 /* Just do the intrastage -> interstage propagation right now,
1880 * since we already know we're in the right type of shader program
1883 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1885 "geometry shader didn't declare primitive input type\n");
1888 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1890 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1892 "geometry shader didn't declare primitive output type\n");
1895 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1897 if (linked_shader
->Geom
.VerticesOut
== 0) {
1899 "geometry shader didn't declare max_vertices\n");
1902 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1904 if (linked_shader
->Geom
.Invocations
== 0)
1905 linked_shader
->Geom
.Invocations
= 1;
1907 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1912 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1913 * qualifiers for the attached compute shaders, and propagate them to the
1914 * linked CS and linked shader program.
1917 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1918 struct gl_shader
*linked_shader
,
1919 struct gl_shader
**shader_list
,
1920 unsigned num_shaders
)
1922 for (int i
= 0; i
< 3; i
++)
1923 linked_shader
->Comp
.LocalSize
[i
] = 0;
1925 /* This function is called for all shader stages, but it only has an effect
1926 * for compute shaders.
1928 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1931 /* From the ARB_compute_shader spec, in the section describing local size
1934 * If multiple compute shaders attached to a single program object
1935 * declare local work-group size, the declarations must be identical;
1936 * otherwise a link-time error results. Furthermore, if a program
1937 * object contains any compute shaders, at least one must contain an
1938 * input layout qualifier specifying the local work sizes of the
1939 * program, or a link-time error will occur.
1941 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1942 struct gl_shader
*shader
= shader_list
[sh
];
1944 if (shader
->Comp
.LocalSize
[0] != 0) {
1945 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1946 for (int i
= 0; i
< 3; i
++) {
1947 if (linked_shader
->Comp
.LocalSize
[i
] !=
1948 shader
->Comp
.LocalSize
[i
]) {
1949 linker_error(prog
, "compute shader defined with conflicting "
1955 for (int i
= 0; i
< 3; i
++)
1956 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1960 /* Just do the intrastage -> interstage propagation right now,
1961 * since we already know we're in the right type of shader program
1964 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1965 linker_error(prog
, "compute shader didn't declare local size\n");
1968 for (int i
= 0; i
< 3; i
++)
1969 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1974 * Combine a group of shaders for a single stage to generate a linked shader
1977 * If this function is supplied a single shader, it is cloned, and the new
1978 * shader is returned.
1980 static struct gl_shader
*
1981 link_intrastage_shaders(void *mem_ctx
,
1982 struct gl_context
*ctx
,
1983 struct gl_shader_program
*prog
,
1984 struct gl_shader
**shader_list
,
1985 unsigned num_shaders
)
1987 struct gl_uniform_block
*uniform_blocks
= NULL
;
1989 /* Check that global variables defined in multiple shaders are consistent.
1991 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1992 if (!prog
->LinkStatus
)
1995 /* Check that interface blocks defined in multiple shaders are consistent.
1997 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1999 if (!prog
->LinkStatus
)
2002 /* Link up uniform blocks defined within this stage. */
2003 const unsigned num_uniform_blocks
=
2004 link_uniform_blocks(mem_ctx
, ctx
, prog
, shader_list
, num_shaders
,
2006 if (!prog
->LinkStatus
)
2009 /* Check that there is only a single definition of each function signature
2010 * across all shaders.
2012 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2013 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2014 ir_function
*const f
= node
->as_function();
2019 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2020 ir_function
*const other
=
2021 shader_list
[j
]->symbols
->get_function(f
->name
);
2023 /* If the other shader has no function (and therefore no function
2024 * signatures) with the same name, skip to the next shader.
2029 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2030 if (!sig
->is_defined
|| sig
->is_builtin())
2033 ir_function_signature
*other_sig
=
2034 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2036 if ((other_sig
!= NULL
) && other_sig
->is_defined
2037 && !other_sig
->is_builtin()) {
2038 linker_error(prog
, "function `%s' is multiply defined\n",
2047 /* Find the shader that defines main, and make a clone of it.
2049 * Starting with the clone, search for undefined references. If one is
2050 * found, find the shader that defines it. Clone the reference and add
2051 * it to the shader. Repeat until there are no undefined references or
2052 * until a reference cannot be resolved.
2054 gl_shader
*main
= NULL
;
2055 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2056 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2057 main
= shader_list
[i
];
2063 linker_error(prog
, "%s shader lacks `main'\n",
2064 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2068 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2069 linked
->ir
= new(linked
) exec_list
;
2070 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2072 linked
->BufferInterfaceBlocks
= uniform_blocks
;
2073 linked
->NumBufferInterfaceBlocks
= num_uniform_blocks
;
2074 ralloc_steal(linked
, linked
->BufferInterfaceBlocks
);
2076 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2077 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2078 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2079 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2080 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2082 populate_symbol_table(linked
);
2084 /* The pointer to the main function in the final linked shader (i.e., the
2085 * copy of the original shader that contained the main function).
2087 ir_function_signature
*const main_sig
=
2088 _mesa_get_main_function_signature(linked
);
2090 /* Move any instructions other than variable declarations or function
2091 * declarations into main.
2093 exec_node
*insertion_point
=
2094 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2097 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2098 if (shader_list
[i
] == main
)
2101 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2102 insertion_point
, true, linked
);
2105 /* Check if any shader needs built-in functions. */
2106 bool need_builtins
= false;
2107 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2108 if (shader_list
[i
]->uses_builtin_functions
) {
2109 need_builtins
= true;
2115 if (need_builtins
) {
2116 /* Make a temporary array one larger than shader_list, which will hold
2117 * the built-in function shader as well.
2119 gl_shader
**linking_shaders
= (gl_shader
**)
2120 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2122 ok
= linking_shaders
!= NULL
;
2125 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2126 _mesa_glsl_initialize_builtin_functions();
2127 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2129 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2131 free(linking_shaders
);
2133 _mesa_error_no_memory(__func__
);
2136 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2141 _mesa_delete_shader(ctx
, linked
);
2145 /* At this point linked should contain all of the linked IR, so
2146 * validate it to make sure nothing went wrong.
2148 validate_ir_tree(linked
->ir
);
2150 /* Set the size of geometry shader input arrays */
2151 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2152 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2153 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2154 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2155 ir
->accept(&input_resize_visitor
);
2159 if (ctx
->Const
.VertexID_is_zero_based
)
2160 lower_vertex_id(linked
);
2162 /* Validate correct usage of barrier() in the tess control shader */
2163 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2164 barrier_use_visitor
visitor(prog
);
2165 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2166 ir
->accept(&visitor
);
2170 /* Make a pass over all variable declarations to ensure that arrays with
2171 * unspecified sizes have a size specified. The size is inferred from the
2172 * max_array_access field.
2174 array_sizing_visitor v
;
2176 v
.fixup_unnamed_interface_types();
2182 * Update the sizes of linked shader uniform arrays to the maximum
2185 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2187 * If one or more elements of an array are active,
2188 * GetActiveUniform will return the name of the array in name,
2189 * subject to the restrictions listed above. The type of the array
2190 * is returned in type. The size parameter contains the highest
2191 * array element index used, plus one. The compiler or linker
2192 * determines the highest index used. There will be only one
2193 * active uniform reported by the GL per uniform array.
2197 update_array_sizes(struct gl_shader_program
*prog
)
2199 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2200 if (prog
->_LinkedShaders
[i
] == NULL
)
2203 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2204 ir_variable
*const var
= node
->as_variable();
2206 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2207 !var
->type
->is_array())
2210 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2211 * will not be eliminated. Since we always do std140, just
2212 * don't resize arrays in UBOs.
2214 * Atomic counters are supposed to get deterministic
2215 * locations assigned based on the declaration ordering and
2216 * sizes, array compaction would mess that up.
2218 * Subroutine uniforms are not removed.
2220 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2221 var
->type
->contains_subroutine())
2224 unsigned int size
= var
->data
.max_array_access
;
2225 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2226 if (prog
->_LinkedShaders
[j
] == NULL
)
2229 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2230 ir_variable
*other_var
= node2
->as_variable();
2234 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2235 other_var
->data
.max_array_access
> size
) {
2236 size
= other_var
->data
.max_array_access
;
2241 if (size
+ 1 != var
->type
->length
) {
2242 /* If this is a built-in uniform (i.e., it's backed by some
2243 * fixed-function state), adjust the number of state slots to
2244 * match the new array size. The number of slots per array entry
2245 * is not known. It seems safe to assume that the total number of
2246 * slots is an integer multiple of the number of array elements.
