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/prog_instruction.h"
75 #include "program/program.h"
77 #include "util/string_to_uint_map.h"
79 #include "link_varyings.h"
80 #include "ir_optimization.h"
81 #include "ir_rvalue_visitor.h"
82 #include "ir_uniform.h"
83 #include "builtin_functions.h"
85 #include "main/shaderobj.h"
86 #include "main/enums.h"
92 * Visitor that determines whether or not a variable is ever written.
94 class find_assignment_visitor
: public ir_hierarchical_visitor
{
96 find_assignment_visitor(const char *name
)
97 : name(name
), found(false)
102 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
104 ir_variable
*const var
= ir
->lhs
->variable_referenced();
106 if (strcmp(name
, var
->name
) == 0) {
111 return visit_continue_with_parent
;
114 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
116 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
117 actual_node
, &ir
->actual_parameters
) {
118 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
119 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
121 if (sig_param
->data
.mode
== ir_var_function_out
||
122 sig_param
->data
.mode
== ir_var_function_inout
) {
123 ir_variable
*var
= param_rval
->variable_referenced();
124 if (var
&& strcmp(name
, var
->name
) == 0) {
131 if (ir
->return_deref
!= NULL
) {
132 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
134 if (strcmp(name
, var
->name
) == 0) {
140 return visit_continue_with_parent
;
143 bool variable_found()
149 const char *name
; /**< Find writes to a variable with this name. */
150 bool found
; /**< Was a write to the variable found? */
155 * Visitor that determines whether or not a variable is ever read.
157 class find_deref_visitor
: public ir_hierarchical_visitor
{
159 find_deref_visitor(const char *name
)
160 : name(name
), found(false)
165 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
167 if (strcmp(this->name
, ir
->var
->name
) == 0) {
172 return visit_continue
;
175 bool variable_found() const
181 const char *name
; /**< Find writes to a variable with this name. */
182 bool found
; /**< Was a write to the variable found? */
187 * A visitor helper that provides methods for updating the types of
188 * ir_dereferences. Classes that update variable types (say, updating
189 * array sizes) will want to use this so that dereference types stay in sync.
191 class deref_type_updater
: public ir_hierarchical_visitor
{
193 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
195 ir
->type
= ir
->var
->type
;
196 return visit_continue
;
199 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
201 const glsl_type
*const vt
= ir
->array
->type
;
203 ir
->type
= vt
->fields
.array
;
204 return visit_continue
;
207 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
209 for (unsigned i
= 0; i
< ir
->record
->type
->length
; i
++) {
210 const struct glsl_struct_field
*field
=
211 &ir
->record
->type
->fields
.structure
[i
];
212 if (strcmp(field
->name
, ir
->field
) == 0) {
213 ir
->type
= field
->type
;
217 return visit_continue
;
222 class array_resize_visitor
: public deref_type_updater
{
224 unsigned num_vertices
;
225 gl_shader_program
*prog
;
226 gl_shader_stage stage
;
228 array_resize_visitor(unsigned num_vertices
,
229 gl_shader_program
*prog
,
230 gl_shader_stage stage
)
232 this->num_vertices
= num_vertices
;
237 virtual ~array_resize_visitor()
242 virtual ir_visitor_status
visit(ir_variable
*var
)
244 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
246 return visit_continue
;
248 unsigned size
= var
->type
->length
;
250 if (stage
== MESA_SHADER_GEOMETRY
) {
251 /* Generate a link error if the shader has declared this array with
254 if (!var
->data
.implicit_sized_array
&&
255 size
&& size
!= this->num_vertices
) {
256 linker_error(this->prog
, "size of array %s declared as %u, "
257 "but number of input vertices is %u\n",
258 var
->name
, size
, this->num_vertices
);
259 return visit_continue
;
262 /* Generate a link error if the shader attempts to access an input
263 * array using an index too large for its actual size assigned at
266 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
267 linker_error(this->prog
, "%s shader accesses element %i of "
268 "%s, but only %i input vertices\n",
269 _mesa_shader_stage_to_string(this->stage
),
270 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
271 return visit_continue
;
275 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
277 var
->data
.max_array_access
= this->num_vertices
- 1;
279 return visit_continue
;
284 * Visitor that determines the highest stream id to which a (geometry) shader
285 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
287 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
289 find_emit_vertex_visitor(int max_allowed
)
290 : max_stream_allowed(max_allowed
),
291 invalid_stream_id(0),
292 invalid_stream_id_from_emit_vertex(false),
293 end_primitive_found(false),
294 uses_non_zero_stream(false)
299 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
301 int stream_id
= ir
->stream_id();
304 invalid_stream_id
= stream_id
;
305 invalid_stream_id_from_emit_vertex
= true;
309 if (stream_id
> max_stream_allowed
) {
310 invalid_stream_id
= stream_id
;
311 invalid_stream_id_from_emit_vertex
= true;
316 uses_non_zero_stream
= true;
318 return visit_continue
;
321 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
323 end_primitive_found
= true;
325 int stream_id
= ir
->stream_id();
328 invalid_stream_id
= stream_id
;
329 invalid_stream_id_from_emit_vertex
= false;
333 if (stream_id
> max_stream_allowed
) {
334 invalid_stream_id
= stream_id
;
335 invalid_stream_id_from_emit_vertex
= false;
340 uses_non_zero_stream
= true;
342 return visit_continue
;
347 return invalid_stream_id
!= 0;
350 const char *error_func()
352 return invalid_stream_id_from_emit_vertex
?
353 "EmitStreamVertex" : "EndStreamPrimitive";
358 return invalid_stream_id
;
363 return uses_non_zero_stream
;
366 bool uses_end_primitive()
368 return end_primitive_found
;
372 int max_stream_allowed
;
373 int invalid_stream_id
;
374 bool invalid_stream_id_from_emit_vertex
;
375 bool end_primitive_found
;
376 bool uses_non_zero_stream
;
379 /* Class that finds array derefs and check if indexes are dynamic. */
380 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
383 dynamic_sampler_array_indexing_visitor() :
384 dynamic_sampler_array_indexing(false)
388 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
390 if (!ir
->variable_referenced())
391 return visit_continue
;
393 if (!ir
->variable_referenced()->type
->contains_sampler())
394 return visit_continue
;
396 if (!ir
->array_index
->constant_expression_value()) {
397 dynamic_sampler_array_indexing
= true;
400 return visit_continue
;
403 bool uses_dynamic_sampler_array_indexing()
405 return dynamic_sampler_array_indexing
;
409 bool dynamic_sampler_array_indexing
;
412 } /* anonymous namespace */
415 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
419 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
421 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
424 prog
->data
->LinkStatus
= false;
429 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
433 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
435 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
442 * Given a string identifying a program resource, break it into a base name
443 * and an optional array index in square brackets.
445 * If an array index is present, \c out_base_name_end is set to point to the
446 * "[" that precedes the array index, and the array index itself is returned
449 * If no array index is present (or if the array index is negative or
450 * mal-formed), \c out_base_name_end, is set to point to the null terminator
451 * at the end of the input string, and -1 is returned.
453 * Only the final array index is parsed; if the string contains other array
454 * indices (or structure field accesses), they are left in the base name.
456 * No attempt is made to check that the base name is properly formed;
457 * typically the caller will look up the base name in a hash table, so
458 * ill-formed base names simply turn into hash table lookup failures.
461 parse_program_resource_name(const GLchar
*name
,
462 const GLchar
**out_base_name_end
)
464 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
466 * "When an integer array element or block instance number is part of
467 * the name string, it will be specified in decimal form without a "+"
468 * or "-" sign or any extra leading zeroes. Additionally, the name
469 * string will not include white space anywhere in the string."
472 const size_t len
= strlen(name
);
473 *out_base_name_end
= name
+ len
;
475 if (len
== 0 || name
[len
-1] != ']')
478 /* Walk backwards over the string looking for a non-digit character. This
479 * had better be the opening bracket for an array index.
481 * Initially, i specifies the location of the ']'. Since the string may
482 * contain only the ']' charcater, walk backwards very carefully.
485 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
488 if ((i
== 0) || name
[i
-1] != '[')
491 long array_index
= strtol(&name
[i
], NULL
, 10);
495 /* Check for leading zero */
496 if (name
[i
] == '0' && name
[i
+1] != ']')
499 *out_base_name_end
= name
+ (i
- 1);
505 link_invalidate_variable_locations(exec_list
*ir
)
507 foreach_in_list(ir_instruction
, node
, ir
) {
508 ir_variable
*const var
= node
->as_variable();
513 /* Only assign locations for variables that lack an explicit location.
514 * Explicit locations are set for all built-in variables, generic vertex
515 * shader inputs (via layout(location=...)), and generic fragment shader
516 * outputs (also via layout(location=...)).
518 if (!var
->data
.explicit_location
) {
519 var
->data
.location
= -1;
520 var
->data
.location_frac
= 0;
523 /* ir_variable::is_unmatched_generic_inout is used by the linker while
524 * connecting outputs from one stage to inputs of the next stage.
526 if (var
->data
.explicit_location
&&
527 var
->data
.location
< VARYING_SLOT_VAR0
) {
528 var
->data
.is_unmatched_generic_inout
= 0;
530 var
->data
.is_unmatched_generic_inout
= 1;
537 * Set clip_distance_array_size based and cull_distance_array_size on the given
540 * Also check for errors based on incorrect usage of gl_ClipVertex and
541 * gl_ClipDistance and gl_CullDistance.
542 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
543 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
545 * Return false if an error was reported.
548 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
549 struct gl_linked_shader
*shader
,
550 struct gl_context
*ctx
,
551 GLuint
*clip_distance_array_size
,
552 GLuint
*cull_distance_array_size
)
554 *clip_distance_array_size
= 0;
555 *cull_distance_array_size
= 0;
557 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
558 /* From section 7.1 (Vertex Shader Special Variables) of the
561 * "It is an error for a shader to statically write both
562 * gl_ClipVertex and gl_ClipDistance."
564 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
565 * gl_ClipVertex nor gl_ClipDistance. However with
566 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
568 find_assignment_visitor
clip_distance("gl_ClipDistance");
569 find_assignment_visitor
cull_distance("gl_CullDistance");
571 clip_distance
.run(shader
->ir
);
572 cull_distance
.run(shader
->ir
);
574 /* From the ARB_cull_distance spec:
576 * It is a compile-time or link-time error for the set of shaders forming
577 * a program to statically read or write both gl_ClipVertex and either
578 * gl_ClipDistance or gl_CullDistance.
580 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
584 find_assignment_visitor
clip_vertex("gl_ClipVertex");
586 clip_vertex
.run(shader
->ir
);
588 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
589 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
590 "and `gl_ClipDistance'\n",
591 _mesa_shader_stage_to_string(shader
->Stage
));
594 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
595 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
596 "and `gl_CullDistance'\n",
597 _mesa_shader_stage_to_string(shader
->Stage
));
602 if (clip_distance
.variable_found()) {
603 ir_variable
*clip_distance_var
=
604 shader
->symbols
->get_variable("gl_ClipDistance");
605 assert(clip_distance_var
);
606 *clip_distance_array_size
= clip_distance_var
->type
->length
;
608 if (cull_distance
.variable_found()) {
609 ir_variable
*cull_distance_var
=
610 shader
->symbols
->get_variable("gl_CullDistance");
611 assert(cull_distance_var
);
612 *cull_distance_array_size
= cull_distance_var
->type
->length
;
614 /* From the ARB_cull_distance spec:
616 * It is a compile-time or link-time error for the set of shaders forming
617 * a program to have the sum of the sizes of the gl_ClipDistance and
618 * gl_CullDistance arrays to be larger than
619 * gl_MaxCombinedClipAndCullDistances.
621 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
622 ctx
->Const
.MaxClipPlanes
) {
623 linker_error(prog
, "%s shader: the combined size of "
624 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
626 "gl_MaxCombinedClipAndCullDistances (%u)",
627 _mesa_shader_stage_to_string(shader
->Stage
),
628 ctx
->Const
.MaxClipPlanes
);
635 * Verify that a vertex shader executable meets all semantic requirements.
637 * Also sets info.clip_distance_array_size and
638 * info.cull_distance_array_size as a side effect.
640 * \param shader Vertex shader executable to be verified
643 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
644 struct gl_linked_shader
*shader
,
645 struct gl_context
*ctx
)
650 /* From the GLSL 1.10 spec, page 48:
652 * "The variable gl_Position is available only in the vertex
653 * language and is intended for writing the homogeneous vertex
654 * position. All executions of a well-formed vertex shader
655 * executable must write a value into this variable. [...] The
656 * variable gl_Position is available only in the vertex
657 * language and is intended for writing the homogeneous vertex
658 * position. All executions of a well-formed vertex shader
659 * executable must write a value into this variable."
661 * while in GLSL 1.40 this text is changed to:
663 * "The variable gl_Position is available only in the vertex
664 * language and is intended for writing the homogeneous vertex
665 * position. It can be written at any time during shader
666 * execution. It may also be read back by a vertex shader
667 * after being written. This value will be used by primitive
668 * assembly, clipping, culling, and other fixed functionality
669 * operations, if present, that operate on primitives after
670 * vertex processing has occurred. Its value is undefined if
671 * the vertex shader executable does not write gl_Position."
673 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
674 * gl_Position is not an error.
676 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
677 find_assignment_visitor
find("gl_Position");
678 find
.run(shader
->ir
);
679 if (!find
.variable_found()) {
682 "vertex shader does not write to `gl_Position'. "
683 "Its value is undefined. \n");
686 "vertex shader does not write to `gl_Position'. \n");
692 analyze_clip_cull_usage(prog
, shader
, ctx
,
693 &shader
->Program
->info
.clip_distance_array_size
,
694 &shader
->Program
->info
.cull_distance_array_size
);
698 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
699 struct gl_linked_shader
*shader
,
700 struct gl_context
*ctx
)
705 analyze_clip_cull_usage(prog
, shader
, ctx
,
706 &shader
->Program
->info
.clip_distance_array_size
,
707 &shader
->Program
->info
.cull_distance_array_size
);
712 * Verify that a fragment shader executable meets all semantic requirements
714 * \param shader Fragment shader executable to be verified
717 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
718 struct gl_linked_shader
*shader
)
723 find_assignment_visitor
frag_color("gl_FragColor");
724 find_assignment_visitor
frag_data("gl_FragData");
726 frag_color
.run(shader
->ir
);
727 frag_data
.run(shader
->ir
);
729 if (frag_color
.variable_found() && frag_data
.variable_found()) {
730 linker_error(prog
, "fragment shader writes to both "
731 "`gl_FragColor' and `gl_FragData'\n");
736 * Verify that a geometry shader executable meets all semantic requirements
738 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
739 * info.cull_distance_array_size as a side effect.
