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
->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 if (ir
->return_deref
!= NULL
) {
121 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
123 if (strcmp(name
, var
->name
) == 0) {
129 return visit_continue_with_parent
;
132 bool variable_found()
138 const char *name
; /**< Find writes to a variable with this name. */
139 bool found
; /**< Was a write to the variable found? */
144 * Visitor that determines whether or not a variable is ever read.
146 class find_deref_visitor
: public ir_hierarchical_visitor
{
148 find_deref_visitor(const char *name
)
149 : name(name
), found(false)
154 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
156 if (strcmp(this->name
, ir
->var
->name
) == 0) {
161 return visit_continue
;
164 bool variable_found() const
170 const char *name
; /**< Find writes to a variable with this name. */
171 bool found
; /**< Was a write to the variable found? */
176 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
180 ralloc_strcat(&prog
->InfoLog
, "error: ");
182 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
185 prog
->LinkStatus
= false;
190 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
194 ralloc_strcat(&prog
->InfoLog
, "error: ");
196 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
203 link_invalidate_variable_locations(gl_shader
*sh
, enum ir_variable_mode mode
,
206 foreach_list(node
, sh
->ir
) {
207 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
209 if ((var
== NULL
) || (var
->mode
!= (unsigned) mode
))
212 /* Only assign locations for generic attributes / varyings / etc.
214 if ((var
->location
>= generic_base
) && !var
->explicit_location
)
221 * Determine the number of attribute slots required for a particular type
223 * This code is here because it implements the language rules of a specific
224 * GLSL version. Since it's a property of the language and not a property of
225 * types in general, it doesn't really belong in glsl_type.
228 count_attribute_slots(const glsl_type
*t
)
230 /* From page 31 (page 37 of the PDF) of the GLSL 1.50 spec:
232 * "A scalar input counts the same amount against this limit as a vec4,
233 * so applications may want to consider packing groups of four
234 * unrelated float inputs together into a vector to better utilize the
235 * capabilities of the underlying hardware. A matrix input will use up
236 * multiple locations. The number of locations used will equal the
237 * number of columns in the matrix."
239 * The spec does not explicitly say how arrays are counted. However, it
240 * should be safe to assume the total number of slots consumed by an array
241 * is the number of entries in the array multiplied by the number of slots
242 * consumed by a single element of the array.
246 return t
->array_size() * count_attribute_slots(t
->element_type());
249 return t
->matrix_columns
;
256 * Verify that a vertex shader executable meets all semantic requirements.
258 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
261 * \param shader Vertex shader executable to be verified
264 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
265 struct gl_shader
*shader
)
270 /* From the GLSL 1.10 spec, page 48:
272 * "The variable gl_Position is available only in the vertex
273 * language and is intended for writing the homogeneous vertex
274 * position. All executions of a well-formed vertex shader
275 * executable must write a value into this variable. [...] The
276 * variable gl_Position is available only in the vertex
277 * language and is intended for writing the homogeneous vertex
278 * position. All executions of a well-formed vertex shader
279 * executable must write a value into this variable."
281 * while in GLSL 1.40 this text is changed to:
283 * "The variable gl_Position is available only in the vertex
284 * language and is intended for writing the homogeneous vertex
285 * position. It can be written at any time during shader
286 * execution. It may also be read back by a vertex shader
287 * after being written. This value will be used by primitive
288 * assembly, clipping, culling, and other fixed functionality
289 * operations, if present, that operate on primitives after
290 * vertex processing has occurred. Its value is undefined if
291 * the vertex shader executable does not write gl_Position."
293 if (prog
->Version
< 140) {
294 find_assignment_visitor
find("gl_Position");
295 find
.run(shader
->ir
);
296 if (!find
.variable_found()) {
297 linker_error(prog
, "vertex shader does not write to `gl_Position'\n");
302 prog
->Vert
.ClipDistanceArraySize
= 0;
304 if (prog
->Version
>= 130) {
305 /* From section 7.1 (Vertex Shader Special Variables) of the
308 * "It is an error for a shader to statically write both
309 * gl_ClipVertex and gl_ClipDistance."
311 find_assignment_visitor
clip_vertex("gl_ClipVertex");
312 find_assignment_visitor
clip_distance("gl_ClipDistance");
314 clip_vertex
.run(shader
->ir
);
315 clip_distance
.run(shader
->ir
);
316 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
317 linker_error(prog
, "vertex shader writes to both `gl_ClipVertex' "
318 "and `gl_ClipDistance'\n");
321 prog
->Vert
.UsesClipDistance
= clip_distance
.variable_found();
322 ir_variable
*clip_distance_var
=
323 shader
->symbols
->get_variable("gl_ClipDistance");
324 if (clip_distance_var
)
325 prog
->Vert
.ClipDistanceArraySize
= clip_distance_var
->type
->length
;
333 * Verify that a fragment shader executable meets all semantic requirements
335 * \param shader Fragment shader executable to be verified
338 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
339 struct gl_shader
*shader
)
344 find_assignment_visitor
frag_color("gl_FragColor");
345 find_assignment_visitor
frag_data("gl_FragData");
347 frag_color
.run(shader
->ir
);
348 frag_data
.run(shader
->ir
);
350 if (frag_color
.variable_found() && frag_data
.variable_found()) {
351 linker_error(prog
, "fragment shader writes to both "
352 "`gl_FragColor' and `gl_FragData'\n");
361 * Generate a string describing the mode of a variable
364 mode_string(const ir_variable
*var
)
368 return (var
->read_only
) ? "global constant" : "global variable";
370 case ir_var_uniform
: return "uniform";
371 case ir_var_in
: return "shader input";
372 case ir_var_out
: return "shader output";
373 case ir_var_inout
: return "shader inout";
375 case ir_var_const_in
:
376 case ir_var_temporary
:
378 assert(!"Should not get here.");
379 return "invalid variable";
385 * Perform validation of global variables used across multiple shaders
388 cross_validate_globals(struct gl_shader_program
*prog
,
389 struct gl_shader
**shader_list
,
390 unsigned num_shaders
,
393 /* Examine all of the uniforms in all of the shaders and cross validate
396 glsl_symbol_table variables
;
397 for (unsigned i
= 0; i
< num_shaders
; i
++) {
398 if (shader_list
[i
] == NULL
)
401 foreach_list(node
, shader_list
[i
]->ir
) {
402 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
407 if (uniforms_only
&& (var
->mode
!= ir_var_uniform
))
410 /* Don't cross validate temporaries that are at global scope. These
411 * will eventually get pulled into the shaders 'main'.
413 if (var
->mode
== ir_var_temporary
)
416 /* If a global with this name has already been seen, verify that the
417 * new instance has the same type. In addition, if the globals have
418 * initializers, the values of the initializers must be the same.
420 ir_variable
*const existing
= variables
.get_variable(var
->name
);
421 if (existing
!= NULL
) {
422 if (var
->type
!= existing
->type
) {
423 /* Consider the types to be "the same" if both types are arrays
424 * of the same type and one of the arrays is implicitly sized.
425 * In addition, set the type of the linked variable to the
426 * explicitly sized array.
428 if (var
->type
->is_array()
429 && existing
->type
->is_array()
430 && (var
->type
->fields
.array
== existing
->type
->fields
.array
)
431 && ((var
->type
->length
== 0)
432 || (existing
->type
->length
== 0))) {
433 if (var
->type
->length
!= 0) {
434 existing
->type
= var
->type
;
437 linker_error(prog
, "%s `%s' declared as type "
438 "`%s' and type `%s'\n",
440 var
->name
, var
->type
->name
,
441 existing
->type
->name
);
446 if (var
->explicit_location
) {
447 if (existing
->explicit_location
448 && (var
->location
!= existing
->location
)) {
449 linker_error(prog
, "explicit locations for %s "
450 "`%s' have differing values\n",
451 mode_string(var
), var
->name
);
455 existing
->location
= var
->location
;
456 existing
->explicit_location
= true;
459 /* Validate layout qualifiers for gl_FragDepth.
461 * From the AMD/ARB_conservative_depth specs:
463 * "If gl_FragDepth is redeclared in any fragment shader in a
464 * program, it must be redeclared in all fragment shaders in
465 * that program that have static assignments to
466 * gl_FragDepth. All redeclarations of gl_FragDepth in all
467 * fragment shaders in a single program must have the same set
470 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
471 bool layout_declared
= var
->depth_layout
!= ir_depth_layout_none
;
472 bool layout_differs
=
473 var
->depth_layout
!= existing
->depth_layout
;
475 if (layout_declared
&& layout_differs
) {
477 "All redeclarations of gl_FragDepth in all "
478 "fragment shaders in a single program must have "
479 "the same set of qualifiers.");
482 if (var
->used
&& layout_differs
) {
484 "If gl_FragDepth is redeclared with a layout "
485 "qualifier in any fragment shader, it must be "
486 "redeclared with the same layout qualifier in "
487 "all fragment shaders that have assignments to "
492 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
494 * "If a shared global has multiple initializers, the
495 * initializers must all be constant expressions, and they
496 * must all have the same value. Otherwise, a link error will
497 * result. (A shared global having only one initializer does
498 * not require that initializer to be a constant expression.)"
