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 temporaries that are at global scope. These
889 * will eventually get pulled into the shaders 'main'.
891 if (var
->data
.mode
== ir_var_temporary
)
894 /* If a global with this name has already been seen, verify that the
895 * new instance has the same type. In addition, if the globals have
896 * initializers, the values of the initializers must be the same.
898 ir_variable
*const existing
= variables
->get_variable(var
->name
);
899 if (existing
!= NULL
) {
900 /* Check if types match. Interface blocks have some special
901 * rules so we handle those elsewhere.
903 if (var
->type
!= existing
->type
&&
904 !var
->is_interface_instance()) {
905 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
906 if (var
->type
->is_record() && existing
->type
->is_record()
907 && existing
->type
->record_compare(var
->type
)) {
908 existing
->type
= var
->type
;
910 /* If it is an unsized array in a Shader Storage Block,
911 * two different shaders can access to different elements.
912 * Because of that, they might be converted to different
913 * sized arrays, then check that they are compatible but
914 * ignore the array size.
916 if (!(var
->data
.mode
== ir_var_shader_storage
&&
917 var
->data
.from_ssbo_unsized_array
&&
918 existing
->data
.mode
== ir_var_shader_storage
&&
919 existing
->data
.from_ssbo_unsized_array
&&
920 var
->type
->gl_type
== existing
->type
->gl_type
)) {
921 linker_error(prog
, "%s `%s' declared as type "
922 "`%s' and type `%s'\n",
924 var
->name
, var
->type
->name
,
925 existing
->type
->name
);
932 if (var
->data
.explicit_location
) {
933 if (existing
->data
.explicit_location
934 && (var
->data
.location
!= existing
->data
.location
)) {
935 linker_error(prog
, "explicit locations for %s "
936 "`%s' have differing values\n",
937 mode_string(var
), var
->name
);
941 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
942 linker_error(prog
, "explicit components for %s `%s' have "
943 "differing values\n", mode_string(var
), var
->name
);
947 existing
->data
.location
= var
->data
.location
;
948 existing
->data
.explicit_location
= true;
950 /* Check if uniform with implicit location was marked explicit
951 * by earlier shader stage. If so, mark it explicit in this stage
952 * too to make sure later processing does not treat it as
955 if (existing
->data
.explicit_location
) {
956 var
->data
.location
= existing
->data
.location
;
957 var
->data
.explicit_location
= true;
961 /* From the GLSL 4.20 specification:
962 * "A link error will result if two compilation units in a program
963 * specify different integer-constant bindings for the same
964 * opaque-uniform name. However, it is not an error to specify a
965 * binding on some but not all declarations for the same name"
967 if (var
->data
.explicit_binding
) {
968 if (existing
->data
.explicit_binding
&&
969 var
->data
.binding
!= existing
->data
.binding
) {
970 linker_error(prog
, "explicit bindings for %s "
971 "`%s' have differing values\n",
972 mode_string(var
), var
->name
);
976 existing
->data
.binding
= var
->data
.binding
;
977 existing
->data
.explicit_binding
= true;
980 if (var
->type
->contains_atomic() &&
981 var
->data
.offset
!= existing
->data
.offset
) {
982 linker_error(prog
, "offset specifications for %s "
983 "`%s' have differing values\n",
984 mode_string(var
), var
->name
);
988 /* Validate layout qualifiers for gl_FragDepth.
990 * From the AMD/ARB_conservative_depth specs:
992 * "If gl_FragDepth is redeclared in any fragment shader in a
993 * program, it must be redeclared in all fragment shaders in
994 * that program that have static assignments to
995 * gl_FragDepth. All redeclarations of gl_FragDepth in all
996 * fragment shaders in a single program must have the same set
999 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1000 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1001 bool layout_differs
=
1002 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1004 if (layout_declared
&& layout_differs
) {
1006 "All redeclarations of gl_FragDepth in all "
1007 "fragment shaders in a single program must have "
1008 "the same set of qualifiers.\n");
1011 if (var
->data
.used
&& layout_differs
) {
1013 "If gl_FragDepth is redeclared with a layout "
1014 "qualifier in any fragment shader, it must be "
1015 "redeclared with the same layout qualifier in "
1016 "all fragment shaders that have assignments to "
1021 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1023 * "If a shared global has multiple initializers, the
1024 * initializers must all be constant expressions, and they
1025 * must all have the same value. Otherwise, a link error will
1026 * result. (A shared global having only one initializer does
1027 * not require that initializer to be a constant expression.)"
1029 * Previous to 4.20 the GLSL spec simply said that initializers
1030 * must have the same value. In this case of non-constant
1031 * initializers, this was impossible to determine. As a result,
1032 * no vendor actually implemented that behavior. The 4.20
1033 * behavior matches the implemented behavior of at least one other
1034 * vendor, so we'll implement that for all GLSL versions.
1036 if (var
->constant_initializer
!= NULL
) {
1037 if (existing
->constant_initializer
!= NULL
) {
1038 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1039 linker_error(prog
, "initializers for %s "
1040 "`%s' have differing values\n",
1041 mode_string(var
), var
->name
);
1045 /* If the first-seen instance of a particular uniform did
1046 * not have an initializer but a later instance does,
1047 * replace the former with the later.
1049 variables
->replace_variable(existing
->name
, var
);
1053 if (var
->data
.has_initializer
) {
1054 if (existing
->data
.has_initializer
1055 && (var
->constant_initializer
== NULL
1056 || existing
->constant_initializer
== NULL
)) {
1058 "shared global variable `%s' has multiple "
1059 "non-constant initializers.\n",
1065 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1066 linker_error(prog
, "declarations for %s `%s' have "
1067 "mismatching invariant qualifiers\n",
1068 mode_string(var
), var
->name
);
1071 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1072 linker_error(prog
, "declarations for %s `%s' have "
1073 "mismatching centroid qualifiers\n",
1074 mode_string(var
), var
->name
);
1077 if (existing
->data
.sample
!= var
->data
.sample
) {
1078 linker_error(prog
, "declarations for %s `%s` have "
1079 "mismatching sample qualifiers\n",
1080 mode_string(var
), var
->name
);
1083 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1084 linker_error(prog
, "declarations for %s `%s` have "
1085 "mismatching image format qualifiers\n",
1086 mode_string(var
), var
->name
);
1090 /* Only in GLSL ES 3.10, the precision qualifier should not match
1091 * between block members defined in matched block names within a
1094 * In GLSL ES 3.00 and ES 3.20, precision qualifier for each block
1095 * member should match.
1097 if (prog
->IsES
&& (prog
->data
->Version
!= 310 ||
1098 !var
->get_interface_type()) &&
1099 existing
->data
.precision
!= var
->data
.precision
) {
1100 linker_error(prog
, "declarations for %s `%s` have "
1101 "mismatching precision qualifiers\n",
1102 mode_string(var
), var
->name
);
1106 variables
->add_variable(var
);
1112 * Perform validation of uniforms used across multiple shader stages
1115 cross_validate_uniforms(struct gl_shader_program
*prog
)
1117 glsl_symbol_table variables
;
1118 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1119 if (prog
->_LinkedShaders
[i
] == NULL
)
1122 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1128 * Accumulates the array of buffer blocks and checks that all definitions of
1129 * blocks agree on their contents.
1132 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1135 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1136 struct gl_uniform_block
*blks
= NULL
;
1137 unsigned *num_blks
= validate_ssbo
? &prog
->data
->NumShaderStorageBlocks
:
1138 &prog
->data
->NumUniformBlocks
;
1140 unsigned max_num_buffer_blocks
= 0;
1141 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1142 if (prog
->_LinkedShaders
[i
]) {
1143 if (validate_ssbo
) {
1144 max_num_buffer_blocks
+=
1145 prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1147 max_num_buffer_blocks
+=
1148 prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1153 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1154 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1156 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1157 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1158 InterfaceBlockStageIndex
[i
][j
] = -1;
1163 unsigned sh_num_blocks
;
1164 struct gl_uniform_block
**sh_blks
;
1165 if (validate_ssbo
) {
1166 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ssbos
;
1167 sh_blks
= sh
->Program
->sh
.ShaderStorageBlocks
;
1169 sh_num_blocks
= prog
->_LinkedShaders
[i
]->Program
->info
.num_ubos
;
1170 sh_blks
= sh
->Program
->sh
.UniformBlocks
;
1173 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1174 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1178 linker_error(prog
, "buffer block `%s' has mismatching "
1179 "definitions\n", sh_blks
[j
]->Name
);
1181 for (unsigned k
= 0; k
<= i
; k
++) {
1182 delete[] InterfaceBlockStageIndex
[k
];
1187 InterfaceBlockStageIndex
[i
][index
] = j
;
1191 /* Update per stage block pointers to point to the program list.
1192 * FIXME: We should be able to free the per stage blocks here.
1194 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1195 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1196 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1198 if (stage_index
!= -1) {
1199 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1201 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1202 sh
->Program
->sh
.ShaderStorageBlocks
:
1203 sh
->Program
->sh
.UniformBlocks
;
1205 blks
[j
].stageref
|= sh_blks
[stage_index
]->stageref
;
1206 sh_blks
[stage_index
] = &blks
[j
];
1211 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1212 delete[] InterfaceBlockStageIndex
[i
];
1216 prog
->data
->ShaderStorageBlocks
= blks
;
1218 prog
->data
->UniformBlocks
= blks
;
1225 * Populates a shaders symbol table with all global declarations
1228 populate_symbol_table(gl_linked_shader
*sh
)
1230 sh
->symbols
= new(sh
) glsl_symbol_table
;
1232 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1236 if ((func
= inst
->as_function()) != NULL
) {
1237 sh
->symbols
->add_function(func
);
1238 } else if ((var
= inst
->as_variable()) != NULL
) {
1239 if (var
->data
.mode
!= ir_var_temporary
)
1240 sh
->symbols
->add_variable(var
);
1247 * Remap variables referenced in an instruction tree
1249 * This is used when instruction trees are cloned from one shader and placed in
1250 * another. These trees will contain references to \c ir_variable nodes that
1251 * do not exist in the target shader. This function finds these \c ir_variable
1252 * references and replaces the references with matching variables in the target
1255 * If there is no matching variable in the target shader, a clone of the
1256 * \c ir_variable is made and added to the target shader. The new variable is
1257 * added to \b both the instruction stream and the symbol table.
1259 * \param inst IR tree that is to be processed.
1260 * \param symbols Symbol table containing global scope symbols in the
1262 * \param instructions Instruction stream where new variable declarations
1266 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1269 class remap_visitor
: public ir_hierarchical_visitor
{
1271 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1273 this->target
= target
;
1274 this->symbols
= target
->symbols
;
1275 this->instructions
= target
->ir
;
1276 this->temps
= temps
;
1279 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1281 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1282 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1283 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1285 assert(var
!= NULL
);
1287 return visit_continue
;
1290 ir_variable
*const existing
=
1291 this->symbols
->get_variable(ir
->var
->name
);
1292 if (existing
!= NULL
)
1295 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1297 this->symbols
->add_variable(copy
);
1298 this->instructions
->push_head(copy
);
1302 return visit_continue
;
1306 struct gl_linked_shader
*target
;
1307 glsl_symbol_table
*symbols
;
1308 exec_list
*instructions
;
1312 remap_visitor
v(target
, temps
);
1319 * Move non-declarations from one instruction stream to another
1321 * The intended usage pattern of this function is to pass the pointer to the
1322 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1323 * pointer) for \c last and \c false for \c make_copies on the first
1324 * call. Successive calls pass the return value of the previous call for
1325 * \c last and \c true for \c make_copies.
1327 * \param instructions Source instruction stream
1328 * \param last Instruction after which new instructions should be
1329 * inserted in the target instruction stream
1330 * \param make_copies Flag selecting whether instructions in \c instructions
1331 * should be copied (via \c ir_instruction::clone) into the
1332 * target list or moved.
1335 * The new "last" instruction in the target instruction stream. This pointer
1336 * is suitable for use as the \c last parameter of a later call to this
1340 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1341 bool make_copies
, gl_linked_shader
*target
)
1343 hash_table
*temps
= NULL
;
1346 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1347 _mesa_key_pointer_equal
);
1349 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1350 if (inst
->as_function())
1353 ir_variable
*var
= inst
->as_variable();
1354 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1357 assert(inst
->as_assignment()
1359 || inst
->as_if() /* for initializers with the ?: operator */
1360 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1363 inst
= inst
->clone(target
, NULL
);
1366 _mesa_hash_table_insert(temps
, var
, inst
);
1368 remap_variables(inst
, target
, temps
);
1373 last
->insert_after(inst
);
1378 _mesa_hash_table_destroy(temps
, NULL
);
1385 * This class is only used in link_intrastage_shaders() below but declaring
1386 * it inside that function leads to compiler warnings with some versions of
1389 class array_sizing_visitor
: public deref_type_updater
{
1391 array_sizing_visitor()
1392 : mem_ctx(ralloc_context(NULL
)),
1393 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1394 _mesa_key_pointer_equal
))
1398 ~array_sizing_visitor()
1400 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1401 ralloc_free(this->mem_ctx
);
1404 virtual ir_visitor_status
visit(ir_variable
*var
)
1406 const glsl_type
*type_without_array
;
1407 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1408 fixup_type(&var
->type
, var
->data
.max_array_access
,
1409 var
->data
.from_ssbo_unsized_array
,
1410 &implicit_sized_array
);
1411 var
->data
.implicit_sized_array
= implicit_sized_array
;
1412 type_without_array
= var
->type
->without_array();
1413 if (var
->type
->is_interface()) {
1414 if (interface_contains_unsized_arrays(var
->type
)) {
1415 const glsl_type
*new_type
=
1416 resize_interface_members(var
->type
,
1417 var
->get_max_ifc_array_access(),
1418 var
->is_in_shader_storage_block());
1419 var
->type
= new_type
;
1420 var
->change_interface_type(new_type
);
1422 } else if (type_without_array
->is_interface()) {
1423 if (interface_contains_unsized_arrays(type_without_array
)) {
1424 const glsl_type
*new_type
=
1425 resize_interface_members(type_without_array
,
1426 var
->get_max_ifc_array_access(),
1427 var
->is_in_shader_storage_block());
1428 var
->change_interface_type(new_type
);
1429 var
->type
= update_interface_members_array(var
->type
, new_type
);
1431 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1432 /* Store a pointer to the variable in the unnamed_interfaces
1436 _mesa_hash_table_search(this->unnamed_interfaces
,
1439 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1441 if (interface_vars
== NULL
) {
1442 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1444 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1447 unsigned index
= ifc_type
->field_index(var
->name
);
1448 assert(index
< ifc_type
->length
);
1449 assert(interface_vars
[index
] == NULL
);
1450 interface_vars
[index
] = var
;
1452 return visit_continue
;
1456 * For each unnamed interface block that was discovered while running the
1457 * visitor, adjust the interface type to reflect the newly assigned array
1458 * sizes, and fix up the ir_variable nodes to point to the new interface
1461 void fixup_unnamed_interface_types()
1463 hash_table_call_foreach(this->unnamed_interfaces
,
1464 fixup_unnamed_interface_type
, NULL
);
1469 * If the type pointed to by \c type represents an unsized array, replace
1470 * it with a sized array whose size is determined by max_array_access.
