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>
67 #include "main/core.h"
68 #include "glsl_symbol_table.h"
69 #include "glsl_parser_extras.h"
72 #include "program/hash_table.h"
74 #include "link_varyings.h"
75 #include "ir_optimization.h"
76 #include "ir_rvalue_visitor.h"
77 #include "ir_uniform.h"
80 #include "main/shaderobj.h"
81 #include "main/enums.h"
84 void linker_error(gl_shader_program
*, const char *, ...);
89 * Visitor that determines whether or not a variable is ever written.
91 class find_assignment_visitor
: public ir_hierarchical_visitor
{
93 find_assignment_visitor(const char *name
)
94 : name(name
), found(false)
99 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
101 ir_variable
*const var
= ir
->lhs
->variable_referenced();
103 if (strcmp(name
, var
->name
) == 0) {
108 return visit_continue_with_parent
;
111 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
113 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
114 actual_node
, &ir
->actual_parameters
) {
115 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
116 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
118 if (sig_param
->data
.mode
== ir_var_function_out
||
119 sig_param
->data
.mode
== ir_var_function_inout
) {
120 ir_variable
*var
= param_rval
->variable_referenced();
121 if (var
&& strcmp(name
, var
->name
) == 0) {
128 if (ir
->return_deref
!= NULL
) {
129 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
131 if (strcmp(name
, var
->name
) == 0) {
137 return visit_continue_with_parent
;
140 bool variable_found()
146 const char *name
; /**< Find writes to a variable with this name. */
147 bool found
; /**< Was a write to the variable found? */
152 * Visitor that determines whether or not a variable is ever read.
154 class find_deref_visitor
: public ir_hierarchical_visitor
{
156 find_deref_visitor(const char *name
)
157 : name(name
), found(false)
162 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
164 if (strcmp(this->name
, ir
->var
->name
) == 0) {
169 return visit_continue
;
172 bool variable_found() const
178 const char *name
; /**< Find writes to a variable with this name. */
179 bool found
; /**< Was a write to the variable found? */
183 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
185 unsigned num_vertices
;
186 gl_shader_program
*prog
;
188 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
190 this->num_vertices
= num_vertices
;
194 virtual ~geom_array_resize_visitor()
199 virtual ir_visitor_status
visit(ir_variable
*var
)
201 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
202 return visit_continue
;
204 unsigned size
= var
->type
->length
;
206 /* Generate a link error if the shader has declared this array with an
209 if (size
&& size
!= this->num_vertices
) {
210 linker_error(this->prog
, "size of array %s declared as %u, "
211 "but number of input vertices is %u\n",
212 var
->name
, size
, this->num_vertices
);
213 return visit_continue
;
216 /* Generate a link error if the shader attempts to access an input
217 * array using an index too large for its actual size assigned at link
220 if (var
->data
.max_array_access
>= this->num_vertices
) {
221 linker_error(this->prog
, "geometry shader accesses element %i of "
222 "%s, but only %i input vertices\n",
223 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
224 return visit_continue
;
227 var
->type
= glsl_type::get_array_instance(var
->type
->element_type(),
229 var
->data
.max_array_access
= this->num_vertices
- 1;
231 return visit_continue
;
234 /* Dereferences of input variables need to be updated so that their type
235 * matches the newly assigned type of the variable they are accessing. */
236 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
238 ir
->type
= ir
->var
->type
;
239 return visit_continue
;
242 /* Dereferences of 2D input arrays need to be updated so that their type
243 * matches the newly assigned type of the array they are accessing. */
244 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
246 const glsl_type
*const vt
= ir
->array
->type
;
248 ir
->type
= vt
->element_type();
249 return visit_continue
;
254 * Visitor that determines the highest stream id to which a (geometry) shader
255 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
257 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
259 find_emit_vertex_visitor(int max_allowed
)
260 : max_stream_allowed(max_allowed
),
261 invalid_stream_id(0),
262 invalid_stream_id_from_emit_vertex(false),
263 end_primitive_found(false),
264 uses_non_zero_stream(false)
269 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
271 int stream_id
= ir
->stream_id();
274 invalid_stream_id
= stream_id
;
275 invalid_stream_id_from_emit_vertex
= true;
279 if (stream_id
> max_stream_allowed
) {
280 invalid_stream_id
= stream_id
;
281 invalid_stream_id_from_emit_vertex
= true;
286 uses_non_zero_stream
= true;
288 return visit_continue
;
291 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
293 end_primitive_found
= true;
295 int stream_id
= ir
->stream_id();
298 invalid_stream_id
= stream_id
;
299 invalid_stream_id_from_emit_vertex
= false;
303 if (stream_id
> max_stream_allowed
) {
304 invalid_stream_id
= stream_id
;
305 invalid_stream_id_from_emit_vertex
= false;
310 uses_non_zero_stream
= true;
312 return visit_continue
;
317 return invalid_stream_id
!= 0;
320 const char *error_func()
322 return invalid_stream_id_from_emit_vertex
?
323 "EmitStreamVertex" : "EndStreamPrimitive";
328 return invalid_stream_id
;
333 return uses_non_zero_stream
;
336 bool uses_end_primitive()
338 return end_primitive_found
;
342 int max_stream_allowed
;
343 int invalid_stream_id
;
344 bool invalid_stream_id_from_emit_vertex
;
345 bool end_primitive_found
;
346 bool uses_non_zero_stream
;
349 } /* anonymous namespace */
352 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
356 ralloc_strcat(&prog
->InfoLog
, "error: ");
358 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
361 prog
->LinkStatus
= false;
366 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
370 ralloc_strcat(&prog
->InfoLog
, "warning: ");
372 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
379 * Given a string identifying a program resource, break it into a base name
380 * and an optional array index in square brackets.
382 * If an array index is present, \c out_base_name_end is set to point to the
383 * "[" that precedes the array index, and the array index itself is returned
386 * If no array index is present (or if the array index is negative or
387 * mal-formed), \c out_base_name_end, is set to point to the null terminator
388 * at the end of the input string, and -1 is returned.
390 * Only the final array index is parsed; if the string contains other array
391 * indices (or structure field accesses), they are left in the base name.
393 * No attempt is made to check that the base name is properly formed;
394 * typically the caller will look up the base name in a hash table, so
395 * ill-formed base names simply turn into hash table lookup failures.
398 parse_program_resource_name(const GLchar
*name
,
399 const GLchar
**out_base_name_end
)
401 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
403 * "When an integer array element or block instance number is part of
404 * the name string, it will be specified in decimal form without a "+"
405 * or "-" sign or any extra leading zeroes. Additionally, the name
406 * string will not include white space anywhere in the string."
409 const size_t len
= strlen(name
);
410 *out_base_name_end
= name
+ len
;
412 if (len
== 0 || name
[len
-1] != ']')
415 /* Walk backwards over the string looking for a non-digit character. This
416 * had better be the opening bracket for an array index.
418 * Initially, i specifies the location of the ']'. Since the string may
419 * contain only the ']' charcater, walk backwards very carefully.
422 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
425 if ((i
== 0) || name
[i
-1] != '[')
428 long array_index
= strtol(&name
[i
], NULL
, 10);
432 *out_base_name_end
= name
+ (i
- 1);
438 link_invalidate_variable_locations(exec_list
*ir
)
440 foreach_in_list(ir_instruction
, node
, ir
) {
441 ir_variable
*const var
= node
->as_variable();
446 /* Only assign locations for variables that lack an explicit location.
447 * Explicit locations are set for all built-in variables, generic vertex
448 * shader inputs (via layout(location=...)), and generic fragment shader
449 * outputs (also via layout(location=...)).
451 if (!var
->data
.explicit_location
) {
452 var
->data
.location
= -1;
453 var
->data
.location_frac
= 0;
456 /* ir_variable::is_unmatched_generic_inout is used by the linker while
457 * connecting outputs from one stage to inputs of the next stage.
459 * There are two implicit assumptions here. First, we assume that any
460 * built-in variable (i.e., non-generic in or out) will have
461 * explicit_location set. Second, we assume that any generic in or out
462 * will not have explicit_location set.
464 * This second assumption will only be valid until
465 * GL_ARB_separate_shader_objects is supported. When that extension is
466 * implemented, this function will need some modifications.
468 if (!var
->data
.explicit_location
) {
469 var
->data
.is_unmatched_generic_inout
= 1;
471 var
->data
.is_unmatched_generic_inout
= 0;
478 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
480 * Also check for errors based on incorrect usage of gl_ClipVertex and
483 * Return false if an error was reported.
