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
= vertices_per_prim(shader
->info
.Geom
.InputType
);
752 prog
->Geom
.VerticesIn
= num_vertices
;
754 analyze_clip_cull_usage(prog
, shader
, ctx
,
755 &shader
->Program
->info
.clip_distance_array_size
,
756 &shader
->Program
->info
.cull_distance_array_size
);
760 * Check if geometry shaders emit to non-zero streams and do corresponding
764 validate_geometry_shader_emissions(struct gl_context
*ctx
,
765 struct gl_shader_program
*prog
)
767 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
770 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
771 emit_vertex
.run(sh
->ir
);
772 if (emit_vertex
.error()) {
773 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
774 "stream parameter are in the range [0, %d].\n",
775 emit_vertex
.error_func(),
776 emit_vertex
.error_stream(),
777 ctx
->Const
.MaxVertexStreams
- 1);
779 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
780 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
782 /* From the ARB_gpu_shader5 spec:
784 * "Multiple vertex streams are supported only if the output primitive
785 * type is declared to be "points". A program will fail to link if it
786 * contains a geometry shader calling EmitStreamVertex() or
787 * EndStreamPrimitive() if its output primitive type is not "points".
789 * However, in the same spec:
791 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
792 * with <stream> set to zero."
796 * "The function EndPrimitive() is equivalent to calling
797 * EndStreamPrimitive() with <stream> set to zero."
799 * Since we can call EmitVertex() and EndPrimitive() when we output
800 * primitives other than points, calling EmitStreamVertex(0) or
801 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
802 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
803 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
806 if (prog
->Geom
.UsesStreams
&& sh
->info
.Geom
.OutputType
!= GL_POINTS
) {
807 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
808 "with n>0 requires point output\n");
814 validate_intrastage_arrays(struct gl_shader_program
*prog
,
815 ir_variable
*const var
,
816 ir_variable
*const existing
)
818 /* Consider the types to be "the same" if both types are arrays
819 * of the same type and one of the arrays is implicitly sized.
820 * In addition, set the type of the linked variable to the
821 * explicitly sized array.
823 if (var
->type
->is_array() && existing
->type
->is_array()) {
824 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
825 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
826 if (var
->type
->length
!= 0) {
827 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
828 linker_error(prog
, "%s `%s' declared as type "
829 "`%s' but outermost dimension has an index"
832 var
->name
, var
->type
->name
,
833 existing
->data
.max_array_access
);
835 existing
->type
= var
->type
;
837 } else if (existing
->type
->length
!= 0) {
838 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
839 !existing
->data
.from_ssbo_unsized_array
) {
840 linker_error(prog
, "%s `%s' declared as type "
841 "`%s' but outermost dimension has an index"
844 var
->name
, existing
->type
->name
,
845 var
->data
.max_array_access
);
850 /* The arrays of structs could have different glsl_type pointers but
851 * they are actually the same type. Use record_compare() to check that.
853 if (existing
->type
->fields
.array
->is_record() &&
854 var
->type
->fields
.array
->is_record() &&
855 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
864 * Perform validation of global variables used across multiple shaders
867 cross_validate_globals(struct gl_shader_program
*prog
,
868 struct exec_list
*ir
, glsl_symbol_table
*variables
,
871 foreach_in_list(ir_instruction
, node
, ir
) {
872 ir_variable
*const var
= node
->as_variable();
877 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
880 /* don't cross validate subroutine uniforms */
881 if (var
->type
->contains_subroutine())
884 /* Don't cross validate temporaries that are at global scope. These
885 * will eventually get pulled into the shaders 'main'.
887 if (var
->data
.mode
== ir_var_temporary
)
890 /* If a global with this name has already been seen, verify that the
891 * new instance has the same type. In addition, if the globals have
892 * initializers, the values of the initializers must be the same.
894 ir_variable
*const existing
= variables
->get_variable(var
->name
);
895 if (existing
!= NULL
) {
896 /* Check if types match. Interface blocks have some special
897 * rules so we handle those elsewhere.
899 if (var
->type
!= existing
->type
&&
900 !var
->is_interface_instance()) {
901 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
902 if (var
->type
->is_record() && existing
->type
->is_record()
903 && existing
->type
->record_compare(var
->type
)) {
904 existing
->type
= var
->type
;
906 /* If it is an unsized array in a Shader Storage Block,
907 * two different shaders can access to different elements.
908 * Because of that, they might be converted to different
909 * sized arrays, then check that they are compatible but
910 * ignore the array size.
912 if (!(var
->data
.mode
== ir_var_shader_storage
&&
913 var
->data
.from_ssbo_unsized_array
&&
914 existing
->data
.mode
== ir_var_shader_storage
&&
915 existing
->data
.from_ssbo_unsized_array
&&
916 var
->type
->gl_type
== existing
->type
->gl_type
)) {
917 linker_error(prog
, "%s `%s' declared as type "
918 "`%s' and type `%s'\n",
920 var
->name
, var
->type
->name
,
921 existing
->type
->name
);
928 if (var
->data
.explicit_location
) {
929 if (existing
->data
.explicit_location
930 && (var
->data
.location
!= existing
->data
.location
)) {
931 linker_error(prog
, "explicit locations for %s "
932 "`%s' have differing values\n",
933 mode_string(var
), var
->name
);
937 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
938 linker_error(prog
, "explicit components for %s `%s' have "
939 "differing values\n", mode_string(var
), var
->name
);
943 existing
->data
.location
= var
->data
.location
;
944 existing
->data
.explicit_location
= true;
946 /* Check if uniform with implicit location was marked explicit
947 * by earlier shader stage. If so, mark it explicit in this stage
948 * too to make sure later processing does not treat it as
951 if (existing
->data
.explicit_location
) {
952 var
->data
.location
= existing
->data
.location
;
953 var
->data
.explicit_location
= true;
957 /* From the GLSL 4.20 specification:
958 * "A link error will result if two compilation units in a program
959 * specify different integer-constant bindings for the same
960 * opaque-uniform name. However, it is not an error to specify a
961 * binding on some but not all declarations for the same name"
963 if (var
->data
.explicit_binding
) {
964 if (existing
->data
.explicit_binding
&&
965 var
->data
.binding
!= existing
->data
.binding
) {
966 linker_error(prog
, "explicit bindings for %s "
967 "`%s' have differing values\n",
968 mode_string(var
), var
->name
);
972 existing
->data
.binding
= var
->data
.binding
;
973 existing
->data
.explicit_binding
= true;
976 if (var
->type
->contains_atomic() &&
977 var
->data
.offset
!= existing
->data
.offset
) {
978 linker_error(prog
, "offset specifications for %s "
979 "`%s' have differing values\n",
980 mode_string(var
), var
->name
);
984 /* Validate layout qualifiers for gl_FragDepth.
986 * From the AMD/ARB_conservative_depth specs:
988 * "If gl_FragDepth is redeclared in any fragment shader in a
989 * program, it must be redeclared in all fragment shaders in
990 * that program that have static assignments to
991 * gl_FragDepth. All redeclarations of gl_FragDepth in all
992 * fragment shaders in a single program must have the same set
995 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
996 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
997 bool layout_differs
=
998 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1000 if (layout_declared
&& layout_differs
) {
1002 "All redeclarations of gl_FragDepth in all "
1003 "fragment shaders in a single program must have "
1004 "the same set of qualifiers.\n");
1007 if (var
->data
.used
&& layout_differs
) {
1009 "If gl_FragDepth is redeclared with a layout "
1010 "qualifier in any fragment shader, it must be "
1011 "redeclared with the same layout qualifier in "
1012 "all fragment shaders that have assignments to "
1017 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1019 * "If a shared global has multiple initializers, the
1020 * initializers must all be constant expressions, and they
1021 * must all have the same value. Otherwise, a link error will
1022 * result. (A shared global having only one initializer does
1023 * not require that initializer to be a constant expression.)"
1025 * Previous to 4.20 the GLSL spec simply said that initializers
1026 * must have the same value. In this case of non-constant
1027 * initializers, this was impossible to determine. As a result,
1028 * no vendor actually implemented that behavior. The 4.20
1029 * behavior matches the implemented behavior of at least one other
1030 * vendor, so we'll implement that for all GLSL versions.
1032 if (var
->constant_initializer
!= NULL
) {
1033 if (existing
->constant_initializer
!= NULL
) {
1034 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1035 linker_error(prog
, "initializers for %s "
1036 "`%s' have differing values\n",
1037 mode_string(var
), var
->name
);
1041 /* If the first-seen instance of a particular uniform did
1042 * not have an initializer but a later instance does,
1043 * replace the former with the later.
1045 variables
->replace_variable(existing
->name
, var
);
1049 if (var
->data
.has_initializer
) {
1050 if (existing
->data
.has_initializer
1051 && (var
->constant_initializer
== NULL
1052 || existing
->constant_initializer
== NULL
)) {
1054 "shared global variable `%s' has multiple "
1055 "non-constant initializers.\n",
1061 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1062 linker_error(prog
, "declarations for %s `%s' have "
1063 "mismatching invariant qualifiers\n",
1064 mode_string(var
), var
->name
);
1067 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1068 linker_error(prog
, "declarations for %s `%s' have "
1069 "mismatching centroid qualifiers\n",
1070 mode_string(var
), var
->name
);
1073 if (existing
->data
.sample
!= var
->data
.sample
) {
1074 linker_error(prog
, "declarations for %s `%s` have "
1075 "mismatching sample qualifiers\n",
1076 mode_string(var
), var
->name
);
1079 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1080 linker_error(prog
, "declarations for %s `%s` have "
1081 "mismatching image format qualifiers\n",
1082 mode_string(var
), var
->name
);
1086 /* Only in GLSL ES 3.10, the precision qualifier should not match
1087 * between block members defined in matched block names within a
1090 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1091 * member should match.
1093 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1094 !var
->get_interface_type()) &&
1095 existing
->data
.precision
!= var
->data
.precision
) {
1096 linker_error(prog
, "declarations for %s `%s` have "
1097 "mismatching precision qualifiers\n",
1098 mode_string(var
), var
->name
);
1102 variables
->add_variable(var
);
1108 * Perform validation of uniforms used across multiple shader stages
1111 cross_validate_uniforms(struct gl_shader_program
*prog
)
1113 glsl_symbol_table variables
;
1114 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1115 if (prog
->_LinkedShaders
[i
] == NULL
)
1118 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1124 * Accumulates the array of buffer blocks and checks that all definitions of
1125 * blocks agree on their contents.
1128 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1131 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1132 struct gl_uniform_block
*blks
= NULL
;
1133 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1134 &prog
->data
->NumUniformBlocks
;
1136 unsigned max_num_buffer_blocks
= 0;
1137 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1138 if (prog
->_LinkedShaders
[i
]) {
1139 if (validate_ssbo
) {
1140 max_num_buffer_blocks
+=
1141 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1143 max_num_buffer_blocks
+=
1144 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1149 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1150 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1152 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1153 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1154 InterfaceBlockStageIndex
[i
][j
] = -1;
1159 unsigned sh_num_blocks
;
1160 struct gl_uniform_block
**sh_blks
;
1161 if (validate_ssbo
) {
1162 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1163 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1165 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1166 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1169 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1170 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1174 linker_error(prog
, "buffer block `%s' has mismatching "
1175 "definitions\n", sh_blks
[j
]->Name
);
1177 for (unsigned k
= 0; k
<= i
; k
++) {
1178 delete[] InterfaceBlockStageIndex
[k
];
1183 InterfaceBlockStageIndex
[i
][index
] = j
;
1187 /* Update per stage block pointers to point to the program list.
1188 * FIXME: We should be able to free the per stage blocks here.
1190 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1191 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1192 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1194 if (stage_index
!= -1) {
1195 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1197 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1198 sh
->Program
->sh
.ShaderStorageBlocks
:
1199 sh
->Program
->sh
.UniformBlocks
;
1201 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1202 sh_blks
[stage_index
] = &blks
[j
];
1207 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1208 delete[] InterfaceBlockStageIndex
[i
];
1212 prog
->data
->ShaderStorageBlocks
= blks
;
1214 prog
->data
->UniformBlocks
= blks
;
1221 * Populates a shaders symbol table with all global declarations
1224 populate_symbol_table(gl_linked_shader
*sh
)
1226 sh
->symbols
= new(sh
) glsl_symbol_table
;
1228 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1232 if ((func
= inst
->as_function()) != NULL
) {
1233 sh
->symbols
->add_function(func
);
1234 } else if ((var
= inst
->as_variable()) != NULL
) {
1235 if (var
->data
.mode
!= ir_var_temporary
)
1236 sh
->symbols
->add_variable(var
);
1243 * Remap variables referenced in an instruction tree
1245 * This is used when instruction trees are cloned from one shader and placed in
1246 * another. These trees will contain references to \c ir_variable nodes that
1247 * do not exist in the target shader. This function finds these \c ir_variable
1248 * references and replaces the references with matching variables in the target
1251 * If there is no matching variable in the target shader, a clone of the
1252 * \c ir_variable is made and added to the target shader. The new variable is
1253 * added to \b both the instruction stream and the symbol table.