741 * \param shader Geometry shader executable to be verified
744 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
745 struct gl_linked_shader
*shader
,
746 struct gl_context
*ctx
)
751 unsigned num_vertices
=
752 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
753 prog
->Geom
.VerticesIn
= num_vertices
;
755 analyze_clip_cull_usage(prog
, shader
, ctx
,
756 &shader
->Program
->info
.clip_distance_array_size
,
757 &shader
->Program
->info
.cull_distance_array_size
);
761 * Check if geometry shaders emit to non-zero streams and do corresponding
765 validate_geometry_shader_emissions(struct gl_context
*ctx
,
766 struct gl_shader_program
*prog
)
768 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
771 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
772 emit_vertex
.run(sh
->ir
);
773 if (emit_vertex
.error()) {
774 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
775 "stream parameter are in the range [0, %d].\n",
776 emit_vertex
.error_func(),
777 emit_vertex
.error_stream(),
778 ctx
->Const
.MaxVertexStreams
- 1);
780 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
781 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
783 /* From the ARB_gpu_shader5 spec:
785 * "Multiple vertex streams are supported only if the output primitive
786 * type is declared to be "points". A program will fail to link if it
787 * contains a geometry shader calling EmitStreamVertex() or
788 * EndStreamPrimitive() if its output primitive type is not "points".
790 * However, in the same spec:
792 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
793 * with <stream> set to zero."
797 * "The function EndPrimitive() is equivalent to calling
798 * EndStreamPrimitive() with <stream> set to zero."
800 * Since we can call EmitVertex() and EndPrimitive() when we output
801 * primitives other than points, calling EmitStreamVertex(0) or
802 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
803 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
804 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
807 if (prog
->Geom
.UsesStreams
&&
808 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
809 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
810 "with n>0 requires point output\n");
816 validate_intrastage_arrays(struct gl_shader_program
*prog
,
817 ir_variable
*const var
,
818 ir_variable
*const existing
)
820 /* Consider the types to be "the same" if both types are arrays
821 * of the same type and one of the arrays is implicitly sized.
822 * In addition, set the type of the linked variable to the
823 * explicitly sized array.
825 if (var
->type
->is_array() && existing
->type
->is_array()) {
826 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
827 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
828 if (var
->type
->length
!= 0) {
829 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
830 linker_error(prog
, "%s `%s' declared as type "
831 "`%s' but outermost dimension has an index"
834 var
->name
, var
->type
->name
,
835 existing
->data
.max_array_access
);
837 existing
->type
= var
->type
;
839 } else if (existing
->type
->length
!= 0) {
840 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
841 !existing
->data
.from_ssbo_unsized_array
) {
842 linker_error(prog
, "%s `%s' declared as type "
843 "`%s' but outermost dimension has an index"
846 var
->name
, existing
->type
->name
,
847 var
->data
.max_array_access
);
852 /* The arrays of structs could have different glsl_type pointers but
853 * they are actually the same type. Use record_compare() to check that.
855 if (existing
->type
->fields
.array
->is_record() &&
856 var
->type
->fields
.array
->is_record() &&
857 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
866 * Perform validation of global variables used across multiple shaders
869 cross_validate_globals(struct gl_shader_program
*prog
,
870 struct exec_list
*ir
, glsl_symbol_table
*variables
,
873 foreach_in_list(ir_instruction
, node
, ir
) {
874 ir_variable
*const var
= node
->as_variable();
879 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
882 /* don't cross validate subroutine uniforms */
883 if (var
->type
->contains_subroutine())
886 /* Don't cross validate temporaries that are at global scope. These
887 * will eventually get pulled into the shaders 'main'.
889 if (var
->data
.mode
== ir_var_temporary
)
892 /* If a global with this name has already been seen, verify that the
893 * new instance has the same type. In addition, if the globals have
894 * initializers, the values of the initializers must be the same.
896 ir_variable
*const existing
= variables
->get_variable(var
->name
);
897 if (existing
!= NULL
) {
898 /* Check if types match. Interface blocks have some special
899 * rules so we handle those elsewhere.
901 if (var
->type
!= existing
->type
&&
902 !var
->is_interface_instance()) {
903 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
904 if (var
->type
->is_record() && existing
->type
->is_record()
905 && existing
->type
->record_compare(var
->type
)) {
906 existing
->type
= var
->type
;
908 /* If it is an unsized array in a Shader Storage Block,
909 * two different shaders can access to different elements.
910 * Because of that, they might be converted to different
911 * sized arrays, then check that they are compatible but
912 * ignore the array size.
914 if (!(var
->data
.mode
== ir_var_shader_storage
&&
915 var
->data
.from_ssbo_unsized_array
&&
916 existing
->data
.mode
== ir_var_shader_storage
&&
917 existing
->data
.from_ssbo_unsized_array
&&
918 var
->type
->gl_type
== existing
->type
->gl_type
)) {
919 linker_error(prog
, "%s `%s' declared as type "
920 "`%s' and type `%s'\n",
922 var
->name
, var
->type
->name
,
923 existing
->type
->name
);
930 if (var
->data
.explicit_location
) {
931 if (existing
->data
.explicit_location
932 && (var
->data
.location
!= existing
->data
.location
)) {
933 linker_error(prog
, "explicit locations for %s "
934 "`%s' have differing values\n",
935 mode_string(var
), var
->name
);
939 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
940 linker_error(prog
, "explicit components for %s `%s' have "
941 "differing values\n", mode_string(var
), var
->name
);
945 existing
->data
.location
= var
->data
.location
;
946 existing
->data
.explicit_location
= true;
948 /* Check if uniform with implicit location was marked explicit
949 * by earlier shader stage. If so, mark it explicit in this stage
950 * too to make sure later processing does not treat it as
953 if (existing
->data
.explicit_location
) {
954 var
->data
.location
= existing
->data
.location
;
955 var
->data
.explicit_location
= true;
959 /* From the GLSL 4.20 specification:
960 * "A link error will result if two compilation units in a program
961 * specify different integer-constant bindings for the same
962 * opaque-uniform name. However, it is not an error to specify a
963 * binding on some but not all declarations for the same name"
965 if (var
->data
.explicit_binding
) {
966 if (existing
->data
.explicit_binding
&&
967 var
->data
.binding
!= existing
->data
.binding
) {
968 linker_error(prog
, "explicit bindings for %s "
969 "`%s' have differing values\n",
970 mode_string(var
), var
->name
);
974 existing
->data
.binding
= var
->data
.binding
;
975 existing
->data
.explicit_binding
= true;
978 if (var
->type
->contains_atomic() &&
979 var
->data
.offset
!= existing
->data
.offset
) {
980 linker_error(prog
, "offset specifications for %s "
981 "`%s' have differing values\n",
982 mode_string(var
), var
->name
);
986 /* Validate layout qualifiers for gl_FragDepth.
988 * From the AMD/ARB_conservative_depth specs:
990 * "If gl_FragDepth is redeclared in any fragment shader in a
991 * program, it must be redeclared in all fragment shaders in
992 * that program that have static assignments to
993 * gl_FragDepth. All redeclarations of gl_FragDepth in all
994 * fragment shaders in a single program must have the same set
997 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
998 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
999 bool layout_differs
=
1000 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1002 if (layout_declared
&& layout_differs
) {
1004 "All redeclarations of gl_FragDepth in all "
1005 "fragment shaders in a single program must have "
1006 "the same set of qualifiers.\n");
1009 if (var
->data
.used
&& layout_differs
) {
1011 "If gl_FragDepth is redeclared with a layout "
1012 "qualifier in any fragment shader, it must be "
1013 "redeclared with the same layout qualifier in "
1014 "all fragment shaders that have assignments to "
1019 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1021 * "If a shared global has multiple initializers, the
1022 * initializers must all be constant expressions, and they
1023 * must all have the same value. Otherwise, a link error will
1024 * result. (A shared global having only one initializer does
1025 * not require that initializer to be a constant expression.)"
1027 * Previous to 4.20 the GLSL spec simply said that initializers
1028 * must have the same value. In this case of non-constant
1029 * initializers, this was impossible to determine. As a result,
1030 * no vendor actually implemented that behavior. The 4.20
1031 * behavior matches the implemented behavior of at least one other
1032 * vendor, so we'll implement that for all GLSL versions.
1034 if (var
->constant_initializer
!= NULL
) {
1035 if (existing
->constant_initializer
!= NULL
) {
1036 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1037 linker_error(prog
, "initializers for %s "
1038 "`%s' have differing values\n",
1039 mode_string(var
), var
->name
);
1043 /* If the first-seen instance of a particular uniform did
1044 * not have an initializer but a later instance does,
1045 * replace the former with the later.
1047 variables
->replace_variable(existing
->name
, var
);
1051 if (var
->data
.has_initializer
) {
1052 if (existing
->data
.has_initializer
1053 && (var
->constant_initializer
== NULL
1054 || existing
->constant_initializer
== NULL
)) {
1056 "shared global variable `%s' has multiple "
1057 "non-constant initializers.\n",
1063 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1064 linker_error(prog
, "declarations for %s `%s' have "
1065 "mismatching invariant qualifiers\n",
1066 mode_string(var
), var
->name
);
1069 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1070 linker_error(prog
, "declarations for %s `%s' have "
1071 "mismatching centroid qualifiers\n",
1072 mode_string(var
), var
->name
);
1075 if (existing
->data
.sample
!= var
->data
.sample
) {
1076 linker_error(prog
, "declarations for %s `%s` have "
1077 "mismatching sample qualifiers\n",
1078 mode_string(var
), var
->name
);
1081 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1082 linker_error(prog
, "declarations for %s `%s` have "
1083 "mismatching image format qualifiers\n",
1084 mode_string(var
), var
->name
);
1088 /* Only in GLSL ES 3.10, the precision qualifier should not match
1089 * between block members defined in matched block names within a
1092 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1093 * member should match.
1095 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1096 !var
->get_interface_type()) &&
1097 existing
->data
.precision
!= var
->data
.precision
) {
1098 linker_error(prog
, "declarations for %s `%s` have "
1099 "mismatching precision qualifiers\n",
1100 mode_string(var
), var
->name
);
1104 variables
->add_variable(var
);
1110 * Perform validation of uniforms used across multiple shader stages
1113 cross_validate_uniforms(struct gl_shader_program
*prog
)
1115 glsl_symbol_table variables
;
1116 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1117 if (prog
->_LinkedShaders
[i
] == NULL
)
1120 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1126 * Accumulates the array of buffer blocks and checks that all definitions of
1127 * blocks agree on their contents.
1130 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1133 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1134 struct gl_uniform_block
*blks
= NULL
;
1135 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1136 &prog
->data
->NumUniformBlocks
;
1138 unsigned max_num_buffer_blocks
= 0;
1139 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1140 if (prog
->_LinkedShaders
[i
]) {
1141 if (validate_ssbo
) {
1142 max_num_buffer_blocks
+=
1143 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1145 max_num_buffer_blocks
+=
1146 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1151 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1152 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1154 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1155 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1156 InterfaceBlockStageIndex
[i
][j
] = -1;
1161 unsigned sh_num_blocks
;
1162 struct gl_uniform_block
**sh_blks
;
1163 if (validate_ssbo
) {
1164 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1165 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1167 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1168 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1171 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1172 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1176 linker_error(prog
, "buffer block `%s' has mismatching "
1177 "definitions\n", sh_blks
[j
]->Name
);
1179 for (unsigned k
= 0; k
<= i
; k
++) {
1180 delete[] InterfaceBlockStageIndex
[k
];
1185 InterfaceBlockStageIndex
[i
][index
] = j
;
1189 /* Update per stage block pointers to point to the program list.
1190 * FIXME: We should be able to free the per stage blocks here.
1192 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1193 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1194 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1196 if (stage_index
!= -1) {
1197 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1199 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1200 sh
->Program
->sh
.ShaderStorageBlocks
:
1201 sh
->Program
->sh
.UniformBlocks
;
1203 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1204 sh_blks
[stage_index
] = &blks
[j
];
1209 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1210 delete[] InterfaceBlockStageIndex
[i
];
1214 prog
->data
->ShaderStorageBlocks
= blks
;
1216 prog
->data
->UniformBlocks
= blks
;
1223 * Populates a shaders symbol table with all global declarations
1226 populate_symbol_table(gl_linked_shader
*sh
)
1228 sh
->symbols
= new(sh
) glsl_symbol_table
;
1230 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1234 if ((func
= inst
->as_function()) != NULL
) {
1235 sh
->symbols
->add_function(func
);
1236 } else if ((var
= inst
->as_variable()) != NULL
) {
1237 if (var
->data
.mode
!= ir_var_temporary
)
1238 sh
->symbols
->add_variable(var
);
1245 * Remap variables referenced in an instruction tree
1247 * This is used when instruction trees are cloned from one shader and placed in
1248 * another. These trees will contain references to \c ir_variable nodes that
1249 * do not exist in the target shader. This function finds these \c ir_variable
1250 * references and replaces the references with matching variables in the target
1253 * If there is no matching variable in the target shader, a clone of the
1254 * \c ir_variable is made and added to the target shader. The new variable is
1255 * added to \b both the instruction stream and the symbol table.
1257 * \param inst IR tree that is to be processed.
1258 * \param symbols Symbol table containing global scope symbols in the
1260 * \param instructions Instruction stream where new variable declarations
1264 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1267 class remap_visitor
: public ir_hierarchical_visitor
{
1269 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1271 this->target
= target
;
1272 this->symbols
= target
->symbols
;
1273 this->instructions
= target
->ir
;
1274 this->temps
= temps
;
1277 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1279 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1280 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1281 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1283 assert(var
!= NULL
);
1285 return visit_continue
;
1288 ir_variable
*const existing
=
1289 this->symbols
->get_variable(ir
->var
->name
);
1290 if (existing
!= NULL
)
1293 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1295 this->symbols
->add_variable(copy
);
1296 this->instructions
->push_head(copy
);
1300 return visit_continue
;
1304 struct gl_linked_shader
*target
;
1305 glsl_symbol_table
*symbols
;
1306 exec_list
*instructions
;
1310 remap_visitor
v(target
, temps
);
1317 * Move non-declarations from one instruction stream to another
1319 * The intended usage pattern of this function is to pass the pointer to the
1320 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1321 * pointer) for \c last and \c false for \c make_copies on the first
1322 * call. Successive calls pass the return value of the previous call for
1323 * \c last and \c true for \c make_copies.
