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 * Update the sizes of linked shader uniform arrays to the maximum
1028 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1030 * If one or more elements of an array are active,
1031 * GetActiveUniform will return the name of the array in name,
1032 * subject to the restrictions listed above. The type of the array
1033 * is returned in type. The size parameter contains the highest
1034 * array element index used, plus one. The compiler or linker
1035 * determines the highest index used. There will be only one
1036 * active uniform reported by the GL per uniform array.
1040 update_array_sizes(struct gl_shader_program
*prog
)
1042 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1043 if (prog
->_LinkedShaders
[i
] == NULL
)
1046 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1047 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1049 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1050 var
->mode
!= ir_var_in
&&
1051 var
->mode
!= ir_var_out
) ||
1052 !var
->type
->is_array())
1055 unsigned int size
= var
->max_array_access
;
1056 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1057 if (prog
->_LinkedShaders
[j
] == NULL
)
1060 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1061 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1065 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1066 other_var
->max_array_access
> size
) {
1067 size
= other_var
->max_array_access
;
1072 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1073 /* If this is a built-in uniform (i.e., it's backed by some
1074 * fixed-function state), adjust the number of state slots to
1075 * match the new array size. The number of slots per array entry
1076 * is not known. It seems safe to assume that the total number of
1077 * slots is an integer multiple of the number of array elements.
1078 * Determine the number of slots per array element by dividing by
1079 * the old (total) size.
1081 if (var
->num_state_slots
> 0) {
1082 var
->num_state_slots
= (size
+ 1)
1083 * (var
->num_state_slots
/ var
->type
->length
);
1086 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1088 /* FINISHME: We should update the types of array
1089 * dereferences of this variable now.
1097 * Find a contiguous set of available bits in a bitmask.
1099 * \param used_mask Bits representing used (1) and unused (0) locations
1100 * \param needed_count Number of contiguous bits needed.
1103 * Base location of the available bits on success or -1 on failure.
1106 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1108 unsigned needed_mask
= (1 << needed_count
) - 1;
1109 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1111 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1112 * cannot optimize possibly infinite loops" for the loop below.
1114 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1117 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1118 if ((needed_mask
& ~used_mask
) == needed_mask
)
1129 * Assign locations for either VS inputs for FS outputs
1131 * \param prog Shader program whose variables need locations assigned
1132 * \param target_index Selector for the program target to receive location
1133 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1134 * \c MESA_SHADER_FRAGMENT.
1135 * \param max_index Maximum number of generic locations. This corresponds
1136 * to either the maximum number of draw buffers or the
1137 * maximum number of generic attributes.
1140 * If locations are successfully assigned, true is returned. Otherwise an
1141 * error is emitted to the shader link log and false is returned.
1144 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1145 unsigned target_index
,
1148 /* Mark invalid locations as being used.
1150 unsigned used_locations
= (max_index
>= 32)
1151 ? ~0 : ~((1 << max_index
) - 1);
1153 assert((target_index
== MESA_SHADER_VERTEX
)
1154 || (target_index
== MESA_SHADER_FRAGMENT
));
1156 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1160 /* Operate in a total of four passes.
1162 * 1. Invalidate the location assignments for all vertex shader inputs.
1164 * 2. Assign locations for inputs that have user-defined (via
1165 * glBindVertexAttribLocation) locations and outputs that have
1166 * user-defined locations (via glBindFragDataLocation).
1168 * 3. Sort the attributes without assigned locations by number of slots
1169 * required in decreasing order. Fragmentation caused by attribute
1170 * locations assigned by the application may prevent large attributes
1171 * from having enough contiguous space.
1173 * 4. Assign locations to any inputs without assigned locations.
1176 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1177 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1179 const enum ir_variable_mode direction
=
1180 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1183 link_invalidate_variable_locations(sh
, direction
, generic_base
);
1185 /* Temporary storage for the set of attributes that need locations assigned.
