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"
71 #include "program/hash_table.h"
73 #include "link_varyings.h"
74 #include "ir_optimization.h"
77 #include "main/shaderobj.h"
81 * Visitor that determines whether or not a variable is ever written.
83 class find_assignment_visitor
: public ir_hierarchical_visitor
{
85 find_assignment_visitor(const char *name
)
86 : name(name
), found(false)
91 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
93 ir_variable
*const var
= ir
->lhs
->variable_referenced();
95 if (strcmp(name
, var
->name
) == 0) {
100 return visit_continue_with_parent
;
103 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
105 exec_list_iterator sig_iter
= ir
->callee
->parameters
.iterator();
106 foreach_iter(exec_list_iterator
, iter
, *ir
) {
107 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
108 ir_variable
*sig_param
= (ir_variable
*)sig_iter
.get();
110 if (sig_param
->mode
== ir_var_function_out
||
111 sig_param
->mode
== ir_var_function_inout
) {
112 ir_variable
*var
= param_rval
->variable_referenced();
113 if (var
&& strcmp(name
, var
->name
) == 0) {
121 if (ir
->return_deref
!= NULL
) {
122 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
124 if (strcmp(name
, var
->name
) == 0) {
130 return visit_continue_with_parent
;
133 bool variable_found()
139 const char *name
; /**< Find writes to a variable with this name. */
140 bool found
; /**< Was a write to the variable found? */
145 * Visitor that determines whether or not a variable is ever read.
147 class find_deref_visitor
: public ir_hierarchical_visitor
{
149 find_deref_visitor(const char *name
)
150 : name(name
), found(false)
155 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
157 if (strcmp(this->name
, ir
->var
->name
) == 0) {
162 return visit_continue
;
165 bool variable_found() const
171 const char *name
; /**< Find writes to a variable with this name. */
172 bool found
; /**< Was a write to the variable found? */
177 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
181 ralloc_strcat(&prog
->InfoLog
, "error: ");
183 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
186 prog
->LinkStatus
= false;
191 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
195 ralloc_strcat(&prog
->InfoLog
, "error: ");
197 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
204 * Given a string identifying a program resource, break it into a base name
205 * and an optional array index in square brackets.
207 * If an array index is present, \c out_base_name_end is set to point to the
208 * "[" that precedes the array index, and the array index itself is returned
211 * If no array index is present (or if the array index is negative or
212 * mal-formed), \c out_base_name_end, is set to point to the null terminator
213 * at the end of the input string, and -1 is returned.
215 * Only the final array index is parsed; if the string contains other array
216 * indices (or structure field accesses), they are left in the base name.
218 * No attempt is made to check that the base name is properly formed;
219 * typically the caller will look up the base name in a hash table, so
220 * ill-formed base names simply turn into hash table lookup failures.
223 parse_program_resource_name(const GLchar
*name
,
224 const GLchar
**out_base_name_end
)
226 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
228 * "When an integer array element or block instance number is part of
229 * the name string, it will be specified in decimal form without a "+"
230 * or "-" sign or any extra leading zeroes. Additionally, the name
231 * string will not include white space anywhere in the string."
234 const size_t len
= strlen(name
);
235 *out_base_name_end
= name
+ len
;
237 if (len
== 0 || name
[len
-1] != ']')
240 /* Walk backwards over the string looking for a non-digit character. This
241 * had better be the opening bracket for an array index.
243 * Initially, i specifies the location of the ']'. Since the string may
244 * contain only the ']' charcater, walk backwards very carefully.
247 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
250 if ((i
== 0) || name
[i
-1] != '[')
253 long array_index
= strtol(&name
[i
], NULL
, 10);
257 *out_base_name_end
= name
+ (i
- 1);
263 link_invalidate_variable_locations(gl_shader
*sh
, int input_base
,
266 foreach_list(node
, sh
->ir
) {
267 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
274 case ir_var_shader_in
:
277 case ir_var_shader_out
:
284 /* Only assign locations for generic attributes / varyings / etc.
286 if ((var
->location
>= base
) && !var
->explicit_location
)
289 if ((var
->location
== -1) && !var
->explicit_location
) {
290 var
->is_unmatched_generic_inout
= 1;
291 var
->location_frac
= 0;
293 var
->is_unmatched_generic_inout
= 0;
300 * Determine the number of attribute slots required for a particular type
302 * This code is here because it implements the language rules of a specific
303 * GLSL version. Since it's a property of the language and not a property of
304 * types in general, it doesn't really belong in glsl_type.
307 count_attribute_slots(const glsl_type
*t
)
309 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
311 * "A scalar input counts the same amount against this limit as a vec4,
312 * so applications may want to consider packing groups of four
313 * unrelated float inputs together into a vector to better utilize the
314 * capabilities of the underlying hardware. A matrix input will use up
315 * multiple locations. The number of locations used will equal the
316 * number of columns in the matrix."
318 * The spec does not explicitly say how arrays are counted. However, it
319 * should be safe to assume the total number of slots consumed by an array
320 * is the number of entries in the array multiplied by the number of slots
321 * consumed by a single element of the array.
325 return t
->array_size() * count_attribute_slots(t
->element_type());
328 return t
->matrix_columns
;
335 * Verify that a vertex shader executable meets all semantic requirements.