500 * Previous to 4.20 the GLSL spec simply said that initializers
501 * must have the same value. In this case of non-constant
502 * initializers, this was impossible to determine. As a result,
503 * no vendor actually implemented that behavior. The 4.20
504 * behavior matches the implemented behavior of at least one other
505 * vendor, so we'll implement that for all GLSL versions.
507 if (var
->constant_initializer
!= NULL
) {
508 if (existing
->constant_initializer
!= NULL
) {
509 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
510 linker_error(prog
, "initializers for %s "
511 "`%s' have differing values\n",
512 mode_string(var
), var
->name
);
516 /* If the first-seen instance of a particular uniform did not
517 * have an initializer but a later instance does, copy the
518 * initializer to the version stored in the symbol table.
520 /* FINISHME: This is wrong. The constant_value field should
521 * FINISHME: not be modified! Imagine a case where a shader
522 * FINISHME: without an initializer is linked in two different
523 * FINISHME: programs with shaders that have differing
524 * FINISHME: initializers. Linking with the first will
525 * FINISHME: modify the shader, and linking with the second
526 * FINISHME: will fail.
528 existing
->constant_initializer
=
529 var
->constant_initializer
->clone(ralloc_parent(existing
),
534 if (var
->has_initializer
) {
535 if (existing
->has_initializer
536 && (var
->constant_initializer
== NULL
537 || existing
->constant_initializer
== NULL
)) {
539 "shared global variable `%s' has multiple "
540 "non-constant initializers.\n",
545 /* Some instance had an initializer, so keep track of that. In
546 * this location, all sorts of initializers (constant or
547 * otherwise) will propagate the existence to the variable
548 * stored in the symbol table.
550 existing
->has_initializer
= true;
553 if (existing
->invariant
!= var
->invariant
) {
554 linker_error(prog
, "declarations for %s `%s' have "
555 "mismatching invariant qualifiers\n",
556 mode_string(var
), var
->name
);
559 if (existing
->centroid
!= var
->centroid
) {
560 linker_error(prog
, "declarations for %s `%s' have "
561 "mismatching centroid qualifiers\n",
562 mode_string(var
), var
->name
);
566 variables
.add_variable(var
);
575 * Perform validation of uniforms used across multiple shader stages
578 cross_validate_uniforms(struct gl_shader_program
*prog
)
580 return cross_validate_globals(prog
, prog
->_LinkedShaders
,
581 MESA_SHADER_TYPES
, true);
585 * Accumulates the array of prog->UniformBlocks and checks that all
586 * definitons of blocks agree on their contents.
589 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
591 unsigned max_num_uniform_blocks
= 0;
592 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
593 if (prog
->_LinkedShaders
[i
])
594 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
597 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
598 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
600 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
601 max_num_uniform_blocks
);
602 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
603 prog
->UniformBlockStageIndex
[i
][j
] = -1;
608 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
609 int index
= link_cross_validate_uniform_block(prog
,
610 &prog
->UniformBlocks
,
611 &prog
->NumUniformBlocks
,
612 &sh
->UniformBlocks
[j
]);
615 linker_error(prog
, "uniform block `%s' has mismatching definitions",
616 sh
->UniformBlocks
[j
].Name
);
620 prog
->UniformBlockStageIndex
[i
][index
] = j
;
628 * Validate that outputs from one stage match inputs of another
631 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
632 gl_shader
*producer
, gl_shader
*consumer
)
634 glsl_symbol_table parameters
;
635 /* FINISHME: Figure these out dynamically. */
636 const char *const producer_stage
= "vertex";
637 const char *const consumer_stage
= "fragment";
639 /* Find all shader outputs in the "producer" stage.
641 foreach_list(node
, producer
->ir
) {
642 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
644 /* FINISHME: For geometry shaders, this should also look for inout
645 * FINISHME: variables.
647 if ((var
== NULL
) || (var
->mode
!= ir_var_out
))
650 parameters
.add_variable(var
);
654 /* Find all shader inputs in the "consumer" stage. Any variables that have
655 * matching outputs already in the symbol table must have the same type and
658 foreach_list(node
, consumer
->ir
) {
659 ir_variable
*const input
= ((ir_instruction
*) node
)->as_variable();
661 /* FINISHME: For geometry shaders, this should also look for inout
662 * FINISHME: variables.
664 if ((input
== NULL
) || (input
->mode
!= ir_var_in
))
667 ir_variable
*const output
= parameters
.get_variable(input
->name
);
668 if (output
!= NULL
) {
669 /* Check that the types match between stages.
671 if (input
->type
!= output
->type
) {
672 /* There is a bit of a special case for gl_TexCoord. This
673 * built-in is unsized by default. Applications that variable
674 * access it must redeclare it with a size. There is some
675 * language in the GLSL spec that implies the fragment shader
676 * and vertex shader do not have to agree on this size. Other
677 * driver behave this way, and one or two applications seem to
680 * Neither declaration needs to be modified here because the array
681 * sizes are fixed later when update_array_sizes is called.
683 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
685 * "Unlike user-defined varying variables, the built-in
686 * varying variables don't have a strict one-to-one
687 * correspondence between the vertex language and the
688 * fragment language."
690 if (!output
->type
->is_array()
691 || (strncmp("gl_", output
->name
, 3) != 0)) {
693 "%s shader output `%s' declared as type `%s', "
694 "but %s shader input declared as type `%s'\n",
695 producer_stage
, output
->name
,
697 consumer_stage
, input
->type
->name
);
702 /* Check that all of the qualifiers match between stages.
704 if (input
->centroid
!= output
->centroid
) {
706 "%s shader output `%s' %s centroid qualifier, "
707 "but %s shader input %s centroid qualifier\n",
710 (output
->centroid
) ? "has" : "lacks",
712 (input
->centroid
) ? "has" : "lacks");
716 if (input
->invariant
!= output
->invariant
) {
718 "%s shader output `%s' %s invariant qualifier, "
719 "but %s shader input %s invariant qualifier\n",
722 (output
->invariant
) ? "has" : "lacks",
724 (input
->invariant
) ? "has" : "lacks");
728 if (input
->interpolation
!= output
->interpolation
) {
730 "%s shader output `%s' specifies %s "
731 "interpolation qualifier, "
732 "but %s shader input specifies %s "
733 "interpolation qualifier\n",
736 output
->interpolation_string(),
738 input
->interpolation_string());
749 * Populates a shaders symbol table with all global declarations
752 populate_symbol_table(gl_shader
*sh
)
754 sh
->symbols
= new(sh
) glsl_symbol_table
;
756 foreach_list(node
, sh
->ir
) {
757 ir_instruction
*const inst
= (ir_instruction
*) node
;
761 if ((func
= inst
->as_function()) != NULL
) {
762 sh
->symbols
->add_function(func
);
763 } else if ((var
= inst
->as_variable()) != NULL
) {
764 sh
->symbols
->add_variable(var
);
771 * Remap variables referenced in an instruction tree
773 * This is used when instruction trees are cloned from one shader and placed in
774 * another. These trees will contain references to \c ir_variable nodes that
775 * do not exist in the target shader. This function finds these \c ir_variable
776 * references and replaces the references with matching variables in the target
779 * If there is no matching variable in the target shader, a clone of the
780 * \c ir_variable is made and added to the target shader. The new variable is
781 * added to \b both the instruction stream and the symbol table.
783 * \param inst IR tree that is to be processed.
784 * \param symbols Symbol table containing global scope symbols in the
786 * \param instructions Instruction stream where new variable declarations
790 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
793 class remap_visitor
: public ir_hierarchical_visitor
{
795 remap_visitor(struct gl_shader
*target
,
798 this->target
= target
;
799 this->symbols
= target
->symbols
;
800 this->instructions
= target
->ir
;
804 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
806 if (ir
->var
->mode
== ir_var_temporary
) {
807 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
811 return visit_continue
;
814 ir_variable
*const existing
=
815 this->symbols
->get_variable(ir
->var
->name
);
816 if (existing
!= NULL
)
819 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
821 this->symbols
->add_variable(copy
);
822 this->instructions
->push_head(copy
);
826 return visit_continue
;
830 struct gl_shader
*target
;
831 glsl_symbol_table
*symbols
;
832 exec_list
*instructions
;
836 remap_visitor
v(target
, temps
);
843 * Move non-declarations from one instruction stream to another
845 * The intended usage pattern of this function is to pass the pointer to the
846 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
847 * pointer) for \c last and \c false for \c make_copies on the first
848 * call. Successive calls pass the return value of the previous call for
849 * \c last and \c true for \c make_copies.
