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 * GLSL ES 3.00 is similar to GLSL 1.40--failing to write to gl_Position is
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()) {
569 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
574 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
575 &prog
->Vert
.ClipDistanceArraySize
);
580 * Verify that a fragment shader executable meets all semantic requirements
582 * \param shader Fragment shader executable to be verified
585 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
586 struct gl_shader
*shader
)
591 find_assignment_visitor
frag_color("gl_FragColor");
592 find_assignment_visitor
frag_data("gl_FragData");
594 frag_color
.run(shader
->ir
);
595 frag_data
.run(shader
->ir
);
597 if (frag_color
.variable_found() && frag_data
.variable_found()) {
598 linker_error(prog
, "fragment shader writes to both "
599 "`gl_FragColor' and `gl_FragData'\n");
604 * Verify that a geometry shader executable meets all semantic requirements
606 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
607 * prog->Geom.ClipDistanceArraySize as a side effect.
609 * \param shader Geometry shader executable to be verified
612 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
613 struct gl_shader
*shader
)
618 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
619 prog
->Geom
.VerticesIn
= num_vertices
;
621 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
622 &prog
->Geom
.ClipDistanceArraySize
);
626 * Check if geometry shaders emit to non-zero streams and do corresponding
630 validate_geometry_shader_emissions(struct gl_context
*ctx
,
631 struct gl_shader_program
*prog
)
633 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
634 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
635 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
636 if (emit_vertex
.error()) {
637 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
638 "stream parameter are in the range [0, %d].",
639 emit_vertex
.error_func(),
640 emit_vertex
.error_stream(),
641 ctx
->Const
.MaxVertexStreams
- 1);
643 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
644 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
646 /* From the ARB_gpu_shader5 spec:
648 * "Multiple vertex streams are supported only if the output primitive
649 * type is declared to be "points". A program will fail to link if it
650 * contains a geometry shader calling EmitStreamVertex() or
651 * EndStreamPrimitive() if its output primitive type is not "points".
653 * However, in the same spec:
655 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
656 * with <stream> set to zero."
660 * "The function EndPrimitive() is equivalent to calling
661 * EndStreamPrimitive() with <stream> set to zero."
663 * Since we can call EmitVertex() and EndPrimitive() when we output
664 * primitives other than points, calling EmitStreamVertex(0) or
665 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
666 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
667 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
670 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
671 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
672 "with n>0 requires point output");
679 * Perform validation of global variables used across multiple shaders
682 cross_validate_globals(struct gl_shader_program
*prog
,
683 struct gl_shader
**shader_list
,
684 unsigned num_shaders
,
687 /* Examine all of the uniforms in all of the shaders and cross validate
690 glsl_symbol_table variables
;
691 for (unsigned i
= 0; i
< num_shaders
; i
++) {
692 if (shader_list
[i
] == NULL
)
695 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
696 ir_variable
*const var
= node
->as_variable();
701 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
))
704 /* Don't cross validate temporaries that are at global scope. These
705 * will eventually get pulled into the shaders 'main'.
707 if (var
->data
.mode
== ir_var_temporary
)
710 /* If a global with this name has already been seen, verify that the
711 * new instance has the same type. In addition, if the globals have
712 * initializers, the values of the initializers must be the same.
714 ir_variable
*const existing
= variables
.get_variable(var
->name
);
715 if (existing
!= NULL
) {
716 if (var
->type
!= existing
->type
) {
717 /* Consider the types to be "the same" if both types are arrays
718 * of the same type and one of the arrays is implicitly sized.
719 * In addition, set the type of the linked variable to the
720 * explicitly sized array.
722 if (var
->type
->is_array()
723 && existing
->type
->is_array()
724 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
725 && ((var
->type
->length
== 0)
726 || (existing
->type
->length
== 0))) {
727 if (var
->type
->length
!= 0) {
728 existing
->type
= var
->type
;
730 } else if (var
->type
->is_record()
731 && existing
->type
->is_record()
732 && existing
->type
->record_compare(var
->type
)) {
733 existing
->type
= var
->type
;
735 linker_error(prog
, "%s `%s' declared as type "
736 "`%s' and type `%s'\n",
738 var
->name
, var
->type
->name
,
739 existing
->type
->name
);
744 if (var
->data
.explicit_location
) {
745 if (existing
->data
.explicit_location
746 && (var
->data
.location
!= existing
->data
.location
)) {
747 linker_error(prog
, "explicit locations for %s "
748 "`%s' have differing values\n",
749 mode_string(var
), var
->name
);
753 existing
->data
.location
= var
->data
.location
;
754 existing
->data
.explicit_location
= true;
757 /* From the GLSL 4.20 specification:
758 * "A link error will result if two compilation units in a program
759 * specify different integer-constant bindings for the same
760 * opaque-uniform name. However, it is not an error to specify a
761 * binding on some but not all declarations for the same name"
763 if (var
->data
.explicit_binding
) {
764 if (existing
->data
.explicit_binding
&&
765 var
->data
.binding
!= existing
->data
.binding
) {
766 linker_error(prog
, "explicit bindings for %s "
767 "`%s' have differing values\n",
768 mode_string(var
), var
->name
);
772 existing
->data
.binding
= var
->data
.binding
;
773 existing
->data
.explicit_binding
= true;
776 if (var
->type
->contains_atomic() &&
777 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
778 linker_error(prog
, "offset specifications for %s "
779 "`%s' have differing values\n",
780 mode_string(var
), var
->name
);
784 /* Validate layout qualifiers for gl_FragDepth.
786 * From the AMD/ARB_conservative_depth specs:
788 * "If gl_FragDepth is redeclared in any fragment shader in a
789 * program, it must be redeclared in all fragment shaders in
790 * that program that have static assignments to
791 * gl_FragDepth. All redeclarations of gl_FragDepth in all
792 * fragment shaders in a single program must have the same set
795 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
796 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
797 bool layout_differs
=
798 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
800 if (layout_declared
&& layout_differs
) {
802 "All redeclarations of gl_FragDepth in all "
803 "fragment shaders in a single program must have "
804 "the same set of qualifiers.");
807 if (var
->data
.used
&& layout_differs
) {
809 "If gl_FragDepth is redeclared with a layout "
810 "qualifier in any fragment shader, it must be "
811 "redeclared with the same layout qualifier in "
812 "all fragment shaders that have assignments to "
817 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
819 * "If a shared global has multiple initializers, the
820 * initializers must all be constant expressions, and they
821 * must all have the same value. Otherwise, a link error will
822 * result. (A shared global having only one initializer does
823 * not require that initializer to be a constant expression.)"
825 * Previous to 4.20 the GLSL spec simply said that initializers
826 * must have the same value. In this case of non-constant
827 * initializers, this was impossible to determine. As a result,
828 * no vendor actually implemented that behavior. The 4.20
829 * behavior matches the implemented behavior of at least one other
830 * vendor, so we'll implement that for all GLSL versions.