1255 * \param inst IR tree that is to be processed.
1256 * \param symbols Symbol table containing global scope symbols in the
1258 * \param instructions Instruction stream where new variable declarations
1262 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1265 class remap_visitor
: public ir_hierarchical_visitor
{
1267 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1269 this->target
= target
;
1270 this->symbols
= target
->symbols
;
1271 this->instructions
= target
->ir
;
1272 this->temps
= temps
;
1275 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1277 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1278 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1279 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1281 assert(var
!= NULL
);
1283 return visit_continue
;
1286 ir_variable
*const existing
=
1287 this->symbols
->get_variable(ir
->var
->name
);
1288 if (existing
!= NULL
)
1291 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1293 this->symbols
->add_variable(copy
);
1294 this->instructions
->push_head(copy
);
1298 return visit_continue
;
1302 struct gl_linked_shader
*target
;
1303 glsl_symbol_table
*symbols
;
1304 exec_list
*instructions
;
1308 remap_visitor
v(target
, temps
);
1315 * Move non-declarations from one instruction stream to another
1317 * The intended usage pattern of this function is to pass the pointer to the
1318 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1319 * pointer) for \c last and \c false for \c make_copies on the first
1320 * call. Successive calls pass the return value of the previous call for
1321 * \c last and \c true for \c make_copies.
1323 * \param instructions Source instruction stream
1324 * \param last Instruction after which new instructions should be
1325 * inserted in the target instruction stream
1326 * \param make_copies Flag selecting whether instructions in \c instructions
1327 * should be copied (via \c ir_instruction::clone) into the
1328 * target list or moved.
1331 * The new "last" instruction in the target instruction stream. This pointer
1332 * is suitable for use as the \c last parameter of a later call to this
1336 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1337 bool make_copies
, gl_linked_shader
*target
)
1339 hash_table
*temps
= NULL
;
1342 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1343 _mesa_key_pointer_equal
);
1345 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1346 if (inst
->as_function())
1349 ir_variable
*var
= inst
->as_variable();
1350 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1353 assert(inst
->as_assignment()
1355 || inst
->as_if() /* for initializers with the ?: operator */
1356 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1359 inst
= inst
->clone(target
, NULL
);
1362 _mesa_hash_table_insert(temps
, var
, inst
);
1364 remap_variables(inst
, target
, temps
);
1369 last
->insert_after(inst
);
1374 _mesa_hash_table_destroy(temps
, NULL
);
1381 * This class is only used in link_intrastage_shaders() below but declaring
1382 * it inside that function leads to compiler warnings with some versions of
1385 class array_sizing_visitor
: public deref_type_updater
{
1387 array_sizing_visitor()
1388 : mem_ctx(ralloc_context(NULL
)),
1389 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1390 _mesa_key_pointer_equal
))
1394 ~array_sizing_visitor()
1396 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1397 ralloc_free(this->mem_ctx
);
1400 virtual ir_visitor_status
visit(ir_variable
*var
)
1402 const glsl_type
*type_without_array
;
1403 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1404 fixup_type(&var
->type
, var
->data
.max_array_access
,
1405 var
->data
.from_ssbo_unsized_array
,
1406 &implicit_sized_array
);
1407 var
->data
.implicit_sized_array
= implicit_sized_array
;
1408 type_without_array
= var
->type
->without_array();
1409 if (var
->type
->is_interface()) {
1410 if (interface_contains_unsized_arrays(var
->type
)) {
1411 const glsl_type
*new_type
=
1412 resize_interface_members(var
->type
,
1413 var
->get_max_ifc_array_access(),
1414 var
->is_in_shader_storage_block());
1415 var
->type
= new_type
;
1416 var
->change_interface_type(new_type
);
1418 } else if (type_without_array
->is_interface()) {
1419 if (interface_contains_unsized_arrays(type_without_array
)) {
1420 const glsl_type
*new_type
=
1421 resize_interface_members(type_without_array
,
1422 var
->get_max_ifc_array_access(),
1423 var
->is_in_shader_storage_block());
1424 var
->change_interface_type(new_type
);
1425 var
->type
= update_interface_members_array(var
->type
, new_type
);
1427 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1428 /* Store a pointer to the variable in the unnamed_interfaces
1432 _mesa_hash_table_search(this->unnamed_interfaces
,
1435 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1437 if (interface_vars
== NULL
) {
1438 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1440 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1443 unsigned index
= ifc_type
->field_index(var
->name
);
1444 assert(index
< ifc_type
->length
);
1445 assert(interface_vars
[index
] == NULL
);
1446 interface_vars
[index
] = var
;
1448 return visit_continue
;
1452 * For each unnamed interface block that was discovered while running the
1453 * visitor, adjust the interface type to reflect the newly assigned array
1454 * sizes, and fix up the ir_variable nodes to point to the new interface
1457 void fixup_unnamed_interface_types()
1459 hash_table_call_foreach(this->unnamed_interfaces
,
1460 fixup_unnamed_interface_type
, NULL
);
1465 * If the type pointed to by \c type represents an unsized array, replace
1466 * it with a sized array whose size is determined by max_array_access.
1468 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1469 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1471 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1472 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1473 max_array_access
+ 1);
1474 *implicit_sized
= true;
1475 assert(*type
!= NULL
);
1479 static const glsl_type
*
1480 update_interface_members_array(const glsl_type
*type
,
1481 const glsl_type
*new_interface_type
)
1483 const glsl_type
*element_type
= type
->fields
.array
;
1484 if (element_type
->is_array()) {
1485 const glsl_type
*new_array_type
=
1486 update_interface_members_array(element_type
, new_interface_type
);
1487 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1489 return glsl_type::get_array_instance(new_interface_type
,
1495 * Determine whether the given interface type contains unsized arrays (if
1496 * it doesn't, array_sizing_visitor doesn't need to process it).
1498 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1500 for (unsigned i
= 0; i
< type
->length
; i
++) {
1501 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1502 if (elem_type
->is_unsized_array())
1509 * Create a new interface type based on the given type, with unsized arrays
1510 * replaced by sized arrays whose size is determined by
1511 * max_ifc_array_access.
1513 static const glsl_type
*
1514 resize_interface_members(const glsl_type
*type
,
1515 const int *max_ifc_array_access
,
1518 unsigned num_fields
= type
->length
;
1519 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1520 memcpy(fields
, type
->fields
.structure
,
1521 num_fields
* sizeof(*fields
));
1522 for (unsigned i
= 0; i
< num_fields
; i
++) {
1523 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1524 /* If SSBO last member is unsized array, we don't replace it by a sized
1527 if (is_ssbo
&& i
== (num_fields
- 1))
1528 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1529 true, &implicit_sized_array
);
1531 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1532 false, &implicit_sized_array
);
1533 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1535 glsl_interface_packing packing
=
1536 (glsl_interface_packing
) type
->interface_packing
;
1537 bool row_major
= (bool) type
->interface_row_major
;
1538 const glsl_type
*new_ifc_type
=
1539 glsl_type::get_interface_instance(fields
, num_fields
,
1540 packing
, row_major
, type
->name
);
1542 return new_ifc_type
;
1545 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1548 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1549 ir_variable
**interface_vars
= (ir_variable
**) data
;
1550 unsigned num_fields
= ifc_type
->length
;
1551 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1552 memcpy(fields
, ifc_type
->fields
.structure
,
1553 num_fields
* sizeof(*fields
));
1554 bool interface_type_changed
= false;
1555 for (unsigned i
= 0; i
< num_fields
; i
++) {
1556 if (interface_vars
[i
] != NULL
&&
1557 fields
[i
].type
!= interface_vars
[i
]->type
) {
1558 fields
[i
].type
= interface_vars
[i
]->type
;
1559 interface_type_changed
= true;
1562 if (!interface_type_changed
) {
1566 glsl_interface_packing packing
=
1567 (glsl_interface_packing
) ifc_type
->interface_packing
;
1568 bool row_major
= (bool) ifc_type
->interface_row_major
;
1569 const glsl_type
*new_ifc_type
=
1570 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1571 row_major
, ifc_type
->name
);
1573 for (unsigned i
= 0; i
< num_fields
; i
++) {
1574 if (interface_vars
[i
] != NULL
)
1575 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1580 * Memory context used to allocate the data in \c unnamed_interfaces.
1585 * Hash table from const glsl_type * to an array of ir_variable *'s
1586 * pointing to the ir_variables constituting each unnamed interface block.
1588 hash_table
*unnamed_interfaces
;
1592 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1593 struct gl_shader_program
*prog
)
1595 /* We will validate doubles at a later stage */
1596 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1597 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1598 "multiple of 4 or if its applied to a type that is "
1599 "or contains a double a multiple of 8.",
1600 prog
->TransformFeedback
.BufferStride
[idx
]);
1604 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1605 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1606 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1607 "limit has been exceeded.");
1615 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1619 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1620 struct gl_shader_program
*prog
,
1621 struct gl_linked_shader
*linked_shader
,
1622 struct gl_shader
**shader_list
,
1623 unsigned num_shaders
)
1625 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1626 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1629 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1630 struct gl_shader
*shader
= shader_list
[i
];
1632 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1633 if (shader
->TransformFeedbackBufferStride
[j
]) {
1634 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1635 prog
->TransformFeedback
.BufferStride
[j
] =
1636 shader
->TransformFeedbackBufferStride
[j
];
1637 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1639 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1640 shader
->TransformFeedbackBufferStride
[j
]){
1642 "intrastage shaders defined with conflicting "
1643 "xfb_stride for buffer %d (%d and %d)\n", j
,
1644 prog
->TransformFeedback
.BufferStride
[j
],
1645 shader
->TransformFeedbackBufferStride
[j
]);
1654 * Performs the cross-validation of tessellation control shader vertices and
1655 * layout qualifiers for the attached tessellation control shaders,
1656 * and propagates them to the linked TCS and linked shader program.
1659 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1660 struct gl_program
*gl_prog
,
1661 struct gl_shader
**shader_list
,
1662 unsigned num_shaders
)
1664 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1667 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1669 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1671 * "All tessellation control shader layout declarations in a program
1672 * must specify the same output patch vertex count. There must be at
1673 * least one layout qualifier specifying an output patch vertex count
1674 * in any program containing tessellation control shaders; however,
1675 * such a declaration is not required in all tessellation control
1679 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1680 struct gl_shader
*shader
= shader_list
[i
];
1682 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1683 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1684 gl_prog
->info
.tess
.tcs_vertices_out
!=
1685 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1686 linker_error(prog
, "tessellation control shader defined with "
1687 "conflicting output vertex count (%d and %d)\n",
1688 gl_prog
->info
.tess
.tcs_vertices_out
,
1689 shader
->info
.TessCtrl
.VerticesOut
);
1692 gl_prog
->info
.tess
.tcs_vertices_out
=
1693 shader
->info
.TessCtrl
.VerticesOut
;
1697 /* Just do the intrastage -> interstage propagation right now,
1698 * since we already know we're in the right type of shader program
1701 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1702 linker_error(prog
, "tessellation control shader didn't declare "
1703 "vertices out layout qualifier\n");
1710 * Performs the cross-validation of tessellation evaluation shader
1711 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1712 * for the attached tessellation evaluation shaders, and propagates them
1713 * to the linked TES and linked shader program.
1716 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1717 struct gl_linked_shader
*linked_shader
,
1718 struct gl_shader
**shader_list
,
1719 unsigned num_shaders
)
1721 linked_shader
->info
.TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1722 linked_shader
->info
.TessEval
.Spacing
= TESS_SPACING_UNSPECIFIED
;
1723 linked_shader
->info
.TessEval
.VertexOrder
= 0;
1724 linked_shader
->info
.TessEval
.PointMode
= -1;
1726 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1729 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1731 * "At least one tessellation evaluation shader (compilation unit) in
1732 * a program must declare a primitive mode in its input layout.
1733 * Declaration vertex spacing, ordering, and point mode identifiers is
1734 * optional. It is not required that all tessellation evaluation
1735 * shaders in a program declare a primitive mode. If spacing or
1736 * vertex ordering declarations are omitted, the tessellation
1737 * primitive generator will use equal spacing or counter-clockwise
1738 * vertex ordering, respectively. If a point mode declaration is
1739 * omitted, the tessellation primitive generator will produce lines or
1740 * triangles according to the primitive mode."
