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"
76 #include "util/mesa-sha1.h"
78 #include "util/string_to_uint_map.h"
80 #include "link_varyings.h"
81 #include "ir_optimization.h"
82 #include "ir_rvalue_visitor.h"
83 #include "ir_uniform.h"
84 #include "builtin_functions.h"
85 #include "shader_cache.h"
87 #include "main/shaderobj.h"
88 #include "main/enums.h"
94 * Visitor that determines whether or not a variable is ever written.
96 class find_assignment_visitor
: public ir_hierarchical_visitor
{
98 find_assignment_visitor(const char *name
)
99 : name(name
), found(false)
104 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
106 ir_variable
*const var
= ir
->lhs
->variable_referenced();
108 if (strcmp(name
, var
->name
) == 0) {
113 return visit_continue_with_parent
;
116 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
118 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
119 actual_node
, &ir
->actual_parameters
) {
120 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
121 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
123 if (sig_param
->data
.mode
== ir_var_function_out
||
124 sig_param
->data
.mode
== ir_var_function_inout
) {
125 ir_variable
*var
= param_rval
->variable_referenced();
126 if (var
&& strcmp(name
, var
->name
) == 0) {
133 if (ir
->return_deref
!= NULL
) {
134 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
136 if (strcmp(name
, var
->name
) == 0) {
142 return visit_continue_with_parent
;
145 bool variable_found()
151 const char *name
; /**< Find writes to a variable with this name. */
152 bool found
; /**< Was a write to the variable found? */
157 * Visitor that determines whether or not a variable is ever read.
159 class find_deref_visitor
: public ir_hierarchical_visitor
{
161 find_deref_visitor(const char *name
)
162 : name(name
), found(false)
167 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
169 if (strcmp(this->name
, ir
->var
->name
) == 0) {
174 return visit_continue
;
177 bool variable_found() const
183 const char *name
; /**< Find writes to a variable with this name. */
184 bool found
; /**< Was a write to the variable found? */
189 * A visitor helper that provides methods for updating the types of
190 * ir_dereferences. Classes that update variable types (say, updating
191 * array sizes) will want to use this so that dereference types stay in sync.
193 class deref_type_updater
: public ir_hierarchical_visitor
{
195 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
197 ir
->type
= ir
->var
->type
;
198 return visit_continue
;
201 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
203 const glsl_type
*const vt
= ir
->array
->type
;
205 ir
->type
= vt
->fields
.array
;
206 return visit_continue
;
209 virtual ir_visitor_status
visit_leave(ir_dereference_record
*ir
)
211 for (unsigned i
= 0; i
< ir
->record
->type
->length
; i
++) {
212 const struct glsl_struct_field
*field
=
213 &ir
->record
->type
->fields
.structure
[i
];
214 if (strcmp(field
->name
, ir
->field
) == 0) {
215 ir
->type
= field
->type
;
219 return visit_continue
;
224 class array_resize_visitor
: public deref_type_updater
{
226 unsigned num_vertices
;
227 gl_shader_program
*prog
;
228 gl_shader_stage stage
;
230 array_resize_visitor(unsigned num_vertices
,
231 gl_shader_program
*prog
,
232 gl_shader_stage stage
)
234 this->num_vertices
= num_vertices
;
239 virtual ~array_resize_visitor()
244 virtual ir_visitor_status
visit(ir_variable
*var
)
246 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
248 return visit_continue
;
250 unsigned size
= var
->type
->length
;
252 if (stage
== MESA_SHADER_GEOMETRY
) {
253 /* Generate a link error if the shader has declared this array with
256 if (!var
->data
.implicit_sized_array
&&
257 size
&& size
!= this->num_vertices
) {
258 linker_error(this->prog
, "size of array %s declared as %u, "
259 "but number of input vertices is %u\n",
260 var
->name
, size
, this->num_vertices
);
261 return visit_continue
;
264 /* Generate a link error if the shader attempts to access an input
265 * array using an index too large for its actual size assigned at
268 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
269 linker_error(this->prog
, "%s shader accesses element %i of "
270 "%s, but only %i input vertices\n",
271 _mesa_shader_stage_to_string(this->stage
),
272 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
273 return visit_continue
;
277 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
279 var
->data
.max_array_access
= this->num_vertices
- 1;
281 return visit_continue
;
286 * Visitor that determines the highest stream id to which a (geometry) shader
287 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
289 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
291 find_emit_vertex_visitor(int max_allowed
)
292 : max_stream_allowed(max_allowed
),
293 invalid_stream_id(0),
294 invalid_stream_id_from_emit_vertex(false),
295 end_primitive_found(false),
296 uses_non_zero_stream(false)
301 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
303 int stream_id
= ir
->stream_id();
306 invalid_stream_id
= stream_id
;
307 invalid_stream_id_from_emit_vertex
= true;
311 if (stream_id
> max_stream_allowed
) {
312 invalid_stream_id
= stream_id
;
313 invalid_stream_id_from_emit_vertex
= true;
318 uses_non_zero_stream
= true;
320 return visit_continue
;
323 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
325 end_primitive_found
= true;
327 int stream_id
= ir
->stream_id();
330 invalid_stream_id
= stream_id
;
331 invalid_stream_id_from_emit_vertex
= false;
335 if (stream_id
> max_stream_allowed
) {
336 invalid_stream_id
= stream_id
;
337 invalid_stream_id_from_emit_vertex
= false;
342 uses_non_zero_stream
= true;
344 return visit_continue
;
349 return invalid_stream_id
!= 0;
352 const char *error_func()
354 return invalid_stream_id_from_emit_vertex
?
355 "EmitStreamVertex" : "EndStreamPrimitive";
360 return invalid_stream_id
;
365 return uses_non_zero_stream
;
368 bool uses_end_primitive()
370 return end_primitive_found
;
374 int max_stream_allowed
;
375 int invalid_stream_id
;
376 bool invalid_stream_id_from_emit_vertex
;
377 bool end_primitive_found
;
378 bool uses_non_zero_stream
;
381 /* Class that finds array derefs and check if indexes are dynamic. */
382 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
385 dynamic_sampler_array_indexing_visitor() :
386 dynamic_sampler_array_indexing(false)
390 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
392 if (!ir
->variable_referenced())
393 return visit_continue
;
395 if (!ir
->variable_referenced()->type
->contains_sampler())
396 return visit_continue
;
398 if (!ir
->array_index
->constant_expression_value()) {
399 dynamic_sampler_array_indexing
= true;
402 return visit_continue
;
405 bool uses_dynamic_sampler_array_indexing()
407 return dynamic_sampler_array_indexing
;
411 bool dynamic_sampler_array_indexing
;
414 } /* anonymous namespace */
417 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
421 ralloc_strcat(&prog
->data
->InfoLog
, "error: ");
423 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
426 prog
->data
->LinkStatus
= linking_failure
;
431 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
435 ralloc_strcat(&prog
->data
->InfoLog
, "warning: ");
437 ralloc_vasprintf_append(&prog
->data
->InfoLog
, fmt
, ap
);
444 * Given a string identifying a program resource, break it into a base name
445 * and an optional array index in square brackets.
447 * If an array index is present, \c out_base_name_end is set to point to the
448 * "[" that precedes the array index, and the array index itself is returned
451 * If no array index is present (or if the array index is negative or
452 * mal-formed), \c out_base_name_end, is set to point to the null terminator
453 * at the end of the input string, and -1 is returned.
455 * Only the final array index is parsed; if the string contains other array
456 * indices (or structure field accesses), they are left in the base name.
458 * No attempt is made to check that the base name is properly formed;
459 * typically the caller will look up the base name in a hash table, so
460 * ill-formed base names simply turn into hash table lookup failures.
463 parse_program_resource_name(const GLchar
*name
,
464 const GLchar
**out_base_name_end
)
466 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
468 * "When an integer array element or block instance number is part of
469 * the name string, it will be specified in decimal form without a "+"
470 * or "-" sign or any extra leading zeroes. Additionally, the name
471 * string will not include white space anywhere in the string."
474 const size_t len
= strlen(name
);
475 *out_base_name_end
= name
+ len
;
477 if (len
== 0 || name
[len
-1] != ']')
480 /* Walk backwards over the string looking for a non-digit character. This
481 * had better be the opening bracket for an array index.
483 * Initially, i specifies the location of the ']'. Since the string may
484 * contain only the ']' charcater, walk backwards very carefully.
487 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
490 if ((i
== 0) || name
[i
-1] != '[')
493 long array_index
= strtol(&name
[i
], NULL
, 10);
497 /* Check for leading zero */
498 if (name
[i
] == '0' && name
[i
+1] != ']')
501 *out_base_name_end
= name
+ (i
- 1);
507 link_invalidate_variable_locations(exec_list
*ir
)
509 foreach_in_list(ir_instruction
, node
, ir
) {
510 ir_variable
*const var
= node
->as_variable();
515 /* Only assign locations for variables that lack an explicit location.
516 * Explicit locations are set for all built-in variables, generic vertex
517 * shader inputs (via layout(location=...)), and generic fragment shader
518 * outputs (also via layout(location=...)).
520 if (!var
->data
.explicit_location
) {
521 var
->data
.location
= -1;
522 var
->data
.location_frac
= 0;
525 /* ir_variable::is_unmatched_generic_inout is used by the linker while
526 * connecting outputs from one stage to inputs of the next stage.
528 if (var
->data
.explicit_location
&&
529 var
->data
.location
< VARYING_SLOT_VAR0
) {
530 var
->data
.is_unmatched_generic_inout
= 0;
532 var
->data
.is_unmatched_generic_inout
= 1;
539 * Set clip_distance_array_size based and cull_distance_array_size on the given
542 * Also check for errors based on incorrect usage of gl_ClipVertex and
543 * gl_ClipDistance and gl_CullDistance.
544 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
545 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
547 * Return false if an error was reported.
550 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
551 struct gl_linked_shader
*shader
,
552 struct gl_context
*ctx
,
553 GLuint
*clip_distance_array_size
,
554 GLuint
*cull_distance_array_size
)
556 *clip_distance_array_size
= 0;
557 *cull_distance_array_size
= 0;
559 if (prog
->data
->Version
>= (prog
->IsES
? 300 : 130)) {
560 /* From section 7.1 (Vertex Shader Special Variables) of the
563 * "It is an error for a shader to statically write both
564 * gl_ClipVertex and gl_ClipDistance."
566 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
567 * gl_ClipVertex nor gl_ClipDistance. However with
568 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
570 find_assignment_visitor
clip_distance("gl_ClipDistance");
571 find_assignment_visitor
cull_distance("gl_CullDistance");
573 clip_distance
.run(shader
->ir
);
574 cull_distance
.run(shader
->ir
);
576 /* From the ARB_cull_distance spec:
578 * It is a compile-time or link-time error for the set of shaders forming
579 * a program to statically read or write both gl_ClipVertex and either
580 * gl_ClipDistance or gl_CullDistance.
582 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
586 find_assignment_visitor
clip_vertex("gl_ClipVertex");
588 clip_vertex
.run(shader
->ir
);
590 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
591 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
592 "and `gl_ClipDistance'\n",
593 _mesa_shader_stage_to_string(shader
->Stage
));
596 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
597 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
598 "and `gl_CullDistance'\n",
599 _mesa_shader_stage_to_string(shader
->Stage
));
604 if (clip_distance
.variable_found()) {
605 ir_variable
*clip_distance_var
=
606 shader
->symbols
->get_variable("gl_ClipDistance");
607 assert(clip_distance_var
);
608 *clip_distance_array_size
= clip_distance_var
->type
->length
;
610 if (cull_distance
.variable_found()) {
611 ir_variable
*cull_distance_var
=
612 shader
->symbols
->get_variable("gl_CullDistance");
613 assert(cull_distance_var
);
614 *cull_distance_array_size
= cull_distance_var
->type
->length
;
616 /* From the ARB_cull_distance spec:
618 * It is a compile-time or link-time error for the set of shaders forming
619 * a program to have the sum of the sizes of the gl_ClipDistance and
620 * gl_CullDistance arrays to be larger than
621 * gl_MaxCombinedClipAndCullDistances.
623 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
624 ctx
->Const
.MaxClipPlanes
) {
625 linker_error(prog
, "%s shader: the combined size of "
626 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
628 "gl_MaxCombinedClipAndCullDistances (%u)",
629 _mesa_shader_stage_to_string(shader
->Stage
),
630 ctx
->Const
.MaxClipPlanes
);
637 * Verify that a vertex shader executable meets all semantic requirements.
639 * Also sets info.clip_distance_array_size and
640 * info.cull_distance_array_size as a side effect.
642 * \param shader Vertex shader executable to be verified
645 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
646 struct gl_linked_shader
*shader
,
647 struct gl_context
*ctx
)
652 /* From the GLSL 1.10 spec, page 48:
654 * "The variable gl_Position is available only in the vertex
655 * language and is intended for writing the homogeneous vertex
656 * position. All executions of a well-formed vertex shader
657 * executable must write a value into this variable. [...] The
658 * variable gl_Position is available only in the vertex
659 * language and is intended for writing the homogeneous vertex
660 * position. All executions of a well-formed vertex shader
661 * executable must write a value into this variable."
663 * while in GLSL 1.40 this text is changed to:
665 * "The variable gl_Position is available only in the vertex
666 * language and is intended for writing the homogeneous vertex
667 * position. It can be written at any time during shader
668 * execution. It may also be read back by a vertex shader
669 * after being written. This value will be used by primitive
670 * assembly, clipping, culling, and other fixed functionality
671 * operations, if present, that operate on primitives after
672 * vertex processing has occurred. Its value is undefined if
673 * the vertex shader executable does not write gl_Position."
675 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
676 * gl_Position is not an error.
678 if (prog
->data
->Version
< (prog
->IsES
? 300 : 140)) {
679 find_assignment_visitor
find("gl_Position");
680 find
.run(shader
->ir
);
681 if (!find
.variable_found()) {
684 "vertex shader does not write to `gl_Position'. "
685 "Its value is undefined. \n");
688 "vertex shader does not write to `gl_Position'. \n");
694 analyze_clip_cull_usage(prog
, shader
, ctx
,
695 &shader
->Program
->info
.clip_distance_array_size
,
696 &shader
->Program
->info
.cull_distance_array_size
);
700 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
701 struct gl_linked_shader
*shader
,
702 struct gl_context
*ctx
)
707 analyze_clip_cull_usage(prog
, shader
, ctx
,
708 &shader
->Program
->info
.clip_distance_array_size
,
709 &shader
->Program
->info
.cull_distance_array_size
);
714 * Verify that a fragment shader executable meets all semantic requirements
716 * \param shader Fragment shader executable to be verified
719 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
720 struct gl_linked_shader
*shader
)
725 find_assignment_visitor
frag_color("gl_FragColor");
726 find_assignment_visitor
frag_data("gl_FragData");
728 frag_color
.run(shader
->ir
);
729 frag_data
.run(shader
->ir
);
731 if (frag_color
.variable_found() && frag_data
.variable_found()) {
732 linker_error(prog
, "fragment shader writes to both "
733 "`gl_FragColor' and `gl_FragData'\n");
738 * Verify that a geometry shader executable meets all semantic requirements
740 * Also sets prog->Geom.VerticesIn, and info.clip_distance_array_sizeand
741 * info.cull_distance_array_size as a side effect.