1472 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1473 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1475 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1476 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1477 max_array_access
+ 1);
1478 *implicit_sized
= true;
1479 assert(*type
!= NULL
);
1483 static const glsl_type
*
1484 update_interface_members_array(const glsl_type
*type
,
1485 const glsl_type
*new_interface_type
)
1487 const glsl_type
*element_type
= type
->fields
.array
;
1488 if (element_type
->is_array()) {
1489 const glsl_type
*new_array_type
=
1490 update_interface_members_array(element_type
, new_interface_type
);
1491 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1493 return glsl_type::get_array_instance(new_interface_type
,
1499 * Determine whether the given interface type contains unsized arrays (if
1500 * it doesn't, array_sizing_visitor doesn't need to process it).
1502 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1504 for (unsigned i
= 0; i
< type
->length
; i
++) {
1505 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1506 if (elem_type
->is_unsized_array())
1513 * Create a new interface type based on the given type, with unsized arrays
1514 * replaced by sized arrays whose size is determined by
1515 * max_ifc_array_access.
1517 static const glsl_type
*
1518 resize_interface_members(const glsl_type
*type
,
1519 const int *max_ifc_array_access
,
1522 unsigned num_fields
= type
->length
;
1523 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1524 memcpy(fields
, type
->fields
.structure
,
1525 num_fields
* sizeof(*fields
));
1526 for (unsigned i
= 0; i
< num_fields
; i
++) {
1527 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1528 /* If SSBO last member is unsized array, we don't replace it by a sized
1531 if (is_ssbo
&& i
== (num_fields
- 1))
1532 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1533 true, &implicit_sized_array
);
1535 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1536 false, &implicit_sized_array
);
1537 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1539 glsl_interface_packing packing
=
1540 (glsl_interface_packing
) type
->interface_packing
;
1541 bool row_major
= (bool) type
->interface_row_major
;
1542 const glsl_type
*new_ifc_type
=
1543 glsl_type::get_interface_instance(fields
, num_fields
,
1544 packing
, row_major
, type
->name
);
1546 return new_ifc_type
;
1549 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1552 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1553 ir_variable
**interface_vars
= (ir_variable
**) data
;
1554 unsigned num_fields
= ifc_type
->length
;
1555 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1556 memcpy(fields
, ifc_type
->fields
.structure
,
1557 num_fields
* sizeof(*fields
));
1558 bool interface_type_changed
= false;
1559 for (unsigned i
= 0; i
< num_fields
; i
++) {
1560 if (interface_vars
[i
] != NULL
&&
1561 fields
[i
].type
!= interface_vars
[i
]->type
) {
1562 fields
[i
].type
= interface_vars
[i
]->type
;
1563 interface_type_changed
= true;
1566 if (!interface_type_changed
) {
1570 glsl_interface_packing packing
=
1571 (glsl_interface_packing
) ifc_type
->interface_packing
;
1572 bool row_major
= (bool) ifc_type
->interface_row_major
;
1573 const glsl_type
*new_ifc_type
=
1574 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1575 row_major
, ifc_type
->name
);
1577 for (unsigned i
= 0; i
< num_fields
; i
++) {
1578 if (interface_vars
[i
] != NULL
)
1579 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1584 * Memory context used to allocate the data in \c unnamed_interfaces.
1589 * Hash table from const glsl_type * to an array of ir_variable *'s
1590 * pointing to the ir_variables constituting each unnamed interface block.
1592 hash_table
*unnamed_interfaces
;
1596 validate_xfb_buffer_stride(struct gl_context
*ctx
, unsigned idx
,
1597 struct gl_shader_program
*prog
)
1599 /* We will validate doubles at a later stage */
1600 if (prog
->TransformFeedback
.BufferStride
[idx
] % 4) {
1601 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1602 "multiple of 4 or if its applied to a type that is "
1603 "or contains a double a multiple of 8.",
1604 prog
->TransformFeedback
.BufferStride
[idx
]);
1608 if (prog
->TransformFeedback
.BufferStride
[idx
] / 4 >
1609 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1610 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1611 "limit has been exceeded.");
1619 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1623 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1624 struct gl_shader_program
*prog
,
1625 struct gl_linked_shader
*linked_shader
,
1626 struct gl_shader
**shader_list
,
1627 unsigned num_shaders
)
1629 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1630 prog
->TransformFeedback
.BufferStride
[i
] = 0;
1633 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1634 struct gl_shader
*shader
= shader_list
[i
];
1636 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1637 if (shader
->TransformFeedbackBufferStride
[j
]) {
1638 if (prog
->TransformFeedback
.BufferStride
[j
] == 0) {
1639 prog
->TransformFeedback
.BufferStride
[j
] =
1640 shader
->TransformFeedbackBufferStride
[j
];
1641 if (!validate_xfb_buffer_stride(ctx
, j
, prog
))
1643 } else if (prog
->TransformFeedback
.BufferStride
[j
] !=
1644 shader
->TransformFeedbackBufferStride
[j
]){
1646 "intrastage shaders defined with conflicting "
1647 "xfb_stride for buffer %d (%d and %d)\n", j
,
1648 prog
->TransformFeedback
.BufferStride
[j
],
1649 shader
->TransformFeedbackBufferStride
[j
]);
1658 * Performs the cross-validation of tessellation control shader vertices and
1659 * layout qualifiers for the attached tessellation control shaders,
1660 * and propagates them to the linked TCS and linked shader program.
1663 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1664 struct gl_program
*gl_prog
,
1665 struct gl_shader
**shader_list
,
1666 unsigned num_shaders
)
1668 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_CTRL
)
1671 gl_prog
->info
.tess
.tcs_vertices_out
= 0;
1673 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1675 * "All tessellation control shader layout declarations in a program
1676 * must specify the same output patch vertex count. There must be at
1677 * least one layout qualifier specifying an output patch vertex count
1678 * in any program containing tessellation control shaders; however,
1679 * such a declaration is not required in all tessellation control
1683 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1684 struct gl_shader
*shader
= shader_list
[i
];
1686 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1687 if (gl_prog
->info
.tess
.tcs_vertices_out
!= 0 &&
1688 gl_prog
->info
.tess
.tcs_vertices_out
!=
1689 (unsigned) shader
->info
.TessCtrl
.VerticesOut
) {
1690 linker_error(prog
, "tessellation control shader defined with "
1691 "conflicting output vertex count (%d and %d)\n",
1692 gl_prog
->info
.tess
.tcs_vertices_out
,
1693 shader
->info
.TessCtrl
.VerticesOut
);
1696 gl_prog
->info
.tess
.tcs_vertices_out
=
1697 shader
->info
.TessCtrl
.VerticesOut
;
1701 /* Just do the intrastage -> interstage propagation right now,
1702 * since we already know we're in the right type of shader program
1705 if (gl_prog
->info
.tess
.tcs_vertices_out
== 0) {
1706 linker_error(prog
, "tessellation control shader didn't declare "
1707 "vertices out layout qualifier\n");
1714 * Performs the cross-validation of tessellation evaluation shader
1715 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1716 * for the attached tessellation evaluation shaders, and propagates them
1717 * to the linked TES and linked shader program.
1720 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1721 struct gl_program
*gl_prog
,
1722 struct gl_shader
**shader_list
,
1723 unsigned num_shaders
)
1725 if (gl_prog
->info
.stage
!= MESA_SHADER_TESS_EVAL
)
1728 int point_mode
= -1;
1729 unsigned vertex_order
= 0;
1731 gl_prog
->info
.tess
.primitive_mode
= PRIM_UNKNOWN
;
1732 gl_prog
->info
.tess
.spacing
= TESS_SPACING_UNSPECIFIED
;
1734 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1736 * "At least one tessellation evaluation shader (compilation unit) in
1737 * a program must declare a primitive mode in its input layout.
1738 * Declaration vertex spacing, ordering, and point mode identifiers is
1739 * optional. It is not required that all tessellation evaluation
1740 * shaders in a program declare a primitive mode. If spacing or
1741 * vertex ordering declarations are omitted, the tessellation
1742 * primitive generator will use equal spacing or counter-clockwise
1743 * vertex ordering, respectively. If a point mode declaration is
1744 * omitted, the tessellation primitive generator will produce lines or
1745 * triangles according to the primitive mode."
1748 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1749 struct gl_shader
*shader
= shader_list
[i
];
1751 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1752 if (gl_prog
->info
.tess
.primitive_mode
!= PRIM_UNKNOWN
&&
1753 gl_prog
->info
.tess
.primitive_mode
!=
1754 shader
->info
.TessEval
.PrimitiveMode
) {
1755 linker_error(prog
, "tessellation evaluation shader defined with "
1756 "conflicting input primitive modes.\n");
1759 gl_prog
->info
.tess
.primitive_mode
=
1760 shader
->info
.TessEval
.PrimitiveMode
;
1763 if (shader
->info
.TessEval
.Spacing
!= 0) {
1764 if (gl_prog
->info
.tess
.spacing
!= 0 && gl_prog
->info
.tess
.spacing
!=
1765 shader
->info
.TessEval
.Spacing
) {
1766 linker_error(prog
, "tessellation evaluation shader defined with "
1767 "conflicting vertex spacing.\n");
1770 gl_prog
->info
.tess
.spacing
= shader
->info
.TessEval
.Spacing
;
1773 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1774 if (vertex_order
!= 0 &&
1775 vertex_order
!= shader
->info
.TessEval
.VertexOrder
) {
1776 linker_error(prog
, "tessellation evaluation shader defined with "
1777 "conflicting ordering.\n");
1780 vertex_order
= shader
->info
.TessEval
.VertexOrder
;
1783 if (shader
->info
.TessEval
.PointMode
!= -1) {
1784 if (point_mode
!= -1 &&
1785 point_mode
!= shader
->info
.TessEval
.PointMode
) {
1786 linker_error(prog
, "tessellation evaluation shader defined with "
1787 "conflicting point modes.\n");
1790 point_mode
= shader
->info
.TessEval
.PointMode
;
1795 /* Just do the intrastage -> interstage propagation right now,
1796 * since we already know we're in the right type of shader program
1799 if (gl_prog
->info
.tess
.primitive_mode
== PRIM_UNKNOWN
) {
1801 "tessellation evaluation shader didn't declare input "
1802 "primitive modes.\n");
1806 if (gl_prog
->info
.tess
.spacing
== TESS_SPACING_UNSPECIFIED
)
1807 gl_prog
->info
.tess
.spacing
= TESS_SPACING_EQUAL
;
1809 if (vertex_order
== 0 || vertex_order
== GL_CCW
)
1810 gl_prog
->info
.tess
.ccw
= true;
1812 gl_prog
->info
.tess
.ccw
= false;
1815 if (point_mode
== -1 || point_mode
== GL_FALSE
)
1816 gl_prog
->info
.tess
.point_mode
= false;
1818 gl_prog
->info
.tess
.point_mode
= true;
1823 * Performs the cross-validation of layout qualifiers specified in
1824 * redeclaration of gl_FragCoord for the attached fragment shaders,
1825 * and propagates them to the linked FS and linked shader program.
1828 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1829 struct gl_linked_shader
*linked_shader
,
1830 struct gl_shader
**shader_list
,
1831 unsigned num_shaders
)
1833 bool redeclares_gl_fragcoord
= false;
1834 bool uses_gl_fragcoord
= false;
1835 bool origin_upper_left
= false;
1836 bool pixel_center_integer
= false;
1838 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1839 (prog
->data
->Version
< 150 &&
1840 !prog
->ARB_fragment_coord_conventions_enable
))
1843 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1844 struct gl_shader
*shader
= shader_list
[i
];
1845 /* From the GLSL 1.50 spec, page 39:
1847 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1848 * it must be redeclared in all the fragment shaders in that program
1849 * that have a static use gl_FragCoord."
