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 "ir_optimization.h"
76 #include "main/shaderobj.h"
80 * Visitor that determines whether or not a variable is ever written.
82 class find_assignment_visitor
: public ir_hierarchical_visitor
{
84 find_assignment_visitor(const char *name
)
85 : name(name
), found(false)
90 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
92 ir_variable
*const var
= ir
->lhs
->variable_referenced();
94 if (strcmp(name
, var
->name
) == 0) {
99 return visit_continue_with_parent
;
102 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
104 exec_list_iterator sig_iter
= ir
->get_callee()->parameters
.iterator();
105 foreach_iter(exec_list_iterator
, iter
, *ir
) {
106 ir_rvalue
*param_rval
= (ir_rvalue
*)iter
.get();
107 ir_variable
*sig_param
= (ir_variable
*)sig_iter
.get();
109 if (sig_param
->mode
== ir_var_out
||
110 sig_param
->mode
== ir_var_inout
) {
111 ir_variable
*var
= param_rval
->variable_referenced();
112 if (var
&& strcmp(name
, var
->name
) == 0) {
120 return visit_continue_with_parent
;
123 bool variable_found()
129 const char *name
; /**< Find writes to a variable with this name. */
130 bool found
; /**< Was a write to the variable found? */
135 * Visitor that determines whether or not a variable is ever read.
137 class find_deref_visitor
: public ir_hierarchical_visitor
{
139 find_deref_visitor(const char *name
)
140 : name(name
), found(false)
145 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
147 if (strcmp(this->name
, ir
->var
->name
) == 0) {
152 return visit_continue
;
155 bool variable_found() const
161 const char *name
; /**< Find writes to a variable with this name. */
162 bool found
; /**< Was a write to the variable found? */
167 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
171 ralloc_strcat(&prog
->InfoLog
, "error: ");
173 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
176 prog
->LinkStatus
= false;
181 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
185 ralloc_strcat(&prog
->InfoLog
, "error: ");
187 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
194 link_invalidate_variable_locations(gl_shader
*sh
, enum ir_variable_mode mode
,
197 foreach_list(node
, sh
->ir
) {
198 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
200 if ((var
== NULL
) || (var
->mode
!= (unsigned) mode
))
203 /* Only assign locations for generic attributes / varyings / etc.
205 if ((var
->location
>= generic_base
) && !var
->explicit_location
)
212 * Determine the number of attribute slots required for a particular type
214 * This code is here because it implements the language rules of a specific
215 * GLSL version. Since it's a property of the language and not a property of
216 * types in general, it doesn't really belong in glsl_type.
219 count_attribute_slots(const glsl_type
*t
)
221 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
223 * "A scalar input counts the same amount against this limit as a vec4,
224 * so applications may want to consider packing groups of four
225 * unrelated float inputs together into a vector to better utilize the
226 * capabilities of the underlying hardware. A matrix input will use up
227 * multiple locations. The number of locations used will equal the
228 * number of columns in the matrix."
230 * The spec does not explicitly say how arrays are counted. However, it
231 * should be safe to assume the total number of slots consumed by an array
232 * is the number of entries in the array multiplied by the number of slots
233 * consumed by a single element of the array.
237 return t
->array_size() * count_attribute_slots(t
->element_type());
240 return t
->matrix_columns
;
247 * Verify that a vertex shader executable meets all semantic requirements.
249 * Also sets prog->Vert.UsesClipDistance as a side effect.
251 * \param shader Vertex shader executable to be verified
254 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
255 struct gl_shader
*shader
)
260 find_assignment_visitor
find("gl_Position");
261 find
.run(shader
->ir
);
262 if (!find
.variable_found()) {
263 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
267 if (prog
->Version
>= 130) {
268 /* From section 7.1 (Vertex Shader Special Variables) of the
271 * "It is an error for a shader to statically write both
272 * gl_ClipVertex and gl_ClipDistance."
274 find_assignment_visitor
clip_vertex("gl_ClipVertex");
275 find_assignment_visitor
clip_distance("gl_ClipDistance");
277 clip_vertex
.run(shader
->ir
);
278 clip_distance
.run(shader
->ir
);
279 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
280 linker_error(prog
, "vertex shader writes to both `gl_ClipVertex' "
281 "and `gl_ClipDistance'\n");
284 prog
->Vert
.UsesClipDistance
= clip_distance
.variable_found();
292 * Verify that a fragment shader executable meets all semantic requirements
294 * \param shader Fragment shader executable to be verified
297 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
298 struct gl_shader
*shader
)
303 find_assignment_visitor
frag_color("gl_FragColor");
304 find_assignment_visitor
frag_data("gl_FragData");
306 frag_color
.run(shader
->ir
);
307 frag_data
.run(shader
->ir
);
309 if (frag_color
.variable_found() && frag_data
.variable_found()) {
310 linker_error(prog
, "fragment shader writes to both "
311 "`gl_FragColor' and `gl_FragData'\n");
320 * Generate a string describing the mode of a variable
323 mode_string(const ir_variable
*var
)
327 return (var
->read_only
) ? "global constant" : "global variable";
329 case ir_var_uniform
: return "uniform";
330 case ir_var_in
: return "shader input";
331 case ir_var_out
: return "shader output";
332 case ir_var_inout
: return "shader inout";
334 case ir_var_const_in
:
335 case ir_var_temporary
:
337 assert(!"Should not get here.");
338 return "invalid variable";
344 * Perform validation of global variables used across multiple shaders
347 cross_validate_globals(struct gl_shader_program
*prog
,
348 struct gl_shader
**shader_list
,
349 unsigned num_shaders
,
352 /* Examine all of the uniforms in all of the shaders and cross validate
355 glsl_symbol_table variables
;
356 for (unsigned i
= 0; i
< num_shaders
; i
++) {
357 if (shader_list
[i
] == NULL
)
360 foreach_list(node
, shader_list
[i
]->ir
) {
361 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
366 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
369 /* Don't cross validate temporaries that are at global scope. These
370 * will eventually get pulled into the shaders 'main'.
