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
, var
->max_ifc_array_access
);
1177 var
->type
= new_type
;
1178 var
->change_interface_type(new_type
);
1180 } else if (var
->type
->is_array() &&
1181 var
->type
->fields
.array
->is_interface()) {
1182 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1183 const glsl_type
*new_type
=
1184 resize_interface_members(var
->type
->fields
.array
,
1185 var
->max_ifc_array_access
);
1186 var
->change_interface_type(new_type
);
1188 glsl_type::get_array_instance(new_type
, var
->type
->length
);
1190 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1191 /* Store a pointer to the variable in the unnamed_interfaces
1194 ir_variable
**interface_vars
= (ir_variable
**)
1195 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1196 if (interface_vars
== NULL
) {
1197 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1199 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1202 unsigned index
= ifc_type
->field_index(var
->name
);
1203 assert(index
< ifc_type
->length
);
1204 assert(interface_vars
[index
] == NULL
);
1205 interface_vars
[index
] = var
;
1207 return visit_continue
;
1211 * For each unnamed interface block that was discovered while running the
1212 * visitor, adjust the interface type to reflect the newly assigned array
1213 * sizes, and fix up the ir_variable nodes to point to the new interface
1216 void fixup_unnamed_interface_types()
1218 hash_table_call_foreach(this->unnamed_interfaces
,
1219 fixup_unnamed_interface_type
, NULL
);
1224 * If the type pointed to by \c type represents an unsized array, replace
1225 * it with a sized array whose size is determined by max_array_access.
1227 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1229 if ((*type
)->is_unsized_array()) {
1230 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1231 max_array_access
+ 1);
1232 assert(*type
!= NULL
);
1237 * Determine whether the given interface type contains unsized arrays (if
1238 * it doesn't, array_sizing_visitor doesn't need to process it).
1240 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1242 for (unsigned i
= 0; i
< type
->length
; i
++) {
1243 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1244 if (elem_type
->is_unsized_array())
1251 * Create a new interface type based on the given type, with unsized arrays
1252 * replaced by sized arrays whose size is determined by
1253 * max_ifc_array_access.
1255 static const glsl_type
*
1256 resize_interface_members(const glsl_type
*type
,
1257 const unsigned *max_ifc_array_access
)
1259 unsigned num_fields
= type
->length
;
1260 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1261 memcpy(fields
, type
->fields
.structure
,
1262 num_fields
* sizeof(*fields
));
1263 for (unsigned i
= 0; i
< num_fields
; i
++) {
1264 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1266 glsl_interface_packing packing
=
1267 (glsl_interface_packing
) type
->interface_packing
;
1268 const glsl_type
*new_ifc_type
=
1269 glsl_type::get_interface_instance(fields
, num_fields
,
1270 packing
, type
->name
);
1272 return new_ifc_type
;
1275 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1278 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1279 ir_variable
**interface_vars
= (ir_variable
**) data
;
1280 unsigned num_fields
= ifc_type
->length
;
1281 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1282 memcpy(fields
, ifc_type
->fields
.structure
,
1283 num_fields
* sizeof(*fields
));
1284 bool interface_type_changed
= false;
1285 for (unsigned i
= 0; i
< num_fields
; i
++) {
1286 if (interface_vars
[i
] != NULL
&&
1287 fields
[i
].type
!= interface_vars
[i
]->type
) {
1288 fields
[i
].type
= interface_vars
[i
]->type
;
1289 interface_type_changed
= true;
1292 if (!interface_type_changed
) {
1296 glsl_interface_packing packing
=
1297 (glsl_interface_packing
) ifc_type
->interface_packing
;
1298 const glsl_type
*new_ifc_type
=
1299 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1302 for (unsigned i
= 0; i
< num_fields
; i
++) {
1303 if (interface_vars
[i
] != NULL
)
1304 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1309 * Memory context used to allocate the data in \c unnamed_interfaces.
1314 * Hash table from const glsl_type * to an array of ir_variable *'s
1315 * pointing to the ir_variables constituting each unnamed interface block.
1317 hash_table
*unnamed_interfaces
;
1321 * Performs the cross-validation of layout qualifiers specified in
1322 * redeclaration of gl_FragCoord for the attached fragment shaders,
1323 * and propagates them to the linked FS and linked shader program.
1326 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1327 struct gl_shader
*linked_shader
,
1328 struct gl_shader
**shader_list
,
1329 unsigned num_shaders
)
1331 linked_shader
->redeclares_gl_fragcoord
= false;
1332 linked_shader
->uses_gl_fragcoord
= false;
1333 linked_shader
->origin_upper_left
= false;
1334 linked_shader
->pixel_center_integer
= false;
1336 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1337 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1340 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1341 struct gl_shader
*shader
= shader_list
[i
];
1342 /* From the GLSL 1.50 spec, page 39:
1344 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1345 * it must be redeclared in all the fragment shaders in that program
1346 * that have a static use gl_FragCoord."
1348 * Exclude the case when one of the 'linked_shader' or 'shader' redeclares
1349 * gl_FragCoord with no layout qualifiers but the other one doesn't
1350 * redeclare it. If we strictly follow GLSL 1.50 spec's language, it
1351 * should be a link error. But, generating link error for this case will
1352 * be a wrong behaviour which spec didn't intend to do and it could also
1353 * break some applications.