2247 * Determine the number of slots per array element by dividing by
2248 * the old (total) size.
2250 const unsigned num_slots
= var
->get_num_state_slots();
2251 if (num_slots
> 0) {
2252 var
->set_num_state_slots((size
+ 1)
2253 * (num_slots
/ var
->type
->length
));
2256 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2258 /* FINISHME: We should update the types of array
2259 * dereferences of this variable now.
2267 * Resize tessellation evaluation per-vertex inputs to the size of
2268 * tessellation control per-vertex outputs.
2271 resize_tes_inputs(struct gl_context
*ctx
,
2272 struct gl_shader_program
*prog
)
2274 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2277 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2278 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2280 /* If no control shader is present, then the TES inputs are statically
2281 * sized to MaxPatchVertices; the actual size of the arrays won't be
2282 * known until draw time.
2284 const int num_vertices
= tcs
2285 ? tcs
->TessCtrl
.VerticesOut
2286 : ctx
->Const
.MaxPatchVertices
;
2288 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2289 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2290 ir
->accept(&input_resize_visitor
);
2294 /* Convert the gl_PatchVerticesIn system value into a constant, since
2295 * the value is known at this point.
2297 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2298 ir_variable
*var
= ir
->as_variable();
2299 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2300 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2301 void *mem_ctx
= ralloc_parent(var
);
2302 var
->data
.mode
= ir_var_auto
;
2303 var
->data
.location
= 0;
2304 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2311 * Find a contiguous set of available bits in a bitmask.
2313 * \param used_mask Bits representing used (1) and unused (0) locations
2314 * \param needed_count Number of contiguous bits needed.
2317 * Base location of the available bits on success or -1 on failure.
2320 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2322 unsigned needed_mask
= (1 << needed_count
) - 1;
2323 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2325 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2326 * cannot optimize possibly infinite loops" for the loop below.
2328 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2331 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2332 if ((needed_mask
& ~used_mask
) == needed_mask
)
2343 * Assign locations for either VS inputs or FS outputs
2345 * \param prog Shader program whose variables need locations assigned
2346 * \param constants Driver specific constant values for the program.
2347 * \param target_index Selector for the program target to receive location
2348 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2349 * \c MESA_SHADER_FRAGMENT.
2352 * If locations are successfully assigned, true is returned. Otherwise an
2353 * error is emitted to the shader link log and false is returned.
2356 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2357 struct gl_constants
*constants
,
2358 unsigned target_index
)
2360 /* Maximum number of generic locations. This corresponds to either the
2361 * maximum number of draw buffers or the maximum number of generic
2364 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2365 constants
->Program
[target_index
].MaxAttribs
:
2366 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2368 /* Mark invalid locations as being used.
2370 unsigned used_locations
= (max_index
>= 32)
2371 ? ~0 : ~((1 << max_index
) - 1);
2372 unsigned double_storage_locations
= 0;
2374 assert((target_index
== MESA_SHADER_VERTEX
)
2375 || (target_index
== MESA_SHADER_FRAGMENT
));
2377 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2381 /* Operate in a total of four passes.
2383 * 1. Invalidate the location assignments for all vertex shader inputs.
2385 * 2. Assign locations for inputs that have user-defined (via
2386 * glBindVertexAttribLocation) locations and outputs that have
2387 * user-defined locations (via glBindFragDataLocation).
2389 * 3. Sort the attributes without assigned locations by number of slots
2390 * required in decreasing order. Fragmentation caused by attribute
2391 * locations assigned by the application may prevent large attributes
2392 * from having enough contiguous space.
2394 * 4. Assign locations to any inputs without assigned locations.
2397 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2398 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2400 const enum ir_variable_mode direction
=
2401 (target_index
== MESA_SHADER_VERTEX
)
2402 ? ir_var_shader_in
: ir_var_shader_out
;
2405 /* Temporary storage for the set of attributes that need locations assigned.
2411 /* Used below in the call to qsort. */
2412 static int compare(const void *a
, const void *b
)
2414 const temp_attr
*const l
= (const temp_attr
*) a
;
2415 const temp_attr
*const r
= (const temp_attr
*) b
;
2417 /* Reversed because we want a descending order sort below. */
2418 return r
->slots
- l
->slots
;
2422 unsigned num_attr
= 0;
2424 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2425 ir_variable
*const var
= node
->as_variable();
2427 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2430 if (var
->data
.explicit_location
) {
2431 var
->data
.is_unmatched_generic_inout
= 0;
2432 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2433 || (var
->data
.location
< 0)) {
2435 "invalid explicit location %d specified for `%s'\n",
2436 (var
->data
.location
< 0)
2437 ? var
->data
.location
2438 : var
->data
.location
- generic_base
,
2442 } else if (target_index
== MESA_SHADER_VERTEX
) {
2445 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2446 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2447 var
->data
.location
= binding
;
2448 var
->data
.is_unmatched_generic_inout
= 0;
2450 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2454 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2455 assert(binding
>= FRAG_RESULT_DATA0
);
2456 var
->data
.location
= binding
;
2457 var
->data
.is_unmatched_generic_inout
= 0;
2459 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2460 var
->data
.index
= index
;
2465 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2467 * "Output binding assignments will cause LinkProgram to fail:
2469 * If the program has an active output assigned to a location greater
2470 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2471 * an active output assigned an index greater than or equal to one;"
2473 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2474 var
->data
.location
- generic_base
>=
2475 (int) constants
->MaxDualSourceDrawBuffers
) {
2477 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2478 "with index %u for %s\n",
2479 var
->data
.location
- generic_base
, var
->data
.index
,
2484 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
? true : false);
2486 /* If the variable is not a built-in and has a location statically
2487 * assigned in the shader (presumably via a layout qualifier), make sure
2488 * that it doesn't collide with other assigned locations. Otherwise,
2489 * add it to the list of variables that need linker-assigned locations.
2491 if (var
->data
.location
!= -1) {
2492 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2493 /* From page 61 of the OpenGL 4.0 spec:
2495 * "LinkProgram will fail if the attribute bindings assigned
2496 * by BindAttribLocation do not leave not enough space to
2497 * assign a location for an active matrix attribute or an
2498 * active attribute array, both of which require multiple
2499 * contiguous generic attributes."
2501 * I think above text prohibits the aliasing of explicit and
2502 * automatic assignments. But, aliasing is allowed in manual
2503 * assignments of attribute locations. See below comments for
2506 * From OpenGL 4.0 spec, page 61:
2508 * "It is possible for an application to bind more than one
2509 * attribute name to the same location. This is referred to as
2510 * aliasing. This will only work if only one of the aliased
2511 * attributes is active in the executable program, or if no
2512 * path through the shader consumes more than one attribute of
2513 * a set of attributes aliased to the same location. A link
2514 * error can occur if the linker determines that every path
2515 * through the shader consumes multiple aliased attributes,
2516 * but implementations are not required to generate an error
2519 * From GLSL 4.30 spec, page 54:
2521 * "A program will fail to link if any two non-vertex shader
2522 * input variables are assigned to the same location. For
2523 * vertex shaders, multiple input variables may be assigned
2524 * to the same location using either layout qualifiers or via
2525 * the OpenGL API. However, such aliasing is intended only to
2526 * support vertex shaders where each execution path accesses
2527 * at most one input per each location. Implementations are
2528 * permitted, but not required, to generate link-time errors
2529 * if they detect that every path through the vertex shader
2530 * executable accesses multiple inputs assigned to any single
2531 * location. For all shader types, a program will fail to link
2532 * if explicit location assignments leave the linker unable
2533 * to find space for other variables without explicit
2536 * From OpenGL ES 3.0 spec, page 56:
2538 * "Binding more than one attribute name to the same location
2539 * is referred to as aliasing, and is not permitted in OpenGL
2540 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2541 * fail when this condition exists. However, aliasing is
2542 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2543 * This will only work if only one of the aliased attributes
2544 * is active in the executable program, or if no path through
2545 * the shader consumes more than one attribute of a set of
2546 * attributes aliased to the same location. A link error can
2547 * occur if the linker determines that every path through the
2548 * shader consumes multiple aliased attributes, but implemen-
2549 * tations are not required to generate an error in this case."