1743 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1744 struct gl_shader
*shader
= shader_list
[i
];
1746 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1747 if (linked_shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1748 linked_shader
->info
.TessEval
.PrimitiveMode
!=
1749 shader
->info
.TessEval
.PrimitiveMode
) {
1750 linker_error(prog
, "tessellation evaluation shader defined with "
1751 "conflicting input primitive modes.\n");
1754 linked_shader
->info
.TessEval
.PrimitiveMode
= shader
->info
.TessEval
.PrimitiveMode
;
1757 if (shader
->info
.TessEval
.Spacing
!= 0) {
1758 if (linked_shader
->info
.TessEval
.Spacing
!= 0 &&
1759 linked_shader
->info
.TessEval
.Spacing
!=
1760 shader
->info
.TessEval
.Spacing
) {
1761 linker_error(prog
, "tessellation evaluation shader defined with "
1762 "conflicting vertex spacing.\n");
1765 linked_shader
->info
.TessEval
.Spacing
= shader
->info
.TessEval
.Spacing
;
1768 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1769 if (linked_shader
->info
.TessEval
.VertexOrder
!= 0 &&
1770 linked_shader
->info
.TessEval
.VertexOrder
!=
1771 shader
->info
.TessEval
.VertexOrder
) {
1772 linker_error(prog
, "tessellation evaluation shader defined with "
1773 "conflicting ordering.\n");
1776 linked_shader
->info
.TessEval
.VertexOrder
=
1777 shader
->info
.TessEval
.VertexOrder
;
1780 if (shader
->info
.TessEval
.PointMode
!= -1) {
1781 if (linked_shader
->info
.TessEval
.PointMode
!= -1 &&
1782 linked_shader
->info
.TessEval
.PointMode
!=
1783 shader
->info
.TessEval
.PointMode
) {
1784 linker_error(prog
, "tessellation evaluation shader defined with "
1785 "conflicting point modes.\n");
1788 linked_shader
->info
.TessEval
.PointMode
=
1789 shader
->info
.TessEval
.PointMode
;
1794 /* Just do the intrastage -> interstage propagation right now,
1795 * since we already know we're in the right type of shader program
1798 if (linked_shader
->info
.TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1800 "tessellation evaluation shader didn't declare input "
1801 "primitive modes.\n");
1805 if (linked_shader
->info
.TessEval
.Spacing
== TESS_SPACING_UNSPECIFIED
)
1806 linked_shader
->info
.TessEval
.Spacing
= TESS_SPACING_EQUAL
;
1808 if (linked_shader
->info
.TessEval
.VertexOrder
== 0)
1809 linked_shader
->info
.TessEval
.VertexOrder
= GL_CCW
;
1811 if (linked_shader
->info
.TessEval
.PointMode
== -1)
1812 linked_shader
->info
.TessEval
.PointMode
= GL_FALSE
;
1817 * Performs the cross-validation of layout qualifiers specified in
1818 * redeclaration of gl_FragCoord for the attached fragment shaders,
1819 * and propagates them to the linked FS and linked shader program.
1822 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1823 struct gl_linked_shader
*linked_shader
,
1824 struct gl_shader
**shader_list
,
1825 unsigned num_shaders
)
1827 bool redeclares_gl_fragcoord
= false;
1828 bool uses_gl_fragcoord
= false;
1829 bool origin_upper_left
= false;
1830 bool pixel_center_integer
= false;
1832 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1833 (prog
->data
->Version
< 150 &&
1834 !prog
->ARB_fragment_coord_conventions_enable
))
1837 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1838 struct gl_shader
*shader
= shader_list
[i
];
1839 /* From the GLSL 1.50 spec, page 39:
1841 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1842 * it must be redeclared in all the fragment shaders in that program
1843 * that have a static use gl_FragCoord."
1845 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1846 shader
->uses_gl_fragcoord
)
1847 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1848 uses_gl_fragcoord
)) {
1849 linker_error(prog
, "fragment shader defined with conflicting "
1850 "layout qualifiers for gl_FragCoord\n");
1853 /* From the GLSL 1.50 spec, page 39:
1855 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1856 * single program must have the same set of qualifiers."
1858 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1859 (shader
->origin_upper_left
!= origin_upper_left
||
1860 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1861 linker_error(prog
, "fragment shader defined with conflicting "
1862 "layout qualifiers for gl_FragCoord\n");
1865 /* Update the linked shader state. Note that uses_gl_fragcoord should
1866 * accumulate the results. The other values should replace. If there
1867 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1868 * are already known to be the same.
1870 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1871 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1872 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1873 origin_upper_left
= shader
->origin_upper_left
;
1874 pixel_center_integer
= shader
->pixel_center_integer
;
1877 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1878 shader
->EarlyFragmentTests
;
1879 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1880 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1881 shader
->PostDepthCoverage
;
1883 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1888 * Performs the cross-validation of geometry shader max_vertices and
1889 * primitive type layout qualifiers for the attached geometry shaders,
1890 * and propagates them to the linked GS and linked shader program.
1893 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1894 struct gl_linked_shader
*linked_shader
,
1895 struct gl_shader
**shader_list
,
1896 unsigned num_shaders
)
1898 linked_shader
->info
.Geom
.VerticesOut
= -1;
1899 linked_shader
->info
.Geom
.Invocations
= 0;
1900 linked_shader
->info
.Geom
.InputType
= PRIM_UNKNOWN
;
1901 linked_shader
->info
.Geom
.OutputType
= PRIM_UNKNOWN
;
1903 /* No in/out qualifiers defined for anything but GLSL 1.50+
1904 * geometry shaders so far.
1906 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
||
1907 prog
->data
->Version
< 150)
1910 /* From the GLSL 1.50 spec, page 46:
1912 * "All geometry shader output layout declarations in a program
1913 * must declare the same layout and same value for
1914 * max_vertices. There must be at least one geometry output
1915 * layout declaration somewhere in a program, but not all
1916 * geometry shaders (compilation units) are required to
1920 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1921 struct gl_shader
*shader
= shader_list
[i
];
1923 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1924 if (linked_shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
&&
1925 linked_shader
->info
.Geom
.InputType
!=
1926 shader
->info
.Geom
.InputType
) {
1927 linker_error(prog
, "geometry shader defined with conflicting "
1931 linked_shader
->info
.Geom
.InputType
= shader
->info
.Geom
.InputType
;
1934 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1935 if (linked_shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
&&
1936 linked_shader
->info
.Geom
.OutputType
!=
1937 shader
->info
.Geom
.OutputType
) {
1938 linker_error(prog
, "geometry shader defined with conflicting "
1942 linked_shader
->info
.Geom
.OutputType
= shader
->info
.Geom
.OutputType
;
1945 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1946 if (linked_shader
->info
.Geom
.VerticesOut
!= -1 &&
1947 linked_shader
->info
.Geom
.VerticesOut
!=
1948 shader
->info
.Geom
.VerticesOut
) {
1949 linker_error(prog
, "geometry shader defined with conflicting "
1950 "output vertex count (%d and %d)\n",
1951 linked_shader
->info
.Geom
.VerticesOut
,
1952 shader
->info
.Geom
.VerticesOut
);
1955 linked_shader
->info
.Geom
.VerticesOut
= shader
->info
.Geom
.VerticesOut
;
1958 if (shader
->info
.Geom
.Invocations
!= 0) {
1959 if (linked_shader
->info
.Geom
.Invocations
!= 0 &&
1960 linked_shader
->info
.Geom
.Invocations
!=
1961 shader
->info
.Geom
.Invocations
) {
1962 linker_error(prog
, "geometry shader defined with conflicting "
1963 "invocation count (%d and %d)\n",
1964 linked_shader
->info
.Geom
.Invocations
,
1965 shader
->info
.Geom
.Invocations
);
1968 linked_shader
->info
.Geom
.Invocations
= shader
->info
.Geom
.Invocations
;
1972 /* Just do the intrastage -> interstage propagation right now,
1973 * since we already know we're in the right type of shader program
1976 if (linked_shader
->info
.Geom
.InputType
== PRIM_UNKNOWN
) {
1978 "geometry shader didn't declare primitive input type\n");
1982 if (linked_shader
->info
.Geom
.OutputType
== PRIM_UNKNOWN
) {
1984 "geometry shader didn't declare primitive output type\n");
1988 if (linked_shader
->info
.Geom
.VerticesOut
== -1) {
1990 "geometry shader didn't declare max_vertices\n");
1994 if (linked_shader
->info
.Geom
.Invocations
== 0)
1995 linked_shader
->info
.Geom
.Invocations
= 1;
2000 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2001 * qualifiers for the attached compute shaders, and propagate them to the
2002 * linked CS and linked shader program.
2005 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2006 struct gl_linked_shader
*linked_shader
,
2007 struct gl_shader
**shader_list
,
2008 unsigned num_shaders
)
2010 for (int i
= 0; i
< 3; i
++)
2011 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
2013 linked_shader
->info
.Comp
.LocalSizeVariable
= false;
2015 /* This function is called for all shader stages, but it only has an effect
2016 * for compute shaders.
2018 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
2021 /* From the ARB_compute_shader spec, in the section describing local size
2024 * If multiple compute shaders attached to a single program object
2025 * declare local work-group size, the declarations must be identical;
2026 * otherwise a link-time error results. Furthermore, if a program
2027 * object contains any compute shaders, at least one must contain an
2028 * input layout qualifier specifying the local work sizes of the
2029 * program, or a link-time error will occur.
2031 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2032 struct gl_shader
*shader
= shader_list
[sh
];
2034 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2035 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2036 for (int i
= 0; i
< 3; i
++) {
2037 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2038 shader
->info
.Comp
.LocalSize
[i
]) {
2039 linker_error(prog
, "compute shader defined with conflicting "
2045 for (int i
= 0; i
< 3; i
++) {
2046 linked_shader
->info
.Comp
.LocalSize
[i
] =
2047 shader
->info
.Comp
.LocalSize
[i
];
2049 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2050 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2051 /* The ARB_compute_variable_group_size spec says:
2053 * If one compute shader attached to a program declares a
2054 * variable local group size and a second compute shader
2055 * attached to the same program declares a fixed local group
2056 * size, a link-time error results.
2058 linker_error(prog
, "compute shader defined with both fixed and "
2059 "variable local group size\n");
2062 linked_shader
->info
.Comp
.LocalSizeVariable
= true;
2066 /* Just do the intrastage -> interstage propagation right now,
2067 * since we already know we're in the right type of shader program
2070 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0 &&
2071 !linked_shader
->info
.Comp
.LocalSizeVariable
) {
2072 linker_error(prog
, "compute shader must contain a fixed or a variable "
2073 "local group size\n");
2076 for (int i
= 0; i
< 3; i
++)
2077 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2079 prog
->Comp
.LocalSizeVariable
=
2080 linked_shader
->info
.Comp
.LocalSizeVariable
;
2085 * Combine a group of shaders for a single stage to generate a linked shader
2088 * If this function is supplied a single shader, it is cloned, and the new
2089 * shader is returned.
2091 struct gl_linked_shader
*
2092 link_intrastage_shaders(void *mem_ctx
,
2093 struct gl_context
*ctx
,
2094 struct gl_shader_program
*prog
,
2095 struct gl_shader
**shader_list
,
2096 unsigned num_shaders
,
2097 bool allow_missing_main
)
2099 struct gl_uniform_block
*ubo_blocks
= NULL
;
2100 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2101 unsigned num_ubo_blocks
= 0;
2102 unsigned num_ssbo_blocks
= 0;
2104 /* Check that global variables defined in multiple shaders are consistent.
2106 glsl_symbol_table variables
;
2107 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2108 if (shader_list
[i
] == NULL
)
2110 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2113 if (!prog
->data
->LinkStatus
)
2116 /* Check that interface blocks defined in multiple shaders are consistent.
2118 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2120 if (!prog
->data
->LinkStatus
)
2123 /* Check that there is only a single definition of each function signature
2124 * across all shaders.
2126 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2127 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2128 ir_function
*const f
= node
->as_function();
2133 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2134 ir_function
*const other
=
2135 shader_list
[j
]->symbols
->get_function(f
->name
);
2137 /* If the other shader has no function (and therefore no function
2138 * signatures) with the same name, skip to the next shader.
2143 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2144 if (!sig
->is_defined
)
2147 ir_function_signature
*other_sig
=
2148 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2150 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2151 linker_error(prog
, "function `%s' is multiply defined\n",
2160 /* Find the shader that defines main, and make a clone of it.
2162 * Starting with the clone, search for undefined references. If one is
2163 * found, find the shader that defines it. Clone the reference and add
2164 * it to the shader. Repeat until there are no undefined references or
2165 * until a reference cannot be resolved.
2167 gl_shader
*main
= NULL
;
2168 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2169 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2170 main
= shader_list
[i
];
2175 if (main
== NULL
&& allow_missing_main
)
2176 main
= shader_list
[0];
2179 linker_error(prog
, "%s shader lacks `main'\n",
2180 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2184 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2185 linked
->Stage
= shader_list
[0]->Stage
;
2187 /* Create program and attach it to the linked shader */
2188 struct gl_program
*gl_prog
=
2189 ctx
->Driver
.NewProgram(ctx
,
2190 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2193 prog
->data
->LinkStatus
= false;
2194 _mesa_delete_linked_shader(ctx
, linked
);
2198 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2200 /* Don't use _mesa_reference_program() just take ownership */
2201 linked
->Program
= gl_prog
;
2203 linked
->ir
= new(linked
) exec_list
;
2204 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2206 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2207 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2208 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2209 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2210 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2211 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2214 populate_symbol_table(linked
);
2216 /* The pointer to the main function in the final linked shader (i.e., the
2217 * copy of the original shader that contained the main function).