743 * \param shader Geometry shader executable to be verified
746 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
747 struct gl_linked_shader
*shader
,
748 struct gl_context
*ctx
)
753 unsigned num_vertices
=
754 vertices_per_prim(shader
->Program
->info
.gs
.input_primitive
);
755 prog
->Geom
.VerticesIn
= num_vertices
;
757 analyze_clip_cull_usage(prog
, shader
, ctx
,
758 &shader
->Program
->info
.clip_distance_array_size
,
759 &shader
->Program
->info
.cull_distance_array_size
);
763 * Check if geometry shaders emit to non-zero streams and do corresponding
767 validate_geometry_shader_emissions(struct gl_context
*ctx
,
768 struct gl_shader_program
*prog
)
770 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
773 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
774 emit_vertex
.run(sh
->ir
);
775 if (emit_vertex
.error()) {
776 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
777 "stream parameter are in the range [0, %d].\n",
778 emit_vertex
.error_func(),
779 emit_vertex
.error_stream(),
780 ctx
->Const
.MaxVertexStreams
- 1);
782 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
783 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
785 /* From the ARB_gpu_shader5 spec:
787 * "Multiple vertex streams are supported only if the output primitive
788 * type is declared to be "points". A program will fail to link if it
789 * contains a geometry shader calling EmitStreamVertex() or
790 * EndStreamPrimitive() if its output primitive type is not "points".
792 * However, in the same spec:
794 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
795 * with <stream> set to zero."
799 * "The function EndPrimitive() is equivalent to calling
800 * EndStreamPrimitive() with <stream> set to zero."
802 * Since we can call EmitVertex() and EndPrimitive() when we output
803 * primitives other than points, calling EmitStreamVertex(0) or
804 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
805 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
806 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
809 if (prog
->Geom
.UsesStreams
&&
810 sh
->Program
->info
.gs
.output_primitive
!= GL_POINTS
) {
811 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
812 "with n>0 requires point output\n");
818 validate_intrastage_arrays(struct gl_shader_program
*prog
,
819 ir_variable
*const var
,
820 ir_variable
*const existing
)
822 /* Consider the types to be "the same" if both types are arrays
823 * of the same type and one of the arrays is implicitly sized.
824 * In addition, set the type of the linked variable to the
825 * explicitly sized array.
827 if (var
->type
->is_array() && existing
->type
->is_array()) {
828 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
829 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
830 if (var
->type
->length
!= 0) {
831 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
832 linker_error(prog
, "%s `%s' declared as type "
833 "`%s' but outermost dimension has an index"
836 var
->name
, var
->type
->name
,
837 existing
->data
.max_array_access
);
839 existing
->type
= var
->type
;
841 } else if (existing
->type
->length
!= 0) {
842 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
843 !existing
->data
.from_ssbo_unsized_array
) {
844 linker_error(prog
, "%s `%s' declared as type "
845 "`%s' but outermost dimension has an index"
848 var
->name
, existing
->type
->name
,
849 var
->data
.max_array_access
);
854 /* The arrays of structs could have different glsl_type pointers but
855 * they are actually the same type. Use record_compare() to check that.
857 if (existing
->type
->fields
.array
->is_record() &&
858 var
->type
->fields
.array
->is_record() &&
859 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
868 * Perform validation of global variables used across multiple shaders
871 cross_validate_globals(struct gl_shader_program
*prog
,
872 struct exec_list
*ir
, glsl_symbol_table
*variables
,
875 foreach_in_list(ir_instruction
, node
, ir
) {
876 ir_variable
*const var
= node
->as_variable();
881 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
884 /* don't cross validate subroutine uniforms */
885 if (var
->type
->contains_subroutine())
888 /* Don't cross validate interface instances. These are only relevant
889 * inside a shader. The cross validation is done at the Interface Block
892 if (var
->is_interface_instance())
895 /* Don't cross validate temporaries that are at global scope. These
896 * will eventually get pulled into the shaders 'main'.
898 if (var
->data
.mode
== ir_var_temporary
)
901 /* If a global with this name has already been seen, verify that the
902 * new instance has the same type. In addition, if the globals have
903 * initializers, the values of the initializers must be the same.
905 ir_variable
*const existing
= variables
->get_variable(var
->name
);
906 if (existing
!= NULL
) {
907 /* Check if types match. */
908 if (var
->type
!= existing
->type
) {
909 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
910 if (var
->type
->is_record() && existing
->type
->is_record()
911 && existing
->type
->record_compare(var
->type
)) {
912 existing
->type
= var
->type
;
914 /* If it is an unsized array in a Shader Storage Block,
915 * two different shaders can access to different elements.
916 * Because of that, they might be converted to different
917 * sized arrays, then check that they are compatible but
918 * ignore the array size.
920 if (!(var
->data
.mode
== ir_var_shader_storage
&&
921 var
->data
.from_ssbo_unsized_array
&&
922 existing
->data
.mode
== ir_var_shader_storage
&&
923 existing
->data
.from_ssbo_unsized_array
&&
924 var
->type
->gl_type
== existing
->type
->gl_type
)) {
925 linker_error(prog
, "%s `%s' declared as type "
926 "`%s' and type `%s'\n",
928 var
->name
, var
->type
->name
,
929 existing
->type
->name
);
936 if (var
->data
.explicit_location
) {
937 if (existing
->data
.explicit_location
938 && (var
->data
.location
!= existing
->data
.location
)) {
939 linker_error(prog
, "explicit locations for %s "
940 "`%s' have differing values\n",
941 mode_string(var
), var
->name
);
945 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
946 linker_error(prog
, "explicit components for %s `%s' have "
947 "differing values\n", mode_string(var
), var
->name
);
951 existing
->data
.location
= var
->data
.location
;
952 existing
->data
.explicit_location
= true;
954 /* Check if uniform with implicit location was marked explicit
955 * by earlier shader stage. If so, mark it explicit in this stage
956 * too to make sure later processing does not treat it as
959 if (existing
->data
.explicit_location
) {
960 var
->data
.location
= existing
->data
.location
;
961 var
->data
.explicit_location
= true;
965 /* From the GLSL 4.20 specification:
966 * "A link error will result if two compilation units in a program
967 * specify different integer-constant bindings for the same
968 * opaque-uniform name. However, it is not an error to specify a
969 * binding on some but not all declarations for the same name"
971 if (var
->data
.explicit_binding
) {
972 if (existing
->data
.explicit_binding
&&
973 var
->data
.binding
!= existing
->data
.binding
) {
974 linker_error(prog
, "explicit bindings for %s "
975 "`%s' have differing values\n",
976 mode_string(var
), var
->name
);
980 existing
->data
.binding
= var
->data
.binding
;
981 existing
->data
.explicit_binding
= true;
984 if (var
->type
->contains_atomic() &&
985 var
->data
.offset
!= existing
->data
.offset
) {
986 linker_error(prog
, "offset specifications for %s "
987 "`%s' have differing values\n",
988 mode_string(var
), var
->name
);
992 /* Validate layout qualifiers for gl_FragDepth.
994 * From the AMD/ARB_conservative_depth specs:
996 * "If gl_FragDepth is redeclared in any fragment shader in a
997 * program, it must be redeclared in all fragment shaders in
998 * that program that have static assignments to
999 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1000 * fragment shaders in a single program must have the same set
1003 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1004 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1005 bool layout_differs
=
1006 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1008 if (layout_declared
&& layout_differs
) {
1010 "All redeclarations of gl_FragDepth in all "
1011 "fragment shaders in a single program must have "
1012 "the same set of qualifiers.\n");
1015 if (var
->data
.used
&& layout_differs
) {
1017 "If gl_FragDepth is redeclared with a layout "
1018 "qualifier in any fragment shader, it must be "
1019 "redeclared with the same layout qualifier in "
1020 "all fragment shaders that have assignments to "
1025 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1027 * "If a shared global has multiple initializers, the
1028 * initializers must all be constant expressions, and they
1029 * must all have the same value. Otherwise, a link error will
1030 * result. (A shared global having only one initializer does
1031 * not require that initializer to be a constant expression.)"
1033 * Previous to 4.20 the GLSL spec simply said that initializers
1034 * must have the same value. In this case of non-constant
1035 * initializers, this was impossible to determine. As a result,
1036 * no vendor actually implemented that behavior. The 4.20
1037 * behavior matches the implemented behavior of at least one other
1038 * vendor, so we'll implement that for all GLSL versions.
1040 if (var
->constant_initializer
!= NULL
) {
1041 if (existing
->constant_initializer
!= NULL
) {
1042 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1043 linker_error(prog
, "initializers for %s "
1044 "`%s' have differing values\n",
1045 mode_string(var
), var
->name
);
1049 /* If the first-seen instance of a particular uniform did
1050 * not have an initializer but a later instance does,
1051 * replace the former with the later.
1053 variables
->replace_variable(existing
->name
, var
);
1057 if (var
->data
.has_initializer
) {
1058 if (existing
->data
.has_initializer
1059 && (var
->constant_initializer
== NULL
1060 || existing
->constant_initializer
== NULL
)) {
1062 "shared global variable `%s' has multiple "
1063 "non-constant initializers.\n",
1069 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1070 linker_error(prog
, "declarations for %s `%s' have "
1071 "mismatching invariant qualifiers\n",
1072 mode_string(var
), var
->name
);
1075 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1076 linker_error(prog
, "declarations for %s `%s' have "
1077 "mismatching centroid qualifiers\n",
1078 mode_string(var
), var
->name
);
1081 if (existing
->data
.sample
!= var
->data
.sample
) {
1082 linker_error(prog
, "declarations for %s `%s` have "
1083 "mismatching sample qualifiers\n",
1084 mode_string(var
), var
->name
);
1087 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1088 linker_error(prog
, "declarations for %s `%s` have "
1089 "mismatching image format qualifiers\n",
1090 mode_string(var
), var
->name
);
1094 /* Only in GLSL ES 3.10, the precision qualifier should not match
1095 * between block members defined in matched block names within a
1098 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1099 * member should match.
1101 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1102 !var
->get_interface_type()) &&
1103 existing
->data
.precision
!= var
->data
.precision
) {
1104 linker_error(prog
, "declarations for %s `%s` have "
1105 "mismatching precision qualifiers\n",
1106 mode_string(var
), var
->name
);
1110 variables
->add_variable(var
);
1116 * Perform validation of uniforms used across multiple shader stages
1119 cross_validate_uniforms(struct gl_shader_program
*prog
)
1121 glsl_symbol_table variables
;
1122 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1123 if (prog
->_LinkedShaders
[i
] == NULL
)
1126 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1132 * Accumulates the array of buffer blocks and checks that all definitions of
1133 * blocks agree on their contents.
1136 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1139 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1140 struct gl_uniform_block
*blks
= NULL
;
1141 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1142 &prog
->data
->NumUniformBlocks
;
1144 unsigned max_num_buffer_blocks
= 0;
1145 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1146 if (prog
->_LinkedShaders
[i
]) {
1147 if (validate_ssbo
) {
1148 max_num_buffer_blocks
+=
1149 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1151 max_num_buffer_blocks
+=
1152 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1157 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1158 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1160 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1161 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1162 InterfaceBlockStageIndex
[i
][j
] = -1;
1167 unsigned sh_num_blocks
;
1168 struct gl_uniform_block
**sh_blks
;
1169 if (validate_ssbo
) {
1170 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1171 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1173 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1174 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1177 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1178 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1182 linker_error(prog
, "buffer block `%s' has mismatching "
1183 "definitions\n", sh_blks
[j
]->Name
);
1185 for (unsigned k
= 0; k
<= i
; k
++) {
1186 delete[] InterfaceBlockStageIndex
[k
];
1189 /* Reset the block count. This will help avoid various segfaults
1190 * from api calls that assume the array exists due to the count
1197 InterfaceBlockStageIndex
[i
][index
] = j
;
1201 /* Update per stage block pointers to point to the program list.
1202 * FIXME: We should be able to free the per stage blocks here.
1204 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1205 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1206 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1208 if (stage_index
!= -1) {
1209 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1211 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1212 sh
->Program
->sh
.ShaderStorageBlocks
:
1213 sh
->Program
->sh
.UniformBlocks
;
1215 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1216 sh_blks
[stage_index
] = &blks
[j
];
1221 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1222 delete[] InterfaceBlockStageIndex
[i
];
1226 prog
->data
->ShaderStorageBlocks
= blks
;
1228 prog
->data
->UniformBlocks
= blks
;
1235 * Populates a shaders symbol table with all global declarations
1238 populate_symbol_table(gl_linked_shader
*sh
)
1240 sh
->symbols
= new(sh
) glsl_symbol_table
;
1242 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1246 if ((func
= inst
->as_function()) != NULL
) {
1247 sh
->symbols
->add_function(func
);
1248 } else if ((var
= inst
->as_variable()) != NULL
) {
1249 if (var
->data
.mode
!= ir_var_temporary
)
1250 sh
->symbols
->add_variable(var
);
1257 * Remap variables referenced in an instruction tree
1259 * This is used when instruction trees are cloned from one shader and placed in
1260 * another. These trees will contain references to \c ir_variable nodes that
1261 * do not exist in the target shader. This function finds these \c ir_variable
1262 * references and replaces the references with matching variables in the target
1265 * If there is no matching variable in the target shader, a clone of the
1266 * \c ir_variable is made and added to the target shader. The new variable is
1267 * added to \b both the instruction stream and the symbol table.
1269 * \param inst IR tree that is to be processed.
