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"
79 #include "main/shaderobj.h"
80 #include "main/enums.h"
83 void linker_error(gl_shader_program
*, const char *, ...);
86 * Visitor that determines whether or not a variable is ever written.
88 class find_assignment_visitor
: public ir_hierarchical_visitor
{
90 find_assignment_visitor(const char *name
)
91 : name(name
), found(false)
96 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
98 ir_variable
*const var
= ir
->lhs
->variable_referenced();
100 if (strcmp(name
, var
->name
) == 0) {
105 return visit_continue_with_parent
;
108 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
110 exec_list_iterator sig_iter
= ir
->callee
->parameters
.iterator();
111 foreach_iter(exec_list_iterator
, iter
, *ir
) {
112 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
113 ir_variable
*sig_param
= (ir_variable
*)sig_iter
.get();
115 if (sig_param
->mode
== ir_var_function_out
||
116 sig_param
->mode
== ir_var_function_inout
) {
117 ir_variable
*var
= param_rval
->variable_referenced();
118 if (var
&& strcmp(name
, var
->name
) == 0) {
126 if (ir
->return_deref
!= NULL
) {
127 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
129 if (strcmp(name
, var
->name
) == 0) {
135 return visit_continue_with_parent
;
138 bool variable_found()
144 const char *name
; /**< Find writes to a variable with this name. */
145 bool found
; /**< Was a write to the variable found? */
150 * Visitor that determines whether or not a variable is ever read.
152 class find_deref_visitor
: public ir_hierarchical_visitor
{
154 find_deref_visitor(const char *name
)
155 : name(name
), found(false)
160 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
162 if (strcmp(this->name
, ir
->var
->name
) == 0) {
167 return visit_continue
;
170 bool variable_found() const
176 const char *name
; /**< Find writes to a variable with this name. */
177 bool found
; /**< Was a write to the variable found? */
181 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
183 unsigned num_vertices
;
184 gl_shader_program
*prog
;
186 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
188 this->num_vertices
= num_vertices
;
192 virtual ~geom_array_resize_visitor()
197 virtual ir_visitor_status
visit(ir_variable
*var
)
199 if (!var
->type
->is_array() || var
->mode
!= ir_var_shader_in
)
200 return visit_continue
;
202 unsigned size
= var
->type
->length
;
204 /* Generate a link error if the shader has declared this array with an
207 if (size
&& size
!= this->num_vertices
) {
208 linker_error(this->prog
, "size of array %s declared as %u, "
209 "but number of input vertices is %u\n",
210 var
->name
, size
, this->num_vertices
);
211 return visit_continue
;
214 /* Generate a link error if the shader attempts to access an input
215 * array using an index too large for its actual size assigned at link
218 if (var
->max_array_access
>= this->num_vertices
) {
219 linker_error(this->prog
, "geometry shader accesses element %i of "
220 "%s, but only %i input vertices\n",
221 var
->max_array_access
, var
->name
, this->num_vertices
);
222 return visit_continue
;
225 var
->type
= glsl_type::get_array_instance(var
->type
->element_type(),
227 var
->max_array_access
= this->num_vertices
- 1;
229 return visit_continue
;
232 /* Dereferences of input variables need to be updated so that their type
233 * matches the newly assigned type of the variable they are accessing. */
234 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
236 ir
->type
= ir
->var
->type
;
237 return visit_continue
;
240 /* Dereferences of 2D input arrays need to be updated so that their type
241 * matches the newly assigned type of the array they are accessing. */
242 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
244 const glsl_type
*const vt
= ir
->array
->type
;
246 ir
->type
= vt
->element_type();
247 return visit_continue
;
253 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
257 ralloc_strcat(&prog
->InfoLog
, "error: ");
259 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
262 prog
->LinkStatus
= false;
267 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
271 ralloc_strcat(&prog
->InfoLog
, "error: ");
273 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
280 * Given a string identifying a program resource, break it into a base name
281 * and an optional array index in square brackets.
283 * If an array index is present, \c out_base_name_end is set to point to the
284 * "[" that precedes the array index, and the array index itself is returned
287 * If no array index is present (or if the array index is negative or
288 * mal-formed), \c out_base_name_end, is set to point to the null terminator
289 * at the end of the input string, and -1 is returned.
291 * Only the final array index is parsed; if the string contains other array
292 * indices (or structure field accesses), they are left in the base name.
294 * No attempt is made to check that the base name is properly formed;
295 * typically the caller will look up the base name in a hash table, so
296 * ill-formed base names simply turn into hash table lookup failures.
299 parse_program_resource_name(const GLchar
*name
,
300 const GLchar
**out_base_name_end
)
302 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
304 * "When an integer array element or block instance number is part of
305 * the name string, it will be specified in decimal form without a "+"
306 * or "-" sign or any extra leading zeroes. Additionally, the name
307 * string will not include white space anywhere in the string."
310 const size_t len
= strlen(name
);
311 *out_base_name_end
= name
+ len
;
313 if (len
== 0 || name
[len
-1] != ']')
316 /* Walk backwards over the string looking for a non-digit character. This
317 * had better be the opening bracket for an array index.
319 * Initially, i specifies the location of the ']'. Since the string may
320 * contain only the ']' charcater, walk backwards very carefully.
323 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
326 if ((i
== 0) || name
[i
-1] != '[')
329 long array_index
= strtol(&name
[i
], NULL
, 10);
333 *out_base_name_end
= name
+ (i
- 1);
339 link_invalidate_variable_locations(gl_shader
*sh
, int input_base
,
342 foreach_list(node
, sh
->ir
) {
343 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
350 case ir_var_shader_in
:
353 case ir_var_shader_out
:
360 /* Only assign locations for generic attributes / varyings / etc.
362 if ((var
->location
>= base
) && !var
->explicit_location
)
365 if ((var
->location
== -1) && !var
->explicit_location
) {
366 var
->is_unmatched_generic_inout
= 1;
367 var
->location_frac
= 0;
369 var
->is_unmatched_generic_inout
= 0;
376 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
378 * Also check for errors based on incorrect usage of gl_ClipVertex and
381 * Return false if an error was reported.
384 analyze_clip_usage(const char *shader_type
, struct gl_shader_program
*prog
,
385 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
386 GLuint
*ClipDistanceArraySize
)
388 *ClipDistanceArraySize
= 0;
390 if (!prog
->IsES
&& prog
->Version
>= 130) {
391 /* From section 7.1 (Vertex Shader Special Variables) of the
394 * "It is an error for a shader to statically write both
395 * gl_ClipVertex and gl_ClipDistance."