2219 ir_function_signature
*const main_sig
=
2220 _mesa_get_main_function_signature(linked
->symbols
);
2222 /* Move any instructions other than variable declarations or function
2223 * declarations into main.
2225 if (main_sig
!= NULL
) {
2226 exec_node
*insertion_point
=
2227 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2230 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2231 if (shader_list
[i
] == main
)
2234 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2235 insertion_point
, true, linked
);
2239 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2240 _mesa_delete_linked_shader(ctx
, linked
);
2244 /* Make a pass over all variable declarations to ensure that arrays with
2245 * unspecified sizes have a size specified. The size is inferred from the
2246 * max_array_access field.
2248 array_sizing_visitor v
;
2250 v
.fixup_unnamed_interface_types();
2252 /* Link up uniform blocks defined within this stage. */
2253 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2254 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2256 if (!prog
->data
->LinkStatus
) {
2257 _mesa_delete_linked_shader(ctx
, linked
);
2261 /* Copy ubo blocks to linked shader list */
2262 linked
->Program
->sh
.UniformBlocks
=
2263 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2264 ralloc_steal(linked
, ubo_blocks
);
2265 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2266 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2268 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2270 /* Copy ssbo blocks to linked shader list */
2271 linked
->Program
->sh
.ShaderStorageBlocks
=
2272 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2273 ralloc_steal(linked
, ssbo_blocks
);
2274 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2275 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2277 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2279 /* At this point linked should contain all of the linked IR, so
2280 * validate it to make sure nothing went wrong.
2282 validate_ir_tree(linked
->ir
);
2284 /* Set the size of geometry shader input arrays */
2285 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2286 unsigned num_vertices
= vertices_per_prim(linked
->info
.Geom
.InputType
);
2287 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2288 MESA_SHADER_GEOMETRY
);
2289 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2290 ir
->accept(&input_resize_visitor
);
2294 if (ctx
->Const
.VertexID_is_zero_based
)
2295 lower_vertex_id(linked
);
2298 /* Compute the source checksum. */
2299 linked
->SourceChecksum
= 0;
2300 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2301 if (shader_list
[i
] == NULL
)
2303 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2311 * Update the sizes of linked shader uniform arrays to the maximum
2314 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2316 * If one or more elements of an array are active,
2317 * GetActiveUniform will return the name of the array in name,
2318 * subject to the restrictions listed above. The type of the array
2319 * is returned in type. The size parameter contains the highest
2320 * array element index used, plus one. The compiler or linker
2321 * determines the highest index used. There will be only one
2322 * active uniform reported by the GL per uniform array.
2326 update_array_sizes(struct gl_shader_program
*prog
)
2328 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2329 if (prog
->_LinkedShaders
[i
] == NULL
)
2332 bool types_were_updated
= false;
2334 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2335 ir_variable
*const var
= node
->as_variable();
2337 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2338 !var
->type
->is_array())
2341 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2342 * will not be eliminated. Since we always do std140, just
2343 * don't resize arrays in UBOs.
2345 * Atomic counters are supposed to get deterministic
2346 * locations assigned based on the declaration ordering and
2347 * sizes, array compaction would mess that up.
2349 * Subroutine uniforms are not removed.
2351 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2352 var
->type
->contains_subroutine() || var
->constant_initializer
)
2355 int size
= var
->data
.max_array_access
;
2356 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2357 if (prog
->_LinkedShaders
[j
] == NULL
)
2360 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2361 ir_variable
*other_var
= node2
->as_variable();
2365 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2366 other_var
->data
.max_array_access
> size
) {
2367 size
= other_var
->data
.max_array_access
;
2372 if (size
+ 1 != (int)var
->type
->length
) {
2373 /* If this is a built-in uniform (i.e., it's backed by some
2374 * fixed-function state), adjust the number of state slots to
2375 * match the new array size. The number of slots per array entry
2376 * is not known. It seems safe to assume that the total number of
2377 * slots is an integer multiple of the number of array elements.
2378 * Determine the number of slots per array element by dividing by
2379 * the old (total) size.
2381 const unsigned num_slots
= var
->get_num_state_slots();
2382 if (num_slots
> 0) {
2383 var
->set_num_state_slots((size
+ 1)
2384 * (num_slots
/ var
->type
->length
));
2387 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2389 types_were_updated
= true;
2393 /* Update the types of dereferences in case we changed any. */
2394 if (types_were_updated
) {
2395 deref_type_updater v
;
2396 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2402 * Resize tessellation evaluation per-vertex inputs to the size of
2403 * tessellation control per-vertex outputs.
2406 resize_tes_inputs(struct gl_context
*ctx
,
2407 struct gl_shader_program
*prog
)
2409 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2412 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2413 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2415 /* If no control shader is present, then the TES inputs are statically
2416 * sized to MaxPatchVertices; the actual size of the arrays won't be
2417 * known until draw time.
2419 const int num_vertices
= tcs
2420 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2421 : ctx
->Const
.MaxPatchVertices
;
2423 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2424 MESA_SHADER_TESS_EVAL
);
2425 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2426 ir
->accept(&input_resize_visitor
);
2429 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2430 /* Convert the gl_PatchVerticesIn system value into a constant, since
2431 * the value is known at this point.
2433 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2434 ir_variable
*var
= ir
->as_variable();
2435 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2436 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2437 void *mem_ctx
= ralloc_parent(var
);
2438 var
->data
.location
= 0;
2439 var
->data
.explicit_location
= false;
2441 var
->data
.mode
= ir_var_auto
;
2442 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2444 var
->data
.mode
= ir_var_uniform
;
2445 var
->data
.how_declared
= ir_var_hidden
;
2446 var
->allocate_state_slots(1);
2447 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2448 slot0
->swizzle
= SWIZZLE_XXXX
;
2449 slot0
->tokens
[0] = STATE_INTERNAL
;
2450 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2451 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2452 slot0
->tokens
[i
] = 0;
2460 * Find a contiguous set of available bits in a bitmask.
2462 * \param used_mask Bits representing used (1) and unused (0) locations
2463 * \param needed_count Number of contiguous bits needed.
2466 * Base location of the available bits on success or -1 on failure.
2469 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2471 unsigned needed_mask
= (1 << needed_count
) - 1;
2472 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2474 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2475 * cannot optimize possibly infinite loops" for the loop below.
2477 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2480 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2481 if ((needed_mask
& ~used_mask
) == needed_mask
)
2492 * Assign locations for either VS inputs or FS outputs
2494 * \param mem_ctx Temporary ralloc context used for linking
2495 * \param prog Shader program whose variables need locations assigned
2496 * \param constants Driver specific constant values for the program.
2497 * \param target_index Selector for the program target to receive location
2498 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2499 * \c MESA_SHADER_FRAGMENT.
2502 * If locations are successfully assigned, true is returned. Otherwise an
2503 * error is emitted to the shader link log and false is returned.
2506 assign_attribute_or_color_locations(void *mem_ctx
,
2507 gl_shader_program
*prog
,
2508 struct gl_constants
*constants
,
2509 unsigned target_index
)
2511 /* Maximum number of generic locations. This corresponds to either the
2512 * maximum number of draw buffers or the maximum number of generic
2515 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2516 constants
->Program
[target_index
].MaxAttribs
:
2517 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2519 /* Mark invalid locations as being used.
2521 unsigned used_locations
= (max_index
>= 32)
2522 ? ~0 : ~((1 << max_index
) - 1);
2523 unsigned double_storage_locations
= 0;
2525 assert((target_index
== MESA_SHADER_VERTEX
)
2526 || (target_index
== MESA_SHADER_FRAGMENT
));
2528 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2532 /* Operate in a total of four passes.
2534 * 1. Invalidate the location assignments for all vertex shader inputs.
2536 * 2. Assign locations for inputs that have user-defined (via
2537 * glBindVertexAttribLocation) locations and outputs that have
2538 * user-defined locations (via glBindFragDataLocation).
2540 * 3. Sort the attributes without assigned locations by number of slots
2541 * required in decreasing order. Fragmentation caused by attribute
2542 * locations assigned by the application may prevent large attributes
2543 * from having enough contiguous space.
2545 * 4. Assign locations to any inputs without assigned locations.
2548 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2549 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2551 const enum ir_variable_mode direction
=
2552 (target_index
== MESA_SHADER_VERTEX
)
2553 ? ir_var_shader_in
: ir_var_shader_out
;
2556 /* Temporary storage for the set of attributes that need locations assigned.
2562 /* Used below in the call to qsort. */
2563 static int compare(const void *a
, const void *b
)
2565 const temp_attr
*const l
= (const temp_attr
*) a
;
2566 const temp_attr
*const r
= (const temp_attr
*) b
;
2568 /* Reversed because we want a descending order sort below. */
2569 return r
->slots
- l
->slots
;
2572 assert(max_index
<= 32);
2574 /* Temporary array for the set of attributes that have locations assigned.
2576 ir_variable
*assigned
[16];
2578 unsigned num_attr
= 0;
2579 unsigned assigned_attr
= 0;
2581 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2582 ir_variable
*const var
= node
->as_variable();
2584 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2587 if (var
->data
.explicit_location
) {
2588 var
->data
.is_unmatched_generic_inout
= 0;
2589 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2590 || (var
->data
.location
< 0)) {
2592 "invalid explicit location %d specified for `%s'\n",
2593 (var
->data
.location
< 0)
2594 ? var
->data
.location
2595 : var
->data
.location
- generic_base
,
2599 } else if (target_index
== MESA_SHADER_VERTEX
) {
2602 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2603 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2604 var
->data
.location
= binding
;
2605 var
->data
.is_unmatched_generic_inout
= 0;
2607 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2610 const char *name
= var
->name
;
2611 const glsl_type
*type
= var
->type
;
2614 /* Check if there's a binding for the variable name */
2615 if (prog
->FragDataBindings
->get(binding
, name
)) {
2616 assert(binding
>= FRAG_RESULT_DATA0
);
2617 var
->data
.location
= binding
;
2618 var
->data
.is_unmatched_generic_inout
= 0;
2620 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2621 var
->data
.index
= index
;
2626 /* If not, but it's an array type, look for name[0] */
2627 if (type
->is_array()) {
2628 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2629 type
= type
->fields
.array
;
2637 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2640 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2642 * "Output binding assignments will cause LinkProgram to fail:
2644 * If the program has an active output assigned to a location greater
2645 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2646 * an active output assigned an index greater than or equal to one;"
2648 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2649 var
->data
.location
- generic_base
>=
2650 (int) constants
->MaxDualSourceDrawBuffers
) {
2652 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2653 "with index %u for %s\n",
2654 var
->data
.location
- generic_base
, var
->data
.index
,
2659 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2661 /* If the variable is not a built-in and has a location statically
2662 * assigned in the shader (presumably via a layout qualifier), make sure
2663 * that it doesn't collide with other assigned locations. Otherwise,
2664 * add it to the list of variables that need linker-assigned locations.
2666 if (var
->data
.location
!= -1) {
2667 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2668 /* From page 61 of the OpenGL 4.0 spec:
2670 * "LinkProgram will fail if the attribute bindings assigned
2671 * by BindAttribLocation do not leave not enough space to
2672 * assign a location for an active matrix attribute or an
2673 * active attribute array, both of which require multiple
2674 * contiguous generic attributes."
2676 * I think above text prohibits the aliasing of explicit and
2677 * automatic assignments. But, aliasing is allowed in manual
2678 * assignments of attribute locations. See below comments for
2681 * From OpenGL 4.0 spec, page 61:
2683 * "It is possible for an application to bind more than one
2684 * attribute name to the same location. This is referred to as
2685 * aliasing. This will only work if only one of the aliased
2686 * attributes is active in the executable program, or if no
2687 * path through the shader consumes more than one attribute of
2688 * a set of attributes aliased to the same location. A link
2689 * error can occur if the linker determines that every path
2690 * through the shader consumes multiple aliased attributes,
2691 * but implementations are not required to generate an error
2694 * From GLSL 4.30 spec, page 54:
2696 * "A program will fail to link if any two non-vertex shader
2697 * input variables are assigned to the same location. For
2698 * vertex shaders, multiple input variables may be assigned
2699 * to the same location using either layout qualifiers or via
2700 * the OpenGL API. However, such aliasing is intended only to
2701 * support vertex shaders where each execution path accesses
2702 * at most one input per each location. Implementations are
2703 * permitted, but not required, to generate link-time errors
2704 * if they detect that every path through the vertex shader
2705 * executable accesses multiple inputs assigned to any single
2706 * location. For all shader types, a program will fail to link
2707 * if explicit location assignments leave the linker unable
2708 * to find space for other variables without explicit
2711 * From OpenGL ES 3.0 spec, page 56:
2713 * "Binding more than one attribute name to the same location
2714 * is referred to as aliasing, and is not permitted in OpenGL
2715 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2716 * fail when this condition exists. However, aliasing is
2717 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2718 * This will only work if only one of the aliased attributes
2719 * is active in the executable program, or if no path through
2720 * the shader consumes more than one attribute of a set of
2721 * attributes aliased to the same location. A link error can
2722 * occur if the linker determines that every path through the
2723 * shader consumes multiple aliased attributes, but implemen-
2724 * tations are not required to generate an error in this case."