2551 * After looking at above references from OpenGL, OpenGL ES and
2552 * GLSL specifications, we allow aliasing of vertex input variables
2553 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2555 * NOTE: This is not required by the spec but its worth mentioning
2556 * here that we're not doing anything to make sure that no path
2557 * through the vertex shader executable accesses multiple inputs
2558 * assigned to any single location.
2561 /* Mask representing the contiguous slots that will be used by
2564 const unsigned attr
= var
->data
.location
- generic_base
;
2565 const unsigned use_mask
= (1 << slots
) - 1;
2566 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2567 ? "vertex shader input" : "fragment shader output";
2569 /* Generate a link error if the requested locations for this
2570 * attribute exceed the maximum allowed attribute location.
2572 if (attr
+ slots
> max_index
) {
2574 "insufficient contiguous locations "
2575 "available for %s `%s' %d %d %d\n", string
,
2576 var
->name
, used_locations
, use_mask
, attr
);
2580 /* Generate a link error if the set of bits requested for this
2581 * attribute overlaps any previously allocated bits.
2583 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2584 if (target_index
== MESA_SHADER_FRAGMENT
||
2585 (prog
->IsES
&& prog
->Version
>= 300)) {
2587 "overlapping location is assigned "
2588 "to %s `%s' %d %d %d\n", string
,
2589 var
->name
, used_locations
, use_mask
, attr
);
2592 linker_warning(prog
,
2593 "overlapping location is assigned "
2594 "to %s `%s' %d %d %d\n", string
,
2595 var
->name
, used_locations
, use_mask
, attr
);
2599 used_locations
|= (use_mask
<< attr
);
2601 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2603 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2604 * active attribute variables may fail to link, unless
2605 * device-dependent optimizations are able to make the program
2606 * fit within available hardware resources. For the purposes
2607 * of this test, attribute variables of the type dvec3, dvec4,
2608 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2609 * count as consuming twice as many attributes as equivalent
2610 * single-precision types. While these types use the same number
2611 * of generic attributes as their single-precision equivalents,
2612 * implementations are permitted to consume two single-precision
2613 * vectors of internal storage for each three- or four-component
2614 * double-precision vector."
2616 * Mark this attribute slot as taking up twice as much space
2617 * so we can count it properly against limits. According to
2618 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2619 * is optional behavior, but it seems preferable.
2621 if (var
->type
->without_array()->is_dual_slot_double())
2622 double_storage_locations
|= (use_mask
<< attr
);
2628 to_assign
[num_attr
].slots
= slots
;
2629 to_assign
[num_attr
].var
= var
;
2633 if (target_index
== MESA_SHADER_VERTEX
) {
2634 unsigned total_attribs_size
=
2635 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2636 _mesa_bitcount(double_storage_locations
);
2637 if (total_attribs_size
> max_index
) {
2639 "attempt to use %d vertex attribute slots only %d available ",
2640 total_attribs_size
, max_index
);
2645 /* If all of the attributes were assigned locations by the application (or
2646 * are built-in attributes with fixed locations), return early. This should
2647 * be the common case.
2652 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2654 if (target_index
== MESA_SHADER_VERTEX
) {
2655 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2656 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2657 * reserved to prevent it from being automatically allocated below.
2659 find_deref_visitor
find("gl_Vertex");
2661 if (find
.variable_found())
2662 used_locations
|= (1 << 0);
2665 for (unsigned i
= 0; i
< num_attr
; i
++) {
2666 /* Mask representing the contiguous slots that will be used by this
2669 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2671 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2674 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2675 ? "vertex shader input" : "fragment shader output";
2678 "insufficient contiguous locations "
2679 "available for %s `%s'\n",
2680 string
, to_assign
[i
].var
->name
);
2684 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2685 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2686 used_locations
|= (use_mask
<< location
);
2693 * Match explicit locations of outputs to inputs and deactivate the
2694 * unmatch flag if found so we don't optimise them away.
2697 match_explicit_outputs_to_inputs(struct gl_shader_program
*prog
,
2698 gl_shader
*producer
,
2699 gl_shader
*consumer
)
2701 glsl_symbol_table parameters
;
2702 ir_variable
*explicit_locations
[MAX_VARYING
] = { NULL
};
2704 /* Find all shader outputs in the "producer" stage.
2706 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2707 ir_variable
*const var
= node
->as_variable();
2709 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2712 if (var
->data
.explicit_location
&&
2713 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2714 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2715 if (explicit_locations
[idx
] == NULL
)
2716 explicit_locations
[idx
] = var
;
2720 /* Match inputs to outputs */
2721 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2722 ir_variable
*const input
= node
->as_variable();
2724 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2727 ir_variable
*output
= NULL
;
2728 if (input
->data
.explicit_location
2729 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2730 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
];
2732 if (output
!= NULL
){
2733 input
->data
.is_unmatched_generic_inout
= 0;
2734 output
->data
.is_unmatched_generic_inout
= 0;
2741 * Store the gl_FragDepth layout in the gl_shader_program struct.
2744 store_fragdepth_layout(struct gl_shader_program
*prog
)
2746 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2750 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2752 /* We don't look up the gl_FragDepth symbol directly because if
2753 * gl_FragDepth is not used in the shader, it's removed from the IR.
2754 * However, the symbol won't be removed from the symbol table.
2756 * We're only interested in the cases where the variable is NOT removed
2759 foreach_in_list(ir_instruction
, node
, ir
) {
2760 ir_variable
*const var
= node
->as_variable();
2762 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2766 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2767 switch (var
->data
.depth_layout
) {
2768 case ir_depth_layout_none
:
2769 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2771 case ir_depth_layout_any
:
2772 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2774 case ir_depth_layout_greater
:
2775 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2777 case ir_depth_layout_less
:
2778 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2780 case ir_depth_layout_unchanged
:
2781 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2792 * Validate the resources used by a program versus the implementation limits
2795 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2797 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2798 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2803 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2804 linker_error(prog
, "Too many %s shader texture samplers\n",
2805 _mesa_shader_stage_to_string(i
));
2808 if (sh
->num_uniform_components
>
2809 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2810 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2811 linker_warning(prog
, "Too many %s shader default uniform block "
2812 "components, but the driver will try to optimize "
2813 "them out; this is non-portable out-of-spec "
2815 _mesa_shader_stage_to_string(i
));
2817 linker_error(prog
, "Too many %s shader default uniform block "
2819 _mesa_shader_stage_to_string(i
));
2823 if (sh
->num_combined_uniform_components
>
2824 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2825 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2826 linker_warning(prog
, "Too many %s shader uniform components, "
2827 "but the driver will try to optimize them out; "
2828 "this is non-portable out-of-spec behavior\n",
2829 _mesa_shader_stage_to_string(i
));
2831 linker_error(prog
, "Too many %s shader uniform components\n",
2832 _mesa_shader_stage_to_string(i
));
2837 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2838 unsigned total_uniform_blocks
= 0;
2839 unsigned shader_blocks
[MESA_SHADER_STAGES
] = {0};
2840 unsigned total_shader_storage_blocks
= 0;
2842 for (unsigned i
= 0; i
< prog
->NumBufferInterfaceBlocks
; i
++) {
2843 /* Don't check SSBOs for Uniform Block Size */
2844 if (!prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2845 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2846 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2847 prog
->BufferInterfaceBlocks
[i
].Name
,
2848 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2849 ctx
->Const
.MaxUniformBlockSize
);
2852 if (prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2853 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxShaderStorageBlockSize
) {
2854 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
2855 prog
->BufferInterfaceBlocks
[i
].Name
,
2856 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2857 ctx
->Const
.MaxShaderStorageBlockSize
);
2860 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2861 if (prog
->InterfaceBlockStageIndex
[j
][i
] != -1) {
2862 struct gl_shader
*sh
= prog
->_LinkedShaders
[j
];
2863 int stage_index
= prog
->InterfaceBlockStageIndex
[j
][i
];
2864 if (sh
&& sh
->BufferInterfaceBlocks
[stage_index
].IsShaderStorage
) {
2866 total_shader_storage_blocks
++;
2869 total_uniform_blocks
++;
2874 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2875 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2876 total_uniform_blocks
,
2877 ctx
->Const
.MaxCombinedUniformBlocks
);
2879 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2880 const unsigned max_uniform_blocks
=
2881 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2882 if (blocks
[i
] > max_uniform_blocks
) {
2883 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2884 _mesa_shader_stage_to_string(i
),
2886 max_uniform_blocks
);
2892 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
2893 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
2894 total_shader_storage_blocks
,
2895 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
2897 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2898 const unsigned max_shader_storage_blocks
=
2899 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
2900 if (shader_blocks
[i
] > max_shader_storage_blocks
) {
2901 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
2902 _mesa_shader_stage_to_string(i
),
2904 max_shader_storage_blocks
);
2913 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
2915 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2916 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2921 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
2922 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
2928 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
2929 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
2930 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
2931 if (fn
->types
[k
] == uni
->type
) {
2937 uni
->num_compatible_subroutines
= count
;
2943 check_subroutine_resources(struct gl_shader_program
*prog
)
2945 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2946 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2949 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
2950 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
2951 _mesa_shader_stage_to_string(i
));
2956 * Validate shader image resources.