397 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
398 * gl_ClipVertex nor gl_ClipDistance.
400 find_assignment_visitor
clip_vertex("gl_ClipVertex");
401 find_assignment_visitor
clip_distance("gl_ClipDistance");
403 clip_vertex
.run(shader
->ir
);
404 clip_distance
.run(shader
->ir
);
405 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
406 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
407 "and `gl_ClipDistance'\n", shader_type
);
410 *UsesClipDistance
= clip_distance
.variable_found();
411 ir_variable
*clip_distance_var
=
412 shader
->symbols
->get_variable("gl_ClipDistance");
413 if (clip_distance_var
)
414 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
416 *UsesClipDistance
= false;
422 * Verify that a vertex shader executable meets all semantic requirements.
424 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
427 * \param shader Vertex shader executable to be verified
430 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
431 struct gl_shader
*shader
)
436 /* From the GLSL 1.10 spec, page 48:
438 * "The variable gl_Position is available only in the vertex
439 * language and is intended for writing the homogeneous vertex
440 * position. All executions of a well-formed vertex shader
441 * executable must write a value into this variable. [...] The
442 * variable gl_Position is available only in the vertex
443 * language and is intended for writing the homogeneous vertex
444 * position. All executions of a well-formed vertex shader
445 * executable must write a value into this variable."
447 * while in GLSL 1.40 this text is changed to:
449 * "The variable gl_Position is available only in the vertex
450 * language and is intended for writing the homogeneous vertex
451 * position. It can be written at any time during shader
452 * execution. It may also be read back by a vertex shader
453 * after being written. This value will be used by primitive
454 * assembly, clipping, culling, and other fixed functionality
455 * operations, if present, that operate on primitives after
456 * vertex processing has occurred. Its value is undefined if
457 * the vertex shader executable does not write gl_Position."
459 * GLSL ES 3.00 is similar to GLSL 1.40--failing to write to gl_Position is
462 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
463 find_assignment_visitor
find("gl_Position");
464 find
.run(shader
->ir
);
465 if (!find
.variable_found()) {
466 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
471 analyze_clip_usage("vertex", prog
, shader
, &prog
->Vert
.UsesClipDistance
,
472 &prog
->Vert
.ClipDistanceArraySize
);
477 * Verify that a fragment shader executable meets all semantic requirements
479 * \param shader Fragment shader executable to be verified
482 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
483 struct gl_shader
*shader
)
488 find_assignment_visitor
frag_color("gl_FragColor");
489 find_assignment_visitor
frag_data("gl_FragData");
491 frag_color
.run(shader
->ir
);
492 frag_data
.run(shader
->ir
);
494 if (frag_color
.variable_found() && frag_data
.variable_found()) {
495 linker_error(prog
, "fragment shader writes to both "
496 "`gl_FragColor' and `gl_FragData'\n");
501 * Verify that a geometry shader executable meets all semantic requirements
503 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
504 * prog->Geom.ClipDistanceArraySize as a side effect.
506 * \param shader Geometry shader executable to be verified
509 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
510 struct gl_shader
*shader
)
515 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
516 prog
->Geom
.VerticesIn
= num_vertices
;
518 analyze_clip_usage("geometry", prog
, shader
, &prog
->Geom
.UsesClipDistance
,
519 &prog
->Geom
.ClipDistanceArraySize
);
524 * Generate a string describing the mode of a variable
527 mode_string(const ir_variable
*var
)
531 return (var
->read_only
) ? "global constant" : "global variable";
533 case ir_var_uniform
: return "uniform";
534 case ir_var_shader_in
: return "shader input";
535 case ir_var_shader_out
: return "shader output";
537 case ir_var_const_in
:
538 case ir_var_temporary
:
540 assert(!"Should not get here.");
541 return "invalid variable";
547 * Perform validation of global variables used across multiple shaders
550 cross_validate_globals(struct gl_shader_program
*prog
,
551 struct gl_shader
**shader_list
,
552 unsigned num_shaders
,
555 /* Examine all of the uniforms in all of the shaders and cross validate
558 glsl_symbol_table variables
;
559 for (unsigned i
= 0; i
< num_shaders
; i
++) {
560 if (shader_list
[i
] == NULL
)
563 foreach_list(node
, shader_list
[i
]->ir
) {
564 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
569 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
572 /* Don't cross validate temporaries that are at global scope. These
573 * will eventually get pulled into the shaders 'main'.
575 if (var
->mode
== ir_var_temporary
)
578 /* If a global with this name has already been seen, verify that the
579 * new instance has the same type. In addition, if the globals have
580 * initializers, the values of the initializers must be the same.
582 ir_variable
*const existing
= variables
.get_variable(var
->name
);
583 if (existing
!= NULL
) {
584 if (var
->type
!= existing
->type
) {
585 /* Consider the types to be "the same" if both types are arrays
586 * of the same type and one of the arrays is implicitly sized.
587 * In addition, set the type of the linked variable to the
588 * explicitly sized array.
590 if (var
->type
->is_array()
591 && existing
->type
->is_array()
592 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
593 && ((var
->type
->length
== 0)
594 || (existing
->type
->length
== 0))) {
595 if (var
->type
->length
!= 0) {
596 existing
->type
= var
->type
;
599 linker_error(prog
, "%s `%s' declared as type "
600 "`%s' and type `%s'\n",
602 var
->name
, var
->type
->name
,
603 existing
->type
->name
);
608 if (var
->explicit_location
) {
609 if (existing
->explicit_location
610 && (var
->location
!= existing
->location
)) {
611 linker_error(prog
, "explicit locations for %s "
612 "`%s' have differing values\n",
613 mode_string(var
), var
->name
);
617 existing
->location
= var
->location
;
618 existing
->explicit_location
= true;
621 /* From the GLSL 4.20 specification:
622 * "A link error will result if two compilation units in a program
623 * specify different integer-constant bindings for the same
624 * opaque-uniform name. However, it is not an error to specify a
625 * binding on some but not all declarations for the same name"
627 if (var
->explicit_binding
) {
628 if (existing
->explicit_binding
&&
629 var
->binding
!= existing
->binding
) {
630 linker_error(prog
, "explicit bindings for %s "
631 "`%s' have differing values\n",
632 mode_string(var
), var
->name
);
636 existing
->binding
= var
->binding
;
637 existing
->explicit_binding
= true;
640 /* Validate layout qualifiers for gl_FragDepth.