2726 * After looking at above references from OpenGL, OpenGL ES and
2727 * GLSL specifications, we allow aliasing of vertex input variables
2728 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2730 * NOTE: This is not required by the spec but its worth mentioning
2731 * here that we're not doing anything to make sure that no path
2732 * through the vertex shader executable accesses multiple inputs
2733 * assigned to any single location.
2736 /* Mask representing the contiguous slots that will be used by
2739 const unsigned attr
= var
->data
.location
- generic_base
;
2740 const unsigned use_mask
= (1 << slots
) - 1;
2741 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2742 ? "vertex shader input" : "fragment shader output";
2744 /* Generate a link error if the requested locations for this
2745 * attribute exceed the maximum allowed attribute location.
2747 if (attr
+ slots
> max_index
) {
2749 "insufficient contiguous locations "
2750 "available for %s `%s' %d %d %d\n", string
,
2751 var
->name
, used_locations
, use_mask
, attr
);
2755 /* Generate a link error if the set of bits requested for this
2756 * attribute overlaps any previously allocated bits.
2758 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2759 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2760 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2763 * "Additionally, for fragment shader outputs, if two
2764 * variables are placed within the same location, they
2765 * must have the same underlying type (floating-point or
2766 * integer). No component aliasing of output variables or
2767 * members is allowed.
2769 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2770 unsigned assigned_slots
=
2771 assigned
[i
]->type
->count_attribute_slots(false);
2772 unsigned assig_attr
=
2773 assigned
[i
]->data
.location
- generic_base
;
2774 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2776 if ((assigned_use_mask
<< assig_attr
) &
2777 (use_mask
<< attr
)) {
2779 const glsl_type
*assigned_type
=
2780 assigned
[i
]->type
->without_array();
2781 const glsl_type
*type
= var
->type
->without_array();
2782 if (assigned_type
->base_type
!= type
->base_type
) {
2783 linker_error(prog
, "types do not match for aliased"
2784 " %ss %s and %s\n", string
,
2785 assigned
[i
]->name
, var
->name
);
2789 unsigned assigned_component_mask
=
2790 ((1 << assigned_type
->vector_elements
) - 1) <<
2791 assigned
[i
]->data
.location_frac
;
2792 unsigned component_mask
=
2793 ((1 << type
->vector_elements
) - 1) <<
2794 var
->data
.location_frac
;
2795 if (assigned_component_mask
& component_mask
) {
2796 linker_error(prog
, "overlapping component is "
2797 "assigned to %ss %s and %s "
2799 string
, assigned
[i
]->name
, var
->name
,
2800 var
->data
.location_frac
);
2805 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2806 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2807 linker_error(prog
, "overlapping location is assigned "
2808 "to %s `%s' %d %d %d\n", string
, var
->name
,
2809 used_locations
, use_mask
, attr
);
2812 linker_warning(prog
, "overlapping location is assigned "
2813 "to %s `%s' %d %d %d\n", string
, var
->name
,
2814 used_locations
, use_mask
, attr
);
2818 used_locations
|= (use_mask
<< attr
);
2820 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2822 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2823 * active attribute variables may fail to link, unless
2824 * device-dependent optimizations are able to make the program
2825 * fit within available hardware resources. For the purposes
2826 * of this test, attribute variables of the type dvec3, dvec4,
2827 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2828 * count as consuming twice as many attributes as equivalent
2829 * single-precision types. While these types use the same number
2830 * of generic attributes as their single-precision equivalents,
2831 * implementations are permitted to consume two single-precision
2832 * vectors of internal storage for each three- or four-component
2833 * double-precision vector."
2835 * Mark this attribute slot as taking up twice as much space
2836 * so we can count it properly against limits. According to
2837 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2838 * is optional behavior, but it seems preferable.
2840 if (var
->type
->without_array()->is_dual_slot())
2841 double_storage_locations
|= (use_mask
<< attr
);
2844 assigned
[assigned_attr
] = var
;
2850 if (num_attr
>= max_index
) {
2851 linker_error(prog
, "too many %s (max %u)",
2852 target_index
== MESA_SHADER_VERTEX
?
2853 "vertex shader inputs" : "fragment shader outputs",
2857 to_assign
[num_attr
].slots
= slots
;
2858 to_assign
[num_attr
].var
= var
;
2862 if (target_index
== MESA_SHADER_VERTEX
) {
2863 unsigned total_attribs_size
=
2864 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2865 _mesa_bitcount(double_storage_locations
);
2866 if (total_attribs_size
> max_index
) {
2868 "attempt to use %d vertex attribute slots only %d available ",
2869 total_attribs_size
, max_index
);
2874 /* If all of the attributes were assigned locations by the application (or
2875 * are built-in attributes with fixed locations), return early. This should
2876 * be the common case.
2881 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2883 if (target_index
== MESA_SHADER_VERTEX
) {
2884 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2885 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2886 * reserved to prevent it from being automatically allocated below.
2888 find_deref_visitor
find("gl_Vertex");
2890 if (find
.variable_found())
2891 used_locations
|= (1 << 0);
2894 for (unsigned i
= 0; i
< num_attr
; i
++) {
2895 /* Mask representing the contiguous slots that will be used by this
2898 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2900 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2903 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2904 ? "vertex shader input" : "fragment shader output";
2907 "insufficient contiguous locations "
2908 "available for %s `%s'\n",
2909 string
, to_assign
[i
].var
->name
);
2913 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2914 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2915 used_locations
|= (use_mask
<< location
);
2917 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2918 double_storage_locations
|= (use_mask
<< location
);
2921 /* Now that we have all the locations, from the GL 4.5 core spec, section
2922 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2923 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2924 * as equivalent single-precision types.
2926 if (target_index
== MESA_SHADER_VERTEX
) {
2927 unsigned total_attribs_size
=
2928 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2929 _mesa_bitcount(double_storage_locations
);
2930 if (total_attribs_size
> max_index
) {
2932 "attempt to use %d vertex attribute slots only %d available ",
2933 total_attribs_size
, max_index
);
2942 * Match explicit locations of outputs to inputs and deactivate the
2943 * unmatch flag if found so we don't optimise them away.
2946 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2947 gl_linked_shader
*consumer
)
2949 glsl_symbol_table parameters
;
2950 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2953 /* Find all shader outputs in the "producer" stage.
2955 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2956 ir_variable
*const var
= node
->as_variable();
2958 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2961 if (var
->data
.explicit_location
&&
2962 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2963 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2964 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2965 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2969 /* Match inputs to outputs */
2970 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2971 ir_variable
*const input
= node
->as_variable();
2973 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2976 ir_variable
*output
= NULL
;
2977 if (input
->data
.explicit_location
2978 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2979 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2980 [input
->data
.location_frac
];
2982 if (output
!= NULL
){
2983 input
->data
.is_unmatched_generic_inout
= 0;
2984 output
->data
.is_unmatched_generic_inout
= 0;
2991 * Store the gl_FragDepth layout in the gl_shader_program struct.
2994 store_fragdepth_layout(struct gl_shader_program
*prog
)
2996 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3000 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3002 /* We don't look up the gl_FragDepth symbol directly because if
3003 * gl_FragDepth is not used in the shader, it's removed from the IR.
3004 * However, the symbol won't be removed from the symbol table.
3006 * We're only interested in the cases where the variable is NOT removed
3009 foreach_in_list(ir_instruction
, node
, ir
) {
3010 ir_variable
*const var
= node
->as_variable();
3012 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3016 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3017 switch (var
->data
.depth_layout
) {
3018 case ir_depth_layout_none
:
3019 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3021 case ir_depth_layout_any
:
3022 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3024 case ir_depth_layout_greater
:
3025 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3027 case ir_depth_layout_less
:
3028 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3030 case ir_depth_layout_unchanged
:
3031 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3042 * Validate the resources used by a program versus the implementation limits
3045 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3047 unsigned total_uniform_blocks
= 0;
3048 unsigned total_shader_storage_blocks
= 0;
3050 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3051 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3056 if (sh
->Program
->info
.num_textures
>
3057 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3058 linker_error(prog
, "Too many %s shader texture samplers\n",
3059 _mesa_shader_stage_to_string(i
));
3062 if (sh
->num_uniform_components
>
3063 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3064 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3065 linker_warning(prog
, "Too many %s shader default uniform block "
3066 "components, but the driver will try to optimize "
3067 "them out; this is non-portable out-of-spec "
3069 _mesa_shader_stage_to_string(i
));
3071 linker_error(prog
, "Too many %s shader default uniform block "
3073 _mesa_shader_stage_to_string(i
));
3077 if (sh
->num_combined_uniform_components
>
3078 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3079 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3080 linker_warning(prog
, "Too many %s shader uniform components, "
3081 "but the driver will try to optimize them out; "
3082 "this is non-portable out-of-spec behavior\n",
3083 _mesa_shader_stage_to_string(i
));
3085 linker_error(prog
, "Too many %s shader uniform components\n",
3086 _mesa_shader_stage_to_string(i
));
3090 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3091 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3093 const unsigned max_uniform_blocks
=
3094 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3095 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3096 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3097 _mesa_shader_stage_to_string(i
),
3098 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3101 const unsigned max_shader_storage_blocks
=
3102 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3103 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3104 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3105 _mesa_shader_stage_to_string(i
),
3106 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3110 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3111 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3112 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3115 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3116 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3117 total_shader_storage_blocks
,
3118 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3121 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3122 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3123 ctx
->Const
.MaxUniformBlockSize
) {
3124 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3125 prog
->data
->UniformBlocks
[i
].Name
,
3126 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3127 ctx
->Const
.MaxUniformBlockSize
);
3131 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3132 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3133 ctx
->Const
.MaxShaderStorageBlockSize
) {
3134 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3135 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3136 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3137 ctx
->Const
.MaxShaderStorageBlockSize
);
3143 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3145 unsigned mask
= prog
->data
->linked_stages
;
3147 const int i
= u_bit_scan(&mask
);
3148 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3150 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3151 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3154 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3160 if (p
->sh
.NumSubroutineFunctions
== 0) {
3161 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3164 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3165 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3166 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3167 if (fn
->types
[k
] == uni
->type
) {
3173 uni
->num_compatible_subroutines
= count
;
3179 check_subroutine_resources(struct gl_shader_program
*prog
)
3181 unsigned mask
= prog
->data
->linked_stages
;
3183 const int i
= u_bit_scan(&mask
);
3184 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3186 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3187 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3188 _mesa_shader_stage_to_string(i
));
3193 * Validate shader image resources.
3196 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3198 unsigned total_image_units
= 0;
3199 unsigned fragment_outputs
= 0;
3200 unsigned total_shader_storage_blocks
= 0;
3202 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3205 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3206 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3209 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3210 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3211 _mesa_shader_stage_to_string(i
),
3212 sh
->Program
->info
.num_images
,
3213 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3215 total_image_units
+= sh
->Program
->info
.num_images
;
3216 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3218 if (i
== MESA_SHADER_FRAGMENT
) {
3219 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3220 ir_variable
*var
= node
->as_variable();
3221 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3222 /* since there are no double fs outputs - pass false */
3223 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3229 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3230 linker_error(prog
, "Too many combined image uniforms\n");
3232 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3233 ctx
->Const
.MaxCombinedShaderOutputResources
)
3234 linker_error(prog
, "Too many combined image uniforms, shader storage "
3235 " buffers and fragment outputs\n");
3240 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3241 * for a variable, checks for overlaps between other uniforms using explicit
3245 reserve_explicit_locations(struct gl_shader_program
*prog
,
3246 string_to_uint_map
*map
, ir_variable
*var
)
3248 unsigned slots
= var
->type
->uniform_locations();
3249 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3250 unsigned return_value
= slots
;
3252 /* Resize remap table if locations do not fit in the current one. */
3253 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3254 prog
->UniformRemapTable
=
3255 reralloc(prog
, prog
->UniformRemapTable
,
3256 gl_uniform_storage
*,
3259 if (!prog
->UniformRemapTable
) {
3260 linker_error(prog
, "Out of memory during linking.\n");
3264 /* Initialize allocated space. */
3265 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3266 prog
->UniformRemapTable
[i
] = NULL
;
3268 prog
->NumUniformRemapTable
= max_loc
+ 1;
3271 for (unsigned i
= 0; i
< slots
; i
++) {
3272 unsigned loc
= var
->data
.location
+ i
;
3274 /* Check if location is already used. */
3275 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3277 /* Possibly same uniform from a different stage, this is ok. */
3279 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3284 /* ARB_explicit_uniform_location specification states:
3286 * "No two default-block uniform variables in the program can have
3287 * the same location, even if they are unused, otherwise a compiler
3288 * or linker error will be generated."