486 analyze_clip_usage(struct gl_shader_program
*prog
,
487 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
488 GLuint
*ClipDistanceArraySize
)
490 *ClipDistanceArraySize
= 0;
492 if (!prog
->IsES
&& prog
->Version
>= 130) {
493 /* From section 7.1 (Vertex Shader Special Variables) of the
496 * "It is an error for a shader to statically write both
497 * gl_ClipVertex and gl_ClipDistance."
499 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
500 * gl_ClipVertex nor gl_ClipDistance.
502 find_assignment_visitor
clip_vertex("gl_ClipVertex");
503 find_assignment_visitor
clip_distance("gl_ClipDistance");
505 clip_vertex
.run(shader
->ir
);
506 clip_distance
.run(shader
->ir
);
507 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
508 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
509 "and `gl_ClipDistance'\n",
510 _mesa_shader_stage_to_string(shader
->Stage
));
513 *UsesClipDistance
= clip_distance
.variable_found();
514 ir_variable
*clip_distance_var
=
515 shader
->symbols
->get_variable("gl_ClipDistance");
516 if (clip_distance_var
)
517 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
519 *UsesClipDistance
= false;
525 * Verify that a vertex shader executable meets all semantic requirements.
527 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
530 * \param shader Vertex shader executable to be verified
533 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
534 struct gl_shader
*shader
)
539 /* From the GLSL 1.10 spec, page 48:
541 * "The variable gl_Position is available only in the vertex
542 * language and is intended for writing the homogeneous vertex
543 * position. All executions of a well-formed vertex shader
544 * executable must write a value into this variable. [...] The
545 * variable gl_Position is available only in the vertex
546 * language and is intended for writing the homogeneous vertex
547 * position. All executions of a well-formed vertex shader
548 * executable must write a value into this variable."
550 * while in GLSL 1.40 this text is changed to:
552 * "The variable gl_Position is available only in the vertex
553 * language and is intended for writing the homogeneous vertex
554 * position. It can be written at any time during shader
555 * execution. It may also be read back by a vertex shader
556 * after being written. This value will be used by primitive
557 * assembly, clipping, culling, and other fixed functionality
558 * operations, if present, that operate on primitives after
559 * vertex processing has occurred. Its value is undefined if
560 * the vertex shader executable does not write gl_Position."
562 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
563 * gl_Position is not an error.
565 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
566 find_assignment_visitor
find("gl_Position");
567 find
.run(shader
->ir
);
568 if (!find
.variable_found()) {
571 "vertex shader does not write to `gl_Position'."
572 "It's value is undefined. \n");
575 "vertex shader does not write to `gl_Position'. \n");
581 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
582 &prog
->Vert
.ClipDistanceArraySize
);
587 * Verify that a fragment shader executable meets all semantic requirements
589 * \param shader Fragment shader executable to be verified
592 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
593 struct gl_shader
*shader
)
598 find_assignment_visitor
frag_color("gl_FragColor");
599 find_assignment_visitor
frag_data("gl_FragData");
601 frag_color
.run(shader
->ir
);
602 frag_data
.run(shader
->ir
);
604 if (frag_color
.variable_found() && frag_data
.variable_found()) {
605 linker_error(prog
, "fragment shader writes to both "
606 "`gl_FragColor' and `gl_FragData'\n");
611 * Verify that a geometry shader executable meets all semantic requirements
613 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
614 * prog->Geom.ClipDistanceArraySize as a side effect.
616 * \param shader Geometry shader executable to be verified
619 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
620 struct gl_shader
*shader
)
625 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
626 prog
->Geom
.VerticesIn
= num_vertices
;
628 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
629 &prog
->Geom
.ClipDistanceArraySize
);
633 * Check if geometry shaders emit to non-zero streams and do corresponding
637 validate_geometry_shader_emissions(struct gl_context
*ctx
,
638 struct gl_shader_program
*prog
)
640 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
641 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
642 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
643 if (emit_vertex
.error()) {
644 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
645 "stream parameter are in the range [0, %d].",
646 emit_vertex
.error_func(),
647 emit_vertex
.error_stream(),
648 ctx
->Const
.MaxVertexStreams
- 1);
650 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
651 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
653 /* From the ARB_gpu_shader5 spec:
655 * "Multiple vertex streams are supported only if the output primitive
656 * type is declared to be "points". A program will fail to link if it
657 * contains a geometry shader calling EmitStreamVertex() or
658 * EndStreamPrimitive() if its output primitive type is not "points".
660 * However, in the same spec:
662 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
663 * with <stream> set to zero."
667 * "The function EndPrimitive() is equivalent to calling
668 * EndStreamPrimitive() with <stream> set to zero."
670 * Since we can call EmitVertex() and EndPrimitive() when we output
671 * primitives other than points, calling EmitStreamVertex(0) or
672 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
673 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
674 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
677 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
678 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
679 "with n>0 requires point output");
686 * Perform validation of global variables used across multiple shaders
689 cross_validate_globals(struct gl_shader_program
*prog
,
690 struct gl_shader
**shader_list
,
691 unsigned num_shaders
,
694 /* Examine all of the uniforms in all of the shaders and cross validate
697 glsl_symbol_table variables
;
698 for (unsigned i
= 0; i
< num_shaders
; i
++) {
699 if (shader_list
[i
] == NULL
)
702 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
703 ir_variable
*const var
= node
->as_variable();
708 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
))
711 /* Don't cross validate temporaries that are at global scope. These
712 * will eventually get pulled into the shaders 'main'.
714 if (var
->data
.mode
== ir_var_temporary
)
717 /* If a global with this name has already been seen, verify that the
718 * new instance has the same type. In addition, if the globals have
719 * initializers, the values of the initializers must be the same.
721 ir_variable
*const existing
= variables
.get_variable(var
->name
);
722 if (existing
!= NULL
) {
723 if (var
->type
!= existing
->type
) {
724 /* Consider the types to be "the same" if both types are arrays
725 * of the same type and one of the arrays is implicitly sized.
726 * In addition, set the type of the linked variable to the
727 * explicitly sized array.
729 if (var
->type
->is_array()
730 && existing
->type
->is_array()
731 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
732 && ((var
->type
->length
== 0)
733 || (existing
->type
->length
== 0))) {
734 if (var
->type
->length
!= 0) {
735 existing
->type
= var
->type
;
737 } else if (var
->type
->is_record()
738 && existing
->type
->is_record()
739 && existing
->type
->record_compare(var
->type
)) {
740 existing
->type
= var
->type
;
742 linker_error(prog
, "%s `%s' declared as type "
743 "`%s' and type `%s'\n",
745 var
->name
, var
->type
->name
,
746 existing
->type
->name
);
751 if (var
->data
.explicit_location
) {
752 if (existing
->data
.explicit_location
753 && (var
->data
.location
!= existing
->data
.location
)) {
754 linker_error(prog
, "explicit locations for %s "
755 "`%s' have differing values\n",
756 mode_string(var
), var
->name
);
760 existing
->data
.location
= var
->data
.location
;
761 existing
->data
.explicit_location
= true;
764 /* From the GLSL 4.20 specification:
765 * "A link error will result if two compilation units in a program
766 * specify different integer-constant bindings for the same
767 * opaque-uniform name. However, it is not an error to specify a
768 * binding on some but not all declarations for the same name"
770 if (var
->data
.explicit_binding
) {
771 if (existing
->data
.explicit_binding
&&
772 var
->data
.binding
!= existing
->data
.binding
) {
773 linker_error(prog
, "explicit bindings for %s "
774 "`%s' have differing values\n",
775 mode_string(var
), var
->name
);
779 existing
->data
.binding
= var
->data
.binding
;
780 existing
->data
.explicit_binding
= true;
783 if (var
->type
->contains_atomic() &&
784 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
785 linker_error(prog
, "offset specifications for %s "
786 "`%s' have differing values\n",
787 mode_string(var
), var
->name
);
791 /* Validate layout qualifiers for gl_FragDepth.