1191 /* Used below in the call to qsort. */
1192 static int compare(const void *a
, const void *b
)
1194 const temp_attr
*const l
= (const temp_attr
*) a
;
1195 const temp_attr
*const r
= (const temp_attr
*) b
;
1197 /* Reversed because we want a descending order sort below. */
1198 return r
->slots
- l
->slots
;
1202 unsigned num_attr
= 0;
1204 foreach_list(node
, sh
->ir
) {
1205 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1207 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1210 if (var
->explicit_location
) {
1211 if ((var
->location
>= (int)(max_index
+ generic_base
))
1212 || (var
->location
< 0)) {
1214 "invalid explicit location %d specified for `%s'\n",
1216 ? var
->location
: var
->location
- generic_base
,
1220 } else if (target_index
== MESA_SHADER_VERTEX
) {
1223 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1224 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1225 var
->location
= binding
;
1227 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1230 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1231 assert(binding
>= FRAG_RESULT_DATA0
);
1232 var
->location
= binding
;
1236 /* If the variable is not a built-in and has a location statically
1237 * assigned in the shader (presumably via a layout qualifier), make sure
1238 * that it doesn't collide with other assigned locations. Otherwise,
1239 * add it to the list of variables that need linker-assigned locations.
1241 const unsigned slots
= count_attribute_slots(var
->type
);
1242 if (var
->location
!= -1) {
1243 if (var
->location
>= generic_base
) {
1244 /* From page 61 of the OpenGL 4.0 spec:
1246 * "LinkProgram will fail if the attribute bindings assigned
1247 * by BindAttribLocation do not leave not enough space to
1248 * assign a location for an active matrix attribute or an
1249 * active attribute array, both of which require multiple
1250 * contiguous generic attributes."
1252 * Previous versions of the spec contain similar language but omit
1253 * the bit about attribute arrays.
1255 * Page 61 of the OpenGL 4.0 spec also says:
1257 * "It is possible for an application to bind more than one
1258 * attribute name to the same location. This is referred to as
1259 * aliasing. This will only work if only one of the aliased
1260 * attributes is active in the executable program, or if no
1261 * path through the shader consumes more than one attribute of
1262 * a set of attributes aliased to the same location. A link
1263 * error can occur if the linker determines that every path
1264 * through the shader consumes multiple aliased attributes,
1265 * but implementations are not required to generate an error
1268 * These two paragraphs are either somewhat contradictory, or I
1269 * don't fully understand one or both of them.
1271 /* FINISHME: The code as currently written does not support
1272 * FINISHME: attribute location aliasing (see comment above).
1274 /* Mask representing the contiguous slots that will be used by
1277 const unsigned attr
= var
->location
- generic_base
;
1278 const unsigned use_mask
= (1 << slots
) - 1;
1280 /* Generate a link error if the set of bits requested for this
1281 * attribute overlaps any previously allocated bits.
1283 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1285 "insufficient contiguous attribute locations "
1286 "available for vertex shader input `%s'",
1291 used_locations
|= (use_mask
<< attr
);
1297 to_assign
[num_attr
].slots
= slots
;
1298 to_assign
[num_attr
].var
= var
;
1302 /* If all of the attributes were assigned locations by the application (or
1303 * are built-in attributes with fixed locations), return early. This should
1304 * be the common case.
1309 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1311 if (target_index
== MESA_SHADER_VERTEX
) {
1312 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1313 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1314 * reserved to prevent it from being automatically allocated below.
1316 find_deref_visitor
find("gl_Vertex");
1318 if (find
.variable_found())
1319 used_locations
|= (1 << 0);
1322 for (unsigned i
= 0; i
< num_attr
; i
++) {
1323 /* Mask representing the contiguous slots that will be used by this
1326 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1328 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1331 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1332 ? "vertex shader input" : "fragment shader output";
1335 "insufficient contiguous attribute locations "
1336 "available for %s `%s'",
1337 string
, to_assign
[i
].var
->name
);
1341 to_assign
[i
].var
->location
= generic_base
+ location
;
1342 used_locations
|= (use_mask
<< location
);
1350 * Demote shader inputs and outputs that are not used in other stages
1353 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1355 foreach_list(node
, sh
->ir
) {
1356 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1358 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1361 /* A shader 'in' or 'out' variable is only really an input or output if
1362 * its value is used by other shader stages. This will cause the variable
1363 * to have a location assigned.