1355 if ((linked_shader
->redeclares_gl_fragcoord
1356 && !shader
->redeclares_gl_fragcoord
1357 && shader
->uses_gl_fragcoord
1358 && (linked_shader
->origin_upper_left
1359 || linked_shader
->pixel_center_integer
))
1360 || (shader
->redeclares_gl_fragcoord
1361 && !linked_shader
->redeclares_gl_fragcoord
1362 && linked_shader
->uses_gl_fragcoord
1363 && (shader
->origin_upper_left
1364 || shader
->pixel_center_integer
))) {
1365 linker_error(prog
, "fragment shader defined with conflicting "
1366 "layout qualifiers for gl_FragCoord\n");
1369 /* From the GLSL 1.50 spec, page 39:
1371 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1372 * single program must have the same set of qualifiers."
1374 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1375 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1376 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1377 linker_error(prog
, "fragment shader defined with conflicting "
1378 "layout qualifiers for gl_FragCoord\n");
1381 /* Update the linked shader state. Note that uses_gl_fragcoord should
1382 * accumulate the results. The other values should replace. If there
1383 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1384 * are already known to be the same.
1386 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1387 linked_shader
->redeclares_gl_fragcoord
=
1388 shader
->redeclares_gl_fragcoord
;
1389 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1390 || shader
->uses_gl_fragcoord
;
1391 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1392 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1398 * Performs the cross-validation of geometry shader max_vertices and
1399 * primitive type layout qualifiers for the attached geometry shaders,
1400 * and propagates them to the linked GS and linked shader program.
1403 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1404 struct gl_shader
*linked_shader
,
1405 struct gl_shader
**shader_list
,
1406 unsigned num_shaders
)
1408 linked_shader
->Geom
.VerticesOut
= 0;
1409 linked_shader
->Geom
.Invocations
= 0;
1410 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1411 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1413 /* No in/out qualifiers defined for anything but GLSL 1.50+
1414 * geometry shaders so far.
1416 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1419 /* From the GLSL 1.50 spec, page 46:
1421 * "All geometry shader output layout declarations in a program
1422 * must declare the same layout and same value for
1423 * max_vertices. There must be at least one geometry output
1424 * layout declaration somewhere in a program, but not all
1425 * geometry shaders (compilation units) are required to
1429 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1430 struct gl_shader
*shader
= shader_list
[i
];
1432 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1433 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1434 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1435 linker_error(prog
, "geometry shader defined with conflicting "
1439 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1442 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1443 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1444 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1445 linker_error(prog
, "geometry shader defined with conflicting "
1449 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1452 if (shader
->Geom
.VerticesOut
!= 0) {
1453 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1454 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1455 linker_error(prog
, "geometry shader defined with conflicting "
1456 "output vertex count (%d and %d)\n",
1457 linked_shader
->Geom
.VerticesOut
,
1458 shader
->Geom
.VerticesOut
);
1461 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1464 if (shader
->Geom
.Invocations
!= 0) {
1465 if (linked_shader
->Geom
.Invocations
!= 0 &&
1466 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1467 linker_error(prog
, "geometry shader defined with conflicting "
1468 "invocation count (%d and %d)\n",
1469 linked_shader
->Geom
.Invocations
,
1470 shader
->Geom
.Invocations
);
1473 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1477 /* Just do the intrastage -> interstage propagation right now,
1478 * since we already know we're in the right type of shader program
1481 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1483 "geometry shader didn't declare primitive input type\n");
1486 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1488 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1490 "geometry shader didn't declare primitive output type\n");
1493 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1495 if (linked_shader
->Geom
.VerticesOut
== 0) {
1497 "geometry shader didn't declare max_vertices\n");
1500 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1502 if (linked_shader
->Geom
.Invocations
== 0)
1503 linked_shader
->Geom
.Invocations
= 1;
1505 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1510 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1511 * qualifiers for the attached compute shaders, and propagate them to the
1512 * linked CS and linked shader program.
1515 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1516 struct gl_shader
*linked_shader
,
1517 struct gl_shader
**shader_list
,
1518 unsigned num_shaders
)
1520 for (int i
= 0; i
< 3; i
++)
1521 linked_shader
->Comp
.LocalSize
[i
] = 0;
1523 /* This function is called for all shader stages, but it only has an effect
1524 * for compute shaders.
1526 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1529 /* From the ARB_compute_shader spec, in the section describing local size
1532 * If multiple compute shaders attached to a single program object
1533 * declare local work-group size, the declarations must be identical;
1534 * otherwise a link-time error results. Furthermore, if a program
1535 * object contains any compute shaders, at least one must contain an
1536 * input layout qualifier specifying the local work sizes of the
1537 * program, or a link-time error will occur.
1539 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1540 struct gl_shader
*shader
= shader_list
[sh
];
1542 if (shader
->Comp
.LocalSize
[0] != 0) {
1543 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1544 for (int i
= 0; i
< 3; i
++) {
1545 if (linked_shader
->Comp
.LocalSize
[i
] !=
1546 shader
->Comp
.LocalSize
[i
]) {
1547 linker_error(prog
, "compute shader defined with conflicting "
1553 for (int i
= 0; i
< 3; i
++)
1554 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1558 /* Just do the intrastage -> interstage propagation right now,
1559 * since we already know we're in the right type of shader program
1562 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1563 linker_error(prog
, "compute shader didn't declare local size\n");
1566 for (int i
= 0; i
< 3; i
++)
1567 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1572 * Combine a group of shaders for a single stage to generate a linked shader
1575 * If this function is supplied a single shader, it is cloned, and the new
1576 * shader is returned.