337 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
340 * \param shader Vertex shader executable to be verified
343 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
344 struct gl_shader
*shader
)
349 /* From the GLSL 1.10 spec, page 48:
351 * "The variable gl_Position is available only in the vertex
352 * language and is intended for writing the homogeneous vertex
353 * position. All executions of a well-formed vertex shader
354 * executable must write a value into this variable. [...] The
355 * variable gl_Position is available only in the vertex
356 * language and is intended for writing the homogeneous vertex
357 * position. All executions of a well-formed vertex shader
358 * executable must write a value into this variable."
360 * while in GLSL 1.40 this text is changed to:
362 * "The variable gl_Position is available only in the vertex
363 * language and is intended for writing the homogeneous vertex
364 * position. It can be written at any time during shader
365 * execution. It may also be read back by a vertex shader
366 * after being written. This value will be used by primitive
367 * assembly, clipping, culling, and other fixed functionality
368 * operations, if present, that operate on primitives after
369 * vertex processing has occurred. Its value is undefined if
370 * the vertex shader executable does not write gl_Position."
372 * GLSL ES 3.00 is similar to GLSL 1.40--failing to write to gl_Position is
375 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
376 find_assignment_visitor
find("gl_Position");
377 find
.run(shader
->ir
);
378 if (!find
.variable_found()) {
379 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
384 prog
->Vert
.ClipDistanceArraySize
= 0;
386 if (!prog
->IsES
&& prog
->Version
>= 130) {
387 /* From section 7.1 (Vertex Shader Special Variables) of the
390 * "It is an error for a shader to statically write both
391 * gl_ClipVertex and gl_ClipDistance."
393 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
394 * gl_ClipVertex nor gl_ClipDistance.
396 find_assignment_visitor
clip_vertex("gl_ClipVertex");
397 find_assignment_visitor
clip_distance("gl_ClipDistance");
399 clip_vertex
.run(shader
->ir
);
400 clip_distance
.run(shader
->ir
);
401 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
402 linker_error(prog
, "vertex shader writes to both `gl_ClipVertex' "
403 "and `gl_ClipDistance'\n");
406 prog
->Vert
.UsesClipDistance
= clip_distance
.variable_found();
407 ir_variable
*clip_distance_var
=
408 shader
->symbols
->get_variable("gl_ClipDistance");
409 if (clip_distance_var
)
410 prog
->Vert
.ClipDistanceArraySize
= clip_distance_var
->type
->length
;
418 * Verify that a fragment shader executable meets all semantic requirements
420 * \param shader Fragment shader executable to be verified
423 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
424 struct gl_shader
*shader
)
429 find_assignment_visitor
frag_color("gl_FragColor");
430 find_assignment_visitor
frag_data("gl_FragData");
432 frag_color
.run(shader
->ir
);
433 frag_data
.run(shader
->ir
);
435 if (frag_color
.variable_found() && frag_data
.variable_found()) {
436 linker_error(prog
, "fragment shader writes to both "
437 "`gl_FragColor' and `gl_FragData'\n");
446 * Generate a string describing the mode of a variable
449 mode_string(const ir_variable
*var
)
453 return (var
->read_only
) ? "global constant" : "global variable";
455 case ir_var_uniform
: return "uniform";
456 case ir_var_shader_in
: return "shader input";
457 case ir_var_shader_out
: return "shader output";
459 case ir_var_const_in
:
460 case ir_var_temporary
:
462 assert(!"Should not get here.");
463 return "invalid variable";
469 * Perform validation of global variables used across multiple shaders
472 cross_validate_globals(struct gl_shader_program
*prog
,
473 struct gl_shader
**shader_list
,
474 unsigned num_shaders
,
477 /* Examine all of the uniforms in all of the shaders and cross validate
480 glsl_symbol_table variables
;
481 for (unsigned i
= 0; i
< num_shaders
; i
++) {
482 if (shader_list
[i
] == NULL
)
485 foreach_list(node
, shader_list
[i
]->ir
) {
486 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
491 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
494 /* Don't cross validate temporaries that are at global scope. These
495 * will eventually get pulled into the shaders 'main'.
497 if (var
->mode
== ir_var_temporary
)
500 /* If a global with this name has already been seen, verify that the
501 * new instance has the same type. In addition, if the globals have
502 * initializers, the values of the initializers must be the same.
504 ir_variable
*const existing
= variables
.get_variable(var
->name
);
505 if (existing
!= NULL
) {
506 if (var
->type
!= existing
->type
) {
507 /* Consider the types to be "the same" if both types are arrays
508 * of the same type and one of the arrays is implicitly sized.
509 * In addition, set the type of the linked variable to the
510 * explicitly sized array.
512 if (var
->type
->is_array()
513 && existing
->type
->is_array()
514 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
515 && ((var
->type
->length
== 0)
516 || (existing
->type
->length
== 0))) {
517 if (var
->type
->length
!= 0) {
518 existing
->type
= var
->type
;
521 linker_error(prog
, "%s `%s' declared as type "
522 "`%s' and type `%s'\n",
524 var
->name
, var
->type
->name
,
525 existing
->type
->name
);
530 if (var
->explicit_location
) {
531 if (existing
->explicit_location
532 && (var
->location
!= existing
->location
)) {
533 linker_error(prog
, "explicit locations for %s "
534 "`%s' have differing values\n",
535 mode_string(var
), var
->name
);
539 existing
->location
= var
->location
;
540 existing
->explicit_location
= true;
543 /* Validate layout qualifiers for gl_FragDepth.