793 * From the AMD/ARB_conservative_depth specs:
795 * "If gl_FragDepth is redeclared in any fragment shader in a
796 * program, it must be redeclared in all fragment shaders in
797 * that program that have static assignments to
798 * gl_FragDepth. All redeclarations of gl_FragDepth in all
799 * fragment shaders in a single program must have the same set
802 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
803 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
804 bool layout_differs
=
805 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
807 if (layout_declared
&& layout_differs
) {
809 "All redeclarations of gl_FragDepth in all "
810 "fragment shaders in a single program must have "
811 "the same set of qualifiers.");
814 if (var
->data
.used
&& layout_differs
) {
816 "If gl_FragDepth is redeclared with a layout "
817 "qualifier in any fragment shader, it must be "
818 "redeclared with the same layout qualifier in "
819 "all fragment shaders that have assignments to "
824 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
826 * "If a shared global has multiple initializers, the
827 * initializers must all be constant expressions, and they
828 * must all have the same value. Otherwise, a link error will
829 * result. (A shared global having only one initializer does
830 * not require that initializer to be a constant expression.)"
832 * Previous to 4.20 the GLSL spec simply said that initializers
833 * must have the same value. In this case of non-constant
834 * initializers, this was impossible to determine. As a result,
835 * no vendor actually implemented that behavior. The 4.20
836 * behavior matches the implemented behavior of at least one other
837 * vendor, so we'll implement that for all GLSL versions.
839 if (var
->constant_initializer
!= NULL
) {
840 if (existing
->constant_initializer
!= NULL
) {
841 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
842 linker_error(prog
, "initializers for %s "
843 "`%s' have differing values\n",
844 mode_string(var
), var
->name
);
848 /* If the first-seen instance of a particular uniform did not
849 * have an initializer but a later instance does, copy the
850 * initializer to the version stored in the symbol table.
852 /* FINISHME: This is wrong. The constant_value field should
853 * FINISHME: not be modified! Imagine a case where a shader
854 * FINISHME: without an initializer is linked in two different
855 * FINISHME: programs with shaders that have differing
856 * FINISHME: initializers. Linking with the first will
857 * FINISHME: modify the shader, and linking with the second
858 * FINISHME: will fail.
860 existing
->constant_initializer
=
861 var
->constant_initializer
->clone(ralloc_parent(existing
),
866 if (var
->data
.has_initializer
) {
867 if (existing
->data
.has_initializer
868 && (var
->constant_initializer
== NULL
869 || existing
->constant_initializer
== NULL
)) {
871 "shared global variable `%s' has multiple "
872 "non-constant initializers.\n",
877 /* Some instance had an initializer, so keep track of that. In
878 * this location, all sorts of initializers (constant or
879 * otherwise) will propagate the existence to the variable
880 * stored in the symbol table.
882 existing
->data
.has_initializer
= true;
885 if (existing
->data
.invariant
!= var
->data
.invariant
) {
886 linker_error(prog
, "declarations for %s `%s' have "
887 "mismatching invariant qualifiers\n",
888 mode_string(var
), var
->name
);
891 if (existing
->data
.centroid
!= var
->data
.centroid
) {
892 linker_error(prog
, "declarations for %s `%s' have "
893 "mismatching centroid qualifiers\n",
894 mode_string(var
), var
->name
);
897 if (existing
->data
.sample
!= var
->data
.sample
) {
898 linker_error(prog
, "declarations for %s `%s` have "
899 "mismatching sample qualifiers\n",
900 mode_string(var
), var
->name
);
904 variables
.add_variable(var
);
911 * Perform validation of uniforms used across multiple shader stages
914 cross_validate_uniforms(struct gl_shader_program
*prog
)
916 cross_validate_globals(prog
, prog
->_LinkedShaders
,
917 MESA_SHADER_STAGES
, true);
921 * Accumulates the array of prog->UniformBlocks and checks that all
922 * definitons of blocks agree on their contents.
925 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
927 unsigned max_num_uniform_blocks
= 0;
928 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
929 if (prog
->_LinkedShaders
[i
])
930 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
933 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
934 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
936 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
937 max_num_uniform_blocks
);
938 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
939 prog
->UniformBlockStageIndex
[i
][j
] = -1;
944 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
945 int index
= link_cross_validate_uniform_block(prog
,
946 &prog
->UniformBlocks
,
947 &prog
->NumUniformBlocks
,
948 &sh
->UniformBlocks
[j
]);
951 linker_error(prog
, "uniform block `%s' has mismatching definitions",
952 sh
->UniformBlocks
[j
].Name
);
956 prog
->UniformBlockStageIndex
[i
][index
] = j
;
965 * Populates a shaders symbol table with all global declarations
968 populate_symbol_table(gl_shader
*sh
)
970 sh
->symbols
= new(sh
) glsl_symbol_table
;
972 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
976 if ((func
= inst
->as_function()) != NULL
) {
977 sh
->symbols
->add_function(func
);
978 } else if ((var
= inst
->as_variable()) != NULL
) {
979 sh
->symbols
->add_variable(var
);
986 * Remap variables referenced in an instruction tree
988 * This is used when instruction trees are cloned from one shader and placed in
989 * another. These trees will contain references to \c ir_variable nodes that
990 * do not exist in the target shader. This function finds these \c ir_variable
991 * references and replaces the references with matching variables in the target
994 * If there is no matching variable in the target shader, a clone of the
995 * \c ir_variable is made and added to the target shader. The new variable is
996 * added to \b both the instruction stream and the symbol table.
998 * \param inst IR tree that is to be processed.
999 * \param symbols Symbol table containing global scope symbols in the
1001 * \param instructions Instruction stream where new variable declarations
1005 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1008 class remap_visitor
: public ir_hierarchical_visitor
{
1010 remap_visitor(struct gl_shader
*target
,
1013 this->target
= target
;
1014 this->symbols
= target
->symbols
;
1015 this->instructions
= target
->ir
;
1016 this->temps
= temps
;
1019 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1021 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1022 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1024 assert(var
!= NULL
);
1026 return visit_continue
;
1029 ir_variable
*const existing
=
1030 this->symbols
->get_variable(ir
->var
->name
);
1031 if (existing
!= NULL
)
1034 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1036 this->symbols
->add_variable(copy
);
1037 this->instructions
->push_head(copy
);
1041 return visit_continue
;
1045 struct gl_shader
*target
;
1046 glsl_symbol_table
*symbols
;
1047 exec_list
*instructions
;
1051 remap_visitor
v(target
, temps
);
1058 * Move non-declarations from one instruction stream to another
1060 * The intended usage pattern of this function is to pass the pointer to the
1061 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1062 * pointer) for \c last and \c false for \c make_copies on the first
1063 * call. Successive calls pass the return value of the previous call for
1064 * \c last and \c true for \c make_copies.
1066 * \param instructions Source instruction stream
1067 * \param last Instruction after which new instructions should be
1068 * inserted in the target instruction stream
1069 * \param make_copies Flag selecting whether instructions in \c instructions
1070 * should be copied (via \c ir_instruction::clone) into the
1071 * target list or moved.
1074 * The new "last" instruction in the target instruction stream. This pointer
1075 * is suitable for use as the \c last parameter of a later call to this
1079 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1080 bool make_copies
, gl_shader
*target
)
1082 hash_table
*temps
= NULL
;
1085 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1086 hash_table_pointer_compare
);
1088 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1089 if (inst
->as_function())
1092 ir_variable
*var
= inst
->as_variable();
1093 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1096 assert(inst
->as_assignment()
1098 || inst
->as_if() /* for initializers with the ?: operator */
1099 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1102 inst
= inst
->clone(target
, NULL
);
1105 hash_table_insert(temps
, inst
, var
);
1107 remap_variables(inst
, target
, temps
);
1112 last
->insert_after(inst
);
1117 hash_table_dtor(temps
);
1123 * Get the function signature for main from a shader
1125 ir_function_signature
*
1126 link_get_main_function_signature(gl_shader
*sh
)
1128 ir_function
*const f
= sh
->symbols
->get_function("main");
1130 exec_list void_parameters
;
1132 /* Look for the 'void main()' signature and ensure that it's defined.
1133 * This keeps the linker from accidentally pick a shader that just
1134 * contains a prototype for main.
1136 * We don't have to check for multiple definitions of main (in multiple
1137 * shaders) because that would have already been caught above.