1851 if ((redeclares_gl_fragcoord
&& !shader
->redeclares_gl_fragcoord
&&
1852 shader
->uses_gl_fragcoord
)
1853 || (shader
->redeclares_gl_fragcoord
&& !redeclares_gl_fragcoord
&&
1854 uses_gl_fragcoord
)) {
1855 linker_error(prog
, "fragment shader defined with conflicting "
1856 "layout qualifiers for gl_FragCoord\n");
1859 /* From the GLSL 1.50 spec, page 39:
1861 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1862 * single program must have the same set of qualifiers."
1864 if (redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
&&
1865 (shader
->origin_upper_left
!= origin_upper_left
||
1866 shader
->pixel_center_integer
!= pixel_center_integer
)) {
1867 linker_error(prog
, "fragment shader defined with conflicting "
1868 "layout qualifiers for gl_FragCoord\n");
1871 /* Update the linked shader state. Note that uses_gl_fragcoord should
1872 * accumulate the results. The other values should replace. If there
1873 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1874 * are already known to be the same.
1876 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1877 redeclares_gl_fragcoord
= shader
->redeclares_gl_fragcoord
;
1878 uses_gl_fragcoord
|= shader
->uses_gl_fragcoord
;
1879 origin_upper_left
= shader
->origin_upper_left
;
1880 pixel_center_integer
= shader
->pixel_center_integer
;
1883 linked_shader
->Program
->info
.fs
.early_fragment_tests
|=
1884 shader
->EarlyFragmentTests
;
1885 linked_shader
->Program
->info
.fs
.inner_coverage
|= shader
->InnerCoverage
;
1886 linked_shader
->Program
->info
.fs
.post_depth_coverage
|=
1887 shader
->PostDepthCoverage
;
1889 linked_shader
->Program
->sh
.fs
.BlendSupport
|= shader
->BlendSupport
;
1894 * Performs the cross-validation of geometry shader max_vertices and
1895 * primitive type layout qualifiers for the attached geometry shaders,
1896 * and propagates them to the linked GS and linked shader program.
1899 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1900 struct gl_program
*gl_prog
,
1901 struct gl_shader
**shader_list
,
1902 unsigned num_shaders
)
1904 /* No in/out qualifiers defined for anything but GLSL 1.50+
1905 * geometry shaders so far.
1907 if (gl_prog
->info
.stage
!= MESA_SHADER_GEOMETRY
||
1908 prog
->data
->Version
< 150)
1911 int vertices_out
= -1;
1913 gl_prog
->info
.gs
.invocations
= 0;
1914 gl_prog
->info
.gs
.input_primitive
= PRIM_UNKNOWN
;
1915 gl_prog
->info
.gs
.output_primitive
= PRIM_UNKNOWN
;
1917 /* From the GLSL 1.50 spec, page 46:
1919 * "All geometry shader output layout declarations in a program
1920 * must declare the same layout and same value for
1921 * max_vertices. There must be at least one geometry output
1922 * layout declaration somewhere in a program, but not all
1923 * geometry shaders (compilation units) are required to
1927 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1928 struct gl_shader
*shader
= shader_list
[i
];
1930 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1931 if (gl_prog
->info
.gs
.input_primitive
!= PRIM_UNKNOWN
&&
1932 gl_prog
->info
.gs
.input_primitive
!=
1933 shader
->info
.Geom
.InputType
) {
1934 linker_error(prog
, "geometry shader defined with conflicting "
1938 gl_prog
->info
.gs
.input_primitive
= shader
->info
.Geom
.InputType
;
1941 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1942 if (gl_prog
->info
.gs
.output_primitive
!= PRIM_UNKNOWN
&&
1943 gl_prog
->info
.gs
.output_primitive
!=
1944 shader
->info
.Geom
.OutputType
) {
1945 linker_error(prog
, "geometry shader defined with conflicting "
1949 gl_prog
->info
.gs
.output_primitive
= shader
->info
.Geom
.OutputType
;
1952 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1953 if (vertices_out
!= -1 &&
1954 vertices_out
!= shader
->info
.Geom
.VerticesOut
) {
1955 linker_error(prog
, "geometry shader defined with conflicting "
1956 "output vertex count (%d and %d)\n",
1957 vertices_out
, shader
->info
.Geom
.VerticesOut
);
1960 vertices_out
= shader
->info
.Geom
.VerticesOut
;
1963 if (shader
->info
.Geom
.Invocations
!= 0) {
1964 if (gl_prog
->info
.gs
.invocations
!= 0 &&
1965 gl_prog
->info
.gs
.invocations
!=
1966 (unsigned) shader
->info
.Geom
.Invocations
) {
1967 linker_error(prog
, "geometry shader defined with conflicting "
1968 "invocation count (%d and %d)\n",
1969 gl_prog
->info
.gs
.invocations
,
1970 shader
->info
.Geom
.Invocations
);
1973 gl_prog
->info
.gs
.invocations
= shader
->info
.Geom
.Invocations
;
1977 /* Just do the intrastage -> interstage propagation right now,
1978 * since we already know we're in the right type of shader program
1981 if (gl_prog
->info
.gs
.input_primitive
== PRIM_UNKNOWN
) {
1983 "geometry shader didn't declare primitive input type\n");
1987 if (gl_prog
->info
.gs
.output_primitive
== PRIM_UNKNOWN
) {
1989 "geometry shader didn't declare primitive output type\n");
1993 if (vertices_out
== -1) {
1995 "geometry shader didn't declare max_vertices\n");
1998 gl_prog
->info
.gs
.vertices_out
= vertices_out
;
2001 if (gl_prog
->info
.gs
.invocations
== 0)
2002 gl_prog
->info
.gs
.invocations
= 1;
2007 * Perform cross-validation of compute shader local_size_{x,y,z} layout
2008 * qualifiers for the attached compute shaders, and propagate them to the
2009 * linked CS and linked shader program.
2012 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
2013 struct gl_program
*gl_prog
,
2014 struct gl_shader
**shader_list
,
2015 unsigned num_shaders
)
2017 /* This function is called for all shader stages, but it only has an effect
2018 * for compute shaders.
2020 if (gl_prog
->info
.stage
!= MESA_SHADER_COMPUTE
)
2023 for (int i
= 0; i
< 3; i
++)
2024 gl_prog
->info
.cs
.local_size
[i
] = 0;
2026 gl_prog
->info
.cs
.local_size_variable
= false;
2028 /* From the ARB_compute_shader spec, in the section describing local size
2031 * If multiple compute shaders attached to a single program object
2032 * declare local work-group size, the declarations must be identical;
2033 * otherwise a link-time error results. Furthermore, if a program
2034 * object contains any compute shaders, at least one must contain an
2035 * input layout qualifier specifying the local work sizes of the
2036 * program, or a link-time error will occur.
2038 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2039 struct gl_shader
*shader
= shader_list
[sh
];
2041 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2042 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2043 for (int i
= 0; i
< 3; i
++) {
2044 if (gl_prog
->info
.cs
.local_size
[i
] !=
2045 shader
->info
.Comp
.LocalSize
[i
]) {
2046 linker_error(prog
, "compute shader defined with conflicting "
2052 for (int i
= 0; i
< 3; i
++) {
2053 gl_prog
->info
.cs
.local_size
[i
] =
2054 shader
->info
.Comp
.LocalSize
[i
];
2056 } else if (shader
->info
.Comp
.LocalSizeVariable
) {
2057 if (gl_prog
->info
.cs
.local_size
[0] != 0) {
2058 /* The ARB_compute_variable_group_size spec says:
2060 * If one compute shader attached to a program declares a
2061 * variable local group size and a second compute shader
2062 * attached to the same program declares a fixed local group
2063 * size, a link-time error results.
2065 linker_error(prog
, "compute shader defined with both fixed and "
2066 "variable local group size\n");
2069 gl_prog
->info
.cs
.local_size_variable
= true;
2073 /* Just do the intrastage -> interstage propagation right now,
2074 * since we already know we're in the right type of shader program
2077 if (gl_prog
->info
.cs
.local_size
[0] == 0 &&
2078 !gl_prog
->info
.cs
.local_size_variable
) {
2079 linker_error(prog
, "compute shader must contain a fixed or a variable "
2080 "local group size\n");
2087 * Combine a group of shaders for a single stage to generate a linked shader
2090 * If this function is supplied a single shader, it is cloned, and the new
2091 * shader is returned.
2093 struct gl_linked_shader
*
2094 link_intrastage_shaders(void *mem_ctx
,
2095 struct gl_context
*ctx
,
2096 struct gl_shader_program
*prog
,
2097 struct gl_shader
**shader_list
,
2098 unsigned num_shaders
,
2099 bool allow_missing_main
)
2101 struct gl_uniform_block
*ubo_blocks
= NULL
;
2102 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2103 unsigned num_ubo_blocks
= 0;
2104 unsigned num_ssbo_blocks
= 0;
2106 /* Check that global variables defined in multiple shaders are consistent.
2108 glsl_symbol_table variables
;
2109 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2110 if (shader_list
[i
] == NULL
)
2112 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2115 if (!prog
->data
->LinkStatus
)
2118 /* Check that interface blocks defined in multiple shaders are consistent.
2120 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2122 if (!prog
->data
->LinkStatus
)
2125 /* Check that there is only a single definition of each function signature
2126 * across all shaders.
2128 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2129 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2130 ir_function
*const f
= node
->as_function();
2135 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2136 ir_function
*const other
=
2137 shader_list
[j
]->symbols
->get_function(f
->name
);
2139 /* If the other shader has no function (and therefore no function
2140 * signatures) with the same name, skip to the next shader.
2145 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2146 if (!sig
->is_defined
)
2149 ir_function_signature
*other_sig
=
2150 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2152 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2153 linker_error(prog
, "function `%s' is multiply defined\n",
2162 /* Find the shader that defines main, and make a clone of it.
2164 * Starting with the clone, search for undefined references. If one is
2165 * found, find the shader that defines it. Clone the reference and add
2166 * it to the shader. Repeat until there are no undefined references or
2167 * until a reference cannot be resolved.
2169 gl_shader
*main
= NULL
;
2170 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2171 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2172 main
= shader_list
[i
];
2177 if (main
== NULL
&& allow_missing_main
)
2178 main
= shader_list
[0];
2181 linker_error(prog
, "%s shader lacks `main'\n",
2182 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2186 gl_linked_shader
*linked
= rzalloc(NULL
, struct gl_linked_shader
);
2187 linked
->Stage
= shader_list
[0]->Stage
;
2189 /* Create program and attach it to the linked shader */
2190 struct gl_program
*gl_prog
=
2191 ctx
->Driver
.NewProgram(ctx
,
2192 _mesa_shader_stage_to_program(shader_list
[0]->Stage
),
2195 prog
->data
->LinkStatus
= linking_failure
;
2196 _mesa_delete_linked_shader(ctx
, linked
);
2200 _mesa_reference_shader_program_data(ctx
, &gl_prog
->sh
.data
, prog
->data
);
2202 /* Don't use _mesa_reference_program() just take ownership */
2203 linked
->Program
= gl_prog
;
2205 linked
->ir
= new(linked
) exec_list
;
2206 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2208 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2209 link_tcs_out_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2210 link_tes_in_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2211 link_gs_inout_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2212 link_cs_input_layout_qualifiers(prog
, gl_prog
, shader_list
, num_shaders
);
2213 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2216 populate_symbol_table(linked
);
2218 /* The pointer to the main function in the final linked shader (i.e., the
2219 * copy of the original shader that contained the main function).
2221 ir_function_signature
*const main_sig
=
2222 _mesa_get_main_function_signature(linked
->symbols
);
2224 /* Move any instructions other than variable declarations or function
2225 * declarations into main.
2227 if (main_sig
!= NULL
) {
2228 exec_node
*insertion_point
=
2229 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2232 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2233 if (shader_list
[i
] == main
)
2236 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2237 insertion_point
, true, linked
);
2241 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2242 _mesa_delete_linked_shader(ctx
, linked
);
2246 /* Make a pass over all variable declarations to ensure that arrays with
2247 * unspecified sizes have a size specified. The size is inferred from the
2248 * max_array_access field.
2250 array_sizing_visitor v
;
2252 v
.fixup_unnamed_interface_types();
2254 /* Link up uniform blocks defined within this stage. */
2255 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2256 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2258 if (!prog
->data
->LinkStatus
) {
2259 _mesa_delete_linked_shader(ctx
, linked
);
2263 /* Copy ubo blocks to linked shader list */
2264 linked
->Program
->sh
.UniformBlocks
=
2265 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2266 ralloc_steal(linked
, ubo_blocks
);
2267 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2268 linked
->Program
->sh
.UniformBlocks
[i
] = &ubo_blocks
[i
];
2270 linked
->Program
->info
.num_ubos
= num_ubo_blocks
;
2272 /* Copy ssbo blocks to linked shader list */
2273 linked
->Program
->sh
.ShaderStorageBlocks
=
2274 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2275 ralloc_steal(linked
, ssbo_blocks
);
2276 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2277 linked
->Program
->sh
.ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2279 linked
->Program
->info
.num_ssbos
= num_ssbo_blocks
;
2281 /* At this point linked should contain all of the linked IR, so
2282 * validate it to make sure nothing went wrong.