1325 * \param instructions Source instruction stream
1326 * \param last Instruction after which new instructions should be
1327 * inserted in the target instruction stream
1328 * \param make_copies Flag selecting whether instructions in \c instructions
1329 * should be copied (via \c ir_instruction::clone) into the
1330 * target list or moved.
1333 * The new "last" instruction in the target instruction stream. This pointer
1334 * is suitable for use as the \c last parameter of a later call to this
1338 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1339 bool make_copies
, gl_linked_shader
*target
)
1341 hash_table
*temps
= NULL
;
1344 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1345 _mesa_key_pointer_equal
);
1347 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1348 if (inst
->as_function())
1351 ir_variable
*var
= inst
->as_variable();
1352 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1355 assert(inst
->as_assignment()
1357 || inst
->as_if() /* for initializers with the ?: operator */
1358 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1361 inst
= inst
->clone(target
, NULL
);
1364 _mesa_hash_table_insert(temps
, var
, inst
);
1366 remap_variables(inst
, target
, temps
);
1371 last
->insert_after(inst
);
1376 _mesa_hash_table_destroy(temps
, NULL
);
1383 * This class is only used in link_intrastage_shaders() below but declaring
1384 * it inside that function leads to compiler warnings with some versions of
1387 class array_sizing_visitor
: public deref_type_updater
{
1389 array_sizing_visitor()
1390 : mem_ctx(ralloc_context(NULL
)),
1391 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1392 _mesa_key_pointer_equal
))
1396 ~array_sizing_visitor()
1398 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1399 ralloc_free(this->mem_ctx
);
1402 virtual ir_visitor_status
visit(ir_variable
*var
)
1404 const glsl_type
*type_without_array
;
1405 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1406 fixup_type(&var
->type
, var
->data
.max_array_access
,
1407 var
->data
.from_ssbo_unsized_array
,
1408 &implicit_sized_array
);
1409 var
->data
.implicit_sized_array
= implicit_sized_array
;
1410 type_without_array
= var
->type
->without_array();
1411 if (var
->type
->is_interface()) {
1412 if (interface_contains_unsized_arrays(var
->type
)) {
1413 const glsl_type
*new_type
=
1414 resize_interface_members(var
->type
,
1415 var
->get_max_ifc_array_access(),
1416 var
->is_in_shader_storage_block());
1417 var
->type
= new_type
;
1418 var
->change_interface_type(new_type
);
1420 } else if (type_without_array
->is_interface()) {
1421 if (interface_contains_unsized_arrays(type_without_array
)) {
1422 const glsl_type
*new_type
=
1423 resize_interface_members(type_without_array
,
1424 var
->get_max_ifc_array_access(),
1425 var
->is_in_shader_storage_block());
1426 var
->change_interface_type(new_type
);
1427 var
->type
= update_interface_members_array(var
->type
, new_type
);
1429 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1430 /* Store a pointer to the variable in the unnamed_interfaces
1434 _mesa_hash_table_search(this->unnamed_interfaces
,
1437 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1439 if (interface_vars
== NULL
) {
1440 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1442 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1445 unsigned index
= ifc_type
->field_index(var
->name
);
1446 assert(index
< ifc_type
->length
);
1447 assert(interface_vars
[index
] == NULL
);
1448 interface_vars
[index
] = var
;
1450 return visit_continue
;
1454 * For each unnamed interface block that was discovered while running the
1455 * visitor, adjust the interface type to reflect the newly assigned array
1456 * sizes, and fix up the ir_variable nodes to point to the new interface
1459 void fixup_unnamed_interface_types()
1461 hash_table_call_foreach(this->unnamed_interfaces
,
1462 fixup_unnamed_interface_type
, NULL
);
1467 * If the type pointed to by \c type represents an unsized array, replace
1468 * it with a sized array whose size is determined by max_array_access.
1470 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1471 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1473 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1474 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1475 max_array_access
+ 1);
1476 *implicit_sized
= true;
1477 assert(*type
!= NULL
);
1481 static const glsl_type
*
1482 update_interface_members_array(const glsl_type
*type
,
1483 const glsl_type
*new_interface_type
)
1485 const glsl_type
*element_type
= type
->fields
.array
;
1486 if (element_type
->is_array()) {
1487 const glsl_type
*new_array_type
=
1488 update_interface_members_array(element_type
, new_interface_type
);
1489 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1491 return glsl_type::get_array_instance(new_interface_type
,
1497 * Determine whether the given interface type contains unsized arrays (if
1498 * it doesn't, array_sizing_visitor doesn't need to process it).
1500 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1502 for (unsigned i
= 0; i
< type
->length
; i
++) {
1503 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1504 if (elem_type
->is_unsized_array())
1511 * Create a new interface type based on the given type, with unsized arrays
1512 * replaced by sized arrays whose size is determined by
1513 * max_ifc_array_access.
1515 static const glsl_type
*
1516 resize_interface_members(const glsl_type
*type
,
1517 const int *max_ifc_array_access
,
1520 unsigned num_fields
= type
->length
;
1521 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1522 memcpy(fields
, type
->fields
.structure
,
1523 num_fields
* sizeof(*fields
));
1524 for (unsigned i
= 0; i
< num_fields
; i
++) {
1525 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1526 /* If SSBO last member is unsized array, we don't replace it by a sized
1529 if (is_ssbo
&& i
== (num_fields
- 1))
1530 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1531 true, &implicit_sized_array
);
1533 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1534 false, &implicit_sized_array
);
1535 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1537 glsl_interface_packing packing
=
1538 (glsl_interface_packing
) type
->interface_packing
;
1539 bool row_major
= (bool) type
->interface_row_major
;
1540 const glsl_type
*new_ifc_type
=
1541 glsl_type::get_interface_instance(fields
, num_fields
,
1542 packing
, row_major
, type
->name
);
1544 return new_ifc_type
;
1547 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1550 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1551 ir_variable
**interface_vars
= (ir_variable
**) data
;
1552 unsigned num_fields
= ifc_type
->length
;
1553 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1554 memcpy(fields
, ifc_type
->fields
.structure
,
1555 num_fields
* sizeof(*fields
));
1556 bool interface_type_changed
= false;
1557 for (unsigned i
= 0; i
< num_fields
; i
++) {
1558 if (interface_vars
[i
] != NULL
&&
1559 fields
[i
].type
!= interface_vars
[i
]->type
) {
1560 fields
[i
].type
= interface_vars
[i
]->type
;
1561 interface_type_changed
= true;
1564 if (!interface_type_changed
) {
1568 glsl_interface_packing packing
=
1569 (glsl_interface_packing
) ifc_type
->interface_packing
;
1570 bool row_major
= (bool) ifc_type
->interface_row_major
;
1571 const glsl_type
*new_ifc_type
=
1572 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1573 row_major
, ifc_type
->name
);
1575 for (unsigned i
= 0; i
< num_fields
; i
++) {
1576 if (interface_vars
[i
] != NULL
)
1577 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1582 * Memory context used to allocate the data in \c unnamed_interfaces.
1587 * Hash table from const glsl_type * to an array of ir_variable *'s
1588 * pointing to the ir_variables constituting each unnamed interface block.
1590 hash_table
*unnamed_interfaces
;
1594 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1595 struct gl_shader_program
*prog
)
1597 /* We will validate doubles at a later stage */
1598 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1599 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1600 "multiple of 4 or if its applied to a type that is "
1601 "or contains a double a multiple of 8.",
1602 prog
->TransformFeedback
.BufferStride
[idx
]);
1606 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1607 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1608 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1609 "limit has been exceeded.");
1617 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1621 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1622 struct gl_shader_program
*prog
,
1623 struct gl_linked_shader
*linked_shader
,
1624 struct gl_shader
**shader_list
,
1625 unsigned num_shaders
)
1627 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1628 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1631 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1632 struct gl_shader
*shader
= shader_list
[i
];
1634 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1635 if (shader
->TransformFeedbackBufferStride
[j
]) {
1636 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1637 prog
->TransformFeedback
.BufferStride
[j
] =
1638 shader
->TransformFeedbackBufferStride
[j
];
1639 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1641 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1642 shader
->TransformFeedbackBufferStride
[j
]){
1644 "intrastage shaders defined with conflicting "
1645 "xfb_stride for buffer %d (%d and %d)\n", j
,
1646 prog
->TransformFeedback
.BufferStride
[j
],
1647 shader
->TransformFeedbackBufferStride
[j
]);
1656 * Performs the cross-validation of tessellation control shader vertices and
1657 * layout qualifiers for the attached tessellation control shaders,
1658 * and propagates them to the linked TCS and linked shader program.
1661 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1662 struct gl_program
*gl_prog
,
1663 struct gl_shader
**shader_list
,
1664 unsigned num_shaders
)
1666 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1669 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1671 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1673 * "All tessellation control shader layout declarations in a program
1674 * must specify the same output patch vertex count. There must be at
1675 * least one layout qualifier specifying an output patch vertex count
1676 * in any program containing tessellation control shaders; however,
1677 * such a declaration is not required in all tessellation control
1681 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1682 struct gl_shader
*shader
= shader_list
[i
];
1684 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1685 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1686 gl_prog
->info
.tess
.tcs_vertices_out
!=
1687 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1688 linker_error(prog
, "tessellation control shader defined with "
1689 "conflicting output vertex count (%d and %d)\n",
1690 gl_prog
->info
.tess
.tcs_vertices_out
,
1691 shader
->info
.TessCtrl
.VerticesOut
);
1694 gl_prog
->info
.tess
.tcs_vertices_out
=
1695 shader
->info
.TessCtrl
.VerticesOut
;
1699 /* Just do the intrastage -> interstage propagation right now,
1700 * since we already know we're in the right type of shader program
1703 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1704 linker_error(prog
, "tessellation control shader didn't declare "
1705 "vertices out layout qualifier\n");
1712 * Performs the cross-validation of tessellation evaluation shader
1713 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1714 * for the attached tessellation evaluation shaders, and propagates them
1715 * to the linked TES and linked shader program.
1718 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1719 struct gl_program
*gl_prog
,
1720 struct gl_shader
**shader_list
,
1721 unsigned num_shaders
)
1723 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1726 int point_mode
= -1;
1727 unsigned vertex_order
= 0;
1729 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1730 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1732 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1734 * "At least one tessellation evaluation shader (compilation unit) in
1735 * a program must declare a primitive mode in its input layout.
1736 * Declaration vertex spacing, ordering, and point mode identifiers is
1737 * optional. It is not required that all tessellation evaluation
1738 * shaders in a program declare a primitive mode. If spacing or
1739 * vertex ordering declarations are omitted, the tessellation
1740 * primitive generator will use equal spacing or counter-clockwise
1741 * vertex ordering, respectively. If a point mode declaration is
1742 * omitted, the tessellation primitive generator will produce lines or
1743 * triangles according to the primitive mode."
1746 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1747 struct gl_shader
*shader
= shader_list
[i
];
1749 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1750 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1751 gl_prog
->info
.tess
.primitive_mode
!=
1752 shader
->info
.TessEval
.PrimitiveMode
) {
1753 linker_error(prog
, "tessellation evaluation shader defined with "
1754 "conflicting input primitive modes.\n");
1757 gl_prog
->info
.tess
.primitive_mode
=
1758 shader
->info
.TessEval
.PrimitiveMode
;
1761 if (shader
->info
.TessEval
.Spacing
!= 0) {
1762 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1763 shader
->info
.TessEval
.Spacing
) {
1764 linker_error(prog
, "tessellation evaluation shader defined with "
1765 "conflicting vertex spacing.\n");
1768 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1771 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1772 if (vertex_order
!= 0 &&
1773 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1774 linker_error(prog
, "tessellation evaluation shader defined with "
1775 "conflicting ordering.\n");
1778 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1781 if (shader
->info
.TessEval
.PointMode
!= -1) {
1782 if (point_mode
!= -1 &&
1783 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1784 linker_error(prog
, "tessellation evaluation shader defined with "
1785 "conflicting point modes.\n");
1788 point_mode
= shader
->info
.TessEval
.PointMode
;
1793 /* Just do the intrastage -> interstage propagation right now,
1794 * since we already know we're in the right type of shader program
1797 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1799 "tessellation evaluation shader didn't declare input "
1800 "primitive modes.\n");
1804 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1805 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1807 if (vertex_order
== 0)
1808 gl_prog
->info
.tess
.ccw
= true;
1810 gl_prog
->info
.tess
.ccw
= false;
1813 if (point_mode
== -1)
1814 gl_prog
->info
.tess
.point_mode
= false;
1816 gl_prog
->info
.tess
.point_mode
= true;
1821 * Performs the cross-validation of layout qualifiers specified in
1822 * redeclaration of gl_FragCoord for the attached fragment shaders,
1823 * and propagates them to the linked FS and linked shader program.
1826 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1827 struct gl_linked_shader
*linked_shader
,
1828 struct gl_shader
**shader_list
,
1829 unsigned num_shaders
)
1831 bool redeclares_gl_fragcoord
= false;
1832 bool uses_gl_fragcoord
= false;
1833 bool origin_upper_left
= false;
1834 bool pixel_center_integer
= false;
1836 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1837 (prog
->data
->Version
< 150 &&
1838 !prog
->ARB_fragment_coord_conventions_enable
))
1841 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1842 struct gl_shader
*shader
= shader_list
[i
];
1843 /* From the GLSL 1.50 spec, page 39:
1845 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1846 * it must be redeclared in all the fragment shaders in that program
1847 * that have a static use gl_FragCoord."
1849 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1850 shader
->uses_gl_fragcoord
)
1851 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1852 uses_gl_fragcoord
)) {
1853 linker_error(prog
, "fragment shader defined with conflicting "
1854 "layout qualifiers for gl_FragCoord\n");
1857 /* From the GLSL 1.50 spec, page 39:
1859 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1860 * single program must have the same set of qualifiers."
1862 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1863 (shader
->origin_upper_left
!= origin_upper_left
||
1864 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1865 linker_error(prog
, "fragment shader defined with conflicting "
1866 "layout qualifiers for gl_FragCoord\n");
1869 /* Update the linked shader state. Note that uses_gl_fragcoord should
1870 * accumulate the results. The other values should replace. If there
1871 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1872 * are already known to be the same.
1874 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1875 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1876 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1877 origin_upper_left
= shader
->origin_upper_left
;
1878 pixel_center_integer
= shader
->pixel_center_integer
;
1881 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1882 shader
->EarlyFragmentTests
;
1883 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1884 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1885 shader
->PostDepthCoverage
;
1887 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1892 * Performs the cross-validation of geometry shader max_vertices and
1893 * primitive type layout qualifiers for the attached geometry shaders,
1894 * and propagates them to the linked GS and linked shader program.