1139 ir_function_signature
*sig
=
1140 f
->matching_signature(NULL
, &void_parameters
, false);
1141 if ((sig
!= NULL
) && sig
->is_defined
) {
1151 * This class is only used in link_intrastage_shaders() below but declaring
1152 * it inside that function leads to compiler warnings with some versions of
1155 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1157 array_sizing_visitor()
1158 : mem_ctx(ralloc_context(NULL
)),
1159 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1160 hash_table_pointer_compare
))
1164 ~array_sizing_visitor()
1166 hash_table_dtor(this->unnamed_interfaces
);
1167 ralloc_free(this->mem_ctx
);
1170 virtual ir_visitor_status
visit(ir_variable
*var
)
1172 fixup_type(&var
->type
, var
->data
.max_array_access
);
1173 if (var
->type
->is_interface()) {
1174 if (interface_contains_unsized_arrays(var
->type
)) {
1175 const glsl_type
*new_type
=
1176 resize_interface_members(var
->type
,
1177 var
->get_max_ifc_array_access());
1178 var
->type
= new_type
;
1179 var
->change_interface_type(new_type
);
1181 } else if (var
->type
->is_array() &&
1182 var
->type
->fields
.array
->is_interface()) {
1183 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1184 const glsl_type
*new_type
=
1185 resize_interface_members(var
->type
->fields
.array
,
1186 var
->get_max_ifc_array_access());
1187 var
->change_interface_type(new_type
);
1189 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1191 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1192 /* Store a pointer to the variable in the unnamed_interfaces
1195 ir_variable
**interface_vars
= (ir_variable
**)
1196 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1197 if (interface_vars
== NULL
) {
1198 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1200 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1203 unsigned index
= ifc_type
->field_index(var
->name
);
1204 assert(index
< ifc_type
->length
);
1205 assert(interface_vars
[index
] == NULL
);
1206 interface_vars
[index
] = var
;
1208 return visit_continue
;
1212 * For each unnamed interface block that was discovered while running the
1213 * visitor, adjust the interface type to reflect the newly assigned array
1214 * sizes, and fix up the ir_variable nodes to point to the new interface
1217 void fixup_unnamed_interface_types()
1219 hash_table_call_foreach(this->unnamed_interfaces
,
1220 fixup_unnamed_interface_type
, NULL
);
1225 * If the type pointed to by \c type represents an unsized array, replace
1226 * it with a sized array whose size is determined by max_array_access.
1228 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1230 if ((*type
)->is_unsized_array()) {
1231 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1232 max_array_access
+ 1);
1233 assert(*type
!= NULL
);
1238 * Determine whether the given interface type contains unsized arrays (if
1239 * it doesn't, array_sizing_visitor doesn't need to process it).
1241 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1243 for (unsigned i
= 0; i
< type
->length
; i
++) {
1244 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1245 if (elem_type
->is_unsized_array())
1252 * Create a new interface type based on the given type, with unsized arrays
1253 * replaced by sized arrays whose size is determined by
1254 * max_ifc_array_access.
1256 static const glsl_type
*
1257 resize_interface_members(const glsl_type
*type
,
1258 const unsigned *max_ifc_array_access
)
1260 unsigned num_fields
= type
->length
;
1261 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1262 memcpy(fields
, type
->fields
.structure
,
1263 num_fields
* sizeof(*fields
));
1264 for (unsigned i
= 0; i
< num_fields
; i
++) {
1265 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1267 glsl_interface_packing packing
=
1268 (glsl_interface_packing
) type
->interface_packing
;
1269 const glsl_type
*new_ifc_type
=
1270 glsl_type::get_interface_instance(fields
, num_fields
,
1271 packing
, type
->name
);
1273 return new_ifc_type
;
1276 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1279 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1280 ir_variable
**interface_vars
= (ir_variable
**) data
;
1281 unsigned num_fields
= ifc_type
->length
;
1282 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1283 memcpy(fields
, ifc_type
->fields
.structure
,
1284 num_fields
* sizeof(*fields
));
1285 bool interface_type_changed
= false;
1286 for (unsigned i
= 0; i
< num_fields
; i
++) {
1287 if (interface_vars
[i
] != NULL
&&
1288 fields
[i
].type
!= interface_vars
[i
]->type
) {
1289 fields
[i
].type
= interface_vars
[i
]->type
;
1290 interface_type_changed
= true;
1293 if (!interface_type_changed
) {
1297 glsl_interface_packing packing
=
1298 (glsl_interface_packing
) ifc_type
->interface_packing
;
1299 const glsl_type
*new_ifc_type
=
1300 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1303 for (unsigned i
= 0; i
< num_fields
; i
++) {
1304 if (interface_vars
[i
] != NULL
)
1305 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1310 * Memory context used to allocate the data in \c unnamed_interfaces.
1315 * Hash table from const glsl_type * to an array of ir_variable *'s
1316 * pointing to the ir_variables constituting each unnamed interface block.
1318 hash_table
*unnamed_interfaces
;
1322 * Performs the cross-validation of layout qualifiers specified in
1323 * redeclaration of gl_FragCoord for the attached fragment shaders,
1324 * and propagates them to the linked FS and linked shader program.
1327 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1328 struct gl_shader
*linked_shader
,
1329 struct gl_shader
**shader_list
,
1330 unsigned num_shaders
)
1332 linked_shader
->redeclares_gl_fragcoord
= false;
1333 linked_shader
->uses_gl_fragcoord
= false;
1334 linked_shader
->origin_upper_left
= false;
1335 linked_shader
->pixel_center_integer
= false;
1337 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1338 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1341 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1342 struct gl_shader
*shader
= shader_list
[i
];
1343 /* From the GLSL 1.50 spec, page 39:
1345 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1346 * it must be redeclared in all the fragment shaders in that program
1347 * that have a static use gl_FragCoord."
1349 * Exclude the case when one of the 'linked_shader' or 'shader' redeclares
1350 * gl_FragCoord with no layout qualifiers but the other one doesn't
1351 * redeclare it. If we strictly follow GLSL 1.50 spec's language, it
1352 * should be a link error. But, generating link error for this case will
1353 * be a wrong behaviour which spec didn't intend to do and it could also
1354 * break some applications.
1356 if ((linked_shader
->redeclares_gl_fragcoord
1357 && !shader
->redeclares_gl_fragcoord
1358 && shader
->uses_gl_fragcoord
1359 && (linked_shader
->origin_upper_left
1360 || linked_shader
->pixel_center_integer
))
1361 || (shader
->redeclares_gl_fragcoord
1362 && !linked_shader
->redeclares_gl_fragcoord
1363 && linked_shader
->uses_gl_fragcoord
1364 && (shader
->origin_upper_left
1365 || shader
->pixel_center_integer
))) {
1366 linker_error(prog
, "fragment shader defined with conflicting "
1367 "layout qualifiers for gl_FragCoord\n");
1370 /* From the GLSL 1.50 spec, page 39:
1372 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1373 * single program must have the same set of qualifiers."
1375 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1376 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1377 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1378 linker_error(prog
, "fragment shader defined with conflicting "
1379 "layout qualifiers for gl_FragCoord\n");
1382 /* Update the linked shader state. Note that uses_gl_fragcoord should
1383 * accumulate the results. The other values should replace. If there
1384 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1385 * are already known to be the same.
1387 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1388 linked_shader
->redeclares_gl_fragcoord
=
1389 shader
->redeclares_gl_fragcoord
;
1390 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1391 || shader
->uses_gl_fragcoord
;
1392 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1393 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1399 * Performs the cross-validation of geometry shader max_vertices and
1400 * primitive type layout qualifiers for the attached geometry shaders,
1401 * and propagates them to the linked GS and linked shader program.
1404 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1405 struct gl_shader
*linked_shader
,
1406 struct gl_shader
**shader_list
,
1407 unsigned num_shaders
)
1409 linked_shader
->Geom
.VerticesOut
= 0;
1410 linked_shader
->Geom
.Invocations
= 0;
1411 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1412 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1414 /* No in/out qualifiers defined for anything but GLSL 1.50+
1415 * geometry shaders so far.