642 * From the AMD/ARB_conservative_depth specs:
644 * "If gl_FragDepth is redeclared in any fragment shader in a
645 * program, it must be redeclared in all fragment shaders in
646 * that program that have static assignments to
647 * gl_FragDepth. All redeclarations of gl_FragDepth in all
648 * fragment shaders in a single program must have the same set
651 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
652 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
653 bool layout_differs
=
654 var
->depth_layout
!= existing
->depth_layout
;
656 if (layout_declared
&& layout_differs
) {
658 "All redeclarations of gl_FragDepth in all "
659 "fragment shaders in a single program must have "
660 "the same set of qualifiers.");
663 if (var
->used
&& layout_differs
) {
665 "If gl_FragDepth is redeclared with a layout "
666 "qualifier in any fragment shader, it must be "
667 "redeclared with the same layout qualifier in "
668 "all fragment shaders that have assignments to "
673 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
675 * "If a shared global has multiple initializers, the
676 * initializers must all be constant expressions, and they
677 * must all have the same value. Otherwise, a link error will
678 * result. (A shared global having only one initializer does
679 * not require that initializer to be a constant expression.)"
681 * Previous to 4.20 the GLSL spec simply said that initializers
682 * must have the same value. In this case of non-constant
683 * initializers, this was impossible to determine. As a result,
684 * no vendor actually implemented that behavior. The 4.20
685 * behavior matches the implemented behavior of at least one other
686 * vendor, so we'll implement that for all GLSL versions.
688 if (var
->constant_initializer
!= NULL
) {
689 if (existing
->constant_initializer
!= NULL
) {
690 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
691 linker_error(prog
, "initializers for %s "
692 "`%s' have differing values\n",
693 mode_string(var
), var
->name
);
697 /* If the first-seen instance of a particular uniform did not
698 * have an initializer but a later instance does, copy the
699 * initializer to the version stored in the symbol table.
701 /* FINISHME: This is wrong. The constant_value field should
702 * FINISHME: not be modified! Imagine a case where a shader
703 * FINISHME: without an initializer is linked in two different
704 * FINISHME: programs with shaders that have differing
705 * FINISHME: initializers. Linking with the first will
706 * FINISHME: modify the shader, and linking with the second
707 * FINISHME: will fail.
709 existing
->constant_initializer
=
710 var
->constant_initializer
->clone(ralloc_parent(existing
),
715 if (var
->has_initializer
) {
716 if (existing
->has_initializer
717 && (var
->constant_initializer
== NULL
718 || existing
->constant_initializer
== NULL
)) {
720 "shared global variable `%s' has multiple "
721 "non-constant initializers.\n",
726 /* Some instance had an initializer, so keep track of that. In
727 * this location, all sorts of initializers (constant or
728 * otherwise) will propagate the existence to the variable
729 * stored in the symbol table.
731 existing
->has_initializer
= true;
734 if (existing
->invariant
!= var
->invariant
) {
735 linker_error(prog
, "declarations for %s `%s' have "
736 "mismatching invariant qualifiers\n",
737 mode_string(var
), var
->name
);
740 if (existing
->centroid
!= var
->centroid
) {
741 linker_error(prog
, "declarations for %s `%s' have "
742 "mismatching centroid qualifiers\n",
743 mode_string(var
), var
->name
);
747 variables
.add_variable(var
);
754 * Perform validation of uniforms used across multiple shader stages
757 cross_validate_uniforms(struct gl_shader_program
*prog
)
759 cross_validate_globals(prog
, prog
->_LinkedShaders
,
760 MESA_SHADER_TYPES
, true);
764 * Accumulates the array of prog->UniformBlocks and checks that all
765 * definitons of blocks agree on their contents.
768 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
770 unsigned max_num_uniform_blocks
= 0;
771 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
772 if (prog
->_LinkedShaders
[i
])
773 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
776 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
777 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
779 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
780 max_num_uniform_blocks
);
781 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
782 prog
->UniformBlockStageIndex
[i
][j
] = -1;
787 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
788 int index
= link_cross_validate_uniform_block(prog
,
789 &prog
->UniformBlocks
,
790 &prog
->NumUniformBlocks
,
791 &sh
->UniformBlocks
[j
]);
794 linker_error(prog
, "uniform block `%s' has mismatching definitions",
795 sh
->UniformBlocks
[j
].Name
);
799 prog
->UniformBlockStageIndex
[i
][index
] = j
;
808 * Populates a shaders symbol table with all global declarations
811 populate_symbol_table(gl_shader
*sh
)
813 sh
->symbols
= new(sh
) glsl_symbol_table
;
815 foreach_list(node
, sh
->ir
) {
816 ir_instruction
*const inst
= (ir_instruction
*) node
;
820 if ((func
= inst
->as_function()) != NULL
) {
821 sh
->symbols
->add_function(func
);
822 } else if ((var
= inst
->as_variable()) != NULL
) {
823 sh
->symbols
->add_variable(var
);
830 * Remap variables referenced in an instruction tree
832 * This is used when instruction trees are cloned from one shader and placed in
833 * another. These trees will contain references to \c ir_variable nodes that
834 * do not exist in the target shader. This function finds these \c ir_variable
835 * references and replaces the references with matching variables in the target
838 * If there is no matching variable in the target shader, a clone of the
839 * \c ir_variable is made and added to the target shader. The new variable is
840 * added to \b both the instruction stream and the symbol table.
842 * \param inst IR tree that is to be processed.
843 * \param symbols Symbol table containing global scope symbols in the
845 * \param instructions Instruction stream where new variable declarations
849 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
852 class remap_visitor
: public ir_hierarchical_visitor
{
854 remap_visitor(struct gl_shader
*target
,
857 this->target
= target
;
858 this->symbols
= target
->symbols
;
859 this->instructions
= target
->ir
;
863 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
865 if (ir
->var
->mode
== ir_var_temporary
) {
866 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
870 return visit_continue
;
873 ir_variable
*const existing
=
874 this->symbols
->get_variable(ir
->var
->name
);
875 if (existing
!= NULL
)
878 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
880 this->symbols
->add_variable(copy
);
881 this->instructions
->push_head(copy
);
885 return visit_continue
;
889 struct gl_shader
*target
;
890 glsl_symbol_table
*symbols
;
891 exec_list
*instructions
;
895 remap_visitor
v(target
, temps
);
902 * Move non-declarations from one instruction stream to another
904 * The intended usage pattern of this function is to pass the pointer to the
905 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
906 * pointer) for \c last and \c false for \c make_copies on the first
907 * call. Successive calls pass the return value of the previous call for
908 * \c last and \c true for \c make_copies.