851 * \param instructions Source instruction stream
852 * \param last Instruction after which new instructions should be
853 * inserted in the target instruction stream
854 * \param make_copies Flag selecting whether instructions in \c instructions
855 * should be copied (via \c ir_instruction::clone) into the
856 * target list or moved.
859 * The new "last" instruction in the target instruction stream. This pointer
860 * is suitable for use as the \c last parameter of a later call to this
864 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
865 bool make_copies
, gl_shader
*target
)
867 hash_table
*temps
= NULL
;
870 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
871 hash_table_pointer_compare
);
873 foreach_list_safe(node
, instructions
) {
874 ir_instruction
*inst
= (ir_instruction
*) node
;
876 if (inst
->as_function())
879 ir_variable
*var
= inst
->as_variable();
880 if ((var
!= NULL
) && (var
->mode
!= ir_var_temporary
))
883 assert(inst
->as_assignment()
885 || ((var
!= NULL
) && (var
->mode
== ir_var_temporary
)));
888 inst
= inst
->clone(target
, NULL
);
891 hash_table_insert(temps
, inst
, var
);
893 remap_variables(inst
, target
, temps
);
898 last
->insert_after(inst
);
903 hash_table_dtor(temps
);
909 * Get the function signature for main from a shader
911 static ir_function_signature
*
912 get_main_function_signature(gl_shader
*sh
)
914 ir_function
*const f
= sh
->symbols
->get_function("main");
916 exec_list void_parameters
;
918 /* Look for the 'void main()' signature and ensure that it's defined.
919 * This keeps the linker from accidentally pick a shader that just
920 * contains a prototype for main.
922 * We don't have to check for multiple definitions of main (in multiple
923 * shaders) because that would have already been caught above.
925 ir_function_signature
*sig
= f
->matching_signature(&void_parameters
);
926 if ((sig
!= NULL
) && sig
->is_defined
) {
936 * This class is only used in link_intrastage_shaders() below but declaring
937 * it inside that function leads to compiler warnings with some versions of
940 class array_sizing_visitor
: public ir_hierarchical_visitor
{
942 virtual ir_visitor_status
visit(ir_variable
*var
)
944 if (var
->type
->is_array() && (var
->type
->length
== 0)) {
945 const glsl_type
*type
=
946 glsl_type::get_array_instance(var
->type
->fields
.array
,
947 var
->max_array_access
+ 1);
948 assert(type
!= NULL
);
951 return visit_continue
;
956 * Combine a group of shaders for a single stage to generate a linked shader
959 * If this function is supplied a single shader, it is cloned, and the new
960 * shader is returned.
962 static struct gl_shader
*
963 link_intrastage_shaders(void *mem_ctx
,
964 struct gl_context
*ctx
,
965 struct gl_shader_program
*prog
,
966 struct gl_shader
**shader_list
,
967 unsigned num_shaders
)
969 struct gl_uniform_block
*uniform_blocks
= NULL
;
970 unsigned num_uniform_blocks
= 0;
972 /* Check that global variables defined in multiple shaders are consistent.
974 if (!cross_validate_globals(prog
, shader_list
, num_shaders
, false))
977 /* Check that uniform blocks between shaders for a stage agree. */
978 for (unsigned i
= 0; i
< num_shaders
; i
++) {
979 struct gl_shader
*sh
= shader_list
[i
];
981 for (unsigned j
= 0; j
< shader_list
[i
]->NumUniformBlocks
; j
++) {
982 link_assign_uniform_block_offsets(shader_list
[i
]);
984 int index
= link_cross_validate_uniform_block(mem_ctx
,
987 &sh
->UniformBlocks
[j
]);
989 linker_error(prog
, "uniform block `%s' has mismatching definitions",
990 sh
->UniformBlocks
[j
].Name
);
996 /* Check that there is only a single definition of each function signature
997 * across all shaders.
999 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1000 foreach_list(node
, shader_list
[i
]->ir
) {
1001 ir_function
*const f
= ((ir_instruction
*) node
)->as_function();
1006 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1007 ir_function
*const other
=
1008 shader_list
[j
]->symbols
->get_function(f
->name
);
1010 /* If the other shader has no function (and therefore no function
1011 * signatures) with the same name, skip to the next shader.
1016 foreach_iter (exec_list_iterator
, iter
, *f
) {
1017 ir_function_signature
*sig
=
1018 (ir_function_signature
*) iter
.get();
1020 if (!sig
->is_defined
|| sig
->is_builtin
)
1023 ir_function_signature
*other_sig
=
1024 other
->exact_matching_signature(& sig
->parameters
);
1026 if ((other_sig
!= NULL
) && other_sig
->is_defined
1027 && !other_sig
->is_builtin
) {
1028 linker_error(prog
, "function `%s' is multiply defined",
1037 /* Find the shader that defines main, and make a clone of it.
1039 * Starting with the clone, search for undefined references. If one is
1040 * found, find the shader that defines it. Clone the reference and add
1041 * it to the shader. Repeat until there are no undefined references or
1042 * until a reference cannot be resolved.
1044 gl_shader
*main
= NULL
;
1045 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1046 if (get_main_function_signature(shader_list
[i
]) != NULL
) {
1047 main
= shader_list
[i
];
1053 linker_error(prog
, "%s shader lacks `main'\n",
1054 (shader_list
[0]->Type
== GL_VERTEX_SHADER
)
1055 ? "vertex" : "fragment");
1059 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
1060 linked
->ir
= new(linked
) exec_list
;
1061 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
1063 linked
->UniformBlocks
= uniform_blocks
;
1064 linked
->NumUniformBlocks
= num_uniform_blocks
;
1065 ralloc_steal(linked
, linked
->UniformBlocks
);
1067 populate_symbol_table(linked
);
1069 /* The a pointer to the main function in the final linked shader (i.e., the
1070 * copy of the original shader that contained the main function).
1072 ir_function_signature
*const main_sig
= get_main_function_signature(linked
);
1074 /* Move any instructions other than variable declarations or function
1075 * declarations into main.
1077 exec_node
*insertion_point
=
1078 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
1081 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1082 if (shader_list
[i
] == main
)
1085 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
1086 insertion_point
, true, linked
);
1089 /* Resolve initializers for global variables in the linked shader.
1091 unsigned num_linking_shaders
= num_shaders
;
1092 for (unsigned i
= 0; i
< num_shaders
; i
++)
1093 num_linking_shaders
+= shader_list
[i
]->num_builtins_to_link
;
1095 gl_shader
**linking_shaders
=
1096 (gl_shader
**) calloc(num_linking_shaders
, sizeof(gl_shader
*));
1098 memcpy(linking_shaders
, shader_list
,
1099 sizeof(linking_shaders
[0]) * num_shaders
);
1101 unsigned idx
= num_shaders
;
1102 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1103 memcpy(&linking_shaders
[idx
], shader_list
[i
]->builtins_to_link
,
1104 sizeof(linking_shaders
[0]) * shader_list
[i
]->num_builtins_to_link
);
1105 idx
+= shader_list
[i
]->num_builtins_to_link
;
1108 assert(idx
== num_linking_shaders
);
1110 if (!link_function_calls(prog
, linked
, linking_shaders
,
1111 num_linking_shaders
)) {
1112 ctx
->Driver
.DeleteShader(ctx
, linked
);
1116 free(linking_shaders
);
1119 /* At this point linked should contain all of the linked IR, so
1120 * validate it to make sure nothing went wrong.
1123 validate_ir_tree(linked
->ir
);
1126 /* Make a pass over all variable declarations to ensure that arrays with
1127 * unspecified sizes have a size specified. The size is inferred from the
1128 * max_array_access field.
1130 if (linked
!= NULL
) {
1131 array_sizing_visitor v
;
1140 * Update the sizes of linked shader uniform arrays to the maximum
1143 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
1145 * If one or more elements of an array are active,
1146 * GetActiveUniform will return the name of the array in name,
1147 * subject to the restrictions listed above. The type of the array
1148 * is returned in type. The size parameter contains the highest
1149 * array element index used, plus one. The compiler or linker
1150 * determines the highest index used. There will be only one
1151 * active uniform reported by the GL per uniform array.