1897 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1898 struct gl_program
*gl_prog
,
1899 struct gl_shader
**shader_list
,
1900 unsigned num_shaders
)
1902 /* No in/out qualifiers defined for anything but GLSL 1.50+
1903 * geometry shaders so far.
1905 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
1906 prog
->data
->Version
< 150)
1909 int vertices_out
= -1;
1911 gl_prog
->info
.gs
.invocations
= 0;
1912 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
1913 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
1915 /* From the GLSL 1.50 spec, page 46:
1917 * "All geometry shader output layout declarations in a program
1918 * must declare the same layout and same value for
1919 * max_vertices. There must be at least one geometry output
1920 * layout declaration somewhere in a program, but not all
1921 * geometry shaders (compilation units) are required to
1925 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1926 struct gl_shader
*shader
= shader_list
[i
];
1928 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1929 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
1930 gl_prog
->info
.gs
.input_primitive
!=
1931 shader
->info
.Geom
.InputType
) {
1932 linker_error(prog
, "geometry shader defined with conflicting "
1936 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
1939 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1940 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
1941 gl_prog
->info
.gs
.output_primitive
!=
1942 shader
->info
.Geom
.OutputType
) {
1943 linker_error(prog
, "geometry shader defined with conflicting "
1947 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
1950 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1951 if (vertices_out
!= -1 &&
1952 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
1953 linker_error(prog
, "geometry shader defined with conflicting "
1954 "output vertex count (%d and %d)\n",
1955 vertices_out
, shader
->info
.Geom
.VerticesOut
);
1958 vertices_out
= shader
->info
.Geom
.VerticesOut
;
1961 if (shader
->info
.Geom
.Invocations
!= 0) {
1962 if (gl_prog
->info
.gs
.invocations
!= 0 &&
1963 gl_prog
->info
.gs
.invocations
!=
1964 (unsigned) shader
->info
.Geom
.Invocations
) {
1965 linker_error(prog
, "geometry shader defined with conflicting "
1966 "invocation count (%d and %d)\n",
1967 gl_prog
->info
.gs
.invocations
,
1968 shader
->info
.Geom
.Invocations
);
1971 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
1975 /* Just do the intrastage -> interstage propagation right now,
1976 * since we already know we're in the right type of shader program
1979 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
1981 "geometry shader didn't declare primitive input type\n");
1985 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
1987 "geometry shader didn't declare primitive output type\n");
1991 if (vertices_out
== -1) {
1993 "geometry shader didn't declare max_vertices\n");
1996 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
1999 if (gl_prog
->info
.gs
.invocations
== 0)
2000 gl_prog
->info
.gs
.invocations
= 1;
2005 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2006 * qualifiers for the attached compute shaders, and propagate them to the
2007 * linked CS and linked shader program.
2010 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2011 struct gl_linked_shader
*linked_shader
,
2012 struct gl_shader
**shader_list
,
2013 unsigned num_shaders
)
2015 for (int i
= 0; i
< 3; i
++)
2016 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
2018 linked_shader
->info
.Comp
.LocalSizeVariable
= false;
2020 /* This function is called for all shader stages, but it only has an effect
2021 * for compute shaders.
2023 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2026 /* From the ARB_compute_shader spec, in the section describing local size
2029 * If multiple compute shaders attached to a single program object
2030 * declare local work-group size, the declarations must be identical;
2031 * otherwise a link-time error results. Furthermore, if a program
2032 * object contains any compute shaders, at least one must contain an
2033 * input layout qualifier specifying the local work sizes of the
2034 * program, or a link-time error will occur.
2036 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2037 struct gl_shader
*shader
= shader_list
[sh
];
2039 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2040 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2041 for (int i
= 0; i
< 3; i
++) {
2042 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2043 shader
->info
.Comp
.LocalSize
[i
]) {
2044 linker_error(prog
, "compute shader defined with conflicting "
2050 for (int i
= 0; i
< 3; i
++) {
2051 linked_shader
->info
.Comp
.LocalSize
[i
] =
2052 shader
->info
.Comp
.LocalSize
[i
];
2054 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2055 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2056 /* The ARB_compute_variable_group_size spec says:
2058 * If one compute shader attached to a program declares a
2059 * variable local group size and a second compute shader
2060 * attached to the same program declares a fixed local group
2061 * size, a link-time error results.
2063 linker_error(prog
, "compute shader defined with both fixed and "
2064 "variable local group size\n");
2067 linked_shader
->info
.Comp
.LocalSizeVariable
= true;
2071 /* Just do the intrastage -> interstage propagation right now,
2072 * since we already know we're in the right type of shader program
2075 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0 &&
2076 !linked_shader
->info
.Comp
.LocalSizeVariable
) {
2077 linker_error(prog
, "compute shader must contain a fixed or a variable "
2078 "local group size\n");
2081 for (int i
= 0; i
< 3; i
++)
2082 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2084 prog
->Comp
.LocalSizeVariable
=
2085 linked_shader
->info
.Comp
.LocalSizeVariable
;
2090 * Combine a group of shaders for a single stage to generate a linked shader
2093 * If this function is supplied a single shader, it is cloned, and the new
2094 * shader is returned.
2096 struct gl_linked_shader
*
2097 link_intrastage_shaders(void *mem_ctx
,
2098 struct gl_context
*ctx
,
2099 struct gl_shader_program
*prog
,
2100 struct gl_shader
**shader_list
,
2101 unsigned num_shaders
,
2102 bool allow_missing_main
)
2104 struct gl_uniform_block
*ubo_blocks
= NULL
;
2105 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2106 unsigned num_ubo_blocks
= 0;
2107 unsigned num_ssbo_blocks
= 0;
2109 /* Check that global variables defined in multiple shaders are consistent.
2111 glsl_symbol_table variables
;
2112 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2113 if (shader_list
[i
] == NULL
)
2115 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2118 if (!prog
->data
->LinkStatus
)
2121 /* Check that interface blocks defined in multiple shaders are consistent.
2123 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2125 if (!prog
->data
->LinkStatus
)
2128 /* Check that there is only a single definition of each function signature
2129 * across all shaders.
2131 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2132 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2133 ir_function
*const f
= node
->as_function();
2138 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2139 ir_function
*const other
=
2140 shader_list
[j
]->symbols
->get_function(f
->name
);
2142 /* If the other shader has no function (and therefore no function
2143 * signatures) with the same name, skip to the next shader.
2148 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2149 if (!sig
->is_defined
)
2152 ir_function_signature
*other_sig
=
2153 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2155 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2156 linker_error(prog
, "function `%s' is multiply defined\n",
2165 /* Find the shader that defines main, and make a clone of it.
2167 * Starting with the clone, search for undefined references. If one is
2168 * found, find the shader that defines it. Clone the reference and add
2169 * it to the shader. Repeat until there are no undefined references or
2170 * until a reference cannot be resolved.
2172 gl_shader
*main
= NULL
;
2173 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2174 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2175 main
= shader_list
[i
];
2180 if (main
== NULL
&& allow_missing_main
)
2181 main
= shader_list
[0];
2184 linker_error(prog
, "%s shader lacks `main'\n",
2185 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2189 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2190 linked
->Stage
= shader_list
[0]->Stage
;
2192 /* Create program and attach it to the linked shader */
2193 struct gl_program
*gl_prog
=
2194 ctx
->Driver
.NewProgram(ctx
,
2195 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2198 prog
->data
->LinkStatus
= false;
2199 _mesa_delete_linked_shader(ctx
, linked
);
2203 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2205 /* Don't use _mesa_reference_program() just take ownership */
2206 linked
->Program
= gl_prog
;
2208 linked
->ir
= new(linked
) exec_list
;
2209 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2211 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2212 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2213 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2214 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2215 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2216 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2219 populate_symbol_table(linked
);
2221 /* The pointer to the main function in the final linked shader (i.e., the
2222 * copy of the original shader that contained the main function).
2224 ir_function_signature
*const main_sig
=
2225 _mesa_get_main_function_signature(linked
->symbols
);
2227 /* Move any instructions other than variable declarations or function
2228 * declarations into main.
2230 if (main_sig
!= NULL
) {
2231 exec_node
*insertion_point
=
2232 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2235 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2236 if (shader_list
[i
] == main
)
2239 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2240 insertion_point
, true, linked
);
2244 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2245 _mesa_delete_linked_shader(ctx
, linked
);
2249 /* Make a pass over all variable declarations to ensure that arrays with
2250 * unspecified sizes have a size specified. The size is inferred from the
2251 * max_array_access field.
2253 array_sizing_visitor v
;
2255 v
.fixup_unnamed_interface_types();
2257 /* Link up uniform blocks defined within this stage. */
2258 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2259 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2261 if (!prog
->data
->LinkStatus
) {
2262 _mesa_delete_linked_shader(ctx
, linked
);
2266 /* Copy ubo blocks to linked shader list */
2267 linked
->Program
->sh
.UniformBlocks
=
2268 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2269 ralloc_steal(linked
, ubo_blocks
);
2270 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2271 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2273 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2275 /* Copy ssbo blocks to linked shader list */
2276 linked
->Program
->sh
.ShaderStorageBlocks
=
2277 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2278 ralloc_steal(linked
, ssbo_blocks
);
2279 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2280 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2282 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2284 /* At this point linked should contain all of the linked IR, so
2285 * validate it to make sure nothing went wrong.
2287 validate_ir_tree(linked
->ir
);
2289 /* Set the size of geometry shader input arrays */
2290 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2291 unsigned num_vertices
=
2292 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2293 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2294 MESA_SHADER_GEOMETRY
);
2295 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2296 ir
->accept(&input_resize_visitor
);
2300 if (ctx
->Const
.VertexID_is_zero_based
)
2301 lower_vertex_id(linked
);
2304 /* Compute the source checksum. */
2305 linked
->SourceChecksum
= 0;
2306 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2307 if (shader_list
[i
] == NULL
)
2309 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2317 * Update the sizes of linked shader uniform arrays to the maximum
2320 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2322 * If one or more elements of an array are active,
2323 * GetActiveUniform will return the name of the array in name,
2324 * subject to the restrictions listed above. The type of the array
2325 * is returned in type. The size parameter contains the highest
2326 * array element index used, plus one. The compiler or linker
2327 * determines the highest index used. There will be only one
2328 * active uniform reported by the GL per uniform array.
2332 update_array_sizes(struct gl_shader_program
*prog
)
2334 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2335 if (prog
->_LinkedShaders
[i
] == NULL
)
2338 bool types_were_updated
= false;
2340 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2341 ir_variable
*const var
= node
->as_variable();
2343 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2344 !var
->type
->is_array())
2347 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2348 * will not be eliminated. Since we always do std140, just
2349 * don't resize arrays in UBOs.
2351 * Atomic counters are supposed to get deterministic
2352 * locations assigned based on the declaration ordering and
2353 * sizes, array compaction would mess that up.
2355 * Subroutine uniforms are not removed.
2357 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2358 var
->type
->contains_subroutine() || var
->constant_initializer
)
2361 int size
= var
->data
.max_array_access
;
2362 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2363 if (prog
->_LinkedShaders
[j
] == NULL
)
2366 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2367 ir_variable
*other_var
= node2
->as_variable();
2371 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2372 other_var
->data
.max_array_access
> size
) {
2373 size
= other_var
->data
.max_array_access
;
2378 if (size
+ 1 != (int)var
->type
->length
) {
2379 /* If this is a built-in uniform (i.e., it's backed by some
2380 * fixed-function state), adjust the number of state slots to
2381 * match the new array size. The number of slots per array entry
2382 * is not known. It seems safe to assume that the total number of
2383 * slots is an integer multiple of the number of array elements.
2384 * Determine the number of slots per array element by dividing by
2385 * the old (total) size.
2387 const unsigned num_slots
= var
->get_num_state_slots();
2388 if (num_slots
> 0) {
2389 var
->set_num_state_slots((size
+ 1)
2390 * (num_slots
/ var
->type
->length
));
2393 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2395 types_were_updated
= true;
2399 /* Update the types of dereferences in case we changed any. */
2400 if (types_were_updated
) {
2401 deref_type_updater v
;
2402 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2408 * Resize tessellation evaluation per-vertex inputs to the size of
2409 * tessellation control per-vertex outputs.
2412 resize_tes_inputs(struct gl_context
*ctx
,
2413 struct gl_shader_program
*prog
)
2415 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2418 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2419 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2421 /* If no control shader is present, then the TES inputs are statically
2422 * sized to MaxPatchVertices; the actual size of the arrays won't be
2423 * known until draw time.
2425 const int num_vertices
= tcs
2426 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2427 : ctx
->Const
.MaxPatchVertices
;
2429 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2430 MESA_SHADER_TESS_EVAL
);
2431 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2432 ir
->accept(&input_resize_visitor
);
2435 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2436 /* Convert the gl_PatchVerticesIn system value into a constant, since
2437 * the value is known at this point.
2439 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2440 ir_variable
*var
= ir
->as_variable();
2441 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2442 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2443 void *mem_ctx
= ralloc_parent(var
);
2444 var
->data
.location
= 0;
2445 var
->data
.explicit_location
= false;
2447 var
->data
.mode
= ir_var_auto
;
2448 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2450 var
->data
.mode
= ir_var_uniform
;
2451 var
->data
.how_declared
= ir_var_hidden
;
2452 var
->allocate_state_slots(1);
2453 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2454 slot0
->swizzle
= SWIZZLE_XXXX
;
2455 slot0
->tokens
[0] = STATE_INTERNAL
;
2456 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2457 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2458 slot0
->tokens
[i
] = 0;
2466 * Find a contiguous set of available bits in a bitmask.
2468 * \param used_mask Bits representing used (1) and unused (0) locations
2469 * \param needed_count Number of contiguous bits needed.
2472 * Base location of the available bits on success or -1 on failure.
2475 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2477 unsigned needed_mask
= (1 << needed_count
) - 1;
2478 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2480 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2481 * cannot optimize possibly infinite loops" for the loop below.
2483 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2486 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2487 if ((needed_mask
& ~used_mask
) == needed_mask
)
2498 * Assign locations for either VS inputs or FS outputs
2500 * \param mem_ctx Temporary ralloc context used for linking
2501 * \param prog Shader program whose variables need locations assigned
2502 * \param constants Driver specific constant values for the program.