910 * \param instructions Source instruction stream
911 * \param last Instruction after which new instructions should be
912 * inserted in the target instruction stream
913 * \param make_copies Flag selecting whether instructions in \c instructions
914 * should be copied (via \c ir_instruction::clone) into the
915 * target list or moved.
918 * The new "last" instruction in the target instruction stream. This pointer
919 * is suitable for use as the \c last parameter of a later call to this
923 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
924 bool make_copies
, gl_shader
*target
)
926 hash_table
*temps
= NULL
;
929 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
930 hash_table_pointer_compare
);
932 foreach_list_safe(node
, instructions
) {
933 ir_instruction
*inst
= (ir_instruction
*) node
;
935 if (inst
->as_function())
938 ir_variable
*var
= inst
->as_variable();
939 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
942 assert(inst
->as_assignment()
944 || inst
->as_if() /* for initializers with the ?: operator */
945 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
948 inst
= inst
->clone(target
, NULL
);
951 hash_table_insert(temps
, inst
, var
);
953 remap_variables(inst
, target
, temps
);
958 last
->insert_after(inst
);
963 hash_table_dtor(temps
);
969 * Get the function signature for main from a shader
971 static ir_function_signature
*
972 get_main_function_signature(gl_shader
*sh
)
974 ir_function
*const f
= sh
->symbols
->get_function("main");
976 exec_list void_parameters
;
978 /* Look for the 'void main()' signature and ensure that it's defined.
979 * This keeps the linker from accidentally pick a shader that just
980 * contains a prototype for main.
982 * We don't have to check for multiple definitions of main (in multiple
983 * shaders) because that would have already been caught above.
985 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
986 if ((sig
!= NULL
) && sig
->is_defined
) {
996 * This class is only used in link_intrastage_shaders() below but declaring
997 * it inside that function leads to compiler warnings with some versions of
1000 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1002 virtual ir_visitor_status
visit(ir_variable
*var
)
1004 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
1005 const glsl_type
*type
=
1006 glsl_type::get_array_instance(var
->type
->fields
.array
,
1007 var
->max_array_access
+ 1);
1008 assert(type
!= NULL
);
1011 return visit_continue
;
1016 * Performs the cross-validation of geometry shader max_vertices and
1017 * primitive type layout qualifiers for the attached geometry shaders,
1018 * and propagates them to the linked GS and linked shader program.
1021 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1022 struct gl_shader
*linked_shader
,
1023 struct gl_shader
**shader_list
,
1024 unsigned num_shaders
)
1026 linked_shader
->Geom
.VerticesOut
= 0;
1027 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1028 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1030 /* No in/out qualifiers defined for anything but GLSL 1.50+
1031 * geometry shaders so far.
1033 if (linked_shader
->Type
!= GL_GEOMETRY_SHADER
|| prog
->Version
< 150)
1036 /* From the GLSL 1.50 spec, page 46:
1038 * "All geometry shader output layout declarations in a program
1039 * must declare the same layout and same value for
1040 * max_vertices. There must be at least one geometry output
1041 * layout declaration somewhere in a program, but not all
1042 * geometry shaders (compilation units) are required to
1046 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1047 struct gl_shader
*shader
= shader_list
[i
];
1049 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1050 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1051 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1052 linker_error(prog
, "geometry shader defined with conflicting "
1056 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1059 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1060 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1061 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1062 linker_error(prog
, "geometry shader defined with conflicting "
1066 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1069 if (shader
->Geom
.VerticesOut
!= 0) {
1070 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1071 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1072 linker_error(prog
, "geometry shader defined with conflicting "
1073 "output vertex count (%d and %d)\n",
1074 linked_shader
->Geom
.VerticesOut
,
1075 shader
->Geom
.VerticesOut
);
1078 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1082 /* Just do the intrastage -> interstage propagation right now,
1083 * since we already know we're in the right type of shader program
1086 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1088 "geometry shader didn't declare primitive input type\n");
1091 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1093 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1095 "geometry shader didn't declare primitive output type\n");
1098 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1100 if (linked_shader
->Geom
.VerticesOut
== 0) {
1102 "geometry shader didn't declare max_vertices\n");
1105 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1109 * Combine a group of shaders for a single stage to generate a linked shader
1112 * If this function is supplied a single shader, it is cloned, and the new
1113 * shader is returned.
1115 static struct gl_shader
*
1116 link_intrastage_shaders(void *mem_ctx
,
1117 struct gl_context
*ctx
,
1118 struct gl_shader_program
*prog
,
1119 struct gl_shader
**shader_list
,
1120 unsigned num_shaders
)
1122 struct gl_uniform_block
*uniform_blocks
= NULL
;
1124 /* Check that global variables defined in multiple shaders are consistent.
1126 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1127 if (!prog
->LinkStatus
)
1130 /* Check that interface blocks defined in multiple shaders are consistent.
1132 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1134 if (!prog
->LinkStatus
)
1137 /* Link up uniform blocks defined within this stage. */
1138 const unsigned num_uniform_blocks
=
1139 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1142 /* Check that there is only a single definition of each function signature
1143 * across all shaders.
1145 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1146 foreach_list(node
, shader_list
[i
]->ir
) {
1147 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
1152 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1153 ir_function
*const other
=
1154 shader_list
[j
]->symbols
->get_function(f
->name
);
1156 /* If the other shader has no function (and therefore no function
1157 * signatures) with the same name, skip to the next shader.
1162 foreach_iter (exec_list_iterator
, iter
, *f
) {
1163 ir_function_signature
*sig
=
1164 (ir_function_signature
*) iter
.get();
1166 if (!sig
->is_defined
|| sig
->is_builtin
)
1169 ir_function_signature
*other_sig
=
1170 other
->exact_matching_signature(& sig
->parameters
);
1172 if ((other_sig
!= NULL
) && other_sig
->is_defined
1173 && !other_sig
->is_builtin
) {
1174 linker_error(prog
, "function `%s' is multiply defined",
1183 /* Find the shader that defines main, and make a clone of it.
1185 * Starting with the clone, search for undefined references. If one is
1186 * found, find the shader that defines it. Clone the reference and add
1187 * it to the shader. Repeat until there are no undefined references or
1188 * until a reference cannot be resolved.