3291 "location qualifier for uniform %s overlaps "
3292 "previously used location\n",
3297 /* Initialize location as inactive before optimization
3298 * rounds and location assignment.
3300 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3303 /* Note, base location used for arrays. */
3304 map
->put(var
->data
.location
, var
->name
);
3306 return return_value
;
3310 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3311 struct gl_program
*p
,
3314 unsigned slots
= var
->type
->uniform_locations();
3315 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3317 /* Resize remap table if locations do not fit in the current one. */
3318 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3319 p
->sh
.SubroutineUniformRemapTable
=
3320 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3321 gl_uniform_storage
*,
3324 if (!p
->sh
.SubroutineUniformRemapTable
) {
3325 linker_error(prog
, "Out of memory during linking.\n");
3329 /* Initialize allocated space. */
3330 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3331 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3333 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3336 for (unsigned i
= 0; i
< slots
; i
++) {
3337 unsigned loc
= var
->data
.location
+ i
;
3339 /* Check if location is already used. */
3340 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3342 /* ARB_explicit_uniform_location specification states:
3343 * "No two subroutine uniform variables can have the same location
3344 * in the same shader stage, otherwise a compiler or linker error
3345 * will be generated."
3348 "location qualifier for uniform %s overlaps "
3349 "previously used location\n",
3354 /* Initialize location as inactive before optimization
3355 * rounds and location assignment.
3357 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3363 * Check and reserve all explicit uniform locations, called before
3364 * any optimizations happen to handle also inactive uniforms and
3365 * inactive array elements that may get trimmed away.
3368 check_explicit_uniform_locations(struct gl_context
*ctx
,
3369 struct gl_shader_program
*prog
)
3371 prog
->NumExplicitUniformLocations
= 0;
3373 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3376 /* This map is used to detect if overlapping explicit locations
3377 * occur with the same uniform (from different stage) or a different one.
3379 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3382 linker_error(prog
, "Out of memory during linking.\n");
3386 unsigned entries_total
= 0;
3387 unsigned mask
= prog
->data
->linked_stages
;
3389 const int i
= u_bit_scan(&mask
);
3390 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3392 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3393 ir_variable
*var
= node
->as_variable();
3394 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3397 if (var
->data
.explicit_location
) {
3399 if (var
->type
->without_array()->is_subroutine())
3400 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3402 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3406 entries_total
+= slots
;
3417 struct empty_uniform_block
*current_block
= NULL
;
3419 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3420 /* We found empty space in UniformRemapTable. */
3421 if (prog
->UniformRemapTable
[i
] == NULL
) {
3422 /* We've found the beginning of a new continous block of empty slots */
3423 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3424 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3425 current_block
->start
= i
;
3426 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3427 ¤t_block
->link
);
3430 /* The current block continues, so we simply increment its slots */
3431 current_block
->slots
++;
3436 prog
->NumExplicitUniformLocations
= entries_total
;
3440 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3441 GLenum type
, const char *name
)
3443 bool found_interface
= false;
3444 unsigned block_name_len
= 0;
3445 const char *block_name_dot
= strchr(name
, '.');
3447 /* These rules only apply to buffer variables. So we return
3448 * true for the rest of types.
3450 if (type
!= GL_BUFFER_VARIABLE
)
3453 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3454 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3455 block_name_len
= strlen(block_name
);
3457 const char *block_square_bracket
= strchr(block_name
, '[');
3458 if (block_square_bracket
) {
3459 /* The block is part of an array of named interfaces,
3460 * for the name comparison we ignore the "[x]" part.
3462 block_name_len
-= strlen(block_square_bracket
);
3465 if (block_name_dot
) {
3466 /* Check if the variable name starts with the interface
3467 * name. The interface name (if present) should have the
3468 * length than the interface block name we are comparing to.
3470 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3471 if (len
!= block_name_len
)
3475 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3476 found_interface
= true;
3481 /* We remove the interface name from the buffer variable name,
3482 * including the dot that follows it.
3484 if (found_interface
)
3485 name
= name
+ block_name_len
+ 1;
3487 /* The ARB_program_interface_query spec says:
3489 * "For an active shader storage block member declared as an array, an
3490 * entry will be generated only for the first array element, regardless
3491 * of its type. For arrays of aggregate types, the enumeration rules
3492 * are applied recursively for the single enumerated array element."
3494 const char *struct_first_dot
= strchr(name
, '.');
3495 const char *first_square_bracket
= strchr(name
, '[');
3497 /* The buffer variable is on top level and it is not an array */
3498 if (!first_square_bracket
) {
3500 /* The shader storage block member is a struct, then generate the entry */
3501 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3504 /* Shader storage block member is an array, only generate an entry for the
3505 * first array element.
3507 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3515 add_program_resource(struct gl_shader_program
*prog
,
3516 struct set
*resource_set
,
3517 GLenum type
, const void *data
, uint8_t stages
)
3521 /* If resource already exists, do not add it again. */
3522 if (_mesa_set_search(resource_set
, data
))
3525 prog
->data
->ProgramResourceList
=
3527 prog
->data
->ProgramResourceList
,
3528 gl_program_resource
,
3529 prog
->data
->NumProgramResourceList
+ 1);
3531 if (!prog
->data
->ProgramResourceList
) {
3532 linker_error(prog
, "Out of memory during linking.\n");
3536 struct gl_program_resource
*res
=
3537 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3541 res
->StageReferences
= stages
;
3543 prog
->data
->NumProgramResourceList
++;
3545 _mesa_set_add(resource_set
, data
);
3550 /* Function checks if a variable var is a packed varying and
3551 * if given name is part of packed varying's list.
3553 * If a variable is a packed varying, it has a name like
3554 * 'packed:a,b,c' where a, b and c are separate variables.
3557 included_in_packed_varying(ir_variable
*var
, const char *name
)
3559 if (strncmp(var
->name
, "packed:", 7) != 0)
3562 char *list
= strdup(var
->name
+ 7);
3567 char *token
= strtok_r(list
, ",", &saveptr
);
3569 if (strcmp(token
, name
) == 0) {
3573 token
= strtok_r(NULL
, ",", &saveptr
);
3580 * Function builds a stage reference bitmask from variable name.
3583 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3588 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3589 * used for reference mask in gl_program_resource will need to be changed.
3591 assert(MESA_SHADER_STAGES
< 8);
3593 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3594 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3598 /* Shader symbol table may contain variables that have
3599 * been optimized away. Search IR for the variable instead.
3601 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3602 ir_variable
*var
= node
->as_variable();
3604 unsigned baselen
= strlen(var
->name
);
3606 if (included_in_packed_varying(var
, name
)) {
3611 /* Type needs to match if specified, otherwise we might
3612 * pick a variable with same name but different interface.
3614 if (var
->data
.mode
!= mode
)
3617 if (strncmp(var
->name
, name
, baselen
) == 0) {
3618 /* Check for exact name matches but also check for arrays and
3621 if (name
[baselen
] == '\0' ||
3622 name
[baselen
] == '[' ||
3623 name
[baselen
] == '.') {
3635 * Create gl_shader_variable from ir_variable class.
3637 static gl_shader_variable
*
3638 create_shader_variable(struct gl_shader_program
*shProg
,
3639 const ir_variable
*in
,
3640 const char *name
, const glsl_type
*type
,
3641 const glsl_type
*interface_type
,
3642 bool use_implicit_location
, int location
,
3643 const glsl_type
*outermost_struct_type
)
3645 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3649 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3650 * expect to see gl_VertexID in the program resource list. Pretend.
3652 if (in
->data
.mode
== ir_var_system_value
&&
3653 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3654 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3655 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3656 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3657 (in
->data
.mode
== ir_var_system_value
&&
3658 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3659 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3660 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3661 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3662 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3663 (in
->data
.mode
== ir_var_system_value
&&
3664 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3665 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3666 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3668 out
->name
= ralloc_strdup(shProg
, name
);
3674 /* The ARB_program_interface_query spec says:
3676 * "Not all active variables are assigned valid locations; the
3677 * following variables will have an effective location of -1:
3679 * * uniforms declared as atomic counters;
3681 * * members of a uniform block;
3683 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3685 * * inputs or outputs not declared with a "location" layout
3686 * qualifier, except for vertex shader inputs and fragment shader
3689 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3690 is_gl_identifier(in
->name
) ||
3691 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3694 out
->location
= location
;
3698 out
->outermost_struct_type
= outermost_struct_type
;
3699 out
->interface_type
= interface_type
;
3700 out
->component
= in
->data
.location_frac
;
3701 out
->index
= in
->data
.index
;
3702 out
->patch
= in
->data
.patch
;
3703 out
->mode
= in
->data
.mode
;
3704 out
->interpolation
= in
->data
.interpolation
;
3705 out
->explicit_location
= in
->data
.explicit_location
;
3706 out
->precision
= in
->data
.precision
;
3711 static const glsl_type
*
3712 resize_to_max_patch_vertices(const struct gl_context
*ctx
,
3713 const glsl_type
*type
)
3718 return glsl_type::get_array_instance(type
->fields
.array
,
3719 ctx
->Const
.MaxPatchVertices
);
3723 add_shader_variable(const struct gl_context
*ctx
,
3724 struct gl_shader_program
*shProg
,
3725 struct set
*resource_set
,
3726 unsigned stage_mask
,
3727 GLenum programInterface
, ir_variable
*var
,
3728 const char *name
, const glsl_type
*type
,
3729 bool use_implicit_location
, int location
,
3730 const glsl_type
*outermost_struct_type
= NULL
)
3732 const glsl_type
*interface_type
= var
->get_interface_type();
3734 if (outermost_struct_type
== NULL
) {
3735 /* Unsized (non-patch) TCS output/TES input arrays are implicitly
3736 * sized to gl_MaxPatchVertices. Internally, we shrink them to a
3739 * This can cause trouble with SSO programs. Since the TCS declares
3740 * the number of output vertices, we can always shrink TCS output
3741 * arrays. However, the TES might not be linked with a TCS, in
3742 * which case it won't know the size of the patch. In other words,
3743 * the TCS and TES may disagree on the (smaller) array sizes. This
3744 * can result in the resource names differing across stages, causing
3745 * SSO validation failures and other cascading issues.
3747 * Expanding the array size to the full gl_MaxPatchVertices fixes
3748 * these issues. It's also what program interface queries expect,
3749 * as that is the official size of the array.
3751 if (var
->data
.tess_varying_implicit_sized_array
) {
3752 type
= resize_to_max_patch_vertices(ctx
, type
);
3753 interface_type
= resize_to_max_patch_vertices(ctx
, interface_type
);
3756 if (var
->data
.from_named_ifc_block
) {
3757 const char *interface_name
= interface_type
->name
;
3759 if (interface_type
->is_array()) {
3760 /* Issue #16 of the ARB_program_interface_query spec says:
3762 * "* If a variable is a member of an interface block without an
3763 * instance name, it is enumerated using just the variable name.
3765 * * If a variable is a member of an interface block with an
3766 * instance name, it is enumerated as "BlockName.Member", where
3767 * "BlockName" is the name of the interface block (not the
3768 * instance name) and "Member" is the name of the variable."
3770 * In particular, it indicates that it should be "BlockName",
3771 * not "BlockName[array length]". The conformance suite and
3772 * dEQP both require this behavior.
3774 * Here, we unwrap the extra array level added by named interface
3775 * block array lowering so we have the correct variable type. We
3776 * also unwrap the interface type when constructing the name.
3778 * We leave interface_type the same so that ES 3.x SSO pipeline
3779 * validation can enforce the rules requiring array length to
3780 * match on interface blocks.
3782 type
= type
->fields
.array
;
3784 interface_name
= interface_type
->fields
.array
->name
;
3787 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3791 switch (type
->base_type
) {
3792 case GLSL_TYPE_STRUCT
: {
3793 /* The ARB_program_interface_query spec says:
3795 * "For an active variable declared as a structure, a separate entry
3796 * will be generated for each active structure member. The name of
3797 * each entry is formed by concatenating the name of the structure,
3798 * the "." character, and the name of the structure member. If a
3799 * structure member to enumerate is itself a structure or array,
3800 * these enumeration rules are applied recursively."
3802 if (outermost_struct_type
== NULL
)
3803 outermost_struct_type
= type
;
3805 unsigned field_location
= location
;
3806 for (unsigned i
= 0; i
< type
->length
; i
++) {
3807 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3808 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3809 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3810 stage_mask
, programInterface
,
3811 var
, field_name
, field
->type
,
3812 use_implicit_location
, field_location
,
3813 outermost_struct_type
))
3816 field_location
+= field
->type
->count_attribute_slots(false);
3822 /* The ARB_program_interface_query spec says:
3824 * "For an active variable declared as a single instance of a basic
3825 * type, a single entry will be generated, using the variable name
3826 * from the shader source."