2284 validate_ir_tree(linked
->ir
);
2286 /* Set the size of geometry shader input arrays */
2287 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2288 unsigned num_vertices
=
2289 vertices_per_prim(gl_prog
->info
.gs
.input_primitive
);
2290 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2291 MESA_SHADER_GEOMETRY
);
2292 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2293 ir
->accept(&input_resize_visitor
);
2297 if (ctx
->Const
.VertexID_is_zero_based
)
2298 lower_vertex_id(linked
);
2301 /* Compute the source checksum. */
2302 linked
->SourceChecksum
= 0;
2303 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2304 if (shader_list
[i
] == NULL
)
2306 linked
->SourceChecksum
^= shader_list
[i
]->SourceChecksum
;
2314 * Update the sizes of linked shader uniform arrays to the maximum
2317 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2319 * If one or more elements of an array are active,
2320 * GetActiveUniform will return the name of the array in name,
2321 * subject to the restrictions listed above. The type of the array
2322 * is returned in type. The size parameter contains the highest
2323 * array element index used, plus one. The compiler or linker
2324 * determines the highest index used. There will be only one
2325 * active uniform reported by the GL per uniform array.
2329 update_array_sizes(struct gl_shader_program
*prog
)
2331 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2332 if (prog
->_LinkedShaders
[i
] == NULL
)
2335 bool types_were_updated
= false;
2337 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2338 ir_variable
*const var
= node
->as_variable();
2340 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2341 !var
->type
->is_array())
2344 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2345 * will not be eliminated. Since we always do std140, just
2346 * don't resize arrays in UBOs.
2348 * Atomic counters are supposed to get deterministic
2349 * locations assigned based on the declaration ordering and
2350 * sizes, array compaction would mess that up.
2352 * Subroutine uniforms are not removed.
2354 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2355 var
->type
->contains_subroutine() || var
->constant_initializer
)
2358 int size
= var
->data
.max_array_access
;
2359 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2360 if (prog
->_LinkedShaders
[j
] == NULL
)
2363 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2364 ir_variable
*other_var
= node2
->as_variable();
2368 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2369 other_var
->data
.max_array_access
> size
) {
2370 size
= other_var
->data
.max_array_access
;
2375 if (size
+ 1 != (int)var
->type
->length
) {
2376 /* If this is a built-in uniform (i.e., it's backed by some
2377 * fixed-function state), adjust the number of state slots to
2378 * match the new array size. The number of slots per array entry
2379 * is not known. It seems safe to assume that the total number of
2380 * slots is an integer multiple of the number of array elements.
2381 * Determine the number of slots per array element by dividing by
2382 * the old (total) size.
2384 const unsigned num_slots
= var
->get_num_state_slots();
2385 if (num_slots
> 0) {
2386 var
->set_num_state_slots((size
+ 1)
2387 * (num_slots
/ var
->type
->length
));
2390 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2392 types_were_updated
= true;
2396 /* Update the types of dereferences in case we changed any. */
2397 if (types_were_updated
) {
2398 deref_type_updater v
;
2399 v
.run(prog
->_LinkedShaders
[i
]->ir
);
2405 * Resize tessellation evaluation per-vertex inputs to the size of
2406 * tessellation control per-vertex outputs.
2409 resize_tes_inputs(struct gl_context
*ctx
,
2410 struct gl_shader_program
*prog
)
2412 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2415 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2416 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2418 /* If no control shader is present, then the TES inputs are statically
2419 * sized to MaxPatchVertices; the actual size of the arrays won't be
2420 * known until draw time.
2422 const int num_vertices
= tcs
2423 ? tcs
->Program
->info
.tess
.tcs_vertices_out
2424 : ctx
->Const
.MaxPatchVertices
;
2426 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2427 MESA_SHADER_TESS_EVAL
);
2428 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2429 ir
->accept(&input_resize_visitor
);
2432 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2433 /* Convert the gl_PatchVerticesIn system value into a constant, since
2434 * the value is known at this point.
2436 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2437 ir_variable
*var
= ir
->as_variable();
2438 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2439 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2440 void *mem_ctx
= ralloc_parent(var
);
2441 var
->data
.location
= 0;
2442 var
->data
.explicit_location
= false;
2444 var
->data
.mode
= ir_var_auto
;
2445 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2447 var
->data
.mode
= ir_var_uniform
;
2448 var
->data
.how_declared
= ir_var_hidden
;
2449 var
->allocate_state_slots(1);
2450 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2451 slot0
->swizzle
= SWIZZLE_XXXX
;
2452 slot0
->tokens
[0] = STATE_INTERNAL
;
2453 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2454 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2455 slot0
->tokens
[i
] = 0;
2463 * Find a contiguous set of available bits in a bitmask.
2465 * \param used_mask Bits representing used (1) and unused (0) locations
2466 * \param needed_count Number of contiguous bits needed.
2469 * Base location of the available bits on success or -1 on failure.
2472 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2474 unsigned needed_mask
= (1 << needed_count
) - 1;
2475 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2477 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2478 * cannot optimize possibly infinite loops" for the loop below.
2480 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2483 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2484 if ((needed_mask
& ~used_mask
) == needed_mask
)
2495 * Assign locations for either VS inputs or FS outputs
2497 * \param mem_ctx Temporary ralloc context used for linking
2498 * \param prog Shader program whose variables need locations assigned
2499 * \param constants Driver specific constant values for the program.
2500 * \param target_index Selector for the program target to receive location
2501 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2502 * \c MESA_SHADER_FRAGMENT.
2505 * If locations are successfully assigned, true is returned. Otherwise an
2506 * error is emitted to the shader link log and false is returned.
2509 assign_attribute_or_color_locations(void *mem_ctx
,
2510 gl_shader_program
*prog
,
2511 struct gl_constants
*constants
,
2512 unsigned target_index
)
2514 /* Maximum number of generic locations. This corresponds to either the
2515 * maximum number of draw buffers or the maximum number of generic
2518 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2519 constants
->Program
[target_index
].MaxAttribs
:
2520 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2522 /* Mark invalid locations as being used.
2524 unsigned used_locations
= (max_index
>= 32)
2525 ? ~0 : ~((1 << max_index
) - 1);
2526 unsigned double_storage_locations
= 0;
2528 assert((target_index
== MESA_SHADER_VERTEX
)
2529 || (target_index
== MESA_SHADER_FRAGMENT
));
2531 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2535 /* Operate in a total of four passes.
2537 * 1. Invalidate the location assignments for all vertex shader inputs.
2539 * 2. Assign locations for inputs that have user-defined (via
2540 * glBindVertexAttribLocation) locations and outputs that have
2541 * user-defined locations (via glBindFragDataLocation).
2543 * 3. Sort the attributes without assigned locations by number of slots
2544 * required in decreasing order. Fragmentation caused by attribute
2545 * locations assigned by the application may prevent large attributes
2546 * from having enough contiguous space.
2548 * 4. Assign locations to any inputs without assigned locations.
2551 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2552 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2554 const enum ir_variable_mode direction
=
2555 (target_index
== MESA_SHADER_VERTEX
)
2556 ? ir_var_shader_in
: ir_var_shader_out
;
2559 /* Temporary storage for the set of attributes that need locations assigned.
2565 /* Used below in the call to qsort. */
2566 static int compare(const void *a
, const void *b
)
2568 const temp_attr
*const l
= (const temp_attr
*) a
;
2569 const temp_attr
*const r
= (const temp_attr
*) b
;
2571 /* Reversed because we want a descending order sort below. */
2572 return r
->slots
- l
->slots
;
2575 assert(max_index
<= 32);
2577 /* Temporary array for the set of attributes that have locations assigned.
2579 ir_variable
*assigned
[16];
2581 unsigned num_attr
= 0;
2582 unsigned assigned_attr
= 0;
2584 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2585 ir_variable
*const var
= node
->as_variable();
2587 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2590 if (var
->data
.explicit_location
) {
2591 var
->data
.is_unmatched_generic_inout
= 0;
2592 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2593 || (var
->data
.location
< 0)) {
2595 "invalid explicit location %d specified for `%s'\n",
2596 (var
->data
.location
< 0)
2597 ? var
->data
.location
2598 : var
->data
.location
- generic_base
,
2602 } else if (target_index
== MESA_SHADER_VERTEX
) {
2605 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2606 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2607 var
->data
.location
= binding
;
2608 var
->data
.is_unmatched_generic_inout
= 0;
2610 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2613 const char *name
= var
->name
;
2614 const glsl_type
*type
= var
->type
;
2617 /* Check if there's a binding for the variable name */
2618 if (prog
->FragDataBindings
->get(binding
, name
)) {
2619 assert(binding
>= FRAG_RESULT_DATA0
);
2620 var
->data
.location
= binding
;
2621 var
->data
.is_unmatched_generic_inout
= 0;
2623 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2624 var
->data
.index
= index
;
2629 /* If not, but it's an array type, look for name[0] */
2630 if (type
->is_array()) {
2631 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2632 type
= type
->fields
.array
;
2640 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2643 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2645 * "Output binding assignments will cause LinkProgram to fail:
2647 * If the program has an active output assigned to a location greater
2648 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2649 * an active output assigned an index greater than or equal to one;"
2651 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2652 var
->data
.location
- generic_base
>=
2653 (int) constants
->MaxDualSourceDrawBuffers
) {
2655 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2656 "with index %u for %s\n",
2657 var
->data
.location
- generic_base
, var
->data
.index
,
2662 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2664 /* If the variable is not a built-in and has a location statically
2665 * assigned in the shader (presumably via a layout qualifier), make sure
2666 * that it doesn't collide with other assigned locations. Otherwise,
2667 * add it to the list of variables that need linker-assigned locations.
2669 if (var
->data
.location
!= -1) {
2670 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2671 /* From page 61 of the OpenGL 4.0 spec:
2673 * "LinkProgram will fail if the attribute bindings assigned
2674 * by BindAttribLocation do not leave not enough space to
2675 * assign a location for an active matrix attribute or an
2676 * active attribute array, both of which require multiple
2677 * contiguous generic attributes."
2679 * I think above text prohibits the aliasing of explicit and
2680 * automatic assignments. But, aliasing is allowed in manual
2681 * assignments of attribute locations. See below comments for
2684 * From OpenGL 4.0 spec, page 61:
2686 * "It is possible for an application to bind more than one
2687 * attribute name to the same location. This is referred to as
2688 * aliasing. This will only work if only one of the aliased
2689 * attributes is active in the executable program, or if no
2690 * path through the shader consumes more than one attribute of
2691 * a set of attributes aliased to the same location. A link
2692 * error can occur if the linker determines that every path
2693 * through the shader consumes multiple aliased attributes,
2694 * but implementations are not required to generate an error
2697 * From GLSL 4.30 spec, page 54:
2699 * "A program will fail to link if any two non-vertex shader
2700 * input variables are assigned to the same location. For
2701 * vertex shaders, multiple input variables may be assigned
2702 * to the same location using either layout qualifiers or via
2703 * the OpenGL API. However, such aliasing is intended only to
2704 * support vertex shaders where each execution path accesses
2705 * at most one input per each location. Implementations are
2706 * permitted, but not required, to generate link-time errors
2707 * if they detect that every path through the vertex shader
2708 * executable accesses multiple inputs assigned to any single
2709 * location. For all shader types, a program will fail to link
2710 * if explicit location assignments leave the linker unable
2711 * to find space for other variables without explicit
2714 * From OpenGL ES 3.0 spec, page 56:
2716 * "Binding more than one attribute name to the same location
2717 * is referred to as aliasing, and is not permitted in OpenGL
2718 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2719 * fail when this condition exists. However, aliasing is
2720 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2721 * This will only work if only one of the aliased attributes
2722 * is active in the executable program, or if no path through
2723 * the shader consumes more than one attribute of a set of
2724 * attributes aliased to the same location. A link error can
2725 * occur if the linker determines that every path through the
2726 * shader consumes multiple aliased attributes, but implemen-
2727 * tations are not required to generate an error in this case."
2729 * After looking at above references from OpenGL, OpenGL ES and
2730 * GLSL specifications, we allow aliasing of vertex input variables
2731 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2733 * NOTE: This is not required by the spec but its worth mentioning
2734 * here that we're not doing anything to make sure that no path
2735 * through the vertex shader executable accesses multiple inputs
2736 * assigned to any single location.
2739 /* Mask representing the contiguous slots that will be used by
2742 const unsigned attr
= var
->data
.location
- generic_base
;
2743 const unsigned use_mask
= (1 << slots
) - 1;
2744 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2745 ? "vertex shader input" : "fragment shader output";
2747 /* Generate a link error if the requested locations for this
2748 * attribute exceed the maximum allowed attribute location.
2750 if (attr
+ slots
> max_index
) {
2752 "insufficient contiguous locations "
2753 "available for %s `%s' %d %d %d\n", string
,
2754 var
->name
, used_locations
, use_mask
, attr
);
2758 /* Generate a link error if the set of bits requested for this
2759 * attribute overlaps any previously allocated bits.
2761 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2762 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2763 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2766 * "Additionally, for fragment shader outputs, if two
2767 * variables are placed within the same location, they
2768 * must have the same underlying type (floating-point or
2769 * integer). No component aliasing of output variables or
2770 * members is allowed.