372 if (var
->mode
== ir_var_temporary
)
375 /* If a global with this name has already been seen, verify that the
376 * new instance has the same type. In addition, if the globals have
377 * initializers, the values of the initializers must be the same.
379 ir_variable
*const existing
= variables
.get_variable(var
->name
);
380 if (existing
!= NULL
) {
381 if (var
->type
!= existing
->type
) {
382 /* Consider the types to be "the same" if both types are arrays
383 * of the same type and one of the arrays is implicitly sized.
384 * In addition, set the type of the linked variable to the
385 * explicitly sized array.
387 if (var
->type
->is_array()
388 && existing
->type
->is_array()
389 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
390 && ((var
->type
->length
== 0)
391 || (existing
->type
->length
== 0))) {
392 if (var
->type
->length
!= 0) {
393 existing
->type
= var
->type
;
396 linker_error(prog
, "%s `%s' declared as type "
397 "`%s' and type `%s'\n",
399 var
->name
, var
->type
->name
,
400 existing
->type
->name
);
405 if (var
->explicit_location
) {
406 if (existing
->explicit_location
407 && (var
->location
!= existing
->location
)) {
408 linker_error(prog
, "explicit locations for %s "
409 "`%s' have differing values\n",
410 mode_string(var
), var
->name
);
414 existing
->location
= var
->location
;
415 existing
->explicit_location
= true;
418 /* Validate layout qualifiers for gl_FragDepth.
420 * From the AMD/ARB_conservative_depth specs:
422 * "If gl_FragDepth is redeclared in any fragment shader in a
423 * program, it must be redeclared in all fragment shaders in
424 * that program that have static assignments to
425 * gl_FragDepth. All redeclarations of gl_FragDepth in all
426 * fragment shaders in a single program must have the same set
429 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
430 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
431 bool layout_differs
=
432 var
->depth_layout
!= existing
->depth_layout
;
434 if (layout_declared
&& layout_differs
) {
436 "All redeclarations of gl_FragDepth in all "
437 "fragment shaders in a single program must have "
438 "the same set of qualifiers.");
441 if (var
->used
&& layout_differs
) {
443 "If gl_FragDepth is redeclared with a layout "
444 "qualifier in any fragment shader, it must be "
445 "redeclared with the same layout qualifier in "
446 "all fragment shaders that have assignments to "
451 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
453 * "If a shared global has multiple initializers, the
454 * initializers must all be constant expressions, and they
455 * must all have the same value. Otherwise, a link error will
456 * result. (A shared global having only one initializer does
457 * not require that initializer to be a constant expression.)"
459 * Previous to 4.20 the GLSL spec simply said that initializers
460 * must have the same value. In this case of non-constant
461 * initializers, this was impossible to determine. As a result,
462 * no vendor actually implemented that behavior. The 4.20
463 * behavior matches the implemented behavior of at least one other
464 * vendor, so we'll implement that for all GLSL versions.
466 if (var
->constant_initializer
!= NULL
) {
467 if (existing
->constant_initializer
!= NULL
) {
468 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
469 linker_error(prog
, "initializers for %s "
470 "`%s' have differing values\n",
471 mode_string(var
), var
->name
);
475 /* If the first-seen instance of a particular uniform did not
476 * have an initializer but a later instance does, copy the
477 * initializer to the version stored in the symbol table.
479 /* FINISHME: This is wrong. The constant_value field should
480 * FINISHME: not be modified! Imagine a case where a shader
481 * FINISHME: without an initializer is linked in two different
482 * FINISHME: programs with shaders that have differing
483 * FINISHME: initializers. Linking with the first will
484 * FINISHME: modify the shader, and linking with the second
485 * FINISHME: will fail.
487 existing
->constant_initializer
=
488 var
->constant_initializer
->clone(ralloc_parent(existing
),
493 if (var
->has_initializer
) {
494 if (existing
->has_initializer
495 && (var
->constant_initializer
== NULL
496 || existing
->constant_initializer
== NULL
)) {
498 "shared global variable `%s' has multiple "
499 "non-constant initializers.\n",
504 /* Some instance had an initializer, so keep track of that. In
505 * this location, all sorts of initializers (constant or
506 * otherwise) will propagate the existence to the variable
507 * stored in the symbol table.
509 existing
->has_initializer
= true;
512 if (existing
->invariant
!= var
->invariant
) {
513 linker_error(prog
, "declarations for %s `%s' have "
514 "mismatching invariant qualifiers\n",
515 mode_string(var
), var
->name
);
518 if (existing
->centroid
!= var
->centroid
) {
519 linker_error(prog
, "declarations for %s `%s' have "
520 "mismatching centroid qualifiers\n",
521 mode_string(var
), var
->name
);
525 variables
.add_variable(var
);
534 * Perform validation of uniforms used across multiple shader stages
537 cross_validate_uniforms(struct gl_shader_program
*prog
)
539 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
540 MESA_SHADER_TYPES
, true);
545 * Validate that outputs from one stage match inputs of another
548 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
549 gl_shader
*producer
, gl_shader
*consumer
)
551 glsl_symbol_table parameters
;
552 /* FINISHME: Figure these out dynamically. */
553 const char *const producer_stage
= "vertex";
554 const char *const consumer_stage
= "fragment";
556 /* Find all shader outputs in the "producer" stage.