1365 if (var
->location
== -1) {
1366 var
->mode
= ir_var_auto
;
1373 * Data structure tracking information about a transform feedback declaration
1376 class tfeedback_decl
1379 bool init(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1380 const void *mem_ctx
, const char *input
);
1381 static bool is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
);
1382 bool assign_location(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1383 ir_variable
*output_var
);
1384 bool store(struct gl_shader_program
*prog
,
1385 struct gl_transform_feedback_info
*info
, unsigned buffer
,
1386 unsigned varying
) const;
1390 * True if assign_location() has been called for this object.
1392 bool is_assigned() const
1394 return this->location
!= -1;
1398 * Determine whether this object refers to the variable var.
1400 bool matches_var(ir_variable
*var
) const
1402 return strcmp(var
->name
, this->var_name
) == 0;
1406 * The total number of varying components taken up by this variable. Only
1407 * valid if is_assigned() is true.
1409 unsigned num_components() const
1411 return this->vector_elements
* this->matrix_columns
;
1416 * The name that was supplied to glTransformFeedbackVaryings. Used for
1417 * error reporting and glGetTransformFeedbackVarying().
1419 const char *orig_name
;
1422 * The name of the variable, parsed from orig_name.
1424 const char *var_name
;
1427 * True if the declaration in orig_name represents an array.
1429 bool is_subscripted
;
1432 * If is_subscripted is true, the subscript that was specified in orig_name.
1434 unsigned array_subscript
;
1437 * Which component to extract from the vertex shader output location that
1438 * the linker assigned to this variable. -1 if all components should be
1441 int single_component
;
1444 * The vertex shader output location that the linker assigned for this
1445 * variable. -1 if a location hasn't been assigned yet.
1450 * If location != -1, the number of vector elements in this variable, or 1
1451 * if this variable is a scalar.
1453 unsigned vector_elements
;
1456 * If location != -1, the number of matrix columns in this variable, or 1
1457 * if this variable is not a matrix.
1459 unsigned matrix_columns
;
1461 /** Type of the varying returned by glGetTransformFeedbackVarying() */
1465 * If location != -1, the size that should be returned by
1466 * glGetTransformFeedbackVarying().
1473 * Initialize this object based on a string that was passed to
1474 * glTransformFeedbackVaryings. If there is a parse error, the error is
1475 * reported using linker_error(), and false is returned.
1478 tfeedback_decl::init(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1479 const void *mem_ctx
, const char *input
)
1481 /* We don't have to be pedantic about what is a valid GLSL variable name,
1482 * because any variable with an invalid name can't exist in the IR anyway.
1485 this->location
= -1;
1486 this->orig_name
= input
;
1487 this->single_component
= -1;
1489 const char *bracket
= strrchr(input
, '[');
1492 this->var_name
= ralloc_strndup(mem_ctx
, input
, bracket
- input
);
1493 if (sscanf(bracket
, "[%u]", &this->array_subscript
) != 1) {
1494 linker_error(prog
, "Cannot parse transform feedback varying %s", input
);
1497 this->is_subscripted
= true;
1499 this->var_name
= ralloc_strdup(mem_ctx
, input
);
1500 this->is_subscripted
= false;
1503 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, we need
1504 * to convert a request for gl_ClipDistance[n] into a request for a
1505 * component of gl_ClipDistanceMESA[n/4].
1507 if (ctx
->ShaderCompilerOptions
[MESA_SHADER_VERTEX
].LowerClipDistance
&&
1508 strcmp(this->var_name
, "gl_ClipDistance") == 0) {
1509 this->var_name
= "gl_ClipDistanceMESA";
1510 if (this->is_subscripted
) {
1511 this->single_component
= this->array_subscript
% 4;
1512 this->array_subscript
/= 4;
1521 * Determine whether two tfeedback_decl objects refer to the same variable and
1522 * array index (if applicable).
1525 tfeedback_decl::is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
)
1527 if (strcmp(x
.var_name
, y
.var_name
) != 0)
1529 if (x
.is_subscripted
!= y
.is_subscripted
)
1531 if (x
.is_subscripted
&& x
.array_subscript
!= y
.array_subscript
)
1533 if (x
.single_component
!= y
.single_component
)
1540 * Assign a location for this tfeedback_decl object based on the location
1541 * assignment in output_var.