2772 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2773 unsigned assigned_slots
=
2774 assigned
[i
]->type
->count_attribute_slots(false);
2775 unsigned assig_attr
=
2776 assigned
[i
]->data
.location
- generic_base
;
2777 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2779 if ((assigned_use_mask
<< assig_attr
) &
2780 (use_mask
<< attr
)) {
2782 const glsl_type
*assigned_type
=
2783 assigned
[i
]->type
->without_array();
2784 const glsl_type
*type
= var
->type
->without_array();
2785 if (assigned_type
->base_type
!= type
->base_type
) {
2786 linker_error(prog
, "types do not match for aliased"
2787 " %ss %s and %s\n", string
,
2788 assigned
[i
]->name
, var
->name
);
2792 unsigned assigned_component_mask
=
2793 ((1 << assigned_type
->vector_elements
) - 1) <<
2794 assigned
[i
]->data
.location_frac
;
2795 unsigned component_mask
=
2796 ((1 << type
->vector_elements
) - 1) <<
2797 var
->data
.location_frac
;
2798 if (assigned_component_mask
& component_mask
) {
2799 linker_error(prog
, "overlapping component is "
2800 "assigned to %ss %s and %s "
2802 string
, assigned
[i
]->name
, var
->name
,
2803 var
->data
.location_frac
);
2808 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2809 (prog
->IsES
&& prog
->data
->Version
>= 300)) {
2810 linker_error(prog
, "overlapping location is assigned "
2811 "to %s `%s' %d %d %d\n", string
, var
->name
,
2812 used_locations
, use_mask
, attr
);
2815 linker_warning(prog
, "overlapping location is assigned "
2816 "to %s `%s' %d %d %d\n", string
, var
->name
,
2817 used_locations
, use_mask
, attr
);
2821 used_locations
|= (use_mask
<< attr
);
2823 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2825 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2826 * active attribute variables may fail to link, unless
2827 * device-dependent optimizations are able to make the program
2828 * fit within available hardware resources. For the purposes
2829 * of this test, attribute variables of the type dvec3, dvec4,
2830 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2831 * count as consuming twice as many attributes as equivalent
2832 * single-precision types. While these types use the same number
2833 * of generic attributes as their single-precision equivalents,
2834 * implementations are permitted to consume two single-precision
2835 * vectors of internal storage for each three- or four-component
2836 * double-precision vector."
2838 * Mark this attribute slot as taking up twice as much space
2839 * so we can count it properly against limits. According to
2840 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2841 * is optional behavior, but it seems preferable.
2843 if (var
->type
->without_array()->is_dual_slot())
2844 double_storage_locations
|= (use_mask
<< attr
);
2847 assigned
[assigned_attr
] = var
;
2853 if (num_attr
>= max_index
) {
2854 linker_error(prog
, "too many %s (max %u)",
2855 target_index
== MESA_SHADER_VERTEX
?
2856 "vertex shader inputs" : "fragment shader outputs",
2860 to_assign
[num_attr
].slots
= slots
;
2861 to_assign
[num_attr
].var
= var
;
2865 if (target_index
== MESA_SHADER_VERTEX
) {
2866 unsigned total_attribs_size
=
2867 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2868 _mesa_bitcount(double_storage_locations
);
2869 if (total_attribs_size
> max_index
) {
2871 "attempt to use %d vertex attribute slots only %d available ",
2872 total_attribs_size
, max_index
);
2877 /* If all of the attributes were assigned locations by the application (or
2878 * are built-in attributes with fixed locations), return early. This should
2879 * be the common case.
2884 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2886 if (target_index
== MESA_SHADER_VERTEX
) {
2887 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2888 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2889 * reserved to prevent it from being automatically allocated below.
2891 find_deref_visitor
find("gl_Vertex");
2893 if (find
.variable_found())
2894 used_locations
|= (1 << 0);
2897 for (unsigned i
= 0; i
< num_attr
; i
++) {
2898 /* Mask representing the contiguous slots that will be used by this
2901 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2903 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2906 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2907 ? "vertex shader input" : "fragment shader output";
2910 "insufficient contiguous locations "
2911 "available for %s `%s'\n",
2912 string
, to_assign
[i
].var
->name
);
2916 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2917 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2918 used_locations
|= (use_mask
<< location
);
2920 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2921 double_storage_locations
|= (use_mask
<< location
);
2924 /* Now that we have all the locations, from the GL 4.5 core spec, section
2925 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2926 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2927 * as equivalent single-precision types.
2929 if (target_index
== MESA_SHADER_VERTEX
) {
2930 unsigned total_attribs_size
=
2931 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2932 _mesa_bitcount(double_storage_locations
);
2933 if (total_attribs_size
> max_index
) {
2935 "attempt to use %d vertex attribute slots only %d available ",
2936 total_attribs_size
, max_index
);
2945 * Match explicit locations of outputs to inputs and deactivate the
2946 * unmatch flag if found so we don't optimise them away.
2949 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2950 gl_linked_shader
*consumer
)
2952 glsl_symbol_table parameters
;
2953 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2956 /* Find all shader outputs in the "producer" stage.
2958 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2959 ir_variable
*const var
= node
->as_variable();
2961 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2964 if (var
->data
.explicit_location
&&
2965 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2966 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2967 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2968 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2972 /* Match inputs to outputs */
2973 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2974 ir_variable
*const input
= node
->as_variable();
2976 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2979 ir_variable
*output
= NULL
;
2980 if (input
->data
.explicit_location
2981 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2982 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2983 [input
->data
.location_frac
];
2985 if (output
!= NULL
){
2986 input
->data
.is_unmatched_generic_inout
= 0;
2987 output
->data
.is_unmatched_generic_inout
= 0;
2994 * Store the gl_FragDepth layout in the gl_shader_program struct.
2997 store_fragdepth_layout(struct gl_shader_program
*prog
)
2999 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3003 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
3005 /* We don't look up the gl_FragDepth symbol directly because if
3006 * gl_FragDepth is not used in the shader, it's removed from the IR.
3007 * However, the symbol won't be removed from the symbol table.
3009 * We're only interested in the cases where the variable is NOT removed
3012 foreach_in_list(ir_instruction
, node
, ir
) {
3013 ir_variable
*const var
= node
->as_variable();
3015 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
3019 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
3020 switch (var
->data
.depth_layout
) {
3021 case ir_depth_layout_none
:
3022 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
3024 case ir_depth_layout_any
:
3025 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
3027 case ir_depth_layout_greater
:
3028 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
3030 case ir_depth_layout_less
:
3031 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
3033 case ir_depth_layout_unchanged
:
3034 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
3045 * Validate the resources used by a program versus the implementation limits
3048 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3050 unsigned total_uniform_blocks
= 0;
3051 unsigned total_shader_storage_blocks
= 0;
3053 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3054 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3059 if (sh
->Program
->info
.num_textures
>
3060 ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
3061 linker_error(prog
, "Too many %s shader texture samplers\n",
3062 _mesa_shader_stage_to_string(i
));
3065 if (sh
->num_uniform_components
>
3066 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
3067 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3068 linker_warning(prog
, "Too many %s shader default uniform block "
3069 "components, but the driver will try to optimize "
3070 "them out; this is non-portable out-of-spec "
3072 _mesa_shader_stage_to_string(i
));
3074 linker_error(prog
, "Too many %s shader default uniform block "
3076 _mesa_shader_stage_to_string(i
));
3080 if (sh
->num_combined_uniform_components
>
3081 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
3082 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
3083 linker_warning(prog
, "Too many %s shader uniform components, "
3084 "but the driver will try to optimize them out; "
3085 "this is non-portable out-of-spec behavior\n",
3086 _mesa_shader_stage_to_string(i
));
3088 linker_error(prog
, "Too many %s shader uniform components\n",
3089 _mesa_shader_stage_to_string(i
));
3093 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3094 total_uniform_blocks
+= sh
->Program
->info
.num_ubos
;
3096 const unsigned max_uniform_blocks
=
3097 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3098 if (max_uniform_blocks
< sh
->Program
->info
.num_ubos
) {
3099 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3100 _mesa_shader_stage_to_string(i
),
3101 sh
->Program
->info
.num_ubos
, max_uniform_blocks
);
3104 const unsigned max_shader_storage_blocks
=
3105 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3106 if (max_shader_storage_blocks
< sh
->Program
->info
.num_ssbos
) {
3107 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3108 _mesa_shader_stage_to_string(i
),
3109 sh
->Program
->info
.num_ssbos
, max_shader_storage_blocks
);
3113 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3114 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3115 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3118 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3119 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3120 total_shader_storage_blocks
,
3121 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3124 for (unsigned i
= 0; i
< prog
->data
->NumUniformBlocks
; i
++) {
3125 if (prog
->data
->UniformBlocks
[i
].UniformBufferSize
>
3126 ctx
->Const
.MaxUniformBlockSize
) {
3127 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3128 prog
->data
->UniformBlocks
[i
].Name
,
3129 prog
->data
->UniformBlocks
[i
].UniformBufferSize
,
3130 ctx
->Const
.MaxUniformBlockSize
);
3134 for (unsigned i
= 0; i
< prog
->data
->NumShaderStorageBlocks
; i
++) {
3135 if (prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3136 ctx
->Const
.MaxShaderStorageBlockSize
) {
3137 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3138 prog
->data
->ShaderStorageBlocks
[i
].Name
,
3139 prog
->data
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3140 ctx
->Const
.MaxShaderStorageBlockSize
);
3146 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3148 unsigned mask
= prog
->data
->linked_stages
;
3150 const int i
= u_bit_scan(&mask
);
3151 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3153 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineUniformRemapTable
; j
++) {
3154 if (p
->sh
.SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3157 struct gl_uniform_storage
*uni
= p
->sh
.SubroutineUniformRemapTable
[j
];
3163 if (p
->sh
.NumSubroutineFunctions
== 0) {
3164 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3167 for (unsigned f
= 0; f
< p
->sh
.NumSubroutineFunctions
; f
++) {
3168 struct gl_subroutine_function
*fn
= &p
->sh
.SubroutineFunctions
[f
];
3169 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3170 if (fn
->types
[k
] == uni
->type
) {
3176 uni
->num_compatible_subroutines
= count
;
3182 check_subroutine_resources(struct gl_shader_program
*prog
)
3184 unsigned mask
= prog
->data
->linked_stages
;
3186 const int i
= u_bit_scan(&mask
);
3187 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3189 if (p
->sh
.NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
) {
3190 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3191 _mesa_shader_stage_to_string(i
));
3196 * Validate shader image resources.
3199 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3201 unsigned total_image_units
= 0;
3202 unsigned fragment_outputs
= 0;
3203 unsigned total_shader_storage_blocks
= 0;
3205 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3208 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3209 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3212 if (sh
->Program
->info
.num_images
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3213 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3214 _mesa_shader_stage_to_string(i
),
3215 sh
->Program
->info
.num_images
,
3216 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3218 total_image_units
+= sh
->Program
->info
.num_images
;
3219 total_shader_storage_blocks
+= sh
->Program
->info
.num_ssbos
;
3221 if (i
== MESA_SHADER_FRAGMENT
) {
3222 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3223 ir_variable
*var
= node
->as_variable();
3224 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3225 /* since there are no double fs outputs - pass false */
3226 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3232 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3233 linker_error(prog
, "Too many combined image uniforms\n");
3235 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3236 ctx
->Const
.MaxCombinedShaderOutputResources
)
3237 linker_error(prog
, "Too many combined image uniforms, shader storage "
3238 " buffers and fragment outputs\n");
3243 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3244 * for a variable, checks for overlaps between other uniforms using explicit
3248 reserve_explicit_locations(struct gl_shader_program
*prog
,
3249 string_to_uint_map
*map
, ir_variable
*var
)
3251 unsigned slots
= var
->type
->uniform_locations();
3252 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3253 unsigned return_value
= slots
;
3255 /* Resize remap table if locations do not fit in the current one. */
3256 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3257 prog
->UniformRemapTable
=
3258 reralloc(prog
, prog
->UniformRemapTable
,
3259 gl_uniform_storage
*,
3262 if (!prog
->UniformRemapTable
) {
3263 linker_error(prog
, "Out of memory during linking.\n");
3267 /* Initialize allocated space. */
3268 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3269 prog
->UniformRemapTable
[i
] = NULL
;
3271 prog
->NumUniformRemapTable
= max_loc
+ 1;
3274 for (unsigned i
= 0; i
< slots
; i
++) {
3275 unsigned loc
= var
->data
.location
+ i
;
3277 /* Check if location is already used. */
3278 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3280 /* Possibly same uniform from a different stage, this is ok. */
3282 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3287 /* ARB_explicit_uniform_location specification states:
3289 * "No two default-block uniform variables in the program can have
3290 * the same location, even if they are unused, otherwise a compiler
3291 * or linker error will be generated."
3294 "location qualifier for uniform %s overlaps "
3295 "previously used location\n",
3300 /* Initialize location as inactive before optimization
3301 * rounds and location assignment.
3303 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3306 /* Note, base location used for arrays. */
3307 map
->put(var
->data
.location
, var
->name
);
3309 return return_value
;
3313 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3314 struct gl_program
*p
,
3317 unsigned slots
= var
->type
->uniform_locations();
3318 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3320 /* Resize remap table if locations do not fit in the current one. */
3321 if (max_loc
+ 1 > p
->sh
.NumSubroutineUniformRemapTable
) {
3322 p
->sh
.SubroutineUniformRemapTable
=
3323 reralloc(p
, p
->sh
.SubroutineUniformRemapTable
,
3324 gl_uniform_storage
*,
3327 if (!p
->sh
.SubroutineUniformRemapTable
) {
3328 linker_error(prog
, "Out of memory during linking.\n");
3332 /* Initialize allocated space. */
3333 for (unsigned i
= p
->sh
.NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3334 p
->sh
.SubroutineUniformRemapTable
[i
] = NULL
;
3336 p
->sh
.NumSubroutineUniformRemapTable
= max_loc
+ 1;
3339 for (unsigned i
= 0; i
< slots
; i
++) {
3340 unsigned loc
= var
->data
.location
+ i
;
3342 /* Check if location is already used. */
3343 if (p
->sh
.SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3345 /* ARB_explicit_uniform_location specification states:
3346 * "No two subroutine uniform variables can have the same location
3347 * in the same shader stage, otherwise a compiler or linker error
3348 * will be generated."
3351 "location qualifier for uniform %s overlaps "
3352 "previously used location\n",
3357 /* Initialize location as inactive before optimization
3358 * rounds and location assignment.