558 foreach_list(node
, producer
->ir
) {
559 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
561 /* FINISHME: For geometry shaders, this should also look for inout
562 * FINISHME: variables.
564 if ((var
== NULL
) || (var
->mode
!= ir_var_out
))
567 parameters
.add_variable(var
);
571 /* Find all shader inputs in the "consumer" stage. Any variables that have
572 * matching outputs already in the symbol table must have the same type and
575 foreach_list(node
, consumer
->ir
) {
576 ir_variable
*const input
= ((ir_instruction
*) node
)->as_variable();
578 /* FINISHME: For geometry shaders, this should also look for inout
579 * FINISHME: variables.
581 if ((input
== NULL
) || (input
->mode
!= ir_var_in
))
584 ir_variable
*const output
= parameters
.get_variable(input
->name
);
585 if (output
!= NULL
) {
586 /* Check that the types match between stages.
588 if (input
->type
!= output
->type
) {
589 /* There is a bit of a special case for gl_TexCoord. This
590 * built-in is unsized by default. Applications that variable
591 * access it must redeclare it with a size. There is some
592 * language in the GLSL spec that implies the fragment shader
593 * and vertex shader do not have to agree on this size. Other
594 * driver behave this way, and one or two applications seem to
597 * Neither declaration needs to be modified here because the array
598 * sizes are fixed later when update_array_sizes is called.
600 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
602 * "Unlike user-defined varying variables, the built-in
603 * varying variables don't have a strict one-to-one
604 * correspondence between the vertex language and the
605 * fragment language."
607 if (!output
->type
->is_array()
608 || (strncmp("gl_", output
->name
, 3) != 0)) {
610 "%s shader output `%s' declared as type `%s', "
611 "but %s shader input declared as type `%s'\n",
612 producer_stage
, output
->name
,
614 consumer_stage
, input
->type
->name
);
619 /* Check that all of the qualifiers match between stages.
621 if (input
->centroid
!= output
->centroid
) {
623 "%s shader output `%s' %s centroid qualifier, "
624 "but %s shader input %s centroid qualifier\n",
627 (output
->centroid
) ? "has" : "lacks",
629 (input
->centroid
) ? "has" : "lacks");
633 if (input
->invariant
!= output
->invariant
) {
635 "%s shader output `%s' %s invariant qualifier, "
636 "but %s shader input %s invariant qualifier\n",
639 (output
->invariant
) ? "has" : "lacks",
641 (input
->invariant
) ? "has" : "lacks");
645 if (input
->interpolation
!= output
->interpolation
) {
647 "%s shader output `%s' specifies %s "
648 "interpolation qualifier, "
649 "but %s shader input specifies %s "
650 "interpolation qualifier\n",
653 output
->interpolation_string(),
655 input
->interpolation_string());
666 * Populates a shaders symbol table with all global declarations
669 populate_symbol_table(gl_shader
*sh
)
671 sh
->symbols
= new(sh
) glsl_symbol_table
;
673 foreach_list(node
, sh
->ir
) {
674 ir_instruction
*const inst
= (ir_instruction
*) node
;
678 if ((func
= inst
->as_function()) != NULL
) {
679 sh
->symbols
->add_function(func
);
680 } else if ((var
= inst
->as_variable()) != NULL
) {
681 sh
->symbols
->add_variable(var
);
688 * Remap variables referenced in an instruction tree
690 * This is used when instruction trees are cloned from one shader and placed in
691 * another. These trees will contain references to \c ir_variable nodes that
692 * do not exist in the target shader. This function finds these \c ir_variable
693 * references and replaces the references with matching variables in the target
696 * If there is no matching variable in the target shader, a clone of the
697 * \c ir_variable is made and added to the target shader. The new variable is
698 * added to \b both the instruction stream and the symbol table.
700 * \param inst IR tree that is to be processed.
701 * \param symbols Symbol table containing global scope symbols in the
703 * \param instructions Instruction stream where new variable declarations
707 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
710 class remap_visitor
: public ir_hierarchical_visitor
{
712 remap_visitor(struct gl_shader
*target
,
715 this->target
= target
;
716 this->symbols
= target
->symbols
;
717 this->instructions
= target
->ir
;
721 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
723 if (ir
->var
->mode
== ir_var_temporary
) {
724 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
728 return visit_continue
;
731 ir_variable
*const existing
=
732 this->symbols
->get_variable(ir
->var
->name
);
733 if (existing
!= NULL
)
736 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
738 this->symbols
->add_variable(copy
);
739 this->instructions
->push_head(copy
);
743 return visit_continue
;
747 struct gl_shader
*target
;
748 glsl_symbol_table
*symbols
;
749 exec_list
*instructions
;
753 remap_visitor
v(target
, temps
);
760 * Move non-declarations from one instruction stream to another
762 * The intended usage pattern of this function is to pass the pointer to the
763 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
764 * pointer) for \c last and \c false for \c make_copies on the first
765 * call. Successive calls pass the return value of the previous call for
766 * \c last and \c true for \c make_copies.