545 * From the AMD/ARB_conservative_depth specs:
547 * "If gl_FragDepth is redeclared in any fragment shader in a
548 * program, it must be redeclared in all fragment shaders in
549 * that program that have static assignments to
550 * gl_FragDepth. All redeclarations of gl_FragDepth in all
551 * fragment shaders in a single program must have the same set
554 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
555 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
556 bool layout_differs
=
557 var
->depth_layout
!= existing
->depth_layout
;
559 if (layout_declared
&& layout_differs
) {
561 "All redeclarations of gl_FragDepth in all "
562 "fragment shaders in a single program must have "
563 "the same set of qualifiers.");
566 if (var
->used
&& layout_differs
) {
568 "If gl_FragDepth is redeclared with a layout "
569 "qualifier in any fragment shader, it must be "
570 "redeclared with the same layout qualifier in "
571 "all fragment shaders that have assignments to "
576 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
578 * "If a shared global has multiple initializers, the
579 * initializers must all be constant expressions, and they
580 * must all have the same value. Otherwise, a link error will
581 * result. (A shared global having only one initializer does
582 * not require that initializer to be a constant expression.)"
584 * Previous to 4.20 the GLSL spec simply said that initializers
585 * must have the same value. In this case of non-constant
586 * initializers, this was impossible to determine. As a result,
587 * no vendor actually implemented that behavior. The 4.20
588 * behavior matches the implemented behavior of at least one other
589 * vendor, so we'll implement that for all GLSL versions.
591 if (var
->constant_initializer
!= NULL
) {
592 if (existing
->constant_initializer
!= NULL
) {
593 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
594 linker_error(prog
, "initializers for %s "
595 "`%s' have differing values\n",
596 mode_string(var
), var
->name
);
600 /* If the first-seen instance of a particular uniform did not
601 * have an initializer but a later instance does, copy the
602 * initializer to the version stored in the symbol table.
604 /* FINISHME: This is wrong. The constant_value field should
605 * FINISHME: not be modified! Imagine a case where a shader
606 * FINISHME: without an initializer is linked in two different
607 * FINISHME: programs with shaders that have differing
608 * FINISHME: initializers. Linking with the first will
609 * FINISHME: modify the shader, and linking with the second
610 * FINISHME: will fail.
612 existing
->constant_initializer
=
613 var
->constant_initializer
->clone(ralloc_parent(existing
),
618 if (var
->has_initializer
) {
619 if (existing
->has_initializer
620 && (var
->constant_initializer
== NULL
621 || existing
->constant_initializer
== NULL
)) {
623 "shared global variable `%s' has multiple "
624 "non-constant initializers.\n",
629 /* Some instance had an initializer, so keep track of that. In
630 * this location, all sorts of initializers (constant or
631 * otherwise) will propagate the existence to the variable
632 * stored in the symbol table.
634 existing
->has_initializer
= true;
637 if (existing
->invariant
!= var
->invariant
) {
638 linker_error(prog
, "declarations for %s `%s' have "
639 "mismatching invariant qualifiers\n",
640 mode_string(var
), var
->name
);
643 if (existing
->centroid
!= var
->centroid
) {
644 linker_error(prog
, "declarations for %s `%s' have "
645 "mismatching centroid qualifiers\n",
646 mode_string(var
), var
->name
);
650 variables
.add_variable(var
);
659 * Perform validation of uniforms used across multiple shader stages
662 cross_validate_uniforms(struct gl_shader_program
*prog
)
664 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
665 MESA_SHADER_TYPES
, true);
669 * Accumulates the array of prog->UniformBlocks and checks that all
670 * definitons of blocks agree on their contents.
673 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
675 unsigned max_num_uniform_blocks
= 0;
676 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
677 if (prog
->_LinkedShaders
[i
])
678 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
681 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
682 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
684 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
685 max_num_uniform_blocks
);
686 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
687 prog
->UniformBlockStageIndex
[i
][j
] = -1;
692 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
693 int index
= link_cross_validate_uniform_block(prog
,
694 &prog
->UniformBlocks
,
695 &prog
->NumUniformBlocks
,
696 &sh
->UniformBlocks
[j
]);
699 linker_error(prog
, "uniform block `%s' has mismatching definitions",
700 sh
->UniformBlocks
[j
].Name
);
704 prog
->UniformBlockStageIndex
[i
][index
] = j
;
713 * Populates a shaders symbol table with all global declarations
716 populate_symbol_table(gl_shader
*sh
)
718 sh
->symbols
= new(sh
) glsl_symbol_table
;
720 foreach_list(node
, sh
->ir
) {
721 ir_instruction
*const inst
= (ir_instruction
*) node
;
725 if ((func
= inst
->as_function()) != NULL
) {
726 sh
->symbols
->add_function(func
);
727 } else if ((var
= inst
->as_variable()) != NULL
) {
728 sh
->symbols
->add_variable(var
);
735 * Remap variables referenced in an instruction tree
737 * This is used when instruction trees are cloned from one shader and placed in
738 * another. These trees will contain references to \c ir_variable nodes that
739 * do not exist in the target shader. This function finds these \c ir_variable
740 * references and replaces the references with matching variables in the target
743 * If there is no matching variable in the target shader, a clone of the
744 * \c ir_variable is made and added to the target shader. The new variable is
745 * added to \b both the instruction stream and the symbol table.
747 * \param inst IR tree that is to be processed.