1270 * \param symbols Symbol table containing global scope symbols in the
1272 * \param instructions Instruction stream where new variable declarations
1276 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1279 class remap_visitor
: public ir_hierarchical_visitor
{
1281 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1283 this->target
= target
;
1284 this->symbols
= target
->symbols
;
1285 this->instructions
= target
->ir
;
1286 this->temps
= temps
;
1289 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1291 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1292 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1293 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1295 assert(var
!= NULL
);
1297 return visit_continue
;
1300 ir_variable
*const existing
=
1301 this->symbols
->get_variable(ir
->var
->name
);
1302 if (existing
!= NULL
)
1305 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1307 this->symbols
->add_variable(copy
);
1308 this->instructions
->push_head(copy
);
1312 return visit_continue
;
1316 struct gl_linked_shader
*target
;
1317 glsl_symbol_table
*symbols
;
1318 exec_list
*instructions
;
1322 remap_visitor
v(target
, temps
);
1329 * Move non-declarations from one instruction stream to another
1331 * The intended usage pattern of this function is to pass the pointer to the
1332 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1333 * pointer) for \c last and \c false for \c make_copies on the first
1334 * call. Successive calls pass the return value of the previous call for
1335 * \c last and \c true for \c make_copies.
1337 * \param instructions Source instruction stream
1338 * \param last Instruction after which new instructions should be
1339 * inserted in the target instruction stream
1340 * \param make_copies Flag selecting whether instructions in \c instructions
1341 * should be copied (via \c ir_instruction::clone) into the
1342 * target list or moved.
1345 * The new "last" instruction in the target instruction stream. This pointer
1346 * is suitable for use as the \c last parameter of a later call to this
1350 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1351 bool make_copies
, gl_linked_shader
*target
)
1353 hash_table
*temps
= NULL
;
1356 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1357 _mesa_key_pointer_equal
);
1359 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1360 if (inst
->as_function())
1363 ir_variable
*var
= inst
->as_variable();
1364 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1367 assert(inst
->as_assignment()
1369 || inst
->as_if() /* for initializers with the ?: operator */
1370 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1373 inst
= inst
->clone(target
, NULL
);
1376 _mesa_hash_table_insert(temps
, var
, inst
);
1378 remap_variables(inst
, target
, temps
);
1383 last
->insert_after(inst
);
1388 _mesa_hash_table_destroy(temps
, NULL
);
1395 * This class is only used in link_intrastage_shaders() below but declaring
1396 * it inside that function leads to compiler warnings with some versions of
1399 class array_sizing_visitor
: public deref_type_updater
{
1401 array_sizing_visitor()
1402 : mem_ctx(ralloc_context(NULL
)),
1403 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1404 _mesa_key_pointer_equal
))
1408 ~array_sizing_visitor()
1410 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1411 ralloc_free(this->mem_ctx
);
1414 virtual ir_visitor_status
visit(ir_variable
*var
)
1416 const glsl_type
*type_without_array
;
1417 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1418 fixup_type(&var
->type
, var
->data
.max_array_access
,
1419 var
->data
.from_ssbo_unsized_array
,
1420 &implicit_sized_array
);
1421 var
->data
.implicit_sized_array
= implicit_sized_array
;
1422 type_without_array
= var
->type
->without_array();
1423 if (var
->type
->is_interface()) {
1424 if (interface_contains_unsized_arrays(var
->type
)) {
1425 const glsl_type
*new_type
=
1426 resize_interface_members(var
->type
,
1427 var
->get_max_ifc_array_access(),
1428 var
->is_in_shader_storage_block());
1429 var
->type
= new_type
;
1430 var
->change_interface_type(new_type
);
1432 } else if (type_without_array
->is_interface()) {
1433 if (interface_contains_unsized_arrays(type_without_array
)) {
1434 const glsl_type
*new_type
=
1435 resize_interface_members(type_without_array
,
1436 var
->get_max_ifc_array_access(),
1437 var
->is_in_shader_storage_block());
1438 var
->change_interface_type(new_type
);
1439 var
->type
= update_interface_members_array(var
->type
, new_type
);
1441 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1442 /* Store a pointer to the variable in the unnamed_interfaces
1446 _mesa_hash_table_search(this->unnamed_interfaces
,
1449 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1451 if (interface_vars
== NULL
) {
1452 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1454 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1457 unsigned index
= ifc_type
->field_index(var
->name
);
1458 assert(index
< ifc_type
->length
);
1459 assert(interface_vars
[index
] == NULL
);
1460 interface_vars
[index
] = var
;
1462 return visit_continue
;
1466 * For each unnamed interface block that was discovered while running the
1467 * visitor, adjust the interface type to reflect the newly assigned array
1468 * sizes, and fix up the ir_variable nodes to point to the new interface
1471 void fixup_unnamed_interface_types()
1473 hash_table_call_foreach(this->unnamed_interfaces
,
1474 fixup_unnamed_interface_type
, NULL
);
1479 * If the type pointed to by \c type represents an unsized array, replace
1480 * it with a sized array whose size is determined by max_array_access.
1482 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1483 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1485 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1486 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1487 max_array_access
+ 1);
1488 *implicit_sized
= true;
1489 assert(*type
!= NULL
);
1493 static const glsl_type
*
1494 update_interface_members_array(const glsl_type
*type
,
1495 const glsl_type
*new_interface_type
)
1497 const glsl_type
*element_type
= type
->fields
.array
;
1498 if (element_type
->is_array()) {
1499 const glsl_type
*new_array_type
=
1500 update_interface_members_array(element_type
, new_interface_type
);
1501 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1503 return glsl_type::get_array_instance(new_interface_type
,
1509 * Determine whether the given interface type contains unsized arrays (if
1510 * it doesn't, array_sizing_visitor doesn't need to process it).
1512 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1514 for (unsigned i
= 0; i
< type
->length
; i
++) {
1515 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1516 if (elem_type
->is_unsized_array())
1523 * Create a new interface type based on the given type, with unsized arrays
1524 * replaced by sized arrays whose size is determined by
1525 * max_ifc_array_access.
1527 static const glsl_type
*
1528 resize_interface_members(const glsl_type
*type
,
1529 const int *max_ifc_array_access
,
1532 unsigned num_fields
= type
->length
;
1533 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1534 memcpy(fields
, type
->fields
.structure
,
1535 num_fields
* sizeof(*fields
));
1536 for (unsigned i
= 0; i
< num_fields
; i
++) {
1537 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1538 /* If SSBO last member is unsized array, we don't replace it by a sized
1541 if (is_ssbo
&& i
== (num_fields
- 1))
1542 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1543 true, &implicit_sized_array
);
1545 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1546 false, &implicit_sized_array
);
1547 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1549 glsl_interface_packing packing
=
1550 (glsl_interface_packing
) type
->interface_packing
;
1551 bool row_major
= (bool) type
->interface_row_major
;
1552 const glsl_type
*new_ifc_type
=
1553 glsl_type::get_interface_instance(fields
, num_fields
,
1554 packing
, row_major
, type
->name
);
1556 return new_ifc_type
;
1559 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1562 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1563 ir_variable
**interface_vars
= (ir_variable
**) data
;
1564 unsigned num_fields
= ifc_type
->length
;
1565 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1566 memcpy(fields
, ifc_type
->fields
.structure
,
1567 num_fields
* sizeof(*fields
));
1568 bool interface_type_changed
= false;
1569 for (unsigned i
= 0; i
< num_fields
; i
++) {
1570 if (interface_vars
[i
] != NULL
&&
1571 fields
[i
].type
!= interface_vars
[i
]->type
) {
1572 fields
[i
].type
= interface_vars
[i
]->type
;
1573 interface_type_changed
= true;
1576 if (!interface_type_changed
) {
1580 glsl_interface_packing packing
=
1581 (glsl_interface_packing
) ifc_type
->interface_packing
;
1582 bool row_major
= (bool) ifc_type
->interface_row_major
;
1583 const glsl_type
*new_ifc_type
=
1584 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1585 row_major
, ifc_type
->name
);
1587 for (unsigned i
= 0; i
< num_fields
; i
++) {
1588 if (interface_vars
[i
] != NULL
)
1589 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1594 * Memory context used to allocate the data in \c unnamed_interfaces.
1599 * Hash table from const glsl_type * to an array of ir_variable *'s
1600 * pointing to the ir_variables constituting each unnamed interface block.
1602 hash_table
*unnamed_interfaces
;
1606 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1607 struct gl_shader_program
*prog
)
1609 /* We will validate doubles at a later stage */
1610 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1611 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1612 "multiple of 4 or if its applied to a type that is "
1613 "or contains a double a multiple of 8.",
1614 prog
->TransformFeedback
.BufferStride
[idx
]);
1618 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1619 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1620 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1621 "limit has been exceeded.");
1629 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1633 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1634 struct gl_shader_program
*prog
,
1635 struct gl_linked_shader
*linked_shader
,
1636 struct gl_shader
**shader_list
,
1637 unsigned num_shaders
)
1639 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1640 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1643 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1644 struct gl_shader
*shader
= shader_list
[i
];
1646 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1647 if (shader
->TransformFeedbackBufferStride
[j
]) {
1648 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1649 prog
->TransformFeedback
.BufferStride
[j
] =
1650 shader
->TransformFeedbackBufferStride
[j
];
1651 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1653 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1654 shader
->TransformFeedbackBufferStride
[j
]){
1656 "intrastage shaders defined with conflicting "
1657 "xfb_stride for buffer %d (%d and %d)\n", j
,
1658 prog
->TransformFeedback
.BufferStride
[j
],
1659 shader
->TransformFeedbackBufferStride
[j
]);
1668 * Performs the cross-validation of tessellation control shader vertices and
1669 * layout qualifiers for the attached tessellation control shaders,
1670 * and propagates them to the linked TCS and linked shader program.
1673 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1674 struct gl_program
*gl_prog
,
1675 struct gl_shader
**shader_list
,
1676 unsigned num_shaders
)
1678 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1681 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1683 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1685 * "All tessellation control shader layout declarations in a program
1686 * must specify the same output patch vertex count. There must be at
1687 * least one layout qualifier specifying an output patch vertex count
1688 * in any program containing tessellation control shaders; however,
1689 * such a declaration is not required in all tessellation control
1693 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1694 struct gl_shader
*shader
= shader_list
[i
];
1696 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1697 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1698 gl_prog
->info
.tess
.tcs_vertices_out
!=
1699 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1700 linker_error(prog
, "tessellation control shader defined with "
1701 "conflicting output vertex count (%d and %d)\n",
1702 gl_prog
->info
.tess
.tcs_vertices_out
,
1703 shader
->info
.TessCtrl
.VerticesOut
);
1706 gl_prog
->info
.tess
.tcs_vertices_out
=
1707 shader
->info
.TessCtrl
.VerticesOut
;
1711 /* Just do the intrastage -> interstage propagation right now,
1712 * since we already know we're in the right type of shader program
1715 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1716 linker_error(prog
, "tessellation control shader didn't declare "
1717 "vertices out layout qualifier\n");
1724 * Performs the cross-validation of tessellation evaluation shader
1725 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1726 * for the attached tessellation evaluation shaders, and propagates them
1727 * to the linked TES and linked shader program.
1730 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1731 struct gl_program
*gl_prog
,
1732 struct gl_shader
**shader_list
,
1733 unsigned num_shaders
)
1735 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1738 int point_mode
= -1;
1739 unsigned vertex_order
= 0;
1741 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1742 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1744 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1746 * "At least one tessellation evaluation shader (compilation unit) in
1747 * a program must declare a primitive mode in its input layout.
1748 * Declaration vertex spacing, ordering, and point mode identifiers is
1749 * optional. It is not required that all tessellation evaluation
1750 * shaders in a program declare a primitive mode. If spacing or
1751 * vertex ordering declarations are omitted, the tessellation
1752 * primitive generator will use equal spacing or counter-clockwise
1753 * vertex ordering, respectively. If a point mode declaration is
1754 * omitted, the tessellation primitive generator will produce lines or
1755 * triangles according to the primitive mode."
1758 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1759 struct gl_shader
*shader
= shader_list
[i
];
1761 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1762 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1763 gl_prog
->info
.tess
.primitive_mode
!=
1764 shader
->info
.TessEval
.PrimitiveMode
) {
1765 linker_error(prog
, "tessellation evaluation shader defined with "
1766 "conflicting input primitive modes.\n");
1769 gl_prog
->info
.tess
.primitive_mode
=
1770 shader
->info
.TessEval
.PrimitiveMode
;
1773 if (shader
->info
.TessEval
.Spacing
!= 0) {
1774 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1775 shader
->info
.TessEval
.Spacing
) {
1776 linker_error(prog
, "tessellation evaluation shader defined with "
1777 "conflicting vertex spacing.\n");
1780 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1783 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1784 if (vertex_order
!= 0 &&
1785 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1786 linker_error(prog
, "tessellation evaluation shader defined with "
1787 "conflicting ordering.\n");
1790 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1793 if (shader
->info
.TessEval
.PointMode
!= -1) {
1794 if (point_mode
!= -1 &&
1795 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1796 linker_error(prog
, "tessellation evaluation shader defined with "
1797 "conflicting point modes.\n");
1800 point_mode
= shader
->info
.TessEval
.PointMode
;
1805 /* Just do the intrastage -> interstage propagation right now,
1806 * since we already know we're in the right type of shader program
1809 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1811 "tessellation evaluation shader didn't declare input "
1812 "primitive modes.\n");
1816 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1817 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1819 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1820 gl_prog
->info
.tess
.ccw
= true;
1822 gl_prog
->info
.tess
.ccw
= false;
1825 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1826 gl_prog
->info
.tess
.point_mode
= false;
1828 gl_prog
->info
.tess
.point_mode
= true;
1833 * Performs the cross-validation of layout qualifiers specified in
1834 * redeclaration of gl_FragCoord for the attached fragment shaders,
1835 * and propagates them to the linked FS and linked shader program.
1838 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1839 struct gl_linked_shader
*linked_shader
,
1840 struct gl_shader
**shader_list
,
1841 unsigned num_shaders
)
1843 bool redeclares_gl_fragcoord
= false;
1844 bool uses_gl_fragcoord
= false;
1845 bool origin_upper_left
= false;
1846 bool pixel_center_integer
= false;
1848 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1849 (prog
->data
->Version
< 150 &&
1850 !prog
->ARB_fragment_coord_conventions_enable
))
1853 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1854 struct gl_shader
*shader
= shader_list
[i
];
1855 /* From the GLSL 1.50 spec, page 39:
1857 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1858 * it must be redeclared in all the fragment shaders in that program
1859 * that have a static use gl_FragCoord."
1861 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1862 shader
->uses_gl_fragcoord
)
1863 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1864 uses_gl_fragcoord
)) {
1865 linker_error(prog
, "fragment shader defined with conflicting "
1866 "layout qualifiers for gl_FragCoord\n");
1869 /* From the GLSL 1.50 spec, page 39:
1871 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1872 * single program must have the same set of qualifiers."
1874 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1875 (shader
->origin_upper_left
!= origin_upper_left
||
1876 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1877 linker_error(prog
, "fragment shader defined with conflicting "
1878 "layout qualifiers for gl_FragCoord\n");
1881 /* Update the linked shader state. Note that uses_gl_fragcoord should
1882 * accumulate the results. The other values should replace. If there
1883 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1884 * are already known to be the same.