2959 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2961 unsigned total_image_units
= 0;
2962 unsigned fragment_outputs
= 0;
2963 unsigned total_shader_storage_blocks
= 0;
2965 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2968 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2969 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2972 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2973 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
2974 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
2975 ctx
->Const
.Program
[i
].MaxImageUniforms
);
2977 total_image_units
+= sh
->NumImages
;
2979 for (unsigned j
= 0; j
< prog
->NumBufferInterfaceBlocks
; j
++) {
2980 int stage_index
= prog
->InterfaceBlockStageIndex
[i
][j
];
2981 if (stage_index
!= -1 && sh
->BufferInterfaceBlocks
[stage_index
].IsShaderStorage
)
2982 total_shader_storage_blocks
++;
2985 if (i
== MESA_SHADER_FRAGMENT
) {
2986 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2987 ir_variable
*var
= node
->as_variable();
2988 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2989 /* since there are no double fs outputs - pass false */
2990 fragment_outputs
+= var
->type
->count_attribute_slots(false);
2996 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2997 linker_error(prog
, "Too many combined image uniforms\n");
2999 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3000 ctx
->Const
.MaxCombinedShaderOutputResources
)
3001 linker_error(prog
, "Too many combined image uniforms, shader storage "
3002 " buffers and fragment outputs\n");
3007 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3008 * for a variable, checks for overlaps between other uniforms using explicit
3012 reserve_explicit_locations(struct gl_shader_program
*prog
,
3013 string_to_uint_map
*map
, ir_variable
*var
)
3015 unsigned slots
= var
->type
->uniform_locations();
3016 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3018 /* Resize remap table if locations do not fit in the current one. */
3019 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3020 prog
->UniformRemapTable
=
3021 reralloc(prog
, prog
->UniformRemapTable
,
3022 gl_uniform_storage
*,
3025 if (!prog
->UniformRemapTable
) {
3026 linker_error(prog
, "Out of memory during linking.\n");
3030 /* Initialize allocated space. */
3031 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3032 prog
->UniformRemapTable
[i
] = NULL
;
3034 prog
->NumUniformRemapTable
= max_loc
+ 1;
3037 for (unsigned i
= 0; i
< slots
; i
++) {
3038 unsigned loc
= var
->data
.location
+ i
;
3040 /* Check if location is already used. */
3041 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3043 /* Possibly same uniform from a different stage, this is ok. */
3045 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
3048 /* ARB_explicit_uniform_location specification states:
3050 * "No two default-block uniform variables in the program can have
3051 * the same location, even if they are unused, otherwise a compiler
3052 * or linker error will be generated."
3055 "location qualifier for uniform %s overlaps "
3056 "previously used location\n",
3061 /* Initialize location as inactive before optimization
3062 * rounds and location assignment.
3064 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3067 /* Note, base location used for arrays. */
3068 map
->put(var
->data
.location
, var
->name
);
3074 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3075 struct gl_shader
*sh
,
3078 unsigned slots
= var
->type
->uniform_locations();
3079 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3081 /* Resize remap table if locations do not fit in the current one. */
3082 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3083 sh
->SubroutineUniformRemapTable
=
3084 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3085 gl_uniform_storage
*,
3088 if (!sh
->SubroutineUniformRemapTable
) {
3089 linker_error(prog
, "Out of memory during linking.\n");
3093 /* Initialize allocated space. */
3094 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3095 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3097 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3100 for (unsigned i
= 0; i
< slots
; i
++) {
3101 unsigned loc
= var
->data
.location
+ i
;
3103 /* Check if location is already used. */
3104 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3106 /* ARB_explicit_uniform_location specification states:
3107 * "No two subroutine uniform variables can have the same location
3108 * in the same shader stage, otherwise a compiler or linker error
3109 * will be generated."
3112 "location qualifier for uniform %s overlaps "
3113 "previously used location\n",
3118 /* Initialize location as inactive before optimization
3119 * rounds and location assignment.
3121 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3127 * Check and reserve all explicit uniform locations, called before
3128 * any optimizations happen to handle also inactive uniforms and
3129 * inactive array elements that may get trimmed away.
3132 check_explicit_uniform_locations(struct gl_context
*ctx
,
3133 struct gl_shader_program
*prog
)
3135 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3138 /* This map is used to detect if overlapping explicit locations
3139 * occur with the same uniform (from different stage) or a different one.
3141 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3144 linker_error(prog
, "Out of memory during linking.\n");
3148 unsigned entries_total
= 0;
3149 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3150 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3155 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3156 ir_variable
*var
= node
->as_variable();
3157 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3160 entries_total
+= var
->type
->uniform_locations();
3162 if (var
->data
.explicit_location
) {
3164 if (var
->type
->without_array()->is_subroutine())
3165 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3167 ret
= reserve_explicit_locations(prog
, uniform_map
, var
);
3176 /* Verify that total amount of entries for explicit and implicit locations
3177 * is less than MAX_UNIFORM_LOCATIONS.
3179 if (entries_total
>= ctx
->Const
.MaxUserAssignableUniformLocations
) {
3180 linker_error(prog
, "count of uniform locations >= MAX_UNIFORM_LOCATIONS"
3181 "(%u >= %u)", entries_total
,
3182 ctx
->Const
.MaxUserAssignableUniformLocations
);
3188 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3189 GLenum type
, const char *name
)
3191 bool found_interface
= false;
3192 unsigned block_name_len
= 0;
3193 const char *block_name_dot
= strchr(name
, '.');
3195 /* These rules only apply to buffer variables. So we return
3196 * true for the rest of types.
3198 if (type
!= GL_BUFFER_VARIABLE
)
3201 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3202 const char *block_name
= shProg
->BufferInterfaceBlocks
[i
].Name
;
3203 block_name_len
= strlen(block_name
);
3205 const char *block_square_bracket
= strchr(block_name
, '[');
3206 if (block_square_bracket
) {
3207 /* The block is part of an array of named interfaces,
3208 * for the name comparison we ignore the "[x]" part.
3210 block_name_len
-= strlen(block_square_bracket
);
3213 if (block_name_dot
) {
3214 /* Check if the variable name starts with the interface
3215 * name. The interface name (if present) should have the
3216 * length than the interface block name we are comparing to.
3218 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3219 if (len
!= block_name_len
)
3223 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3224 found_interface
= true;
3229 /* We remove the interface name from the buffer variable name,
3230 * including the dot that follows it.
3232 if (found_interface
)
3233 name
= name
+ block_name_len
+ 1;
3235 /* From: ARB_program_interface_query extension:
3237 * "For an active shader storage block member declared as an array, an
3238 * entry will be generated only for the first array element, regardless
3239 * of its type. For arrays of aggregate types, the enumeration rules are
3240 * applied recursively for the single enumerated array element.
3242 const char *struct_first_dot
= strchr(name
, '.');
3243 const char *first_square_bracket
= strchr(name
, '[');
3245 /* The buffer variable is on top level and it is not an array */
3246 if (!first_square_bracket
) {
3248 /* The shader storage block member is a struct, then generate the entry */
3249 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3252 /* Shader storage block member is an array, only generate an entry for the
3253 * first array element.