1543 * If an error occurs, the error is reported through linker_error() and false
1547 tfeedback_decl::assign_location(struct gl_context
*ctx
,
1548 struct gl_shader_program
*prog
,
1549 ir_variable
*output_var
)
1551 if (output_var
->type
->is_array()) {
1552 /* Array variable */
1553 const unsigned matrix_cols
=
1554 output_var
->type
->fields
.array
->matrix_columns
;
1556 if (this->is_subscripted
) {
1557 /* Check array bounds. */
1558 if (this->array_subscript
>=
1559 (unsigned) output_var
->type
->array_size()) {
1560 linker_error(prog
, "Transform feedback varying %s has index "
1561 "%i, but the array size is %i.",
1562 this->orig_name
, this->array_subscript
,
1563 output_var
->type
->array_size());
1567 output_var
->location
+ this->array_subscript
* matrix_cols
;
1570 this->location
= output_var
->location
;
1571 this->size
= (unsigned) output_var
->type
->array_size();
1573 this->vector_elements
= output_var
->type
->fields
.array
->vector_elements
;
1574 this->matrix_columns
= matrix_cols
;
1575 this->type
= output_var
->type
->fields
.array
->gl_type
;
1577 /* Regular variable (scalar, vector, or matrix) */
1578 if (this->is_subscripted
) {
1579 linker_error(prog
, "Transform feedback varying %s found, "
1580 "but it's an array ([] expected).",
1584 this->location
= output_var
->location
;
1586 this->vector_elements
= output_var
->type
->vector_elements
;
1587 this->matrix_columns
= output_var
->type
->matrix_columns
;
1588 this->type
= output_var
->type
->gl_type
;
1590 /* From GL_EXT_transform_feedback:
1591 * A program will fail to link if:
1593 * * the total number of components to capture in any varying
1594 * variable in <varyings> is greater than the constant
1595 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
1596 * buffer mode is SEPARATE_ATTRIBS_EXT;
1598 if (prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
&&
1599 this->num_components() >
1600 ctx
->Const
.MaxTransformFeedbackSeparateComponents
) {
1601 linker_error(prog
, "Transform feedback varying %s exceeds "
1602 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
1612 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
1614 * If an error occurs, the error is reported through linker_error() and false
1618 tfeedback_decl::store(struct gl_shader_program
*prog
,
1619 struct gl_transform_feedback_info
*info
,
1620 unsigned buffer
, unsigned varying
) const
1622 if (!this->is_assigned()) {
1623 /* From GL_EXT_transform_feedback:
1624 * A program will fail to link if:
1626 * * any variable name specified in the <varyings> array is not
1627 * declared as an output in the geometry shader (if present) or
1628 * the vertex shader (if no geometry shader is present);
1630 linker_error(prog
, "Transform feedback varying %s undeclared.",
1634 for (unsigned index
= 0; index
< this->size
; ++index
) {
1635 for (unsigned v
= 0; v
< this->matrix_columns
; ++v
) {
1636 unsigned num_components
=
1637 this->single_component
>= 0 ? 1 : this->vector_elements
;
1638 info
->Outputs
[info
->NumOutputs
].OutputRegister
=
1639 this->location
+ v
+ index
* this->matrix_columns
;
1640 info
->Outputs
[info
->NumOutputs
].NumComponents
= num_components
;
1641 info
->Outputs
[info
->NumOutputs
].OutputBuffer
= buffer
;
1642 info
->Outputs
[info
->NumOutputs
].DstOffset
= info
->BufferStride
[buffer
];
1643 info
->Outputs
[info
->NumOutputs
].ComponentOffset
=
1644 this->single_component
>= 0 ? this->single_component
: 0;
1646 info
->BufferStride
[buffer
] += num_components
;
1650 info
->Varyings
[varying
].Name
= ralloc_strdup(prog
, this->orig_name
);
1651 info
->Varyings
[varying
].Type
= this->type
;
1652 info
->Varyings
[varying
].Size
= this->size
;
1660 * Parse all the transform feedback declarations that were passed to
1661 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
1663 * If an error occurs, the error is reported through linker_error() and false
1667 parse_tfeedback_decls(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1668 const void *mem_ctx
, unsigned num_names
,
1669 char **varying_names
, tfeedback_decl
*decls
)
1671 for (unsigned i
= 0; i
< num_names
; ++i
) {
1672 if (!decls
[i
].init(ctx
, prog
, mem_ctx
, varying_names
[i
]))
1674 /* From GL_EXT_transform_feedback:
1675 * A program will fail to link if:
1677 * * any two entries in the <varyings> array specify the same varying
1680 * We interpret this to mean "any two entries in the <varyings> array
1681 * specify the same varying variable and array index", since transform
1682 * feedback of arrays would be useless otherwise.