1417 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1420 /* From the GLSL 1.50 spec, page 46:
1422 * "All geometry shader output layout declarations in a program
1423 * must declare the same layout and same value for
1424 * max_vertices. There must be at least one geometry output
1425 * layout declaration somewhere in a program, but not all
1426 * geometry shaders (compilation units) are required to
1430 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1431 struct gl_shader
*shader
= shader_list
[i
];
1433 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1434 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1435 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1436 linker_error(prog
, "geometry shader defined with conflicting "
1440 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1443 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1444 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1445 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1446 linker_error(prog
, "geometry shader defined with conflicting "
1450 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1453 if (shader
->Geom
.VerticesOut
!= 0) {
1454 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1455 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1456 linker_error(prog
, "geometry shader defined with conflicting "
1457 "output vertex count (%d and %d)\n",
1458 linked_shader
->Geom
.VerticesOut
,
1459 shader
->Geom
.VerticesOut
);
1462 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1465 if (shader
->Geom
.Invocations
!= 0) {
1466 if (linked_shader
->Geom
.Invocations
!= 0 &&
1467 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1468 linker_error(prog
, "geometry shader defined with conflicting "
1469 "invocation count (%d and %d)\n",
1470 linked_shader
->Geom
.Invocations
,
1471 shader
->Geom
.Invocations
);
1474 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1478 /* Just do the intrastage -> interstage propagation right now,
1479 * since we already know we're in the right type of shader program
1482 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1484 "geometry shader didn't declare primitive input type\n");
1487 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1489 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1491 "geometry shader didn't declare primitive output type\n");
1494 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1496 if (linked_shader
->Geom
.VerticesOut
== 0) {
1498 "geometry shader didn't declare max_vertices\n");
1501 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1503 if (linked_shader
->Geom
.Invocations
== 0)
1504 linked_shader
->Geom
.Invocations
= 1;
1506 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1511 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1512 * qualifiers for the attached compute shaders, and propagate them to the
1513 * linked CS and linked shader program.
1516 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1517 struct gl_shader
*linked_shader
,
1518 struct gl_shader
**shader_list
,
1519 unsigned num_shaders
)
1521 for (int i
= 0; i
< 3; i
++)
1522 linked_shader
->Comp
.LocalSize
[i
] = 0;
1524 /* This function is called for all shader stages, but it only has an effect
1525 * for compute shaders.
1527 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1530 /* From the ARB_compute_shader spec, in the section describing local size
1533 * If multiple compute shaders attached to a single program object
1534 * declare local work-group size, the declarations must be identical;
1535 * otherwise a link-time error results. Furthermore, if a program
1536 * object contains any compute shaders, at least one must contain an
1537 * input layout qualifier specifying the local work sizes of the
1538 * program, or a link-time error will occur.
1540 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1541 struct gl_shader
*shader
= shader_list
[sh
];
1543 if (shader
->Comp
.LocalSize
[0] != 0) {
1544 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1545 for (int i
= 0; i
< 3; i
++) {
1546 if (linked_shader
->Comp
.LocalSize
[i
] !=
1547 shader
->Comp
.LocalSize
[i
]) {
1548 linker_error(prog
, "compute shader defined with conflicting "
1554 for (int i
= 0; i
< 3; i
++)
1555 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1559 /* Just do the intrastage -> interstage propagation right now,
1560 * since we already know we're in the right type of shader program
1563 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1564 linker_error(prog
, "compute shader didn't declare local size\n");
1567 for (int i
= 0; i
< 3; i
++)
1568 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1573 * Combine a group of shaders for a single stage to generate a linked shader
1576 * If this function is supplied a single shader, it is cloned, and the new
1577 * shader is returned.
1579 static struct gl_shader
*
1580 link_intrastage_shaders(void *mem_ctx
,
1581 struct gl_context
*ctx
,
1582 struct gl_shader_program
*prog
,
1583 struct gl_shader
**shader_list
,
1584 unsigned num_shaders
)
1586 struct gl_uniform_block
*uniform_blocks
= NULL
;
1588 /* Check that global variables defined in multiple shaders are consistent.
1590 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1591 if (!prog
->LinkStatus
)
1594 /* Check that interface blocks defined in multiple shaders are consistent.
1596 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1598 if (!prog
->LinkStatus
)
1601 /* Link up uniform blocks defined within this stage. */
1602 const unsigned num_uniform_blocks
=
1603 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1605 if (!prog
->LinkStatus
)
1608 /* Check that there is only a single definition of each function signature
1609 * across all shaders.
1611 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1612 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1613 ir_function
*const f
= node
->as_function();
1618 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1619 ir_function
*const other
=
1620 shader_list
[j
]->symbols
->get_function(f
->name
);
1622 /* If the other shader has no function (and therefore no function
1623 * signatures) with the same name, skip to the next shader.
1628 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1629 if (!sig
->is_defined
|| sig
->is_builtin())
1632 ir_function_signature
*other_sig
=
1633 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1635 if ((other_sig
!= NULL
) && other_sig
->is_defined
1636 && !other_sig
->is_builtin()) {
1637 linker_error(prog
, "function `%s' is multiply defined",
1646 /* Find the shader that defines main, and make a clone of it.
1648 * Starting with the clone, search for undefined references. If one is
1649 * found, find the shader that defines it. Clone the reference and add
1650 * it to the shader. Repeat until there are no undefined references or
1651 * until a reference cannot be resolved.
1653 gl_shader
*main
= NULL
;
1654 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1655 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1656 main
= shader_list
[i
];
1662 linker_error(prog
, "%s shader lacks `main'\n",
1663 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1667 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1668 linked
->ir
= new(linked
) exec_list
;
1669 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1671 linked
->UniformBlocks
= uniform_blocks
;
1672 linked
->NumUniformBlocks
= num_uniform_blocks
;
1673 ralloc_steal(linked
, linked
->UniformBlocks
);
1675 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1676 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1677 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1679 populate_symbol_table(linked
);
1681 /* The a pointer to the main function in the final linked shader (i.e., the
1682 * copy of the original shader that contained the main function).
1684 ir_function_signature
*const main_sig
=
1685 link_get_main_function_signature(linked
);
1687 /* Move any instructions other than variable declarations or function
1688 * declarations into main.
1690 exec_node
*insertion_point
=
1691 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1694 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1695 if (shader_list
[i
] == main
)
1698 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1699 insertion_point
, true, linked
);
1702 /* Check if any shader needs built-in functions. */
1703 bool need_builtins
= false;
1704 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1705 if (shader_list
[i
]->uses_builtin_functions
) {
1706 need_builtins
= true;
1712 if (need_builtins
) {
1713 /* Make a temporary array one larger than shader_list, which will hold
1714 * the built-in function shader as well.
1716 gl_shader
**linking_shaders
= (gl_shader
**)
1717 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1719 ok
= linking_shaders
!= NULL
;
1722 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1723 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1725 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1727 free(linking_shaders
);
1729 _mesa_error_no_memory(__func__
);
1732 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1737 ctx
->Driver
.DeleteShader(ctx
, linked
);
1741 /* At this point linked should contain all of the linked IR, so
1742 * validate it to make sure nothing went wrong.
1744 validate_ir_tree(linked
->ir
);
1746 /* Set the size of geometry shader input arrays */
1747 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1748 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1749 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1750 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
1751 ir
->accept(&input_resize_visitor
);
1755 if (ctx
->Const
.VertexID_is_zero_based
)
1756 lower_vertex_id(linked
);
1758 /* Make a pass over all variable declarations to ensure that arrays with
1759 * unspecified sizes have a size specified. The size is inferred from the
1760 * max_array_access field.
1762 array_sizing_visitor v
;
1764 v
.fixup_unnamed_interface_types();
1770 * Update the sizes of linked shader uniform arrays to the maximum
1773 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1775 * If one or more elements of an array are active,
1776 * GetActiveUniform will return the name of the array in name,
1777 * subject to the restrictions listed above. The type of the array
1778 * is returned in type. The size parameter contains the highest
1779 * array element index used, plus one. The compiler or linker
1780 * determines the highest index used. There will be only one
1781 * active uniform reported by the GL per uniform array.
1785 update_array_sizes(struct gl_shader_program
*prog
)
1787 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1788 if (prog
->_LinkedShaders
[i
] == NULL
)
1791 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
1792 ir_variable
*const var
= node
->as_variable();
1794 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1795 !var
->type
->is_array())
1798 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1799 * will not be eliminated. Since we always do std140, just
1800 * don't resize arrays in UBOs.
1802 * Atomic counters are supposed to get deterministic
1803 * locations assigned based on the declaration ordering and
1804 * sizes, array compaction would mess that up.