3255 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3263 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3264 const void *data
, uint8_t stages
)
3268 /* If resource already exists, do not add it again. */
3269 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3270 if (prog
->ProgramResourceList
[i
].Data
== data
)
3273 prog
->ProgramResourceList
=
3275 prog
->ProgramResourceList
,
3276 gl_program_resource
,
3277 prog
->NumProgramResourceList
+ 1);
3279 if (!prog
->ProgramResourceList
) {
3280 linker_error(prog
, "Out of memory during linking.\n");
3284 struct gl_program_resource
*res
=
3285 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3289 res
->StageReferences
= stages
;
3291 prog
->NumProgramResourceList
++;
3296 /* Function checks if a variable var is a packed varying and
3297 * if given name is part of packed varying's list.
3299 * If a variable is a packed varying, it has a name like
3300 * 'packed:a,b,c' where a, b and c are separate variables.
3303 included_in_packed_varying(ir_variable
*var
, const char *name
)
3305 if (strncmp(var
->name
, "packed:", 7) != 0)
3308 char *list
= strdup(var
->name
+ 7);
3313 char *token
= strtok_r(list
, ",", &saveptr
);
3315 if (strcmp(token
, name
) == 0) {
3319 token
= strtok_r(NULL
, ",", &saveptr
);
3326 * Function builds a stage reference bitmask from variable name.
3329 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3334 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3335 * used for reference mask in gl_program_resource will need to be changed.
3337 assert(MESA_SHADER_STAGES
< 8);
3339 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3340 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3344 /* Shader symbol table may contain variables that have
3345 * been optimized away. Search IR for the variable instead.
3347 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3348 ir_variable
*var
= node
->as_variable();
3350 unsigned baselen
= strlen(var
->name
);
3352 if (included_in_packed_varying(var
, name
)) {
3357 /* Type needs to match if specified, otherwise we might
3358 * pick a variable with same name but different interface.
3360 if (var
->data
.mode
!= mode
)
3363 if (strncmp(var
->name
, name
, baselen
) == 0) {
3364 /* Check for exact name matches but also check for arrays and
3367 if (name
[baselen
] == '\0' ||
3368 name
[baselen
] == '[' ||
3369 name
[baselen
] == '.') {
3381 * Create gl_shader_variable from ir_variable class.
3383 static gl_shader_variable
*
3384 create_shader_variable(struct gl_shader_program
*shProg
, const ir_variable
*in
)
3386 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3390 out
->type
= in
->type
;
3391 out
->name
= ralloc_strdup(shProg
, in
->name
);
3396 out
->location
= in
->data
.location
;
3397 out
->index
= in
->data
.index
;
3398 out
->patch
= in
->data
.patch
;
3399 out
->mode
= in
->data
.mode
;
3405 add_interface_variables(struct gl_shader_program
*shProg
,
3406 exec_list
*ir
, GLenum programInterface
)
3408 foreach_in_list(ir_instruction
, node
, ir
) {
3409 ir_variable
*var
= node
->as_variable();
3415 switch (var
->data
.mode
) {
3416 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
3417 * "For GetActiveAttrib, all active vertex shader input variables
3418 * are enumerated, including the special built-in inputs gl_VertexID
3419 * and gl_InstanceID."
3421 case ir_var_system_value
:
3422 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
3423 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
3424 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
3426 /* Mark special built-in inputs referenced by the vertex stage so
3427 * that they are considered active by the shader queries.
3429 mask
= (1 << (MESA_SHADER_VERTEX
));
3431 case ir_var_shader_in
:
3432 if (programInterface
!= GL_PROGRAM_INPUT
)
3435 case ir_var_shader_out
:
3436 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3443 /* Skip packed varyings, packed varyings are handled separately
3444 * by add_packed_varyings.
3446 if (strncmp(var
->name
, "packed:", 7) == 0)
3449 /* Skip fragdata arrays, these are handled separately
3450 * by add_fragdata_arrays.
3452 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3455 gl_shader_variable
*sha_v
= create_shader_variable(shProg
, var
);
3459 if (!add_program_resource(shProg
, programInterface
, sha_v
,
3460 build_stageref(shProg
, sha_v
->name
,
3461 sha_v
->mode
) | mask
))
3468 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
, GLenum type
)
3470 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3473 if (!sh
|| !sh
->packed_varyings
)
3476 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3477 ir_variable
*var
= node
->as_variable();
3479 switch (var
->data
.mode
) {
3480 case ir_var_shader_in
:
3481 iface
= GL_PROGRAM_INPUT
;
3483 case ir_var_shader_out
:
3484 iface
= GL_PROGRAM_OUTPUT
;
3487 unreachable("unexpected type");
3490 if (type
== iface
) {
3491 gl_shader_variable
*sha_v
= create_shader_variable(shProg
, var
);
3494 if (!add_program_resource(shProg
, iface
, sha_v
,
3495 build_stageref(shProg
, sha_v
->name
,
3505 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3507 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3509 if (!sh
|| !sh
->fragdata_arrays
)
3512 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3513 ir_variable
*var
= node
->as_variable();
3515 assert(var
->data
.mode
== ir_var_shader_out
);
3516 gl_shader_variable
*sha_v
= create_shader_variable(shProg
, var
);
3519 if (!add_program_resource(shProg
, GL_PROGRAM_OUTPUT
, sha_v
,
3520 1 << MESA_SHADER_FRAGMENT
))
3528 get_top_level_name(const char *name
)
3530 const char *first_dot
= strchr(name
, '.');
3531 const char *first_square_bracket
= strchr(name
, '[');
3533 /* From ARB_program_interface_query spec:
3535 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying the
3536 * number of active array elements of the top-level shader storage block
3537 * member containing to the active variable is written to <params>. If the
3538 * top-level block member is not declared as an array, the value one is
3539 * written to <params>. If the top-level block member is an array with no
3540 * declared size, the value zero is written to <params>.
3543 /* The buffer variable is on top level.*/
3544 if (!first_square_bracket
&& !first_dot
)
3545 name_size
= strlen(name
);
3546 else if ((!first_square_bracket
||
3547 (first_dot
&& first_dot
< first_square_bracket
)))
3548 name_size
= first_dot
- name
;
3550 name_size
= first_square_bracket
- name
;
3552 return strndup(name
, name_size
);
3556 get_var_name(const char *name
)
3558 const char *first_dot
= strchr(name
, '.');
3561 return strdup(name
);
3563 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3567 is_top_level_shader_storage_block_member(const char* name
,
3568 const char* interface_name
,
3569 const char* field_name
)
3571 bool result
= false;
3573 /* If the given variable is already a top-level shader storage
3574 * block member, then return array_size = 1.
3575 * We could have two possibilities: if we have an instanced
3576 * shader storage block or not instanced.
3578 * For the first, we check create a name as it was in top level and
3579 * compare it with the real name. If they are the same, then
3580 * the variable is already at top-level.
3582 * Full instanced name is: interface name + '.' + var name +
3585 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3586 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3587 if (!full_instanced_name
) {
3588 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3592 snprintf(full_instanced_name
, name_length
, "%s.%s",
3593 interface_name
, field_name
);
3595 /* Check if its top-level shader storage block member of an
3596 * instanced interface block, or of a unnamed interface block.
3598 if (strcmp(name
, full_instanced_name
) == 0 ||
3599 strcmp(name
, field_name
) == 0)
3602 free(full_instanced_name
);
3607 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3608 char *interface_name
, char *var_name
)
3610 /* From GL_ARB_program_interface_query spec:
3612 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer
3613 * identifying the number of active array elements of the top-level
3614 * shader storage block member containing to the active variable is
3615 * written to <params>. If the top-level block member is not
3616 * declared as an array, the value one is written to <params>. If
3617 * the top-level block member is an array with no declared size,
3618 * the value zero is written to <params>.