3360 p
->sh
.SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3366 * Check and reserve all explicit uniform locations, called before
3367 * any optimizations happen to handle also inactive uniforms and
3368 * inactive array elements that may get trimmed away.
3371 check_explicit_uniform_locations(struct gl_context
*ctx
,
3372 struct gl_shader_program
*prog
)
3374 prog
->NumExplicitUniformLocations
= 0;
3376 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3379 /* This map is used to detect if overlapping explicit locations
3380 * occur with the same uniform (from different stage) or a different one.
3382 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3385 linker_error(prog
, "Out of memory during linking.\n");
3389 unsigned entries_total
= 0;
3390 unsigned mask
= prog
->data
->linked_stages
;
3392 const int i
= u_bit_scan(&mask
);
3393 struct gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
3395 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
3396 ir_variable
*var
= node
->as_variable();
3397 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3400 if (var
->data
.explicit_location
) {
3402 if (var
->type
->without_array()->is_subroutine())
3403 ret
= reserve_subroutine_explicit_locations(prog
, p
, var
);
3405 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3409 entries_total
+= slots
;
3420 struct empty_uniform_block
*current_block
= NULL
;
3422 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3423 /* We found empty space in UniformRemapTable. */
3424 if (prog
->UniformRemapTable
[i
] == NULL
) {
3425 /* We've found the beginning of a new continous block of empty slots */
3426 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3427 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3428 current_block
->start
= i
;
3429 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3430 ¤t_block
->link
);
3433 /* The current block continues, so we simply increment its slots */
3434 current_block
->slots
++;
3439 prog
->NumExplicitUniformLocations
= entries_total
;
3443 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3444 GLenum type
, const char *name
)
3446 bool found_interface
= false;
3447 unsigned block_name_len
= 0;
3448 const char *block_name_dot
= strchr(name
, '.');
3450 /* These rules only apply to buffer variables. So we return
3451 * true for the rest of types.
3453 if (type
!= GL_BUFFER_VARIABLE
)
3456 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
3457 const char *block_name
= shProg
->data
->ShaderStorageBlocks
[i
].Name
;
3458 block_name_len
= strlen(block_name
);
3460 const char *block_square_bracket
= strchr(block_name
, '[');
3461 if (block_square_bracket
) {
3462 /* The block is part of an array of named interfaces,
3463 * for the name comparison we ignore the "[x]" part.
3465 block_name_len
-= strlen(block_square_bracket
);
3468 if (block_name_dot
) {
3469 /* Check if the variable name starts with the interface
3470 * name. The interface name (if present) should have the
3471 * length than the interface block name we are comparing to.
3473 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3474 if (len
!= block_name_len
)
3478 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3479 found_interface
= true;
3484 /* We remove the interface name from the buffer variable name,
3485 * including the dot that follows it.
3487 if (found_interface
)
3488 name
= name
+ block_name_len
+ 1;
3490 /* The ARB_program_interface_query spec says:
3492 * "For an active shader storage block member declared as an array, an
3493 * entry will be generated only for the first array element, regardless
3494 * of its type. For arrays of aggregate types, the enumeration rules
3495 * are applied recursively for the single enumerated array element."
3497 const char *struct_first_dot
= strchr(name
, '.');
3498 const char *first_square_bracket
= strchr(name
, '[');
3500 /* The buffer variable is on top level and it is not an array */
3501 if (!first_square_bracket
) {
3503 /* The shader storage block member is a struct, then generate the entry */
3504 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3507 /* Shader storage block member is an array, only generate an entry for the
3508 * first array element.
3510 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3518 add_program_resource(struct gl_shader_program
*prog
,
3519 struct set
*resource_set
,
3520 GLenum type
, const void *data
, uint8_t stages
)
3524 /* If resource already exists, do not add it again. */
3525 if (_mesa_set_search(resource_set
, data
))
3528 prog
->data
->ProgramResourceList
=
3530 prog
->data
->ProgramResourceList
,
3531 gl_program_resource
,
3532 prog
->data
->NumProgramResourceList
+ 1);
3534 if (!prog
->data
->ProgramResourceList
) {
3535 linker_error(prog
, "Out of memory during linking.\n");
3539 struct gl_program_resource
*res
=
3540 &prog
->data
->ProgramResourceList
[prog
->data
->NumProgramResourceList
];
3544 res
->StageReferences
= stages
;
3546 prog
->data
->NumProgramResourceList
++;
3548 _mesa_set_add(resource_set
, data
);
3553 /* Function checks if a variable var is a packed varying and
3554 * if given name is part of packed varying's list.
3556 * If a variable is a packed varying, it has a name like
3557 * 'packed:a,b,c' where a, b and c are separate variables.
3560 included_in_packed_varying(ir_variable
*var
, const char *name
)
3562 if (strncmp(var
->name
, "packed:", 7) != 0)
3565 char *list
= strdup(var
->name
+ 7);
3570 char *token
= strtok_r(list
, ",", &saveptr
);
3572 if (strcmp(token
, name
) == 0) {
3576 token
= strtok_r(NULL
, ",", &saveptr
);
3583 * Function builds a stage reference bitmask from variable name.
3586 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3591 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3592 * used for reference mask in gl_program_resource will need to be changed.
3594 assert(MESA_SHADER_STAGES
< 8);
3596 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3597 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3601 /* Shader symbol table may contain variables that have
3602 * been optimized away. Search IR for the variable instead.
3604 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3605 ir_variable
*var
= node
->as_variable();
3607 unsigned baselen
= strlen(var
->name
);
3609 if (included_in_packed_varying(var
, name
)) {
3614 /* Type needs to match if specified, otherwise we might
3615 * pick a variable with same name but different interface.
3617 if (var
->data
.mode
!= mode
)
3620 if (strncmp(var
->name
, name
, baselen
) == 0) {
3621 /* Check for exact name matches but also check for arrays and
3624 if (name
[baselen
] == '\0' ||
3625 name
[baselen
] == '[' ||
3626 name
[baselen
] == '.') {
3638 * Create gl_shader_variable from ir_variable class.
3640 static gl_shader_variable
*
3641 create_shader_variable(struct gl_shader_program
*shProg
,
3642 const ir_variable
*in
,
3643 const char *name
, const glsl_type
*type
,
3644 const glsl_type
*interface_type
,
3645 bool use_implicit_location
, int location
,
3646 const glsl_type
*outermost_struct_type
)
3648 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3652 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3653 * expect to see gl_VertexID in the program resource list. Pretend.
3655 if (in
->data
.mode
== ir_var_system_value
&&
3656 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3657 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3658 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3659 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3660 (in
->data
.mode
== ir_var_system_value
&&
3661 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3662 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3663 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3664 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3665 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3666 (in
->data
.mode
== ir_var_system_value
&&
3667 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3668 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3669 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3671 out
->name
= ralloc_strdup(shProg
, name
);
3677 /* The ARB_program_interface_query spec says:
3679 * "Not all active variables are assigned valid locations; the
3680 * following variables will have an effective location of -1:
3682 * * uniforms declared as atomic counters;
3684 * * members of a uniform block;
3686 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3688 * * inputs or outputs not declared with a "location" layout
3689 * qualifier, except for vertex shader inputs and fragment shader
3692 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3693 is_gl_identifier(in
->name
) ||
3694 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3697 out
->location
= location
;
3701 out
->outermost_struct_type
= outermost_struct_type
;
3702 out
->interface_type
= interface_type
;
3703 out
->component
= in
->data
.location_frac
;
3704 out
->index
= in
->data
.index
;
3705 out
->patch
= in
->data
.patch
;
3706 out
->mode
= in
->data
.mode
;
3707 out
->interpolation
= in
->data
.interpolation
;
3708 out
->explicit_location
= in
->data
.explicit_location
;
3709 out
->precision
= in
->data
.precision
;
3715 add_shader_variable(const struct gl_context
*ctx
,
3716 struct gl_shader_program
*shProg
,
3717 struct set
*resource_set
,
3718 unsigned stage_mask
,
3719 GLenum programInterface
, ir_variable
*var
,
3720 const char *name
, const glsl_type
*type
,
3721 bool use_implicit_location
, int location
,
3722 const glsl_type
*outermost_struct_type
= NULL
)
3724 const glsl_type
*interface_type
= var
->get_interface_type();
3726 if (outermost_struct_type
== NULL
) {
3727 if (var
->data
.from_named_ifc_block
) {
3728 const char *interface_name
= interface_type
->name
;
3730 if (interface_type
->is_array()) {
3731 /* Issue #16 of the ARB_program_interface_query spec says:
3733 * "* If a variable is a member of an interface block without an
3734 * instance name, it is enumerated using just the variable name.
3736 * * If a variable is a member of an interface block with an
3737 * instance name, it is enumerated as "BlockName.Member", where
3738 * "BlockName" is the name of the interface block (not the
3739 * instance name) and "Member" is the name of the variable."
3741 * In particular, it indicates that it should be "BlockName",
3742 * not "BlockName[array length]". The conformance suite and
3743 * dEQP both require this behavior.
3745 * Here, we unwrap the extra array level added by named interface
3746 * block array lowering so we have the correct variable type. We
3747 * also unwrap the interface type when constructing the name.
3749 * We leave interface_type the same so that ES 3.x SSO pipeline
3750 * validation can enforce the rules requiring array length to
3751 * match on interface blocks.
3753 type
= type
->fields
.array
;
3755 interface_name
= interface_type
->fields
.array
->name
;
3758 name
= ralloc_asprintf(shProg
, "%s.%s", interface_name
, name
);
3762 switch (type
->base_type
) {
3763 case GLSL_TYPE_STRUCT
: {
3764 /* The ARB_program_interface_query spec says:
3766 * "For an active variable declared as a structure, a separate entry
3767 * will be generated for each active structure member. The name of
3768 * each entry is formed by concatenating the name of the structure,
3769 * the "." character, and the name of the structure member. If a
3770 * structure member to enumerate is itself a structure or array,
3771 * these enumeration rules are applied recursively."
3773 if (outermost_struct_type
== NULL
)
3774 outermost_struct_type
= type
;
3776 unsigned field_location
= location
;
3777 for (unsigned i
= 0; i
< type
->length
; i
++) {
3778 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3779 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3780 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3781 stage_mask
, programInterface
,
3782 var
, field_name
, field
->type
,
3783 use_implicit_location
, field_location
,
3784 outermost_struct_type
))
3787 field_location
+= field
->type
->count_attribute_slots(false);
3793 /* The ARB_program_interface_query spec says:
3795 * "For an active variable declared as a single instance of a basic
3796 * type, a single entry will be generated, using the variable name
3797 * from the shader source."
3799 gl_shader_variable
*sha_v
=
3800 create_shader_variable(shProg
, var
, name
, type
, interface_type
,
3801 use_implicit_location
, location
,
3802 outermost_struct_type
);
3806 return add_program_resource(shProg
, resource_set
,
3807 programInterface
, sha_v
, stage_mask
);
3813 add_interface_variables(const struct gl_context
*ctx
,
3814 struct gl_shader_program
*shProg
,
3815 struct set
*resource_set
,
3816 unsigned stage
, GLenum programInterface
)
3818 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3820 foreach_in_list(ir_instruction
, node
, ir
) {
3821 ir_variable
*var
= node
->as_variable();
3823 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3828 switch (var
->data
.mode
) {
3829 case ir_var_system_value
:
3830 case ir_var_shader_in
:
3831 if (programInterface
!= GL_PROGRAM_INPUT
)
3833 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3834 : int(VARYING_SLOT_VAR0
);
3836 case ir_var_shader_out
:
3837 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3839 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3840 : int(VARYING_SLOT_VAR0
);
3846 if (var
->data
.patch
)
3847 loc_bias
= int(VARYING_SLOT_PATCH0
);
3849 /* Skip packed varyings, packed varyings are handled separately
3850 * by add_packed_varyings.
3852 if (strncmp(var
->name
, "packed:", 7) == 0)
3855 /* Skip fragdata arrays, these are handled separately
3856 * by add_fragdata_arrays.
3858 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3861 const bool vs_input_or_fs_output
=
3862 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3863 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3865 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3866 1 << stage
, programInterface
,
3867 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3868 var
->data
.location
- loc_bias
))
3875 add_packed_varyings(const struct gl_context
*ctx
,
3876 struct gl_shader_program
*shProg
,
3877 struct set
*resource_set
,
3878 int stage
, GLenum type
)
3880 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3883 if (!sh
|| !sh
->packed_varyings
)
3886 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3887 ir_variable
*var
= node
->as_variable();
3889 switch (var
->data
.mode
) {
3890 case ir_var_shader_in
:
3891 iface
= GL_PROGRAM_INPUT
;
3893 case ir_var_shader_out
:
3894 iface
= GL_PROGRAM_OUTPUT
;
3897 unreachable("unexpected type");
3900 if (type
== iface
) {
3901 const int stage_mask
=
3902 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3903 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3905 iface
, var
, var
->name
, var
->type
, false,
3906 var
->data
.location
- VARYING_SLOT_VAR0
))
3915 add_fragdata_arrays(const struct gl_context
*ctx
,
3916 struct gl_shader_program
*shProg
,
3917 struct set
*resource_set
)
3919 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3921 if (!sh
|| !sh
->fragdata_arrays
)
3924 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3925 ir_variable
*var
= node
->as_variable();
3927 assert(var
->data
.mode
== ir_var_shader_out
);
3929 if (!add_shader_variable(ctx
, shProg
, resource_set
,
3930 1 << MESA_SHADER_FRAGMENT
,
3931 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3932 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3940 get_top_level_name(const char *name
)
3942 const char *first_dot
= strchr(name
, '.');
3943 const char *first_square_bracket
= strchr(name
, '[');
3946 /* The ARB_program_interface_query spec says:
3948 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3949 * the number of active array elements of the top-level shader storage
3950 * block member containing to the active variable is written to
3951 * <params>. If the top-level block member is not declared as an
3952 * array, the value one is written to <params>. If the top-level block
3953 * member is an array with no declared size, the value zero is written
3957 /* The buffer variable is on top level.*/
3958 if (!first_square_bracket
&& !first_dot
)
3959 name_size
= strlen(name
);
3960 else if ((!first_square_bracket
||
3961 (first_dot
&& first_dot
< first_square_bracket
)))
3962 name_size
= first_dot
- name
;
3964 name_size
= first_square_bracket
- name
;
3966 return strndup(name
, name_size
);
3970 get_var_name(const char *name
)
3972 const char *first_dot
= strchr(name
, '.');
3975 return strdup(name
);
3977 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3981 is_top_level_shader_storage_block_member(const char* name
,
3982 const char* interface_name
,
3983 const char* field_name
)
3985 bool result
= false;
3987 /* If the given variable is already a top-level shader storage
3988 * block member, then return array_size = 1.