768 * \param instructions Source instruction stream
769 * \param last Instruction after which new instructions should be
770 * inserted in the target instruction stream
771 * \param make_copies Flag selecting whether instructions in \c instructions
772 * should be copied (via \c ir_instruction::clone) into the
773 * target list or moved.
776 * The new "last" instruction in the target instruction stream. This pointer
777 * is suitable for use as the \c last parameter of a later call to this
781 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
782 bool make_copies
, gl_shader
*target
)
784 hash_table
*temps
= NULL
;
787 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
788 hash_table_pointer_compare
);
790 foreach_list_safe(node
, instructions
) {
791 ir_instruction
*inst
= (ir_instruction
*) node
;
793 if (inst
->as_function())
796 ir_variable
*var
= inst
->as_variable();
797 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
800 assert(inst
->as_assignment()
801 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
804 inst
= inst
->clone(target
, NULL
);
807 hash_table_insert(temps
, inst
, var
);
809 remap_variables(inst
, target
, temps
);
814 last
->insert_after(inst
);
819 hash_table_dtor(temps
);
825 * Get the function signature for main from a shader
827 static ir_function_signature
*
828 get_main_function_signature(gl_shader
*sh
)
830 ir_function
*const f
= sh
->symbols
->get_function("main");
832 exec_list void_parameters
;
834 /* Look for the 'void main()' signature and ensure that it's defined.
835 * This keeps the linker from accidentally pick a shader that just
836 * contains a prototype for main.
838 * We don't have to check for multiple definitions of main (in multiple
839 * shaders) because that would have already been caught above.
841 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
842 if ((sig
!= NULL
) && sig
->is_defined
) {
852 * Combine a group of shaders for a single stage to generate a linked shader
855 * If this function is supplied a single shader, it is cloned, and the new
856 * shader is returned.
858 static struct gl_shader
*
859 link_intrastage_shaders(void *mem_ctx
,
860 struct gl_context
*ctx
,
861 struct gl_shader_program
*prog
,
862 struct gl_shader
**shader_list
,
863 unsigned num_shaders
)
865 /* Check that global variables defined in multiple shaders are consistent.
867 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
870 /* Check that there is only a single definition of each function signature
871 * across all shaders.
873 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
874 foreach_list(node
, shader_list
[i
]->ir
) {
875 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
880 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
881 ir_function
*const other
=
882 shader_list
[j
]->symbols
->get_function(f
->name
);
884 /* If the other shader has no function (and therefore no function
885 * signatures) with the same name, skip to the next shader.
890 foreach_iter (exec_list_iterator
, iter
, *f
) {
891 ir_function_signature
*sig
=
892 (ir_function_signature
*) iter
.get();
894 if (!sig
->is_defined
|| sig
->is_builtin
)
897 ir_function_signature
*other_sig
=
898 other
->exact_matching_signature(& sig
->parameters
);
900 if ((other_sig
!= NULL
) && other_sig
->is_defined
901 && !other_sig
->is_builtin
) {
902 linker_error(prog
, "function `%s' is multiply defined",
911 /* Find the shader that defines main, and make a clone of it.
913 * Starting with the clone, search for undefined references. If one is
914 * found, find the shader that defines it. Clone the reference and add
915 * it to the shader. Repeat until there are no undefined references or
916 * until a reference cannot be resolved.
918 gl_shader
*main
= NULL
;
919 for (unsigned i
= 0; i
< num_shaders
; i
++) {
920 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
921 main
= shader_list
[i
];
927 linker_error(prog
, "%s shader lacks `main'\n",
928 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
929 ? "vertex" : "fragment");
933 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
934 linked
->ir
= new(linked
) exec_list
;
935 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
937 populate_symbol_table(linked
);
939 /* The a pointer to the main function in the final linked shader (i.e., the
940 * copy of the original shader that contained the main function).
942 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
944 /* Move any instructions other than variable declarations or function
945 * declarations into main.
947 exec_node
*insertion_point
=
948 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
951 for (unsigned i
= 0; i
< num_shaders
; i
++) {
952 if (shader_list
[i
] == main
)
955 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
956 insertion_point
, true, linked
);
959 /* Resolve initializers for global variables in the linked shader.
961 unsigned num_linking_shaders
= num_shaders
;
962 for (unsigned i
= 0; i
< num_shaders
; i
++)
963 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
965 gl_shader
**linking_shaders
=
966 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
968 memcpy(linking_shaders
, shader_list
,
969 sizeof(linking_shaders
[0]) * num_shaders
);
971 unsigned idx
= num_shaders
;
972 for (unsigned i
= 0; i
< num_shaders
; i
++) {
973 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
974 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
975 idx
+= shader_list
[i
]->num_builtins_to_link
;
978 assert(idx
== num_linking_shaders
);
980 if (!link_function_calls(prog
, linked
, linking_shaders
,
981 num_linking_shaders
)) {
982 ctx
->Driver
.DeleteShader(ctx
, linked
);
986 free(linking_shaders
);
989 /* At this point linked should contain all of the linked IR, so
990 * validate it to make sure nothing went wrong.
993 validate_ir_tree(linked
->ir
);
996 /* Make a pass over all variable declarations to ensure that arrays with
997 * unspecified sizes have a size specified. The size is inferred from the
998 * max_array_access field.