748 * \param symbols Symbol table containing global scope symbols in the
750 * \param instructions Instruction stream where new variable declarations
754 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
757 class remap_visitor
: public ir_hierarchical_visitor
{
759 remap_visitor(struct gl_shader
*target
,
762 this->target
= target
;
763 this->symbols
= target
->symbols
;
764 this->instructions
= target
->ir
;
768 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
770 if (ir
->var
->mode
== ir_var_temporary
) {
771 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
775 return visit_continue
;
778 ir_variable
*const existing
=
779 this->symbols
->get_variable(ir
->var
->name
);
780 if (existing
!= NULL
)
783 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
785 this->symbols
->add_variable(copy
);
786 this->instructions
->push_head(copy
);
790 return visit_continue
;
794 struct gl_shader
*target
;
795 glsl_symbol_table
*symbols
;
796 exec_list
*instructions
;
800 remap_visitor
v(target
, temps
);
807 * Move non-declarations from one instruction stream to another
809 * The intended usage pattern of this function is to pass the pointer to the
810 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
811 * pointer) for \c last and \c false for \c make_copies on the first
812 * call. Successive calls pass the return value of the previous call for
813 * \c last and \c true for \c make_copies.
815 * \param instructions Source instruction stream
816 * \param last Instruction after which new instructions should be
817 * inserted in the target instruction stream
818 * \param make_copies Flag selecting whether instructions in \c instructions
819 * should be copied (via \c ir_instruction::clone) into the
820 * target list or moved.
823 * The new "last" instruction in the target instruction stream. This pointer
824 * is suitable for use as the \c last parameter of a later call to this
828 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
829 bool make_copies
, gl_shader
*target
)
831 hash_table
*temps
= NULL
;
834 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
835 hash_table_pointer_compare
);
837 foreach_list_safe(node
, instructions
) {
838 ir_instruction
*inst
= (ir_instruction
*) node
;
840 if (inst
->as_function())
843 ir_variable
*var
= inst
->as_variable();
844 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
847 assert(inst
->as_assignment()
849 || inst
->as_if() /* for initializers with the ?: operator */
850 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
853 inst
= inst
->clone(target
, NULL
);
856 hash_table_insert(temps
, inst
, var
);
858 remap_variables(inst
, target
, temps
);
863 last
->insert_after(inst
);
868 hash_table_dtor(temps
);
874 * Get the function signature for main from a shader
876 static ir_function_signature
*
877 get_main_function_signature(gl_shader
*sh
)
879 ir_function
*const f
= sh
->symbols
->get_function("main");
881 exec_list void_parameters
;
883 /* Look for the 'void main()' signature and ensure that it's defined.
884 * This keeps the linker from accidentally pick a shader that just
885 * contains a prototype for main.
887 * We don't have to check for multiple definitions of main (in multiple
888 * shaders) because that would have already been caught above.
890 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
891 if ((sig
!= NULL
) && sig
->is_defined
) {
901 * This class is only used in link_intrastage_shaders() below but declaring
902 * it inside that function leads to compiler warnings with some versions of
905 class array_sizing_visitor
: public ir_hierarchical_visitor
{
907 virtual ir_visitor_status
visit(ir_variable
*var
)
909 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
910 const glsl_type
*type
=
911 glsl_type::get_array_instance(var
->type
->fields
.array
,
912 var
->max_array_access
+ 1);
913 assert(type
!= NULL
);
916 return visit_continue
;
921 * Combine a group of shaders for a single stage to generate a linked shader
924 * If this function is supplied a single shader, it is cloned, and the new
925 * shader is returned.
927 static struct gl_shader
*
928 link_intrastage_shaders(void *mem_ctx
,
929 struct gl_context
*ctx
,
930 struct gl_shader_program
*prog
,
931 struct gl_shader
**shader_list
,
932 unsigned num_shaders
)
934 struct gl_uniform_block
*uniform_blocks
= NULL
;
936 /* Check that global variables defined in multiple shaders are consistent.
938 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
941 /* Check that uniform blocks between shaders for a stage agree. */
942 const int num_uniform_blocks
=
943 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
945 if (num_uniform_blocks
< 0)
948 /* Check that there is only a single definition of each function signature
949 * across all shaders.
951 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
952 foreach_list(node
, shader_list
[i
]->ir
) {
953 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
958 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
959 ir_function
*const other
=
960 shader_list
[j
]->symbols
->get_function(f
->name
);
962 /* If the other shader has no function (and therefore no function
963 * signatures) with the same name, skip to the next shader.
968 foreach_iter (exec_list_iterator
, iter
, *f
) {
969 ir_function_signature
*sig
=
970 (ir_function_signature
*) iter
.get();
972 if (!sig
->is_defined
|| sig
->is_builtin
)
975 ir_function_signature
*other_sig
=
976 other
->exact_matching_signature(& sig
->parameters
);
978 if ((other_sig
!= NULL
) && other_sig
->is_defined
979 && !other_sig
->is_builtin
) {
980 linker_error(prog
, "function `%s' is multiply defined",
989 /* Find the shader that defines main, and make a clone of it.
991 * Starting with the clone, search for undefined references. If one is
992 * found, find the shader that defines it. Clone the reference and add
993 * it to the shader. Repeat until there are no undefined references or
994 * until a reference cannot be resolved.
996 gl_shader
*main
= NULL
;
997 for (unsigned i
= 0; i
< num_shaders
; i
++) {
998 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
999 main
= shader_list
[i
];
1005 linker_error(prog
, "%s shader lacks `main'\n",
1006 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
1007 ? "vertex" : "fragment");
1011 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1012 linked
->ir
= new(linked
) exec_list
;
1013 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1015 linked
->UniformBlocks
= uniform_blocks
;
1016 linked
->NumUniformBlocks
= num_uniform_blocks
;
1017 ralloc_steal(linked
, linked
->UniformBlocks
);
1019 populate_symbol_table(linked
);
1021 /* The a pointer to the main function in the final linked shader (i.e., the
1022 * copy of the original shader that contained the main function).