2503 * \param target_index Selector for the program target to receive location
2504 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2505 * \c MESA_SHADER_FRAGMENT.
2508 * If locations are successfully assigned, true is returned. Otherwise an
2509 * error is emitted to the shader link log and false is returned.
2512 assign_attribute_or_color_locations(void *mem_ctx
,
2513 gl_shader_program
*prog
,
2514 struct gl_constants
*constants
,
2515 unsigned target_index
)
2517 /* Maximum number of generic locations. This corresponds to either the
2518 * maximum number of draw buffers or the maximum number of generic
2521 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2522 constants
->Program
[target_index
].MaxAttribs
:
2523 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2525 /* Mark invalid locations as being used.
2527 unsigned used_locations
= (max_index
>= 32)
2528 ? ~0 : ~((1 << max_index
) - 1);
2529 unsigned double_storage_locations
= 0;
2531 assert((target_index
== MESA_SHADER_VERTEX
)
2532 || (target_index
== MESA_SHADER_FRAGMENT
));
2534 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2538 /* Operate in a total of four passes.
2540 * 1. Invalidate the location assignments for all vertex shader inputs.
2542 * 2. Assign locations for inputs that have user-defined (via
2543 * glBindVertexAttribLocation) locations and outputs that have
2544 * user-defined locations (via glBindFragDataLocation).
2546 * 3. Sort the attributes without assigned locations by number of slots
2547 * required in decreasing order. Fragmentation caused by attribute
2548 * locations assigned by the application may prevent large attributes
2549 * from having enough contiguous space.
2551 * 4. Assign locations to any inputs without assigned locations.
2554 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2555 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2557 const enum ir_variable_mode direction
=
2558 (target_index
== MESA_SHADER_VERTEX
)
2559 ? ir_var_shader_in
: ir_var_shader_out
;
2562 /* Temporary storage for the set of attributes that need locations assigned.
2568 /* Used below in the call to qsort. */
2569 static int compare(const void *a
, const void *b
)
2571 const temp_attr
*const l
= (const temp_attr
*) a
;
2572 const temp_attr
*const r
= (const temp_attr
*) b
;
2574 /* Reversed because we want a descending order sort below. */
2575 return r
->slots
- l
->slots
;
2578 assert(max_index
<= 32);
2580 /* Temporary array for the set of attributes that have locations assigned.
2582 ir_variable
*assigned
[16];
2584 unsigned num_attr
= 0;
2585 unsigned assigned_attr
= 0;
2587 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2588 ir_variable
*const var
= node
->as_variable();
2590 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2593 if (var
->data
.explicit_location
) {
2594 var
->data
.is_unmatched_generic_inout
= 0;
2595 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2596 || (var
->data
.location
< 0)) {
2598 "invalid explicit location %d specified for `%s'\n",
2599 (var
->data
.location
< 0)
2600 ? var
->data
.location
2601 : var
->data
.location
- generic_base
,
2605 } else if (target_index
== MESA_SHADER_VERTEX
) {
2608 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2609 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2610 var
->data
.location
= binding
;
2611 var
->data
.is_unmatched_generic_inout
= 0;
2613 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2616 const char *name
= var
->name
;
2617 const glsl_type
*type
= var
->type
;
2620 /* Check if there's a binding for the variable name */
2621 if (prog
->FragDataBindings
->get(binding
, name
)) {
2622 assert(binding
>= FRAG_RESULT_DATA0
);
2623 var
->data
.location
= binding
;
2624 var
->data
.is_unmatched_generic_inout
= 0;
2626 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2627 var
->data
.index
= index
;
2632 /* If not, but it's an array type, look for name[0] */
2633 if (type
->is_array()) {
2634 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2635 type
= type
->fields
.array
;
2643 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2646 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2648 * "Output binding assignments will cause LinkProgram to fail:
2650 * If the program has an active output assigned to a location greater
2651 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2652 * an active output assigned an index greater than or equal to one;"
2654 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2655 var
->data
.location
- generic_base
>=
2656 (int) constants
->MaxDualSourceDrawBuffers
) {
2658 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2659 "with index %u for %s\n",
2660 var
->data
.location
- generic_base
, var
->data
.index
,
2665 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2667 /* If the variable is not a built-in and has a location statically
2668 * assigned in the shader (presumably via a layout qualifier), make sure
2669 * that it doesn't collide with other assigned locations. Otherwise,
2670 * add it to the list of variables that need linker-assigned locations.
2672 if (var
->data
.location
!= -1) {
2673 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2674 /* From page 61 of the OpenGL 4.0 spec:
2676 * "LinkProgram will fail if the attribute bindings assigned
2677 * by BindAttribLocation do not leave not enough space to
2678 * assign a location for an active matrix attribute or an
2679 * active attribute array, both of which require multiple
2680 * contiguous generic attributes."
2682 * I think above text prohibits the aliasing of explicit and
2683 * automatic assignments. But, aliasing is allowed in manual
2684 * assignments of attribute locations. See below comments for
2687 * From OpenGL 4.0 spec, page 61:
2689 * "It is possible for an application to bind more than one
2690 * attribute name to the same location. This is referred to as
2691 * aliasing. This will only work if only one of the aliased
2692 * attributes is active in the executable program, or if no
2693 * path through the shader consumes more than one attribute of
2694 * a set of attributes aliased to the same location. A link
2695 * error can occur if the linker determines that every path
2696 * through the shader consumes multiple aliased attributes,
2697 * but implementations are not required to generate an error
2700 * From GLSL 4.30 spec, page 54:
2702 * "A program will fail to link if any two non-vertex shader
2703 * input variables are assigned to the same location. For
2704 * vertex shaders, multiple input variables may be assigned
2705 * to the same location using either layout qualifiers or via
2706 * the OpenGL API. However, such aliasing is intended only to
2707 * support vertex shaders where each execution path accesses
2708 * at most one input per each location. Implementations are
2709 * permitted, but not required, to generate link-time errors
2710 * if they detect that every path through the vertex shader
2711 * executable accesses multiple inputs assigned to any single
2712 * location. For all shader types, a program will fail to link
2713 * if explicit location assignments leave the linker unable
2714 * to find space for other variables without explicit
2717 * From OpenGL ES 3.0 spec, page 56:
2719 * "Binding more than one attribute name to the same location
2720 * is referred to as aliasing, and is not permitted in OpenGL
2721 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2722 * fail when this condition exists. However, aliasing is
2723 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2724 * This will only work if only one of the aliased attributes
2725 * is active in the executable program, or if no path through
2726 * the shader consumes more than one attribute of a set of
2727 * attributes aliased to the same location. A link error can
2728 * occur if the linker determines that every path through the
2729 * shader consumes multiple aliased attributes, but implemen-
2730 * tations are not required to generate an error in this case."
2732 * After looking at above references from OpenGL, OpenGL ES and
2733 * GLSL specifications, we allow aliasing of vertex input variables
2734 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2736 * NOTE: This is not required by the spec but its worth mentioning
2737 * here that we're not doing anything to make sure that no path
2738 * through the vertex shader executable accesses multiple inputs
2739 * assigned to any single location.
2742 /* Mask representing the contiguous slots that will be used by
2745 const unsigned attr
= var
->data
.location
- generic_base
;
2746 const unsigned use_mask
= (1 << slots
) - 1;
2747 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2748 ? "vertex shader input" : "fragment shader output";
2750 /* Generate a link error if the requested locations for this
2751 * attribute exceed the maximum allowed attribute location.
2753 if (attr
+ slots
> max_index
) {
2755 "insufficient contiguous locations "
2756 "available for %s `%s' %d %d %d\n", string
,
2757 var
->name
, used_locations
, use_mask
, attr
);
2761 /* Generate a link error if the set of bits requested for this
2762 * attribute overlaps any previously allocated bits.
2764 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2765 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2766 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2769 * "Additionally, for fragment shader outputs, if two
2770 * variables are placed within the same location, they
2771 * must have the same underlying type (floating-point or
2772 * integer). No component aliasing of output variables or
2773 * members is allowed.
2775 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2776 unsigned assigned_slots
=
2777 assigned
[i
]->type
->count_attribute_slots(false);
2778 unsigned assig_attr
=
2779 assigned
[i
]->data
.location
- generic_base
;
2780 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2782 if ((assigned_use_mask
<< assig_attr
) &
2783 (use_mask
<< attr
)) {
2785 const glsl_type
*assigned_type
=
2786 assigned
[i
]->type
->without_array();
2787 const glsl_type
*type
= var
->type
->without_array();
2788 if (assigned_type
->base_type
!= type
->base_type
) {
2789 linker_error(prog
, "types do not match for aliased"
2790 " %ss %s and %s\n", string
,
2791 assigned
[i
]->name
, var
->name
);
2795 unsigned assigned_component_mask
=
2796 ((1 << assigned_type
->vector_elements
) - 1) <<
2797 assigned
[i
]->data
.location_frac
;
2798 unsigned component_mask
=
2799 ((1 << type
->vector_elements
) - 1) <<
2800 var
->data
.location_frac
;
2801 if (assigned_component_mask
& component_mask
) {
2802 linker_error(prog
, "overlapping component is "
2803 "assigned to %ss %s and %s "
2805 string
, assigned
[i
]->name
, var
->name
,
2806 var
->data
.location_frac
);
2811 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2812 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2813 linker_error(prog
, "overlapping location is assigned "
2814 "to %s `%s' %d %d %d\n", string
, var
->name
,
2815 used_locations
, use_mask
, attr
);
2818 linker_warning(prog
, "overlapping location is assigned "
2819 "to %s `%s' %d %d %d\n", string
, var
->name
,
2820 used_locations
, use_mask
, attr
);
2824 used_locations
|= (use_mask
<< attr
);
2826 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2828 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2829 * active attribute variables may fail to link, unless
2830 * device-dependent optimizations are able to make the program
2831 * fit within available hardware resources. For the purposes
2832 * of this test, attribute variables of the type dvec3, dvec4,
2833 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2834 * count as consuming twice as many attributes as equivalent
2835 * single-precision types. While these types use the same number
2836 * of generic attributes as their single-precision equivalents,
2837 * implementations are permitted to consume two single-precision
2838 * vectors of internal storage for each three- or four-component
2839 * double-precision vector."
2841 * Mark this attribute slot as taking up twice as much space
2842 * so we can count it properly against limits. According to
2843 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2844 * is optional behavior, but it seems preferable.
2846 if (var
->type
->without_array()->is_dual_slot())
2847 double_storage_locations
|= (use_mask
<< attr
);
2850 assigned
[assigned_attr
] = var
;
2856 if (num_attr
>= max_index
) {
2857 linker_error(prog
, "too many %s (max %u)",
2858 target_index
== MESA_SHADER_VERTEX
?
2859 "vertex shader inputs" : "fragment shader outputs",
2863 to_assign
[num_attr
].slots
= slots
;
2864 to_assign
[num_attr
].var
= var
;
2868 if (target_index
== MESA_SHADER_VERTEX
) {
2869 unsigned total_attribs_size
=
2870 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2871 _mesa_bitcount(double_storage_locations
);
2872 if (total_attribs_size
> max_index
) {
2874 "attempt to use %d vertex attribute slots only %d available ",
2875 total_attribs_size
, max_index
);
2880 /* If all of the attributes were assigned locations by the application (or
2881 * are built-in attributes with fixed locations), return early. This should
2882 * be the common case.
2887 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2889 if (target_index
== MESA_SHADER_VERTEX
) {
2890 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2891 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2892 * reserved to prevent it from being automatically allocated below.
2894 find_deref_visitor
find("gl_Vertex");
2896 if (find
.variable_found())
2897 used_locations
|= (1 << 0);
2900 for (unsigned i
= 0; i
< num_attr
; i
++) {
2901 /* Mask representing the contiguous slots that will be used by this
2904 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2906 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2909 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2910 ? "vertex shader input" : "fragment shader output";
2913 "insufficient contiguous locations "
2914 "available for %s `%s'\n",
2915 string
, to_assign
[i
].var
->name
);
2919 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2920 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2921 used_locations
|= (use_mask
<< location
);
2923 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2924 double_storage_locations
|= (use_mask
<< location
);
2927 /* Now that we have all the locations, from the GL 4.5 core spec, section
2928 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2929 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2930 * as equivalent single-precision types.
2932 if (target_index
== MESA_SHADER_VERTEX
) {
2933 unsigned total_attribs_size
=
2934 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2935 _mesa_bitcount(double_storage_locations
);
2936 if (total_attribs_size
> max_index
) {
2938 "attempt to use %d vertex attribute slots only %d available ",
2939 total_attribs_size
, max_index
);
2948 * Match explicit locations of outputs to inputs and deactivate the
2949 * unmatch flag if found so we don't optimise them away.
2952 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2953 gl_linked_shader
*consumer
)
2955 glsl_symbol_table parameters
;
2956 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2959 /* Find all shader outputs in the "producer" stage.
2961 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2962 ir_variable
*const var
= node
->as_variable();
2964 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2967 if (var
->data
.explicit_location
&&
2968 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2969 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2970 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2971 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2975 /* Match inputs to outputs */
2976 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2977 ir_variable
*const input
= node
->as_variable();
2979 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2982 ir_variable
*output
= NULL
;
2983 if (input
->data
.explicit_location
2984 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2985 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2986 [input
->data
.location_frac
];
2988 if (output
!= NULL
){
2989 input
->data
.is_unmatched_generic_inout
= 0;
2990 output
->data
.is_unmatched_generic_inout
= 0;
2997 * Store the gl_FragDepth layout in the gl_shader_program struct.
3000 store_fragdepth_layout(struct gl_shader_program
*prog
)
3002 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3006 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3008 /* We don't look up the gl_FragDepth symbol directly because if
3009 * gl_FragDepth is not used in the shader, it's removed from the IR.
3010 * However, the symbol won't be removed from the symbol table.