832 if (var
->constant_initializer
!= NULL
) {
833 if (existing
->constant_initializer
!= NULL
) {
834 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
835 linker_error(prog
, "initializers for %s "
836 "`%s' have differing values\n",
837 mode_string(var
), var
->name
);
841 /* If the first-seen instance of a particular uniform did not
842 * have an initializer but a later instance does, copy the
843 * initializer to the version stored in the symbol table.
845 /* FINISHME: This is wrong. The constant_value field should
846 * FINISHME: not be modified! Imagine a case where a shader
847 * FINISHME: without an initializer is linked in two different
848 * FINISHME: programs with shaders that have differing
849 * FINISHME: initializers. Linking with the first will
850 * FINISHME: modify the shader, and linking with the second
851 * FINISHME: will fail.
853 existing
->constant_initializer
=
854 var
->constant_initializer
->clone(ralloc_parent(existing
),
859 if (var
->data
.has_initializer
) {
860 if (existing
->data
.has_initializer
861 && (var
->constant_initializer
== NULL
862 || existing
->constant_initializer
== NULL
)) {
864 "shared global variable `%s' has multiple "
865 "non-constant initializers.\n",
870 /* Some instance had an initializer, so keep track of that. In
871 * this location, all sorts of initializers (constant or
872 * otherwise) will propagate the existence to the variable
873 * stored in the symbol table.
875 existing
->data
.has_initializer
= true;
878 if (existing
->data
.invariant
!= var
->data
.invariant
) {
879 linker_error(prog
, "declarations for %s `%s' have "
880 "mismatching invariant qualifiers\n",
881 mode_string(var
), var
->name
);
884 if (existing
->data
.centroid
!= var
->data
.centroid
) {
885 linker_error(prog
, "declarations for %s `%s' have "
886 "mismatching centroid qualifiers\n",
887 mode_string(var
), var
->name
);
890 if (existing
->data
.sample
!= var
->data
.sample
) {
891 linker_error(prog
, "declarations for %s `%s` have "
892 "mismatching sample qualifiers\n",
893 mode_string(var
), var
->name
);
897 variables
.add_variable(var
);
904 * Perform validation of uniforms used across multiple shader stages
907 cross_validate_uniforms(struct gl_shader_program
*prog
)
909 cross_validate_globals(prog
, prog
->_LinkedShaders
,
910 MESA_SHADER_STAGES
, true);
914 * Accumulates the array of prog->UniformBlocks and checks that all
915 * definitons of blocks agree on their contents.
918 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
920 unsigned max_num_uniform_blocks
= 0;
921 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
922 if (prog
->_LinkedShaders
[i
])
923 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
926 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
927 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
929 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
930 max_num_uniform_blocks
);
931 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
932 prog
->UniformBlockStageIndex
[i
][j
] = -1;
937 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
938 int index
= link_cross_validate_uniform_block(prog
,
939 &prog
->UniformBlocks
,
940 &prog
->NumUniformBlocks
,
941 &sh
->UniformBlocks
[j
]);
944 linker_error(prog
, "uniform block `%s' has mismatching definitions",
945 sh
->UniformBlocks
[j
].Name
);
949 prog
->UniformBlockStageIndex
[i
][index
] = j
;
958 * Populates a shaders symbol table with all global declarations
961 populate_symbol_table(gl_shader
*sh
)
963 sh
->symbols
= new(sh
) glsl_symbol_table
;
965 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
969 if ((func
= inst
->as_function()) != NULL
) {
970 sh
->symbols
->add_function(func
);
971 } else if ((var
= inst
->as_variable()) != NULL
) {
972 sh
->symbols
->add_variable(var
);
979 * Remap variables referenced in an instruction tree
981 * This is used when instruction trees are cloned from one shader and placed in
982 * another. These trees will contain references to \c ir_variable nodes that
983 * do not exist in the target shader. This function finds these \c ir_variable
984 * references and replaces the references with matching variables in the target
987 * If there is no matching variable in the target shader, a clone of the
988 * \c ir_variable is made and added to the target shader. The new variable is
989 * added to \b both the instruction stream and the symbol table.
991 * \param inst IR tree that is to be processed.
992 * \param symbols Symbol table containing global scope symbols in the
994 * \param instructions Instruction stream where new variable declarations
998 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1001 class remap_visitor
: public ir_hierarchical_visitor
{
1003 remap_visitor(struct gl_shader
*target
,
1006 this->target
= target
;
1007 this->symbols
= target
->symbols
;
1008 this->instructions
= target
->ir
;
1009 this->temps
= temps
;
1012 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1014 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1015 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1017 assert(var
!= NULL
);
1019 return visit_continue
;
1022 ir_variable
*const existing
=
1023 this->symbols
->get_variable(ir
->var
->name
);
1024 if (existing
!= NULL
)
1027 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1029 this->symbols
->add_variable(copy
);
1030 this->instructions
->push_head(copy
);
1034 return visit_continue
;
1038 struct gl_shader
*target
;
1039 glsl_symbol_table
*symbols
;
1040 exec_list
*instructions
;
1044 remap_visitor
v(target
, temps
);
1051 * Move non-declarations from one instruction stream to another
1053 * The intended usage pattern of this function is to pass the pointer to the
1054 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1055 * pointer) for \c last and \c false for \c make_copies on the first
1056 * call. Successive calls pass the return value of the previous call for
1057 * \c last and \c true for \c make_copies.
1059 * \param instructions Source instruction stream
1060 * \param last Instruction after which new instructions should be
1061 * inserted in the target instruction stream
1062 * \param make_copies Flag selecting whether instructions in \c instructions
1063 * should be copied (via \c ir_instruction::clone) into the
1064 * target list or moved.
1067 * The new "last" instruction in the target instruction stream. This pointer
1068 * is suitable for use as the \c last parameter of a later call to this
1072 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1073 bool make_copies
, gl_shader
*target
)
1075 hash_table
*temps
= NULL
;
1078 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1079 hash_table_pointer_compare
);
1081 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1082 if (inst
->as_function())
1085 ir_variable
*var
= inst
->as_variable();
1086 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1089 assert(inst
->as_assignment()
1091 || inst
->as_if() /* for initializers with the ?: operator */
1092 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1095 inst
= inst
->clone(target
, NULL
);
1098 hash_table_insert(temps
, inst
, var
);
1100 remap_variables(inst
, target
, temps
);
1105 last
->insert_after(inst
);
1110 hash_table_dtor(temps
);
1116 * Get the function signature for main from a shader
1118 static ir_function_signature
*
1119 get_main_function_signature(gl_shader
*sh
)
1121 ir_function
*const f
= sh
->symbols
->get_function("main");
1123 exec_list void_parameters
;
1125 /* Look for the 'void main()' signature and ensure that it's defined.
1126 * This keeps the linker from accidentally pick a shader that just
1127 * contains a prototype for main.
1129 * We don't have to check for multiple definitions of main (in multiple
1130 * shaders) because that would have already been caught above.