1684 for (unsigned j
= 0; j
< i
; ++j
) {
1685 if (tfeedback_decl::is_same(decls
[i
], decls
[j
])) {
1686 linker_error(prog
, "Transform feedback varying %s specified "
1687 "more than once.", varying_names
[i
]);
1697 * Assign a location for a variable that is produced in one pipeline stage
1698 * (the "producer") and consumed in the next stage (the "consumer").
1700 * \param input_var is the input variable declaration in the consumer.
1702 * \param output_var is the output variable declaration in the producer.
1704 * \param input_index is the counter that keeps track of assigned input
1705 * locations in the consumer.
1707 * \param output_index is the counter that keeps track of assigned output
1708 * locations in the producer.
1710 * It is permissible for \c input_var to be NULL (this happens if a variable
1711 * is output by the producer and consumed by transform feedback, but not
1712 * consumed by the consumer).
1714 * If the variable has already been assigned a location, this function has no
1718 assign_varying_location(ir_variable
*input_var
, ir_variable
*output_var
,
1719 unsigned *input_index
, unsigned *output_index
)
1721 if (output_var
->location
!= -1) {
1722 /* Location already assigned. */
1727 assert(input_var
->location
== -1);
1728 input_var
->location
= *input_index
;
1731 output_var
->location
= *output_index
;
1733 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1734 assert(!output_var
->type
->is_record());
1736 if (output_var
->type
->is_array()) {
1737 const unsigned slots
= output_var
->type
->length
1738 * output_var
->type
->fields
.array
->matrix_columns
;
1740 *output_index
+= slots
;
1741 *input_index
+= slots
;
1743 const unsigned slots
= output_var
->type
->matrix_columns
;
1745 *output_index
+= slots
;
1746 *input_index
+= slots
;
1752 * Assign locations for all variables that are produced in one pipeline stage
1753 * (the "producer") and consumed in the next stage (the "consumer").
1755 * Variables produced by the producer may also be consumed by transform
1758 * \param num_tfeedback_decls is the number of declarations indicating
1759 * variables that may be consumed by transform feedback.
1761 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
1762 * representing the result of parsing the strings passed to
1763 * glTransformFeedbackVaryings(). assign_location() will be called for
1764 * each of these objects that matches one of the outputs of the
1767 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
1768 * be NULL. In this case, varying locations are assigned solely based on the
1769 * requirements of transform feedback.
1772 assign_varying_locations(struct gl_context
*ctx
,
1773 struct gl_shader_program
*prog
,
1774 gl_shader
*producer
, gl_shader
*consumer
,
1775 unsigned num_tfeedback_decls
,
1776 tfeedback_decl
*tfeedback_decls
)
1778 /* FINISHME: Set dynamically when geometry shader support is added. */
1779 unsigned output_index
= VERT_RESULT_VAR0
;
1780 unsigned input_index
= FRAG_ATTRIB_VAR0
;
1782 /* Operate in a total of three passes.
1784 * 1. Assign locations for any matching inputs and outputs.
1786 * 2. Mark output variables in the producer that do not have locations as
1787 * not being outputs. This lets the optimizer eliminate them.
1789 * 3. Mark input variables in the consumer that do not have locations as
1790 * not being inputs. This lets the optimizer eliminate them.