3012 * We're only interested in the cases where the variable is NOT removed
3015 foreach_in_list(ir_instruction
, node
, ir
) {
3016 ir_variable
*const var
= node
->as_variable();
3018 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3022 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3023 switch (var
->data
.depth_layout
) {
3024 case ir_depth_layout_none
:
3025 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3027 case ir_depth_layout_any
:
3028 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3030 case ir_depth_layout_greater
:
3031 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3033 case ir_depth_layout_less
:
3034 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3036 case ir_depth_layout_unchanged
:
3037 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3048 * Validate the resources used by a program versus the implementation limits
3051 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3053 unsigned total_uniform_blocks
= 0;
3054 unsigned total_shader_storage_blocks
= 0;
3056 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3057 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3062 if (sh
->Program
->info
.num_textures
>
3063 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3064 linker_error(prog
, "Too many %s shader texture samplers\n",
3065 _mesa_shader_stage_to_string(i
));
3068 if (sh
->num_uniform_components
>
3069 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3070 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3071 linker_warning(prog
, "Too many %s shader default uniform block "
3072 "components, but the driver will try to optimize "
3073 "them out; this is non-portable out-of-spec "
3075 _mesa_shader_stage_to_string(i
));
3077 linker_error(prog
, "Too many %s shader default uniform block "
3079 _mesa_shader_stage_to_string(i
));
3083 if (sh
->num_combined_uniform_components
>
3084 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3085 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3086 linker_warning(prog
, "Too many %s shader uniform components, "
3087 "but the driver will try to optimize them out; "
3088 "this is non-portable out-of-spec behavior\n",
3089 _mesa_shader_stage_to_string(i
));
3091 linker_error(prog
, "Too many %s shader uniform components\n",
3092 _mesa_shader_stage_to_string(i
));
3096 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3097 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3099 const unsigned max_uniform_blocks
=
3100 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3101 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3102 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3103 _mesa_shader_stage_to_string(i
),
3104 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3107 const unsigned max_shader_storage_blocks
=
3108 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3109 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3110 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3111 _mesa_shader_stage_to_string(i
),
3112 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3116 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3117 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3118 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3121 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3122 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3123 total_shader_storage_blocks
,
3124 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3127 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3128 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3129 ctx
->Const
.MaxUniformBlockSize
) {
3130 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3131 prog
->data
->UniformBlocks
[i
].Name
,
3132 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3133 ctx
->Const
.MaxUniformBlockSize
);
3137 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3138 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3139 ctx
->Const
.MaxShaderStorageBlockSize
) {
3140 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3141 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3142 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3143 ctx
->Const
.MaxShaderStorageBlockSize
);
3149 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3151 unsigned mask
= prog
->data
->linked_stages
;
3153 const int i
= u_bit_scan(&mask
);
3154 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3156 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3157 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3160 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3166 if (p
->sh
.NumSubroutineFunctions
== 0) {
3167 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3170 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3171 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3172 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3173 if (fn
->types
[k
] == uni
->type
) {
3179 uni
->num_compatible_subroutines
= count
;
3185 check_subroutine_resources(struct gl_shader_program
*prog
)
3187 unsigned mask
= prog
->data
->linked_stages
;
3189 const int i
= u_bit_scan(&mask
);
3190 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3192 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3193 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3194 _mesa_shader_stage_to_string(i
));
3199 * Validate shader image resources.
3202 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3204 unsigned total_image_units
= 0;
3205 unsigned fragment_outputs
= 0;
3206 unsigned total_shader_storage_blocks
= 0;
3208 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3211 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3212 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3215 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3216 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3217 _mesa_shader_stage_to_string(i
),
3218 sh
->Program
->info
.num_images
,
3219 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3221 total_image_units
+= sh
->Program
->info
.num_images
;
3222 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3224 if (i
== MESA_SHADER_FRAGMENT
) {
3225 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3226 ir_variable
*var
= node
->as_variable();
3227 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3228 /* since there are no double fs outputs - pass false */
3229 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3235 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3236 linker_error(prog
, "Too many combined image uniforms\n");
3238 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3239 ctx
->Const
.MaxCombinedShaderOutputResources
)
3240 linker_error(prog
, "Too many combined image uniforms, shader storage "
3241 " buffers and fragment outputs\n");
3246 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3247 * for a variable, checks for overlaps between other uniforms using explicit
3251 reserve_explicit_locations(struct gl_shader_program
*prog
,
3252 string_to_uint_map
*map
, ir_variable
*var
)
3254 unsigned slots
= var
->type
->uniform_locations();
3255 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3256 unsigned return_value
= slots
;
3258 /* Resize remap table if locations do not fit in the current one. */
3259 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3260 prog
->UniformRemapTable
=
3261 reralloc(prog
, prog
->UniformRemapTable
,
3262 gl_uniform_storage
*,
3265 if (!prog
->UniformRemapTable
) {
3266 linker_error(prog
, "Out of memory during linking.\n");
3270 /* Initialize allocated space. */
3271 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3272 prog
->UniformRemapTable
[i
] = NULL
;
3274 prog
->NumUniformRemapTable
= max_loc
+ 1;
3277 for (unsigned i
= 0; i
< slots
; i
++) {
3278 unsigned loc
= var
->data
.location
+ i
;
3280 /* Check if location is already used. */
3281 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3283 /* Possibly same uniform from a different stage, this is ok. */
3285 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3290 /* ARB_explicit_uniform_location specification states:
3292 * "No two default-block uniform variables in the program can have
3293 * the same location, even if they are unused, otherwise a compiler
3294 * or linker error will be generated."
3297 "location qualifier for uniform %s overlaps "
3298 "previously used location\n",
3303 /* Initialize location as inactive before optimization
3304 * rounds and location assignment.
3306 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3309 /* Note, base location used for arrays. */
3310 map
->put(var
->data
.location
, var
->name
);
3312 return return_value
;
3316 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3317 struct gl_program
*p
,
3320 unsigned slots
= var
->type
->uniform_locations();
3321 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3323 /* Resize remap table if locations do not fit in the current one. */
3324 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3325 p
->sh
.SubroutineUniformRemapTable
=
3326 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3327 gl_uniform_storage
*,
3330 if (!p
->sh
.SubroutineUniformRemapTable
) {
3331 linker_error(prog
, "Out of memory during linking.\n");
3335 /* Initialize allocated space. */
3336 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3337 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3339 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3342 for (unsigned i
= 0; i
< slots
; i
++) {
3343 unsigned loc
= var
->data
.location
+ i
;
3345 /* Check if location is already used. */
3346 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3348 /* ARB_explicit_uniform_location specification states:
3349 * "No two subroutine uniform variables can have the same location
3350 * in the same shader stage, otherwise a compiler or linker error
3351 * will be generated."
3354 "location qualifier for uniform %s overlaps "
3355 "previously used location\n",
3360 /* Initialize location as inactive before optimization
3361 * rounds and location assignment.
3363 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3369 * Check and reserve all explicit uniform locations, called before
3370 * any optimizations happen to handle also inactive uniforms and
3371 * inactive array elements that may get trimmed away.
3374 check_explicit_uniform_locations(struct gl_context
*ctx
,
3375 struct gl_shader_program
*prog
)
3377 prog
->NumExplicitUniformLocations
= 0;
3379 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3382 /* This map is used to detect if overlapping explicit locations
3383 * occur with the same uniform (from different stage) or a different one.
3385 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3388 linker_error(prog
, "Out of memory during linking.\n");
3392 unsigned entries_total
= 0;
3393 unsigned mask
= prog
->data
->linked_stages
;
3395 const int i
= u_bit_scan(&mask
);
3396 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3398 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3399 ir_variable
*var
= node
->as_variable();
3400 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3403 if (var
->data
.explicit_location
) {
3405 if (var
->type
->without_array()->is_subroutine())
3406 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3408 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3412 entries_total
+= slots
;
3423 struct empty_uniform_block
*current_block
= NULL
;
3425 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3426 /* We found empty space in UniformRemapTable. */
3427 if (prog
->UniformRemapTable
[i
] == NULL
) {
3428 /* We've found the beginning of a new continous block of empty slots */
3429 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3430 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3431 current_block
->start
= i
;
3432 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3433 ¤t_block
->link
);
3436 /* The current block continues, so we simply increment its slots */
3437 current_block
->slots
++;
3442 prog
->NumExplicitUniformLocations
= entries_total
;
3446 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3447 GLenum type
, const char *name
)
3449 bool found_interface
= false;
3450 unsigned block_name_len
= 0;
3451 const char *block_name_dot
= strchr(name
, '.');
3453 /* These rules only apply to buffer variables. So we return
3454 * true for the rest of types.
3456 if (type
!= GL_BUFFER_VARIABLE
)
3459 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3460 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3461 block_name_len
= strlen(block_name
);
3463 const char *block_square_bracket
= strchr(block_name
, '[');
3464 if (block_square_bracket
) {
3465 /* The block is part of an array of named interfaces,
3466 * for the name comparison we ignore the "[x]" part.
3468 block_name_len
-= strlen(block_square_bracket
);
3471 if (block_name_dot
) {
3472 /* Check if the variable name starts with the interface
3473 * name. The interface name (if present) should have the
3474 * length than the interface block name we are comparing to.
3476 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3477 if (len
!= block_name_len
)
3481 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3482 found_interface
= true;
3487 /* We remove the interface name from the buffer variable name,
3488 * including the dot that follows it.
3490 if (found_interface
)
3491 name
= name
+ block_name_len
+ 1;
3493 /* The ARB_program_interface_query spec says:
3495 * "For an active shader storage block member declared as an array, an
3496 * entry will be generated only for the first array element, regardless
3497 * of its type. For arrays of aggregate types, the enumeration rules
3498 * are applied recursively for the single enumerated array element."
3500 const char *struct_first_dot
= strchr(name
, '.');
3501 const char *first_square_bracket
= strchr(name
, '[');
3503 /* The buffer variable is on top level and it is not an array */
3504 if (!first_square_bracket
) {
3506 /* The shader storage block member is a struct, then generate the entry */
3507 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3510 /* Shader storage block member is an array, only generate an entry for the
3511 * first array element.
3513 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3521 add_program_resource(struct gl_shader_program
*prog
,
3522 struct set
*resource_set
,
3523 GLenum type
, const void *data
, uint8_t stages
)
3527 /* If resource already exists, do not add it again. */
3528 if (_mesa_set_search(resource_set
, data
))
3531 prog
->data
->ProgramResourceList
=
3533 prog
->data
->ProgramResourceList
,
3534 gl_program_resource
,
3535 prog
->data
->NumProgramResourceList
+ 1);
3537 if (!prog
->data
->ProgramResourceList
) {
3538 linker_error(prog
, "Out of memory during linking.\n");
3542 struct gl_program_resource
*res
=
3543 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3547 res
->StageReferences
= stages
;
3549 prog
->data
->NumProgramResourceList
++;
3551 _mesa_set_add(resource_set
, data
);
3556 /* Function checks if a variable var is a packed varying and
3557 * if given name is part of packed varying's list.
3559 * If a variable is a packed varying, it has a name like
3560 * 'packed:a,b,c' where a, b and c are separate variables.
3563 included_in_packed_varying(ir_variable
*var
, const char *name
)
3565 if (strncmp(var
->name
, "packed:", 7) != 0)
3568 char *list
= strdup(var
->name
+ 7);
3573 char *token
= strtok_r(list
, ",", &saveptr
);
3575 if (strcmp(token
, name
) == 0) {
3579 token
= strtok_r(NULL
, ",", &saveptr
);
3586 * Function builds a stage reference bitmask from variable name.
3589 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3594 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3595 * used for reference mask in gl_program_resource will need to be changed.
3597 assert(MESA_SHADER_STAGES
< 8);
3599 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3600 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3604 /* Shader symbol table may contain variables that have
3605 * been optimized away. Search IR for the variable instead.
3607 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3608 ir_variable
*var
= node
->as_variable();
3610 unsigned baselen
= strlen(var
->name
);
3612 if (included_in_packed_varying(var
, name
)) {
3617 /* Type needs to match if specified, otherwise we might
3618 * pick a variable with same name but different interface.
3620 if (var
->data
.mode
!= mode
)
3623 if (strncmp(var
->name
, name
, baselen
) == 0) {
3624 /* Check for exact name matches but also check for arrays and
3627 if (name
[baselen
] == '\0' ||
3628 name
[baselen
] == '[' ||
3629 name
[baselen
] == '.') {
3641 * Create gl_shader_variable from ir_variable class.
3643 static gl_shader_variable
*
3644 create_shader_variable(struct gl_shader_program
*shProg
,
3645 const ir_variable
*in
,
3646 const char *name
, const glsl_type
*type
,
3647 const glsl_type
*interface_type
,
3648 bool use_implicit_location
, int location
,
3649 const glsl_type
*outermost_struct_type
)
3651 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3655 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3656 * expect to see gl_VertexID in the program resource list. Pretend.
3658 if (in
->data
.mode
== ir_var_system_value
&&
3659 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3660 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3661 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3662 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3663 (in
->data
.mode
== ir_var_system_value
&&
3664 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3665 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3666 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3667 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3668 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3669 (in
->data
.mode
== ir_var_system_value
&&
3670 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3671 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3672 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3674 out
->name
= ralloc_strdup(shProg
, name
);
3680 /* The ARB_program_interface_query spec says:
3682 * "Not all active variables are assigned valid locations; the
3683 * following variables will have an effective location of -1:
3685 * * uniforms declared as atomic counters;
3687 * * members of a uniform block;
3689 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3691 * * inputs or outputs not declared with a "location" layout
3692 * qualifier, except for vertex shader inputs and fragment shader
3695 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3696 is_gl_identifier(in
->name
) ||
3697 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3700 out
->location
= location
;
3704 out
->outermost_struct_type
= outermost_struct_type
;
3705 out
->interface_type
= interface_type
;
3706 out
->component
= in
->data
.location_frac
;
3707 out
->index
= in
->data
.index
;
3708 out
->patch
= in
->data
.patch
;
3709 out
->mode
= in
->data
.mode
;
3710 out
->interpolation
= in
->data
.interpolation
;
3711 out
->explicit_location
= in
->data
.explicit_location
;
3712 out
->precision
= in
->data
.precision
;
3717 static const glsl_type
*
3718 resize_to_max_patch_vertices(const struct gl_context
*ctx
,
3719 const glsl_type
*type
)
3724 return glsl_type::get_array_instance(type
->fields
.array
,
3725 ctx
->Const
.MaxPatchVertices
);
3729 add_shader_variable(const struct gl_context
*ctx
,
3730 struct gl_shader_program
*shProg
,
3731 struct set
*resource_set
,
3732 unsigned stage_mask
,
3733 GLenum programInterface
, ir_variable
*var
,
3734 const char *name
, const glsl_type
*type
,
3735 bool use_implicit_location
, int location
,
3736 const glsl_type
*outermost_struct_type
= NULL
)
3738 const glsl_type
*interface_type
= var
->get_interface_type();
3740 if (outermost_struct_type
== NULL
) {
3741 /* Unsized (non-patch) TCS output/TES input arrays are implicitly
3742 * sized to gl_MaxPatchVertices. Internally, we shrink them to a
3745 * This can cause trouble with SSO programs. Since the TCS declares
3746 * the number of output vertices, we can always shrink TCS output
3747 * arrays. However, the TES might not be linked with a TCS, in
3748 * which case it won't know the size of the patch. In other words,
3749 * the TCS and TES may disagree on the (smaller) array sizes. This
3750 * can result in the resource names differing across stages, causing
3751 * SSO validation failures and other cascading issues.