1155 update_array_sizes(struct gl_shader_program
*prog
)
1157 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
1158 if (prog
->_LinkedShaders
[i
] == NULL
)
1161 foreach_list(node
, prog
->_LinkedShaders
[i
]->ir
) {
1162 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1164 if ((var
== NULL
) || (var
->mode
!= ir_var_uniform
&&
1165 var
->mode
!= ir_var_in
&&
1166 var
->mode
!= ir_var_out
) ||
1167 !var
->type
->is_array())
1170 /* GL_ARB_uniform_buffer_object says that std140 uniforms
1171 * will not be eliminated. Since we always do std140, just
1172 * don't resize arrays in UBOs.
1174 if (var
->uniform_block
!= -1)
1177 unsigned int size
= var
->max_array_access
;
1178 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
1179 if (prog
->_LinkedShaders
[j
] == NULL
)
1182 foreach_list(node2
, prog
->_LinkedShaders
[j
]->ir
) {
1183 ir_variable
*other_var
= ((ir_instruction
*) node2
)->as_variable();
1187 if (strcmp(var
->name
, other_var
->name
) == 0 &&
1188 other_var
->max_array_access
> size
) {
1189 size
= other_var
->max_array_access
;
1194 if (size
+ 1 != var
->type
->fields
.array
->length
) {
1195 /* If this is a built-in uniform (i.e., it's backed by some
1196 * fixed-function state), adjust the number of state slots to
1197 * match the new array size. The number of slots per array entry
1198 * is not known. It seems safe to assume that the total number of
1199 * slots is an integer multiple of the number of array elements.
1200 * Determine the number of slots per array element by dividing by
1201 * the old (total) size.
1203 if (var
->num_state_slots
> 0) {
1204 var
->num_state_slots
= (size
+ 1)
1205 * (var
->num_state_slots
/ var
->type
->length
);
1208 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
1210 /* FINISHME: We should update the types of array
1211 * dereferences of this variable now.
1219 * Find a contiguous set of available bits in a bitmask.
1221 * \param used_mask Bits representing used (1) and unused (0) locations
1222 * \param needed_count Number of contiguous bits needed.
1225 * Base location of the available bits on success or -1 on failure.
1228 find_available_slots(unsigned used_mask
, unsigned needed_count
)
1230 unsigned needed_mask
= (1 << needed_count
) - 1;
1231 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
1233 /* The comparison to 32 is redundant, but without it GCC emits "warning:
1234 * cannot optimize possibly infinite loops" for the loop below.
1236 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
1239 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
1240 if ((needed_mask
& ~used_mask
) == needed_mask
)
1251 * Assign locations for either VS inputs for FS outputs
1253 * \param prog Shader program whose variables need locations assigned
1254 * \param target_index Selector for the program target to receive location
1255 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
1256 * \c MESA_SHADER_FRAGMENT.
1257 * \param max_index Maximum number of generic locations. This corresponds
1258 * to either the maximum number of draw buffers or the
1259 * maximum number of generic attributes.
1262 * If locations are successfully assigned, true is returned. Otherwise an
1263 * error is emitted to the shader link log and false is returned.
1266 assign_attribute_or_color_locations(gl_shader_program
*prog
,
1267 unsigned target_index
,
1270 /* Mark invalid locations as being used.
1272 unsigned used_locations
= (max_index
>= 32)
1273 ? ~0 : ~((1 << max_index
) - 1);
1275 assert((target_index
== MESA_SHADER_VERTEX
)
1276 || (target_index
== MESA_SHADER_FRAGMENT
));
1278 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
1282 /* Operate in a total of four passes.
1284 * 1. Invalidate the location assignments for all vertex shader inputs.
1286 * 2. Assign locations for inputs that have user-defined (via
1287 * glBindVertexAttribLocation) locations and outputs that have
1288 * user-defined locations (via glBindFragDataLocation).
1290 * 3. Sort the attributes without assigned locations by number of slots
1291 * required in decreasing order. Fragmentation caused by attribute
1292 * locations assigned by the application may prevent large attributes
1293 * from having enough contiguous space.
1295 * 4. Assign locations to any inputs without assigned locations.
1298 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
1299 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
1301 const enum ir_variable_mode direction
=
1302 (target_index
== MESA_SHADER_VERTEX
) ? ir_var_in
: ir_var_out
;
1305 link_invalidate_variable_locations(sh
, direction
, generic_base
);
1307 /* Temporary storage for the set of attributes that need locations assigned.
1313 /* Used below in the call to qsort. */
1314 static int compare(const void *a
, const void *b
)
1316 const temp_attr
*const l
= (const temp_attr
*) a
;
1317 const temp_attr
*const r
= (const temp_attr
*) b
;
1319 /* Reversed because we want a descending order sort below. */
1320 return r
->slots
- l
->slots
;
1324 unsigned num_attr
= 0;
1326 foreach_list(node
, sh
->ir
) {
1327 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1329 if ((var
== NULL
) || (var
->mode
!= (unsigned) direction
))
1332 if (var
->explicit_location
) {
1333 if ((var
->location
>= (int)(max_index
+ generic_base
))
1334 || (var
->location
< 0)) {
1336 "invalid explicit location %d specified for `%s'\n",
1338 ? var
->location
: var
->location
- generic_base
,
1342 } else if (target_index
== MESA_SHADER_VERTEX
) {
1345 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
1346 assert(binding
>= VERT_ATTRIB_GENERIC0
);
1347 var
->location
= binding
;
1349 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
1353 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
1354 assert(binding
>= FRAG_RESULT_DATA0
);
1355 var
->location
= binding
;
1357 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
1363 /* If the variable is not a built-in and has a location statically
1364 * assigned in the shader (presumably via a layout qualifier), make sure
1365 * that it doesn't collide with other assigned locations. Otherwise,
1366 * add it to the list of variables that need linker-assigned locations.
1368 const unsigned slots
= count_attribute_slots(var
->type
);
1369 if (var
->location
!= -1) {
1370 if (var
->location
>= generic_base
&& var
->index
< 1) {
1371 /* From page 61 of the OpenGL 4.0 spec:
1373 * "LinkProgram will fail if the attribute bindings assigned
1374 * by BindAttribLocation do not leave not enough space to
1375 * assign a location for an active matrix attribute or an
1376 * active attribute array, both of which require multiple
1377 * contiguous generic attributes."
1379 * Previous versions of the spec contain similar language but omit
1380 * the bit about attribute arrays.
1382 * Page 61 of the OpenGL 4.0 spec also says:
1384 * "It is possible for an application to bind more than one
1385 * attribute name to the same location. This is referred to as
1386 * aliasing. This will only work if only one of the aliased
1387 * attributes is active in the executable program, or if no
1388 * path through the shader consumes more than one attribute of
1389 * a set of attributes aliased to the same location. A link
1390 * error can occur if the linker determines that every path
1391 * through the shader consumes multiple aliased attributes,
1392 * but implementations are not required to generate an error
1395 * These two paragraphs are either somewhat contradictory, or I
1396 * don't fully understand one or both of them.
1398 /* FINISHME: The code as currently written does not support
1399 * FINISHME: attribute location aliasing (see comment above).
1401 /* Mask representing the contiguous slots that will be used by
1404 const unsigned attr
= var
->location
- generic_base
;
1405 const unsigned use_mask
= (1 << slots
) - 1;
1407 /* Generate a link error if the set of bits requested for this
1408 * attribute overlaps any previously allocated bits.
1410 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
1411 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1412 ? "vertex shader input" : "fragment shader output";
1414 "insufficient contiguous locations "
1415 "available for %s `%s' %d %d %d", string
,
1416 var
->name
, used_locations
, use_mask
, attr
);
1420 used_locations
|= (use_mask
<< attr
);
1426 to_assign
[num_attr
].slots
= slots
;
1427 to_assign
[num_attr
].var
= var
;
1431 /* If all of the attributes were assigned locations by the application (or
1432 * are built-in attributes with fixed locations), return early. This should
1433 * be the common case.
1438 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
1440 if (target_index
== MESA_SHADER_VERTEX
) {
1441 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
1442 * only be explicitly assigned by via glBindAttribLocation. Mark it as
1443 * reserved to prevent it from being automatically allocated below.
1445 find_deref_visitor
find("gl_Vertex");
1447 if (find
.variable_found())
1448 used_locations
|= (1 << 0);
1451 for (unsigned i
= 0; i
< num_attr
; i
++) {
1452 /* Mask representing the contiguous slots that will be used by this
1455 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
1457 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
1460 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
1461 ? "vertex shader input" : "fragment shader output";
1464 "insufficient contiguous locations "
1465 "available for %s `%s'",
1466 string
, to_assign
[i
].var
->name
);
1470 to_assign
[i
].var
->location
= generic_base
+ location
;
1471 used_locations
|= (use_mask
<< location
);
1479 * Demote shader inputs and outputs that are not used in other stages
1482 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
1484 foreach_list(node
, sh
->ir
) {
1485 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
1487 if ((var
== NULL
) || (var
->mode
!= int(mode
)))
1490 /* A shader 'in' or 'out' variable is only really an input or output if
1491 * its value is used by other shader stages. This will cause the variable
1492 * to have a location assigned.