1190 gl_shader
*main
= NULL
;
1191 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1192 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
1193 main
= shader_list
[i
];
1199 linker_error(prog
, "%s shader lacks `main'\n",
1200 _mesa_glsl_shader_target_name(shader_list
[0]->Type
));
1204 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1205 linked
->ir
= new(linked
) exec_list
;
1206 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1208 linked
->UniformBlocks
= uniform_blocks
;
1209 linked
->NumUniformBlocks
= num_uniform_blocks
;
1210 ralloc_steal(linked
, linked
->UniformBlocks
);
1212 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
1214 populate_symbol_table(linked
);
1216 /* The a pointer to the main function in the final linked shader (i.e., the
1217 * copy of the original shader that contained the main function).
1219 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1221 /* Move any instructions other than variable declarations or function
1222 * declarations into main.
1224 exec_node
*insertion_point
=
1225 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1228 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1229 if (shader_list
[i
] == main
)
1232 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1233 insertion_point
, true, linked
);
1236 /* Resolve initializers for global variables in the linked shader.
1238 unsigned num_linking_shaders
= num_shaders
;
1239 for (unsigned i
= 0; i
< num_shaders
; i
++)
1240 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
1242 gl_shader
**linking_shaders
=
1243 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
1245 memcpy(linking_shaders
, shader_list
,
1246 sizeof(linking_shaders
[0]) * num_shaders
);
1248 unsigned idx
= num_shaders
;
1249 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1250 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
1251 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
1252 idx
+= shader_list
[i
]->num_builtins_to_link
;
1255 assert(idx
== num_linking_shaders
);
1257 if (!link_function_calls(prog
, linked
, linking_shaders
,
1258 num_linking_shaders
)) {
1259 ctx
->Driver
.DeleteShader(ctx
, linked
);
1260 free(linking_shaders
);
1264 free(linking_shaders
);
1266 /* At this point linked should contain all of the linked IR, so
1267 * validate it to make sure nothing went wrong.
1269 validate_ir_tree(linked
->ir
);
1271 /* Set the size of geometry shader input arrays */
1272 if (linked
->Type
== GL_GEOMETRY_SHADER
) {
1273 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
1274 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
1275 foreach_iter(exec_list_iterator
, iter
, *linked
->ir
) {
1276 ir_instruction
*ir
= (ir_instruction
*)iter
.get();
1277 ir
->accept(&input_resize_visitor
);
1281 /* Make a pass over all variable declarations to ensure that arrays with
1282 * unspecified sizes have a size specified. The size is inferred from the
1283 * max_array_access field.
1285 array_sizing_visitor v
;
1292 * Update the sizes of linked shader uniform arrays to the maximum
1295 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1297 * If one or more elements of an array are active,
1298 * GetActiveUniform will return the name of the array in name,
1299 * subject to the restrictions listed above. The type of the array
1300 * is returned in type. The size parameter contains the highest
1301 * array element index used, plus one. The compiler or linker
1302 * determines the highest index used. There will be only one
1303 * active uniform reported by the GL per uniform array.
1307 update_array_sizes(struct gl_shader_program
*prog
)
1309 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1310 if (prog
->_LinkedShaders
[i
] == NULL
)
1313 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1314 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1316 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
) ||
1317 !var
->type
->is_array())
1320 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1321 * will not be eliminated. Since we always do std140, just
1322 * don't resize arrays in UBOs.
1324 if (var
->is_in_uniform_block())
1327 unsigned int size
= var
->max_array_access
;
1328 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1329 if (prog
->_LinkedShaders
[j
] == NULL
)
1332 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1333 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1337 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1338 other_var
->max_array_access
> size
) {
1339 size
= other_var
->max_array_access
;
1344 if (size
+ 1 != var
->type
->length
) {
1345 /* If this is a built-in uniform (i.e., it's backed by some
1346 * fixed-function state), adjust the number of state slots to
1347 * match the new array size. The number of slots per array entry
1348 * is not known. It seems safe to assume that the total number of
1349 * slots is an integer multiple of the number of array elements.
1350 * Determine the number of slots per array element by dividing by
1351 * the old (total) size.
1353 if (var
->num_state_slots
> 0) {
1354 var
->num_state_slots
= (size
+ 1)
1355 * (var
->num_state_slots
/ var
->type
->length
);
1358 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1360 /* FINISHME: We should update the types of array
1361 * dereferences of this variable now.
1369 * Find a contiguous set of available bits in a bitmask.
1371 * \param used_mask Bits representing used (1) and unused (0) locations
1372 * \param needed_count Number of contiguous bits needed.
1375 * Base location of the available bits on success or -1 on failure.
1378 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1380 unsigned needed_mask
= (1 << needed_count
) - 1;
1381 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1383 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1384 * cannot optimize possibly infinite loops" for the loop below.
1386 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1389 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1390 if ((needed_mask
& ~used_mask
) == needed_mask
)
1401 * Assign locations for either VS inputs for FS outputs
1403 * \param prog Shader program whose variables need locations assigned
1404 * \param target_index Selector for the program target to receive location
1405 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1406 * \c MESA_SHADER_FRAGMENT.
1407 * \param max_index Maximum number of generic locations. This corresponds
1408 * to either the maximum number of draw buffers or the
1409 * maximum number of generic attributes.
1412 * If locations are successfully assigned, true is returned. Otherwise an
1413 * error is emitted to the shader link log and false is returned.
1416 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1417 unsigned target_index
,
1420 /* Mark invalid locations as being used.
1422 unsigned used_locations
= (max_index
>= 32)
1423 ? ~0 : ~((1 << max_index
) - 1);
1425 assert((target_index
== MESA_SHADER_VERTEX
)
1426 || (target_index
== MESA_SHADER_FRAGMENT
));
1428 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1432 /* Operate in a total of four passes.
1434 * 1. Invalidate the location assignments for all vertex shader inputs.
1436 * 2. Assign locations for inputs that have user-defined (via
1437 * glBindVertexAttribLocation) locations and outputs that have
1438 * user-defined locations (via glBindFragDataLocation).
1440 * 3. Sort the attributes without assigned locations by number of slots
1441 * required in decreasing order. Fragmentation caused by attribute
1442 * locations assigned by the application may prevent large attributes
1443 * from having enough contiguous space.
1445 * 4. Assign locations to any inputs without assigned locations.