1578 static struct gl_shader
*
1579 link_intrastage_shaders(void *mem_ctx
,
1580 struct gl_context
*ctx
,
1581 struct gl_shader_program
*prog
,
1582 struct gl_shader
**shader_list
,
1583 unsigned num_shaders
)
1585 struct gl_uniform_block
*uniform_blocks
= NULL
;
1587 /* Check that global variables defined in multiple shaders are consistent.
1589 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1590 if (!prog
->LinkStatus
)
1593 /* Check that interface blocks defined in multiple shaders are consistent.
1595 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1597 if (!prog
->LinkStatus
)
1600 /* Link up uniform blocks defined within this stage. */
1601 const unsigned num_uniform_blocks
=
1602 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1604 if (!prog
->LinkStatus
)
1607 /* Check that there is only a single definition of each function signature
1608 * across all shaders.
1610 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1611 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1612 ir_function
*const f
= node
->as_function();
1617 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1618 ir_function
*const other
=
1619 shader_list
[j
]->symbols
->get_function(f
->name
);
1621 /* If the other shader has no function (and therefore no function
1622 * signatures) with the same name, skip to the next shader.
1627 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1628 if (!sig
->is_defined
|| sig
->is_builtin())
1631 ir_function_signature
*other_sig
=
1632 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1634 if ((other_sig
!= NULL
) && other_sig
->is_defined
1635 && !other_sig
->is_builtin()) {
1636 linker_error(prog
, "function `%s' is multiply defined",
1645 /* Find the shader that defines main, and make a clone of it.
1647 * Starting with the clone, search for undefined references. If one is
1648 * found, find the shader that defines it. Clone the reference and add
1649 * it to the shader. Repeat until there are no undefined references or
1650 * until a reference cannot be resolved.
1652 gl_shader
*main
= NULL
;
1653 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1654 if (link_get_main_function_signature(shader_list
[i
]) != NULL
) {
1655 main
= shader_list
[i
];
1661 linker_error(prog
, "%s shader lacks `main'\n",
1662 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
1666 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1667 linked
->ir
= new(linked
) exec_list
;
1668 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1670 linked
->UniformBlocks
= uniform_blocks
;
1671 linked
->NumUniformBlocks
= num_uniform_blocks
;
1672 ralloc_steal(linked
, linked
->UniformBlocks
);
1674 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1675 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1676 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1678 populate_symbol_table(linked
);
1680 /* The a pointer to the main function in the final linked shader (i.e., the
1681 * copy of the original shader that contained the main function).
1683 ir_function_signature
*const main_sig
=
1684 link_get_main_function_signature(linked
);
1686 /* Move any instructions other than variable declarations or function
1687 * declarations into main.
1689 exec_node
*insertion_point
=
1690 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1693 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1694 if (shader_list
[i
] == main
)
1697 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1698 insertion_point
, true, linked
);
1701 /* Check if any shader needs built-in functions. */
1702 bool need_builtins
= false;
1703 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1704 if (shader_list
[i
]->uses_builtin_functions
) {
1705 need_builtins
= true;
1711 if (need_builtins
) {
1712 /* Make a temporary array one larger than shader_list, which will hold
1713 * the built-in function shader as well.
1715 gl_shader
**linking_shaders
= (gl_shader
**)
1716 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
1717 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
1718 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
1720 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
1722 free(linking_shaders
);
1724 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
1729 ctx
->Driver
.DeleteShader(ctx
, linked
);
1733 /* At this point linked should contain all of the linked IR, so
1734 * validate it to make sure nothing went wrong.
1736 validate_ir_tree(linked
->ir
);
1738 /* Set the size of geometry shader input arrays */
1739 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
1740 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1741 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1742 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
1743 ir
->accept(&input_resize_visitor
);
1747 if (ctx
->Const
.VertexID_is_zero_based
)
1748 lower_vertex_id(linked
);
1750 /* Make a pass over all variable declarations to ensure that arrays with
1751 * unspecified sizes have a size specified. The size is inferred from the
1752 * max_array_access field.
1754 array_sizing_visitor v
;
1756 v
.fixup_unnamed_interface_types();
1762 * Update the sizes of linked shader uniform arrays to the maximum
1765 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1767 * If one or more elements of an array are active,
1768 * GetActiveUniform will return the name of the array in name,
1769 * subject to the restrictions listed above. The type of the array
1770 * is returned in type. The size parameter contains the highest
1771 * array element index used, plus one. The compiler or linker
1772 * determines the highest index used. There will be only one
1773 * active uniform reported by the GL per uniform array.
1777 update_array_sizes(struct gl_shader_program
*prog
)
1779 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1780 if (prog
->_LinkedShaders
[i
] == NULL
)
1783 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
1784 ir_variable
*const var
= node
->as_variable();
1786 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
1787 !var
->type
->is_array())
1790 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1791 * will not be eliminated. Since we always do std140, just
1792 * don't resize arrays in UBOs.