1024 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1026 /* Move any instructions other than variable declarations or function
1027 * declarations into main.
1029 exec_node
*insertion_point
=
1030 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1033 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1034 if (shader_list
[i
] == main
)
1037 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1038 insertion_point
, true, linked
);
1041 /* Resolve initializers for global variables in the linked shader.
1043 unsigned num_linking_shaders
= num_shaders
;
1044 for (unsigned i
= 0; i
< num_shaders
; i
++)
1045 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
1047 gl_shader
**linking_shaders
=
1048 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
1050 memcpy(linking_shaders
, shader_list
,
1051 sizeof(linking_shaders
[0]) * num_shaders
);
1053 unsigned idx
= num_shaders
;
1054 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1055 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
1056 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
1057 idx
+= shader_list
[i
]->num_builtins_to_link
;
1060 assert(idx
== num_linking_shaders
);
1062 if (!link_function_calls(prog
, linked
, linking_shaders
,
1063 num_linking_shaders
)) {
1064 ctx
->Driver
.DeleteShader(ctx
, linked
);
1068 free(linking_shaders
);
1071 /* At this point linked should contain all of the linked IR, so
1072 * validate it to make sure nothing went wrong.
1075 validate_ir_tree(linked
->ir
);
1078 /* Make a pass over all variable declarations to ensure that arrays with
1079 * unspecified sizes have a size specified. The size is inferred from the
1080 * max_array_access field.
1082 if (linked
!= NULL
) {
1083 array_sizing_visitor v
;
1092 * Update the sizes of linked shader uniform arrays to the maximum
1095 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1097 * If one or more elements of an array are active,
1098 * GetActiveUniform will return the name of the array in name,
1099 * subject to the restrictions listed above. The type of the array
1100 * is returned in type. The size parameter contains the highest
1101 * array element index used, plus one. The compiler or linker
1102 * determines the highest index used. There will be only one
1103 * active uniform reported by the GL per uniform array.
1107 update_array_sizes(struct gl_shader_program
*prog
)
1109 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1110 if (prog
->_LinkedShaders
[i
] == NULL
)
1113 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1114 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1116 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1117 var
->mode
!= ir_var_shader_in
&&
1118 var
->mode
!= ir_var_shader_out
) ||
1119 !var
->type
->is_array())
1122 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1123 * will not be eliminated. Since we always do std140, just
1124 * don't resize arrays in UBOs.
1126 if (var
->is_in_uniform_block())
1129 unsigned int size
= var
->max_array_access
;
1130 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1131 if (prog
->_LinkedShaders
[j
] == NULL
)
1134 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1135 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1139 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1140 other_var
->max_array_access
> size
) {
1141 size
= other_var
->max_array_access
;
1146 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1147 /* If this is a built-in uniform (i.e., it's backed by some
1148 * fixed-function state), adjust the number of state slots to
1149 * match the new array size. The number of slots per array entry
1150 * is not known. It seems safe to assume that the total number of
1151 * slots is an integer multiple of the number of array elements.
1152 * Determine the number of slots per array element by dividing by
1153 * the old (total) size.
1155 if (var
->num_state_slots
> 0) {
1156 var
->num_state_slots
= (size
+ 1)
1157 * (var
->num_state_slots
/ var
->type
->length
);
1160 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1162 /* FINISHME: We should update the types of array
1163 * dereferences of this variable now.
1171 * Find a contiguous set of available bits in a bitmask.
1173 * \param used_mask Bits representing used (1) and unused (0) locations
1174 * \param needed_count Number of contiguous bits needed.
1177 * Base location of the available bits on success or -1 on failure.
1180 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1182 unsigned needed_mask
= (1 << needed_count
) - 1;
1183 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1185 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1186 * cannot optimize possibly infinite loops" for the loop below.
1188 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1191 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1192 if ((needed_mask
& ~used_mask
) == needed_mask
)
1203 * Assign locations for either VS inputs for FS outputs
1205 * \param prog Shader program whose variables need locations assigned
1206 * \param target_index Selector for the program target to receive location
1207 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1208 * \c MESA_SHADER_FRAGMENT.
1209 * \param max_index Maximum number of generic locations. This corresponds
1210 * to either the maximum number of draw buffers or the
1211 * maximum number of generic attributes.
1214 * If locations are successfully assigned, true is returned. Otherwise an
1215 * error is emitted to the shader link log and false is returned.
1218 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1219 unsigned target_index
,
1222 /* Mark invalid locations as being used.
1224 unsigned used_locations
= (max_index
>= 32)
1225 ? ~0 : ~((1 << max_index
) - 1);
1227 assert((target_index
== MESA_SHADER_VERTEX
)
1228 || (target_index
== MESA_SHADER_FRAGMENT
));
1230 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1234 /* Operate in a total of four passes.
1236 * 1. Invalidate the location assignments for all vertex shader inputs.
1238 * 2. Assign locations for inputs that have user-defined (via
1239 * glBindVertexAttribLocation) locations and outputs that have
1240 * user-defined locations (via glBindFragDataLocation).