1448 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1449 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1451 const enum ir_variable_mode direction
=
1452 (target_index
== MESA_SHADER_VERTEX
)
1453 ? ir_var_shader_in
: ir_var_shader_out
;
1456 /* Temporary storage for the set of attributes that need locations assigned.
1462 /* Used below in the call to qsort. */
1463 static int compare(const void *a
, const void *b
)
1465 const temp_attr
*const l
= (const temp_attr
*) a
;
1466 const temp_attr
*const r
= (const temp_attr
*) b
;
1468 /* Reversed because we want a descending order sort below. */
1469 return r
->slots
- l
->slots
;
1473 unsigned num_attr
= 0;
1475 foreach_list(node
, sh
->ir
) {
1476 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1478 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1481 if (var
->explicit_location
) {
1482 if ((var
->location
>= (int)(max_index
+ generic_base
))
1483 || (var
->location
< 0)) {
1485 "invalid explicit location %d specified for `%s'\n",
1487 ? var
->location
: var
->location
- generic_base
,
1491 } else if (target_index
== MESA_SHADER_VERTEX
) {
1494 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1495 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1496 var
->location
= binding
;
1497 var
->is_unmatched_generic_inout
= 0;
1499 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1503 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1504 assert(binding
>= FRAG_RESULT_DATA0
);
1505 var
->location
= binding
;
1506 var
->is_unmatched_generic_inout
= 0;
1508 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1514 /* If the variable is not a built-in and has a location statically
1515 * assigned in the shader (presumably via a layout qualifier), make sure
1516 * that it doesn't collide with other assigned locations. Otherwise,
1517 * add it to the list of variables that need linker-assigned locations.
1519 const unsigned slots
= var
->type
->count_attribute_slots();
1520 if (var
->location
!= -1) {
1521 if (var
->location
>= generic_base
&& var
->index
< 1) {
1522 /* From page 61 of the OpenGL 4.0 spec:
1524 * "LinkProgram will fail if the attribute bindings assigned
1525 * by BindAttribLocation do not leave not enough space to
1526 * assign a location for an active matrix attribute or an
1527 * active attribute array, both of which require multiple
1528 * contiguous generic attributes."
1530 * Previous versions of the spec contain similar language but omit
1531 * the bit about attribute arrays.
1533 * Page 61 of the OpenGL 4.0 spec also says:
1535 * "It is possible for an application to bind more than one
1536 * attribute name to the same location. This is referred to as
1537 * aliasing. This will only work if only one of the aliased
1538 * attributes is active in the executable program, or if no
1539 * path through the shader consumes more than one attribute of
1540 * a set of attributes aliased to the same location. A link
1541 * error can occur if the linker determines that every path
1542 * through the shader consumes multiple aliased attributes,
1543 * but implementations are not required to generate an error
1546 * These two paragraphs are either somewhat contradictory, or I
1547 * don't fully understand one or both of them.
1549 /* FINISHME: The code as currently written does not support
1550 * FINISHME: attribute location aliasing (see comment above).
1552 /* Mask representing the contiguous slots that will be used by
1555 const unsigned attr
= var
->location
- generic_base
;
1556 const unsigned use_mask
= (1 << slots
) - 1;
1558 /* Generate a link error if the set of bits requested for this
1559 * attribute overlaps any previously allocated bits.
1561 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1562 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1563 ? "vertex shader input" : "fragment shader output";
1565 "insufficient contiguous locations "
1566 "available for %s `%s' %d %d %d", string
,
1567 var
->name
, used_locations
, use_mask
, attr
);
1571 used_locations
|= (use_mask
<< attr
);
1577 to_assign
[num_attr
].slots
= slots
;
1578 to_assign
[num_attr
].var
= var
;
1582 /* If all of the attributes were assigned locations by the application (or
1583 * are built-in attributes with fixed locations), return early. This should
1584 * be the common case.
1589 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1591 if (target_index
== MESA_SHADER_VERTEX
) {
1592 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1593 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1594 * reserved to prevent it from being automatically allocated below.
1596 find_deref_visitor
find("gl_Vertex");
1598 if (find
.variable_found())
1599 used_locations
|= (1 << 0);
1602 for (unsigned i
= 0; i
< num_attr
; i
++) {
1603 /* Mask representing the contiguous slots that will be used by this
1606 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1608 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1611 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1612 ? "vertex shader input" : "fragment shader output";
1615 "insufficient contiguous locations "
1616 "available for %s `%s'",
1617 string
, to_assign
[i
].var
->name
);
1621 to_assign
[i
].var
->location
= generic_base
+ location
;
1622 to_assign
[i
].var
->is_unmatched_generic_inout
= 0;
1623 used_locations
|= (use_mask
<< location
);
1631 * Demote shader inputs and outputs that are not used in other stages
1634 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1636 foreach_list(node
, sh
->ir
) {
1637 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1639 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1642 /* A shader 'in' or 'out' variable is only really an input or output if
1643 * its value is used by other shader stages. This will cause the variable
1644 * to have a location assigned.
1646 if (var
->is_unmatched_generic_inout
) {
1647 var
->mode
= ir_var_auto
;
1654 * Store the gl_FragDepth layout in the gl_shader_program struct.
1657 store_fragdepth_layout(struct gl_shader_program
*prog
)
1659 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1663 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
1665 /* We don't look up the gl_FragDepth symbol directly because if
1666 * gl_FragDepth is not used in the shader, it's removed from the IR.
1667 * However, the symbol won't be removed from the symbol table.