1886 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1887 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1888 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1889 origin_upper_left
= shader
->origin_upper_left
;
1890 pixel_center_integer
= shader
->pixel_center_integer
;
1893 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1894 shader
->EarlyFragmentTests
|| shader
->PostDepthCoverage
;
1895 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1896 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1897 shader
->PostDepthCoverage
;
1899 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1904 * Performs the cross-validation of geometry shader max_vertices and
1905 * primitive type layout qualifiers for the attached geometry shaders,
1906 * and propagates them to the linked GS and linked shader program.
1909 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1910 struct gl_program
*gl_prog
,
1911 struct gl_shader
**shader_list
,
1912 unsigned num_shaders
)
1914 /* No in/out qualifiers defined for anything but GLSL 1.50+
1915 * geometry shaders so far.
1917 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
1918 prog
->data
->Version
< 150)
1921 int vertices_out
= -1;
1923 gl_prog
->info
.gs
.invocations
= 0;
1924 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
1925 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
1927 /* From the GLSL 1.50 spec, page 46:
1929 * "All geometry shader output layout declarations in a program
1930 * must declare the same layout and same value for
1931 * max_vertices. There must be at least one geometry output
1932 * layout declaration somewhere in a program, but not all
1933 * geometry shaders (compilation units) are required to
1937 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1938 struct gl_shader
*shader
= shader_list
[i
];
1940 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1941 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
1942 gl_prog
->info
.gs
.input_primitive
!=
1943 shader
->info
.Geom
.InputType
) {
1944 linker_error(prog
, "geometry shader defined with conflicting "
1948 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
1951 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1952 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
1953 gl_prog
->info
.gs
.output_primitive
!=
1954 shader
->info
.Geom
.OutputType
) {
1955 linker_error(prog
, "geometry shader defined with conflicting "
1959 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
1962 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1963 if (vertices_out
!= -1 &&
1964 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
1965 linker_error(prog
, "geometry shader defined with conflicting "
1966 "output vertex count (%d and %d)\n",
1967 vertices_out
, shader
->info
.Geom
.VerticesOut
);
1970 vertices_out
= shader
->info
.Geom
.VerticesOut
;
1973 if (shader
->info
.Geom
.Invocations
!= 0) {
1974 if (gl_prog
->info
.gs
.invocations
!= 0 &&
1975 gl_prog
->info
.gs
.invocations
!=
1976 (unsigned) shader
->info
.Geom
.Invocations
) {
1977 linker_error(prog
, "geometry shader defined with conflicting "
1978 "invocation count (%d and %d)\n",
1979 gl_prog
->info
.gs
.invocations
,
1980 shader
->info
.Geom
.Invocations
);
1983 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
1987 /* Just do the intrastage -> interstage propagation right now,
1988 * since we already know we're in the right type of shader program
1991 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
1993 "geometry shader didn't declare primitive input type\n");
1997 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
1999 "geometry shader didn't declare primitive output type\n");
2003 if (vertices_out
== -1) {
2005 "geometry shader didn't declare max_vertices\n");
2008 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2011 if (gl_prog
->info
.gs
.invocations
== 0)
2012 gl_prog
->info
.gs
.invocations
= 1;
2017 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2018 * qualifiers for the attached compute shaders, and propagate them to the
2019 * linked CS and linked shader program.
2022 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2023 struct gl_program
*gl_prog
,
2024 struct gl_shader
**shader_list
,
2025 unsigned num_shaders
)
2027 /* This function is called for all shader stages, but it only has an effect
2028 * for compute shaders.
2030 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2033 for (int i
= 0; i
< 3; i
++)
2034 gl_prog
->info
.cs
.local_size
[i
] = 0;
2036 gl_prog
->info
.cs
.local_size_variable
= false;
2038 /* From the ARB_compute_shader spec, in the section describing local size
2041 * If multiple compute shaders attached to a single program object
2042 * declare local work-group size, the declarations must be identical;
2043 * otherwise a link-time error results. Furthermore, if a program
2044 * object contains any compute shaders, at least one must contain an
2045 * input layout qualifier specifying the local work sizes of the
2046 * program, or a link-time error will occur.
2048 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2049 struct gl_shader
*shader
= shader_list
[sh
];
2051 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2052 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2053 for (int i
= 0; i
< 3; i
++) {
2054 if (gl_prog
->info
.cs
.local_size
[i
] !=
2055 shader
->info
.Comp
.LocalSize
[i
]) {
2056 linker_error(prog
, "compute shader defined with conflicting "
2062 for (int i
= 0; i
< 3; i
++) {
2063 gl_prog
->info
.cs
.local_size
[i
] =
2064 shader
->info
.Comp
.LocalSize
[i
];
2066 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2067 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2068 /* The ARB_compute_variable_group_size spec says:
2070 * If one compute shader attached to a program declares a
2071 * variable local group size and a second compute shader
2072 * attached to the same program declares a fixed local group
2073 * size, a link-time error results.
2075 linker_error(prog
, "compute shader defined with both fixed and "
2076 "variable local group size\n");
2079 gl_prog
->info
.cs
.local_size_variable
= true;
2083 /* Just do the intrastage -> interstage propagation right now,
2084 * since we already know we're in the right type of shader program
2087 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2088 !gl_prog
->info
.cs
.local_size_variable
) {
2089 linker_error(prog
, "compute shader must contain a fixed or a variable "
2090 "local group size\n");
2097 * Combine a group of shaders for a single stage to generate a linked shader
2100 * If this function is supplied a single shader, it is cloned, and the new
2101 * shader is returned.
2103 struct gl_linked_shader
*
2104 link_intrastage_shaders(void *mem_ctx
,
2105 struct gl_context
*ctx
,
2106 struct gl_shader_program
*prog
,
2107 struct gl_shader
**shader_list
,
2108 unsigned num_shaders
,
2109 bool allow_missing_main
)
2111 struct gl_uniform_block
*ubo_blocks
= NULL
;
2112 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2113 unsigned num_ubo_blocks
= 0;
2114 unsigned num_ssbo_blocks
= 0;
2116 /* Check that global variables defined in multiple shaders are consistent.
2118 glsl_symbol_table variables
;
2119 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2120 if (shader_list
[i
] == NULL
)
2122 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2125 if (!prog
->data
->LinkStatus
)
2128 /* Check that interface blocks defined in multiple shaders are consistent.
2130 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2132 if (!prog
->data
->LinkStatus
)
2135 /* Check that there is only a single definition of each function signature
2136 * across all shaders.
2138 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2139 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2140 ir_function
*const f
= node
->as_function();
2145 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2146 ir_function
*const other
=
2147 shader_list
[j
]->symbols
->get_function(f
->name
);
2149 /* If the other shader has no function (and therefore no function
2150 * signatures) with the same name, skip to the next shader.
2155 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2156 if (!sig
->is_defined
)
2159 ir_function_signature
*other_sig
=
2160 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2162 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2163 linker_error(prog
, "function `%s' is multiply defined\n",
2172 /* Find the shader that defines main, and make a clone of it.
2174 * Starting with the clone, search for undefined references. If one is
2175 * found, find the shader that defines it. Clone the reference and add
2176 * it to the shader. Repeat until there are no undefined references or
2177 * until a reference cannot be resolved.
2179 gl_shader
*main
= NULL
;
2180 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2181 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2182 main
= shader_list
[i
];
2187 if (main
== NULL
&& allow_missing_main
)
2188 main
= shader_list
[0];
2191 linker_error(prog
, "%s shader lacks `main'\n",
2192 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2196 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2197 linked
->Stage
= shader_list
[0]->Stage
;
2199 /* Create program and attach it to the linked shader */
2200 struct gl_program
*gl_prog
=
2201 ctx
->Driver
.NewProgram(ctx
,
2202 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2205 prog
->data
->LinkStatus
= linking_failure
;
2206 _mesa_delete_linked_shader(ctx
, linked
);
2210 if (!prog
->data
->cache_fallback
)
2211 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2213 /* Don't use _mesa_reference_program() just take ownership */
2214 linked
->Program
= gl_prog
;
2216 linked
->ir
= new(linked
) exec_list
;
2217 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2219 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2220 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2221 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2222 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2223 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2224 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2227 populate_symbol_table(linked
);
2229 /* The pointer to the main function in the final linked shader (i.e., the
2230 * copy of the original shader that contained the main function).
2232 ir_function_signature
*const main_sig
=
2233 _mesa_get_main_function_signature(linked
->symbols
);
2235 /* Move any instructions other than variable declarations or function
2236 * declarations into main.
2238 if (main_sig
!= NULL
) {
2239 exec_node
*insertion_point
=
2240 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2243 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2244 if (shader_list
[i
] == main
)
2247 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2248 insertion_point
, true, linked
);
2252 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2253 _mesa_delete_linked_shader(ctx
, linked
);
2257 /* Make a pass over all variable declarations to ensure that arrays with
2258 * unspecified sizes have a size specified. The size is inferred from the
2259 * max_array_access field.
2261 array_sizing_visitor v
;
2263 v
.fixup_unnamed_interface_types();
2265 if (!prog
->data
->cache_fallback
) {
2266 /* Link up uniform blocks defined within this stage. */
2267 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2268 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2270 if (!prog
->data
->LinkStatus
) {
2271 _mesa_delete_linked_shader(ctx
, linked
);
2275 /* Copy ubo blocks to linked shader list */
2276 linked
->Program
->sh
.UniformBlocks
=
2277 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2278 ralloc_steal(linked
, ubo_blocks
);
2279 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2280 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2282 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2284 /* Copy ssbo blocks to linked shader list */
2285 linked
->Program
->sh
.ShaderStorageBlocks
=
2286 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2287 ralloc_steal(linked
, ssbo_blocks
);
2288 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2289 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2291 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2294 /* At this point linked should contain all of the linked IR, so
2295 * validate it to make sure nothing went wrong.
2297 validate_ir_tree(linked
->ir
);
2299 /* Set the size of geometry shader input arrays */
2300 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2301 unsigned num_vertices
=
2302 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2303 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2304 MESA_SHADER_GEOMETRY
);
2305 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2306 ir
->accept(&input_resize_visitor
);
2310 if (ctx
->Const
.VertexID_is_zero_based
)
2311 lower_vertex_id(linked
);
2314 /* Compute the source checksum. */
2315 linked
->SourceChecksum
= 0;
2316 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2317 if (shader_list
[i
] == NULL
)
2319 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2327 * Update the sizes of linked shader uniform arrays to the maximum
2330 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2332 * If one or more elements of an array are active,
2333 * GetActiveUniform will return the name of the array in name,
2334 * subject to the restrictions listed above. The type of the array
2335 * is returned in type. The size parameter contains the highest
2336 * array element index used, plus one. The compiler or linker
2337 * determines the highest index used. There will be only one
2338 * active uniform reported by the GL per uniform array.
2342 update_array_sizes(struct gl_shader_program
*prog
)
2344 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2345 if (prog
->_LinkedShaders
[i
] == NULL
)
2348 bool types_were_updated
= false;
2350 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2351 ir_variable
*const var
= node
->as_variable();
2353 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2354 !var
->type
->is_array())
2357 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2358 * will not be eliminated. Since we always do std140, just
2359 * don't resize arrays in UBOs.
2361 * Atomic counters are supposed to get deterministic
2362 * locations assigned based on the declaration ordering and
2363 * sizes, array compaction would mess that up.
2365 * Subroutine uniforms are not removed.
2367 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2368 var
->type
->contains_subroutine() || var
->constant_initializer
)
2371 int size
= var
->data
.max_array_access
;
2372 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2373 if (prog
->_LinkedShaders
[j
] == NULL
)
2376 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2377 ir_variable
*other_var
= node2
->as_variable();
2381 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2382 other_var
->data
.max_array_access
> size
) {
2383 size
= other_var
->data
.max_array_access
;
2388 if (size
+ 1 != (int)var
->type
->length
) {
2389 /* If this is a built-in uniform (i.e., it's backed by some
2390 * fixed-function state), adjust the number of state slots to
2391 * match the new array size. The number of slots per array entry
2392 * is not known. It seems safe to assume that the total number of
2393 * slots is an integer multiple of the number of array elements.
2394 * Determine the number of slots per array element by dividing by
2395 * the old (total) size.
2397 const unsigned num_slots
= var
->get_num_state_slots();
2398 if (num_slots
> 0) {
2399 var
->set_num_state_slots((size
+ 1)
2400 * (num_slots
/ var
->type
->length
));
2403 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2405 types_were_updated
= true;
2409 /* Update the types of dereferences in case we changed any. */
2410 if (types_were_updated
) {
2411 deref_type_updater v
;
2412 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2418 * Resize tessellation evaluation per-vertex inputs to the size of
2419 * tessellation control per-vertex outputs.
2422 resize_tes_inputs(struct gl_context
*ctx
,
2423 struct gl_shader_program
*prog
)
2425 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2428 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2429 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2431 /* If no control shader is present, then the TES inputs are statically
2432 * sized to MaxPatchVertices; the actual size of the arrays won't be
2433 * known until draw time.
2435 const int num_vertices
= tcs
2436 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2437 : ctx
->Const
.MaxPatchVertices
;
2439 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2440 MESA_SHADER_TESS_EVAL
);
2441 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2442 ir
->accept(&input_resize_visitor
);
2445 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2446 /* Convert the gl_PatchVerticesIn system value into a constant, since
2447 * the value is known at this point.
2449 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2450 ir_variable
*var
= ir
->as_variable();
2451 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2452 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2453 void *mem_ctx
= ralloc_parent(var
);
2454 var
->data
.location
= 0;
2455 var
->data
.explicit_location
= false;
2457 var
->data
.mode
= ir_var_auto
;
2458 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2460 var
->data
.mode
= ir_var_uniform
;
2461 var
->data
.how_declared
= ir_var_hidden
;
2462 var
->allocate_state_slots(1);
2463 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2464 slot0
->swizzle
= SWIZZLE_XXXX
;
2465 slot0
->tokens
[0] = STATE_INTERNAL
;
2466 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2467 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2468 slot0
->tokens
[i
] = 0;
2476 * Find a contiguous set of available bits in a bitmask.
2478 * \param used_mask Bits representing used (1) and unused (0) locations
2479 * \param needed_count Number of contiguous bits needed.
2482 * Base location of the available bits on success or -1 on failure.
2485 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2487 unsigned needed_mask
= (1 << needed_count
) - 1;
2488 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2490 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2491 * cannot optimize possibly infinite loops" for the loop below.