1132 ir_function_signature
*sig
=
1133 f
->matching_signature(NULL
, &void_parameters
, false);
1134 if ((sig
!= NULL
) && sig
->is_defined
) {
1144 * This class is only used in link_intrastage_shaders() below but declaring
1145 * it inside that function leads to compiler warnings with some versions of
1148 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1150 array_sizing_visitor()
1151 : mem_ctx(ralloc_context(NULL
)),
1152 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1153 hash_table_pointer_compare
))
1157 ~array_sizing_visitor()
1159 hash_table_dtor(this->unnamed_interfaces
);
1160 ralloc_free(this->mem_ctx
);
1163 virtual ir_visitor_status
visit(ir_variable
*var
)
1165 fixup_type(&var
->type
, var
->data
.max_array_access
);
1166 if (var
->type
->is_interface()) {
1167 if (interface_contains_unsized_arrays(var
->type
)) {
1168 const glsl_type
*new_type
=
1169 resize_interface_members(var
->type
, var
->max_ifc_array_access
);
1170 var
->type
= new_type
;
1171 var
->change_interface_type(new_type
);
1173 } else if (var
->type
->is_array() &&
1174 var
->type
->fields
.array
->is_interface()) {
1175 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1176 const glsl_type
*new_type
=
1177 resize_interface_members(var
->type
->fields
.array
,
1178 var
->max_ifc_array_access
);
1179 var
->change_interface_type(new_type
);
1181 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1183 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1184 /* Store a pointer to the variable in the unnamed_interfaces
1187 ir_variable
**interface_vars
= (ir_variable
**)
1188 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1189 if (interface_vars
== NULL
) {
1190 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1192 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1195 unsigned index
= ifc_type
->field_index(var
->name
);
1196 assert(index
< ifc_type
->length
);
1197 assert(interface_vars
[index
] == NULL
);
1198 interface_vars
[index
] = var
;
1200 return visit_continue
;
1204 * For each unnamed interface block that was discovered while running the
1205 * visitor, adjust the interface type to reflect the newly assigned array
1206 * sizes, and fix up the ir_variable nodes to point to the new interface
1209 void fixup_unnamed_interface_types()
1211 hash_table_call_foreach(this->unnamed_interfaces
,
1212 fixup_unnamed_interface_type
, NULL
);
1217 * If the type pointed to by \c type represents an unsized array, replace
1218 * it with a sized array whose size is determined by max_array_access.
1220 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1222 if ((*type
)->is_unsized_array()) {
1223 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1224 max_array_access
+ 1);
1225 assert(*type
!= NULL
);
1230 * Determine whether the given interface type contains unsized arrays (if
1231 * it doesn't, array_sizing_visitor doesn't need to process it).
1233 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1235 for (unsigned i
= 0; i
< type
->length
; i
++) {
1236 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1237 if (elem_type
->is_unsized_array())
1244 * Create a new interface type based on the given type, with unsized arrays
1245 * replaced by sized arrays whose size is determined by
1246 * max_ifc_array_access.
1248 static const glsl_type
*
1249 resize_interface_members(const glsl_type
*type
,
1250 const unsigned *max_ifc_array_access
)
1252 unsigned num_fields
= type
->length
;
1253 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1254 memcpy(fields
, type
->fields
.structure
,
1255 num_fields
* sizeof(*fields
));
1256 for (unsigned i
= 0; i
< num_fields
; i
++) {
1257 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1259 glsl_interface_packing packing
=
1260 (glsl_interface_packing
) type
->interface_packing
;
1261 const glsl_type
*new_ifc_type
=
1262 glsl_type::get_interface_instance(fields
, num_fields
,
1263 packing
, type
->name
);
1265 return new_ifc_type
;
1268 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1271 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1272 ir_variable
**interface_vars
= (ir_variable
**) data
;
1273 unsigned num_fields
= ifc_type
->length
;
1274 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1275 memcpy(fields
, ifc_type
->fields
.structure
,
1276 num_fields
* sizeof(*fields
));
1277 bool interface_type_changed
= false;
1278 for (unsigned i
= 0; i
< num_fields
; i
++) {
1279 if (interface_vars
[i
] != NULL
&&
1280 fields
[i
].type
!= interface_vars
[i
]->type
) {
1281 fields
[i
].type
= interface_vars
[i
]->type
;
1282 interface_type_changed
= true;
1285 if (!interface_type_changed
) {
1289 glsl_interface_packing packing
=
1290 (glsl_interface_packing
) ifc_type
->interface_packing
;
1291 const glsl_type
*new_ifc_type
=
1292 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1295 for (unsigned i
= 0; i
< num_fields
; i
++) {
1296 if (interface_vars
[i
] != NULL
)
1297 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1302 * Memory context used to allocate the data in \c unnamed_interfaces.
1307 * Hash table from const glsl_type * to an array of ir_variable *'s
1308 * pointing to the ir_variables constituting each unnamed interface block.
1310 hash_table
*unnamed_interfaces
;
1314 * Performs the cross-validation of layout qualifiers specified in
1315 * redeclaration of gl_FragCoord for the attached fragment shaders,
1316 * and propagates them to the linked FS and linked shader program.
1319 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1320 struct gl_shader
*linked_shader
,
1321 struct gl_shader
**shader_list
,
1322 unsigned num_shaders
)
1324 linked_shader
->redeclares_gl_fragcoord
= false;
1325 linked_shader
->uses_gl_fragcoord
= false;
1326 linked_shader
->origin_upper_left
= false;
1327 linked_shader
->pixel_center_integer
= false;
1329 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1330 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1333 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1334 struct gl_shader
*shader
= shader_list
[i
];
1335 /* From the GLSL 1.50 spec, page 39:
1337 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1338 * it must be redeclared in all the fragment shaders in that program
1339 * that have a static use gl_FragCoord."
1341 * Exclude the case when one of the 'linked_shader' or 'shader' redeclares
1342 * gl_FragCoord with no layout qualifiers but the other one doesn't
1343 * redeclare it. If we strictly follow GLSL 1.50 spec's language, it
1344 * should be a link error. But, generating link error for this case will
1345 * be a wrong behaviour which spec didn't intend to do and it could also
1346 * break some applications.
1348 if ((linked_shader
->redeclares_gl_fragcoord
1349 && !shader
->redeclares_gl_fragcoord
1350 && shader
->uses_gl_fragcoord
1351 && (linked_shader
->origin_upper_left
1352 || linked_shader
->pixel_center_integer
))
1353 || (shader
->redeclares_gl_fragcoord
1354 && !linked_shader
->redeclares_gl_fragcoord
1355 && linked_shader
->uses_gl_fragcoord
1356 && (shader
->origin_upper_left
1357 || shader
->pixel_center_integer
))) {
1358 linker_error(prog
, "fragment shader defined with conflicting "
1359 "layout qualifiers for gl_FragCoord\n");
1362 /* From the GLSL 1.50 spec, page 39:
1364 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1365 * single program must have the same set of qualifiers."
1367 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1368 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1369 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1370 linker_error(prog
, "fragment shader defined with conflicting "
1371 "layout qualifiers for gl_FragCoord\n");
1374 /* Update the linked shader state. Note that uses_gl_fragcoord should
1375 * accumulate the results. The other values should replace. If there
1376 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1377 * are already known to be the same.