1242 * 3. Sort the attributes without assigned locations by number of slots
1243 * required in decreasing order. Fragmentation caused by attribute
1244 * locations assigned by the application may prevent large attributes
1245 * from having enough contiguous space.
1247 * 4. Assign locations to any inputs without assigned locations.
1250 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1251 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1253 const enum ir_variable_mode direction
=
1254 (target_index
== MESA_SHADER_VERTEX
)
1255 ? ir_var_shader_in
: ir_var_shader_out
;
1258 /* Temporary storage for the set of attributes that need locations assigned.
1264 /* Used below in the call to qsort. */
1265 static int compare(const void *a
, const void *b
)
1267 const temp_attr
*const l
= (const temp_attr
*) a
;
1268 const temp_attr
*const r
= (const temp_attr
*) b
;
1270 /* Reversed because we want a descending order sort below. */
1271 return r
->slots
- l
->slots
;
1275 unsigned num_attr
= 0;
1277 foreach_list(node
, sh
->ir
) {
1278 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1280 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1283 if (var
->explicit_location
) {
1284 if ((var
->location
>= (int)(max_index
+ generic_base
))
1285 || (var
->location
< 0)) {
1287 "invalid explicit location %d specified for `%s'\n",
1289 ? var
->location
: var
->location
- generic_base
,
1293 } else if (target_index
== MESA_SHADER_VERTEX
) {
1296 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1297 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1298 var
->location
= binding
;
1299 var
->is_unmatched_generic_inout
= 0;
1301 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1305 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1306 assert(binding
>= FRAG_RESULT_DATA0
);
1307 var
->location
= binding
;
1308 var
->is_unmatched_generic_inout
= 0;
1310 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1316 /* If the variable is not a built-in and has a location statically
1317 * assigned in the shader (presumably via a layout qualifier), make sure
1318 * that it doesn't collide with other assigned locations. Otherwise,
1319 * add it to the list of variables that need linker-assigned locations.
1321 const unsigned slots
= count_attribute_slots(var
->type
);
1322 if (var
->location
!= -1) {
1323 if (var
->location
>= generic_base
&& var
->index
< 1) {
1324 /* From page 61 of the OpenGL 4.0 spec:
1326 * "LinkProgram will fail if the attribute bindings assigned
1327 * by BindAttribLocation do not leave not enough space to
1328 * assign a location for an active matrix attribute or an
1329 * active attribute array, both of which require multiple
1330 * contiguous generic attributes."
1332 * Previous versions of the spec contain similar language but omit
1333 * the bit about attribute arrays.
1335 * Page 61 of the OpenGL 4.0 spec also says:
1337 * "It is possible for an application to bind more than one
1338 * attribute name to the same location. This is referred to as
1339 * aliasing. This will only work if only one of the aliased
1340 * attributes is active in the executable program, or if no
1341 * path through the shader consumes more than one attribute of
1342 * a set of attributes aliased to the same location. A link
1343 * error can occur if the linker determines that every path
1344 * through the shader consumes multiple aliased attributes,
1345 * but implementations are not required to generate an error
1348 * These two paragraphs are either somewhat contradictory, or I
1349 * don't fully understand one or both of them.
1351 /* FINISHME: The code as currently written does not support
1352 * FINISHME: attribute location aliasing (see comment above).
1354 /* Mask representing the contiguous slots that will be used by
1357 const unsigned attr
= var
->location
- generic_base
;
1358 const unsigned use_mask
= (1 << slots
) - 1;
1360 /* Generate a link error if the set of bits requested for this
1361 * attribute overlaps any previously allocated bits.
1363 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1364 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1365 ? "vertex shader input" : "fragment shader output";
1367 "insufficient contiguous locations "
1368 "available for %s `%s' %d %d %d", string
,
1369 var
->name
, used_locations
, use_mask
, attr
);
1373 used_locations
|= (use_mask
<< attr
);
1379 to_assign
[num_attr
].slots
= slots
;
1380 to_assign
[num_attr
].var
= var
;
1384 /* If all of the attributes were assigned locations by the application (or
1385 * are built-in attributes with fixed locations), return early. This should
1386 * be the common case.
1391 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1393 if (target_index
== MESA_SHADER_VERTEX
) {
1394 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1395 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1396 * reserved to prevent it from being automatically allocated below.
1398 find_deref_visitor
find("gl_Vertex");
1400 if (find
.variable_found())
1401 used_locations
|= (1 << 0);
1404 for (unsigned i
= 0; i
< num_attr
; i
++) {
1405 /* Mask representing the contiguous slots that will be used by this
1408 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1410 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1413 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1414 ? "vertex shader input" : "fragment shader output";
1417 "insufficient contiguous locations "
1418 "available for %s `%s'",
1419 string
, to_assign
[i
].var
->name
);
1423 to_assign
[i
].var
->location
= generic_base
+ location
;
1424 to_assign
[i
].var
->is_unmatched_generic_inout
= 0;
1425 used_locations
|= (use_mask
<< location
);
1433 * Demote shader inputs and outputs that are not used in other stages
1436 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1438 foreach_list(node
, sh
->ir
) {
1439 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1441 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1444 /* A shader 'in' or 'out' variable is only really an input or output if
1445 * its value is used by other shader stages. This will cause the variable
1446 * to have a location assigned.
1448 if (var
->is_unmatched_generic_inout
) {
1449 var
->mode
= ir_var_auto
;
1456 * Store the gl_FragDepth layout in the gl_shader_program struct.