2493 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2496 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2497 if ((needed_mask
& ~used_mask
) == needed_mask
)
2508 * Assign locations for either VS inputs or FS outputs
2510 * \param mem_ctx Temporary ralloc context used for linking
2511 * \param prog Shader program whose variables need locations assigned
2512 * \param constants Driver specific constant values for the program.
2513 * \param target_index Selector for the program target to receive location
2514 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2515 * \c MESA_SHADER_FRAGMENT.
2518 * If locations are successfully assigned, true is returned. Otherwise an
2519 * error is emitted to the shader link log and false is returned.
2522 assign_attribute_or_color_locations(void *mem_ctx
,
2523 gl_shader_program
*prog
,
2524 struct gl_constants
*constants
,
2525 unsigned target_index
)
2527 /* Maximum number of generic locations. This corresponds to either the
2528 * maximum number of draw buffers or the maximum number of generic
2531 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2532 constants
->Program
[target_index
].MaxAttribs
:
2533 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2535 /* Mark invalid locations as being used.
2537 unsigned used_locations
= (max_index
>= 32)
2538 ? ~0 : ~((1 << max_index
) - 1);
2539 unsigned double_storage_locations
= 0;
2541 assert((target_index
== MESA_SHADER_VERTEX
)
2542 || (target_index
== MESA_SHADER_FRAGMENT
));
2544 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2548 /* Operate in a total of four passes.
2550 * 1. Invalidate the location assignments for all vertex shader inputs.
2552 * 2. Assign locations for inputs that have user-defined (via
2553 * glBindVertexAttribLocation) locations and outputs that have
2554 * user-defined locations (via glBindFragDataLocation).
2556 * 3. Sort the attributes without assigned locations by number of slots
2557 * required in decreasing order. Fragmentation caused by attribute
2558 * locations assigned by the application may prevent large attributes
2559 * from having enough contiguous space.
2561 * 4. Assign locations to any inputs without assigned locations.
2564 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2565 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2567 const enum ir_variable_mode direction
=
2568 (target_index
== MESA_SHADER_VERTEX
)
2569 ? ir_var_shader_in
: ir_var_shader_out
;
2572 /* Temporary storage for the set of attributes that need locations assigned.
2578 /* Used below in the call to qsort. */
2579 static int compare(const void *a
, const void *b
)
2581 const temp_attr
*const l
= (const temp_attr
*) a
;
2582 const temp_attr
*const r
= (const temp_attr
*) b
;
2584 /* Reversed because we want a descending order sort below. */
2585 return r
->slots
- l
->slots
;
2588 assert(max_index
<= 32);
2590 /* Temporary array for the set of attributes that have locations assigned.
2592 ir_variable
*assigned
[16];
2594 unsigned num_attr
= 0;
2595 unsigned assigned_attr
= 0;
2597 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2598 ir_variable
*const var
= node
->as_variable();
2600 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2603 if (var
->data
.explicit_location
) {
2604 var
->data
.is_unmatched_generic_inout
= 0;
2605 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2606 || (var
->data
.location
< 0)) {
2608 "invalid explicit location %d specified for `%s'\n",
2609 (var
->data
.location
< 0)
2610 ? var
->data
.location
2611 : var
->data
.location
- generic_base
,
2615 } else if (target_index
== MESA_SHADER_VERTEX
) {
2618 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2619 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2620 var
->data
.location
= binding
;
2621 var
->data
.is_unmatched_generic_inout
= 0;
2623 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2626 const char *name
= var
->name
;
2627 const glsl_type
*type
= var
->type
;
2630 /* Check if there's a binding for the variable name */
2631 if (prog
->FragDataBindings
->get(binding
, name
)) {
2632 assert(binding
>= FRAG_RESULT_DATA0
);
2633 var
->data
.location
= binding
;
2634 var
->data
.is_unmatched_generic_inout
= 0;
2636 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2637 var
->data
.index
= index
;
2642 /* If not, but it's an array type, look for name[0] */
2643 if (type
->is_array()) {
2644 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2645 type
= type
->fields
.array
;
2653 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2656 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2658 * "Output binding assignments will cause LinkProgram to fail:
2660 * If the program has an active output assigned to a location greater
2661 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2662 * an active output assigned an index greater than or equal to one;"
2664 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2665 var
->data
.location
- generic_base
>=
2666 (int) constants
->MaxDualSourceDrawBuffers
) {
2668 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2669 "with index %u for %s\n",
2670 var
->data
.location
- generic_base
, var
->data
.index
,
2675 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2677 /* If the variable is not a built-in and has a location statically
2678 * assigned in the shader (presumably via a layout qualifier), make sure
2679 * that it doesn't collide with other assigned locations. Otherwise,
2680 * add it to the list of variables that need linker-assigned locations.
2682 if (var
->data
.location
!= -1) {
2683 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2684 /* From page 61 of the OpenGL 4.0 spec:
2686 * "LinkProgram will fail if the attribute bindings assigned
2687 * by BindAttribLocation do not leave not enough space to
2688 * assign a location for an active matrix attribute or an
2689 * active attribute array, both of which require multiple
2690 * contiguous generic attributes."
2692 * I think above text prohibits the aliasing of explicit and
2693 * automatic assignments. But, aliasing is allowed in manual
2694 * assignments of attribute locations. See below comments for
2697 * From OpenGL 4.0 spec, page 61:
2699 * "It is possible for an application to bind more than one
2700 * attribute name to the same location. This is referred to as
2701 * aliasing. This will only work if only one of the aliased
2702 * attributes is active in the executable program, or if no
2703 * path through the shader consumes more than one attribute of
2704 * a set of attributes aliased to the same location. A link
2705 * error can occur if the linker determines that every path
2706 * through the shader consumes multiple aliased attributes,
2707 * but implementations are not required to generate an error
2710 * From GLSL 4.30 spec, page 54:
2712 * "A program will fail to link if any two non-vertex shader
2713 * input variables are assigned to the same location. For
2714 * vertex shaders, multiple input variables may be assigned
2715 * to the same location using either layout qualifiers or via
2716 * the OpenGL API. However, such aliasing is intended only to
2717 * support vertex shaders where each execution path accesses
2718 * at most one input per each location. Implementations are
2719 * permitted, but not required, to generate link-time errors
2720 * if they detect that every path through the vertex shader
2721 * executable accesses multiple inputs assigned to any single
2722 * location. For all shader types, a program will fail to link
2723 * if explicit location assignments leave the linker unable
2724 * to find space for other variables without explicit
2727 * From OpenGL ES 3.0 spec, page 56:
2729 * "Binding more than one attribute name to the same location
2730 * is referred to as aliasing, and is not permitted in OpenGL
2731 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2732 * fail when this condition exists. However, aliasing is
2733 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2734 * This will only work if only one of the aliased attributes
2735 * is active in the executable program, or if no path through
2736 * the shader consumes more than one attribute of a set of
2737 * attributes aliased to the same location. A link error can
2738 * occur if the linker determines that every path through the
2739 * shader consumes multiple aliased attributes, but implemen-
2740 * tations are not required to generate an error in this case."
2742 * After looking at above references from OpenGL, OpenGL ES and
2743 * GLSL specifications, we allow aliasing of vertex input variables
2744 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2746 * NOTE: This is not required by the spec but its worth mentioning
2747 * here that we're not doing anything to make sure that no path
2748 * through the vertex shader executable accesses multiple inputs
2749 * assigned to any single location.
2752 /* Mask representing the contiguous slots that will be used by
2755 const unsigned attr
= var
->data
.location
- generic_base
;
2756 const unsigned use_mask
= (1 << slots
) - 1;
2757 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2758 ? "vertex shader input" : "fragment shader output";
2760 /* Generate a link error if the requested locations for this
2761 * attribute exceed the maximum allowed attribute location.
2763 if (attr
+ slots
> max_index
) {
2765 "insufficient contiguous locations "
2766 "available for %s `%s' %d %d %d\n", string
,
2767 var
->name
, used_locations
, use_mask
, attr
);
2771 /* Generate a link error if the set of bits requested for this
2772 * attribute overlaps any previously allocated bits.
2774 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2775 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2776 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2779 * "Additionally, for fragment shader outputs, if two
2780 * variables are placed within the same location, they
2781 * must have the same underlying type (floating-point or
2782 * integer). No component aliasing of output variables or
2783 * members is allowed.
2785 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2786 unsigned assigned_slots
=
2787 assigned
[i
]->type
->count_attribute_slots(false);
2788 unsigned assig_attr
=
2789 assigned
[i
]->data
.location
- generic_base
;
2790 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2792 if ((assigned_use_mask
<< assig_attr
) &
2793 (use_mask
<< attr
)) {
2795 const glsl_type
*assigned_type
=
2796 assigned
[i
]->type
->without_array();
2797 const glsl_type
*type
= var
->type
->without_array();
2798 if (assigned_type
->base_type
!= type
->base_type
) {
2799 linker_error(prog
, "types do not match for aliased"
2800 " %ss %s and %s\n", string
,
2801 assigned
[i
]->name
, var
->name
);
2805 unsigned assigned_component_mask
=
2806 ((1 << assigned_type
->vector_elements
) - 1) <<
2807 assigned
[i
]->data
.location_frac
;
2808 unsigned component_mask
=
2809 ((1 << type
->vector_elements
) - 1) <<
2810 var
->data
.location_frac
;
2811 if (assigned_component_mask
& component_mask
) {
2812 linker_error(prog
, "overlapping component is "
2813 "assigned to %ss %s and %s "
2815 string
, assigned
[i
]->name
, var
->name
,
2816 var
->data
.location_frac
);
2821 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2822 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2823 linker_error(prog
, "overlapping location is assigned "
2824 "to %s `%s' %d %d %d\n", string
, var
->name
,
2825 used_locations
, use_mask
, attr
);
2828 linker_warning(prog
, "overlapping location is assigned "
2829 "to %s `%s' %d %d %d\n", string
, var
->name
,
2830 used_locations
, use_mask
, attr
);
2834 used_locations
|= (use_mask
<< attr
);
2836 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2838 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2839 * active attribute variables may fail to link, unless
2840 * device-dependent optimizations are able to make the program
2841 * fit within available hardware resources. For the purposes
2842 * of this test, attribute variables of the type dvec3, dvec4,
2843 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2844 * count as consuming twice as many attributes as equivalent
2845 * single-precision types. While these types use the same number
2846 * of generic attributes as their single-precision equivalents,
2847 * implementations are permitted to consume two single-precision
2848 * vectors of internal storage for each three- or four-component
2849 * double-precision vector."
2851 * Mark this attribute slot as taking up twice as much space
2852 * so we can count it properly against limits. According to
2853 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2854 * is optional behavior, but it seems preferable.
2856 if (var
->type
->without_array()->is_dual_slot())
2857 double_storage_locations
|= (use_mask
<< attr
);
2860 assigned
[assigned_attr
] = var
;
2866 if (num_attr
>= max_index
) {
2867 linker_error(prog
, "too many %s (max %u)",
2868 target_index
== MESA_SHADER_VERTEX
?
2869 "vertex shader inputs" : "fragment shader outputs",
2873 to_assign
[num_attr
].slots
= slots
;
2874 to_assign
[num_attr
].var
= var
;
2878 if (target_index
== MESA_SHADER_VERTEX
) {
2879 unsigned total_attribs_size
=
2880 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2881 _mesa_bitcount(double_storage_locations
);
2882 if (total_attribs_size
> max_index
) {
2884 "attempt to use %d vertex attribute slots only %d available ",
2885 total_attribs_size
, max_index
);
2890 /* If all of the attributes were assigned locations by the application (or
2891 * are built-in attributes with fixed locations), return early. This should
2892 * be the common case.
2897 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2899 if (target_index
== MESA_SHADER_VERTEX
) {
2900 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2901 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2902 * reserved to prevent it from being automatically allocated below.
2904 find_deref_visitor
find("gl_Vertex");
2906 if (find
.variable_found())
2907 used_locations
|= (1 << 0);
2910 for (unsigned i
= 0; i
< num_attr
; i
++) {
2911 /* Mask representing the contiguous slots that will be used by this
2914 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2916 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2919 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2920 ? "vertex shader input" : "fragment shader output";
2923 "insufficient contiguous locations "
2924 "available for %s `%s'\n",
2925 string
, to_assign
[i
].var
->name
);
2929 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2930 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2931 used_locations
|= (use_mask
<< location
);
2933 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2934 double_storage_locations
|= (use_mask
<< location
);
2937 /* Now that we have all the locations, from the GL 4.5 core spec, section
2938 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2939 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2940 * as equivalent single-precision types.
2942 if (target_index
== MESA_SHADER_VERTEX
) {
2943 unsigned total_attribs_size
=
2944 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2945 _mesa_bitcount(double_storage_locations
);
2946 if (total_attribs_size
> max_index
) {
2948 "attempt to use %d vertex attribute slots only %d available ",
2949 total_attribs_size
, max_index
);
2958 * Match explicit locations of outputs to inputs and deactivate the
2959 * unmatch flag if found so we don't optimise them away.
2962 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2963 gl_linked_shader
*consumer
)
2965 glsl_symbol_table parameters
;
2966 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2969 /* Find all shader outputs in the "producer" stage.
2971 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2972 ir_variable
*const var
= node
->as_variable();
2974 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2977 if (var
->data
.explicit_location
&&
2978 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2979 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2980 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2981 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2985 /* Match inputs to outputs */
2986 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2987 ir_variable
*const input
= node
->as_variable();
2989 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2992 ir_variable
*output
= NULL
;
2993 if (input
->data
.explicit_location
2994 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2995 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2996 [input
->data
.location_frac
];
2998 if (output
!= NULL
){
2999 input
->data
.is_unmatched_generic_inout
= 0;
3000 output
->data
.is_unmatched_generic_inout
= 0;
3007 * Store the gl_FragDepth layout in the gl_shader_program struct.
3010 store_fragdepth_layout(struct gl_shader_program
*prog
)
3012 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3016 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3018 /* We don't look up the gl_FragDepth symbol directly because if
3019 * gl_FragDepth is not used in the shader, it's removed from the IR.
3020 * However, the symbol won't be removed from the symbol table.