1000 if (linked
!= NULL
) {
1001 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1003 virtual ir_visitor_status
visit(ir_variable
*var
)
1005 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
1006 const glsl_type
*type
=
1007 glsl_type::get_array_instance(var
->type
->fields
.array
,
1008 var
->max_array_access
+ 1);
1010 assert(type
!= NULL
);
1014 return visit_continue
;
1025 struct uniform_node
{
1027 struct gl_uniform
*u
;
1032 * Update the sizes of linked shader uniform arrays to the maximum
1035 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1037 * If one or more elements of an array are active,
1038 * GetActiveUniform will return the name of the array in name,
1039 * subject to the restrictions listed above. The type of the array
1040 * is returned in type. The size parameter contains the highest
1041 * array element index used, plus one. The compiler or linker
1042 * determines the highest index used. There will be only one
1043 * active uniform reported by the GL per uniform array.
1047 update_array_sizes(struct gl_shader_program
*prog
)
1049 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1050 if (prog
->_LinkedShaders
[i
] == NULL
)
1053 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1054 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1056 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1057 var
->mode
!= ir_var_in
&&
1058 var
->mode
!= ir_var_out
) ||
1059 !var
->type
->is_array())
1062 unsigned int size
= var
->max_array_access
;
1063 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1064 if (prog
->_LinkedShaders
[j
] == NULL
)
1067 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1068 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1072 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1073 other_var
->max_array_access
> size
) {
1074 size
= other_var
->max_array_access
;
1079 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1080 /* If this is a built-in uniform (i.e., it's backed by some
1081 * fixed-function state), adjust the number of state slots to
1082 * match the new array size. The number of slots per array entry
1083 * is not known. It seems safe to assume that the total number of
1084 * slots is an integer multiple of the number of array elements.
1085 * Determine the number of slots per array element by dividing by
1086 * the old (total) size.
1088 if (var
->num_state_slots
> 0) {
1089 var
->num_state_slots
= (size
+ 1)
1090 * (var
->num_state_slots
/ var
->type
->length
);
1093 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1095 /* FINISHME: We should update the types of array
1096 * dereferences of this variable now.
1104 add_uniform(void *mem_ctx
, exec_list
*uniforms
, struct hash_table
*ht
,
1105 const char *name
, const glsl_type
*type
, GLenum shader_type
,
1106 unsigned *next_shader_pos
, unsigned *total_uniforms
)
1108 if (type
->is_record()) {
1109 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1110 const glsl_type
*field_type
= type
->fields
.structure
[i
].type
;
1111 char *field_name
= ralloc_asprintf(mem_ctx
, "%s.%s", name
,
1112 type
->fields
.structure
[i
].name
);
1114 add_uniform(mem_ctx
, uniforms
, ht
, field_name
, field_type
,
1115 shader_type
, next_shader_pos
, total_uniforms
);
1118 uniform_node
*n
= (uniform_node
*) hash_table_find(ht
, name
);
1119 unsigned int vec4_slots
;
1120 const glsl_type
*array_elem_type
= NULL
;
1122 if (type
->is_array()) {
1123 array_elem_type
= type
->fields
.array
;
1124 /* Array of structures. */
1125 if (array_elem_type
->is_record()) {
1126 for (unsigned int i
= 0; i
< type
->length
; i
++) {
1127 char *elem_name
= ralloc_asprintf(mem_ctx
, "%s[%d]", name
, i
);
1128 add_uniform(mem_ctx
, uniforms
, ht
, elem_name
, array_elem_type
,
1129 shader_type
, next_shader_pos
, total_uniforms
);
1135 /* Fix the storage size of samplers at 1 vec4 each. Be sure to pad out
1136 * vectors to vec4 slots.
1138 if (type
->is_array()) {
1139 if (array_elem_type
->is_sampler())
1140 vec4_slots
= type
->length
;
1142 vec4_slots
= type
->length
* array_elem_type
->matrix_columns
;
1143 } else if (type
->is_sampler()) {
1146 vec4_slots
= type
->matrix_columns
;
1150 n
= (uniform_node
*) calloc(1, sizeof(struct uniform_node
));
1151 n
->u
= (gl_uniform
*) calloc(1, sizeof(struct gl_uniform
));
1152 n
->slots
= vec4_slots
;
1154 n
->u
->Name
= strdup(name
);
1159 (*total_uniforms
)++;
1161 hash_table_insert(ht
, n
, name
);
1162 uniforms
->push_tail(& n
->link
);
1165 switch (shader_type
) {
1166 case GL_VERTEX_SHADER
:
1167 n
->u
->VertPos
= *next_shader_pos
;
1169 case GL_FRAGMENT_SHADER
:
1170 n
->u
->FragPos
= *next_shader_pos
;
1172 case GL_GEOMETRY_SHADER
:
1173 n
->u
->GeomPos
= *next_shader_pos
;
1177 (*next_shader_pos
) += vec4_slots
;
1182 assign_uniform_locations(struct gl_shader_program
*prog
)
1186 unsigned total_uniforms
= 0;
1187 hash_table
*ht
= hash_table_ctor(32, hash_table_string_hash
,
1188 hash_table_string_compare
);
1189 void *mem_ctx
= ralloc_context(NULL
);
1191 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1192 if (prog
->_LinkedShaders
[i
] == NULL
)
1195 unsigned next_position
= 0;
1197 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1198 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1200 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
))
1203 if (strncmp(var
->name
, "gl_", 3) == 0) {
1204 /* At the moment, we don't allocate uniform locations for
1205 * builtin uniforms. It's permitted by spec, and we'll
1206 * likely switch to doing that at some point, but not yet.