1794 * Atomic counters are supposed to get deterministic
1795 * locations assigned based on the declaration ordering and
1796 * sizes, array compaction would mess that up.
1798 if (var
->is_in_uniform_block() || var
->type
->contains_atomic())
1801 unsigned int size
= var
->data
.max_array_access
;
1802 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
1803 if (prog
->_LinkedShaders
[j
] == NULL
)
1806 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
1807 ir_variable
*other_var
= node2
->as_variable();
1811 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1812 other_var
->data
.max_array_access
> size
) {
1813 size
= other_var
->data
.max_array_access
;
1818 if (size
+ 1 != var
->type
->length
) {
1819 /* If this is a built-in uniform (i.e., it's backed by some
1820 * fixed-function state), adjust the number of state slots to
1821 * match the new array size. The number of slots per array entry
1822 * is not known. It seems safe to assume that the total number of
1823 * slots is an integer multiple of the number of array elements.
1824 * Determine the number of slots per array element by dividing by
1825 * the old (total) size.
1827 if (var
->num_state_slots
> 0) {
1828 var
->num_state_slots
= (size
+ 1)
1829 * (var
->num_state_slots
/ var
->type
->length
);
1832 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1834 /* FINISHME: We should update the types of array
1835 * dereferences of this variable now.
1843 * Find a contiguous set of available bits in a bitmask.
1845 * \param used_mask Bits representing used (1) and unused (0) locations
1846 * \param needed_count Number of contiguous bits needed.
1849 * Base location of the available bits on success or -1 on failure.
1852 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1854 unsigned needed_mask
= (1 << needed_count
) - 1;
1855 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1857 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1858 * cannot optimize possibly infinite loops" for the loop below.
1860 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1863 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1864 if ((needed_mask
& ~used_mask
) == needed_mask
)
1875 * Assign locations for either VS inputs for FS outputs
1877 * \param prog Shader program whose variables need locations assigned
1878 * \param target_index Selector for the program target to receive location
1879 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1880 * \c MESA_SHADER_FRAGMENT.
1881 * \param max_index Maximum number of generic locations. This corresponds
1882 * to either the maximum number of draw buffers or the
1883 * maximum number of generic attributes.
1886 * If locations are successfully assigned, true is returned. Otherwise an
1887 * error is emitted to the shader link log and false is returned.
1890 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1891 unsigned target_index
,
1894 /* Mark invalid locations as being used.
1896 unsigned used_locations
= (max_index
>= 32)
1897 ? ~0 : ~((1 << max_index
) - 1);
1899 assert((target_index
== MESA_SHADER_VERTEX
)
1900 || (target_index
== MESA_SHADER_FRAGMENT
));
1902 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1906 /* Operate in a total of four passes.
1908 * 1. Invalidate the location assignments for all vertex shader inputs.
1910 * 2. Assign locations for inputs that have user-defined (via
1911 * glBindVertexAttribLocation) locations and outputs that have
1912 * user-defined locations (via glBindFragDataLocation).
1914 * 3. Sort the attributes without assigned locations by number of slots
1915 * required in decreasing order. Fragmentation caused by attribute
1916 * locations assigned by the application may prevent large attributes
1917 * from having enough contiguous space.
1919 * 4. Assign locations to any inputs without assigned locations.
1922 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1923 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1925 const enum ir_variable_mode direction
=
1926 (target_index
== MESA_SHADER_VERTEX
)
1927 ? ir_var_shader_in
: ir_var_shader_out
;
1930 /* Temporary storage for the set of attributes that need locations assigned.
1936 /* Used below in the call to qsort. */
1937 static int compare(const void *a
, const void *b
)
1939 const temp_attr
*const l
= (const temp_attr
*) a
;
1940 const temp_attr
*const r
= (const temp_attr
*) b
;
1942 /* Reversed because we want a descending order sort below. */
1943 return r
->slots
- l
->slots
;
1947 unsigned num_attr
= 0;
1949 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
1950 ir_variable
*const var
= node
->as_variable();
1952 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
1955 if (var
->data
.explicit_location
) {
1956 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
1957 || (var
->data
.location
< 0)) {
1959 "invalid explicit location %d specified for `%s'\n",
1960 (var
->data
.location
< 0)
1961 ? var
->data
.location
1962 : var
->data
.location
- generic_base
,
1966 } else if (target_index
== MESA_SHADER_VERTEX
) {
1969 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1970 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1971 var
->data
.location
= binding
;
1972 var
->data
.is_unmatched_generic_inout
= 0;
1974 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1978 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1979 assert(binding
>= FRAG_RESULT_DATA0
);
1980 var
->data
.location
= binding
;
1981 var
->data
.is_unmatched_generic_inout
= 0;
1983 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1984 var
->data
.index
= index
;
1989 /* If the variable is not a built-in and has a location statically
1990 * assigned in the shader (presumably via a layout qualifier), make sure
1991 * that it doesn't collide with other assigned locations. Otherwise,
1992 * add it to the list of variables that need linker-assigned locations.
1994 const unsigned slots
= var
->type
->count_attribute_slots();
1995 if (var
->data
.location
!= -1) {
1996 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
1997 /* From page 61 of the OpenGL 4.0 spec:
1999 * "LinkProgram will fail if the attribute bindings assigned
2000 * by BindAttribLocation do not leave not enough space to
2001 * assign a location for an active matrix attribute or an
2002 * active attribute array, both of which require multiple
2003 * contiguous generic attributes."
