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>
68 #include "main/core.h"
69 #include "glsl_symbol_table.h"
70 #include "glsl_parser_extras.h"
73 #include "program/hash_table.h"
75 #include "link_varyings.h"
76 #include "ir_optimization.h"
77 #include "ir_rvalue_visitor.h"
78 #include "ir_uniform.h"
80 #include "main/shaderobj.h"
81 #include "main/enums.h"
84 void linker_error(gl_shader_program
*, const char *, ...);
89 * Visitor that determines whether or not a variable is ever written.
91 class find_assignment_visitor
: public ir_hierarchical_visitor
{
93 find_assignment_visitor(const char *name
)
94 : name(name
), found(false)
99 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
101 ir_variable
*const var
= ir
->lhs
->variable_referenced();
103 if (strcmp(name
, var
->name
) == 0) {
108 return visit_continue_with_parent
;
111 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
113 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
114 actual_node
, &ir
->actual_parameters
) {
115 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
116 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
118 if (sig_param
->data
.mode
== ir_var_function_out
||
119 sig_param
->data
.mode
== ir_var_function_inout
) {
120 ir_variable
*var
= param_rval
->variable_referenced();
121 if (var
&& strcmp(name
, var
->name
) == 0) {
128 if (ir
->return_deref
!= NULL
) {
129 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
131 if (strcmp(name
, var
->name
) == 0) {
137 return visit_continue_with_parent
;
140 bool variable_found()
146 const char *name
; /**< Find writes to a variable with this name. */
147 bool found
; /**< Was a write to the variable found? */
152 * Visitor that determines whether or not a variable is ever read.
154 class find_deref_visitor
: public ir_hierarchical_visitor
{
156 find_deref_visitor(const char *name
)
157 : name(name
), found(false)
162 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
164 if (strcmp(this->name
, ir
->var
->name
) == 0) {
169 return visit_continue
;
172 bool variable_found() const
178 const char *name
; /**< Find writes to a variable with this name. */
179 bool found
; /**< Was a write to the variable found? */
183 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
185 unsigned num_vertices
;
186 gl_shader_program
*prog
;
188 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
190 this->num_vertices
= num_vertices
;
194 virtual ~geom_array_resize_visitor()
199 virtual ir_visitor_status
visit(ir_variable
*var
)
201 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
202 return visit_continue
;
204 unsigned size
= var
->type
->length
;
206 /* Generate a link error if the shader has declared this array with an
209 if (size
&& size
!= this->num_vertices
) {
210 linker_error(this->prog
, "size of array %s declared as %u, "
211 "but number of input vertices is %u\n",
212 var
->name
, size
, this->num_vertices
);
213 return visit_continue
;
216 /* Generate a link error if the shader attempts to access an input
217 * array using an index too large for its actual size assigned at link
220 if (var
->data
.max_array_access
>= this->num_vertices
) {
221 linker_error(this->prog
, "geometry shader accesses element %i of "
222 "%s, but only %i input vertices\n",
223 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
224 return visit_continue
;
227 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
229 var
->data
.max_array_access
= this->num_vertices
- 1;
231 return visit_continue
;
234 /* Dereferences of input variables need to be updated so that their type
235 * matches the newly assigned type of the variable they are accessing. */
236 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
238 ir
->type
= ir
->var
->type
;
239 return visit_continue
;
242 /* Dereferences of 2D input arrays need to be updated so that their type
243 * matches the newly assigned type of the array they are accessing. */
244 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
246 const glsl_type
*const vt
= ir
->array
->type
;
248 ir
->type
= vt
->fields
.array
;
249 return visit_continue
;
253 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
255 unsigned num_vertices
;
256 gl_shader_program
*prog
;
258 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
260 this->num_vertices
= num_vertices
;
264 virtual ~tess_eval_array_resize_visitor()
269 virtual ir_visitor_status
visit(ir_variable
*var
)
271 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
272 return visit_continue
;
274 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
276 var
->data
.max_array_access
= this->num_vertices
- 1;
278 return visit_continue
;
281 /* Dereferences of input variables need to be updated so that their type
282 * matches the newly assigned type of the variable they are accessing. */
283 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
285 ir
->type
= ir
->var
->type
;
286 return visit_continue
;
289 /* Dereferences of 2D input arrays need to be updated so that their type
290 * matches the newly assigned type of the array they are accessing. */
291 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
293 const glsl_type
*const vt
= ir
->array
->type
;
295 ir
->type
= vt
->fields
.array
;
296 return visit_continue
;
300 class barrier_use_visitor
: public ir_hierarchical_visitor
{
302 barrier_use_visitor(gl_shader_program
*prog
)
303 : prog(prog
), in_main(false), after_return(false), control_flow(0)
307 virtual ~barrier_use_visitor()
312 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
314 if (strcmp(ir
->name
, "main") == 0)
317 return visit_continue
;
320 virtual ir_visitor_status
visit_leave(ir_function
*)
323 after_return
= false;
324 return visit_continue
;
327 virtual ir_visitor_status
visit_leave(ir_return
*)
330 return visit_continue
;
333 virtual ir_visitor_status
visit_enter(ir_if
*)
336 return visit_continue
;
339 virtual ir_visitor_status
visit_leave(ir_if
*)
342 return visit_continue
;
345 virtual ir_visitor_status
visit_enter(ir_loop
*)
348 return visit_continue
;
351 virtual ir_visitor_status
visit_leave(ir_loop
*)
354 return visit_continue
;
357 /* FINISHME: `switch` is not expressed at the IR level -- it's already
358 * been lowered to a mess of `if`s. We'll correctly disallow any use of
359 * barrier() in a conditional path within the switch, but not in a path
360 * which is always hit.