3620 if (is_top_level_shader_storage_block_member(uni
->name
,
3624 else if (field
->type
->is_unsized_array())
3626 else if (field
->type
->is_array())
3627 return field
->type
->length
;
3633 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3634 const glsl_struct_field
*field
, char *interface_name
,
3637 /* From GL_ARB_program_interface_query:
3639 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3640 * identifying the stride between array elements of the top-level
3641 * shader storage block member containing the active variable is
3642 * written to <params>. For top-level block members declared as
3643 * arrays, the value written is the difference, in basic machine
3644 * units, between the offsets of the active variable for
3645 * consecutive elements in the top-level array. For top-level
3646 * block members not declared as an array, zero is written to
3649 if (field
->type
->is_array()) {
3650 const enum glsl_matrix_layout matrix_layout
=
3651 glsl_matrix_layout(field
->matrix_layout
);
3652 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3653 const glsl_type
*array_type
= field
->type
->fields
.array
;
3655 if (is_top_level_shader_storage_block_member(uni
->name
,
3660 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3661 if (array_type
->is_record() || array_type
->is_array())
3662 return glsl_align(array_type
->std140_size(row_major
), 16);
3664 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3666 return array_type
->std430_array_stride(row_major
);
3673 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3674 struct gl_uniform_storage
*uni
)
3676 int block_index
= uni
->block_index
;
3677 int array_size
= -1;
3678 int array_stride
= -1;
3679 char *var_name
= get_top_level_name(uni
->name
);
3680 char *interface_name
=
3681 get_top_level_name(shProg
->BufferInterfaceBlocks
[block_index
].Name
);
3683 if (strcmp(var_name
, interface_name
) == 0) {
3684 /* Deal with instanced array of SSBOs */
3685 char *temp_name
= get_var_name(uni
->name
);
3687 linker_error(shProg
, "Out of memory during linking.\n");
3688 goto write_top_level_array_size_and_stride
;
3691 var_name
= get_top_level_name(temp_name
);
3694 linker_error(shProg
, "Out of memory during linking.\n");
3695 goto write_top_level_array_size_and_stride
;
3699 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
3700 if (shProg
->Shaders
[i
] == NULL
)
3703 const gl_shader
*stage
= shProg
->Shaders
[i
];
3704 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
3705 ir_variable
*var
= node
->as_variable();
3706 if (!var
|| !var
->get_interface_type() ||
3707 var
->data
.mode
!= ir_var_shader_storage
)
3710 const glsl_type
*interface
= var
->get_interface_type();
3712 if (strcmp(interface_name
, interface
->name
) != 0)
3715 for (unsigned i
= 0; i
< interface
->length
; i
++) {
3716 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
3717 if (strcmp(field
->name
, var_name
) != 0)
3720 array_stride
= get_array_stride(uni
, interface
, field
,
3721 interface_name
, var_name
);
3722 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
3723 goto write_top_level_array_size_and_stride
;
3727 write_top_level_array_size_and_stride
:
3728 free(interface_name
);
3730 uni
->top_level_array_stride
= array_stride
;
3731 uni
->top_level_array_size
= array_size
;
3735 * Builds up a list of program resources that point to existing
3739 build_program_resource_list(struct gl_shader_program
*shProg
)
3741 /* Rebuild resource list. */
3742 if (shProg
->ProgramResourceList
) {
3743 ralloc_free(shProg
->ProgramResourceList
);
3744 shProg
->ProgramResourceList
= NULL
;
3745 shProg
->NumProgramResourceList
= 0;
3748 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
3750 /* Determine first input and final output stage. These are used to
3751 * detect which variables should be enumerated in the resource list
3752 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
3754 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3755 if (!shProg
->_LinkedShaders
[i
])
3757 if (input_stage
== MESA_SHADER_STAGES
)
3762 /* Empty shader, no resources. */
3763 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
3766 /* Program interface needs to expose varyings in case of SSO. */
3767 if (shProg
->SeparateShader
) {
3768 if (!add_packed_varyings(shProg
, input_stage
, GL_PROGRAM_INPUT
))
3771 if (!add_packed_varyings(shProg
, output_stage
, GL_PROGRAM_OUTPUT
))
3775 if (!add_fragdata_arrays(shProg
))
3778 /* Add inputs and outputs to the resource list. */
3779 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
]->ir
,
3783 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
]->ir
,
3787 /* Add transform feedback varyings. */
3788 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
3789 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
3790 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
3791 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
3797 /* Add uniforms from uniform storage. */
3798 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3799 /* Do not add uniforms internally used by Mesa. */
3800 if (shProg
->UniformStorage
[i
].hidden
)
3804 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
3807 /* Add stagereferences for uniforms in a uniform block. */
3808 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3809 if (block_index
!= -1) {
3810 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
3811 if (shProg
->InterfaceBlockStageIndex
[j
][block_index
] != -1)
3812 stageref
|= (1 << j
);
3816 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
3817 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
3818 if (!should_add_buffer_variable(shProg
, type
,
3819 shProg
->UniformStorage
[i
].name
))
3822 if (is_shader_storage
) {
3823 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
3826 if (!add_program_resource(shProg
, type
,
3827 &shProg
->UniformStorage
[i
], stageref
))
3831 /* Add program uniform blocks and shader storage blocks. */
3832 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3833 bool is_shader_storage
= shProg
->BufferInterfaceBlocks
[i
].IsShaderStorage
;
3834 GLenum type
= is_shader_storage
? GL_SHADER_STORAGE_BLOCK
: GL_UNIFORM_BLOCK
;
3835 if (!add_program_resource(shProg
, type
,
3836 &shProg
->BufferInterfaceBlocks
[i
], 0))
3840 /* Add atomic counter buffers. */
3841 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3842 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3843 &shProg
->AtomicBuffers
[i
], 0))
3847 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3849 if (!shProg
->UniformStorage
[i
].hidden
)
3852 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
3853 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
3854 !shProg
->UniformStorage
[i
].type
->is_subroutine())
3857 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
3858 /* add shader subroutines */
3859 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
3864 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3865 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3871 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
3872 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3873 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
3880 * This check is done to make sure we allow only constant expression
3881 * indexing and "constant-index-expression" (indexing with an expression
3882 * that includes loop induction variable).
3885 validate_sampler_array_indexing(struct gl_context
*ctx
,
3886 struct gl_shader_program
*prog
)
3888 dynamic_sampler_array_indexing_visitor v
;
3889 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3890 if (prog
->_LinkedShaders
[i
] == NULL
)
3893 bool no_dynamic_indexing
=
3894 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
3896 /* Search for array derefs in shader. */
3897 v
.run(prog
->_LinkedShaders
[i
]->ir
);
3898 if (v
.uses_dynamic_sampler_array_indexing()) {
3899 const char *msg
= "sampler arrays indexed with non-constant "
3900 "expressions is forbidden in GLSL %s %u";
3901 /* Backend has indicated that it has no dynamic indexing support. */
3902 if (no_dynamic_indexing
) {
3903 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3906 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3914 link_assign_subroutine_types(struct gl_shader_program
*prog
)
3916 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3917 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3922 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3923 ir_function
*fn
= node
->as_function();
3927 if (fn
->is_subroutine
)
3928 sh
->NumSubroutineUniformTypes
++;
3930 if (!fn
->num_subroutine_types
)
3933 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
3934 struct gl_subroutine_function
,
3935 sh
->NumSubroutineFunctions
+ 1);
3936 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
3937 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
3938 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
3939 ralloc_array(sh
, const struct glsl_type
*,
3940 fn
->num_subroutine_types
);
3942 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
3945 * "Each subroutine with an index qualifier in the shader must be
3946 * given a unique index, otherwise a compile or link error will be
3949 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3950 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
3951 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
3952 linker_error(prog
, "each subroutine index qualifier in the "
3953 "shader must be unique\n");
3957 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
3958 fn
->subroutine_index
;
3960 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
3961 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
3962 sh
->NumSubroutineFunctions
++;
3965 /* Assign index for subroutines without an explicit index*/
3967 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3968 while (sh
->SubroutineFunctions
[j
].index
== -1) {
3969 for (unsigned k
= 0; k
< sh
->NumSubroutineFunctions
; k
++) {
3970 if (sh
->SubroutineFunctions
[k
].index
== index
)
3972 else if (k
== sh
->NumSubroutineFunctions
- 1)
3973 sh
->SubroutineFunctions
[j
].index
= index
;
3982 split_ubos_and_ssbos(void *mem_ctx
,
3983 struct gl_uniform_block
*blocks
,
3984 unsigned num_blocks
,
3985 struct gl_uniform_block
***ubos
,
3987 unsigned **ubo_interface_block_indices
,
3988 struct gl_uniform_block
***ssbos
,
3989 unsigned *num_ssbos
,
3990 unsigned **ssbo_interface_block_indices
)
3992 unsigned num_ubo_blocks
= 0;
3993 unsigned num_ssbo_blocks
= 0;
3995 for (unsigned i
= 0; i
< num_blocks
; i
++) {
3996 if (blocks
[i
].IsShaderStorage
)
4002 *ubos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ubo_blocks
);
4005 *ssbos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ssbo_blocks
);
4008 if (ubo_interface_block_indices
)
4009 *ubo_interface_block_indices
=
4010 ralloc_array(mem_ctx
, unsigned, num_ubo_blocks
);
4012 if (ssbo_interface_block_indices
)
4013 *ssbo_interface_block_indices
=
4014 ralloc_array(mem_ctx
, unsigned, num_ssbo_blocks
);
4016 for (unsigned i
= 0; i
< num_blocks
; i
++) {
4017 if (blocks
[i
].IsShaderStorage
) {
4018 (*ssbos
)[*num_ssbos
] = &blocks
[i
];
4019 if (ssbo_interface_block_indices
)
4020 (*ssbo_interface_block_indices
)[*num_ssbos
] = i
;
4023 (*ubos
)[*num_ubos
] = &blocks
[i
];
4024 if (ubo_interface_block_indices
)
4025 (*ubo_interface_block_indices
)[*num_ubos
] = i
;
4030 assert(*num_ubos
+ *num_ssbos
== num_blocks
);
4034 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4036 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4038 foreach_in_list(ir_instruction
, node
, ir
) {
4039 ir_variable
*const var
= node
->as_variable();
4041 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4044 /* Don't set always active on builtins that haven't been redeclared */
4045 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4048 var
->data
.always_active_io
= true;
4053 * When separate shader programs are enabled, only input/outputs between
4054 * the stages of a multi-stage separate program can be safely removed
4055 * from the shader interface. Other inputs/outputs must remain active.