1806 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1809 unsigned int size
= var
->data
.max_array_access
;
1810 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1811 if (prog
->_LinkedShaders
[j
] == NULL
)
1814 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
1815 ir_variable
*other_var
= node2
->as_variable();
1819 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1820 other_var
->data
.max_array_access
> size
) {
1821 size
= other_var
->data
.max_array_access
;
1826 if (size
+ 1 != var
->type
->length
) {
1827 /* If this is a built-in uniform (i.e., it's backed by some
1828 * fixed-function state), adjust the number of state slots to
1829 * match the new array size. The number of slots per array entry
1830 * is not known. It seems safe to assume that the total number of
1831 * slots is an integer multiple of the number of array elements.
1832 * Determine the number of slots per array element by dividing by
1833 * the old (total) size.
1835 if (var
->num_state_slots
> 0) {
1836 var
->num_state_slots
= (size
+ 1)
1837 * (var
->num_state_slots
/ var
->type
->length
);
1840 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1842 /* FINISHME: We should update the types of array
1843 * dereferences of this variable now.
1851 * Find a contiguous set of available bits in a bitmask.
1853 * \param used_mask Bits representing used (1) and unused (0) locations
1854 * \param needed_count Number of contiguous bits needed.
1857 * Base location of the available bits on success or -1 on failure.
1860 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1862 unsigned needed_mask
= (1 << needed_count
) - 1;
1863 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1865 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1866 * cannot optimize possibly infinite loops" for the loop below.
1868 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1871 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1872 if ((needed_mask
& ~used_mask
) == needed_mask
)
1883 * Assign locations for either VS inputs for FS outputs
1885 * \param prog Shader program whose variables need locations assigned
1886 * \param target_index Selector for the program target to receive location
1887 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1888 * \c MESA_SHADER_FRAGMENT.
1889 * \param max_index Maximum number of generic locations. This corresponds
1890 * to either the maximum number of draw buffers or the
1891 * maximum number of generic attributes.
1894 * If locations are successfully assigned, true is returned. Otherwise an
1895 * error is emitted to the shader link log and false is returned.
1898 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1899 unsigned target_index
,
1902 /* Mark invalid locations as being used.
1904 unsigned used_locations
= (max_index
>= 32)
1905 ? ~0 : ~((1 << max_index
) - 1);
1907 assert((target_index
== MESA_SHADER_VERTEX
)
1908 || (target_index
== MESA_SHADER_FRAGMENT
));
1910 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1914 /* Operate in a total of four passes.
1916 * 1. Invalidate the location assignments for all vertex shader inputs.
1918 * 2. Assign locations for inputs that have user-defined (via
1919 * glBindVertexAttribLocation) locations and outputs that have
1920 * user-defined locations (via glBindFragDataLocation).
1922 * 3. Sort the attributes without assigned locations by number of slots
1923 * required in decreasing order. Fragmentation caused by attribute
1924 * locations assigned by the application may prevent large attributes
1925 * from having enough contiguous space.
1927 * 4. Assign locations to any inputs without assigned locations.
1930 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1931 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1933 const enum ir_variable_mode direction
=
1934 (target_index
== MESA_SHADER_VERTEX
)
1935 ? ir_var_shader_in
: ir_var_shader_out
;
1938 /* Temporary storage for the set of attributes that need locations assigned.
1944 /* Used below in the call to qsort. */
1945 static int compare(const void *a
, const void *b
)
1947 const temp_attr
*const l
= (const temp_attr
*) a
;
1948 const temp_attr
*const r
= (const temp_attr
*) b
;
1950 /* Reversed because we want a descending order sort below. */
1951 return r
->slots
- l
->slots
;
1955 unsigned num_attr
= 0;
1957 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
1958 ir_variable
*const var
= node
->as_variable();
1960 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
1963 if (var
->data
.explicit_location
) {
1964 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
1965 || (var
->data
.location
< 0)) {
1967 "invalid explicit location %d specified for `%s'\n",
1968 (var
->data
.location
< 0)
1969 ? var
->data
.location
1970 : var
->data
.location
- generic_base
,
1974 } else if (target_index
== MESA_SHADER_VERTEX
) {
1977 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1978 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1979 var
->data
.location
= binding
;
1980 var
->data
.is_unmatched_generic_inout
= 0;
1982 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1986 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1987 assert(binding
>= FRAG_RESULT_DATA0
);
1988 var
->data
.location
= binding
;
1989 var
->data
.is_unmatched_generic_inout
= 0;
1991 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1992 var
->data
.index
= index
;
1997 /* If the variable is not a built-in and has a location statically
1998 * assigned in the shader (presumably via a layout qualifier), make sure
1999 * that it doesn't collide with other assigned locations. Otherwise,
2000 * add it to the list of variables that need linker-assigned locations.
2002 const unsigned slots
= var
->type
->count_attribute_slots();
2003 if (var
->data
.location
!= -1) {
2004 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2005 /* From page 61 of the OpenGL 4.0 spec:
2007 * "LinkProgram will fail if the attribute bindings assigned
2008 * by BindAttribLocation do not leave not enough space to
2009 * assign a location for an active matrix attribute or an
2010 * active attribute array, both of which require multiple
2011 * contiguous generic attributes."
2013 * I think above text prohibits the aliasing of explicit and
2014 * automatic assignments. But, aliasing is allowed in manual
2015 * assignments of attribute locations. See below comments for
2018 * From OpenGL 4.0 spec, page 61:
2020 * "It is possible for an application to bind more than one
2021 * attribute name to the same location. This is referred to as
2022 * aliasing. This will only work if only one of the aliased
2023 * attributes is active in the executable program, or if no
2024 * path through the shader consumes more than one attribute of
2025 * a set of attributes aliased to the same location. A link
2026 * error can occur if the linker determines that every path
2027 * through the shader consumes multiple aliased attributes,
2028 * but implementations are not required to generate an error
2031 * From GLSL 4.30 spec, page 54:
2033 * "A program will fail to link if any two non-vertex shader
2034 * input variables are assigned to the same location. For
2035 * vertex shaders, multiple input variables may be assigned
2036 * to the same location using either layout qualifiers or via
2037 * the OpenGL API. However, such aliasing is intended only to
2038 * support vertex shaders where each execution path accesses
2039 * at most one input per each location. Implementations are
2040 * permitted, but not required, to generate link-time errors
2041 * if they detect that every path through the vertex shader
2042 * executable accesses multiple inputs assigned to any single
2043 * location. For all shader types, a program will fail to link
2044 * if explicit location assignments leave the linker unable
2045 * to find space for other variables without explicit
2048 * From OpenGL ES 3.0 spec, page 56:
2050 * "Binding more than one attribute name to the same location
2051 * is referred to as aliasing, and is not permitted in OpenGL
2052 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2053 * fail when this condition exists. However, aliasing is
2054 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2055 * This will only work if only one of the aliased attributes
2056 * is active in the executable program, or if no path through
2057 * the shader consumes more than one attribute of a set of
2058 * attributes aliased to the same location. A link error can
2059 * occur if the linker determines that every path through the
2060 * shader consumes multiple aliased attributes, but implemen-
2061 * tations are not required to generate an error in this case."
2063 * After looking at above references from OpenGL, OpenGL ES and
2064 * GLSL specifications, we allow aliasing of vertex input variables
2065 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2067 * NOTE: This is not required by the spec but its worth mentioning
2068 * here that we're not doing anything to make sure that no path
2069 * through the vertex shader executable accesses multiple inputs
2070 * assigned to any single location.
2073 /* Mask representing the contiguous slots that will be used by
2076 const unsigned attr
= var
->data
.location
- generic_base
;
2077 const unsigned use_mask
= (1 << slots
) - 1;
2078 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2079 ? "vertex shader input" : "fragment shader output";
2081 /* Generate a link error if the requested locations for this
2082 * attribute exceed the maximum allowed attribute location.
2084 if (attr
+ slots
> max_index
) {
2086 "insufficient contiguous locations "
2087 "available for %s `%s' %d %d %d", string
,
2088 var
->name
, used_locations
, use_mask
, attr
);
2092 /* Generate a link error if the set of bits requested for this
2093 * attribute overlaps any previously allocated bits.
2095 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2096 if (target_index
== MESA_SHADER_FRAGMENT
||
2097 (prog
->IsES
&& prog
->Version
>= 300)) {
2099 "overlapping location is assigned "
2100 "to %s `%s' %d %d %d\n", string
,
2101 var
->name
, used_locations
, use_mask
, attr
);
2104 linker_warning(prog
,
2105 "overlapping location is assigned "
2106 "to %s `%s' %d %d %d\n", string
,
2107 var
->name
, used_locations
, use_mask
, attr
);
2111 used_locations
|= (use_mask
<< attr
);
2117 to_assign
[num_attr
].slots
= slots
;
2118 to_assign
[num_attr
].var
= var
;
2122 /* If all of the attributes were assigned locations by the application (or
2123 * are built-in attributes with fixed locations), return early. This should
2124 * be the common case.