363 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
365 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
366 /* Use of barrier(); determine if it is legal: */
368 linker_error(prog
, "Builtin barrier() may only be used in main");
373 linker_error(prog
, "Builtin barrier() may not be used after return");
377 if (control_flow
!= 0) {
378 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
382 return visit_continue
;
386 gl_shader_program
*prog
;
387 bool in_main
, after_return
;
392 * Visitor that determines the highest stream id to which a (geometry) shader
393 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
395 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
397 find_emit_vertex_visitor(int max_allowed
)
398 : max_stream_allowed(max_allowed
),
399 invalid_stream_id(0),
400 invalid_stream_id_from_emit_vertex(false),
401 end_primitive_found(false),
402 uses_non_zero_stream(false)
407 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
409 int stream_id
= ir
->stream_id();
412 invalid_stream_id
= stream_id
;
413 invalid_stream_id_from_emit_vertex
= true;
417 if (stream_id
> max_stream_allowed
) {
418 invalid_stream_id
= stream_id
;
419 invalid_stream_id_from_emit_vertex
= true;
424 uses_non_zero_stream
= true;
426 return visit_continue
;
429 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
431 end_primitive_found
= true;
433 int stream_id
= ir
->stream_id();
436 invalid_stream_id
= stream_id
;
437 invalid_stream_id_from_emit_vertex
= false;
441 if (stream_id
> max_stream_allowed
) {
442 invalid_stream_id
= stream_id
;
443 invalid_stream_id_from_emit_vertex
= false;
448 uses_non_zero_stream
= true;
450 return visit_continue
;
455 return invalid_stream_id
!= 0;
458 const char *error_func()
460 return invalid_stream_id_from_emit_vertex
?
461 "EmitStreamVertex" : "EndStreamPrimitive";
466 return invalid_stream_id
;
471 return uses_non_zero_stream
;
474 bool uses_end_primitive()
476 return end_primitive_found
;
480 int max_stream_allowed
;
481 int invalid_stream_id
;
482 bool invalid_stream_id_from_emit_vertex
;
483 bool end_primitive_found
;
484 bool uses_non_zero_stream
;
487 /* Class that finds array derefs and check if indexes are dynamic. */
488 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
491 dynamic_sampler_array_indexing_visitor() :
492 dynamic_sampler_array_indexing(false)
496 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
498 if (!ir
->variable_referenced())
499 return visit_continue
;
501 if (!ir
->variable_referenced()->type
->contains_sampler())
502 return visit_continue
;
504 if (!ir
->array_index
->constant_expression_value()) {
505 dynamic_sampler_array_indexing
= true;
508 return visit_continue
;
511 bool uses_dynamic_sampler_array_indexing()
513 return dynamic_sampler_array_indexing
;
517 bool dynamic_sampler_array_indexing
;
520 } /* anonymous namespace */
523 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
527 ralloc_strcat(&prog
->InfoLog
, "error: ");
529 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
532 prog
->LinkStatus
= false;
537 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
541 ralloc_strcat(&prog
->InfoLog
, "warning: ");
543 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
550 * Given a string identifying a program resource, break it into a base name
551 * and an optional array index in square brackets.
553 * If an array index is present, \c out_base_name_end is set to point to the
554 * "[" that precedes the array index, and the array index itself is returned
557 * If no array index is present (or if the array index is negative or
558 * mal-formed), \c out_base_name_end, is set to point to the null terminator
559 * at the end of the input string, and -1 is returned.