2005 * I think above text prohibits the aliasing of explicit and
2006 * automatic assignments. But, aliasing is allowed in manual
2007 * assignments of attribute locations. See below comments for
2010 * From OpenGL 4.0 spec, page 61:
2012 * "It is possible for an application to bind more than one
2013 * attribute name to the same location. This is referred to as
2014 * aliasing. This will only work if only one of the aliased
2015 * attributes is active in the executable program, or if no
2016 * path through the shader consumes more than one attribute of
2017 * a set of attributes aliased to the same location. A link
2018 * error can occur if the linker determines that every path
2019 * through the shader consumes multiple aliased attributes,
2020 * but implementations are not required to generate an error
2023 * From GLSL 4.30 spec, page 54:
2025 * "A program will fail to link if any two non-vertex shader
2026 * input variables are assigned to the same location. For
2027 * vertex shaders, multiple input variables may be assigned
2028 * to the same location using either layout qualifiers or via
2029 * the OpenGL API. However, such aliasing is intended only to
2030 * support vertex shaders where each execution path accesses
2031 * at most one input per each location. Implementations are
2032 * permitted, but not required, to generate link-time errors
2033 * if they detect that every path through the vertex shader
2034 * executable accesses multiple inputs assigned to any single
2035 * location. For all shader types, a program will fail to link
2036 * if explicit location assignments leave the linker unable
2037 * to find space for other variables without explicit
2040 * From OpenGL ES 3.0 spec, page 56:
2042 * "Binding more than one attribute name to the same location
2043 * is referred to as aliasing, and is not permitted in OpenGL
2044 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2045 * fail when this condition exists. However, aliasing is
2046 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2047 * This will only work if only one of the aliased attributes
2048 * is active in the executable program, or if no path through
2049 * the shader consumes more than one attribute of a set of
2050 * attributes aliased to the same location. A link error can
2051 * occur if the linker determines that every path through the
2052 * shader consumes multiple aliased attributes, but implemen-
2053 * tations are not required to generate an error in this case."
2055 * After looking at above references from OpenGL, OpenGL ES and
2056 * GLSL specifications, we allow aliasing of vertex input variables
2057 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2059 * NOTE: This is not required by the spec but its worth mentioning
2060 * here that we're not doing anything to make sure that no path
2061 * through the vertex shader executable accesses multiple inputs
2062 * assigned to any single location.
2065 /* Mask representing the contiguous slots that will be used by
2068 const unsigned attr
= var
->data
.location
- generic_base
;
2069 const unsigned use_mask
= (1 << slots
) - 1;
2070 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2071 ? "vertex shader input" : "fragment shader output";
2073 /* Generate a link error if the requested locations for this
2074 * attribute exceed the maximum allowed attribute location.
2076 if (attr
+ slots
> max_index
) {
2078 "insufficient contiguous locations "
2079 "available for %s `%s' %d %d %d", string
,
2080 var
->name
, used_locations
, use_mask
, attr
);
2084 /* Generate a link error if the set of bits requested for this
2085 * attribute overlaps any previously allocated bits.
2087 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2088 if (target_index
== MESA_SHADER_FRAGMENT
||
2089 (prog
->IsES
&& prog
->Version
>= 300)) {
2091 "overlapping location is assigned "
2092 "to %s `%s' %d %d %d\n", string
,
2093 var
->name
, used_locations
, use_mask
, attr
);
2096 linker_warning(prog
,
2097 "overlapping location is assigned "
2098 "to %s `%s' %d %d %d\n", string
,
2099 var
->name
, used_locations
, use_mask
, attr
);
2103 used_locations
|= (use_mask
<< attr
);
2109 to_assign
[num_attr
].slots
= slots
;
2110 to_assign
[num_attr
].var
= var
;
2114 /* If all of the attributes were assigned locations by the application (or
2115 * are built-in attributes with fixed locations), return early. This should
2116 * be the common case.
2121 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2123 if (target_index
== MESA_SHADER_VERTEX
) {
2124 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2125 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2126 * reserved to prevent it from being automatically allocated below.
2128 find_deref_visitor
find("gl_Vertex");
2130 if (find
.variable_found())
2131 used_locations
|= (1 << 0);
2134 for (unsigned i
= 0; i
< num_attr
; i
++) {
2135 /* Mask representing the contiguous slots that will be used by this
2138 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2140 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2143 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2144 ? "vertex shader input" : "fragment shader output";
2147 "insufficient contiguous locations "
2148 "available for %s `%s'",
2149 string
, to_assign
[i
].var
->name
);
2153 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2154 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2155 used_locations
|= (use_mask
<< location
);
2163 * Demote shader inputs and outputs that are not used in other stages
2166 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2168 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2169 ir_variable
*const var
= node
->as_variable();
2171 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2174 /* A shader 'in' or 'out' variable is only really an input or output if
2175 * its value is used by other shader stages. This will cause the variable
2176 * to have a location assigned.
2178 if (var
->data
.is_unmatched_generic_inout
) {
2179 var
->data
.mode
= ir_var_auto
;
2186 * Store the gl_FragDepth layout in the gl_shader_program struct.