2129 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2131 if (target_index
== MESA_SHADER_VERTEX
) {
2132 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2133 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2134 * reserved to prevent it from being automatically allocated below.
2136 find_deref_visitor
find("gl_Vertex");
2138 if (find
.variable_found())
2139 used_locations
|= (1 << 0);
2142 for (unsigned i
= 0; i
< num_attr
; i
++) {
2143 /* Mask representing the contiguous slots that will be used by this
2146 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2148 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2151 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2152 ? "vertex shader input" : "fragment shader output";
2155 "insufficient contiguous locations "
2156 "available for %s `%s'",
2157 string
, to_assign
[i
].var
->name
);
2161 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2162 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2163 used_locations
|= (use_mask
<< location
);
2171 * Demote shader inputs and outputs that are not used in other stages
2174 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2176 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2177 ir_variable
*const var
= node
->as_variable();
2179 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2182 /* A shader 'in' or 'out' variable is only really an input or output if
2183 * its value is used by other shader stages. This will cause the variable
2184 * to have a location assigned.
2186 if (var
->data
.is_unmatched_generic_inout
) {
2187 var
->data
.mode
= ir_var_auto
;
2194 * Store the gl_FragDepth layout in the gl_shader_program struct.
2197 store_fragdepth_layout(struct gl_shader_program
*prog
)
2199 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2203 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2205 /* We don't look up the gl_FragDepth symbol directly because if
2206 * gl_FragDepth is not used in the shader, it's removed from the IR.
2207 * However, the symbol won't be removed from the symbol table.
2209 * We're only interested in the cases where the variable is NOT removed
2212 foreach_in_list(ir_instruction
, node
, ir
) {
2213 ir_variable
*const var
= node
->as_variable();
2215 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2219 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2220 switch (var
->data
.depth_layout
) {
2221 case ir_depth_layout_none
:
2222 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2224 case ir_depth_layout_any
:
2225 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2227 case ir_depth_layout_greater
:
2228 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2230 case ir_depth_layout_less
:
2231 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2233 case ir_depth_layout_unchanged
:
2234 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2245 * Validate the resources used by a program versus the implementation limits
2248 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2250 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2251 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2256 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2257 linker_error(prog
, "Too many %s shader texture samplers",
2258 _mesa_shader_stage_to_string(i
));
2261 if (sh
->num_uniform_components
>
2262 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2263 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2264 linker_warning(prog
, "Too many %s shader default uniform block "
2265 "components, but the driver will try to optimize "
2266 "them out; this is non-portable out-of-spec "
2268 _mesa_shader_stage_to_string(i
));
2270 linker_error(prog
, "Too many %s shader default uniform block "
2272 _mesa_shader_stage_to_string(i
));
2276 if (sh
->num_combined_uniform_components
>
2277 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2278 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2279 linker_warning(prog
, "Too many %s shader uniform components, "
2280 "but the driver will try to optimize them out; "
2281 "this is non-portable out-of-spec behavior\n",
2282 _mesa_shader_stage_to_string(i
));
2284 linker_error(prog
, "Too many %s shader uniform components",
2285 _mesa_shader_stage_to_string(i
));
2290 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2291 unsigned total_uniform_blocks
= 0;
2293 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2294 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2295 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2297 total_uniform_blocks
++;
2301 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2302 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
2303 prog
->NumUniformBlocks
,
2304 ctx
->Const
.MaxCombinedUniformBlocks
);
2306 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2307 const unsigned max_uniform_blocks
=
2308 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2309 if (blocks
[i
] > max_uniform_blocks
) {
2310 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
2311 _mesa_shader_stage_to_string(i
),
2313 max_uniform_blocks
);
2322 * Validate shader image resources.
2325 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2327 unsigned total_image_units
= 0;
2328 unsigned fragment_outputs
= 0;
2330 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2333 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2334 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2337 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2338 linker_error(prog
, "Too many %s shader image uniforms",
2339 _mesa_shader_stage_to_string(i
));
2341 total_image_units
+= sh
->NumImages
;
2343 if (i
== MESA_SHADER_FRAGMENT
) {
2344 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2345 ir_variable
*var
= node
->as_variable();
2346 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2347 fragment_outputs
+= var
->type
->count_attribute_slots();
2353 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2354 linker_error(prog
, "Too many combined image uniforms");
2356 if (total_image_units
+ fragment_outputs
>
2357 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2358 linker_error(prog
, "Too many combined image uniforms and fragment outputs");
2363 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2364 * for a variable, checks for overlaps between other uniforms using explicit
2368 reserve_explicit_locations(struct gl_shader_program
*prog
,
2369 string_to_uint_map
*map
, ir_variable
*var
)
2371 unsigned slots
= var
->type
->uniform_locations();
2372 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2374 /* Resize remap table if locations do not fit in the current one. */
2375 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2376 prog
->UniformRemapTable
=
2377 reralloc(prog
, prog
->UniformRemapTable
,
2378 gl_uniform_storage
*,
2381 if (!prog
->UniformRemapTable
) {
2382 linker_error(prog
, "Out of memory during linking.");
2386 /* Initialize allocated space. */
2387 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2388 prog
->UniformRemapTable
[i
] = NULL
;
2390 prog
->NumUniformRemapTable
= max_loc
+ 1;
2393 for (unsigned i
= 0; i
< slots
; i
++) {
2394 unsigned loc
= var
->data
.location
+ i
;
2396 /* Check if location is already used. */
2397 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2399 /* Possibly same uniform from a different stage, this is ok. */
2401 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2404 /* ARB_explicit_uniform_location specification states:
2406 * "No two default-block uniform variables in the program can have
2407 * the same location, even if they are unused, otherwise a compiler
2408 * or linker error will be generated."
2411 "location qualifier for uniform %s overlaps"
2412 "previously used location",
2417 /* Initialize location as inactive before optimization
2418 * rounds and location assignment.
2420 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2423 /* Note, base location used for arrays. */
2424 map
->put(var
->data
.location
, var
->name
);
2430 * Check and reserve all explicit uniform locations, called before
2431 * any optimizations happen to handle also inactive uniforms and
2432 * inactive array elements that may get trimmed away.
2435 check_explicit_uniform_locations(struct gl_context
*ctx
,
2436 struct gl_shader_program
*prog
)
2438 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2441 /* This map is used to detect if overlapping explicit locations
2442 * occur with the same uniform (from different stage) or a different one.
2444 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2447 linker_error(prog
, "Out of memory during linking.");
2451 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2452 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2457 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2458 ir_variable
*var
= node
->as_variable();
2459 if ((var
&& var
->data
.mode
== ir_var_uniform
) &&
2460 var
->data
.explicit_location
) {
2461 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2473 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2475 tfeedback_decl
*tfeedback_decls
= NULL
;
2476 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2478 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2480 prog
->LinkStatus
= true; /* All error paths will set this to false */
2481 prog
->Validated
= false;
2482 prog
->_Used
= false;
2484 ralloc_free(prog
->InfoLog
);
2485 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2487 ralloc_free(prog
->UniformBlocks
);
2488 prog
->UniformBlocks
= NULL
;
2489 prog
->NumUniformBlocks
= 0;
2490 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2491 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2492 prog
->UniformBlockStageIndex
[i
] = NULL
;
2495 ralloc_free(prog
->AtomicBuffers
);
2496 prog
->AtomicBuffers
= NULL
;
2497 prog
->NumAtomicBuffers
= 0;
2498 prog
->ARB_fragment_coord_conventions_enable
= false;
2500 /* Separate the shaders into groups based on their type.
2502 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2503 unsigned num_shaders
[MESA_SHADER_STAGES
];
2505 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2506 shader_list
[i
] = (struct gl_shader
**)
2507 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2511 unsigned min_version
= UINT_MAX
;
2512 unsigned max_version
= 0;
2513 const bool is_es_prog
=
2514 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2515 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2516 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2517 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2519 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2520 linker_error(prog
, "all shaders must use same shading "
2521 "language version\n");
2525 prog
->ARB_fragment_coord_conventions_enable
|=
2526 prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
;
2528 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2529 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2530 num_shaders
[shader_type
]++;
2533 /* In desktop GLSL, different shader versions may be linked together. In
2534 * GLSL ES, all shader versions must be the same.