3828 gl_shader_variable
*sha_v
=
3829 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3830 use_implicit_location
, location
,
3831 outermost_struct_type
);
3835 return add_program_resource(shProg
, resource_set
,
3836 programInterface
, sha_v
, stage_mask
);
3842 add_interface_variables(const struct gl_context
*ctx
,
3843 struct gl_shader_program
*shProg
,
3844 struct set
*resource_set
,
3845 unsigned stage
, GLenum programInterface
)
3847 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3849 foreach_in_list(ir_instruction
, node
, ir
) {
3850 ir_variable
*var
= node
->as_variable();
3852 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3857 switch (var
->data
.mode
) {
3858 case ir_var_system_value
:
3859 case ir_var_shader_in
:
3860 if (programInterface
!= GL_PROGRAM_INPUT
)
3862 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3863 : int(VARYING_SLOT_VAR0
);
3865 case ir_var_shader_out
:
3866 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3868 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3869 : int(VARYING_SLOT_VAR0
);
3875 if (var
->data
.patch
)
3876 loc_bias
= int(VARYING_SLOT_PATCH0
);
3878 /* Skip packed varyings, packed varyings are handled separately
3879 * by add_packed_varyings.
3881 if (strncmp(var
->name
, "packed:", 7) == 0)
3884 /* Skip fragdata arrays, these are handled separately
3885 * by add_fragdata_arrays.
3887 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3890 const bool vs_input_or_fs_output
=
3891 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3892 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3894 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3895 1 << stage
, programInterface
,
3896 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3897 var
->data
.location
- loc_bias
))
3904 add_packed_varyings(const struct gl_context
*ctx
,
3905 struct gl_shader_program
*shProg
,
3906 struct set
*resource_set
,
3907 int stage
, GLenum type
)
3909 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3912 if (!sh
|| !sh
->packed_varyings
)
3915 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3916 ir_variable
*var
= node
->as_variable();
3918 switch (var
->data
.mode
) {
3919 case ir_var_shader_in
:
3920 iface
= GL_PROGRAM_INPUT
;
3922 case ir_var_shader_out
:
3923 iface
= GL_PROGRAM_OUTPUT
;
3926 unreachable("unexpected type");
3929 if (type
== iface
) {
3930 const int stage_mask
=
3931 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3932 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3934 iface
, var
, var
->name
, var
->type
, false,
3935 var
->data
.location
- VARYING_SLOT_VAR0
))
3944 add_fragdata_arrays(const struct gl_context
*ctx
,
3945 struct gl_shader_program
*shProg
,
3946 struct set
*resource_set
)
3948 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3950 if (!sh
|| !sh
->fragdata_arrays
)
3953 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3954 ir_variable
*var
= node
->as_variable();
3956 assert(var
->data
.mode
== ir_var_shader_out
);
3958 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3959 1 << MESA_SHADER_FRAGMENT
,
3960 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3961 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3969 get_top_level_name(const char *name
)
3971 const char *first_dot
= strchr(name
, '.');
3972 const char *first_square_bracket
= strchr(name
, '[');
3975 /* The ARB_program_interface_query spec says:
3977 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3978 * the number of active array elements of the top-level shader storage
3979 * block member containing to the active variable is written to
3980 * <params>. If the top-level block member is not declared as an
3981 * array, the value one is written to <params>. If the top-level block
3982 * member is an array with no declared size, the value zero is written
3986 /* The buffer variable is on top level.*/
3987 if (!first_square_bracket
&& !first_dot
)
3988 name_size
= strlen(name
);
3989 else if ((!first_square_bracket
||
3990 (first_dot
&& first_dot
< first_square_bracket
)))
3991 name_size
= first_dot
- name
;
3993 name_size
= first_square_bracket
- name
;
3995 return strndup(name
, name_size
);
3999 get_var_name(const char *name
)
4001 const char *first_dot
= strchr(name
, '.');
4004 return strdup(name
);
4006 return strndup(first_dot
+1, strlen(first_dot
) - 1);
4010 is_top_level_shader_storage_block_member(const char* name
,
4011 const char* interface_name
,
4012 const char* field_name
)
4014 bool result
= false;
4016 /* If the given variable is already a top-level shader storage
4017 * block member, then return array_size = 1.
4018 * We could have two possibilities: if we have an instanced
4019 * shader storage block or not instanced.
4021 * For the first, we check create a name as it was in top level and
4022 * compare it with the real name. If they are the same, then
4023 * the variable is already at top-level.
4025 * Full instanced name is: interface name + '.' + var name +
4028 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4029 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4030 if (!full_instanced_name
) {
4031 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4035 snprintf(full_instanced_name
, name_length
, "%s.%s",
4036 interface_name
, field_name
);
4038 /* Check if its top-level shader storage block member of an
4039 * instanced interface block, or of a unnamed interface block.
4041 if (strcmp(name
, full_instanced_name
) == 0 ||
4042 strcmp(name
, field_name
) == 0)
4045 free(full_instanced_name
);
4050 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4051 char *interface_name
, char *var_name
)
4053 /* The ARB_program_interface_query spec says:
4055 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4056 * the number of active array elements of the top-level shader storage
4057 * block member containing to the active variable is written to
4058 * <params>. If the top-level block member is not declared as an
4059 * array, the value one is written to <params>. If the top-level block
4060 * member is an array with no declared size, the value zero is written
4063 if (is_top_level_shader_storage_block_member(uni
->name
,
4067 else if (field
->type
->is_unsized_array())
4069 else if (field
->type
->is_array())
4070 return field
->type
->length
;
4076 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4077 const glsl_struct_field
*field
, char *interface_name
,
4080 /* The ARB_program_interface_query spec says:
4082 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4083 * identifying the stride between array elements of the top-level
4084 * shader storage block member containing the active variable is
4085 * written to <params>. For top-level block members declared as
4086 * arrays, the value written is the difference, in basic machine units,
4087 * between the offsets of the active variable for consecutive elements
4088 * in the top-level array. For top-level block members not declared as
4089 * an array, zero is written to <params>."
4091 if (field
->type
->is_array()) {
4092 const enum glsl_matrix_layout matrix_layout
=
4093 glsl_matrix_layout(field
->matrix_layout
);
4094 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4095 const glsl_type
*array_type
= field
->type
->fields
.array
;
4097 if (is_top_level_shader_storage_block_member(uni
->name
,
4102 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4103 if (array_type
->is_record() || array_type
->is_array())
4104 return glsl_align(array_type
->std140_size(row_major
), 16);
4106 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4108 return array_type
->std430_array_stride(row_major
);
4115 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4116 struct gl_uniform_storage
*uni
)
4118 int block_index
= uni
->block_index
;
4119 int array_size
= -1;
4120 int array_stride
= -1;
4121 char *var_name
= get_top_level_name(uni
->name
);
4122 char *interface_name
=
4123 get_top_level_name(uni
->is_shader_storage
?
4124 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4125 shProg
->data
->UniformBlocks
[block_index
].Name
);
4127 if (strcmp(var_name
, interface_name
) == 0) {
4128 /* Deal with instanced array of SSBOs */
4129 char *temp_name
= get_var_name(uni
->name
);
4131 linker_error(shProg
, "Out of memory during linking.\n");
4132 goto write_top_level_array_size_and_stride
;
4135 var_name
= get_top_level_name(temp_name
);
4138 linker_error(shProg
, "Out of memory during linking.\n");
4139 goto write_top_level_array_size_and_stride
;
4143 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4144 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4148 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4149 ir_variable
*var
= node
->as_variable();
4150 if (!var
|| !var
->get_interface_type() ||
4151 var
->data
.mode
!= ir_var_shader_storage
)
4154 const glsl_type
*interface
= var
->get_interface_type();
4156 if (strcmp(interface_name
, interface
->name
) != 0)
4159 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4160 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4161 if (strcmp(field
->name
, var_name
) != 0)
4164 array_stride
= get_array_stride(uni
, interface
, field
,
4165 interface_name
, var_name
);
4166 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4167 goto write_top_level_array_size_and_stride
;
4171 write_top_level_array_size_and_stride
:
4172 free(interface_name
);
4174 uni
->top_level_array_stride
= array_stride
;
4175 uni
->top_level_array_size
= array_size
;
4179 * Builds up a list of program resources that point to existing
4183 build_program_resource_list(struct gl_context
*ctx
,
4184 struct gl_shader_program
*shProg
)
4186 /* Rebuild resource list. */
4187 if (shProg
->data
->ProgramResourceList
) {
4188 ralloc_free(shProg
->data
->ProgramResourceList
);
4189 shProg
->data
->ProgramResourceList
= NULL
;
4190 shProg
->data
->NumProgramResourceList
= 0;
4193 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4195 /* Determine first input and final output stage. These are used to
4196 * detect which variables should be enumerated in the resource list
4197 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4199 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4200 if (!shProg
->_LinkedShaders
[i
])
4202 if (input_stage
== MESA_SHADER_STAGES
)
4207 /* Empty shader, no resources. */
4208 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4211 struct set
*resource_set
= _mesa_set_create(NULL
,
4213 _mesa_key_pointer_equal
);
4215 /* Program interface needs to expose varyings in case of SSO. */
4216 if (shProg
->SeparateShader
) {
4217 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4218 input_stage
, GL_PROGRAM_INPUT
))
4221 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4222 output_stage
, GL_PROGRAM_OUTPUT
))
4226 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4229 /* Add inputs and outputs to the resource list. */
4230 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4231 input_stage
, GL_PROGRAM_INPUT
))
4234 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4235 output_stage
, GL_PROGRAM_OUTPUT
))
4238 if (shProg
->last_vert_prog
) {
4239 struct gl_transform_feedback_info
*linked_xfb
=
4240 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4242 /* Add transform feedback varyings. */
4243 if (linked_xfb
->NumVarying
> 0) {
4244 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4245 if (!add_program_resource(shProg
, resource_set
,
4246 GL_TRANSFORM_FEEDBACK_VARYING
,
4247 &linked_xfb
->Varyings
[i
], 0))
4252 /* Add transform feedback buffers. */
4253 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4254 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4255 linked_xfb
->Buffers
[i
].Binding
= i
;
4256 if (!add_program_resource(shProg
, resource_set
,
4257 GL_TRANSFORM_FEEDBACK_BUFFER
,
4258 &linked_xfb
->Buffers
[i
], 0))
4264 /* Add uniforms from uniform storage. */
4265 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4266 /* Do not add uniforms internally used by Mesa. */
4267 if (shProg
->data
->UniformStorage
[i
].hidden
)
4271 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4274 /* Add stagereferences for uniforms in a uniform block. */
4275 bool is_shader_storage
=
4276 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4277 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4278 if (block_index
!= -1) {
4279 stageref
|= is_shader_storage
?
4280 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4281 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4284 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4285 if (!should_add_buffer_variable(shProg
, type
,
4286 shProg
->data
->UniformStorage
[i
].name
))
4289 if (is_shader_storage
) {
4290 calculate_array_size_and_stride(shProg
,
4291 &shProg
->data
->UniformStorage
[i
]);
4294 if (!add_program_resource(shProg
, resource_set
, type
,
4295 &shProg
->data
->UniformStorage
[i
], stageref
))
4299 /* Add program uniform blocks. */
4300 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4301 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4302 &shProg
->data
->UniformBlocks
[i
], 0))
4306 /* Add program shader storage blocks. */
4307 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4308 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4309 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4313 /* Add atomic counter buffers. */
4314 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4315 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4316 &shProg
->data
->AtomicBuffers
[i
], 0))
4320 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4322 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4325 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4326 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4327 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4330 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4331 /* add shader subroutines */
4332 if (!add_program_resource(shProg
, resource_set
,
4333 type
, &shProg
->data
->UniformStorage
[i
], 0))
4338 unsigned mask
= shProg
->data
->linked_stages
;
4340 const int i
= u_bit_scan(&mask
);
4341 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4343 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4344 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4345 if (!add_program_resource(shProg
, resource_set
,
4346 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4351 _mesa_set_destroy(resource_set
, NULL
);
4355 * This check is done to make sure we allow only constant expression
4356 * indexing and "constant-index-expression" (indexing with an expression
4357 * that includes loop induction variable).