1211 var
->location
= next_position
;
1212 add_uniform(mem_ctx
, &uniforms
, ht
, var
->name
, var
->type
,
1213 prog
->_LinkedShaders
[i
]->Type
,
1214 &next_position
, &total_uniforms
);
1218 ralloc_free(mem_ctx
);
1220 gl_uniform_list
*ul
= (gl_uniform_list
*)
1221 calloc(1, sizeof(gl_uniform_list
));
1223 ul
->Size
= total_uniforms
;
1224 ul
->NumUniforms
= total_uniforms
;
1225 ul
->Uniforms
= (gl_uniform
*) calloc(total_uniforms
, sizeof(gl_uniform
));
1229 for (uniform_node
*node
= (uniform_node
*) uniforms
.head
1230 ; node
->link
.next
!= NULL
1232 next
= (uniform_node
*) node
->link
.next
;
1234 node
->link
.remove();
1235 memcpy(&ul
->Uniforms
[idx
], node
->u
, sizeof(gl_uniform
));
1242 hash_table_dtor(ht
);
1244 prog
->Uniforms
= ul
;
1249 * Find a contiguous set of available bits in a bitmask.
1251 * \param used_mask Bits representing used (1) and unused (0) locations
1252 * \param needed_count Number of contiguous bits needed.
1255 * Base location of the available bits on success or -1 on failure.
1258 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1260 unsigned needed_mask
= (1 << needed_count
) - 1;
1261 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1263 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1264 * cannot optimize possibly infinite loops" for the loop below.
1266 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1269 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1270 if ((needed_mask
& ~used_mask
) == needed_mask
)
1281 * Assign locations for either VS inputs for FS outputs
1283 * \param prog Shader program whose variables need locations assigned
1284 * \param target_index Selector for the program target to receive location
1285 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1286 * \c MESA_SHADER_FRAGMENT.
1287 * \param max_index Maximum number of generic locations. This corresponds
1288 * to either the maximum number of draw buffers or the
1289 * maximum number of generic attributes.
1292 * If locations are successfully assigned, true is returned. Otherwise an
1293 * error is emitted to the shader link log and false is returned.
1296 * Locations set via \c glBindFragDataLocation are not currently supported.
1297 * Only locations assigned automatically by the linker, explicitly set by a
1298 * layout qualifier, or explicitly set by a built-in variable (e.g., \c
1299 * gl_FragColor) are supported for fragment shaders.
1302 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1303 unsigned target_index
,
1306 /* Mark invalid locations as being used.
1308 unsigned used_locations
= (max_index
>= 32)
1309 ? ~0 : ~((1 << max_index
) - 1);
1311 assert((target_index
== MESA_SHADER_VERTEX
)
1312 || (target_index
== MESA_SHADER_FRAGMENT
));
1314 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1318 /* Operate in a total of four passes.
1320 * 1. Invalidate the location assignments for all vertex shader inputs.
1322 * 2. Assign locations for inputs that have user-defined (via
1323 * glBindVertexAttribLocation) locations.
1325 * 3. Sort the attributes without assigned locations by number of slots
1326 * required in decreasing order. Fragmentation caused by attribute
1327 * locations assigned by the application may prevent large attributes
1328 * from having enough contiguous space.
1330 * 4. Assign locations to any inputs without assigned locations.
1333 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1334 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1336 const enum ir_variable_mode direction
=
1337 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1340 link_invalidate_variable_locations(sh
, direction
, generic_base
);
1342 /* Temporary storage for the set of attributes that need locations assigned.
1348 /* Used below in the call to qsort. */
1349 static int compare(const void *a
, const void *b
)
1351 const temp_attr
*const l
= (const temp_attr
*) a
;
1352 const temp_attr
*const r
= (const temp_attr
*) b
;
1354 /* Reversed because we want a descending order sort below. */
1355 return r
->slots
- l
->slots
;
1359 unsigned num_attr
= 0;
1361 foreach_list(node
, sh
->ir
) {
1362 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1364 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1367 if (var
->explicit_location
) {
1368 if ((var
->location
>= (int)(max_index
+ generic_base
))
1369 || (var
->location
< 0)) {
1371 "invalid explicit location %d specified for `%s'\n",
1373 ? var
->location
: var
->location
- generic_base
,
1377 } else if (target_index
== MESA_SHADER_VERTEX
) {
1380 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1381 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1382 var
->location
= binding
;
1386 /* If the variable is not a built-in and has a location statically
1387 * assigned in the shader (presumably via a layout qualifier), make sure
1388 * that it doesn't collide with other assigned locations. Otherwise,
1389 * add it to the list of variables that need linker-assigned locations.
1391 const unsigned slots
= count_attribute_slots(var
->type
);
1392 if (var
->location
!= -1) {
1393 if (var
->location
>= generic_base
) {
1394 /* From page 61 of the OpenGL 4.0 spec:
1396 * "LinkProgram will fail if the attribute bindings assigned
1397 * by BindAttribLocation do not leave not enough space to
1398 * assign a location for an active matrix attribute or an
1399 * active attribute array, both of which require multiple
1400 * contiguous generic attributes."
1402 * Previous versions of the spec contain similar language but omit
1403 * the bit about attribute arrays.