1494 if (var
->location
== -1) {
1495 var
->mode
= ir_var_auto
;
1502 * Data structure tracking information about a transform feedback declaration
1505 class tfeedback_decl
1508 bool init(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1509 const void *mem_ctx
, const char *input
);
1510 static bool is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
);
1511 bool assign_location(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1512 ir_variable
*output_var
);
1513 bool accumulate_num_outputs(struct gl_shader_program
*prog
, unsigned *count
);
1514 bool store(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1515 struct gl_transform_feedback_info
*info
, unsigned buffer
,
1516 const unsigned max_outputs
) const;
1519 * True if assign_location() has been called for this object.
1521 bool is_assigned() const
1523 return this->location
!= -1;
1526 bool is_next_buffer_separator() const
1528 return this->next_buffer_separator
;
1531 bool is_varying() const
1533 return !this->next_buffer_separator
&& !this->skip_components
;
1537 * Determine whether this object refers to the variable var.
1539 bool matches_var(ir_variable
*var
) const
1541 if (this->is_clip_distance_mesa
)
1542 return strcmp(var
->name
, "gl_ClipDistanceMESA") == 0;
1544 return strcmp(var
->name
, this->var_name
) == 0;
1548 * The total number of varying components taken up by this variable. Only
1549 * valid if is_assigned() is true.
1551 unsigned num_components() const
1553 if (this->is_clip_distance_mesa
)
1556 return this->vector_elements
* this->matrix_columns
* this->size
;
1561 * The name that was supplied to glTransformFeedbackVaryings. Used for
1562 * error reporting and glGetTransformFeedbackVarying().
1564 const char *orig_name
;
1567 * The name of the variable, parsed from orig_name.
1569 const char *var_name
;
1572 * True if the declaration in orig_name represents an array.
1574 bool is_subscripted
;
1577 * If is_subscripted is true, the subscript that was specified in orig_name.
1579 unsigned array_subscript
;
1582 * True if the variable is gl_ClipDistance and the driver lowers
1583 * gl_ClipDistance to gl_ClipDistanceMESA.
1585 bool is_clip_distance_mesa
;
1588 * The vertex shader output location that the linker assigned for this
1589 * variable. -1 if a location hasn't been assigned yet.
1594 * If location != -1, the number of vector elements in this variable, or 1
1595 * if this variable is a scalar.
1597 unsigned vector_elements
;
1600 * If location != -1, the number of matrix columns in this variable, or 1
1601 * if this variable is not a matrix.
1603 unsigned matrix_columns
;
1605 /** Type of the varying returned by glGetTransformFeedbackVarying() */
1609 * If location != -1, the size that should be returned by
1610 * glGetTransformFeedbackVarying().
1615 * How many components to skip. If non-zero, this is
1616 * gl_SkipComponents{1,2,3,4} from ARB_transform_feedback3.
1618 unsigned skip_components
;
1621 * Whether this is gl_NextBuffer from ARB_transform_feedback3.
1623 bool next_buffer_separator
;
1628 * Initialize this object based on a string that was passed to
1629 * glTransformFeedbackVaryings. If there is a parse error, the error is
1630 * reported using linker_error(), and false is returned.
1633 tfeedback_decl::init(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1634 const void *mem_ctx
, const char *input
)
1636 /* We don't have to be pedantic about what is a valid GLSL variable name,
1637 * because any variable with an invalid name can't exist in the IR anyway.
1640 this->location
= -1;
1641 this->orig_name
= input
;
1642 this->is_clip_distance_mesa
= false;
1643 this->skip_components
= 0;
1644 this->next_buffer_separator
= false;
1646 if (ctx
->Extensions
.ARB_transform_feedback3
) {
1647 /* Parse gl_NextBuffer. */
1648 if (strcmp(input
, "gl_NextBuffer") == 0) {
1649 this->next_buffer_separator
= true;
1653 /* Parse gl_SkipComponents. */
1654 if (strcmp(input
, "gl_SkipComponents1") == 0)
1655 this->skip_components
= 1;
1656 else if (strcmp(input
, "gl_SkipComponents2") == 0)
1657 this->skip_components
= 2;
1658 else if (strcmp(input
, "gl_SkipComponents3") == 0)
1659 this->skip_components
= 3;
1660 else if (strcmp(input
, "gl_SkipComponents4") == 0)
1661 this->skip_components
= 4;
1663 if (this->skip_components
)
1667 /* Parse a declaration. */
1668 const char *bracket
= strrchr(input
, '[');
1671 this->var_name
= ralloc_strndup(mem_ctx
, input
, bracket
- input
);
1672 if (sscanf(bracket
, "[%u]", &this->array_subscript
) != 1) {
1673 linker_error(prog
, "Cannot parse transform feedback varying %s", input
);
1676 this->is_subscripted
= true;
1678 this->var_name
= ralloc_strdup(mem_ctx
, input
);
1679 this->is_subscripted
= false;
1682 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this
1683 * class must behave specially to account for the fact that gl_ClipDistance
1684 * is converted from a float[8] to a vec4[2].
1686 if (ctx
->ShaderCompilerOptions
[MESA_SHADER_VERTEX
].LowerClipDistance
&&
1687 strcmp(this->var_name
, "gl_ClipDistance") == 0) {
1688 this->is_clip_distance_mesa
= true;
1696 * Determine whether two tfeedback_decl objects refer to the same variable and
1697 * array index (if applicable).
1700 tfeedback_decl::is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
)
1702 assert(x
.is_varying() && y
.is_varying());
1704 if (strcmp(x
.var_name
, y
.var_name
) != 0)
1706 if (x
.is_subscripted
!= y
.is_subscripted
)
1708 if (x
.is_subscripted
&& x
.array_subscript
!= y
.array_subscript
)
1715 * Assign a location for this tfeedback_decl object based on the location
1716 * assignment in output_var.
1718 * If an error occurs, the error is reported through linker_error() and false
1722 tfeedback_decl::assign_location(struct gl_context
*ctx
,
1723 struct gl_shader_program
*prog
,
1724 ir_variable
*output_var
)
1726 assert(this->is_varying());
1728 if (output_var
->type
->is_array()) {
1729 /* Array variable */
1730 const unsigned matrix_cols
=
1731 output_var
->type
->fields
.array
->matrix_columns
;
1732 unsigned actual_array_size
= this->is_clip_distance_mesa
?