2189 store_fragdepth_layout(struct gl_shader_program
*prog
)
2191 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2195 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2197 /* We don't look up the gl_FragDepth symbol directly because if
2198 * gl_FragDepth is not used in the shader, it's removed from the IR.
2199 * However, the symbol won't be removed from the symbol table.
2201 * We're only interested in the cases where the variable is NOT removed
2204 foreach_in_list(ir_instruction
, node
, ir
) {
2205 ir_variable
*const var
= node
->as_variable();
2207 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2211 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2212 switch (var
->data
.depth_layout
) {
2213 case ir_depth_layout_none
:
2214 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2216 case ir_depth_layout_any
:
2217 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2219 case ir_depth_layout_greater
:
2220 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2222 case ir_depth_layout_less
:
2223 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2225 case ir_depth_layout_unchanged
:
2226 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2237 * Validate the resources used by a program versus the implementation limits
2240 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2242 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2243 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2248 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2249 linker_error(prog
, "Too many %s shader texture samplers",
2250 _mesa_shader_stage_to_string(i
));
2253 if (sh
->num_uniform_components
>
2254 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2255 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2256 linker_warning(prog
, "Too many %s shader default uniform block "
2257 "components, but the driver will try to optimize "
2258 "them out; this is non-portable out-of-spec "
2260 _mesa_shader_stage_to_string(i
));
2262 linker_error(prog
, "Too many %s shader default uniform block "
2264 _mesa_shader_stage_to_string(i
));
2268 if (sh
->num_combined_uniform_components
>
2269 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2270 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2271 linker_warning(prog
, "Too many %s shader uniform components, "
2272 "but the driver will try to optimize them out; "
2273 "this is non-portable out-of-spec behavior\n",
2274 _mesa_shader_stage_to_string(i
));
2276 linker_error(prog
, "Too many %s shader uniform components",
2277 _mesa_shader_stage_to_string(i
));
2282 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2283 unsigned total_uniform_blocks
= 0;
2285 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2286 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2287 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2289 total_uniform_blocks
++;
2293 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2294 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
2295 prog
->NumUniformBlocks
,
2296 ctx
->Const
.MaxCombinedUniformBlocks
);
2298 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2299 const unsigned max_uniform_blocks
=
2300 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2301 if (blocks
[i
] > max_uniform_blocks
) {
2302 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
2303 _mesa_shader_stage_to_string(i
),
2305 max_uniform_blocks
);
2314 * Validate shader image resources.
2317 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2319 unsigned total_image_units
= 0;
2320 unsigned fragment_outputs
= 0;
2322 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2325 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2326 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2329 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2330 linker_error(prog
, "Too many %s shader image uniforms",
2331 _mesa_shader_stage_to_string(i
));
2333 total_image_units
+= sh
->NumImages
;
2335 if (i
== MESA_SHADER_FRAGMENT
) {
2336 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2337 ir_variable
*var
= node
->as_variable();
2338 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2339 fragment_outputs
+= var
->type
->count_attribute_slots();
2345 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2346 linker_error(prog
, "Too many combined image uniforms");
2348 if (total_image_units
+ fragment_outputs
>
2349 ctx
->Const
.MaxCombinedImageUnitsAndFragmentOutputs
)
2350 linker_error(prog
, "Too many combined image uniforms and fragment outputs");
2355 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2356 * for a variable, checks for overlaps between other uniforms using explicit
2360 reserve_explicit_locations(struct gl_shader_program
*prog
,
2361 string_to_uint_map
*map
, ir_variable
*var
)
2363 unsigned slots
= var
->type
->uniform_locations();
2364 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2366 /* Resize remap table if locations do not fit in the current one. */
2367 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2368 prog
->UniformRemapTable
=
2369 reralloc(prog
, prog
->UniformRemapTable
,
2370 gl_uniform_storage
*,
2373 if (!prog
->UniformRemapTable
) {
2374 linker_error(prog
, "Out of memory during linking.");
2378 /* Initialize allocated space. */
2379 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2380 prog
->UniformRemapTable
[i
] = NULL
;
2382 prog
->NumUniformRemapTable
= max_loc
+ 1;
2385 for (unsigned i
= 0; i
< slots
; i
++) {
2386 unsigned loc
= var
->data
.location
+ i
;
2388 /* Check if location is already used. */
2389 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2391 /* Possibly same uniform from a different stage, this is ok. */
2393 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2396 /* ARB_explicit_uniform_location specification states:
2398 * "No two default-block uniform variables in the program can have
2399 * the same location, even if they are unused, otherwise a compiler
2400 * or linker error will be generated."
2403 "location qualifier for uniform %s overlaps"
2404 "previously used location",
2409 /* Initialize location as inactive before optimization
2410 * rounds and location assignment.
2412 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2415 /* Note, base location used for arrays. */
2416 map
->put(var
->data
.location
, var
->name
);
2422 * Check and reserve all explicit uniform locations, called before
2423 * any optimizations happen to handle also inactive uniforms and
2424 * inactive array elements that may get trimmed away.
2427 check_explicit_uniform_locations(struct gl_context
*ctx
,
2428 struct gl_shader_program
*prog
)
2430 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
2433 /* This map is used to detect if overlapping explicit locations
2434 * occur with the same uniform (from different stage) or a different one.