3753 * Expanding the array size to the full gl_MaxPatchVertices fixes
3754 * these issues. It's also what program interface queries expect,
3755 * as that is the official size of the array.
3757 if (var
->data
.tess_varying_implicit_sized_array
) {
3758 type
= resize_to_max_patch_vertices(ctx
, type
);
3759 interface_type
= resize_to_max_patch_vertices(ctx
, interface_type
);
3762 if (var
->data
.from_named_ifc_block
) {
3763 const char *interface_name
= interface_type
->name
;
3765 if (interface_type
->is_array()) {
3766 /* Issue #16 of the ARB_program_interface_query spec says:
3768 * "* If a variable is a member of an interface block without an
3769 * instance name, it is enumerated using just the variable name.
3771 * * If a variable is a member of an interface block with an
3772 * instance name, it is enumerated as "BlockName.Member", where
3773 * "BlockName" is the name of the interface block (not the
3774 * instance name) and "Member" is the name of the variable."
3776 * In particular, it indicates that it should be "BlockName",
3777 * not "BlockName[array length]". The conformance suite and
3778 * dEQP both require this behavior.
3780 * Here, we unwrap the extra array level added by named interface
3781 * block array lowering so we have the correct variable type. We
3782 * also unwrap the interface type when constructing the name.
3784 * We leave interface_type the same so that ES 3.x SSO pipeline
3785 * validation can enforce the rules requiring array length to
3786 * match on interface blocks.
3788 type
= type
->fields
.array
;
3790 interface_name
= interface_type
->fields
.array
->name
;
3793 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3797 switch (type
->base_type
) {
3798 case GLSL_TYPE_STRUCT
: {
3799 /* The ARB_program_interface_query spec says:
3801 * "For an active variable declared as a structure, a separate entry
3802 * will be generated for each active structure member. The name of
3803 * each entry is formed by concatenating the name of the structure,
3804 * the "." character, and the name of the structure member. If a
3805 * structure member to enumerate is itself a structure or array,
3806 * these enumeration rules are applied recursively."
3808 if (outermost_struct_type
== NULL
)
3809 outermost_struct_type
= type
;
3811 unsigned field_location
= location
;
3812 for (unsigned i
= 0; i
< type
->length
; i
++) {
3813 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3814 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3815 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3816 stage_mask
, programInterface
,
3817 var
, field_name
, field
->type
,
3818 use_implicit_location
, field_location
,
3819 outermost_struct_type
))
3822 field_location
+= field
->type
->count_attribute_slots(false);
3828 /* The ARB_program_interface_query spec says:
3830 * "For an active variable declared as a single instance of a basic
3831 * type, a single entry will be generated, using the variable name
3832 * from the shader source."
3834 gl_shader_variable
*sha_v
=
3835 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3836 use_implicit_location
, location
,
3837 outermost_struct_type
);
3841 return add_program_resource(shProg
, resource_set
,
3842 programInterface
, sha_v
, stage_mask
);
3848 add_interface_variables(const struct gl_context
*ctx
,
3849 struct gl_shader_program
*shProg
,
3850 struct set
*resource_set
,
3851 unsigned stage
, GLenum programInterface
)
3853 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3855 foreach_in_list(ir_instruction
, node
, ir
) {
3856 ir_variable
*var
= node
->as_variable();
3858 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3863 switch (var
->data
.mode
) {
3864 case ir_var_system_value
:
3865 case ir_var_shader_in
:
3866 if (programInterface
!= GL_PROGRAM_INPUT
)
3868 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3869 : int(VARYING_SLOT_VAR0
);
3871 case ir_var_shader_out
:
3872 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3874 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3875 : int(VARYING_SLOT_VAR0
);
3881 if (var
->data
.patch
)
3882 loc_bias
= int(VARYING_SLOT_PATCH0
);
3884 /* Skip packed varyings, packed varyings are handled separately
3885 * by add_packed_varyings.
3887 if (strncmp(var
->name
, "packed:", 7) == 0)
3890 /* Skip fragdata arrays, these are handled separately
3891 * by add_fragdata_arrays.
3893 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3896 const bool vs_input_or_fs_output
=
3897 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3898 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3900 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3901 1 << stage
, programInterface
,
3902 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3903 var
->data
.location
- loc_bias
))
3910 add_packed_varyings(const struct gl_context
*ctx
,
3911 struct gl_shader_program
*shProg
,
3912 struct set
*resource_set
,
3913 int stage
, GLenum type
)
3915 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3918 if (!sh
|| !sh
->packed_varyings
)
3921 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3922 ir_variable
*var
= node
->as_variable();
3924 switch (var
->data
.mode
) {
3925 case ir_var_shader_in
:
3926 iface
= GL_PROGRAM_INPUT
;
3928 case ir_var_shader_out
:
3929 iface
= GL_PROGRAM_OUTPUT
;
3932 unreachable("unexpected type");
3935 if (type
== iface
) {
3936 const int stage_mask
=
3937 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3938 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3940 iface
, var
, var
->name
, var
->type
, false,
3941 var
->data
.location
- VARYING_SLOT_VAR0
))
3950 add_fragdata_arrays(const struct gl_context
*ctx
,
3951 struct gl_shader_program
*shProg
,
3952 struct set
*resource_set
)
3954 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3956 if (!sh
|| !sh
->fragdata_arrays
)
3959 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3960 ir_variable
*var
= node
->as_variable();
3962 assert(var
->data
.mode
== ir_var_shader_out
);
3964 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3965 1 << MESA_SHADER_FRAGMENT
,
3966 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3967 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3975 get_top_level_name(const char *name
)
3977 const char *first_dot
= strchr(name
, '.');
3978 const char *first_square_bracket
= strchr(name
, '[');
3981 /* The ARB_program_interface_query spec says:
3983 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3984 * the number of active array elements of the top-level shader storage
3985 * block member containing to the active variable is written to
3986 * <params>. If the top-level block member is not declared as an
3987 * array, the value one is written to <params>. If the top-level block
3988 * member is an array with no declared size, the value zero is written
3992 /* The buffer variable is on top level.*/
3993 if (!first_square_bracket
&& !first_dot
)
3994 name_size
= strlen(name
);
3995 else if ((!first_square_bracket
||
3996 (first_dot
&& first_dot
< first_square_bracket
)))
3997 name_size
= first_dot
- name
;
3999 name_size
= first_square_bracket
- name
;
4001 return strndup(name
, name_size
);
4005 get_var_name(const char *name
)
4007 const char *first_dot
= strchr(name
, '.');
4010 return strdup(name
);
4012 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4016 is_top_level_shader_storage_block_member(const char* name
,
4017 const char* interface_name
,
4018 const char* field_name
)
4020 bool result
= false;
4022 /* If the given variable is already a top-level shader storage
4023 * block member, then return array_size = 1.
4024 * We could have two possibilities: if we have an instanced
4025 * shader storage block or not instanced.
4027 * For the first, we check create a name as it was in top level and
4028 * compare it with the real name. If they are the same, then
4029 * the variable is already at top-level.
4031 * Full instanced name is: interface name + '.' + var name +
4034 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4035 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4036 if (!full_instanced_name
) {
4037 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4041 snprintf(full_instanced_name
, name_length
, "%s.%s",
4042 interface_name
, field_name
);
4044 /* Check if its top-level shader storage block member of an
4045 * instanced interface block, or of a unnamed interface block.
4047 if (strcmp(name
, full_instanced_name
) == 0 ||
4048 strcmp(name
, field_name
) == 0)
4051 free(full_instanced_name
);
4056 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4057 char *interface_name
, char *var_name
)
4059 /* The ARB_program_interface_query spec says:
4061 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4062 * the number of active array elements of the top-level shader storage
4063 * block member containing to the active variable is written to
4064 * <params>. If the top-level block member is not declared as an
4065 * array, the value one is written to <params>. If the top-level block
4066 * member is an array with no declared size, the value zero is written
4069 if (is_top_level_shader_storage_block_member(uni
->name
,
4073 else if (field
->type
->is_unsized_array())
4075 else if (field
->type
->is_array())
4076 return field
->type
->length
;
4082 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4083 const glsl_struct_field
*field
, char *interface_name
,
4086 /* The ARB_program_interface_query spec says:
4088 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4089 * identifying the stride between array elements of the top-level
4090 * shader storage block member containing the active variable is
4091 * written to <params>. For top-level block members declared as
4092 * arrays, the value written is the difference, in basic machine units,
4093 * between the offsets of the active variable for consecutive elements
4094 * in the top-level array. For top-level block members not declared as
4095 * an array, zero is written to <params>."
4097 if (field
->type
->is_array()) {
4098 const enum glsl_matrix_layout matrix_layout
=
4099 glsl_matrix_layout(field
->matrix_layout
);
4100 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4101 const glsl_type
*array_type
= field
->type
->fields
.array
;
4103 if (is_top_level_shader_storage_block_member(uni
->name
,
4108 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4109 if (array_type
->is_record() || array_type
->is_array())
4110 return glsl_align(array_type
->std140_size(row_major
), 16);
4112 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4114 return array_type
->std430_array_stride(row_major
);
4121 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4122 struct gl_uniform_storage
*uni
)
4124 int block_index
= uni
->block_index
;
4125 int array_size
= -1;
4126 int array_stride
= -1;
4127 char *var_name
= get_top_level_name(uni
->name
);
4128 char *interface_name
=
4129 get_top_level_name(uni
->is_shader_storage
?
4130 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4131 shProg
->data
->UniformBlocks
[block_index
].Name
);
4133 if (strcmp(var_name
, interface_name
) == 0) {
4134 /* Deal with instanced array of SSBOs */
4135 char *temp_name
= get_var_name(uni
->name
);
4137 linker_error(shProg
, "Out of memory during linking.\n");
4138 goto write_top_level_array_size_and_stride
;
4141 var_name
= get_top_level_name(temp_name
);
4144 linker_error(shProg
, "Out of memory during linking.\n");
4145 goto write_top_level_array_size_and_stride
;
4149 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4150 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4154 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4155 ir_variable
*var
= node
->as_variable();
4156 if (!var
|| !var
->get_interface_type() ||
4157 var
->data
.mode
!= ir_var_shader_storage
)
4160 const glsl_type
*interface
= var
->get_interface_type();
4162 if (strcmp(interface_name
, interface
->name
) != 0)
4165 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4166 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4167 if (strcmp(field
->name
, var_name
) != 0)
4170 array_stride
= get_array_stride(uni
, interface
, field
,
4171 interface_name
, var_name
);
4172 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4173 goto write_top_level_array_size_and_stride
;
4177 write_top_level_array_size_and_stride
:
4178 free(interface_name
);
4180 uni
->top_level_array_stride
= array_stride
;
4181 uni
->top_level_array_size
= array_size
;
4185 * Builds up a list of program resources that point to existing
4189 build_program_resource_list(struct gl_context
*ctx
,
4190 struct gl_shader_program
*shProg
)
4192 /* Rebuild resource list. */
4193 if (shProg
->data
->ProgramResourceList
) {
4194 ralloc_free(shProg
->data
->ProgramResourceList
);
4195 shProg
->data
->ProgramResourceList
= NULL
;
4196 shProg
->data
->NumProgramResourceList
= 0;
4199 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4201 /* Determine first input and final output stage. These are used to
4202 * detect which variables should be enumerated in the resource list
4203 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4205 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4206 if (!shProg
->_LinkedShaders
[i
])
4208 if (input_stage
== MESA_SHADER_STAGES
)
4213 /* Empty shader, no resources. */
4214 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4217 struct set
*resource_set
= _mesa_set_create(NULL
,
4219 _mesa_key_pointer_equal
);
4221 /* Program interface needs to expose varyings in case of SSO. */
4222 if (shProg
->SeparateShader
) {
4223 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4224 input_stage
, GL_PROGRAM_INPUT
))
4227 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4228 output_stage
, GL_PROGRAM_OUTPUT
))
4232 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4235 /* Add inputs and outputs to the resource list. */
4236 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4237 input_stage
, GL_PROGRAM_INPUT
))
4240 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4241 output_stage
, GL_PROGRAM_OUTPUT
))
4244 if (shProg
->last_vert_prog
) {
4245 struct gl_transform_feedback_info
*linked_xfb
=
4246 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4248 /* Add transform feedback varyings. */
4249 if (linked_xfb
->NumVarying
> 0) {
4250 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4251 if (!add_program_resource(shProg
, resource_set
,
4252 GL_TRANSFORM_FEEDBACK_VARYING
,
4253 &linked_xfb
->Varyings
[i
], 0))
4258 /* Add transform feedback buffers. */
4259 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4260 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4261 linked_xfb
->Buffers
[i
].Binding
= i
;
4262 if (!add_program_resource(shProg
, resource_set
,
4263 GL_TRANSFORM_FEEDBACK_BUFFER
,
4264 &linked_xfb
->Buffers
[i
], 0))
4270 /* Add uniforms from uniform storage. */
4271 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4272 /* Do not add uniforms internally used by Mesa. */
4273 if (shProg
->data
->UniformStorage
[i
].hidden
)
4277 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4280 /* Add stagereferences for uniforms in a uniform block. */
4281 bool is_shader_storage
=
4282 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4283 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4284 if (block_index
!= -1) {
4285 stageref
|= is_shader_storage
?
4286 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4287 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4290 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4291 if (!should_add_buffer_variable(shProg
, type
,
4292 shProg
->data
->UniformStorage
[i
].name
))
4295 if (is_shader_storage
) {
4296 calculate_array_size_and_stride(shProg
,
4297 &shProg
->data
->UniformStorage
[i
]);
4300 if (!add_program_resource(shProg
, resource_set
, type
,
4301 &shProg
->data
->UniformStorage
[i
], stageref
))
4305 /* Add program uniform blocks. */
4306 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4307 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4308 &shProg
->data
->UniformBlocks
[i
], 0))
4312 /* Add program shader storage blocks. */
4313 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4314 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4315 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4319 /* Add atomic counter buffers. */
4320 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4321 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4322 &shProg
->data
->AtomicBuffers
[i
], 0))
4326 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4328 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4331 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4332 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4333 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4336 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4337 /* add shader subroutines */
4338 if (!add_program_resource(shProg
, resource_set
,
4339 type
, &shProg
->data
->UniformStorage
[i
], 0))
4344 unsigned mask
= shProg
->data
->linked_stages
;
4346 const int i
= u_bit_scan(&mask
);
4347 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4349 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4350 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4351 if (!add_program_resource(shProg
, resource_set
,
4352 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4357 _mesa_set_destroy(resource_set
, NULL
);
4361 * This check is done to make sure we allow only constant expression
4362 * indexing and "constant-index-expression" (indexing with an expression
4363 * that includes loop induction variable).