1733 prog
->Vert
.ClipDistanceArraySize
: output_var
->type
->array_size();
1735 if (this->is_subscripted
) {
1736 /* Check array bounds. */
1737 if (this->array_subscript
>= actual_array_size
) {
1738 linker_error(prog
, "Transform feedback varying %s has index "
1739 "%i, but the array size is %u.",
1740 this->orig_name
, this->array_subscript
,
1744 if (this->is_clip_distance_mesa
) {
1746 output_var
->location
+ this->array_subscript
/ 4;
1749 output_var
->location
+ this->array_subscript
* matrix_cols
;
1753 this->location
= output_var
->location
;
1754 this->size
= actual_array_size
;
1756 this->vector_elements
= output_var
->type
->fields
.array
->vector_elements
;
1757 this->matrix_columns
= matrix_cols
;
1758 if (this->is_clip_distance_mesa
)
1759 this->type
= GL_FLOAT
;
1761 this->type
= output_var
->type
->fields
.array
->gl_type
;
1763 /* Regular variable (scalar, vector, or matrix) */
1764 if (this->is_subscripted
) {
1765 linker_error(prog
, "Transform feedback varying %s requested, "
1766 "but %s is not an array.",
1767 this->orig_name
, this->var_name
);
1770 this->location
= output_var
->location
;
1772 this->vector_elements
= output_var
->type
->vector_elements
;
1773 this->matrix_columns
= output_var
->type
->matrix_columns
;
1774 this->type
= output_var
->type
->gl_type
;
1777 /* From GL_EXT_transform_feedback:
1778 * A program will fail to link if:
1780 * * the total number of components to capture in any varying
1781 * variable in <varyings> is greater than the constant
1782 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
1783 * buffer mode is SEPARATE_ATTRIBS_EXT;
1785 if (prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
&&
1786 this->num_components() >
1787 ctx
->Const
.MaxTransformFeedbackSeparateComponents
) {
1788 linker_error(prog
, "Transform feedback varying %s exceeds "
1789 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
1799 tfeedback_decl::accumulate_num_outputs(struct gl_shader_program
*prog
,
1802 if (!this->is_varying()) {
1806 if (!this->is_assigned()) {
1807 /* From GL_EXT_transform_feedback:
1808 * A program will fail to link if:
1810 * * any variable name specified in the <varyings> array is not
1811 * declared as an output in the geometry shader (if present) or
1812 * the vertex shader (if no geometry shader is present);
1814 linker_error(prog
, "Transform feedback varying %s undeclared.",
1819 unsigned translated_size
= this->size
;
1820 if (this->is_clip_distance_mesa
)
1821 translated_size
= (translated_size
+ 3) / 4;
1823 *count
+= translated_size
* this->matrix_columns
;
1830 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
1832 * If an error occurs, the error is reported through linker_error() and false
1836 tfeedback_decl::store(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1837 struct gl_transform_feedback_info
*info
,
1838 unsigned buffer
, const unsigned max_outputs
) const
1840 assert(!this->next_buffer_separator
);
1842 /* Handle gl_SkipComponents. */
1843 if (this->skip_components
) {
1844 info
->BufferStride
[buffer
] += this->skip_components
;
1848 /* From GL_EXT_transform_feedback:
1849 * A program will fail to link if:
1851 * * the total number of components to capture is greater than
1852 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
1853 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
1855 if (prog
->TransformFeedback
.BufferMode
== GL_INTERLEAVED_ATTRIBS
&&
1856 info
->BufferStride
[buffer
] + this->num_components() >
1857 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1858 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1859 "limit has been exceeded.");
1863 unsigned translated_size
= this->size
;
1864 if (this->is_clip_distance_mesa
)
1865 translated_size
= (translated_size
+ 3) / 4;
1866 unsigned components_so_far
= 0;
1867 for (unsigned index
= 0; index
< translated_size
; ++index
) {
1868 for (unsigned v
= 0; v
< this->matrix_columns
; ++v
) {
1869 unsigned num_components
= this->vector_elements
;
1870 assert(info
->NumOutputs
< max_outputs
);
1871 info
->Outputs
[info
->NumOutputs
].ComponentOffset
= 0;
1872 if (this->is_clip_distance_mesa
) {
1873 if (this->is_subscripted
) {
1875 info
->Outputs
[info
->NumOutputs
].ComponentOffset
=
1876 this->array_subscript
% 4;
1878 num_components
= MIN2(4, this->size
- components_so_far
);
1881 info
->Outputs
[info
->NumOutputs
].OutputRegister
=
1882 this->location
+ v
+ index
* this->matrix_columns
;
1883 info
->Outputs
[info
->NumOutputs
].NumComponents
= num_components
;
1884 info
->Outputs
[info
->NumOutputs
].OutputBuffer
= buffer
;
1885 info
->Outputs
[info
->NumOutputs
].DstOffset
= info
->BufferStride
[buffer
];
1887 info
->BufferStride
[buffer
] += num_components
;
1888 components_so_far
+= num_components
;
1891 assert(components_so_far
== this->num_components());
1893 info
->Varyings
[info
->NumVarying
].Name
= ralloc_strdup(prog
, this->orig_name
);
1894 info
->Varyings
[info
->NumVarying
].Type
= this->type
;
1895 info
->Varyings
[info
->NumVarying
].Size
= this->size
;
1903 * Parse all the transform feedback declarations that were passed to
1904 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
1906 * If an error occurs, the error is reported through linker_error() and false
1910 parse_tfeedback_decls(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
1911 const void *mem_ctx
, unsigned num_names
,
1912 char **varying_names
, tfeedback_decl
*decls
)
1914 for (unsigned i
= 0; i
< num_names
; ++i
) {
1915 if (!decls
[i
].init(ctx
, prog
, mem_ctx
, varying_names
[i
]))
1918 if (!decls
[i
].is_varying())
1921 /* From GL_EXT_transform_feedback:
1922 * A program will fail to link if:
1924 * * any two entries in the <varyings> array specify the same varying
1927 * We interpret this to mean "any two entries in the <varyings> array
1928 * specify the same varying variable and array index", since transform
1929 * feedback of arrays would be useless otherwise.
1931 for (unsigned j
= 0; j
< i
; ++j
) {
1932 if (!decls
[j
].is_varying())
1935 if (tfeedback_decl::is_same(decls
[i
], decls
[j
])) {
1936 linker_error(prog
, "Transform feedback varying %s specified "
1937 "more than once.", varying_names
[i
]);
1947 * Assign a location for a variable that is produced in one pipeline stage
1948 * (the "producer") and consumed in the next stage (the "consumer").
1950 * \param input_var is the input variable declaration in the consumer.
1952 * \param output_var is the output variable declaration in the producer.
1954 * \param input_index is the counter that keeps track of assigned input
1955 * locations in the consumer.
1957 * \param output_index is the counter that keeps track of assigned output
1958 * locations in the producer.
1960 * It is permissible for \c input_var to be NULL (this happens if a variable
1961 * is output by the producer and consumed by transform feedback, but not
1962 * consumed by the consumer).
1964 * If the variable has already been assigned a location, this function has no
1968 assign_varying_location(ir_variable
*input_var
, ir_variable
*output_var
,
1969 unsigned *input_index
, unsigned *output_index
)
1971 if (output_var
->location
!= -1) {
1972 /* Location already assigned. */
1977 assert(input_var
->location
== -1);
1978 input_var
->location
= *input_index
;
1981 output_var
->location
= *output_index
;
1983 /* FINISHME: Support for "varying" records in GLSL 1.50. */
1984 assert(!output_var
->type
->is_record());
1986 if (output_var
->type
->is_array()) {
1987 const unsigned slots
= output_var
->type
->length
1988 * output_var
->type
->fields
.array
->matrix_columns
;
1990 *output_index
+= slots
;
1991 *input_index
+= slots
;
1993 const unsigned slots
= output_var
->type
->matrix_columns
;
1995 *output_index
+= slots
;
1996 *input_index
+= slots
;
2002 * Is the given variable a varying variable to be counted against the
2003 * limit in ctx->Const.MaxVarying?
2004 * This includes variables such as texcoords, colors and generic
2005 * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord.
2008 is_varying_var(GLenum shaderType
, const ir_variable
*var
)
2010 /* Only fragment shaders will take a varying variable as an input */
2011 if (shaderType
== GL_FRAGMENT_SHADER
&&
2012 var
->mode
== ir_var_in
&&
2013 var
->explicit_location
) {
2014 switch (var
->location
) {
2015 case FRAG_ATTRIB_WPOS
:
2016 case FRAG_ATTRIB_FACE
:
2017 case FRAG_ATTRIB_PNTC
:
2028 * Assign locations for all variables that are produced in one pipeline stage
2029 * (the "producer") and consumed in the next stage (the "consumer").
2031 * Variables produced by the producer may also be consumed by transform
2034 * \param num_tfeedback_decls is the number of declarations indicating
2035 * variables that may be consumed by transform feedback.
2037 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
2038 * representing the result of parsing the strings passed to
2039 * glTransformFeedbackVaryings(). assign_location() will be called for
2040 * each of these objects that matches one of the outputs of the
2043 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
2044 * be NULL. In this case, varying locations are assigned solely based on the
2045 * requirements of transform feedback.
2048 assign_varying_locations(struct gl_context
*ctx
,
2049 struct gl_shader_program
*prog
,
2050 gl_shader
*producer
, gl_shader
*consumer
,
2051 unsigned num_tfeedback_decls
,
2052 tfeedback_decl
*tfeedback_decls
)
2054 /* FINISHME: Set dynamically when geometry shader support is added. */
2055 unsigned output_index
= VERT_RESULT_VAR0
;
2056 unsigned input_index
= FRAG_ATTRIB_VAR0
;
2058 /* Operate in a total of three passes.
2060 * 1. Assign locations for any matching inputs and outputs.
2062 * 2. Mark output variables in the producer that do not have locations as
2063 * not being outputs. This lets the optimizer eliminate them.
2065 * 3. Mark input variables in the consumer that do not have locations as
2066 * not being inputs. This lets the optimizer eliminate them.