1379 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1380 linked_shader
->redeclares_gl_fragcoord
=
1381 shader
->redeclares_gl_fragcoord
;
1382 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1383 || shader
->uses_gl_fragcoord
;
1384 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1385 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1391 * Performs the cross-validation of geometry shader max_vertices and
1392 * primitive type layout qualifiers for the attached geometry shaders,
1393 * and propagates them to the linked GS and linked shader program.
1396 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1397 struct gl_shader
*linked_shader
,
1398 struct gl_shader
**shader_list
,
1399 unsigned num_shaders
)
1401 linked_shader
->Geom
.VerticesOut
= 0;
1402 linked_shader
->Geom
.Invocations
= 0;
1403 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1404 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1406 /* No in/out qualifiers defined for anything but GLSL 1.50+
1407 * geometry shaders so far.
1409 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1412 /* From the GLSL 1.50 spec, page 46:
1414 * "All geometry shader output layout declarations in a program
1415 * must declare the same layout and same value for
1416 * max_vertices. There must be at least one geometry output
1417 * layout declaration somewhere in a program, but not all
1418 * geometry shaders (compilation units) are required to
1422 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1423 struct gl_shader
*shader
= shader_list
[i
];
1425 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1426 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1427 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1428 linker_error(prog
, "geometry shader defined with conflicting "
1432 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1435 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1436 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1437 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1438 linker_error(prog
, "geometry shader defined with conflicting "
1442 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1445 if (shader
->Geom
.VerticesOut
!= 0) {
1446 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1447 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1448 linker_error(prog
, "geometry shader defined with conflicting "
1449 "output vertex count (%d and %d)\n",
1450 linked_shader
->Geom
.VerticesOut
,
1451 shader
->Geom
.VerticesOut
);
1454 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1457 if (shader
->Geom
.Invocations
!= 0) {
1458 if (linked_shader
->Geom
.Invocations
!= 0 &&
1459 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1460 linker_error(prog
, "geometry shader defined with conflicting "
1461 "invocation count (%d and %d)\n",
1462 linked_shader
->Geom
.Invocations
,
1463 shader
->Geom
.Invocations
);
1466 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1470 /* Just do the intrastage -> interstage propagation right now,
1471 * since we already know we're in the right type of shader program
1474 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1476 "geometry shader didn't declare primitive input type\n");
1479 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1481 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1483 "geometry shader didn't declare primitive output type\n");
1486 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1488 if (linked_shader
->Geom
.VerticesOut
== 0) {
1490 "geometry shader didn't declare max_vertices\n");
1493 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1495 if (linked_shader
->Geom
.Invocations
== 0)
1496 linked_shader
->Geom
.Invocations
= 1;
1498 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1503 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1504 * qualifiers for the attached compute shaders, and propagate them to the
1505 * linked CS and linked shader program.
1508 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1509 struct gl_shader
*linked_shader
,
1510 struct gl_shader
**shader_list
,
1511 unsigned num_shaders
)
1513 for (int i
= 0; i
< 3; i
++)
1514 linked_shader
->Comp
.LocalSize
[i
] = 0;
1516 /* This function is called for all shader stages, but it only has an effect
1517 * for compute shaders.
1519 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1522 /* From the ARB_compute_shader spec, in the section describing local size
1525 * If multiple compute shaders attached to a single program object
1526 * declare local work-group size, the declarations must be identical;
1527 * otherwise a link-time error results. Furthermore, if a program
1528 * object contains any compute shaders, at least one must contain an
1529 * input layout qualifier specifying the local work sizes of the
1530 * program, or a link-time error will occur.
1532 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1533 struct gl_shader
*shader
= shader_list
[sh
];
1535 if (shader
->Comp
.LocalSize
[0] != 0) {
1536 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1537 for (int i
= 0; i
< 3; i
++) {
1538 if (linked_shader
->Comp
.LocalSize
[i
] !=
1539 shader
->Comp
.LocalSize
[i
]) {
1540 linker_error(prog
, "compute shader defined with conflicting "
1546 for (int i
= 0; i
< 3; i
++)
1547 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1551 /* Just do the intrastage -> interstage propagation right now,
1552 * since we already know we're in the right type of shader program
1555 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1556 linker_error(prog
, "compute shader didn't declare local size\n");
1559 for (int i
= 0; i
< 3; i
++)
1560 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1565 * Combine a group of shaders for a single stage to generate a linked shader
1568 * If this function is supplied a single shader, it is cloned, and the new
1569 * shader is returned.
1571 static struct gl_shader
*
1572 link_intrastage_shaders(void *mem_ctx
,
1573 struct gl_context
*ctx
,
1574 struct gl_shader_program
*prog
,
1575 struct gl_shader
**shader_list
,
1576 unsigned num_shaders
)
1578 struct gl_uniform_block
*uniform_blocks
= NULL
;
1580 /* Check that global variables defined in multiple shaders are consistent.
1582 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1583 if (!prog
->LinkStatus
)
1586 /* Check that interface blocks defined in multiple shaders are consistent.
1588 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1590 if (!prog
->LinkStatus
)
1593 /* Link up uniform blocks defined within this stage. */
1594 const unsigned num_uniform_blocks
=
1595 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1597 if (!prog
->LinkStatus
)
1600 /* Check that there is only a single definition of each function signature
1601 * across all shaders.
1603 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1604 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1605 ir_function
*const f
= node
->as_function();
1610 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1611 ir_function
*const other
=
1612 shader_list
[j
]->symbols
->get_function(f
->name
);
1614 /* If the other shader has no function (and therefore no function
1615 * signatures) with the same name, skip to the next shader.
1620 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1621 if (!sig
->is_defined
|| sig
->is_builtin())
1624 ir_function_signature
*other_sig
=
1625 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1627 if ((other_sig
!= NULL
) && other_sig
->is_defined
1628 && !other_sig
->is_builtin()) {
1629 linker_error(prog
, "function `%s' is multiply defined",
1638 /* Find the shader that defines main, and make a clone of it.
1640 * Starting with the clone, search for undefined references. If one is
1641 * found, find the shader that defines it. Clone the reference and add
1642 * it to the shader. Repeat until there are no undefined references or
1643 * until a reference cannot be resolved.
1645 gl_shader
*main
= NULL
;
1646 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1647 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
1648 main
= shader_list
[i
];
1654 linker_error(prog
, "%s shader lacks `main'\n",
1655 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1659 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1660 linked
->ir
= new(linked
) exec_list
;
1661 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1663 linked
->UniformBlocks
= uniform_blocks
;
1664 linked
->NumUniformBlocks
= num_uniform_blocks
;
1665 ralloc_steal(linked
, linked
->UniformBlocks
);
1667 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1668 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1669 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1671 populate_symbol_table(linked
);
1673 /* The a pointer to the main function in the final linked shader (i.e., the
1674 * copy of the original shader that contained the main function).