3989 * We could have two possibilities: if we have an instanced
3990 * shader storage block or not instanced.
3992 * For the first, we check create a name as it was in top level and
3993 * compare it with the real name. If they are the same, then
3994 * the variable is already at top-level.
3996 * Full instanced name is: interface name + '.' + var name +
3999 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
4000 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
4001 if (!full_instanced_name
) {
4002 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
4006 snprintf(full_instanced_name
, name_length
, "%s.%s",
4007 interface_name
, field_name
);
4009 /* Check if its top-level shader storage block member of an
4010 * instanced interface block, or of a unnamed interface block.
4012 if (strcmp(name
, full_instanced_name
) == 0 ||
4013 strcmp(name
, field_name
) == 0)
4016 free(full_instanced_name
);
4021 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
4022 char *interface_name
, char *var_name
)
4024 /* The ARB_program_interface_query spec says:
4026 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
4027 * the number of active array elements of the top-level shader storage
4028 * block member containing to the active variable is written to
4029 * <params>. If the top-level block member is not declared as an
4030 * array, the value one is written to <params>. If the top-level block
4031 * member is an array with no declared size, the value zero is written
4034 if (is_top_level_shader_storage_block_member(uni
->name
,
4038 else if (field
->type
->is_unsized_array())
4040 else if (field
->type
->is_array())
4041 return field
->type
->length
;
4047 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
4048 const glsl_struct_field
*field
, char *interface_name
,
4051 /* The ARB_program_interface_query spec says:
4053 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
4054 * identifying the stride between array elements of the top-level
4055 * shader storage block member containing the active variable is
4056 * written to <params>. For top-level block members declared as
4057 * arrays, the value written is the difference, in basic machine units,
4058 * between the offsets of the active variable for consecutive elements
4059 * in the top-level array. For top-level block members not declared as
4060 * an array, zero is written to <params>."
4062 if (field
->type
->is_array()) {
4063 const enum glsl_matrix_layout matrix_layout
=
4064 glsl_matrix_layout(field
->matrix_layout
);
4065 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
4066 const glsl_type
*array_type
= field
->type
->fields
.array
;
4068 if (is_top_level_shader_storage_block_member(uni
->name
,
4073 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
4074 if (array_type
->is_record() || array_type
->is_array())
4075 return glsl_align(array_type
->std140_size(row_major
), 16);
4077 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
4079 return array_type
->std430_array_stride(row_major
);
4086 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
4087 struct gl_uniform_storage
*uni
)
4089 int block_index
= uni
->block_index
;
4090 int array_size
= -1;
4091 int array_stride
= -1;
4092 char *var_name
= get_top_level_name(uni
->name
);
4093 char *interface_name
=
4094 get_top_level_name(uni
->is_shader_storage
?
4095 shProg
->data
->ShaderStorageBlocks
[block_index
].Name
:
4096 shProg
->data
->UniformBlocks
[block_index
].Name
);
4098 if (strcmp(var_name
, interface_name
) == 0) {
4099 /* Deal with instanced array of SSBOs */
4100 char *temp_name
= get_var_name(uni
->name
);
4102 linker_error(shProg
, "Out of memory during linking.\n");
4103 goto write_top_level_array_size_and_stride
;
4106 var_name
= get_top_level_name(temp_name
);
4109 linker_error(shProg
, "Out of memory during linking.\n");
4110 goto write_top_level_array_size_and_stride
;
4114 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4115 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4119 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4120 ir_variable
*var
= node
->as_variable();
4121 if (!var
|| !var
->get_interface_type() ||
4122 var
->data
.mode
!= ir_var_shader_storage
)
4125 const glsl_type
*interface
= var
->get_interface_type();
4127 if (strcmp(interface_name
, interface
->name
) != 0)
4130 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4131 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4132 if (strcmp(field
->name
, var_name
) != 0)
4135 array_stride
= get_array_stride(uni
, interface
, field
,
4136 interface_name
, var_name
);
4137 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4138 goto write_top_level_array_size_and_stride
;
4142 write_top_level_array_size_and_stride
:
4143 free(interface_name
);
4145 uni
->top_level_array_stride
= array_stride
;
4146 uni
->top_level_array_size
= array_size
;
4150 * Builds up a list of program resources that point to existing
4154 build_program_resource_list(struct gl_context
*ctx
,
4155 struct gl_shader_program
*shProg
)
4157 /* Rebuild resource list. */
4158 if (shProg
->data
->ProgramResourceList
) {
4159 ralloc_free(shProg
->data
->ProgramResourceList
);
4160 shProg
->data
->ProgramResourceList
= NULL
;
4161 shProg
->data
->NumProgramResourceList
= 0;
4164 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4166 /* Determine first input and final output stage. These are used to
4167 * detect which variables should be enumerated in the resource list
4168 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4170 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4171 if (!shProg
->_LinkedShaders
[i
])
4173 if (input_stage
== MESA_SHADER_STAGES
)
4178 /* Empty shader, no resources. */
4179 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4182 struct set
*resource_set
= _mesa_set_create(NULL
,
4184 _mesa_key_pointer_equal
);
4186 /* Program interface needs to expose varyings in case of SSO. */
4187 if (shProg
->SeparateShader
) {
4188 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4189 input_stage
, GL_PROGRAM_INPUT
))
4192 if (!add_packed_varyings(ctx
, shProg
, resource_set
,
4193 output_stage
, GL_PROGRAM_OUTPUT
))
4197 if (!add_fragdata_arrays(ctx
, shProg
, resource_set
))
4200 /* Add inputs and outputs to the resource list. */
4201 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4202 input_stage
, GL_PROGRAM_INPUT
))
4205 if (!add_interface_variables(ctx
, shProg
, resource_set
,
4206 output_stage
, GL_PROGRAM_OUTPUT
))
4209 if (shProg
->last_vert_prog
) {
4210 struct gl_transform_feedback_info
*linked_xfb
=
4211 shProg
->last_vert_prog
->sh
.LinkedTransformFeedback
;
4213 /* Add transform feedback varyings. */
4214 if (linked_xfb
->NumVarying
> 0) {
4215 for (int i
= 0; i
< linked_xfb
->NumVarying
; i
++) {
4216 if (!add_program_resource(shProg
, resource_set
,
4217 GL_TRANSFORM_FEEDBACK_VARYING
,
4218 &linked_xfb
->Varyings
[i
], 0))
4223 /* Add transform feedback buffers. */
4224 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4225 if ((linked_xfb
->ActiveBuffers
>> i
) & 1) {
4226 linked_xfb
->Buffers
[i
].Binding
= i
;
4227 if (!add_program_resource(shProg
, resource_set
,
4228 GL_TRANSFORM_FEEDBACK_BUFFER
,
4229 &linked_xfb
->Buffers
[i
], 0))
4235 /* Add uniforms from uniform storage. */
4236 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4237 /* Do not add uniforms internally used by Mesa. */
4238 if (shProg
->data
->UniformStorage
[i
].hidden
)
4242 build_stageref(shProg
, shProg
->data
->UniformStorage
[i
].name
,
4245 /* Add stagereferences for uniforms in a uniform block. */
4246 bool is_shader_storage
=
4247 shProg
->data
->UniformStorage
[i
].is_shader_storage
;
4248 int block_index
= shProg
->data
->UniformStorage
[i
].block_index
;
4249 if (block_index
!= -1) {
4250 stageref
|= is_shader_storage
?
4251 shProg
->data
->ShaderStorageBlocks
[block_index
].stageref
:
4252 shProg
->data
->UniformBlocks
[block_index
].stageref
;
4255 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4256 if (!should_add_buffer_variable(shProg
, type
,
4257 shProg
->data
->UniformStorage
[i
].name
))
4260 if (is_shader_storage
) {
4261 calculate_array_size_and_stride(shProg
,
4262 &shProg
->data
->UniformStorage
[i
]);
4265 if (!add_program_resource(shProg
, resource_set
, type
,
4266 &shProg
->data
->UniformStorage
[i
], stageref
))
4270 /* Add program uniform blocks. */
4271 for (unsigned i
= 0; i
< shProg
->data
->NumUniformBlocks
; i
++) {
4272 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4273 &shProg
->data
->UniformBlocks
[i
], 0))
4277 /* Add program shader storage blocks. */
4278 for (unsigned i
= 0; i
< shProg
->data
->NumShaderStorageBlocks
; i
++) {
4279 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4280 &shProg
->data
->ShaderStorageBlocks
[i
], 0))
4284 /* Add atomic counter buffers. */
4285 for (unsigned i
= 0; i
< shProg
->data
->NumAtomicBuffers
; i
++) {
4286 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4287 &shProg
->data
->AtomicBuffers
[i
], 0))
4291 for (unsigned i
= 0; i
< shProg
->data
->NumUniformStorage
; i
++) {
4293 if (!shProg
->data
->UniformStorage
[i
].hidden
)
4296 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4297 if (!shProg
->data
->UniformStorage
[i
].opaque
[j
].active
||
4298 !shProg
->data
->UniformStorage
[i
].type
->is_subroutine())
4301 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4302 /* add shader subroutines */
4303 if (!add_program_resource(shProg
, resource_set
,
4304 type
, &shProg
->data
->UniformStorage
[i
], 0))
4309 unsigned mask
= shProg
->data
->linked_stages
;
4311 const int i
= u_bit_scan(&mask
);
4312 struct gl_program
*p
= shProg
->_LinkedShaders
[i
]->Program
;
4314 GLuint type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4315 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4316 if (!add_program_resource(shProg
, resource_set
,
4317 type
, &p
->sh
.SubroutineFunctions
[j
], 0))
4322 _mesa_set_destroy(resource_set
, NULL
);
4326 * This check is done to make sure we allow only constant expression
4327 * indexing and "constant-index-expression" (indexing with an expression
4328 * that includes loop induction variable).
4331 validate_sampler_array_indexing(struct gl_context
*ctx
,
4332 struct gl_shader_program
*prog
)
4334 dynamic_sampler_array_indexing_visitor v
;
4335 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4336 if (prog
->_LinkedShaders
[i
] == NULL
)
4339 bool no_dynamic_indexing
=
4340 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4342 /* Search for array derefs in shader. */
4343 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4344 if (v
.uses_dynamic_sampler_array_indexing()) {
4345 const char *msg
= "sampler arrays indexed with non-constant "
4346 "expressions is forbidden in GLSL %s %u";
4347 /* Backend has indicated that it has no dynamic indexing support. */
4348 if (no_dynamic_indexing
) {
4349 linker_error(prog
, msg
, prog
->IsES
? "ES" : "",
4350 prog
->data
->Version
);
4353 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "",
4354 prog
->data
->Version
);
4362 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4364 unsigned mask
= prog
->data
->linked_stages
;
4366 const int i
= u_bit_scan(&mask
);
4367 gl_program
*p
= prog
->_LinkedShaders
[i
]->Program
;
4369 p
->sh
.MaxSubroutineFunctionIndex
= 0;
4370 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
4371 ir_function
*fn
= node
->as_function();
4375 if (fn
->is_subroutine
)
4376 p
->sh
.NumSubroutineUniformTypes
++;
4378 if (!fn
->num_subroutine_types
)
4381 /* these should have been calculated earlier. */
4382 assert(fn
->subroutine_index
!= -1);
4383 if (p
->sh
.NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4384 linker_error(prog
, "Too many subroutine functions declared.\n");
4387 p
->sh
.SubroutineFunctions
= reralloc(p
, p
->sh
.SubroutineFunctions
,
4388 struct gl_subroutine_function
,
4389 p
->sh
.NumSubroutineFunctions
+ 1);
4390 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].name
= ralloc_strdup(p
, fn
->name
);
4391 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4392 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
=
4393 ralloc_array(p
, const struct glsl_type
*,
4394 fn
->num_subroutine_types
);
4396 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4399 * "Each subroutine with an index qualifier in the shader must be
4400 * given a unique index, otherwise a compile or link error will be
4403 for (unsigned j
= 0; j
< p
->sh
.NumSubroutineFunctions
; j
++) {
4404 if (p
->sh
.SubroutineFunctions
[j
].index
!= -1 &&
4405 p
->sh
.SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4406 linker_error(prog
, "each subroutine index qualifier in the "
4407 "shader must be unique\n");
4411 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].index
=
4412 fn
->subroutine_index
;
4414 if (fn
->subroutine_index
> (int)p
->sh
.MaxSubroutineFunctionIndex
)
4415 p
->sh
.MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4417 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4418 p
->sh
.SubroutineFunctions
[p
->sh
.NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4419 p
->sh
.NumSubroutineFunctions
++;
4425 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4427 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4429 foreach_in_list(ir_instruction
, node
, ir
) {
4430 ir_variable
*const var
= node
->as_variable();
4432 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4435 /* Don't set always active on builtins that haven't been redeclared */
4436 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4439 var
->data
.always_active_io
= true;
4444 * When separate shader programs are enabled, only input/outputs between
4445 * the stages of a multi-stage separate program can be safely removed
4446 * from the shader interface. Other inputs/outputs must remain active.