1793 link_invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
1795 link_invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
1797 foreach_list(node
, producer
->ir
) {
1798 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
1800 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
))
1803 ir_variable
*input_var
=
1804 consumer
? consumer
->symbols
->get_variable(output_var
->name
) : NULL
;
1806 if (input_var
&& input_var
->mode
!= ir_var_in
)
1810 assign_varying_location(input_var
, output_var
, &input_index
,
1814 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
1815 if (!tfeedback_decls
[i
].is_assigned() &&
1816 tfeedback_decls
[i
].matches_var(output_var
)) {
1817 if (output_var
->location
== -1) {
1818 assign_varying_location(input_var
, output_var
, &input_index
,
1821 if (!tfeedback_decls
[i
].assign_location(ctx
, prog
, output_var
))
1827 unsigned varying_vectors
= 0;
1830 foreach_list(node
, consumer
->ir
) {
1831 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1833 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
1836 if (var
->location
== -1) {
1837 if (prog
->Version
<= 120) {
1838 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1840 * Only those varying variables used (i.e. read) in
1841 * the fragment shader executable must be written to
1842 * by the vertex shader executable; declaring
1843 * superfluous varying variables in a vertex shader is
1846 * We interpret this text as meaning that the VS must
1847 * write the variable for the FS to read it. See
1848 * "glsl1-varying read but not written" in piglit.
1851 linker_error(prog
, "fragment shader varying %s not written "
1852 "by vertex shader\n.", var
->name
);
1855 /* An 'in' variable is only really a shader input if its
1856 * value is written by the previous stage.
1858 var
->mode
= ir_var_auto
;
1860 /* The packing rules are used for vertex shader inputs are also
1861 * used for fragment shader inputs.
1863 varying_vectors
+= count_attribute_slots(var
->type
);
1868 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
1869 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
1870 if (ctx
->Const
.GLSLSkipStrictMaxVaryingLimitCheck
) {
1871 linker_warning(prog
, "shader uses too many varying vectors "
1872 "(%u > %u), but the driver will try to optimize "
1873 "them out; this is non-portable out-of-spec "
1875 varying_vectors
, ctx
->Const
.MaxVarying
);
1877 linker_error(prog
, "shader uses too many varying vectors "
1879 varying_vectors
, ctx
->Const
.MaxVarying
);
1884 const unsigned float_components
= varying_vectors
* 4;
1885 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
1886 if (ctx
->Const
.GLSLSkipStrictMaxVaryingLimitCheck
) {
1887 linker_warning(prog
, "shader uses too many varying components "
1888 "(%u > %u), but the driver will try to optimize "
1889 "them out; this is non-portable out-of-spec "
1891 float_components
, ctx
->Const
.MaxVarying
* 4);
1893 linker_error(prog
, "shader uses too many varying components "
1895 float_components
, ctx
->Const
.MaxVarying
* 4);
1906 * Store transform feedback location assignments into
1907 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
1909 * If an error occurs, the error is reported through linker_error() and false
1913 store_tfeedback_info(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1914 unsigned num_tfeedback_decls
,
1915 tfeedback_decl
*tfeedback_decls
)
1917 unsigned total_tfeedback_components
= 0;
1918 bool separate_attribs_mode
=
1919 prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
;
1921 ralloc_free(prog
->LinkedTransformFeedback
.Varyings
);
1923 memset(&prog
->LinkedTransformFeedback
, 0,
1924 sizeof(prog
->LinkedTransformFeedback
));
1926 prog
->LinkedTransformFeedback
.NumBuffers
=
1927 separate_attribs_mode
? num_tfeedback_decls
: 1;
1929 prog
->LinkedTransformFeedback
.Varyings
=
1930 rzalloc_array(prog
->LinkedTransformFeedback
.Varyings
,
1931 struct gl_transform_feedback_varying_info
,
1932 num_tfeedback_decls
);
1934 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
1935 unsigned buffer
= separate_attribs_mode
? i
: 0;
1936 if (!tfeedback_decls
[i
].store(prog
, &prog
->LinkedTransformFeedback
,
1939 total_tfeedback_components
+= tfeedback_decls
[i
].num_components();
1942 /* From GL_EXT_transform_feedback:
1943 * A program will fail to link if:
1945 * * the total number of components to capture is greater than
1946 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
1947 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
1949 if (prog
->TransformFeedback
.BufferMode
== GL_INTERLEAVED_ATTRIBS
&&
1950 total_tfeedback_components
>
1951 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1952 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1953 "limit has been exceeded.");
1961 * Store the gl_FragDepth layout in the gl_shader_program struct.
1964 store_fragdepth_layout(struct gl_shader_program
*prog
)
1966 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
1970 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
1972 /* We don't look up the gl_FragDepth symbol directly because if
1973 * gl_FragDepth is not used in the shader, it's removed from the IR.