1405 * Page 61 of the OpenGL 4.0 spec also says:
1407 * "It is possible for an application to bind more than one
1408 * attribute name to the same location. This is referred to as
1409 * aliasing. This will only work if only one of the aliased
1410 * attributes is active in the executable program, or if no
1411 * path through the shader consumes more than one attribute of
1412 * a set of attributes aliased to the same location. A link
1413 * error can occur if the linker determines that every path
1414 * through the shader consumes multiple aliased attributes,
1415 * but implementations are not required to generate an error
1418 * These two paragraphs are either somewhat contradictory, or I
1419 * don't fully understand one or both of them.
1421 /* FINISHME: The code as currently written does not support
1422 * FINISHME: attribute location aliasing (see comment above).
1424 /* Mask representing the contiguous slots that will be used by
1427 const unsigned attr
= var
->location
- generic_base
;
1428 const unsigned use_mask
= (1 << slots
) - 1;
1430 /* Generate a link error if the set of bits requested for this
1431 * attribute overlaps any previously allocated bits.
1433 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1435 "insufficient contiguous attribute locations "
1436 "available for vertex shader input `%s'",
1441 used_locations
|= (use_mask
<< attr
);
1447 to_assign
[num_attr
].slots
= slots
;
1448 to_assign
[num_attr
].var
= var
;
1452 /* If all of the attributes were assigned locations by the application (or
1453 * are built-in attributes with fixed locations), return early. This should
1454 * be the common case.
1459 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1461 if (target_index
== MESA_SHADER_VERTEX
) {
1462 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1463 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1464 * reserved to prevent it from being automatically allocated below.
1466 find_deref_visitor
find("gl_Vertex");
1468 if (find
.variable_found())
1469 used_locations
|= (1 << 0);
1472 for (unsigned i
= 0; i
< num_attr
; i
++) {
1473 /* Mask representing the contiguous slots that will be used by this
1476 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1478 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1481 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1482 ? "vertex shader input" : "fragment shader output";
1485 "insufficient contiguous attribute locations "
1486 "available for %s `%s'",
1487 string
, to_assign
[i
].var
->name
);
1491 to_assign
[i
].var
->location
= generic_base
+ location
;
1492 used_locations
|= (use_mask
<< location
);
1500 * Demote shader inputs and outputs that are not used in other stages
1503 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1505 foreach_list(node
, sh
->ir
) {
1506 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1508 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1511 /* A shader 'in' or 'out' variable is only really an input or output if
1512 * its value is used by other shader stages. This will cause the variable
1513 * to have a location assigned.
1515 if (var
->location
== -1) {
1516 var
->mode
= ir_var_auto
;
1523 assign_varying_locations(struct gl_context
*ctx
,
1524 struct gl_shader_program
*prog
,
1525 gl_shader
*producer
, gl_shader
*consumer
)
1527 /* FINISHME: Set dynamically when geometry shader support is added. */
1528 unsigned output_index
= VERT_RESULT_VAR0
;
1529 unsigned input_index
= FRAG_ATTRIB_VAR0
;
1531 /* Operate in a total of three passes.
1533 * 1. Assign locations for any matching inputs and outputs.
1535 * 2. Mark output variables in the producer that do not have locations as
1536 * not being outputs. This lets the optimizer eliminate them.
1538 * 3. Mark input variables in the consumer that do not have locations as
1539 * not being inputs. This lets the optimizer eliminate them.
1542 link_invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
1543 link_invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
1545 foreach_list(node
, producer
->ir
) {
1546 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
1548 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
)
1549 || (output_var
->location
!= -1))
1552 ir_variable
*const input_var
=
1553 consumer
->symbols
->get_variable(output_var
->name
);
1555 if ((input_var
== NULL
) || (input_var
->mode
!= ir_var_in
))
1558 assert(input_var
->location
== -1);
1560 output_var
->location
= output_index
;
1561 input_var
->location
= input_index
;
1563 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1564 assert(!output_var
->type
->is_record());
1566 if (output_var
->type
->is_array()) {
1567 const unsigned slots
= output_var
->type
->length
1568 * output_var
->type
->fields
.array
->matrix_columns
;
1570 output_index
+= slots
;
1571 input_index
+= slots
;
1573 const unsigned slots
= output_var
->type
->matrix_columns
;
1575 output_index
+= slots
;
1576 input_index
+= slots
;
1580 unsigned varying_vectors
= 0;
1582 foreach_list(node
, consumer
->ir
) {
1583 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1585 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
1588 if (var
->location
== -1) {
1589 if (prog
->Version
<= 120) {
1590 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1592 * Only those varying variables used (i.e. read) in
1593 * the fragment shader executable must be written to
1594 * by the vertex shader executable; declaring
1595 * superfluous varying variables in a vertex shader is
1598 * We interpret this text as meaning that the VS must
1599 * write the variable for the FS to read it. See
1600 * "glsl1-varying read but not written" in piglit.
1603 linker_error(prog
, "fragment shader varying %s not written "
1604 "by vertex shader\n.", var
->name
);
1607 /* An 'in' variable is only really a shader input if its
1608 * value is written by the previous stage.
1610 var
->mode
= ir_var_auto
;
1612 /* The packing rules are used for vertex shader inputs are also used
1613 * for fragment shader inputs.
1615 varying_vectors
+= count_attribute_slots(var
->type
);
1619 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1620 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
1621 linker_error(prog
, "shader uses too many varying vectors "
1623 varying_vectors
, ctx
->Const
.MaxVarying
);
1627 const unsigned float_components
= varying_vectors
* 4;
1628 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
1629 linker_error(prog
, "shader uses too many varying components "
1631 float_components
, ctx
->Const
.MaxVarying
* 4);
1641 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
1643 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
1645 prog
->LinkStatus
= false;
1646 prog
->Validated
= false;
1647 prog
->_Used
= false;
1649 if (prog
->InfoLog
!= NULL
)
1650 ralloc_free(prog
->InfoLog
);
1652 prog
->InfoLog
= ralloc_strdup(NULL
, "");
1654 /* Separate the shaders into groups based on their type.