2436 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
2439 linker_error(prog
, "Out of memory during linking.");
2443 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2444 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2449 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2450 ir_variable
*var
= node
->as_variable();
2451 if ((var
&& var
->data
.mode
== ir_var_uniform
) &&
2452 var
->data
.explicit_location
) {
2453 if (!reserve_explicit_locations(prog
, uniform_map
, var
)) {
2465 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2467 tfeedback_decl
*tfeedback_decls
= NULL
;
2468 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2470 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2472 prog
->LinkStatus
= true; /* All error paths will set this to false */
2473 prog
->Validated
= false;
2474 prog
->_Used
= false;
2476 ralloc_free(prog
->InfoLog
);
2477 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2479 ralloc_free(prog
->UniformBlocks
);
2480 prog
->UniformBlocks
= NULL
;
2481 prog
->NumUniformBlocks
= 0;
2482 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2483 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2484 prog
->UniformBlockStageIndex
[i
] = NULL
;
2487 ralloc_free(prog
->AtomicBuffers
);
2488 prog
->AtomicBuffers
= NULL
;
2489 prog
->NumAtomicBuffers
= 0;
2490 prog
->ARB_fragment_coord_conventions_enable
= false;
2492 /* Separate the shaders into groups based on their type.
2494 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
2495 unsigned num_shaders
[MESA_SHADER_STAGES
];
2497 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2498 shader_list
[i
] = (struct gl_shader
**)
2499 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
2503 unsigned min_version
= UINT_MAX
;
2504 unsigned max_version
= 0;
2505 const bool is_es_prog
=
2506 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
2507 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2508 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2509 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2511 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
2512 linker_error(prog
, "all shaders must use same shading "
2513 "language version\n");
2517 prog
->ARB_fragment_coord_conventions_enable
|=
2518 prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
;
2520 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
2521 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
2522 num_shaders
[shader_type
]++;
2525 /* In desktop GLSL, different shader versions may be linked together. In
2526 * GLSL ES, all shader versions must be the same.
2528 if (is_es_prog
&& min_version
!= max_version
) {
2529 linker_error(prog
, "all shaders must use same shading "
2530 "language version\n");
2534 prog
->Version
= max_version
;
2535 prog
->IsES
= is_es_prog
;
2537 /* Geometry shaders have to be linked with vertex shaders.
2539 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
2540 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
2541 !prog
->SeparateShader
) {
2542 linker_error(prog
, "Geometry shader must be linked with "
2547 /* Compute shaders have additional restrictions. */
2548 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
2549 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
2550 linker_error(prog
, "Compute shaders may not be linked with any other "
2551 "type of shader\n");
2554 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2555 if (prog
->_LinkedShaders
[i
] != NULL
)
2556 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2558 prog
->_LinkedShaders
[i
] = NULL
;
2561 /* Link all shaders for a particular stage and validate the result.
2563 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
2564 if (num_shaders
[stage
] > 0) {
2565 gl_shader
*const sh
=
2566 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
2567 num_shaders
[stage
]);
2569 if (!prog
->LinkStatus
)
2573 case MESA_SHADER_VERTEX
:
2574 validate_vertex_shader_executable(prog
, sh
);
2576 case MESA_SHADER_GEOMETRY
:
2577 validate_geometry_shader_executable(prog
, sh
);
2579 case MESA_SHADER_FRAGMENT
:
2580 validate_fragment_shader_executable(prog
, sh
);
2583 if (!prog
->LinkStatus
)
2586 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
2590 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
2591 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
2592 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
2593 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
2595 prog
->LastClipDistanceArraySize
= 0; /* Not used */
2597 /* Here begins the inter-stage linking phase. Some initial validation is
2598 * performed, then locations are assigned for uniforms, attributes, and
2601 cross_validate_uniforms(prog
);
2602 if (!prog
->LinkStatus
)
2607 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
2608 if (prog
->_LinkedShaders
[prev
] != NULL
)
2612 check_explicit_uniform_locations(ctx
, prog
);
2613 if (!prog
->LinkStatus
)
2616 /* Validate the inputs of each stage with the output of the preceding
2619 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2620 if (prog
->_LinkedShaders
[i
] == NULL
)
2623 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
2624 prog
->_LinkedShaders
[i
]);
2625 if (!prog
->LinkStatus
)
2628 cross_validate_outputs_to_inputs(prog
,
2629 prog
->_LinkedShaders
[prev
],
2630 prog
->_LinkedShaders
[i
]);
2631 if (!prog
->LinkStatus
)
2637 /* Cross-validate uniform blocks between shader stages */
2638 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
2639 MESA_SHADER_STAGES
);
2640 if (!prog
->LinkStatus
)
2643 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2644 if (prog
->_LinkedShaders
[i
] != NULL
)
2645 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
2648 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2649 * it before optimization because we want most of the checks to get
2650 * dropped thanks to constant propagation.
2652 * This rule also applies to GLSL ES 3.00.
2654 if (max_version
>= (is_es_prog
? 300 : 130)) {
2655 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2657 lower_discard_flow(sh
->ir
);
2661 if (!interstage_cross_validate_uniform_blocks(prog
))
2664 /* Do common optimization before assigning storage for attributes,
2665 * uniforms, and varyings. Later optimization could possibly make
2666 * some of that unused.