4058 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4060 unsigned first
, last
;
4061 assert(prog
->SeparateShader
);
4063 first
= MESA_SHADER_STAGES
;
4066 /* Determine first and last stage. Excluding the compute stage */
4067 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4068 if (!prog
->_LinkedShaders
[i
])
4070 if (first
== MESA_SHADER_STAGES
)
4075 if (first
== MESA_SHADER_STAGES
)
4078 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4079 gl_shader
*sh
= prog
->_LinkedShaders
[stage
];
4083 if (first
== last
) {
4084 /* For a single shader program only allow inputs to the vertex shader
4085 * and outputs from the fragment shader to be removed.
4087 if (stage
!= MESA_SHADER_VERTEX
)
4088 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4089 if (stage
!= MESA_SHADER_FRAGMENT
)
4090 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4092 /* For multi-stage separate shader programs only allow inputs and
4093 * outputs between the shader stages to be removed as well as inputs
4094 * to the vertex shader and outputs from the fragment shader.
4096 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4097 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4098 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4099 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4105 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4107 prog
->LinkStatus
= true; /* All error paths will set this to false */
4108 prog
->Validated
= false;
4109 prog
->_Used
= false;
4111 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4113 * "Linking can fail for a variety of reasons as specified in the
4114 * OpenGL Shading Language Specification, as well as any of the
4115 * following reasons:
4117 * - No shader objects are attached to program."
4119 * The Compatibility Profile specification does not list the error. In
4120 * Compatibility Profile missing shader stages are replaced by
4121 * fixed-function. This applies to the case where all stages are
4124 if (prog
->NumShaders
== 0) {
4125 if (ctx
->API
!= API_OPENGL_COMPAT
)
4126 linker_error(prog
, "no shaders attached to the program\n");
4130 tfeedback_decl
*tfeedback_decls
= NULL
;
4131 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4133 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4135 prog
->ARB_fragment_coord_conventions_enable
= false;
4137 /* Separate the shaders into groups based on their type.
4139 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4140 unsigned num_shaders
[MESA_SHADER_STAGES
];
4142 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4143 shader_list
[i
] = (struct gl_shader
**)
4144 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4148 unsigned min_version
= UINT_MAX
;
4149 unsigned max_version
= 0;
4150 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4151 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4152 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4154 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4155 linker_error(prog
, "all shaders must use same shading "
4156 "language version\n");
4160 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4161 prog
->ARB_fragment_coord_conventions_enable
= true;
4164 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4165 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4166 num_shaders
[shader_type
]++;
4169 /* In desktop GLSL, different shader versions may be linked together. In
4170 * GLSL ES, all shader versions must be the same.
4172 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4173 linker_error(prog
, "all shaders must use same shading "
4174 "language version\n");
4178 prog
->Version
= max_version
;
4179 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4181 /* Some shaders have to be linked with some other shaders present.
4183 if (!prog
->SeparateShader
) {
4184 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4185 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4186 linker_error(prog
, "Geometry shader must be linked with "
4190 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4191 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4192 linker_error(prog
, "Tessellation evaluation shader must be linked "
4193 "with vertex shader\n");
4196 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4197 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4198 linker_error(prog
, "Tessellation control shader must be linked with "
4203 /* The spec is self-contradictory here. It allows linking without a tess
4204 * eval shader, but that can only be used with transform feedback and
4205 * rasterization disabled. However, transform feedback isn't allowed
4206 * with GL_PATCHES, so it can't be used.
4208 * More investigation showed that the idea of transform feedback after
4209 * a tess control shader was dropped, because some hw vendors couldn't
4210 * support tessellation without a tess eval shader, but the linker
4211 * section wasn't updated to reflect that.
4213 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4216 * Do what's reasonable and always require a tess eval shader if a tess
4217 * control shader is present.
4219 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4220 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4221 linker_error(prog
, "Tessellation control shader must be linked with "
4222 "tessellation evaluation shader\n");
4227 /* Compute shaders have additional restrictions. */
4228 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4229 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4230 linker_error(prog
, "Compute shaders may not be linked with any other "
4231 "type of shader\n");
4234 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4235 if (prog
->_LinkedShaders
[i
] != NULL
)
4236 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4238 prog
->_LinkedShaders
[i
] = NULL
;
4241 /* Link all shaders for a particular stage and validate the result.
4243 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4244 if (num_shaders
[stage
] > 0) {
4245 gl_shader
*const sh
=
4246 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4247 num_shaders
[stage
]);
4249 if (!prog
->LinkStatus
) {
4251 _mesa_delete_shader(ctx
, sh
);
4256 case MESA_SHADER_VERTEX
:
4257 validate_vertex_shader_executable(prog
, sh
);
4259 case MESA_SHADER_TESS_CTRL
:
4260 /* nothing to be done */
4262 case MESA_SHADER_TESS_EVAL
:
4263 validate_tess_eval_shader_executable(prog
, sh
);
4265 case MESA_SHADER_GEOMETRY
:
4266 validate_geometry_shader_executable(prog
, sh
);
4268 case MESA_SHADER_FRAGMENT
:
4269 validate_fragment_shader_executable(prog
, sh
);
4272 if (!prog
->LinkStatus
) {
4274 _mesa_delete_shader(ctx
, sh
);
4278 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4282 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
4283 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4284 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
4285 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4286 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
4287 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4289 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4291 /* Here begins the inter-stage linking phase. Some initial validation is
4292 * performed, then locations are assigned for uniforms, attributes, and
4295 cross_validate_uniforms(prog
);
4296 if (!prog
->LinkStatus
)
4299 unsigned first
, last
, prev
;
4301 first
= MESA_SHADER_STAGES
;
4304 /* Determine first and last stage. */
4305 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4306 if (!prog
->_LinkedShaders
[i
])
4308 if (first
== MESA_SHADER_STAGES
)
4313 check_explicit_uniform_locations(ctx
, prog
);
4314 link_assign_subroutine_types(prog
);
4316 if (!prog
->LinkStatus
)
4319 resize_tes_inputs(ctx
, prog
);
4321 /* Validate the inputs of each stage with the output of the preceding
4325 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4326 if (prog
->_LinkedShaders
[i
] == NULL
)
4329 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4330 prog
->_LinkedShaders
[i
]);
4331 if (!prog
->LinkStatus
)
4334 cross_validate_outputs_to_inputs(prog
,
4335 prog
->_LinkedShaders
[prev
],
4336 prog
->_LinkedShaders
[i
]);
4337 if (!prog
->LinkStatus
)
4343 /* Cross-validate uniform blocks between shader stages */
4344 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4345 MESA_SHADER_STAGES
);
4346 if (!prog
->LinkStatus
)
4349 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4350 if (prog
->_LinkedShaders
[i
] != NULL
)
4351 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4354 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4355 * it before optimization because we want most of the checks to get
4356 * dropped thanks to constant propagation.