1676 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1678 /* Move any instructions other than variable declarations or function
1679 * declarations into main.
1681 exec_node
*insertion_point
=
1682 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1685 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1686 if (shader_list
[i
] == main
)
1689 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1690 insertion_point
, true, linked
);
1693 /* Check if any shader needs built-in functions. */
1694 bool need_builtins
= false;
1695 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1696 if (shader_list
[i
]->uses_builtin_functions
) {
1697 need_builtins
= true;
1703 if (need_builtins
) {
1704 /* Make a temporary array one larger than shader_list, which will hold
1705 * the built-in function shader as well.
1707 gl_shader
**linking_shaders
= (gl_shader
**)
1708 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1709 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1710 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1712 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1714 free(linking_shaders
);
1716 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1721 ctx
->Driver
.DeleteShader(ctx
, linked
);
1725 /* At this point linked should contain all of the linked IR, so
1726 * validate it to make sure nothing went wrong.
1728 validate_ir_tree(linked
->ir
);
1730 /* Set the size of geometry shader input arrays */
1731 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1732 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1733 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1734 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
1735 ir
->accept(&input_resize_visitor
);
1739 /* Make a pass over all variable declarations to ensure that arrays with
1740 * unspecified sizes have a size specified. The size is inferred from the
1741 * max_array_access field.
1743 array_sizing_visitor v
;
1745 v
.fixup_unnamed_interface_types();
1751 * Update the sizes of linked shader uniform arrays to the maximum
1754 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1756 * If one or more elements of an array are active,
1757 * GetActiveUniform will return the name of the array in name,
1758 * subject to the restrictions listed above. The type of the array
1759 * is returned in type. The size parameter contains the highest
1760 * array element index used, plus one. The compiler or linker
1761 * determines the highest index used. There will be only one
1762 * active uniform reported by the GL per uniform array.
1766 update_array_sizes(struct gl_shader_program
*prog
)
1768 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1769 if (prog
->_LinkedShaders
[i
] == NULL
)
1772 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
1773 ir_variable
*const var
= node
->as_variable();
1775 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1776 !var
->type
->is_array())
1779 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1780 * will not be eliminated. Since we always do std140, just
1781 * don't resize arrays in UBOs.
1783 * Atomic counters are supposed to get deterministic
1784 * locations assigned based on the declaration ordering and
1785 * sizes, array compaction would mess that up.
1787 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1790 unsigned int size
= var
->data
.max_array_access
;
1791 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1792 if (prog
->_LinkedShaders
[j
] == NULL
)
1795 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
1796 ir_variable
*other_var
= node2
->as_variable();
1800 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1801 other_var
->data
.max_array_access
> size
) {
1802 size
= other_var
->data
.max_array_access
;
1807 if (size
+ 1 != var
->type
->length
) {
1808 /* If this is a built-in uniform (i.e., it's backed by some
1809 * fixed-function state), adjust the number of state slots to
1810 * match the new array size. The number of slots per array entry
1811 * is not known. It seems safe to assume that the total number of
1812 * slots is an integer multiple of the number of array elements.
1813 * Determine the number of slots per array element by dividing by
1814 * the old (total) size.
1816 if (var
->num_state_slots
> 0) {
1817 var
->num_state_slots
= (size
+ 1)
1818 * (var
->num_state_slots
/ var
->type
->length
);
1821 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1823 /* FINISHME: We should update the types of array
1824 * dereferences of this variable now.
1832 * Find a contiguous set of available bits in a bitmask.
1834 * \param used_mask Bits representing used (1) and unused (0) locations
1835 * \param needed_count Number of contiguous bits needed.
1838 * Base location of the available bits on success or -1 on failure.
1841 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1843 unsigned needed_mask
= (1 << needed_count
) - 1;
1844 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1846 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1847 * cannot optimize possibly infinite loops" for the loop below.
1849 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1852 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1853 if ((needed_mask
& ~used_mask
) == needed_mask
)
1864 * Assign locations for either VS inputs for FS outputs
1866 * \param prog Shader program whose variables need locations assigned
1867 * \param target_index Selector for the program target to receive location
1868 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1869 * \c MESA_SHADER_FRAGMENT.
1870 * \param max_index Maximum number of generic locations. This corresponds
1871 * to either the maximum number of draw buffers or the
1872 * maximum number of generic attributes.
1875 * If locations are successfully assigned, true is returned. Otherwise an
1876 * error is emitted to the shader link log and false is returned.
1879 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1880 unsigned target_index
,
1883 /* Mark invalid locations as being used.
1885 unsigned used_locations
= (max_index
>= 32)
1886 ? ~0 : ~((1 << max_index
) - 1);
1888 assert((target_index
== MESA_SHADER_VERTEX
)
1889 || (target_index
== MESA_SHADER_FRAGMENT
));
1891 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1895 /* Operate in a total of four passes.
1897 * 1. Invalidate the location assignments for all vertex shader inputs.
1899 * 2. Assign locations for inputs that have user-defined (via
1900 * glBindVertexAttribLocation) locations and outputs that have
1901 * user-defined locations (via glBindFragDataLocation).
1903 * 3. Sort the attributes without assigned locations by number of slots
1904 * required in decreasing order. Fragmentation caused by attribute
1905 * locations assigned by the application may prevent large attributes
1906 * from having enough contiguous space.
1908 * 4. Assign locations to any inputs without assigned locations.
1911 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1912 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1914 const enum ir_variable_mode direction
=
1915 (target_index
== MESA_SHADER_VERTEX
)
1916 ? ir_var_shader_in
: ir_var_shader_out
;
1919 /* Temporary storage for the set of attributes that need locations assigned.
1925 /* Used below in the call to qsort. */
1926 static int compare(const void *a
, const void *b
)
1928 const temp_attr
*const l
= (const temp_attr
*) a
;
1929 const temp_attr
*const r
= (const temp_attr
*) b
;
1931 /* Reversed because we want a descending order sort below. */
1932 return r
->slots
- l
->slots
;
1936 unsigned num_attr
= 0;
1938 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
1939 ir_variable
*const var
= node
->as_variable();
1941 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
1944 if (var
->data
.explicit_location
) {
1945 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
1946 || (var
->data
.location
< 0)) {
1948 "invalid explicit location %d specified for `%s'\n",
1949 (var
->data
.location
< 0)
1950 ? var
->data
.location
1951 : var
->data
.location
- generic_base
,
1955 } else if (target_index
== MESA_SHADER_VERTEX
) {
1958 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1959 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1960 var
->data
.location
= binding
;
1961 var
->data
.is_unmatched_generic_inout
= 0;
1963 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1967 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1968 assert(binding
>= FRAG_RESULT_DATA0
);
1969 var
->data
.location
= binding
;
1970 var
->data
.is_unmatched_generic_inout
= 0;
1972 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1973 var
->data
.index
= index
;
1978 /* If the variable is not a built-in and has a location statically
1979 * assigned in the shader (presumably via a layout qualifier), make sure
1980 * that it doesn't collide with other assigned locations. Otherwise,
1981 * add it to the list of variables that need linker-assigned locations.