4366 validate_sampler_array_indexing(struct gl_context
*ctx
,
4367 struct gl_shader_program
*prog
)
4369 dynamic_sampler_array_indexing_visitor v
;
4370 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4371 if (prog
->_LinkedShaders
[i
] == NULL
)
4374 bool no_dynamic_indexing
=
4375 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4377 /* Search for array derefs in shader. */
4378 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4379 if (v
.uses_dynamic_sampler_array_indexing()) {
4380 const char *msg
= "sampler arrays indexed with non-constant "
4381 "expressions is forbidden in GLSL %s %u";
4382 /* Backend has indicated that it has no dynamic indexing support. */
4383 if (no_dynamic_indexing
) {
4384 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4385 prog
->data
->Version
);
4388 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4389 prog
->data
->Version
);
4397 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4399 unsigned mask
= prog
->data
->linked_stages
;
4401 const int i
= u_bit_scan(&mask
);
4402 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4404 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4405 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4406 ir_function
*fn
= node
->as_function();
4410 if (fn
->is_subroutine
)
4411 p
->sh
.NumSubroutineUniformTypes
++;
4413 if (!fn
->num_subroutine_types
)
4416 /* these should have been calculated earlier. */
4417 assert(fn
->subroutine_index
!= -1);
4418 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4419 linker_error(prog
, "Too many subroutine functions declared.\n");
4422 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4423 struct gl_subroutine_function
,
4424 p
->sh
.NumSubroutineFunctions
+ 1);
4425 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4426 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4427 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4428 ralloc_array(p
, const struct glsl_type
*,
4429 fn
->num_subroutine_types
);
4431 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4434 * "Each subroutine with an index qualifier in the shader must be
4435 * given a unique index, otherwise a compile or link error will be
4438 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4439 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4440 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4441 linker_error(prog
, "each subroutine index qualifier in the "
4442 "shader must be unique\n");
4446 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4447 fn
->subroutine_index
;
4449 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4450 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4452 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4453 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4454 p
->sh
.NumSubroutineFunctions
++;
4460 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4462 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4464 foreach_in_list(ir_instruction
, node
, ir
) {
4465 ir_variable
*const var
= node
->as_variable();
4467 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4470 /* Don't set always active on builtins that haven't been redeclared */
4471 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4474 var
->data
.always_active_io
= true;
4479 * When separate shader programs are enabled, only input/outputs between
4480 * the stages of a multi-stage separate program can be safely removed
4481 * from the shader interface. Other inputs/outputs must remain active.
4484 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4486 unsigned first
, last
;
4487 assert(prog
->SeparateShader
);
4489 first
= MESA_SHADER_STAGES
;
4492 /* Determine first and last stage. Excluding the compute stage */
4493 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4494 if (!prog
->_LinkedShaders
[i
])
4496 if (first
== MESA_SHADER_STAGES
)
4501 if (first
== MESA_SHADER_STAGES
)
4504 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4505 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4509 if (first
== last
) {
4510 /* For a single shader program only allow inputs to the vertex shader
4511 * and outputs from the fragment shader to be removed.
4513 if (stage
!= MESA_SHADER_VERTEX
)
4514 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4515 if (stage
!= MESA_SHADER_FRAGMENT
)
4516 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4518 /* For multi-stage separate shader programs only allow inputs and
4519 * outputs between the shader stages to be removed as well as inputs
4520 * to the vertex shader and outputs from the fragment shader.
4522 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4523 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4524 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4525 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4531 link_and_validate_uniforms(struct gl_context
*ctx
,
4532 struct gl_shader_program
*prog
)
4534 update_array_sizes(prog
);
4535 link_assign_uniform_locations(prog
, ctx
);
4537 link_assign_atomic_counter_resources(ctx
, prog
);
4538 link_calculate_subroutine_compat(prog
);
4539 check_resources(ctx
, prog
);
4540 check_subroutine_resources(prog
);
4541 check_image_resources(ctx
, prog
);
4542 link_check_atomic_counter_resources(ctx
, prog
);
4546 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4547 struct gl_context
*ctx
,
4548 struct gl_shader_program
*prog
, void *mem_ctx
)
4550 /* Mark all generic shader inputs and outputs as unpaired. */
4551 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4552 if (prog
->_LinkedShaders
[i
] != NULL
) {
4553 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4557 unsigned prev
= first
;
4558 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4559 if (prog
->_LinkedShaders
[i
] == NULL
)
4562 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4563 prog
->_LinkedShaders
[i
]);
4567 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4568 MESA_SHADER_VERTEX
)) {
4572 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4573 MESA_SHADER_FRAGMENT
)) {
4577 prog
->last_vert_prog
= NULL
;
4578 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4579 if (prog
->_LinkedShaders
[i
] == NULL
)
4582 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4586 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4589 link_and_validate_uniforms(ctx
, prog
);
4591 if (!prog
->data
->LinkStatus
)
4594 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4595 if (prog
->_LinkedShaders
[i
] == NULL
)
4598 const struct gl_shader_compiler_options
*options
=
4599 &ctx
->Const
.ShaderCompilerOptions
[i
];
4601 if (options
->LowerBufferInterfaceBlocks
)
4602 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4603 options
->ClampBlockIndicesToArrayBounds
);
4605 if (i
== MESA_SHADER_COMPUTE
)
4606 lower_shared_reference(prog
->_LinkedShaders
[i
],
4607 &prog
->Comp
.SharedSize
);
4609 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4610 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4617 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4619 prog
->data
->LinkStatus
= true; /* All error paths will set this to false */
4620 prog
->data
->Validated
= false;
4622 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4624 * "Linking can fail for a variety of reasons as specified in the
4625 * OpenGL Shading Language Specification, as well as any of the
4626 * following reasons:
4628 * - No shader objects are attached to program."
4630 * The Compatibility Profile specification does not list the error. In
4631 * Compatibility Profile missing shader stages are replaced by
4632 * fixed-function. This applies to the case where all stages are
4635 if (prog
->NumShaders
== 0) {
4636 if (ctx
->API
!= API_OPENGL_COMPAT
)
4637 linker_error(prog
, "no shaders attached to the program\n");
4641 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4643 prog
->ARB_fragment_coord_conventions_enable
= false;
4645 /* Separate the shaders into groups based on their type.
4647 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4648 unsigned num_shaders
[MESA_SHADER_STAGES
];
4650 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4651 shader_list
[i
] = (struct gl_shader
**)
4652 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4656 unsigned min_version
= UINT_MAX
;
4657 unsigned max_version
= 0;
4658 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4659 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4660 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4662 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4663 linker_error(prog
, "all shaders must use same shading "
4664 "language version\n");
4668 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4669 prog
->ARB_fragment_coord_conventions_enable
= true;
4672 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4673 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4674 num_shaders
[shader_type
]++;
4677 /* In desktop GLSL, different shader versions may be linked together. In
4678 * GLSL ES, all shader versions must be the same.
4680 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4681 linker_error(prog
, "all shaders must use same shading "
4682 "language version\n");
4686 prog
->data
->Version
= max_version
;
4687 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4689 /* Some shaders have to be linked with some other shaders present.
4691 if (!prog
->SeparateShader
) {
4692 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4693 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4694 linker_error(prog
, "Geometry shader must be linked with "
4698 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4699 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4700 linker_error(prog
, "Tessellation evaluation shader must be linked "
4701 "with vertex shader\n");
4704 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4705 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4706 linker_error(prog
, "Tessellation control shader must be linked with "
4711 /* The spec is self-contradictory here. It allows linking without a tess
4712 * eval shader, but that can only be used with transform feedback and
4713 * rasterization disabled. However, transform feedback isn't allowed
4714 * with GL_PATCHES, so it can't be used.
4716 * More investigation showed that the idea of transform feedback after
4717 * a tess control shader was dropped, because some hw vendors couldn't
4718 * support tessellation without a tess eval shader, but the linker
4719 * section wasn't updated to reflect that.
4721 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4724 * Do what's reasonable and always require a tess eval shader if a tess
4725 * control shader is present.
4727 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4728 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4729 linker_error(prog
, "Tessellation control shader must be linked with "
4730 "tessellation evaluation shader\n");
4735 /* Compute shaders have additional restrictions. */
4736 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4737 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4738 linker_error(prog
, "Compute shaders may not be linked with any other "
4739 "type of shader\n");
4742 /* Link all shaders for a particular stage and validate the result.
4744 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4745 if (num_shaders
[stage
] > 0) {
4746 gl_linked_shader
*const sh
=
4747 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4748 num_shaders
[stage
], false);
4750 if (!prog
->data
->LinkStatus
) {
4752 _mesa_delete_linked_shader(ctx
, sh
);
4757 case MESA_SHADER_VERTEX
:
4758 validate_vertex_shader_executable(prog
, sh
, ctx
);
4760 case MESA_SHADER_TESS_CTRL
:
4761 /* nothing to be done */
4763 case MESA_SHADER_TESS_EVAL
:
4764 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4766 case MESA_SHADER_GEOMETRY
:
4767 validate_geometry_shader_executable(prog
, sh
, ctx
);
4769 case MESA_SHADER_FRAGMENT
:
4770 validate_fragment_shader_executable(prog
, sh
);
4773 if (!prog
->data
->LinkStatus
) {
4775 _mesa_delete_linked_shader(ctx
, sh
);
4779 prog
->_LinkedShaders
[stage
] = sh
;
4780 prog
->data
->linked_stages
|= 1 << stage
;
4784 /* Here begins the inter-stage linking phase. Some initial validation is
4785 * performed, then locations are assigned for uniforms, attributes, and
4788 cross_validate_uniforms(prog
);
4789 if (!prog
->data
->LinkStatus
)
4792 unsigned first
, last
, prev
;
4794 first
= MESA_SHADER_STAGES
;
4797 /* Determine first and last stage. */
4798 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4799 if (!prog
->_LinkedShaders
[i
])
4801 if (first
== MESA_SHADER_STAGES
)
4806 check_explicit_uniform_locations(ctx
, prog
);
4807 link_assign_subroutine_types(prog
);
4809 if (!prog
->data
->LinkStatus
)
4812 resize_tes_inputs(ctx
, prog
);
4814 /* Validate the inputs of each stage with the output of the preceding
4818 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4819 if (prog
->_LinkedShaders
[i
] == NULL
)
4822 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4823 prog
->_LinkedShaders
[i
]);
4824 if (!prog
->data
->LinkStatus
)
4827 cross_validate_outputs_to_inputs(prog
,
4828 prog
->_LinkedShaders
[prev
],
4829 prog
->_LinkedShaders
[i
]);
4830 if (!prog
->data
->LinkStatus
)
4836 /* Cross-validate uniform blocks between shader stages */
4837 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4838 if (!prog
->data
->LinkStatus
)
4841 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4842 if (prog
->_LinkedShaders
[i
] != NULL
)
4843 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4846 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4847 * it before optimization because we want most of the checks to get
4848 * dropped thanks to constant propagation.
4850 * This rule also applies to GLSL ES 3.00.
4852 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4853 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4855 lower_discard_flow(sh
->ir
);
4859 if (prog
->SeparateShader
)
4860 disable_varying_optimizations_for_sso(prog
);
4863 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4867 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4870 /* Do common optimization before assigning storage for attributes,
4871 * uniforms, and varyings. Later optimization could possibly make
4872 * some of that unused.
4874 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4875 if (prog
->_LinkedShaders
[i
] == NULL
)
4878 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4879 if (!prog
->data
->LinkStatus
)
4882 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4883 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4886 if (ctx
->Const
.LowerTessLevel
) {
4887 lower_tess_level(prog
->_LinkedShaders
[i
]);
4890 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4891 /* Run it just once. */
4892 do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4893 &ctx
->Const
.ShaderCompilerOptions
[i
],
4894 ctx
->Const
.NativeIntegers
);
4896 /* Repeat it until it stops making changes. */
4897 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4898 &ctx
->Const
.ShaderCompilerOptions
[i
],
4899 ctx
->Const
.NativeIntegers
))
4903 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
4904 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4907 /* Validation for special cases where we allow sampler array indexing
4908 * with loop induction variable. This check emits a warning or error
4909 * depending if backend can handle dynamic indexing.
4911 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
4912 (prog
->IsES
&& prog
->data
->Version
< 300)) {
4913 if (!validate_sampler_array_indexing(ctx
, prog
))
4917 /* Check and validate stream emissions in geometry shaders */
4918 validate_geometry_shader_emissions(ctx
, prog
);
4920 store_fragdepth_layout(prog
);
4922 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
4925 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4926 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4927 * anything about shader linking when one of the shaders (vertex or
4928 * fragment shader) is absent. So, the extension shouldn't change the
4929 * behavior specified in GLSL specification.
4931 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4932 * "Linking can fail for a variety of reasons as specified in the
4933 * OpenGL ES Shading Language Specification, as well as any of the
4934 * following reasons:
4938 * * program contains objects to form either a vertex shader or
4939 * fragment shader, and program is not separable, and does not
4940 * contain objects to form both a vertex shader and fragment
4943 * However, the only scenario in 3.1+ where we don't require them both is
4944 * when we have a compute shader. For example:
4946 * - No shaders is a link error.
4947 * - Geom or Tess without a Vertex shader is a link error which means we
4948 * always require a Vertex shader and hence a Fragment shader.
4949 * - Finally a Compute shader linked with any other stage is a link error.
4951 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4952 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4953 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4954 linker_error(prog
, "program lacks a vertex shader\n");
4955 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4956 linker_error(prog
, "program lacks a fragment shader\n");
4961 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4962 free(shader_list
[i
]);
4963 if (prog
->_LinkedShaders
[i
] == NULL
)
4966 /* Do a final validation step to make sure that the IR wasn't
4967 * invalidated by any modifications performed after intrastage linking.
4969 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4971 /* Retain any live IR, but trash the rest. */
4972 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4974 /* The symbol table in the linked shaders may contain references to
4975 * variables that were removed (e.g., unused uniforms). Since it may
4976 * contain junk, there is no possible valid use. Delete it and set the
4979 delete prog
->_LinkedShaders
[i
]->symbols
;
4980 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4983 ralloc_free(mem_ctx
);