4449 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4451 unsigned first
, last
;
4452 assert(prog
->SeparateShader
);
4454 first
= MESA_SHADER_STAGES
;
4457 /* Determine first and last stage. Excluding the compute stage */
4458 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4459 if (!prog
->_LinkedShaders
[i
])
4461 if (first
== MESA_SHADER_STAGES
)
4466 if (first
== MESA_SHADER_STAGES
)
4469 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4470 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4474 if (first
== last
) {
4475 /* For a single shader program only allow inputs to the vertex shader
4476 * and outputs from the fragment shader to be removed.
4478 if (stage
!= MESA_SHADER_VERTEX
)
4479 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4480 if (stage
!= MESA_SHADER_FRAGMENT
)
4481 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4483 /* For multi-stage separate shader programs only allow inputs and
4484 * outputs between the shader stages to be removed as well as inputs
4485 * to the vertex shader and outputs from the fragment shader.
4487 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4488 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4489 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4490 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4496 link_and_validate_uniforms(struct gl_context
*ctx
,
4497 struct gl_shader_program
*prog
)
4499 update_array_sizes(prog
);
4500 link_assign_uniform_locations(prog
, ctx
);
4502 link_assign_atomic_counter_resources(ctx
, prog
);
4503 link_calculate_subroutine_compat(prog
);
4504 check_resources(ctx
, prog
);
4505 check_subroutine_resources(prog
);
4506 check_image_resources(ctx
, prog
);
4507 link_check_atomic_counter_resources(ctx
, prog
);
4511 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4512 struct gl_context
*ctx
,
4513 struct gl_shader_program
*prog
, void *mem_ctx
)
4515 /* Mark all generic shader inputs and outputs as unpaired. */
4516 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4517 if (prog
->_LinkedShaders
[i
] != NULL
) {
4518 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4522 unsigned prev
= first
;
4523 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4524 if (prog
->_LinkedShaders
[i
] == NULL
)
4527 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4528 prog
->_LinkedShaders
[i
]);
4532 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4533 MESA_SHADER_VERTEX
)) {
4537 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4538 MESA_SHADER_FRAGMENT
)) {
4542 prog
->last_vert_prog
= NULL
;
4543 for (int i
= MESA_SHADER_GEOMETRY
; i
>= MESA_SHADER_VERTEX
; i
--) {
4544 if (prog
->_LinkedShaders
[i
] == NULL
)
4547 prog
->last_vert_prog
= prog
->_LinkedShaders
[i
]->Program
;
4551 if (!link_varyings(prog
, first
, last
, ctx
, mem_ctx
))
4554 link_and_validate_uniforms(ctx
, prog
);
4556 if (!prog
->data
->LinkStatus
)
4559 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4560 if (prog
->_LinkedShaders
[i
] == NULL
)
4563 const struct gl_shader_compiler_options
*options
=
4564 &ctx
->Const
.ShaderCompilerOptions
[i
];
4566 if (options
->LowerBufferInterfaceBlocks
)
4567 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4568 options
->ClampBlockIndicesToArrayBounds
);
4570 if (i
== MESA_SHADER_COMPUTE
)
4571 lower_shared_reference(prog
->_LinkedShaders
[i
],
4572 &prog
->Comp
.SharedSize
);
4574 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4575 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4582 linker_optimisation_loop(struct gl_context
*ctx
, exec_list
*ir
,
4585 if (ctx
->Const
.GLSLOptimizeConservatively
) {
4586 /* Run it just once. */
4587 do_common_optimization(ir
, true, false,
4588 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4589 ctx
->Const
.NativeIntegers
);
4591 /* Repeat it until it stops making changes. */
4592 while (do_common_optimization(ir
, true, false,
4593 &ctx
->Const
.ShaderCompilerOptions
[stage
],
4594 ctx
->Const
.NativeIntegers
))
4600 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4602 prog
->data
->LinkStatus
= linking_success
; /* All error paths will set this to false */
4603 prog
->data
->Validated
= false;
4605 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4607 * "Linking can fail for a variety of reasons as specified in the
4608 * OpenGL Shading Language Specification, as well as any of the
4609 * following reasons:
4611 * - No shader objects are attached to program."
4613 * The Compatibility Profile specification does not list the error. In
4614 * Compatibility Profile missing shader stages are replaced by
4615 * fixed-function. This applies to the case where all stages are
4618 if (prog
->NumShaders
== 0) {
4619 if (ctx
->API
!= API_OPENGL_COMPAT
)
4620 linker_error(prog
, "no shaders attached to the program\n");
4624 if (shader_cache_read_program_metadata(ctx
, prog
))
4627 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4629 prog
->ARB_fragment_coord_conventions_enable
= false;
4631 /* Separate the shaders into groups based on their type.
4633 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4634 unsigned num_shaders
[MESA_SHADER_STAGES
];
4636 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4637 shader_list
[i
] = (struct gl_shader
**)
4638 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4642 unsigned min_version
= UINT_MAX
;
4643 unsigned max_version
= 0;
4644 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4645 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4646 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4648 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4649 linker_error(prog
, "all shaders must use same shading "
4650 "language version\n");
4654 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
4655 prog
->ARB_fragment_coord_conventions_enable
= true;
4658 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4659 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4660 num_shaders
[shader_type
]++;
4663 /* In desktop GLSL, different shader versions may be linked together. In
4664 * GLSL ES, all shader versions must be the same.
4666 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4667 linker_error(prog
, "all shaders must use same shading "
4668 "language version\n");
4672 prog
->data
->Version
= max_version
;
4673 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4675 /* Some shaders have to be linked with some other shaders present.
4677 if (!prog
->SeparateShader
) {
4678 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4679 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4680 linker_error(prog
, "Geometry shader must be linked with "
4684 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4685 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4686 linker_error(prog
, "Tessellation evaluation shader must be linked "
4687 "with vertex shader\n");
4690 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4691 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4692 linker_error(prog
, "Tessellation control shader must be linked with "
4697 /* Section 7.3 of the OpenGL ES 3.2 specification says:
4699 * "Linking can fail for [...] any of the following reasons:
4701 * * program contains an object to form a tessellation control
4702 * shader [...] and [...] the program is not separable and
4703 * contains no object to form a tessellation evaluation shader"
4705 * The OpenGL spec is contradictory. It allows linking without a tess
4706 * eval shader, but that can only be used with transform feedback and
4707 * rasterization disabled. However, transform feedback isn't allowed
4708 * with GL_PATCHES, so it can't be used.
4710 * More investigation showed that the idea of transform feedback after
4711 * a tess control shader was dropped, because some hw vendors couldn't
4712 * support tessellation without a tess eval shader, but the linker
4713 * section wasn't updated to reflect that.
4715 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4718 * Do what's reasonable and always require a tess eval shader if a tess
4719 * control shader is present.
4721 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4722 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4723 linker_error(prog
, "Tessellation control shader must be linked with "
4724 "tessellation evaluation shader\n");
4729 /* Compute shaders have additional restrictions. */
4730 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4731 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4732 linker_error(prog
, "Compute shaders may not be linked with any other "
4733 "type of shader\n");
4736 /* Link all shaders for a particular stage and validate the result.
4738 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4739 if (num_shaders
[stage
] > 0) {
4740 gl_linked_shader
*const sh
=
4741 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4742 num_shaders
[stage
], false);
4744 if (!prog
->data
->LinkStatus
) {
4746 _mesa_delete_linked_shader(ctx
, sh
);
4751 case MESA_SHADER_VERTEX
:
4752 validate_vertex_shader_executable(prog
, sh
, ctx
);
4754 case MESA_SHADER_TESS_CTRL
:
4755 /* nothing to be done */
4757 case MESA_SHADER_TESS_EVAL
:
4758 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4760 case MESA_SHADER_GEOMETRY
:
4761 validate_geometry_shader_executable(prog
, sh
, ctx
);
4763 case MESA_SHADER_FRAGMENT
:
4764 validate_fragment_shader_executable(prog
, sh
);
4767 if (!prog
->data
->LinkStatus
) {
4769 _mesa_delete_linked_shader(ctx
, sh
);
4773 prog
->_LinkedShaders
[stage
] = sh
;
4774 prog
->data
->linked_stages
|= 1 << stage
;
4778 /* Here begins the inter-stage linking phase. Some initial validation is
4779 * performed, then locations are assigned for uniforms, attributes, and
4782 cross_validate_uniforms(prog
);
4783 if (!prog
->data
->LinkStatus
)
4786 unsigned first
, last
, prev
;
4788 first
= MESA_SHADER_STAGES
;
4791 /* Determine first and last stage. */
4792 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4793 if (!prog
->_LinkedShaders
[i
])
4795 if (first
== MESA_SHADER_STAGES
)
4800 check_explicit_uniform_locations(ctx
, prog
);
4801 link_assign_subroutine_types(prog
);
4803 if (!prog
->data
->LinkStatus
)
4806 resize_tes_inputs(ctx
, prog
);
4808 /* Validate the inputs of each stage with the output of the preceding
4812 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4813 if (prog
->_LinkedShaders
[i
] == NULL
)
4816 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4817 prog
->_LinkedShaders
[i
]);
4818 if (!prog
->data
->LinkStatus
)
4821 cross_validate_outputs_to_inputs(prog
,
4822 prog
->_LinkedShaders
[prev
],
4823 prog
->_LinkedShaders
[i
]);
4824 if (!prog
->data
->LinkStatus
)
4830 /* Cross-validate uniform blocks between shader stages */
4831 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4832 if (!prog
->data
->LinkStatus
)
4835 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4836 if (prog
->_LinkedShaders
[i
] != NULL
)
4837 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4840 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4841 * it before optimization because we want most of the checks to get
4842 * dropped thanks to constant propagation.
4844 * This rule also applies to GLSL ES 3.00.
4846 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4847 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4849 lower_discard_flow(sh
->ir
);
4853 if (prog
->SeparateShader
)
4854 disable_varying_optimizations_for_sso(prog
);
4857 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4861 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4864 /* Do common optimization before assigning storage for attributes,
4865 * uniforms, and varyings. Later optimization could possibly make
4866 * some of that unused.
4868 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4869 if (prog
->_LinkedShaders
[i
] == NULL
)
4872 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4873 if (!prog
->data
->LinkStatus
)
4876 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4877 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4880 if (ctx
->Const
.LowerTessLevel
) {
4881 lower_tess_level(prog
->_LinkedShaders
[i
]);
4884 /* Call opts before lowering const arrays to uniforms so we can const
4885 * propagate any elements accessed directly.
4887 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
4889 /* Call opts after lowering const arrays to copy propagate things. */
4890 if (lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
))
4891 linker_optimisation_loop(ctx
, prog
->_LinkedShaders
[i
]->ir
, i
);
4893 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4896 /* Validation for special cases where we allow sampler array indexing
4897 * with loop induction variable. This check emits a warning or error
4898 * depending if backend can handle dynamic indexing.
4900 if ((!prog
->IsES
&& prog
->data
->Version
< 130) ||
4901 (prog
->IsES
&& prog
->data
->Version
< 300)) {
4902 if (!validate_sampler_array_indexing(ctx
, prog
))
4906 /* Check and validate stream emissions in geometry shaders */
4907 validate_geometry_shader_emissions(ctx
, prog
);
4909 store_fragdepth_layout(prog
);
4911 if(!link_varyings_and_uniforms(first
, last
, ctx
, prog
, mem_ctx
))
4914 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4915 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4916 * anything about shader linking when one of the shaders (vertex or
4917 * fragment shader) is absent. So, the extension shouldn't change the
4918 * behavior specified in GLSL specification.
4920 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4921 * "Linking can fail for a variety of reasons as specified in the
4922 * OpenGL ES Shading Language Specification, as well as any of the
4923 * following reasons:
4927 * * program contains objects to form either a vertex shader or
4928 * fragment shader, and program is not separable, and does not
4929 * contain objects to form both a vertex shader and fragment
4932 * However, the only scenario in 3.1+ where we don't require them both is
4933 * when we have a compute shader. For example:
4935 * - No shaders is a link error.
4936 * - Geom or Tess without a Vertex shader is a link error which means we
4937 * always require a Vertex shader and hence a Fragment shader.
4938 * - Finally a Compute shader linked with any other stage is a link error.
4940 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4941 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4942 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4943 linker_error(prog
, "program lacks a vertex shader\n");
4944 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4945 linker_error(prog
, "program lacks a fragment shader\n");
4950 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4951 free(shader_list
[i
]);
4952 if (prog
->_LinkedShaders
[i
] == NULL
)
4955 /* Do a final validation step to make sure that the IR wasn't
4956 * invalidated by any modifications performed after intrastage linking.
4958 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4960 /* Retain any live IR, but trash the rest. */
4961 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4963 /* The symbol table in the linked shaders may contain references to
4964 * variables that were removed (e.g., unused uniforms). Since it may
4965 * contain junk, there is no possible valid use. Delete it and set the
4968 delete prog
->_LinkedShaders
[i
]->symbols
;
4969 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4972 ralloc_free(mem_ctx
);