2069 link_invalidate_variable_locations(producer
, ir_var_out
, VERT_RESULT_VAR0
);
2071 link_invalidate_variable_locations(consumer
, ir_var_in
, FRAG_ATTRIB_VAR0
);
2073 foreach_list(node
, producer
->ir
) {
2074 ir_variable
*const output_var
= ((ir_instruction
*) node
)->as_variable();
2076 if ((output_var
== NULL
) || (output_var
->mode
!= ir_var_out
))
2079 ir_variable
*input_var
=
2080 consumer
? consumer
->symbols
->get_variable(output_var
->name
) : NULL
;
2082 if (input_var
&& input_var
->mode
!= ir_var_in
)
2086 assign_varying_location(input_var
, output_var
, &input_index
,
2090 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2091 if (!tfeedback_decls
[i
].is_varying())
2094 if (!tfeedback_decls
[i
].is_assigned() &&
2095 tfeedback_decls
[i
].matches_var(output_var
)) {
2096 if (output_var
->location
== -1) {
2097 assign_varying_location(input_var
, output_var
, &input_index
,
2100 if (!tfeedback_decls
[i
].assign_location(ctx
, prog
, output_var
))
2106 unsigned varying_vectors
= 0;
2109 foreach_list(node
, consumer
->ir
) {
2110 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
2112 if ((var
== NULL
) || (var
->mode
!= ir_var_in
))
2115 if (var
->location
== -1) {
2116 if (prog
->Version
<= 120) {
2117 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
2119 * Only those varying variables used (i.e. read) in
2120 * the fragment shader executable must be written to
2121 * by the vertex shader executable; declaring
2122 * superfluous varying variables in a vertex shader is
2125 * We interpret this text as meaning that the VS must
2126 * write the variable for the FS to read it. See
2127 * "glsl1-varying read but not written" in piglit.
2130 linker_error(prog
, "fragment shader varying %s not written "
2131 "by vertex shader\n.", var
->name
);
2134 /* An 'in' variable is only really a shader input if its
2135 * value is written by the previous stage.
2137 var
->mode
= ir_var_auto
;
2138 } else if (is_varying_var(consumer
->Type
, var
)) {
2139 /* The packing rules are used for vertex shader inputs are also
2140 * used for fragment shader inputs.
2142 varying_vectors
+= count_attribute_slots(var
->type
);
2147 if (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100) {
2148 if (varying_vectors
> ctx
->Const
.MaxVarying
) {
2149 if (ctx
->Const
.GLSLSkipStrictMaxVaryingLimitCheck
) {
2150 linker_warning(prog
, "shader uses too many varying vectors "
2151 "(%u > %u), but the driver will try to optimize "
2152 "them out; this is non-portable out-of-spec "
2154 varying_vectors
, ctx
->Const
.MaxVarying
);
2156 linker_error(prog
, "shader uses too many varying vectors "
2158 varying_vectors
, ctx
->Const
.MaxVarying
);
2163 const unsigned float_components
= varying_vectors
* 4;
2164 if (float_components
> ctx
->Const
.MaxVarying
* 4) {
2165 if (ctx
->Const
.GLSLSkipStrictMaxVaryingLimitCheck
) {
2166 linker_warning(prog
, "shader uses too many varying components "
2167 "(%u > %u), but the driver will try to optimize "
2168 "them out; this is non-portable out-of-spec "
2170 float_components
, ctx
->Const
.MaxVarying
* 4);
2172 linker_error(prog
, "shader uses too many varying components "
2174 float_components
, ctx
->Const
.MaxVarying
* 4);
2185 * Store transform feedback location assignments into
2186 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
2188 * If an error occurs, the error is reported through linker_error() and false
2192 store_tfeedback_info(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
2193 unsigned num_tfeedback_decls
,
2194 tfeedback_decl
*tfeedback_decls
)
2196 bool separate_attribs_mode
=
2197 prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
;
2199 ralloc_free(prog
->LinkedTransformFeedback
.Varyings
);
2200 ralloc_free(prog
->LinkedTransformFeedback
.Outputs
);
2202 memset(&prog
->LinkedTransformFeedback
, 0,
2203 sizeof(prog
->LinkedTransformFeedback
));
2205 prog
->LinkedTransformFeedback
.Varyings
=
2207 struct gl_transform_feedback_varying_info
,
2208 num_tfeedback_decls
);
2210 unsigned num_outputs
= 0;
2211 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
)
2212 if (!tfeedback_decls
[i
].accumulate_num_outputs(prog
, &num_outputs
))
2215 prog
->LinkedTransformFeedback
.Outputs
=
2217 struct gl_transform_feedback_output
,
2220 unsigned num_buffers
= 0;
2222 if (separate_attribs_mode
) {
2223 /* GL_SEPARATE_ATTRIBS */
2224 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2225 if (!tfeedback_decls
[i
].store(ctx
, prog
, &prog
->LinkedTransformFeedback
,
2226 num_buffers
, num_outputs
))
2233 /* GL_INVERLEAVED_ATTRIBS */
2234 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2235 if (tfeedback_decls
[i
].is_next_buffer_separator()) {
2240 if (!tfeedback_decls
[i
].store(ctx
, prog
,
2241 &prog
->LinkedTransformFeedback
,
2242 num_buffers
, num_outputs
))
2248 assert(prog
->LinkedTransformFeedback
.NumOutputs
== num_outputs
);
2250 prog
->LinkedTransformFeedback
.NumBuffers
= num_buffers
;
2255 * Store the gl_FragDepth layout in the gl_shader_program struct.
2258 store_fragdepth_layout(struct gl_shader_program
*prog
)
2260 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2264 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2266 /* We don't look up the gl_FragDepth symbol directly because if
2267 * gl_FragDepth is not used in the shader, it's removed from the IR.
2268 * However, the symbol won't be removed from the symbol table.
2270 * We're only interested in the cases where the variable is NOT removed
2273 foreach_list(node
, ir
) {
2274 ir_variable
*const var
= ((ir_instruction
*) node
)->as_variable();
2276 if (var
== NULL
|| var
->mode
!= ir_var_out
) {
2280 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2281 switch (var
->depth_layout
) {
2282 case ir_depth_layout_none
:
2283 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2285 case ir_depth_layout_any
:
2286 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2288 case ir_depth_layout_greater
:
2289 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2291 case ir_depth_layout_less
:
2292 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2294 case ir_depth_layout_unchanged
:
2295 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2306 * Validate the resources used by a program versus the implementation limits
2309 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2311 static const char *const shader_names
[MESA_SHADER_TYPES
] = {
2312 "vertex", "fragment", "geometry"
2315 const unsigned max_samplers
[MESA_SHADER_TYPES
] = {
2316 ctx
->Const
.MaxVertexTextureImageUnits
,
2317 ctx
->Const
.MaxTextureImageUnits
,
2318 ctx
->Const
.MaxGeometryTextureImageUnits
2321 const unsigned max_uniform_components
[MESA_SHADER_TYPES
] = {
2322 ctx
->Const
.VertexProgram
.MaxUniformComponents
,
2323 ctx
->Const
.FragmentProgram
.MaxUniformComponents
,
2324 0 /* FINISHME: Geometry shaders. */
2327 const unsigned max_uniform_blocks
[MESA_SHADER_TYPES
] = {
2328 ctx
->Const
.VertexProgram
.MaxUniformBlocks
,
2329 ctx
->Const
.FragmentProgram
.MaxUniformBlocks
,
2330 ctx
->Const
.GeometryProgram
.MaxUniformBlocks
,
2333 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2334 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2339 if (sh
->num_samplers
> max_samplers
[i
]) {
2340 linker_error(prog
, "Too many %s shader texture samplers",
2344 if (sh
->num_uniform_components
> max_uniform_components
[i
]) {
2345 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2346 linker_warning(prog
, "Too many %s shader uniform components, "
2347 "but the driver will try to optimize them out; "
2348 "this is non-portable out-of-spec behavior\n",
2351 linker_error(prog
, "Too many %s shader uniform components",
2357 unsigned blocks
[MESA_SHADER_TYPES
] = {0};
2358 unsigned total_uniform_blocks
= 0;
2360 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2361 for (unsigned j
= 0; j
< MESA_SHADER_TYPES
; j
++) {
2362 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2364 total_uniform_blocks
++;
2368 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2369 linker_error(prog
, "Too many combined uniform blocks (%d/%d)",
2370 prog
->NumUniformBlocks
,
2371 ctx
->Const
.MaxCombinedUniformBlocks
);
2373 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2374 if (blocks
[i
] > max_uniform_blocks
[i
]) {
2375 linker_error(prog
, "Too many %s uniform blocks (%d/%d)",
2378 max_uniform_blocks
[i
]);
2385 return prog
->LinkStatus
;
2389 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2391 tfeedback_decl
*tfeedback_decls
= NULL
;
2392 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
2394 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
2396 prog
->LinkStatus
= false;
2397 prog
->Validated
= false;
2398 prog
->_Used
= false;
2400 ralloc_free(prog
->InfoLog
);
2401 prog
->InfoLog
= ralloc_strdup(NULL
, "");
2403 ralloc_free(prog
->UniformBlocks
);
2404 prog
->UniformBlocks
= NULL
;
2405 prog
->NumUniformBlocks
= 0;
2406 for (int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2407 ralloc_free(prog
->UniformBlockStageIndex
[i
]);
2408 prog
->UniformBlockStageIndex
[i
] = NULL
;
2411 /* Separate the shaders into groups based on their type.