1983 const unsigned slots
= var
->type
->count_attribute_slots();
1984 if (var
->data
.location
!= -1) {
1985 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
1986 /* From page 61 of the OpenGL 4.0 spec:
1988 * "LinkProgram will fail if the attribute bindings assigned
1989 * by BindAttribLocation do not leave not enough space to
1990 * assign a location for an active matrix attribute or an
1991 * active attribute array, both of which require multiple
1992 * contiguous generic attributes."
1994 * I think above text prohibits the aliasing of explicit and
1995 * automatic assignments. But, aliasing is allowed in manual
1996 * assignments of attribute locations. See below comments for
1999 * From OpenGL 4.0 spec, page 61:
2001 * "It is possible for an application to bind more than one
2002 * attribute name to the same location. This is referred to as
2003 * aliasing. This will only work if only one of the aliased
2004 * attributes is active in the executable program, or if no
2005 * path through the shader consumes more than one attribute of
2006 * a set of attributes aliased to the same location. A link
2007 * error can occur if the linker determines that every path
2008 * through the shader consumes multiple aliased attributes,
2009 * but implementations are not required to generate an error
2012 * From GLSL 4.30 spec, page 54:
2014 * "A program will fail to link if any two non-vertex shader
2015 * input variables are assigned to the same location. For
2016 * vertex shaders, multiple input variables may be assigned
2017 * to the same location using either layout qualifiers or via
2018 * the OpenGL API. However, such aliasing is intended only to
2019 * support vertex shaders where each execution path accesses
2020 * at most one input per each location. Implementations are
2021 * permitted, but not required, to generate link-time errors
2022 * if they detect that every path through the vertex shader
2023 * executable accesses multiple inputs assigned to any single
2024 * location. For all shader types, a program will fail to link
2025 * if explicit location assignments leave the linker unable
2026 * to find space for other variables without explicit
2029 * From OpenGL ES 3.0 spec, page 56:
2031 * "Binding more than one attribute name to the same location
2032 * is referred to as aliasing, and is not permitted in OpenGL
2033 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2034 * fail when this condition exists. However, aliasing is
2035 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2036 * This will only work if only one of the aliased attributes
2037 * is active in the executable program, or if no path through
2038 * the shader consumes more than one attribute of a set of
2039 * attributes aliased to the same location. A link error can
2040 * occur if the linker determines that every path through the
2041 * shader consumes multiple aliased attributes, but implemen-
2042 * tations are not required to generate an error in this case."
2044 * After looking at above references from OpenGL, OpenGL ES and
2045 * GLSL specifications, we allow aliasing of vertex input variables
2046 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2048 * NOTE: This is not required by the spec but its worth mentioning
2049 * here that we're not doing anything to make sure that no path
2050 * through the vertex shader executable accesses multiple inputs
2051 * assigned to any single location.
2054 /* Mask representing the contiguous slots that will be used by
2057 const unsigned attr
= var
->data
.location
- generic_base
;
2058 const unsigned use_mask
= (1 << slots
) - 1;
2059 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2060 ? "vertex shader input" : "fragment shader output";
2062 /* Generate a link error if the requested locations for this
2063 * attribute exceed the maximum allowed attribute location.
2065 if (attr
+ slots
> max_index
) {
2067 "insufficient contiguous locations "
2068 "available for %s `%s' %d %d %d", string
,
2069 var
->name
, used_locations
, use_mask
, attr
);
2073 /* Generate a link error if the set of bits requested for this
2074 * attribute overlaps any previously allocated bits.
2076 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2077 if (target_index
== MESA_SHADER_FRAGMENT
||
2078 (prog
->IsES
&& prog
->Version
>= 300)) {
2080 "overlapping location is assigned "
2081 "to %s `%s' %d %d %d\n", string
,
2082 var
->name
, used_locations
, use_mask
, attr
);
2085 linker_warning(prog
,
2086 "overlapping location is assigned "
2087 "to %s `%s' %d %d %d\n", string
,
2088 var
->name
, used_locations
, use_mask
, attr
);
2092 used_locations
|= (use_mask
<< attr
);
2098 to_assign
[num_attr
].slots
= slots
;
2099 to_assign
[num_attr
].var
= var
;
2103 /* If all of the attributes were assigned locations by the application (or
2104 * are built-in attributes with fixed locations), return early. This should
2105 * be the common case.
2110 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2112 if (target_index
== MESA_SHADER_VERTEX
) {
2113 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2114 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2115 * reserved to prevent it from being automatically allocated below.
2117 find_deref_visitor
find("gl_Vertex");
2119 if (find
.variable_found())
2120 used_locations
|= (1 << 0);
2123 for (unsigned i
= 0; i
< num_attr
; i
++) {
2124 /* Mask representing the contiguous slots that will be used by this
2127 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2129 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2132 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2133 ? "vertex shader input" : "fragment shader output";
2136 "insufficient contiguous locations "
2137 "available for %s `%s'",
2138 string
, to_assign
[i
].var
->name
);
2142 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2143 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2144 used_locations
|= (use_mask
<< location
);
2152 * Demote shader inputs and outputs that are not used in other stages
2155 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2157 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2158 ir_variable
*const var
= node
->as_variable();
2160 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2163 /* A shader 'in' or 'out' variable is only really an input or output if
2164 * its value is used by other shader stages. This will cause the variable
2165 * to have a location assigned.
2167 if (var
->data
.is_unmatched_generic_inout
) {
2168 var
->data
.mode
= ir_var_auto
;
2175 * Store the gl_FragDepth layout in the gl_shader_program struct.
2178 store_fragdepth_layout(struct gl_shader_program
*prog
)
2180 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2184 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2186 /* We don't look up the gl_FragDepth symbol directly because if
2187 * gl_FragDepth is not used in the shader, it's removed from the IR.
2188 * However, the symbol won't be removed from the symbol table.
2190 * We're only interested in the cases where the variable is NOT removed
2193 foreach_in_list(ir_instruction
, node
, ir
) {
2194 ir_variable
*const var
= node
->as_variable();
2196 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2200 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2201 switch (var
->data
.depth_layout
) {
2202 case ir_depth_layout_none
:
2203 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2205 case ir_depth_layout_any
:
2206 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2208 case ir_depth_layout_greater
:
2209 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2211 case ir_depth_layout_less
:
2212 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2214 case ir_depth_layout_unchanged
:
2215 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2226 * Validate the resources used by a program versus the implementation limits
2229 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2231 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2232 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2237 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2238 linker_error(prog
, "Too many %s shader texture samplers",
2239 _mesa_shader_stage_to_string(i
));
2242 if (sh
->num_uniform_components
>
2243 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2244 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2245 linker_warning(prog
, "Too many %s shader default uniform block "
2246 "components, but the driver will try to optimize "
2247 "them out; this is non-portable out-of-spec "
2249 _mesa_shader_stage_to_string(i
));
2251 linker_error(prog
, "Too many %s shader default uniform block "
2253 _mesa_shader_stage_to_string(i
));
2257 if (sh
->num_combined_uniform_components
>
2258 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2259 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2260 linker_warning(prog
, "Too many %s shader uniform components, "
2261 "but the driver will try to optimize them out; "
2262 "this is non-portable out-of-spec behavior\n",
2263 _mesa_shader_stage_to_string(i
));
2265 linker_error(prog
, "Too many %s shader uniform components",
2266 _mesa_shader_stage_to_string(i
));
2271 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2272 unsigned total_uniform_blocks
= 0;
2274 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2275 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2276 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2278 total_uniform_blocks
++;
2282 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2283 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
2284 prog
->NumUniformBlocks
,
2285 ctx
->Const
.MaxCombinedUniformBlocks
);
2287 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2288 const unsigned max_uniform_blocks
=
2289 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2290 if (blocks
[i
] > max_uniform_blocks
) {
2291 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
2292 _mesa_shader_stage_to_string(i
),
2294 max_uniform_blocks
);
2303 * Validate shader image resources.