1974 * However, the symbol won't be removed from the symbol table.
1976 * We're only interested in the cases where the variable is NOT removed
1979 foreach_list(node
, ir
) {
1980 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1982 if (var
== NULL
|| var
->mode
!= ir_var_out
) {
1986 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1987 switch (var
->depth_layout
) {
1988 case ir_depth_layout_none
:
1989 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
1991 case ir_depth_layout_any
:
1992 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
1994 case ir_depth_layout_greater
:
1995 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
1997 case ir_depth_layout_less
:
1998 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2000 case ir_depth_layout_unchanged
:
2001 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2012 * Validate the resources used by a program versus the implementation limits
2015 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2017 static const char *const shader_names
[MESA_SHADER_TYPES
] = {
2018 "vertex", "fragment", "geometry"
2021 const unsigned max_samplers
[MESA_SHADER_TYPES
] = {
2022 ctx
->Const
.MaxVertexTextureImageUnits
,
2023 ctx
->Const
.MaxTextureImageUnits
,
2024 ctx
->Const
.MaxGeometryTextureImageUnits
2027 const unsigned max_uniform_components
[MESA_SHADER_TYPES
] = {
2028 ctx
->Const
.VertexProgram
.MaxUniformComponents
,
2029 ctx
->Const
.FragmentProgram
.MaxUniformComponents
,
2030 0 /* FINISHME: Geometry shaders. */
2033 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2034 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2039 if (sh
->num_samplers
> max_samplers
[i
]) {
2040 linker_error(prog
, "Too many %s shader texture samplers",
2044 if (sh
->num_uniform_components
> max_uniform_components
[i
]) {
2045 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2046 linker_warning(prog
, "Too many %s shader uniform components, "
2047 "but the driver will try to optimize them out; "
2048 "this is non-portable out-of-spec behavior\n",
2051 linker_error(prog
, "Too many %s shader uniform components",
2057 return prog
->LinkStatus
;
2061 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2063 tfeedback_decl
*tfeedback_decls
= NULL
;
2064 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2066 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2068 prog
->LinkStatus
= false;
2069 prog
->Validated
= false;
2070 prog
->_Used
= false;
2072 if (prog
->InfoLog
!= NULL
)
2073 ralloc_free(prog
->InfoLog
);
2075 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2077 /* Separate the shaders into groups based on their type.
2079 struct gl_shader
**vert_shader_list
;
2080 unsigned num_vert_shaders
= 0;
2081 struct gl_shader
**frag_shader_list
;
2082 unsigned num_frag_shaders
= 0;
2084 vert_shader_list
= (struct gl_shader
**)
2085 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
2086 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
2088 unsigned min_version
= UINT_MAX
;
2089 unsigned max_version
= 0;
2090 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2091 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2092 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2094 switch (prog
->Shaders
[i
]->Type
) {
2095 case GL_VERTEX_SHADER
:
2096 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
2099 case GL_FRAGMENT_SHADER
:
2100 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
2103 case GL_GEOMETRY_SHADER
:
2104 /* FINISHME: Support geometry shaders. */
2105 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
2110 /* Previous to GLSL version 1.30, different compilation units could mix and
2111 * match shading language versions. With GLSL 1.30 and later, the versions
2112 * of all shaders must match.
2114 assert(min_version
>= 100);
2115 assert(max_version
<= 130);
2116 if ((max_version
>= 130 || min_version
== 100)
2117 && min_version
!= max_version
) {
2118 linker_error(prog
, "all shaders must use same shading "
2119 "language version\n");
2123 prog
->Version
= max_version
;
2125 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2126 if (prog
->_LinkedShaders
[i
] != NULL
)
2127 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2129 prog
->_LinkedShaders
[i
] = NULL
;
2132 /* Link all shaders for a particular stage and validate the result.