3022 * We're only interested in the cases where the variable is NOT removed
3025 foreach_in_list(ir_instruction
, node
, ir
) {
3026 ir_variable
*const var
= node
->as_variable();
3028 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3032 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3033 switch (var
->data
.depth_layout
) {
3034 case ir_depth_layout_none
:
3035 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3037 case ir_depth_layout_any
:
3038 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3040 case ir_depth_layout_greater
:
3041 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3043 case ir_depth_layout_less
:
3044 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3046 case ir_depth_layout_unchanged
:
3047 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3058 * Validate the resources used by a program versus the implementation limits
3061 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3063 unsigned total_uniform_blocks
= 0;
3064 unsigned total_shader_storage_blocks
= 0;
3066 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3067 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3072 if (sh
->Program
->info
.num_textures
>
3073 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3074 linker_error(prog
, "Too many %s shader texture samplers\n",
3075 _mesa_shader_stage_to_string(i
));
3078 if (sh
->num_uniform_components
>
3079 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3080 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3081 linker_warning(prog
, "Too many %s shader default uniform block "
3082 "components, but the driver will try to optimize "
3083 "them out; this is non-portable out-of-spec "
3085 _mesa_shader_stage_to_string(i
));
3087 linker_error(prog
, "Too many %s shader default uniform block "
3089 _mesa_shader_stage_to_string(i
));
3093 if (sh
->num_combined_uniform_components
>
3094 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3095 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3096 linker_warning(prog
, "Too many %s shader uniform components, "
3097 "but the driver will try to optimize them out; "
3098 "this is non-portable out-of-spec behavior\n",
3099 _mesa_shader_stage_to_string(i
));
3101 linker_error(prog
, "Too many %s shader uniform components\n",
3102 _mesa_shader_stage_to_string(i
));
3106 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3107 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3109 const unsigned max_uniform_blocks
=
3110 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3111 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3112 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3113 _mesa_shader_stage_to_string(i
),
3114 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3117 const unsigned max_shader_storage_blocks
=
3118 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3119 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3120 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3121 _mesa_shader_stage_to_string(i
),
3122 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3126 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3127 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3128 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3131 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3132 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3133 total_shader_storage_blocks
,
3134 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3137 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3138 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3139 ctx
->Const
.MaxUniformBlockSize
) {
3140 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3141 prog
->data
->UniformBlocks
[i
].Name
,
3142 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3143 ctx
->Const
.MaxUniformBlockSize
);
3147 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3148 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3149 ctx
->Const
.MaxShaderStorageBlockSize
) {
3150 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3151 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3152 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3153 ctx
->Const
.MaxShaderStorageBlockSize
);
3159 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3161 unsigned mask
= prog
->data
->linked_stages
;
3163 const int i
= u_bit_scan(&mask
);
3164 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3166 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3167 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3170 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3176 if (p
->sh
.NumSubroutineFunctions
== 0) {
3177 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3180 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3181 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3182 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3183 if (fn
->types
[k
] == uni
->type
) {
3189 uni
->num_compatible_subroutines
= count
;
3195 check_subroutine_resources(struct gl_shader_program
*prog
)
3197 unsigned mask
= prog
->data
->linked_stages
;
3199 const int i
= u_bit_scan(&mask
);
3200 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3202 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3203 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3204 _mesa_shader_stage_to_string(i
));
3209 * Validate shader image resources.
3212 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3214 unsigned total_image_units
= 0;
3215 unsigned fragment_outputs
= 0;
3216 unsigned total_shader_storage_blocks
= 0;
3218 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3221 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3222 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3225 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3226 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3227 _mesa_shader_stage_to_string(i
),
3228 sh
->Program
->info
.num_images
,
3229 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3231 total_image_units
+= sh
->Program
->info
.num_images
;
3232 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3234 if (i
== MESA_SHADER_FRAGMENT
) {
3235 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3236 ir_variable
*var
= node
->as_variable();
3237 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3238 /* since there are no double fs outputs - pass false */
3239 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3245 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3246 linker_error(prog
, "Too many combined image uniforms\n");
3248 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3249 ctx
->Const
.MaxCombinedShaderOutputResources
)
3250 linker_error(prog
, "Too many combined image uniforms, shader storage "
3251 " buffers and fragment outputs\n");
3256 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3257 * for a variable, checks for overlaps between other uniforms using explicit
3261 reserve_explicit_locations(struct gl_shader_program
*prog
,
3262 string_to_uint_map
*map
, ir_variable
*var
)
3264 unsigned slots
= var
->type
->uniform_locations();
3265 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3266 unsigned return_value
= slots
;
3268 /* Resize remap table if locations do not fit in the current one. */
3269 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3270 prog
->UniformRemapTable
=
3271 reralloc(prog
, prog
->UniformRemapTable
,
3272 gl_uniform_storage
*,
3275 if (!prog
->UniformRemapTable
) {
3276 linker_error(prog
, "Out of memory during linking.\n");
3280 /* Initialize allocated space. */
3281 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3282 prog
->UniformRemapTable
[i
] = NULL
;
3284 prog
->NumUniformRemapTable
= max_loc
+ 1;
3287 for (unsigned i
= 0; i
< slots
; i
++) {
3288 unsigned loc
= var
->data
.location
+ i
;
3290 /* Check if location is already used. */
3291 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3293 /* Possibly same uniform from a different stage, this is ok. */
3295 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3300 /* ARB_explicit_uniform_location specification states:
3302 * "No two default-block uniform variables in the program can have
3303 * the same location, even if they are unused, otherwise a compiler
3304 * or linker error will be generated."
3307 "location qualifier for uniform %s overlaps "
3308 "previously used location\n",
3313 /* Initialize location as inactive before optimization
3314 * rounds and location assignment.
3316 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3319 /* Note, base location used for arrays. */
3320 map
->put(var
->data
.location
, var
->name
);
3322 return return_value
;
3326 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3327 struct gl_program
*p
,
3330 unsigned slots
= var
->type
->uniform_locations();
3331 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3333 /* Resize remap table if locations do not fit in the current one. */
3334 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3335 p
->sh
.SubroutineUniformRemapTable
=
3336 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3337 gl_uniform_storage
*,
3340 if (!p
->sh
.SubroutineUniformRemapTable
) {
3341 linker_error(prog
, "Out of memory during linking.\n");
3345 /* Initialize allocated space. */
3346 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3347 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3349 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3352 for (unsigned i
= 0; i
< slots
; i
++) {
3353 unsigned loc
= var
->data
.location
+ i
;
3355 /* Check if location is already used. */
3356 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3358 /* ARB_explicit_uniform_location specification states:
3359 * "No two subroutine uniform variables can have the same location
3360 * in the same shader stage, otherwise a compiler or linker error
3361 * will be generated."
3364 "location qualifier for uniform %s overlaps "
3365 "previously used location\n",
3370 /* Initialize location as inactive before optimization
3371 * rounds and location assignment.
3373 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3379 * Check and reserve all explicit uniform locations, called before
3380 * any optimizations happen to handle also inactive uniforms and
3381 * inactive array elements that may get trimmed away.
3384 check_explicit_uniform_locations(struct gl_context
*ctx
,
3385 struct gl_shader_program
*prog
)
3387 prog
->NumExplicitUniformLocations
= 0;
3389 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3392 /* This map is used to detect if overlapping explicit locations
3393 * occur with the same uniform (from different stage) or a different one.
3395 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3398 linker_error(prog
, "Out of memory during linking.\n");
3402 unsigned entries_total
= 0;
3403 unsigned mask
= prog
->data
->linked_stages
;
3405 const int i
= u_bit_scan(&mask
);
3406 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3408 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3409 ir_variable
*var
= node
->as_variable();
3410 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3413 if (var
->data
.explicit_location
) {
3415 if (var
->type
->without_array()->is_subroutine())
3416 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3418 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3422 entries_total
+= slots
;
3433 struct empty_uniform_block
*current_block
= NULL
;
3435 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3436 /* We found empty space in UniformRemapTable. */
3437 if (prog
->UniformRemapTable
[i
] == NULL
) {
3438 /* We've found the beginning of a new continous block of empty slots */
3439 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3440 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3441 current_block
->start
= i
;
3442 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3443 ¤t_block
->link
);
3446 /* The current block continues, so we simply increment its slots */
3447 current_block
->slots
++;
3452 prog
->NumExplicitUniformLocations
= entries_total
;
3456 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3457 GLenum type
, const char *name
)
3459 bool found_interface
= false;
3460 unsigned block_name_len
= 0;
3461 const char *block_name_dot
= strchr(name
, '.');
3463 /* These rules only apply to buffer variables. So we return
3464 * true for the rest of types.
3466 if (type
!= GL_BUFFER_VARIABLE
)
3469 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3470 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3471 block_name_len
= strlen(block_name
);
3473 const char *block_square_bracket
= strchr(block_name
, '[');
3474 if (block_square_bracket
) {
3475 /* The block is part of an array of named interfaces,
3476 * for the name comparison we ignore the "[x]" part.
3478 block_name_len
-= strlen(block_square_bracket
);
3481 if (block_name_dot
) {
3482 /* Check if the variable name starts with the interface
3483 * name. The interface name (if present) should have the
3484 * length than the interface block name we are comparing to.
3486 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3487 if (len
!= block_name_len
)
3491 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3492 found_interface
= true;
3497 /* We remove the interface name from the buffer variable name,
3498 * including the dot that follows it.
3500 if (found_interface
)
3501 name
= name
+ block_name_len
+ 1;
3503 /* The ARB_program_interface_query spec says:
3505 * "For an active shader storage block member declared as an array, an
3506 * entry will be generated only for the first array element, regardless
3507 * of its type. For arrays of aggregate types, the enumeration rules
3508 * are applied recursively for the single enumerated array element."
3510 const char *struct_first_dot
= strchr(name
, '.');
3511 const char *first_square_bracket
= strchr(name
, '[');
3513 /* The buffer variable is on top level and it is not an array */
3514 if (!first_square_bracket
) {
3516 /* The shader storage block member is a struct, then generate the entry */
3517 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3520 /* Shader storage block member is an array, only generate an entry for the
3521 * first array element.
3523 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3531 add_program_resource(struct gl_shader_program
*prog
,
3532 struct set
*resource_set
,
3533 GLenum type
, const void *data
, uint8_t stages
)
3537 /* If resource already exists, do not add it again. */
3538 if (_mesa_set_search(resource_set
, data
))
3541 prog
->data
->ProgramResourceList
=
3543 prog
->data
->ProgramResourceList
,
3544 gl_program_resource
,
3545 prog
->data
->NumProgramResourceList
+ 1);
3547 if (!prog
->data
->ProgramResourceList
) {
3548 linker_error(prog
, "Out of memory during linking.\n");
3552 struct gl_program_resource
*res
=
3553 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3557 res
->StageReferences
= stages
;
3559 prog
->data
->NumProgramResourceList
++;
3561 _mesa_set_add(resource_set
, data
);
3566 /* Function checks if a variable var is a packed varying and
3567 * if given name is part of packed varying's list.
3569 * If a variable is a packed varying, it has a name like
3570 * 'packed:a,b,c' where a, b and c are separate variables.
3573 included_in_packed_varying(ir_variable
*var
, const char *name
)
3575 if (strncmp(var
->name
, "packed:", 7) != 0)
3578 char *list
= strdup(var
->name
+ 7);
3583 char *token
= strtok_r(list
, ",", &saveptr
);
3585 if (strcmp(token
, name
) == 0) {
3589 token
= strtok_r(NULL
, ",", &saveptr
);
3596 * Function builds a stage reference bitmask from variable name.
3599 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3604 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3605 * used for reference mask in gl_program_resource will need to be changed.
3607 assert(MESA_SHADER_STAGES
< 8);
3609 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3610 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3614 /* Shader symbol table may contain variables that have
3615 * been optimized away. Search IR for the variable instead.
3617 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3618 ir_variable
*var
= node
->as_variable();
3620 unsigned baselen
= strlen(var
->name
);
3622 if (included_in_packed_varying(var
, name
)) {
3627 /* Type needs to match if specified, otherwise we might
3628 * pick a variable with same name but different interface.
3630 if (var
->data
.mode
!= mode
)
3633 if (strncmp(var
->name
, name
, baselen
) == 0) {
3634 /* Check for exact name matches but also check for arrays and
3637 if (name
[baselen
] == '\0' ||
3638 name
[baselen
] == '[' ||
3639 name
[baselen
] == '.') {
3651 * Create gl_shader_variable from ir_variable class.
3653 static gl_shader_variable
*
3654 create_shader_variable(struct gl_shader_program
*shProg
,
3655 const ir_variable
*in
,
3656 const char *name
, const glsl_type
*type
,
3657 const glsl_type
*interface_type
,
3658 bool use_implicit_location
, int location
,
3659 const glsl_type
*outermost_struct_type
)
3661 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3665 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3666 * expect to see gl_VertexID in the program resource list. Pretend.
3668 if (in
->data
.mode
== ir_var_system_value
&&
3669 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3670 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3671 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3672 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3673 (in
->data
.mode
== ir_var_system_value
&&
3674 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3675 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3676 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3677 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3678 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3679 (in
->data
.mode
== ir_var_system_value
&&
3680 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3681 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3682 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3684 out
->name
= ralloc_strdup(shProg
, name
);
3690 /* The ARB_program_interface_query spec says:
3692 * "Not all active variables are assigned valid locations; the
3693 * following variables will have an effective location of -1:
3695 * * uniforms declared as atomic counters;
3697 * * members of a uniform block;
3699 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3701 * * inputs or outputs not declared with a "location" layout
3702 * qualifier, except for vertex shader inputs and fragment shader
3705 if (in
->type
->is_atomic_uint() || is_gl_identifier(in
->name
) ||
3706 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3709 out
->location
= location
;
3713 out
->outermost_struct_type
= outermost_struct_type
;
3714 out
->interface_type
= interface_type
;
3715 out
->component
= in
->data
.location_frac
;
3716 out
->index
= in
->data
.index
;
3717 out
->patch
= in
->data
.patch
;
3718 out
->mode
= in
->data
.mode
;
3719 out
->interpolation
= in
->data
.interpolation
;
3720 out
->explicit_location
= in
->data
.explicit_location
;
3721 out
->precision
= in
->data
.precision
;
3727 add_shader_variable(const struct gl_context
*ctx
,
3728 struct gl_shader_program
*shProg
,
3729 struct set
*resource_set
,
3730 unsigned stage_mask
,
3731 GLenum programInterface
, ir_variable
*var
,
3732 const char *name
, const glsl_type
*type
,
3733 bool use_implicit_location
, int location
,
3734 const glsl_type
*outermost_struct_type
= NULL
)
3736 const glsl_type
*interface_type
= var
->get_interface_type();
3738 if (outermost_struct_type
== NULL
) {
3739 if (var
->data
.from_named_ifc_block
) {
3740 const char *interface_name
= interface_type
->name
;
3742 if (interface_type
->is_array()) {
3743 /* Issue #16 of the ARB_program_interface_query spec says:
3745 * "* If a variable is a member of an interface block without an
3746 * instance name, it is enumerated using just the variable name.