2306 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2308 unsigned total_image_units
= 0;
2309 unsigned fragment_outputs
= 0;
2311 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2314 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2315 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2318 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2319 linker_error(prog
, "Too many %s shader image uniforms",
2320 _mesa_shader_stage_to_string(i
));
2322 total_image_units
+= sh
->NumImages
;
2324 if (i
== MESA_SHADER_FRAGMENT
) {
2325 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2326 ir_variable
*var
= node
->as_variable();
2327 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2328 fragment_outputs
+= var
->type
->count_attribute_slots();
2334 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2335 linker_error(prog
, "Too many combined image uniforms");
2337 if (total_image_units
+ fragment_outputs
>
2338 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2339 linker_error(prog
, "Too many combined image uniforms and fragment outputs");
2344 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2345 * for a variable, checks for overlaps between other uniforms using explicit
2349 reserve_explicit_locations(struct gl_shader_program
*prog
,
2350 string_to_uint_map
*map
, ir_variable
*var
)
2352 unsigned slots
= var
->type
->uniform_locations();
2353 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2355 /* Resize remap table if locations do not fit in the current one. */
2356 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2357 prog
->UniformRemapTable
=
2358 reralloc(prog
, prog
->UniformRemapTable
,
2359 gl_uniform_storage
*,
2362 if (!prog
->UniformRemapTable
) {
2363 linker_error(prog
, "Out of memory during linking.");
2367 /* Initialize allocated space. */
2368 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2369 prog
->UniformRemapTable
[i
] = NULL
;
2371 prog
->NumUniformRemapTable
= max_loc
+ 1;
2374 for (unsigned i
= 0; i
< slots
; i
++) {
2375 unsigned loc
= var
->data
.location
+ i
;
2377 /* Check if location is already used. */
2378 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2380 /* Possibly same uniform from a different stage, this is ok. */
2382 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2385 /* ARB_explicit_uniform_location specification states:
2387 * "No two default-block uniform variables in the program can have
2388 * the same location, even if they are unused, otherwise a compiler
2389 * or linker error will be generated."
2392 "location qualifier for uniform %s overlaps"
2393 "previously used location",
2398 /* Initialize location as inactive before optimization
2399 * rounds and location assignment.
2401 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2404 /* Note, base location used for arrays. */
2405 map
->put(var
->data
.location
, var
->name
);
2411 * Check and reserve all explicit uniform locations, called before
2412 * any optimizations happen to handle also inactive uniforms and
2413 * inactive array elements that may get trimmed away.
2416 check_explicit_uniform_locations(struct gl_context
*ctx
,
2417 struct gl_shader_program
*prog
)
2419 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2422 /* This map is used to detect if overlapping explicit locations
2423 * occur with the same uniform (from different stage) or a different one.
2425 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2428 linker_error(prog
, "Out of memory during linking.");
2432 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2433 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2438 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2439 ir_variable
*var
= node
->as_variable();
2440 if ((var
&& var
->data
.mode
== ir_var_uniform
) &&
2441 var
->data
.explicit_location
) {
2442 if (!reserve_explicit_locations(prog
, uniform_map
, var
))
2452 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2454 tfeedback_decl
*tfeedback_decls
= NULL
;
2455 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2457 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2459 prog
->LinkStatus
= true; /* All error paths will set this to false */
2460 prog
->Validated
= false;
2461 prog
->_Used
= false;
2463 ralloc_free(prog
->InfoLog
);
2464 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2466 ralloc_free(prog
->UniformBlocks
);
2467 prog
->UniformBlocks
= NULL
;
2468 prog
->NumUniformBlocks
= 0;
2469 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2470 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2471 prog
->UniformBlockStageIndex
[i
] = NULL
;
2474 ralloc_free(prog
->AtomicBuffers
);
2475 prog
->AtomicBuffers
= NULL
;
2476 prog
->NumAtomicBuffers
= 0;
2477 prog
->ARB_fragment_coord_conventions_enable
= false;
2479 /* Separate the shaders into groups based on their type.
2481 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2482 unsigned num_shaders
[MESA_SHADER_STAGES
];
2484 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2485 shader_list
[i
] = (struct gl_shader
**)
2486 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2490 unsigned min_version
= UINT_MAX
;
2491 unsigned max_version
= 0;
2492 const bool is_es_prog
=
2493 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2494 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2495 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2496 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2498 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2499 linker_error(prog
, "all shaders must use same shading "
2500 "language version\n");
2504 prog
->ARB_fragment_coord_conventions_enable
|=
2505 prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
;
2507 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2508 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2509 num_shaders
[shader_type
]++;
2512 /* In desktop GLSL, different shader versions may be linked together. In
2513 * GLSL ES, all shader versions must be the same.
2515 if (is_es_prog
&& min_version
!= max_version
) {
2516 linker_error(prog
, "all shaders must use same shading "
2517 "language version\n");
2521 prog
->Version
= max_version
;
2522 prog
->IsES
= is_es_prog
;
2524 /* Geometry shaders have to be linked with vertex shaders.
2526 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2527 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2528 !prog
->SeparateShader
) {
2529 linker_error(prog
, "Geometry shader must be linked with "
2534 /* Compute shaders have additional restrictions. */
2535 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2536 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2537 linker_error(prog
, "Compute shaders may not be linked with any other "
2538 "type of shader\n");
2541 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2542 if (prog
->_LinkedShaders
[i
] != NULL
)
2543 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2545 prog
->_LinkedShaders
[i
] = NULL
;
2548 /* Link all shaders for a particular stage and validate the result.