4358 * This rule also applies to GLSL ES 3.00.
4360 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4361 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4363 lower_discard_flow(sh
->ir
);
4367 if (prog
->SeparateShader
)
4368 disable_varying_optimizations_for_sso(prog
);
4370 if (!interstage_cross_validate_uniform_blocks(prog
))
4373 /* Do common optimization before assigning storage for attributes,
4374 * uniforms, and varyings. Later optimization could possibly make
4375 * some of that unused.
4377 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4378 if (prog
->_LinkedShaders
[i
] == NULL
)
4381 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4382 if (!prog
->LinkStatus
)
4385 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
4386 lower_clip_distance(prog
->_LinkedShaders
[i
]);
4389 if (ctx
->Const
.LowerTessLevel
) {
4390 lower_tess_level(prog
->_LinkedShaders
[i
]);
4393 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4394 &ctx
->Const
.ShaderCompilerOptions
[i
],
4395 ctx
->Const
.NativeIntegers
))
4398 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4401 /* Validation for special cases where we allow sampler array indexing
4402 * with loop induction variable. This check emits a warning or error
4403 * depending if backend can handle dynamic indexing.
4405 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4406 (prog
->IsES
&& prog
->Version
< 300)) {
4407 if (!validate_sampler_array_indexing(ctx
, prog
))
4411 /* Check and validate stream emissions in geometry shaders */
4412 validate_geometry_shader_emissions(ctx
, prog
);
4414 /* Mark all generic shader inputs and outputs as unpaired. */
4415 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4416 if (prog
->_LinkedShaders
[i
] != NULL
) {
4417 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4422 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4423 if (prog
->_LinkedShaders
[i
] == NULL
)
4426 match_explicit_outputs_to_inputs(prog
, prog
->_LinkedShaders
[prev
],
4427 prog
->_LinkedShaders
[i
]);
4431 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4432 MESA_SHADER_VERTEX
)) {
4436 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4437 MESA_SHADER_FRAGMENT
)) {
4441 if (num_tfeedback_decls
!= 0) {
4442 /* From GL_EXT_transform_feedback:
4443 * A program will fail to link if:
4445 * * the <count> specified by TransformFeedbackVaryingsEXT is
4446 * non-zero, but the program object has no vertex or geometry
4449 if (first
>= MESA_SHADER_FRAGMENT
) {
4450 linker_error(prog
, "Transform feedback varyings specified, but "
4451 "no vertex, tessellation, or geometry shader is "
4456 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4457 prog
->TransformFeedback
.NumVarying
);
4458 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4459 prog
->TransformFeedback
.VaryingNames
,
4464 /* If there is no fragment shader we need to set transform feedback.
4466 * For SSO we need also need to assign output locations, we assign them
4467 * here because we need to do it for both single stage programs and multi
4470 if (last
< MESA_SHADER_FRAGMENT
&&
4471 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4472 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4473 prog
->_LinkedShaders
[last
], NULL
,
4474 num_tfeedback_decls
, tfeedback_decls
))
4478 if (last
<= MESA_SHADER_FRAGMENT
) {
4479 /* Remove unused varyings from the first/last stage unless SSO */
4480 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4481 prog
->_LinkedShaders
[first
],
4483 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4484 prog
->_LinkedShaders
[last
],
4487 /* If the program is made up of only a single stage */
4488 if (first
== last
) {
4490 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4491 if (prog
->SeparateShader
) {
4492 /* Assign input locations for SSO, output locations are already
4495 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4496 NULL
/* producer */,
4498 0 /* num_tfeedback_decls */,
4499 NULL
/* tfeedback_decls */))
4503 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4504 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4507 /* Linking the stages in the opposite order (from fragment to vertex)
4508 * ensures that inter-shader outputs written to in an earlier stage
4509 * are eliminated if they are (transitively) not used in a later
4513 for (int i
= next
- 1; i
>= 0; i
--) {
4514 if (prog
->_LinkedShaders
[i
] == NULL
)
4517 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4518 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4520 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4521 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4525 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4526 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4529 /* This must be done after all dead varyings are eliminated. */
4530 if (!check_against_output_limit(ctx
, prog
, sh_i
))
4532 if (!check_against_input_limit(ctx
, prog
, sh_next
))
4540 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
4543 update_array_sizes(prog
);
4544 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
4545 link_assign_atomic_counter_resources(ctx
, prog
);
4546 store_fragdepth_layout(prog
);
4548 link_calculate_subroutine_compat(prog
);
4549 check_resources(ctx
, prog
);
4550 check_subroutine_resources(prog
);
4551 check_image_resources(ctx
, prog
);
4552 link_check_atomic_counter_resources(ctx
, prog
);
4554 if (!prog
->LinkStatus
)
4557 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4558 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4559 * anything about shader linking when one of the shaders (vertex or
4560 * fragment shader) is absent. So, the extension shouldn't change the
4561 * behavior specified in GLSL specification.
4563 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4564 * "Linking can fail for a variety of reasons as specified in the
4565 * OpenGL ES Shading Language Specification, as well as any of the
4566 * following reasons:
4570 * * program contains objects to form either a vertex shader or
4571 * fragment shader, and program is not separable, and does not
4572 * contain objects to form both a vertex shader and fragment
4575 * However, the only scenario in 3.1+ where we don't require them both is
4576 * when we have a compute shader. For example:
4578 * - No shaders is a link error.
4579 * - Geom or Tess without a Vertex shader is a link error which means we
4580 * always require a Vertex shader and hence a Fragment shader.
4581 * - Finally a Compute shader linked with any other stage is a link error.
4583 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4584 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4585 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4586 linker_error(prog
, "program lacks a vertex shader\n");
4587 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4588 linker_error(prog
, "program lacks a fragment shader\n");
4592 /* Split BufferInterfaceBlocks into UniformBlocks and ShaderStorageBlocks
4593 * for gl_shader_program and gl_shader, so that drivers that need separate
4594 * index spaces for each set can have that.
4596 for (unsigned i
= MESA_SHADER_VERTEX
; i
< MESA_SHADER_STAGES
; i
++) {
4597 if (prog
->_LinkedShaders
[i
] != NULL
) {
4598 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4599 split_ubos_and_ssbos(sh
,
4600 sh
->BufferInterfaceBlocks
,
4601 sh
->NumBufferInterfaceBlocks
,
4603 &sh
->NumUniformBlocks
,
4605 &sh
->ShaderStorageBlocks
,
4606 &sh
->NumShaderStorageBlocks
,
4611 split_ubos_and_ssbos(prog
,
4612 prog
->BufferInterfaceBlocks
,
4613 prog
->NumBufferInterfaceBlocks
,
4614 &prog
->UniformBlocks
,
4615 &prog
->NumUniformBlocks
,
4616 &prog
->UboInterfaceBlockIndex
,
4617 &prog
->ShaderStorageBlocks
,
4618 &prog
->NumShaderStorageBlocks
,
4619 &prog
->SsboInterfaceBlockIndex
);
4621 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4622 if (prog
->_LinkedShaders
[i
] == NULL
)
4625 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerBufferInterfaceBlocks
)
4626 lower_ubo_reference(prog
->_LinkedShaders
[i
]);
4628 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerShaderSharedVariables
)
4629 lower_shared_reference(prog
->_LinkedShaders
[i
],
4630 &prog
->Comp
.SharedSize
);
4632 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4636 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4637 free(shader_list
[i
]);
4638 if (prog
->_LinkedShaders
[i
] == NULL
)
4641 /* Do a final validation step to make sure that the IR wasn't
4642 * invalidated by any modifications performed after intrastage linking.
4644 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4646 /* Retain any live IR, but trash the rest. */
4647 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4649 /* The symbol table in the linked shaders may contain references to
4650 * variables that were removed (e.g., unused uniforms). Since it may
4651 * contain junk, there is no possible valid use. Delete it and set the
4654 delete prog
->_LinkedShaders
[i
]->symbols
;
4655 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4658 ralloc_free(mem_ctx
);