2536 if (is_es_prog
&& min_version
!= max_version
) {
2537 linker_error(prog
, "all shaders must use same shading "
2538 "language version\n");
2542 prog
->Version
= max_version
;
2543 prog
->IsES
= is_es_prog
;
2545 /* Geometry shaders have to be linked with vertex shaders.
2547 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2548 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2549 !prog
->SeparateShader
) {
2550 linker_error(prog
, "Geometry shader must be linked with "
2555 /* Compute shaders have additional restrictions. */
2556 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2557 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2558 linker_error(prog
, "Compute shaders may not be linked with any other "
2559 "type of shader\n");
2562 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2563 if (prog
->_LinkedShaders
[i
] != NULL
)
2564 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2566 prog
->_LinkedShaders
[i
] = NULL
;
2569 /* Link all shaders for a particular stage and validate the result.
2571 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2572 if (num_shaders
[stage
] > 0) {
2573 gl_shader
*const sh
=
2574 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2575 num_shaders
[stage
]);
2577 if (!prog
->LinkStatus
)
2581 case MESA_SHADER_VERTEX
:
2582 validate_vertex_shader_executable(prog
, sh
);
2584 case MESA_SHADER_GEOMETRY
:
2585 validate_geometry_shader_executable(prog
, sh
);
2587 case MESA_SHADER_FRAGMENT
:
2588 validate_fragment_shader_executable(prog
, sh
);
2591 if (!prog
->LinkStatus
)
2594 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2598 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2599 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2600 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2601 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2603 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2605 /* Here begins the inter-stage linking phase. Some initial validation is
2606 * performed, then locations are assigned for uniforms, attributes, and
2609 cross_validate_uniforms(prog
);
2610 if (!prog
->LinkStatus
)
2615 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2616 if (prog
->_LinkedShaders
[prev
] != NULL
)
2620 check_explicit_uniform_locations(ctx
, prog
);
2621 if (!prog
->LinkStatus
)
2624 /* Validate the inputs of each stage with the output of the preceding
2627 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2628 if (prog
->_LinkedShaders
[i
] == NULL
)
2631 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2632 prog
->_LinkedShaders
[i
]);
2633 if (!prog
->LinkStatus
)
2636 cross_validate_outputs_to_inputs(prog
,
2637 prog
->_LinkedShaders
[prev
],
2638 prog
->_LinkedShaders
[i
]);
2639 if (!prog
->LinkStatus
)
2645 /* Cross-validate uniform blocks between shader stages */
2646 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
2647 MESA_SHADER_STAGES
);
2648 if (!prog
->LinkStatus
)
2651 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2652 if (prog
->_LinkedShaders
[i
] != NULL
)
2653 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2656 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2657 * it before optimization because we want most of the checks to get
2658 * dropped thanks to constant propagation.
2660 * This rule also applies to GLSL ES 3.00.
2662 if (max_version
>= (is_es_prog
? 300 : 130)) {
2663 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2665 lower_discard_flow(sh
->ir
);
2669 if (!interstage_cross_validate_uniform_blocks(prog
))
2672 /* Do common optimization before assigning storage for attributes,
2673 * uniforms, and varyings. Later optimization could possibly make
2674 * some of that unused.
2676 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2677 if (prog
->_LinkedShaders
[i
] == NULL
)
2680 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2681 if (!prog
->LinkStatus
)
2684 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
2685 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2688 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
2689 &ctx
->Const
.ShaderCompilerOptions
[i
],
2690 ctx
->Const
.NativeIntegers
))
2694 /* Check and validate stream emissions in geometry shaders */
2695 validate_geometry_shader_emissions(ctx
, prog
);
2697 /* Mark all generic shader inputs and outputs as unpaired. */
2698 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2699 if (prog
->_LinkedShaders
[i
] != NULL
) {
2700 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
2704 /* FINISHME: The value of the max_attribute_index parameter is
2705 * FINISHME: implementation dependent based on the value of
2706 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2707 * FINISHME: at least 16, so hardcode 16 for now.
2709 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2713 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2718 for (first
= 0; first
<= MESA_SHADER_FRAGMENT
; first
++) {
2719 if (prog
->_LinkedShaders
[first
] != NULL
)
2723 if (num_tfeedback_decls
!= 0) {
2724 /* From GL_EXT_transform_feedback:
2725 * A program will fail to link if:
2727 * * the <count> specified by TransformFeedbackVaryingsEXT is
2728 * non-zero, but the program object has no vertex or geometry
2731 if (first
== MESA_SHADER_FRAGMENT
) {
2732 linker_error(prog
, "Transform feedback varyings specified, but "
2733 "no vertex or geometry shader is present.");
2737 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2738 prog
->TransformFeedback
.NumVarying
);
2739 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2740 prog
->TransformFeedback
.VaryingNames
,
2745 /* Linking the stages in the opposite order (from fragment to vertex)
2746 * ensures that inter-shader outputs written to in an earlier stage are
2747 * eliminated if they are (transitively) not used in a later stage.
2750 for (last
= MESA_SHADER_FRAGMENT
; last
>= 0; last
--) {
2751 if (prog
->_LinkedShaders
[last
] != NULL
)
2755 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2756 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2758 if (num_tfeedback_decls
!= 0 || prog
->SeparateShader
) {
2759 /* There was no fragment shader, but we still have to assign varying
2760 * locations for use by transform feedback.
2762 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2764 num_tfeedback_decls
, tfeedback_decls
,
2769 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2770 num_tfeedback_decls
, tfeedback_decls
);
2772 if (!prog
->SeparateShader
)
2773 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2775 /* Eliminate code that is now dead due to unused outputs being demoted.
2777 while (do_dead_code(sh
->ir
, false))
2780 else if (first
== MESA_SHADER_FRAGMENT
) {
2781 /* If the program only contains a fragment shader...
2783 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2785 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2786 num_tfeedback_decls
, tfeedback_decls
);
2788 if (prog
->SeparateShader
) {
2789 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2790 NULL
/* producer */,
2792 0 /* num_tfeedback_decls */,
2793 NULL
/* tfeedback_decls */,
2794 0 /* gs_input_vertices */))
2797 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2799 while (do_dead_code(sh
->ir
, false))
2804 for (int i
= next
- 1; i
>= 0; i
--) {
2805 if (prog
->_LinkedShaders
[i
] == NULL
)
2808 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2809 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2810 unsigned gs_input_vertices
=
2811 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2813 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2814 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2815 tfeedback_decls
, gs_input_vertices
))
2818 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2819 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2822 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2823 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2825 /* Eliminate code that is now dead due to unused outputs being demoted.
2827 while (do_dead_code(sh_i
->ir
, false))
2829 while (do_dead_code(sh_next
->ir
, false))
2832 /* This must be done after all dead varyings are eliminated. */
2833 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2835 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2841 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2844 update_array_sizes(prog
);
2845 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
2846 link_assign_atomic_counter_resources(ctx
, prog
);
2847 store_fragdepth_layout(prog
);
2849 check_resources(ctx
, prog
);
2850 check_image_resources(ctx
, prog
);
2851 link_check_atomic_counter_resources(ctx
, prog
);
2853 if (!prog
->LinkStatus
)
2856 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2857 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
2858 * anything about shader linking when one of the shaders (vertex or
2859 * fragment shader) is absent. So, the extension shouldn't change the
2860 * behavior specified in GLSL specification.
2862 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
2863 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2864 linker_error(prog
, "program lacks a vertex shader\n");
2865 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2866 linker_error(prog
, "program lacks a fragment shader\n");
2870 /* FINISHME: Assign fragment shader output locations. */
2873 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2874 free(shader_list
[i
]);
2875 if (prog
->_LinkedShaders
[i
] == NULL
)
2878 /* Do a final validation step to make sure that the IR wasn't
2879 * invalidated by any modifications performed after intrastage linking.
2881 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
2883 /* Retain any live IR, but trash the rest. */
2884 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2886 /* The symbol table in the linked shaders may contain references to
2887 * variables that were removed (e.g., unused uniforms). Since it may
2888 * contain junk, there is no possible valid use. Delete it and set the
2891 delete prog
->_LinkedShaders
[i
]->symbols
;
2892 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2895 ralloc_free(mem_ctx
);