1459 store_fragdepth_layout(struct gl_shader_program
*prog
)
1461 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1465 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
1467 /* We don't look up the gl_FragDepth symbol directly because if
1468 * gl_FragDepth is not used in the shader, it's removed from the IR.
1469 * However, the symbol won't be removed from the symbol table.
1471 * We're only interested in the cases where the variable is NOT removed
1474 foreach_list(node
, ir
) {
1475 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1477 if (var
== NULL
|| var
->mode
!= ir_var_shader_out
) {
1481 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1482 switch (var
->depth_layout
) {
1483 case ir_depth_layout_none
:
1484 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
1486 case ir_depth_layout_any
:
1487 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
1489 case ir_depth_layout_greater
:
1490 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
1492 case ir_depth_layout_less
:
1493 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
1495 case ir_depth_layout_unchanged
:
1496 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
1507 * Validate the resources used by a program versus the implementation limits
1510 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1512 static const char *const shader_names
[MESA_SHADER_TYPES
] = {
1513 "vertex", "fragment", "geometry"
1516 const unsigned max_samplers
[MESA_SHADER_TYPES
] = {
1517 ctx
->Const
.MaxVertexTextureImageUnits
,
1518 ctx
->Const
.MaxTextureImageUnits
,
1519 ctx
->Const
.MaxGeometryTextureImageUnits
1522 const unsigned max_uniform_components
[MESA_SHADER_TYPES
] = {
1523 ctx
->Const
.VertexProgram
.MaxUniformComponents
,
1524 ctx
->Const
.FragmentProgram
.MaxUniformComponents
,
1525 0 /* FINISHME: Geometry shaders. */
1528 const unsigned max_uniform_blocks
[MESA_SHADER_TYPES
] = {
1529 ctx
->Const
.VertexProgram
.MaxUniformBlocks
,
1530 ctx
->Const
.FragmentProgram
.MaxUniformBlocks
,
1531 ctx
->Const
.GeometryProgram
.MaxUniformBlocks
,
1534 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1535 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1540 if (sh
->num_samplers
> max_samplers
[i
]) {
1541 linker_error(prog
, "Too many %s shader texture samplers",
1545 if (sh
->num_uniform_components
> max_uniform_components
[i
]) {
1546 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
1547 linker_warning(prog
, "Too many %s shader uniform components, "
1548 "but the driver will try to optimize them out; "
1549 "this is non-portable out-of-spec behavior\n",
1552 linker_error(prog
, "Too many %s shader uniform components",
1558 unsigned blocks
[MESA_SHADER_TYPES
] = {0};
1559 unsigned total_uniform_blocks
= 0;
1561 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
1562 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1563 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
1565 total_uniform_blocks
++;
1569 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
1570 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
1571 prog
->NumUniformBlocks
,
1572 ctx
->Const
.MaxCombinedUniformBlocks
);
1574 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1575 if (blocks
[i
] > max_uniform_blocks
[i
]) {
1576 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
1579 max_uniform_blocks
[i
]);
1586 return prog
->LinkStatus
;
1590 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1592 tfeedback_decl
*tfeedback_decls
= NULL
;
1593 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
1595 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1597 prog
->LinkStatus
= false;
1598 prog
->Validated
= false;
1599 prog
->_Used
= false;
1601 ralloc_free(prog
->InfoLog
);
1602 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1604 ralloc_free(prog
->UniformBlocks
);
1605 prog
->UniformBlocks
= NULL
;
1606 prog
->NumUniformBlocks
= 0;
1607 for (int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1608 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
1609 prog
->UniformBlockStageIndex
[i
] = NULL
;
1612 /* Separate the shaders into groups based on their type.
1614 struct gl_shader
**vert_shader_list
;
1615 unsigned num_vert_shaders
= 0;
1616 struct gl_shader
**frag_shader_list
;
1617 unsigned num_frag_shaders
= 0;
1619 vert_shader_list
= (struct gl_shader
**)
1620 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
1621 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
1623 unsigned min_version
= UINT_MAX
;
1624 unsigned max_version
= 0;
1625 const bool is_es_prog
=
1626 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
1627 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1628 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1629 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1631 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
1632 linker_error(prog
, "all shaders must use same shading "
1633 "language version\n");
1637 switch (prog
->Shaders
[i
]->Type
) {
1638 case GL_VERTEX_SHADER
:
1639 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1642 case GL_FRAGMENT_SHADER
:
1643 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1646 case GL_GEOMETRY_SHADER
:
1647 /* FINISHME: Support geometry shaders. */
1648 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
1653 /* Previous to GLSL version 1.30, different compilation units could mix and
1654 * match shading language versions. With GLSL 1.30 and later, the versions
1655 * of all shaders must match.
1657 * GLSL ES has never allowed mixing of shading language versions.
1659 if ((is_es_prog
|| max_version
>= 130)
1660 && min_version
!= max_version
) {
1661 linker_error(prog
, "all shaders must use same shading "
1662 "language version\n");
1666 prog
->Version
= max_version
;
1667 prog
->IsES
= is_es_prog
;
1669 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1670 if (prog
->_LinkedShaders
[i
] != NULL
)
1671 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1673 prog
->_LinkedShaders
[i
] = NULL
;
1676 /* Link all shaders for a particular stage and validate the result.