1656 struct gl_shader
**vert_shader_list
;
1657 unsigned num_vert_shaders
= 0;
1658 struct gl_shader
**frag_shader_list
;
1659 unsigned num_frag_shaders
= 0;
1661 vert_shader_list
= (struct gl_shader
**)
1662 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
1663 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
1665 unsigned min_version
= UINT_MAX
;
1666 unsigned max_version
= 0;
1667 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
1668 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
1669 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
1671 switch (prog
->Shaders
[i
]->Type
) {
1672 case GL_VERTEX_SHADER
:
1673 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
1676 case GL_FRAGMENT_SHADER
:
1677 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
1680 case GL_GEOMETRY_SHADER
:
1681 /* FINISHME: Support geometry shaders. */
1682 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
1687 /* Previous to GLSL version 1.30, different compilation units could mix and
1688 * match shading language versions. With GLSL 1.30 and later, the versions
1689 * of all shaders must match.
1691 assert(min_version
>= 100);
1692 assert(max_version
<= 130);
1693 if ((max_version
>= 130 || min_version
== 100)
1694 && min_version
!= max_version
) {
1695 linker_error(prog
, "all shaders must use same shading "
1696 "language version\n");
1700 prog
->Version
= max_version
;
1702 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1703 if (prog
->_LinkedShaders
[i
] != NULL
)
1704 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
1706 prog
->_LinkedShaders
[i
] = NULL
;
1709 /* Link all shaders for a particular stage and validate the result.
1711 if (num_vert_shaders
> 0) {
1712 gl_shader
*const sh
=
1713 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
1719 if (!validate_vertex_shader_executable(prog
, sh
))
1722 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1726 if (num_frag_shaders
> 0) {
1727 gl_shader
*const sh
=
1728 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
1734 if (!validate_fragment_shader_executable(prog
, sh
))
1737 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
1741 /* Here begins the inter-stage linking phase. Some initial validation is
1742 * performed, then locations are assigned for uniforms, attributes, and
1745 if (cross_validate_uniforms(prog
)) {
1748 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1749 if (prog
->_LinkedShaders
[prev
] != NULL
)
1753 /* Validate the inputs of each stage with the output of the preceding
1756 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1757 if (prog
->_LinkedShaders
[i
] == NULL
)
1760 if (!cross_validate_outputs_to_inputs(prog
,
1761 prog
->_LinkedShaders
[prev
],
1762 prog
->_LinkedShaders
[i
]))
1768 prog
->LinkStatus
= true;
1771 /* Do common optimization before assigning storage for attributes,
1772 * uniforms, and varyings. Later optimization could possibly make
1773 * some of that unused.
1775 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1776 if (prog
->_LinkedShaders
[i
] == NULL
)
1779 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
1780 if (!prog
->LinkStatus
)
1783 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
)
1784 lower_clip_distance(prog
->_LinkedShaders
[i
]->ir
);
1786 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, 32))
1790 /* FINISHME: The value of the max_attribute_index parameter is
1791 * FINISHME: implementation dependent based on the value of
1792 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
1793 * FINISHME: at least 16, so hardcode 16 for now.
1795 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
1799 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, ctx
->Const
.MaxDrawBuffers
)) {
1804 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
1805 if (prog
->_LinkedShaders
[prev
] != NULL
)
1809 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
1810 if (prog
->_LinkedShaders
[i
] == NULL
)
1813 if (!assign_varying_locations(ctx
, prog
,
1814 prog
->_LinkedShaders
[prev
],
1815 prog
->_LinkedShaders
[i
])) {
1822 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
1823 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
1826 /* Eliminate code that is now dead due to unused vertex outputs being
1829 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
, false))
1833 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
1834 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
1836 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1837 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
1838 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
1840 /* Eliminate code that is now dead due to unused geometry outputs being
1843 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
, false))
1847 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
1848 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
1850 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
1852 /* Eliminate code that is now dead due to unused fragment inputs being
1853 * demoted. This shouldn't actually do anything other than remove
1854 * declarations of the (now unused) global variables.
1856 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
, false))
1860 update_array_sizes(prog
);
1861 assign_uniform_locations(prog
);
1863 /* OpenGL ES requires that a vertex shader and a fragment shader both be
1864 * present in a linked program. By checking for use of shading language
1865 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
1867 if (!prog
->InternalSeparateShader
&&
1868 (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100)) {
1869 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
1870 linker_error(prog
, "program lacks a vertex shader\n");
1871 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1872 linker_error(prog
, "program lacks a fragment shader\n");
1876 /* FINISHME: Assign fragment shader output locations. */
1879 free(vert_shader_list
);
1881 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1882 if (prog
->_LinkedShaders
[i
] == NULL
)
1885 /* Retain any live IR, but trash the rest. */
1886 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
1888 /* The symbol table in the linked shaders may contain references to
1889 * variables that were removed (e.g., unused uniforms). Since it may
1890 * contain junk, there is no possible valid use. Delete it and set the
1893 delete prog
->_LinkedShaders
[i
]->symbols
;
1894 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
1897 ralloc_free(mem_ctx
);