2413 struct gl_shader
**vert_shader_list
;
2414 unsigned num_vert_shaders
= 0;
2415 struct gl_shader
**frag_shader_list
;
2416 unsigned num_frag_shaders
= 0;
2418 vert_shader_list
= (struct gl_shader
**)
2419 calloc(2 * prog
->NumShaders
, sizeof(struct gl_shader
*));
2420 frag_shader_list
= &vert_shader_list
[prog
->NumShaders
];
2422 unsigned min_version
= UINT_MAX
;
2423 unsigned max_version
= 0;
2424 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
2425 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
2426 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
2428 switch (prog
->Shaders
[i
]->Type
) {
2429 case GL_VERTEX_SHADER
:
2430 vert_shader_list
[num_vert_shaders
] = prog
->Shaders
[i
];
2433 case GL_FRAGMENT_SHADER
:
2434 frag_shader_list
[num_frag_shaders
] = prog
->Shaders
[i
];
2437 case GL_GEOMETRY_SHADER
:
2438 /* FINISHME: Support geometry shaders. */
2439 assert(prog
->Shaders
[i
]->Type
!= GL_GEOMETRY_SHADER
);
2444 /* Previous to GLSL version 1.30, different compilation units could mix and
2445 * match shading language versions. With GLSL 1.30 and later, the versions
2446 * of all shaders must match.
2448 assert(min_version
>= 100);
2449 assert(max_version
<= 140);
2450 if ((max_version
>= 130 || min_version
== 100)
2451 && min_version
!= max_version
) {
2452 linker_error(prog
, "all shaders must use same shading "
2453 "language version\n");
2457 prog
->Version
= max_version
;
2459 for (unsigned int i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2460 if (prog
->_LinkedShaders
[i
] != NULL
)
2461 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
2463 prog
->_LinkedShaders
[i
] = NULL
;
2466 /* Link all shaders for a particular stage and validate the result.
2468 if (num_vert_shaders
> 0) {
2469 gl_shader
*const sh
=
2470 link_intrastage_shaders(mem_ctx
, ctx
, prog
, vert_shader_list
,
2476 if (!validate_vertex_shader_executable(prog
, sh
))
2479 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2483 if (num_frag_shaders
> 0) {
2484 gl_shader
*const sh
=
2485 link_intrastage_shaders(mem_ctx
, ctx
, prog
, frag_shader_list
,
2491 if (!validate_fragment_shader_executable(prog
, sh
))
2494 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
],
2498 /* Here begins the inter-stage linking phase. Some initial validation is
2499 * performed, then locations are assigned for uniforms, attributes, and
2502 if (cross_validate_uniforms(prog
)) {
2505 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
2506 if (prog
->_LinkedShaders
[prev
] != NULL
)
2510 /* Validate the inputs of each stage with the output of the preceding
2513 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
2514 if (prog
->_LinkedShaders
[i
] == NULL
)
2517 if (!cross_validate_outputs_to_inputs(prog
,
2518 prog
->_LinkedShaders
[prev
],
2519 prog
->_LinkedShaders
[i
]))
2525 prog
->LinkStatus
= true;
2528 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
2529 * it before optimization because we want most of the checks to get
2530 * dropped thanks to constant propagation.
2532 if (max_version
>= 130) {
2533 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2535 lower_discard_flow(sh
->ir
);
2539 if (!interstage_cross_validate_uniform_blocks(prog
))
2542 /* Do common optimization before assigning storage for attributes,
2543 * uniforms, and varyings. Later optimization could possibly make
2544 * some of that unused.
2546 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2547 if (prog
->_LinkedShaders
[i
] == NULL
)
2550 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
2551 if (!prog
->LinkStatus
)
2554 if (ctx
->ShaderCompilerOptions
[i
].LowerClipDistance
)
2555 lower_clip_distance(prog
->_LinkedShaders
[i
]->ir
);
2557 unsigned max_unroll
= ctx
->ShaderCompilerOptions
[i
].MaxUnrollIterations
;
2559 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false, max_unroll
))
2563 /* FINISHME: The value of the max_attribute_index parameter is
2564 * FINISHME: implementation dependent based on the value of
2565 * FINISHME: GL_MAX_VERTEX_ATTRIBS. GL_MAX_VERTEX_ATTRIBS must be
2566 * FINISHME: at least 16, so hardcode 16 for now.
2568 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_VERTEX
, 16)) {
2572 if (!assign_attribute_or_color_locations(prog
, MESA_SHADER_FRAGMENT
, MAX2(ctx
->Const
.MaxDrawBuffers
, ctx
->Const
.MaxDualSourceDrawBuffers
))) {
2577 for (prev
= 0; prev
< MESA_SHADER_TYPES
; prev
++) {
2578 if (prog
->_LinkedShaders
[prev
] != NULL
)
2582 if (num_tfeedback_decls
!= 0) {
2583 /* From GL_EXT_transform_feedback:
2584 * A program will fail to link if:
2586 * * the <count> specified by TransformFeedbackVaryingsEXT is
2587 * non-zero, but the program object has no vertex or geometry
2590 if (prev
>= MESA_SHADER_FRAGMENT
) {
2591 linker_error(prog
, "Transform feedback varyings specified, but "
2592 "no vertex or geometry shader is present.");
2596 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
2597 prog
->TransformFeedback
.NumVarying
);
2598 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
2599 prog
->TransformFeedback
.VaryingNames
,
2604 for (unsigned i
= prev
+ 1; i
< MESA_SHADER_TYPES
; i
++) {
2605 if (prog
->_LinkedShaders
[i
] == NULL
)
2608 if (!assign_varying_locations(
2609 ctx
, prog
, prog
->_LinkedShaders
[prev
], prog
->_LinkedShaders
[i
],
2610 i
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
2617 if (prev
!= MESA_SHADER_FRAGMENT
&& num_tfeedback_decls
!= 0) {
2618 /* There was no fragment shader, but we still have to assign varying
2619 * locations for use by transform feedback.
2621 if (!assign_varying_locations(
2622 ctx
, prog
, prog
->_LinkedShaders
[prev
], NULL
, num_tfeedback_decls
,
2627 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
2630 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] != NULL
) {
2631 demote_shader_inputs_and_outputs(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
],
2634 /* Eliminate code that is now dead due to unused vertex outputs being
2637 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_VERTEX
]->ir
, false))
2641 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
2642 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
2644 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
2645 demote_shader_inputs_and_outputs(sh
, ir_var_inout
);
2646 demote_shader_inputs_and_outputs(sh
, ir_var_out
);
2648 /* Eliminate code that is now dead due to unused geometry outputs being
2651 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
, false))
2655 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] != NULL
) {
2656 gl_shader
*const sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
2658 demote_shader_inputs_and_outputs(sh
, ir_var_in
);
2660 /* Eliminate code that is now dead due to unused fragment inputs being
2661 * demoted. This shouldn't actually do anything other than remove
2662 * declarations of the (now unused) global variables.
2664 while (do_dead_code(prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
, false))
2668 update_array_sizes(prog
);
2669 link_assign_uniform_locations(prog
);
2670 store_fragdepth_layout(prog
);
2672 if (!check_resources(ctx
, prog
))
2675 /* OpenGL ES requires that a vertex shader and a fragment shader both be
2676 * present in a linked program. By checking for use of shading language
2677 * version 1.00, we also catch the GL_ARB_ES2_compatibility case.
2679 if (!prog
->InternalSeparateShader
&&
2680 (ctx
->API
== API_OPENGLES2
|| prog
->Version
== 100)) {
2681 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
2682 linker_error(prog
, "program lacks a vertex shader\n");
2683 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2684 linker_error(prog
, "program lacks a fragment shader\n");
2688 /* FINISHME: Assign fragment shader output locations. */
2691 free(vert_shader_list
);
2693 for (unsigned i
= 0; i
< MESA_SHADER_TYPES
; i
++) {
2694 if (prog
->_LinkedShaders
[i
] == NULL
)
2697 /* Retain any live IR, but trash the rest. */
2698 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
2700 /* The symbol table in the linked shaders may contain references to
2701 * variables that were removed (e.g., unused uniforms). Since it may
2702 * contain junk, there is no possible valid use. Delete it and set the
2705 delete prog
->_LinkedShaders
[i
]->symbols
;
2706 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
2709 ralloc_free(mem_ctx
);