1678 if (num_vert_shaders
> 0) {
1679 gl_shader
*const sh
=
1680 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1686 if (!validate_vertex_shader_executable(prog
, sh
))
1689 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1693 if (num_frag_shaders
> 0) {
1694 gl_shader
*const sh
=
1695 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1701 if (!validate_fragment_shader_executable(prog
, sh
))
1704 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1708 /* Here begins the inter-stage linking phase. Some initial validation is
1709 * performed, then locations are assigned for uniforms, attributes, and
1712 if (cross_validate_uniforms(prog
)) {
1715 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1716 if (prog
->_LinkedShaders
[prev
] != NULL
)
1720 /* Validate the inputs of each stage with the output of the preceding
1723 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1724 if (prog
->_LinkedShaders
[i
] == NULL
)
1727 if (!cross_validate_outputs_to_inputs(prog
,
1728 prog
->_LinkedShaders
[prev
],
1729 prog
->_LinkedShaders
[i
]))
1735 prog
->LinkStatus
= true;
1738 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
1739 * it before optimization because we want most of the checks to get
1740 * dropped thanks to constant propagation.
1742 * This rule also applies to GLSL ES 3.00.
1744 if (max_version
>= (is_es_prog
? 300 : 130)) {
1745 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1747 lower_discard_flow(sh
->ir
);
1751 if (!interstage_cross_validate_uniform_blocks(prog
))
1754 /* Do common optimization before assigning storage for attributes,
1755 * uniforms, and varyings. Later optimization could possibly make
1756 * some of that unused.
1758 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1759 if (prog
->_LinkedShaders
[i
] == NULL
)
1762 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
1763 if (!prog
->LinkStatus
)
1766 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
) {
1767 lower_clip_distance(prog
->_LinkedShaders
[i
]);
1770 unsigned max_unroll
= ctx
->ShaderCompilerOptions
[i
].MaxUnrollIterations
;
1772 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, max_unroll
))
1776 /* Mark all generic shader inputs and outputs as unpaired. */
1777 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1778 link_invalidate_variable_locations(
1779 prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1780 VERT_ATTRIB_GENERIC0
, VARYING_SLOT_VAR0
);
1782 /* FINISHME: Geometry shaders not implemented yet */
1783 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1784 link_invalidate_variable_locations(
1785 prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1786 FRAG_ATTRIB_VAR0
, FRAG_RESULT_DATA0
);
1789 /* FINISHME: The value of the max_attribute_index parameter is
1790 * FINISHME: implementation dependent based on the value of
1791 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1792 * FINISHME: at least 16, so hardcode 16 for now.
1794 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
1798 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
1803 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1804 if (prog
->_LinkedShaders
[prev
] != NULL
)
1808 if (num_tfeedback_decls
!= 0) {
1809 /* From GL_EXT_transform_feedback:
1810 * A program will fail to link if:
1812 * * the <count> specified by TransformFeedbackVaryingsEXT is
1813 * non-zero, but the program object has no vertex or geometry
1816 if (prev
>= MESA_SHADER_FRAGMENT
) {
1817 linker_error(prog
, "Transform feedback varyings specified, but "
1818 "no vertex or geometry shader is present.");
1822 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
1823 prog
->TransformFeedback
.NumVarying
);
1824 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
1825 prog
->TransformFeedback
.VaryingNames
,
1830 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1831 if (prog
->_LinkedShaders
[i
] == NULL
)
1834 if (!assign_varying_locations(
1835 ctx
, mem_ctx
, prog
, prog
->_LinkedShaders
[prev
], prog
->_LinkedShaders
[i
],
1836 i
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
1843 if (prev
!= MESA_SHADER_FRAGMENT
&& num_tfeedback_decls
!= 0) {
1844 /* There was no fragment shader, but we still have to assign varying
1845 * locations for use by transform feedback.
1847 if (!assign_varying_locations(
1848 ctx
, mem_ctx
, prog
, prog
->_LinkedShaders
[prev
], NULL
, num_tfeedback_decls
,
1853 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
1856 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1857 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1860 /* Eliminate code that is now dead due to unused vertex outputs being
1863 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
, false))
1867 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
1868 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
1870 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
1871 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
1873 /* Eliminate code that is now dead due to unused geometry outputs being
1876 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
, false))
1880 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1881 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1883 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
1885 /* Eliminate code that is now dead due to unused fragment inputs being
1886 * demoted. This shouldn't actually do anything other than remove
1887 * declarations of the (now unused) global variables.
1889 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
, false))
1893 update_array_sizes(prog
);
1894 link_assign_uniform_locations(prog
);
1895 store_fragdepth_layout(prog
);
1897 if (!check_resources(ctx
, prog
))
1900 /* OpenGL ES requires that a vertex shader and a fragment shader both be
1901 * present in a linked program. By checking prog->IsES, we also
1902 * catch the GL_ARB_ES2_compatibility case.
1904 if (!prog
->InternalSeparateShader
&&
1905 (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)) {
1906 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
1907 linker_error(prog
, "program lacks a vertex shader\n");
1908 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1909 linker_error(prog
, "program lacks a fragment shader\n");
1913 /* FINISHME: Assign fragment shader output locations. */
1916 free(vert_shader_list
);
1918 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1919 if (prog
->_LinkedShaders
[i
] == NULL
)
1922 /* Retain any live IR, but trash the rest. */
1923 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
1925 /* The symbol table in the linked shaders may contain references to
1926 * variables that were removed (e.g., unused uniforms). Since it may
1927 * contain junk, there is no possible valid use. Delete it and set the
1930 delete prog
->_LinkedShaders
[i
]->symbols
;
1931 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
1934 ralloc_free(mem_ctx
);