1669 * We're only interested in the cases where the variable is NOT removed
1672 foreach_list(node
, ir
) {
1673 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1675 if (var
== NULL
|| var
->mode
!= ir_var_shader_out
) {
1679 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1680 switch (var
->depth_layout
) {
1681 case ir_depth_layout_none
:
1682 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
1684 case ir_depth_layout_any
:
1685 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
1687 case ir_depth_layout_greater
:
1688 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
1690 case ir_depth_layout_less
:
1691 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
1693 case ir_depth_layout_unchanged
:
1694 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
1705 * Validate the resources used by a program versus the implementation limits
1708 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1710 static const char *const shader_names
[MESA_SHADER_TYPES
] = {
1711 "vertex", "geometry", "fragment"
1714 const unsigned max_samplers
[MESA_SHADER_TYPES
] = {
1715 ctx
->Const
.VertexProgram
.MaxTextureImageUnits
,
1716 ctx
->Const
.GeometryProgram
.MaxTextureImageUnits
,
1717 ctx
->Const
.FragmentProgram
.MaxTextureImageUnits
1720 const unsigned max_default_uniform_components
[MESA_SHADER_TYPES
] = {
1721 ctx
->Const
.VertexProgram
.MaxUniformComponents
,
1722 ctx
->Const
.GeometryProgram
.MaxUniformComponents
,
1723 ctx
->Const
.FragmentProgram
.MaxUniformComponents
1726 const unsigned max_combined_uniform_components
[MESA_SHADER_TYPES
] = {
1727 ctx
->Const
.VertexProgram
.MaxCombinedUniformComponents
,
1728 ctx
->Const
.GeometryProgram
.MaxCombinedUniformComponents
,
1729 ctx
->Const
.FragmentProgram
.MaxCombinedUniformComponents
1732 const unsigned max_uniform_blocks
[MESA_SHADER_TYPES
] = {
1733 ctx
->Const
.VertexProgram
.MaxUniformBlocks
,
1734 ctx
->Const
.GeometryProgram
.MaxUniformBlocks
,
1735 ctx
->Const
.FragmentProgram
.MaxUniformBlocks
1738 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1739 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1744 if (sh
->num_samplers
> max_samplers
[i
]) {
1745 linker_error(prog
, "Too many %s shader texture samplers",
1749 if (sh
->num_uniform_components
> max_default_uniform_components
[i
]) {
1750 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
1751 linker_warning(prog
, "Too many %s shader default uniform block "
1752 "components, but the driver will try to optimize "
1753 "them out; this is non-portable out-of-spec "
1757 linker_error(prog
, "Too many %s shader default uniform block "
1763 if (sh
->num_combined_uniform_components
>
1764 max_combined_uniform_components
[i
]) {
1765 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
1766 linker_warning(prog
, "Too many %s shader uniform components, "
1767 "but the driver will try to optimize them out; "
1768 "this is non-portable out-of-spec behavior\n",
1771 linker_error(prog
, "Too many %s shader uniform components",
1777 unsigned blocks
[MESA_SHADER_TYPES
] = {0};
1778 unsigned total_uniform_blocks
= 0;
1780 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
1781 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1782 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
1784 total_uniform_blocks
++;
1788 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
1789 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
1790 prog
->NumUniformBlocks
,
1791 ctx
->Const
.MaxCombinedUniformBlocks
);
1793 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1794 if (blocks
[i
] > max_uniform_blocks
[i
]) {
1795 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
1798 max_uniform_blocks
[i
]);
1807 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1809 tfeedback_decl
*tfeedback_decls
= NULL
;
1810 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
1812 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1814 prog
->LinkStatus
= true; /* All error paths will set this to false */
1815 prog
->Validated
= false;
1816 prog
->_Used
= false;
1818 ralloc_free(prog
->InfoLog
);
1819 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1821 ralloc_free(prog
->UniformBlocks
);
1822 prog
->UniformBlocks
= NULL
;
1823 prog
->NumUniformBlocks
= 0;
1824 for (int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1825 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
1826 prog
->UniformBlockStageIndex
[i
] = NULL
;
1829 /* Separate the shaders into groups based on their type.
1831 struct gl_shader
**vert_shader_list
;
1832 unsigned num_vert_shaders
= 0;
1833 struct gl_shader
**frag_shader_list
;
1834 unsigned num_frag_shaders
= 0;
1835 struct gl_shader
**geom_shader_list
;
1836 unsigned num_geom_shaders
= 0;
1838 vert_shader_list
= (struct gl_shader
**)
1839 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
1840 frag_shader_list
= (struct gl_shader
**)
1841 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
1842 geom_shader_list
= (struct gl_shader
**)
1843 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
1845 unsigned min_version
= UINT_MAX
;
1846 unsigned max_version
= 0;
1847 const bool is_es_prog
=
1848 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
1849 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1850 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1851 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1853 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
1854 linker_error(prog
, "all shaders must use same shading "
1855 "language version\n");
1859 switch (prog
->Shaders
[i
]->Type
) {
1860 case GL_VERTEX_SHADER
:
1861 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1864 case GL_FRAGMENT_SHADER
:
1865 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1868 case GL_GEOMETRY_SHADER
:
1869 geom_shader_list
[num_geom_shaders
] = prog
->Shaders
[i
];
1875 /* Previous to GLSL version 1.30, different compilation units could mix and
1876 * match shading language versions. With GLSL 1.30 and later, the versions
1877 * of all shaders must match.
1879 * GLSL ES has never allowed mixing of shading language versions.
1881 if ((is_es_prog
|| max_version
>= 130)
1882 && min_version
!= max_version
) {
1883 linker_error(prog
, "all shaders must use same shading "
1884 "language version\n");
1888 prog
->Version
= max_version
;
1889 prog
->IsES
= is_es_prog
;
1891 /* Geometry shaders have to be linked with vertex shaders.
1893 if (num_geom_shaders
> 0 && num_vert_shaders
== 0) {
1894 linker_error(prog
, "Geometry shader must be linked with "
1899 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1900 if (prog
->_LinkedShaders
[i
] != NULL
)
1901 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1903 prog
->_LinkedShaders
[i
] = NULL
;
1906 /* Link all shaders for a particular stage and validate the result.
1908 if (num_vert_shaders
> 0) {
1909 gl_shader
*const sh
=
1910 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1913 if (!prog
->LinkStatus
)
1916 validate_vertex_shader_executable(prog
, sh
);
1917 if (!prog
->LinkStatus
)
1920 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1924 if (num_frag_shaders
> 0) {
1925 gl_shader
*const sh
=
1926 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1929 if (!prog
->LinkStatus
)
1932 validate_fragment_shader_executable(prog
, sh
);
1933 if (!prog
->LinkStatus
)
1936 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1940 if (num_geom_shaders
> 0) {
1941 gl_shader
*const sh
=
1942 link_intrastage_shaders(mem_ctx
, ctx
, prog
, geom_shader_list
,
1945 if (!prog
->LinkStatus
)
1948 validate_geometry_shader_executable(prog
, sh
);
1949 if (!prog
->LinkStatus
)
1952 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
],
1956 /* Here begins the inter-stage linking phase. Some initial validation is
1957 * performed, then locations are assigned for uniforms, attributes, and
1960 cross_validate_uniforms(prog
);
1961 if (!prog
->LinkStatus
)
1966 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1967 if (prog
->_LinkedShaders
[prev
] != NULL
)
1971 /* Validate the inputs of each stage with the output of the preceding
1974 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1975 if (prog
->_LinkedShaders
[i
] == NULL
)
1978 validate_interstage_interface_blocks(prog
, prog
->_LinkedShaders
[prev
],
1979 prog
->_LinkedShaders
[i
]);
1980 if (!prog
->LinkStatus
)
1983 cross_validate_outputs_to_inputs(prog
,
1984 prog
->_LinkedShaders
[prev
],
1985 prog
->_LinkedShaders
[i
]);
1986 if (!prog
->LinkStatus
)
1993 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1994 if (prog
->_LinkedShaders
[i
] != NULL
)
1995 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
1998 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
1999 * it before optimization because we want most of the checks to get
2000 * dropped thanks to constant propagation.