2668 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2669 if (prog
->_LinkedShaders
[i
] == NULL
)
2672 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2673 if (!prog
->LinkStatus
)
2676 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
2677 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2680 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
2681 &ctx
->Const
.ShaderCompilerOptions
[i
],
2682 ctx
->Const
.NativeIntegers
))
2686 /* Check and validate stream emissions in geometry shaders */
2687 validate_geometry_shader_emissions(ctx
, prog
);
2689 /* Mark all generic shader inputs and outputs as unpaired. */
2690 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
2691 if (prog
->_LinkedShaders
[i
] != NULL
) {
2692 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
2696 /* FINISHME: The value of the max_attribute_index parameter is
2697 * FINISHME: implementation dependent based on the value of
2698 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2699 * FINISHME: at least 16, so hardcode 16 for now.
2701 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2705 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2710 for (first
= 0; first
<= MESA_SHADER_FRAGMENT
; first
++) {
2711 if (prog
->_LinkedShaders
[first
] != NULL
)
2715 if (num_tfeedback_decls
!= 0) {
2716 /* From GL_EXT_transform_feedback:
2717 * A program will fail to link if:
2719 * * the <count> specified by TransformFeedbackVaryingsEXT is
2720 * non-zero, but the program object has no vertex or geometry
2723 if (first
== MESA_SHADER_FRAGMENT
) {
2724 linker_error(prog
, "Transform feedback varyings specified, but "
2725 "no vertex or geometry shader is present.");
2729 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2730 prog
->TransformFeedback
.NumVarying
);
2731 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2732 prog
->TransformFeedback
.VaryingNames
,
2737 /* Linking the stages in the opposite order (from fragment to vertex)
2738 * ensures that inter-shader outputs written to in an earlier stage are
2739 * eliminated if they are (transitively) not used in a later stage.
2742 for (last
= MESA_SHADER_FRAGMENT
; last
>= 0; last
--) {
2743 if (prog
->_LinkedShaders
[last
] != NULL
)
2747 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2748 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2750 if (num_tfeedback_decls
!= 0 || prog
->SeparateShader
) {
2751 /* There was no fragment shader, but we still have to assign varying
2752 * locations for use by transform feedback.
2754 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2756 num_tfeedback_decls
, tfeedback_decls
,
2761 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2762 num_tfeedback_decls
, tfeedback_decls
);
2764 if (!prog
->SeparateShader
)
2765 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2767 /* Eliminate code that is now dead due to unused outputs being demoted.
2769 while (do_dead_code(sh
->ir
, false))
2772 else if (first
== MESA_SHADER_FRAGMENT
) {
2773 /* If the program only contains a fragment shader...
2775 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2777 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2778 num_tfeedback_decls
, tfeedback_decls
);
2780 if (prog
->SeparateShader
) {
2781 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2782 NULL
/* producer */,
2784 0 /* num_tfeedback_decls */,
2785 NULL
/* tfeedback_decls */,
2786 0 /* gs_input_vertices */))
2789 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2791 while (do_dead_code(sh
->ir
, false))
2796 for (int i
= next
- 1; i
>= 0; i
--) {
2797 if (prog
->_LinkedShaders
[i
] == NULL
)
2800 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2801 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2802 unsigned gs_input_vertices
=
2803 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2805 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2806 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2807 tfeedback_decls
, gs_input_vertices
))
2810 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2811 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2814 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2815 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2817 /* Eliminate code that is now dead due to unused outputs being demoted.
2819 while (do_dead_code(sh_i
->ir
, false))
2821 while (do_dead_code(sh_next
->ir
, false))
2824 /* This must be done after all dead varyings are eliminated. */
2825 if (!check_against_output_limit(ctx
, prog
, sh_i
))
2827 if (!check_against_input_limit(ctx
, prog
, sh_next
))
2833 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2836 update_array_sizes(prog
);
2837 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
2838 link_assign_atomic_counter_resources(ctx
, prog
);
2839 store_fragdepth_layout(prog
);
2841 check_resources(ctx
, prog
);
2842 check_image_resources(ctx
, prog
);
2843 link_check_atomic_counter_resources(ctx
, prog
);
2845 if (!prog
->LinkStatus
)
2848 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2849 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
2850 * anything about shader linking when one of the shaders (vertex or
2851 * fragment shader) is absent. So, the extension shouldn't change the
2852 * behavior specified in GLSL specification.
2854 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
2855 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2856 linker_error(prog
, "program lacks a vertex shader\n");
2857 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2858 linker_error(prog
, "program lacks a fragment shader\n");
2862 /* FINISHME: Assign fragment shader output locations. */
2865 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2866 free(shader_list
[i
]);
2867 if (prog
->_LinkedShaders
[i
] == NULL
)
2870 /* Do a final validation step to make sure that the IR wasn't
2871 * invalidated by any modifications performed after intrastage linking.
2873 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
2875 /* Retain any live IR, but trash the rest. */
2876 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2878 /* The symbol table in the linked shaders may contain references to
2879 * variables that were removed (e.g., unused uniforms). Since it may
2880 * contain junk, there is no possible valid use. Delete it and set the
2883 delete prog
->_LinkedShaders
[i
]->symbols
;
2884 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2887 ralloc_free(mem_ctx
);