2550 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2551 if (num_shaders
[stage
] > 0) {
2552 gl_shader
*const sh
=
2553 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2554 num_shaders
[stage
]);
2556 if (!prog
->LinkStatus
)
2560 case MESA_SHADER_VERTEX
:
2561 validate_vertex_shader_executable(prog
, sh
);
2563 case MESA_SHADER_GEOMETRY
:
2564 validate_geometry_shader_executable(prog
, sh
);
2566 case MESA_SHADER_FRAGMENT
:
2567 validate_fragment_shader_executable(prog
, sh
);
2570 if (!prog
->LinkStatus
)
2573 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2577 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2578 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2579 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2580 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2582 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2584 /* Here begins the inter-stage linking phase. Some initial validation is
2585 * performed, then locations are assigned for uniforms, attributes, and
2588 cross_validate_uniforms(prog
);
2589 if (!prog
->LinkStatus
)
2594 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2595 if (prog
->_LinkedShaders
[prev
] != NULL
)
2599 check_explicit_uniform_locations(ctx
, prog
);
2600 if (!prog
->LinkStatus
)
2603 /* Validate the inputs of each stage with the output of the preceding
2606 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2607 if (prog
->_LinkedShaders
[i
] == NULL
)
2610 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2611 prog
->_LinkedShaders
[i
]);
2612 if (!prog
->LinkStatus
)
2615 cross_validate_outputs_to_inputs(prog
,
2616 prog
->_LinkedShaders
[prev
],
2617 prog
->_LinkedShaders
[i
]);
2618 if (!prog
->LinkStatus
)
2624 /* Cross-validate uniform blocks between shader stages */
2625 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
2626 MESA_SHADER_STAGES
);
2627 if (!prog
->LinkStatus
)
2630 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2631 if (prog
->_LinkedShaders
[i
] != NULL
)
2632 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2635 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2636 * it before optimization because we want most of the checks to get
2637 * dropped thanks to constant propagation.
2639 * This rule also applies to GLSL ES 3.00.
2641 if (max_version
>= (is_es_prog
? 300 : 130)) {
2642 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2644 lower_discard_flow(sh
->ir
);
2648 if (!interstage_cross_validate_uniform_blocks(prog
))
2651 /* Do common optimization before assigning storage for attributes,
2652 * uniforms, and varyings. Later optimization could possibly make
2653 * some of that unused.
2655 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2656 if (prog
->_LinkedShaders
[i
] == NULL
)
2659 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2660 if (!prog
->LinkStatus
)
2663 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
2664 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2667 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
2668 &ctx
->Const
.ShaderCompilerOptions
[i
],
2669 ctx
->Const
.NativeIntegers
))
2673 /* Check and validate stream emissions in geometry shaders */
2674 validate_geometry_shader_emissions(ctx
, prog
);
2676 /* Mark all generic shader inputs and outputs as unpaired. */
2677 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2678 if (prog
->_LinkedShaders
[i
] != NULL
) {
2679 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
2683 /* FINISHME: The value of the max_attribute_index parameter is
2684 * FINISHME: implementation dependent based on the value of
2685 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2686 * FINISHME: at least 16, so hardcode 16 for now.
2688 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2692 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2697 for (first
= 0; first
<= MESA_SHADER_FRAGMENT
; first
++) {
2698 if (prog
->_LinkedShaders
[first
] != NULL
)
2702 if (num_tfeedback_decls
!= 0) {
2703 /* From GL_EXT_transform_feedback:
2704 * A program will fail to link if:
2706 * * the <count> specified by TransformFeedbackVaryingsEXT is
2707 * non-zero, but the program object has no vertex or geometry
2710 if (first
== MESA_SHADER_FRAGMENT
) {
2711 linker_error(prog
, "Transform feedback varyings specified, but "
2712 "no vertex or geometry shader is present.");
2716 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2717 prog
->TransformFeedback
.NumVarying
);
2718 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2719 prog
->TransformFeedback
.VaryingNames
,
2724 /* Linking the stages in the opposite order (from fragment to vertex)
2725 * ensures that inter-shader outputs written to in an earlier stage are
2726 * eliminated if they are (transitively) not used in a later stage.
2729 for (last
= MESA_SHADER_FRAGMENT
; last
>= 0; last
--) {
2730 if (prog
->_LinkedShaders
[last
] != NULL
)
2734 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2735 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2737 if (num_tfeedback_decls
!= 0 || prog
->SeparateShader
) {
2738 /* There was no fragment shader, but we still have to assign varying
2739 * locations for use by transform feedback.
2741 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2743 num_tfeedback_decls
, tfeedback_decls
,
2748 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2749 num_tfeedback_decls
, tfeedback_decls
);
2751 if (!prog
->SeparateShader
)
2752 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2754 /* Eliminate code that is now dead due to unused outputs being demoted.
2756 while (do_dead_code(sh
->ir
, false))
2759 else if (first
== MESA_SHADER_FRAGMENT
) {
2760 /* If the program only contains a fragment shader...
2762 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2764 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2765 num_tfeedback_decls
, tfeedback_decls
);
2767 if (prog
->SeparateShader
) {
2768 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2769 NULL
/* producer */,
2771 0 /* num_tfeedback_decls */,
2772 NULL
/* tfeedback_decls */,
2773 0 /* gs_input_vertices */))
2776 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2778 while (do_dead_code(sh
->ir
, false))
2783 for (int i
= next
- 1; i
>= 0; i
--) {
2784 if (prog
->_LinkedShaders
[i
] == NULL
)
2787 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2788 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2789 unsigned gs_input_vertices
=
2790 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2792 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2793 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2794 tfeedback_decls
, gs_input_vertices
))
2797 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2798 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2801 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2802 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2804 /* Eliminate code that is now dead due to unused outputs being demoted.
2806 while (do_dead_code(sh_i
->ir
, false))
2808 while (do_dead_code(sh_next
->ir
, false))
2811 /* This must be done after all dead varyings are eliminated. */
2812 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2814 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2820 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2823 update_array_sizes(prog
);
2824 link_assign_uniform_locations(prog
);
2825 link_assign_atomic_counter_resources(ctx
, prog
);
2826 store_fragdepth_layout(prog
);
2828 check_resources(ctx
, prog
);
2829 check_image_resources(ctx
, prog
);
2830 link_check_atomic_counter_resources(ctx
, prog
);
2832 if (!prog
->LinkStatus
)
2835 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2836 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
2837 * anything about shader linking when one of the shaders (vertex or
2838 * fragment shader) is absent. So, the extension shouldn't change the
2839 * behavior specified in GLSL specification.
2841 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
2842 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2843 linker_error(prog
, "program lacks a vertex shader\n");
2844 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2845 linker_error(prog
, "program lacks a fragment shader\n");
2849 /* FINISHME: Assign fragment shader output locations. */
2852 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2853 free(shader_list
[i
]);
2854 if (prog
->_LinkedShaders
[i
] == NULL
)
2857 /* Do a final validation step to make sure that the IR wasn't
2858 * invalidated by any modifications performed after intrastage linking.
2860 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
2862 /* Retain any live IR, but trash the rest. */
2863 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2865 /* The symbol table in the linked shaders may contain references to
2866 * variables that were removed (e.g., unused uniforms). Since it may
2867 * contain junk, there is no possible valid use. Delete it and set the
2870 delete prog
->_LinkedShaders
[i
]->symbols
;
2871 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2874 ralloc_free(mem_ctx
);