4360 validate_sampler_array_indexing(struct gl_context
*ctx
,
4361 struct gl_shader_program
*prog
)
4363 dynamic_sampler_array_indexing_visitor v
;
4364 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4365 if (prog
->_LinkedShaders
[i
] == NULL
)
4368 bool no_dynamic_indexing
=
4369 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4371 /* Search for array derefs in shader. */
4372 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4373 if (v
.uses_dynamic_sampler_array_indexing()) {
4374 const char *msg
= "sampler arrays indexed with non-constant "
4375 "expressions is forbidden in GLSL %s %u";
4376 /* Backend has indicated that it has no dynamic indexing support. */
4377 if (no_dynamic_indexing
) {
4378 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4379 prog
->data
->Version
);
4382 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4383 prog
->data
->Version
);
4391 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4393 unsigned mask
= prog
->data
->linked_stages
;
4395 const int i
= u_bit_scan(&mask
);
4396 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4398 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4399 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4400 ir_function
*fn
= node
->as_function();
4404 if (fn
->is_subroutine
)
4405 p
->sh
.NumSubroutineUniformTypes
++;
4407 if (!fn
->num_subroutine_types
)
4410 /* these should have been calculated earlier. */
4411 assert(fn
->subroutine_index
!= -1);
4412 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4413 linker_error(prog
, "Too many subroutine functions declared.\n");
4416 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4417 struct gl_subroutine_function
,
4418 p
->sh
.NumSubroutineFunctions
+ 1);
4419 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4420 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4421 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4422 ralloc_array(p
, const struct glsl_type
*,
4423 fn
->num_subroutine_types
);
4425 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4428 * "Each subroutine with an index qualifier in the shader must be
4429 * given a unique index, otherwise a compile or link error will be
4432 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4433 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4434 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4435 linker_error(prog
, "each subroutine index qualifier in the "
4436 "shader must be unique\n");
4440 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4441 fn
->subroutine_index
;
4443 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4444 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4446 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4447 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4448 p
->sh
.NumSubroutineFunctions
++;
4454 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4456 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4458 foreach_in_list(ir_instruction
, node
, ir
) {
4459 ir_variable
*const var
= node
->as_variable();
4461 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4464 /* Don't set always active on builtins that haven't been redeclared */
4465 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4468 var
->data
.always_active_io
= true;
4473 * When separate shader programs are enabled, only input/outputs between
4474 * the stages of a multi-stage separate program can be safely removed
4475 * from the shader interface. Other inputs/outputs must remain active.
4478 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4480 unsigned first
, last
;
4481 assert(prog
->SeparateShader
);
4483 first
= MESA_SHADER_STAGES
;
4486 /* Determine first and last stage. Excluding the compute stage */
4487 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4488 if (!prog
->_LinkedShaders
[i
])
4490 if (first
== MESA_SHADER_STAGES
)
4495 if (first
== MESA_SHADER_STAGES
)
4498 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4499 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4503 if (first
== last
) {
4504 /* For a single shader program only allow inputs to the vertex shader
4505 * and outputs from the fragment shader to be removed.
4507 if (stage
!= MESA_SHADER_VERTEX
)
4508 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4509 if (stage
!= MESA_SHADER_FRAGMENT
)
4510 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4512 /* For multi-stage separate shader programs only allow inputs and
4513 * outputs between the shader stages to be removed as well as inputs
4514 * to the vertex shader and outputs from the fragment shader.
4516 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4517 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4518 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4519 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4525 link_and_validate_uniforms(struct gl_context
*ctx
,
4526 struct gl_shader_program
*prog
)
4528 update_array_sizes(prog
);
4529 link_assign_uniform_locations(prog
, ctx
);
4531 link_assign_atomic_counter_resources(ctx
, prog
);
4532 link_calculate_subroutine_compat(prog
);
4533 check_resources(ctx
, prog
);
4534 check_subroutine_resources(prog
);
4535 check_image_resources(ctx
, prog
);
4536 link_check_atomic_counter_resources(ctx
, prog
);
4540 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4541 struct gl_context
*ctx
,
4542 struct gl_shader_program
*prog
, void *mem_ctx
)
4544 /* Mark all generic shader inputs and outputs as unpaired. */
4545 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4546 if (prog
->_LinkedShaders
[i
] != NULL
) {
4547 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4551 unsigned prev
= first
;
4552 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4553 if (prog
->_LinkedShaders
[i
] == NULL
)
4556 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4557 prog
->_LinkedShaders
[i
]);
4561 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4562 MESA_SHADER_VERTEX
)) {
4566 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4567 MESA_SHADER_FRAGMENT
)) {
4571 prog
->last_vert_prog
= NULL
;
4572 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4573 if (prog
->_LinkedShaders
[i
] == NULL
)
4576 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4580 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4583 link_and_validate_uniforms(ctx
, prog
);
4585 if (!prog
->data
->LinkStatus
)
4588 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4589 if (prog
->_LinkedShaders
[i
] == NULL
)
4592 const struct gl_shader_compiler_options
*options
=
4593 &ctx
->Const
.ShaderCompilerOptions
[i
];
4595 if (options
->LowerBufferInterfaceBlocks
)
4596 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4597 options
->ClampBlockIndicesToArrayBounds
);
4599 if (i
== MESA_SHADER_COMPUTE
)
4600 lower_shared_reference(prog
->_LinkedShaders
[i
],
4601 &prog
->Comp
.SharedSize
);
4603 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4604 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4611 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4613 prog
->data
->LinkStatus
= true; /* All error paths will set this to false */
4614 prog
->data
->Validated
= false;
4616 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4618 * "Linking can fail for a variety of reasons as specified in the
4619 * OpenGL Shading Language Specification, as well as any of the
4620 * following reasons:
4622 * - No shader objects are attached to program."
4624 * The Compatibility Profile specification does not list the error. In
4625 * Compatibility Profile missing shader stages are replaced by
4626 * fixed-function. This applies to the case where all stages are
4629 if (prog
->NumShaders
== 0) {
4630 if (ctx
->API
!= API_OPENGL_COMPAT
)
4631 linker_error(prog
, "no shaders attached to the program\n");
4635 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4637 prog
->ARB_fragment_coord_conventions_enable
= false;
4639 /* Separate the shaders into groups based on their type.
4641 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4642 unsigned num_shaders
[MESA_SHADER_STAGES
];
4644 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4645 shader_list
[i
] = (struct gl_shader
**)
4646 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4650 unsigned min_version
= UINT_MAX
;
4651 unsigned max_version
= 0;
4652 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4653 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4654 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4656 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4657 linker_error(prog
, "all shaders must use same shading "
4658 "language version\n");
4662 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4663 prog
->ARB_fragment_coord_conventions_enable
= true;
4666 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4667 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4668 num_shaders
[shader_type
]++;
4671 /* In desktop GLSL, different shader versions may be linked together. In
4672 * GLSL ES, all shader versions must be the same.
4674 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4675 linker_error(prog
, "all shaders must use same shading "
4676 "language version\n");
4680 prog
->data
->Version
= max_version
;
4681 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4683 /* Some shaders have to be linked with some other shaders present.
4685 if (!prog
->SeparateShader
) {
4686 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4687 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4688 linker_error(prog
, "Geometry shader must be linked with "
4692 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4693 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4694 linker_error(prog
, "Tessellation evaluation shader must be linked "
4695 "with vertex shader\n");
4698 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4699 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4700 linker_error(prog
, "Tessellation control shader must be linked with "
4705 /* The spec is self-contradictory here. It allows linking without a tess
4706 * eval shader, but that can only be used with transform feedback and
4707 * rasterization disabled. However, transform feedback isn't allowed
4708 * with GL_PATCHES, so it can't be used.
4710 * More investigation showed that the idea of transform feedback after
4711 * a tess control shader was dropped, because some hw vendors couldn't
4712 * support tessellation without a tess eval shader, but the linker
4713 * section wasn't updated to reflect that.
4715 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4718 * Do what's reasonable and always require a tess eval shader if a tess
4719 * control shader is present.
4721 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4722 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4723 linker_error(prog
, "Tessellation control shader must be linked with "
4724 "tessellation evaluation shader\n");
4729 /* Compute shaders have additional restrictions. */
4730 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4731 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4732 linker_error(prog
, "Compute shaders may not be linked with any other "
4733 "type of shader\n");
4736 /* Link all shaders for a particular stage and validate the result.
4738 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4739 if (num_shaders
[stage
] > 0) {
4740 gl_linked_shader
*const sh
=
4741 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4742 num_shaders
[stage
], false);
4744 if (!prog
->data
->LinkStatus
) {
4746 _mesa_delete_linked_shader(ctx
, sh
);
4751 case MESA_SHADER_VERTEX
:
4752 validate_vertex_shader_executable(prog
, sh
, ctx
);
4754 case MESA_SHADER_TESS_CTRL
:
4755 /* nothing to be done */
4757 case MESA_SHADER_TESS_EVAL
:
4758 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4760 case MESA_SHADER_GEOMETRY
:
4761 validate_geometry_shader_executable(prog
, sh
, ctx
);
4763 case MESA_SHADER_FRAGMENT
:
4764 validate_fragment_shader_executable(prog
, sh
);
4767 if (!prog
->data
->LinkStatus
) {
4769 _mesa_delete_linked_shader(ctx
, sh
);
4773 prog
->_LinkedShaders
[stage
] = sh
;
4774 prog
->data
->linked_stages
|= 1 << stage
;
4778 /* Here begins the inter-stage linking phase. Some initial validation is
4779 * performed, then locations are assigned for uniforms, attributes, and
4782 cross_validate_uniforms(prog
);
4783 if (!prog
->data
->LinkStatus
)
4786 unsigned first
, last
, prev
;
4788 first
= MESA_SHADER_STAGES
;
4791 /* Determine first and last stage. */
4792 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4793 if (!prog
->_LinkedShaders
[i
])
4795 if (first
== MESA_SHADER_STAGES
)
4800 check_explicit_uniform_locations(ctx
, prog
);
4801 link_assign_subroutine_types(prog
);
4803 if (!prog
->data
->LinkStatus
)
4806 resize_tes_inputs(ctx
, prog
);
4808 /* Validate the inputs of each stage with the output of the preceding
4812 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4813 if (prog
->_LinkedShaders
[i
] == NULL
)
4816 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4817 prog
->_LinkedShaders
[i
]);
4818 if (!prog
->data
->LinkStatus
)
4821 cross_validate_outputs_to_inputs(prog
,
4822 prog
->_LinkedShaders
[prev
],
4823 prog
->_LinkedShaders
[i
]);
4824 if (!prog
->data
->LinkStatus
)
4830 /* Cross-validate uniform blocks between shader stages */
4831 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4832 if (!prog
->data
->LinkStatus
)
4835 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4836 if (prog
->_LinkedShaders
[i
] != NULL
)
4837 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4840 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4841 * it before optimization because we want most of the checks to get
4842 * dropped thanks to constant propagation.
4844 * This rule also applies to GLSL ES 3.00.
4846 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4847 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4849 lower_discard_flow(sh
->ir
);
4853 if (prog
->SeparateShader
)
4854 disable_varying_optimizations_for_sso(prog
);
4857 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4861 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4864 /* Do common optimization before assigning storage for attributes,
4865 * uniforms, and varyings. Later optimization could possibly make
4866 * some of that unused.
4868 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4869 if (prog
->_LinkedShaders
[i
] == NULL
)
4872 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4873 if (!prog
->data
->LinkStatus
)
4876 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4877 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4880 if (ctx
->Const
.LowerTessLevel
) {
4881 lower_tess_level(prog
->_LinkedShaders
[i
]);
4884 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4885 /* Run it just once. */
4886 do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4887 &ctx
->Const
.ShaderCompilerOptions
[i
],
4888 ctx
->Const
.NativeIntegers
);
4890 /* Repeat it until it stops making changes. */
4891 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4892 &ctx
->Const
.ShaderCompilerOptions
[i
],
4893 ctx
->Const
.NativeIntegers
))
4897 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
4898 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4901 /* Validation for special cases where we allow sampler array indexing
4902 * with loop induction variable. This check emits a warning or error
4903 * depending if backend can handle dynamic indexing.
4905 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
4906 (prog
->IsES
&& prog
->data
->Version
< 300)) {
4907 if (!validate_sampler_array_indexing(ctx
, prog
))
4911 /* Check and validate stream emissions in geometry shaders */
4912 validate_geometry_shader_emissions(ctx
, prog
);
4914 store_fragdepth_layout(prog
);
4916 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
4919 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4920 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4921 * anything about shader linking when one of the shaders (vertex or
4922 * fragment shader) is absent. So, the extension shouldn't change the
4923 * behavior specified in GLSL specification.
4925 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4926 * "Linking can fail for a variety of reasons as specified in the
4927 * OpenGL ES Shading Language Specification, as well as any of the
4928 * following reasons:
4932 * * program contains objects to form either a vertex shader or
4933 * fragment shader, and program is not separable, and does not
4934 * contain objects to form both a vertex shader and fragment
4937 * However, the only scenario in 3.1+ where we don't require them both is
4938 * when we have a compute shader. For example:
4940 * - No shaders is a link error.
4941 * - Geom or Tess without a Vertex shader is a link error which means we
4942 * always require a Vertex shader and hence a Fragment shader.
4943 * - Finally a Compute shader linked with any other stage is a link error.
4945 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4946 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4947 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4948 linker_error(prog
, "program lacks a vertex shader\n");
4949 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4950 linker_error(prog
, "program lacks a fragment shader\n");
4955 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4956 free(shader_list
[i
]);
4957 if (prog
->_LinkedShaders
[i
] == NULL
)
4960 /* Do a final validation step to make sure that the IR wasn't
4961 * invalidated by any modifications performed after intrastage linking.
4963 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4965 /* Retain any live IR, but trash the rest. */
4966 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4968 /* The symbol table in the linked shaders may contain references to
4969 * variables that were removed (e.g., unused uniforms). Since it may
4970 * contain junk, there is no possible valid use. Delete it and set the
4973 delete prog
->_LinkedShaders
[i
]->symbols
;
4974 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4977 ralloc_free(mem_ctx
);