2134 if (num_vert_shaders
> 0) {
2135 gl_shader
*const sh
=
2136 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
2142 if (!validate_vertex_shader_executable(prog
, sh
))
2145 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2149 if (num_frag_shaders
> 0) {
2150 gl_shader
*const sh
=
2151 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
2157 if (!validate_fragment_shader_executable(prog
, sh
))
2160 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
2164 /* Here begins the inter-stage linking phase. Some initial validation is
2165 * performed, then locations are assigned for uniforms, attributes, and
2168 if (cross_validate_uniforms(prog
)) {
2171 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
2172 if (prog
->_LinkedShaders
[prev
] != NULL
)
2176 /* Validate the inputs of each stage with the output of the preceding
2179 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
2180 if (prog
->_LinkedShaders
[i
] == NULL
)
2183 if (!cross_validate_outputs_to_inputs(prog
,
2184 prog
->_LinkedShaders
[prev
],
2185 prog
->_LinkedShaders
[i
]))
2191 prog
->LinkStatus
= true;
2194 /* Do common optimization before assigning storage for attributes,
2195 * uniforms, and varyings. Later optimization could possibly make
2196 * some of that unused.
2198 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2199 if (prog
->_LinkedShaders
[i
] == NULL
)
2202 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2203 if (!prog
->LinkStatus
)
2206 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
)
2207 lower_clip_distance(prog
->_LinkedShaders
[i
]->ir
);
2209 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, 32))
2213 /* FINISHME: The value of the max_attribute_index parameter is
2214 * FINISHME: implementation dependent based on the value of
2215 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2216 * FINISHME: at least 16, so hardcode 16 for now.
2218 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2222 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, ctx
->Const
.MaxDrawBuffers
)) {
2227 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
2228 if (prog
->_LinkedShaders
[prev
] != NULL
)
2232 if (num_tfeedback_decls
!= 0) {
2233 /* From GL_EXT_transform_feedback:
2234 * A program will fail to link if:
2236 * * the <count> specified by TransformFeedbackVaryingsEXT is
2237 * non-zero, but the program object has no vertex or geometry
2240 if (prev
>= MESA_SHADER_FRAGMENT
) {
2241 linker_error(prog
, "Transform feedback varyings specified, but "
2242 "no vertex or geometry shader is present.");
2246 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2247 prog
->TransformFeedback
.NumVarying
);
2248 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2249 prog
->TransformFeedback
.VaryingNames
,
2254 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
2255 if (prog
->_LinkedShaders
[i
] == NULL
)
2258 if (!assign_varying_locations(
2259 ctx
, prog
, prog
->_LinkedShaders
[prev
], prog
->_LinkedShaders
[i
],
2260 i
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2267 if (prev
!= MESA_SHADER_FRAGMENT
&& num_tfeedback_decls
!= 0) {
2268 /* There was no fragment shader, but we still have to assign varying
2269 * locations for use by transform feedback.
2271 if (!assign_varying_locations(
2272 ctx
, prog
, prog
->_LinkedShaders
[prev
], NULL
, num_tfeedback_decls
,
2277 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2280 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
2281 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2284 /* Eliminate code that is now dead due to unused vertex outputs being
2287 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
, false))
2291 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
2292 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
2294 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
2295 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
2296 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
2298 /* Eliminate code that is now dead due to unused geometry outputs being
2301 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
, false))
2305 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
2306 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2308 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
2310 /* Eliminate code that is now dead due to unused fragment inputs being
2311 * demoted. This shouldn't actually do anything other than remove
2312 * declarations of the (now unused) global variables.
2314 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
, false))
2318 update_array_sizes(prog
);
2319 link_assign_uniform_locations(prog
);
2320 store_fragdepth_layout(prog
);
2322 if (!check_resources(ctx
, prog
))
2325 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2326 * present in a linked program. By checking for use of shading language
2327 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
2329 if (!prog
->InternalSeparateShader
&&
2330 (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100)) {
2331 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2332 linker_error(prog
, "program lacks a vertex shader\n");
2333 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2334 linker_error(prog
, "program lacks a fragment shader\n");
2338 /* FINISHME: Assign fragment shader output locations. */
2341 free(vert_shader_list
);
2343 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2344 if (prog
->_LinkedShaders
[i
] == NULL
)
2347 /* Retain any live IR, but trash the rest. */
2348 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2350 /* The symbol table in the linked shaders may contain references to
2351 * variables that were removed (e.g., unused uniforms). Since it may
2352 * contain junk, there is no possible valid use. Delete it and set the
2355 delete prog
->_LinkedShaders
[i
]->symbols
;
2356 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2359 ralloc_free(mem_ctx
);