3748 * * If a variable is a member of an interface block with an
3749 * instance name, it is enumerated as "BlockName.Member", where
3750 * "BlockName" is the name of the interface block (not the
3751 * instance name) and "Member" is the name of the variable."
3753 * In particular, it indicates that it should be "BlockName",
3754 * not "BlockName[array length]". The conformance suite and
3755 * dEQP both require this behavior.
3757 * Here, we unwrap the extra array level added by named interface
3758 * block array lowering so we have the correct variable type. We
3759 * also unwrap the interface type when constructing the name.
3761 * We leave interface_type the same so that ES 3.x SSO pipeline
3762 * validation can enforce the rules requiring array length to
3763 * match on interface blocks.
3765 type
= type
->fields
.array
;
3767 interface_name
= interface_type
->fields
.array
->name
;
3770 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3774 switch (type
->base_type
) {
3775 case GLSL_TYPE_STRUCT
: {
3776 /* The ARB_program_interface_query spec says:
3778 * "For an active variable declared as a structure, a separate entry
3779 * will be generated for each active structure member. The name of
3780 * each entry is formed by concatenating the name of the structure,
3781 * the "." character, and the name of the structure member. If a
3782 * structure member to enumerate is itself a structure or array,
3783 * these enumeration rules are applied recursively."
3785 if (outermost_struct_type
== NULL
)
3786 outermost_struct_type
= type
;
3788 unsigned field_location
= location
;
3789 for (unsigned i
= 0; i
< type
->length
; i
++) {
3790 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3791 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3792 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3793 stage_mask
, programInterface
,
3794 var
, field_name
, field
->type
,
3795 use_implicit_location
, field_location
,
3796 outermost_struct_type
))
3799 field_location
+= field
->type
->count_attribute_slots(false);
3805 /* The ARB_program_interface_query spec says:
3807 * "For an active variable declared as a single instance of a basic
3808 * type, a single entry will be generated, using the variable name
3809 * from the shader source."
3811 gl_shader_variable
*sha_v
=
3812 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3813 use_implicit_location
, location
,
3814 outermost_struct_type
);
3818 return add_program_resource(shProg
, resource_set
,
3819 programInterface
, sha_v
, stage_mask
);
3825 add_interface_variables(const struct gl_context
*ctx
,
3826 struct gl_shader_program
*shProg
,
3827 struct set
*resource_set
,
3828 unsigned stage
, GLenum programInterface
)
3830 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3832 foreach_in_list(ir_instruction
, node
, ir
) {
3833 ir_variable
*var
= node
->as_variable();
3835 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3840 switch (var
->data
.mode
) {
3841 case ir_var_system_value
:
3842 case ir_var_shader_in
:
3843 if (programInterface
!= GL_PROGRAM_INPUT
)
3845 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3846 : int(VARYING_SLOT_VAR0
);
3848 case ir_var_shader_out
:
3849 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3851 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3852 : int(VARYING_SLOT_VAR0
);
3858 if (var
->data
.patch
)
3859 loc_bias
= int(VARYING_SLOT_PATCH0
);
3861 /* Skip packed varyings, packed varyings are handled separately
3862 * by add_packed_varyings.
3864 if (strncmp(var
->name
, "packed:", 7) == 0)
3867 /* Skip fragdata arrays, these are handled separately
3868 * by add_fragdata_arrays.
3870 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3873 const bool vs_input_or_fs_output
=
3874 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3875 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3877 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3878 1 << stage
, programInterface
,
3879 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3880 var
->data
.location
- loc_bias
))
3887 add_packed_varyings(const struct gl_context
*ctx
,
3888 struct gl_shader_program
*shProg
,
3889 struct set
*resource_set
,
3890 int stage
, GLenum type
)
3892 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3895 if (!sh
|| !sh
->packed_varyings
)
3898 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3899 ir_variable
*var
= node
->as_variable();
3901 switch (var
->data
.mode
) {
3902 case ir_var_shader_in
:
3903 iface
= GL_PROGRAM_INPUT
;
3905 case ir_var_shader_out
:
3906 iface
= GL_PROGRAM_OUTPUT
;
3909 unreachable("unexpected type");
3912 if (type
== iface
) {
3913 const int stage_mask
=
3914 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3915 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3917 iface
, var
, var
->name
, var
->type
, false,
3918 var
->data
.location
- VARYING_SLOT_VAR0
))
3927 add_fragdata_arrays(const struct gl_context
*ctx
,
3928 struct gl_shader_program
*shProg
,
3929 struct set
*resource_set
)
3931 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3933 if (!sh
|| !sh
->fragdata_arrays
)
3936 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3937 ir_variable
*var
= node
->as_variable();
3939 assert(var
->data
.mode
== ir_var_shader_out
);
3941 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3942 1 << MESA_SHADER_FRAGMENT
,
3943 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3944 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3952 get_top_level_name(const char *name
)
3954 const char *first_dot
= strchr(name
, '.');
3955 const char *first_square_bracket
= strchr(name
, '[');
3958 /* The ARB_program_interface_query spec says:
3960 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3961 * the number of active array elements of the top-level shader storage
3962 * block member containing to the active variable is written to
3963 * <params>. If the top-level block member is not declared as an
3964 * array, the value one is written to <params>. If the top-level block
3965 * member is an array with no declared size, the value zero is written
3969 /* The buffer variable is on top level.*/
3970 if (!first_square_bracket
&& !first_dot
)
3971 name_size
= strlen(name
);
3972 else if ((!first_square_bracket
||
3973 (first_dot
&& first_dot
< first_square_bracket
)))
3974 name_size
= first_dot
- name
;
3976 name_size
= first_square_bracket
- name
;
3978 return strndup(name
, name_size
);
3982 get_var_name(const char *name
)
3984 const char *first_dot
= strchr(name
, '.');
3987 return strdup(name
);
3989 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3993 is_top_level_shader_storage_block_member(const char* name
,
3994 const char* interface_name
,
3995 const char* field_name
)
3997 bool result
= false;
3999 /* If the given variable is already a top-level shader storage
4000 * block member, then return array_size = 1.
4001 * We could have two possibilities: if we have an instanced
4002 * shader storage block or not instanced.
4004 * For the first, we check create a name as it was in top level and
4005 * compare it with the real name. If they are the same, then
4006 * the variable is already at top-level.
4008 * Full instanced name is: interface name + '.' + var name +
4011 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4012 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4013 if (!full_instanced_name
) {
4014 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4018 snprintf(full_instanced_name
, name_length
, "%s.%s",
4019 interface_name
, field_name
);
4021 /* Check if its top-level shader storage block member of an
4022 * instanced interface block, or of a unnamed interface block.
4024 if (strcmp(name
, full_instanced_name
) == 0 ||
4025 strcmp(name
, field_name
) == 0)
4028 free(full_instanced_name
);
4033 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4034 char *interface_name
, char *var_name
)
4036 /* The ARB_program_interface_query spec says:
4038 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4039 * the number of active array elements of the top-level shader storage
4040 * block member containing to the active variable is written to
4041 * <params>. If the top-level block member is not declared as an
4042 * array, the value one is written to <params>. If the top-level block
4043 * member is an array with no declared size, the value zero is written
4046 if (is_top_level_shader_storage_block_member(uni
->name
,
4050 else if (field
->type
->is_unsized_array())
4052 else if (field
->type
->is_array())
4053 return field
->type
->length
;
4059 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4060 const glsl_struct_field
*field
, char *interface_name
,
4063 /* The ARB_program_interface_query spec says:
4065 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4066 * identifying the stride between array elements of the top-level
4067 * shader storage block member containing the active variable is
4068 * written to <params>. For top-level block members declared as
4069 * arrays, the value written is the difference, in basic machine units,
4070 * between the offsets of the active variable for consecutive elements
4071 * in the top-level array. For top-level block members not declared as
4072 * an array, zero is written to <params>."
4074 if (field
->type
->is_array()) {
4075 const enum glsl_matrix_layout matrix_layout
=
4076 glsl_matrix_layout(field
->matrix_layout
);
4077 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4078 const glsl_type
*array_type
= field
->type
->fields
.array
;
4080 if (is_top_level_shader_storage_block_member(uni
->name
,
4085 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4086 if (array_type
->is_record() || array_type
->is_array())
4087 return glsl_align(array_type
->std140_size(row_major
), 16);
4089 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4091 return array_type
->std430_array_stride(row_major
);
4098 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4099 struct gl_uniform_storage
*uni
)
4101 int block_index
= uni
->block_index
;
4102 int array_size
= -1;
4103 int array_stride
= -1;
4104 char *var_name
= get_top_level_name(uni
->name
);
4105 char *interface_name
=
4106 get_top_level_name(uni
->is_shader_storage
?
4107 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4108 shProg
->data
->UniformBlocks
[block_index
].Name
);
4110 if (strcmp(var_name
, interface_name
) == 0) {
4111 /* Deal with instanced array of SSBOs */
4112 char *temp_name
= get_var_name(uni
->name
);
4114 linker_error(shProg
, "Out of memory during linking.\n");
4115 goto write_top_level_array_size_and_stride
;
4118 var_name
= get_top_level_name(temp_name
);
4121 linker_error(shProg
, "Out of memory during linking.\n");
4122 goto write_top_level_array_size_and_stride
;
4126 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4127 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4131 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4132 ir_variable
*var
= node
->as_variable();
4133 if (!var
|| !var
->get_interface_type() ||
4134 var
->data
.mode
!= ir_var_shader_storage
)
4137 const glsl_type
*interface
= var
->get_interface_type();
4139 if (strcmp(interface_name
, interface
->name
) != 0)
4142 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4143 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4144 if (strcmp(field
->name
, var_name
) != 0)
4147 array_stride
= get_array_stride(uni
, interface
, field
,
4148 interface_name
, var_name
);
4149 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4150 goto write_top_level_array_size_and_stride
;
4154 write_top_level_array_size_and_stride
:
4155 free(interface_name
);
4157 uni
->top_level_array_stride
= array_stride
;
4158 uni
->top_level_array_size
= array_size
;
4162 * Builds up a list of program resources that point to existing
4166 build_program_resource_list(struct gl_context
*ctx
,
4167 struct gl_shader_program
*shProg
)
4169 /* Rebuild resource list. */
4170 if (shProg
->data
->ProgramResourceList
) {
4171 ralloc_free(shProg
->data
->ProgramResourceList
);
4172 shProg
->data
->ProgramResourceList
= NULL
;
4173 shProg
->data
->NumProgramResourceList
= 0;
4176 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4178 /* Determine first input and final output stage. These are used to
4179 * detect which variables should be enumerated in the resource list
4180 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4182 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4183 if (!shProg
->_LinkedShaders
[i
])
4185 if (input_stage
== MESA_SHADER_STAGES
)
4190 /* Empty shader, no resources. */
4191 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4194 struct set
*resource_set
= _mesa_set_create(NULL
,
4196 _mesa_key_pointer_equal
);
4198 /* Program interface needs to expose varyings in case of SSO. */
4199 if (shProg
->SeparateShader
) {
4200 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4201 input_stage
, GL_PROGRAM_INPUT
))
4204 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4205 output_stage
, GL_PROGRAM_OUTPUT
))
4209 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4212 /* Add inputs and outputs to the resource list. */
4213 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4214 input_stage
, GL_PROGRAM_INPUT
))
4217 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4218 output_stage
, GL_PROGRAM_OUTPUT
))
4221 if (shProg
->last_vert_prog
) {
4222 struct gl_transform_feedback_info
*linked_xfb
=
4223 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4225 /* Add transform feedback varyings. */
4226 if (linked_xfb
->NumVarying
> 0) {
4227 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4228 if (!add_program_resource(shProg
, resource_set
,
4229 GL_TRANSFORM_FEEDBACK_VARYING
,
4230 &linked_xfb
->Varyings
[i
], 0))
4235 /* Add transform feedback buffers. */
4236 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4237 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4238 linked_xfb
->Buffers
[i
].Binding
= i
;
4239 if (!add_program_resource(shProg
, resource_set
,
4240 GL_TRANSFORM_FEEDBACK_BUFFER
,
4241 &linked_xfb
->Buffers
[i
], 0))
4247 /* Add uniforms from uniform storage. */
4248 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4249 /* Do not add uniforms internally used by Mesa. */
4250 if (shProg
->data
->UniformStorage
[i
].hidden
)
4254 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4257 /* Add stagereferences for uniforms in a uniform block. */
4258 bool is_shader_storage
=
4259 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4260 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4261 if (block_index
!= -1) {
4262 stageref
|= is_shader_storage
?
4263 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4264 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4267 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4268 if (!should_add_buffer_variable(shProg
, type
,
4269 shProg
->data
->UniformStorage
[i
].name
))
4272 if (is_shader_storage
) {
4273 calculate_array_size_and_stride(shProg
,
4274 &shProg
->data
->UniformStorage
[i
]);
4277 if (!add_program_resource(shProg
, resource_set
, type
,
4278 &shProg
->data
->UniformStorage
[i
], stageref
))
4282 /* Add program uniform blocks. */
4283 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4284 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4285 &shProg
->data
->UniformBlocks
[i
], 0))
4289 /* Add program shader storage blocks. */
4290 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4291 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4292 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4296 /* Add atomic counter buffers. */
4297 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4298 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4299 &shProg
->data
->AtomicBuffers
[i
], 0))
4303 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4305 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4308 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4309 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4310 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4313 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4314 /* add shader subroutines */
4315 if (!add_program_resource(shProg
, resource_set
,
4316 type
, &shProg
->data
->UniformStorage
[i
], 0))
4321 unsigned mask
= shProg
->data
->linked_stages
;
4323 const int i
= u_bit_scan(&mask
);
4324 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4326 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4327 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4328 if (!add_program_resource(shProg
, resource_set
,
4329 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4334 _mesa_set_destroy(resource_set
, NULL
);
4338 * This check is done to make sure we allow only constant expression
4339 * indexing and "constant-index-expression" (indexing with an expression
4340 * that includes loop induction variable).