2002 * This rule also applies to GLSL ES 3.00.
2004 if (max_version
>= (is_es_prog
? 300 : 130)) {
2005 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2007 lower_discard_flow(sh
->ir
);
2011 if (!interstage_cross_validate_uniform_blocks(prog
))
2014 /* Do common optimization before assigning storage for attributes,
2015 * uniforms, and varyings. Later optimization could possibly make
2016 * some of that unused.
2018 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2019 if (prog
->_LinkedShaders
[i
] == NULL
)
2022 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2023 if (!prog
->LinkStatus
)
2026 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
) {
2027 lower_clip_distance(prog
->_LinkedShaders
[i
]);
2030 unsigned max_unroll
= ctx
->ShaderCompilerOptions
[i
].MaxUnrollIterations
;
2032 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, max_unroll
, &ctx
->ShaderCompilerOptions
[i
]))
2036 /* Mark all generic shader inputs and outputs as unpaired. */
2037 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
2038 link_invalidate_variable_locations(
2039 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2040 VERT_ATTRIB_GENERIC0
, VARYING_SLOT_VAR0
);
2042 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
2043 link_invalidate_variable_locations(
2044 prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
],
2045 VARYING_SLOT_VAR0
, VARYING_SLOT_VAR0
);
2047 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
2048 link_invalidate_variable_locations(
2049 prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
2050 VARYING_SLOT_VAR0
, FRAG_RESULT_DATA0
);
2053 /* FINISHME: The value of the max_attribute_index parameter is
2054 * FINISHME: implementation dependent based on the value of
2055 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2056 * FINISHME: at least 16, so hardcode 16 for now.
2058 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2062 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2067 for (first
= 0; first
< MESA_SHADER_TYPES
; first
++) {
2068 if (prog
->_LinkedShaders
[first
] != NULL
)
2072 if (num_tfeedback_decls
!= 0) {
2073 /* From GL_EXT_transform_feedback:
2074 * A program will fail to link if:
2076 * * the <count> specified by TransformFeedbackVaryingsEXT is
2077 * non-zero, but the program object has no vertex or geometry
2080 if (first
== MESA_SHADER_FRAGMENT
) {
2081 linker_error(prog
, "Transform feedback varyings specified, but "
2082 "no vertex or geometry shader is present.");
2086 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2087 prog
->TransformFeedback
.NumVarying
);
2088 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2089 prog
->TransformFeedback
.VaryingNames
,
2094 /* Linking the stages in the opposite order (from fragment to vertex)
2095 * ensures that inter-shader outputs written to in an earlier stage are
2096 * eliminated if they are (transitively) not used in a later stage.
2099 for (last
= MESA_SHADER_TYPES
-1; last
>= 0; last
--) {
2100 if (prog
->_LinkedShaders
[last
] != NULL
)
2104 if (last
>= 0 && last
< MESA_SHADER_FRAGMENT
) {
2105 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
2107 if (num_tfeedback_decls
!= 0) {
2108 /* There was no fragment shader, but we still have to assign varying
2109 * locations for use by transform feedback.
2111 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
2113 num_tfeedback_decls
, tfeedback_decls
,
2118 do_dead_builtin_varyings(ctx
, sh
, NULL
,
2119 num_tfeedback_decls
, tfeedback_decls
);
2121 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
2123 /* Eliminate code that is now dead due to unused outputs being demoted.
2125 while (do_dead_code(sh
->ir
, false))
2128 else if (first
== MESA_SHADER_FRAGMENT
) {
2129 /* If the program only contains a fragment shader...
2131 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
2133 do_dead_builtin_varyings(ctx
, NULL
, sh
,
2134 num_tfeedback_decls
, tfeedback_decls
);
2136 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
2138 while (do_dead_code(sh
->ir
, false))
2143 for (int i
= next
- 1; i
>= 0; i
--) {
2144 if (prog
->_LinkedShaders
[i
] == NULL
)
2147 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
2148 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
2149 unsigned gs_input_vertices
=
2150 next
== MESA_SHADER_GEOMETRY
? prog
->Geom
.VerticesIn
: 0;
2152 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
2153 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2154 tfeedback_decls
, gs_input_vertices
))
2157 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
2158 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2161 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
2162 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
2164 /* Eliminate code that is now dead due to unused outputs being demoted.
2166 while (do_dead_code(sh_i
->ir
, false))
2168 while (do_dead_code(sh_next
->ir
, false))
2171 /* This must be done after all dead varyings are eliminated. */
2172 if (!check_against_varying_limit(ctx
, prog
, sh_next
))
2178 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2181 update_array_sizes(prog
);
2182 link_assign_uniform_locations(prog
);
2183 store_fragdepth_layout(prog
);
2185 check_resources(ctx
, prog
);
2186 if (!prog
->LinkStatus
)
2189 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2190 * present in a linked program. By checking prog->IsES, we also
2191 * catch the GL_ARB_ES2_compatibility case.
2193 if (!prog
->InternalSeparateShader
&&
2194 (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)) {
2195 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2196 linker_error(prog
, "program lacks a vertex shader\n");
2197 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2198 linker_error(prog
, "program lacks a fragment shader\n");
2202 /* FINISHME: Assign fragment shader output locations. */
2205 free(vert_shader_list
);
2206 free(frag_shader_list
);
2207 free(geom_shader_list
);
2209 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2210 if (prog
->_LinkedShaders
[i
] == NULL
)
2213 /* Retain any live IR, but trash the rest. */
2214 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2216 /* The symbol table in the linked shaders may contain references to
2217 * variables that were removed (e.g., unused uniforms). Since it may
2218 * contain junk, there is no possible valid use. Delete it and set the
2221 delete prog
->_LinkedShaders
[i
]->symbols
;
2222 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2225 ralloc_free(mem_ctx
);