4343 validate_sampler_array_indexing(struct gl_context
*ctx
,
4344 struct gl_shader_program
*prog
)
4346 dynamic_sampler_array_indexing_visitor v
;
4347 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4348 if (prog
->_LinkedShaders
[i
] == NULL
)
4351 bool no_dynamic_indexing
=
4352 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4354 /* Search for array derefs in shader. */
4355 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4356 if (v
.uses_dynamic_sampler_array_indexing()) {
4357 const char *msg
= "sampler arrays indexed with non-constant "
4358 "expressions is forbidden in GLSL %s %u";
4359 /* Backend has indicated that it has no dynamic indexing support. */
4360 if (no_dynamic_indexing
) {
4361 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4362 prog
->data
->Version
);
4365 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4366 prog
->data
->Version
);
4374 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4376 unsigned mask
= prog
->data
->linked_stages
;
4378 const int i
= u_bit_scan(&mask
);
4379 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4381 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4382 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4383 ir_function
*fn
= node
->as_function();
4387 if (fn
->is_subroutine
)
4388 p
->sh
.NumSubroutineUniformTypes
++;
4390 if (!fn
->num_subroutine_types
)
4393 /* these should have been calculated earlier. */
4394 assert(fn
->subroutine_index
!= -1);
4395 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4396 linker_error(prog
, "Too many subroutine functions declared.\n");
4399 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4400 struct gl_subroutine_function
,
4401 p
->sh
.NumSubroutineFunctions
+ 1);
4402 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4403 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4404 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4405 ralloc_array(p
, const struct glsl_type
*,
4406 fn
->num_subroutine_types
);
4408 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4411 * "Each subroutine with an index qualifier in the shader must be
4412 * given a unique index, otherwise a compile or link error will be
4415 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4416 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4417 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4418 linker_error(prog
, "each subroutine index qualifier in the "
4419 "shader must be unique\n");
4423 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4424 fn
->subroutine_index
;
4426 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4427 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4429 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4430 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4431 p
->sh
.NumSubroutineFunctions
++;
4437 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4439 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4441 foreach_in_list(ir_instruction
, node
, ir
) {
4442 ir_variable
*const var
= node
->as_variable();
4444 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4447 /* Don't set always active on builtins that haven't been redeclared */
4448 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4451 var
->data
.always_active_io
= true;
4456 * When separate shader programs are enabled, only input/outputs between
4457 * the stages of a multi-stage separate program can be safely removed
4458 * from the shader interface. Other inputs/outputs must remain active.
4461 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4463 unsigned first
, last
;
4464 assert(prog
->SeparateShader
);
4466 first
= MESA_SHADER_STAGES
;
4469 /* Determine first and last stage. Excluding the compute stage */
4470 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4471 if (!prog
->_LinkedShaders
[i
])
4473 if (first
== MESA_SHADER_STAGES
)
4478 if (first
== MESA_SHADER_STAGES
)
4481 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4482 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4486 if (first
== last
) {
4487 /* For a single shader program only allow inputs to the vertex shader
4488 * and outputs from the fragment shader to be removed.
4490 if (stage
!= MESA_SHADER_VERTEX
)
4491 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4492 if (stage
!= MESA_SHADER_FRAGMENT
)
4493 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4495 /* For multi-stage separate shader programs only allow inputs and
4496 * outputs between the shader stages to be removed as well as inputs
4497 * to the vertex shader and outputs from the fragment shader.
4499 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4500 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4501 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4502 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4508 link_and_validate_uniforms(struct gl_context
*ctx
,
4509 struct gl_shader_program
*prog
)
4511 update_array_sizes(prog
);
4512 link_assign_uniform_locations(prog
, ctx
);
4514 if (!prog
->data
->cache_fallback
) {
4515 link_assign_atomic_counter_resources(ctx
, prog
);
4516 link_calculate_subroutine_compat(prog
);
4517 check_resources(ctx
, prog
);
4518 check_subroutine_resources(prog
);
4519 check_image_resources(ctx
, prog
);
4520 link_check_atomic_counter_resources(ctx
, prog
);
4525 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4526 struct gl_context
*ctx
,
4527 struct gl_shader_program
*prog
, void *mem_ctx
)
4529 /* Mark all generic shader inputs and outputs as unpaired. */
4530 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4531 if (prog
->_LinkedShaders
[i
] != NULL
) {
4532 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4536 unsigned prev
= first
;
4537 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4538 if (prog
->_LinkedShaders
[i
] == NULL
)
4541 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4542 prog
->_LinkedShaders
[i
]);
4546 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4547 MESA_SHADER_VERTEX
)) {
4551 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4552 MESA_SHADER_FRAGMENT
)) {
4556 prog
->last_vert_prog
= NULL
;
4557 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4558 if (prog
->_LinkedShaders
[i
] == NULL
)
4561 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4565 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4568 link_and_validate_uniforms(ctx
, prog
);
4570 if (!prog
->data
->LinkStatus
)
4573 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4574 if (prog
->_LinkedShaders
[i
] == NULL
)
4577 const struct gl_shader_compiler_options
*options
=
4578 &ctx
->Const
.ShaderCompilerOptions
[i
];
4580 if (options
->LowerBufferInterfaceBlocks
)
4581 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4582 options
->ClampBlockIndicesToArrayBounds
);
4584 if (i
== MESA_SHADER_COMPUTE
)
4585 lower_shared_reference(prog
->_LinkedShaders
[i
],
4586 &prog
->Comp
.SharedSize
);
4588 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4589 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4596 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4599 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4600 /* Run it just once. */
4601 do_common_optimization(ir
, true, false,
4602 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4603 ctx
->Const
.NativeIntegers
);
4605 /* Repeat it until it stops making changes. */
4606 while (do_common_optimization(ir
, true, false,
4607 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4608 ctx
->Const
.NativeIntegers
))
4614 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4616 prog
->data
->LinkStatus
= linking_success
; /* All error paths will set this to false */
4617 prog
->data
->Validated
= false;
4619 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4621 * "Linking can fail for a variety of reasons as specified in the
4622 * OpenGL Shading Language Specification, as well as any of the
4623 * following reasons:
4625 * - No shader objects are attached to program."
4627 * The Compatibility Profile specification does not list the error. In
4628 * Compatibility Profile missing shader stages are replaced by
4629 * fixed-function. This applies to the case where all stages are
4632 if (prog
->NumShaders
== 0) {
4633 if (ctx
->API
!= API_OPENGL_COMPAT
)
4634 linker_error(prog
, "no shaders attached to the program\n");
4638 #ifdef ENABLE_SHADER_CACHE
4639 /* If transform feedback used on the program then compile all shaders. */
4640 bool skip_cache
= false;
4641 if (prog
->TransformFeedback
.NumVarying
> 0) {
4642 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4643 _mesa_glsl_compile_shader(ctx
, prog
->Shaders
[i
], false, false, true);
4648 if (!skip_cache
&& shader_cache_read_program_metadata(ctx
, prog
))
4652 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4654 prog
->ARB_fragment_coord_conventions_enable
= false;
4656 /* Separate the shaders into groups based on their type.
4658 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4659 unsigned num_shaders
[MESA_SHADER_STAGES
];
4661 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4662 shader_list
[i
] = (struct gl_shader
**)
4663 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4667 unsigned min_version
= UINT_MAX
;
4668 unsigned max_version
= 0;
4669 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4670 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4671 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4673 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4674 linker_error(prog
, "all shaders must use same shading "
4675 "language version\n");
4679 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4680 prog
->ARB_fragment_coord_conventions_enable
= true;
4683 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4684 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4685 num_shaders
[shader_type
]++;
4688 /* In desktop GLSL, different shader versions may be linked together. In
4689 * GLSL ES, all shader versions must be the same.
4691 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4692 linker_error(prog
, "all shaders must use same shading "
4693 "language version\n");
4697 prog
->data
->Version
= max_version
;
4698 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4700 /* Some shaders have to be linked with some other shaders present.
4702 if (!prog
->SeparateShader
) {
4703 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4704 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4705 linker_error(prog
, "Geometry shader must be linked with "
4709 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4710 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4711 linker_error(prog
, "Tessellation evaluation shader must be linked "
4712 "with vertex shader\n");
4715 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4716 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4717 linker_error(prog
, "Tessellation control shader must be linked with "
4722 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4724 * "Linking can fail for [...] any of the following reasons:
4726 * * program contains an object to form a tessellation control
4727 * shader [...] and [...] the program is not separable and
4728 * contains no object to form a tessellation evaluation shader"
4730 * The OpenGL spec is contradictory. It allows linking without a tess
4731 * eval shader, but that can only be used with transform feedback and
4732 * rasterization disabled. However, transform feedback isn't allowed
4733 * with GL_PATCHES, so it can't be used.
4735 * More investigation showed that the idea of transform feedback after
4736 * a tess control shader was dropped, because some hw vendors couldn't
4737 * support tessellation without a tess eval shader, but the linker
4738 * section wasn't updated to reflect that.
4740 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4743 * Do what's reasonable and always require a tess eval shader if a tess
4744 * control shader is present.
4746 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4747 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4748 linker_error(prog
, "Tessellation control shader must be linked with "
4749 "tessellation evaluation shader\n");
4754 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4755 num_shaders
[MESA_SHADER_TESS_CTRL
] == 0) {
4756 linker_error(prog
, "GLSL ES requires non-separable programs "
4757 "containing a tessellation evaluation shader to also "
4758 "be linked with a tessellation control shader\n");
4764 /* Compute shaders have additional restrictions. */
4765 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4766 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4767 linker_error(prog
, "Compute shaders may not be linked with any other "
4768 "type of shader\n");
4771 /* Link all shaders for a particular stage and validate the result.
4773 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4774 if (num_shaders
[stage
] > 0) {
4775 gl_linked_shader
*const sh
=
4776 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4777 num_shaders
[stage
], false);
4779 if (!prog
->data
->LinkStatus
) {
4781 _mesa_delete_linked_shader(ctx
, sh
);
4786 case MESA_SHADER_VERTEX
:
4787 validate_vertex_shader_executable(prog
, sh
, ctx
);
4789 case MESA_SHADER_TESS_CTRL
:
4790 /* nothing to be done */
4792 case MESA_SHADER_TESS_EVAL
:
4793 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4795 case MESA_SHADER_GEOMETRY
:
4796 validate_geometry_shader_executable(prog
, sh
, ctx
);
4798 case MESA_SHADER_FRAGMENT
:
4799 validate_fragment_shader_executable(prog
, sh
);
4802 if (!prog
->data
->LinkStatus
) {
4804 _mesa_delete_linked_shader(ctx
, sh
);
4808 prog
->_LinkedShaders
[stage
] = sh
;
4809 prog
->data
->linked_stages
|= 1 << stage
;
4813 /* Here begins the inter-stage linking phase. Some initial validation is
4814 * performed, then locations are assigned for uniforms, attributes, and
4817 cross_validate_uniforms(prog
);
4818 if (!prog
->data
->LinkStatus
)
4821 unsigned first
, last
, prev
;
4823 first
= MESA_SHADER_STAGES
;
4826 /* Determine first and last stage. */
4827 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4828 if (!prog
->_LinkedShaders
[i
])
4830 if (first
== MESA_SHADER_STAGES
)
4835 if (!prog
->data
->cache_fallback
) {
4836 check_explicit_uniform_locations(ctx
, prog
);
4837 link_assign_subroutine_types(prog
);
4840 if (!prog
->data
->LinkStatus
)
4843 resize_tes_inputs(ctx
, prog
);
4845 /* Validate the inputs of each stage with the output of the preceding
4849 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4850 if (prog
->_LinkedShaders
[i
] == NULL
)
4853 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4854 prog
->_LinkedShaders
[i
]);
4855 if (!prog
->data
->LinkStatus
)
4858 cross_validate_outputs_to_inputs(prog
,
4859 prog
->_LinkedShaders
[prev
],
4860 prog
->_LinkedShaders
[i
]);
4861 if (!prog
->data
->LinkStatus
)
4867 /* Cross-validate uniform blocks between shader stages */
4868 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4869 if (!prog
->data
->LinkStatus
)
4872 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4873 if (prog
->_LinkedShaders
[i
] != NULL
)
4874 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4877 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4878 * it before optimization because we want most of the checks to get
4879 * dropped thanks to constant propagation.
4881 * This rule also applies to GLSL ES 3.00.
4883 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4884 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4886 lower_discard_flow(sh
->ir
);
4890 if (prog
->SeparateShader
)
4891 disable_varying_optimizations_for_sso(prog
);
4893 if (!prog
->data
->cache_fallback
) {
4895 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4899 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4903 /* Do common optimization before assigning storage for attributes,
4904 * uniforms, and varyings. Later optimization could possibly make
4905 * some of that unused.
4907 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4908 if (prog
->_LinkedShaders
[i
] == NULL
)
4911 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4912 if (!prog
->data
->LinkStatus
)
4915 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4916 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4919 if (ctx
->Const
.LowerTessLevel
) {
4920 lower_tess_level(prog
->_LinkedShaders
[i
]);
4923 /* Call opts before lowering const arrays to uniforms so we can const
4924 * propagate any elements accessed directly.
4926 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
4928 /* Call opts after lowering const arrays to copy propagate things. */
4929 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
4930 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
4932 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4935 /* Validation for special cases where we allow sampler array indexing
4936 * with loop induction variable. This check emits a warning or error
4937 * depending if backend can handle dynamic indexing.
4939 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
4940 (prog
->IsES
&& prog
->data
->Version
< 300)) {
4941 if (!validate_sampler_array_indexing(ctx
, prog
))
4945 /* Check and validate stream emissions in geometry shaders */
4946 validate_geometry_shader_emissions(ctx
, prog
);
4948 store_fragdepth_layout(prog
);
4950 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
4953 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4954 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4955 * anything about shader linking when one of the shaders (vertex or
4956 * fragment shader) is absent. So, the extension shouldn't change the
4957 * behavior specified in GLSL specification.
4959 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4960 * "Linking can fail for a variety of reasons as specified in the
4961 * OpenGL ES Shading Language Specification, as well as any of the
4962 * following reasons:
4966 * * program contains objects to form either a vertex shader or
4967 * fragment shader, and program is not separable, and does not
4968 * contain objects to form both a vertex shader and fragment
4971 * However, the only scenario in 3.1+ where we don't require them both is
4972 * when we have a compute shader. For example:
4974 * - No shaders is a link error.
4975 * - Geom or Tess without a Vertex shader is a link error which means we
4976 * always require a Vertex shader and hence a Fragment shader.
4977 * - Finally a Compute shader linked with any other stage is a link error.
4979 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4980 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4981 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4982 linker_error(prog
, "program lacks a vertex shader\n");
4983 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4984 linker_error(prog
, "program lacks a fragment shader\n");
4989 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4990 free(shader_list
[i
]);
4991 if (prog
->_LinkedShaders
[i
] == NULL
)
4994 /* Do a final validation step to make sure that the IR wasn't
4995 * invalidated by any modifications performed after intrastage linking.
4997 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4999 /* Retain any live IR, but trash the rest. */
5000 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
5002 /* The symbol table in the linked shaders may contain references to
5003 * variables that were removed (e.g., unused uniforms). Since it may
5004 * contain junk, there is no possible valid use. Delete it and set the
5007 delete prog
->_LinkedShaders
[i
]->symbols
;
5008 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
5011 ralloc_free(mem_ctx
);