561 * Only the final array index is parsed; if the string contains other array
562 * indices (or structure field accesses), they are left in the base name.
564 * No attempt is made to check that the base name is properly formed;
565 * typically the caller will look up the base name in a hash table, so
566 * ill-formed base names simply turn into hash table lookup failures.
569 parse_program_resource_name(const GLchar
*name
,
570 const GLchar
**out_base_name_end
)
572 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
574 * "When an integer array element or block instance number is part of
575 * the name string, it will be specified in decimal form without a "+"
576 * or "-" sign or any extra leading zeroes. Additionally, the name
577 * string will not include white space anywhere in the string."
580 const size_t len
= strlen(name
);
581 *out_base_name_end
= name
+ len
;
583 if (len
== 0 || name
[len
-1] != ']')
586 /* Walk backwards over the string looking for a non-digit character. This
587 * had better be the opening bracket for an array index.
589 * Initially, i specifies the location of the ']'. Since the string may
590 * contain only the ']' charcater, walk backwards very carefully.
593 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
596 if ((i
== 0) || name
[i
-1] != '[')
599 long array_index
= strtol(&name
[i
], NULL
, 10);
603 /* Check for leading zero */
604 if (name
[i
] == '0' && name
[i
+1] != ']')
607 *out_base_name_end
= name
+ (i
- 1);
613 link_invalidate_variable_locations(exec_list
*ir
)
615 foreach_in_list(ir_instruction
, node
, ir
) {
616 ir_variable
*const var
= node
->as_variable();
621 /* Only assign locations for variables that lack an explicit location.
622 * Explicit locations are set for all built-in variables, generic vertex
623 * shader inputs (via layout(location=...)), and generic fragment shader
624 * outputs (also via layout(location=...)).
626 if (!var
->data
.explicit_location
) {
627 var
->data
.location
= -1;
628 var
->data
.location_frac
= 0;
631 /* ir_variable::is_unmatched_generic_inout is used by the linker while
632 * connecting outputs from one stage to inputs of the next stage.
634 * There are two implicit assumptions here. First, we assume that any
635 * built-in variable (i.e., non-generic in or out) will have
636 * explicit_location set. Second, we assume that any generic in or out
637 * will not have explicit_location set.
639 * This second assumption will only be valid until
640 * GL_ARB_separate_shader_objects is supported. When that extension is
641 * implemented, this function will need some modifications.
643 if (!var
->data
.explicit_location
) {
644 var
->data
.is_unmatched_generic_inout
= 1;
646 var
->data
.is_unmatched_generic_inout
= 0;
653 * Set UsesClipDistance and ClipDistanceArraySize based on the given shader.
655 * Also check for errors based on incorrect usage of gl_ClipVertex and
658 * Return false if an error was reported.
661 analyze_clip_usage(struct gl_shader_program
*prog
,
662 struct gl_shader
*shader
, GLboolean
*UsesClipDistance
,
663 GLuint
*ClipDistanceArraySize
)
665 *ClipDistanceArraySize
= 0;
667 if (!prog
->IsES
&& prog
->Version
>= 130) {
668 /* From section 7.1 (Vertex Shader Special Variables) of the
671 * "It is an error for a shader to statically write both
672 * gl_ClipVertex and gl_ClipDistance."
674 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
675 * gl_ClipVertex nor gl_ClipDistance.
677 find_assignment_visitor
clip_vertex("gl_ClipVertex");
678 find_assignment_visitor
clip_distance("gl_ClipDistance");
680 clip_vertex
.run(shader
->ir
);
681 clip_distance
.run(shader
->ir
);
682 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
683 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
684 "and `gl_ClipDistance'\n",
685 _mesa_shader_stage_to_string(shader
->Stage
));
688 *UsesClipDistance
= clip_distance
.variable_found();
689 ir_variable
*clip_distance_var
=
690 shader
->symbols
->get_variable("gl_ClipDistance");
691 if (clip_distance_var
)
692 *ClipDistanceArraySize
= clip_distance_var
->type
->length
;
694 *UsesClipDistance
= false;
700 * Verify that a vertex shader executable meets all semantic requirements.
702 * Also sets prog->Vert.UsesClipDistance and prog->Vert.ClipDistanceArraySize
705 * \param shader Vertex shader executable to be verified
708 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
709 struct gl_shader
*shader
)
714 /* From the GLSL 1.10 spec, page 48:
716 * "The variable gl_Position is available only in the vertex
717 * language and is intended for writing the homogeneous vertex
718 * position. All executions of a well-formed vertex shader
719 * executable must write a value into this variable. [...] The
720 * variable gl_Position is available only in the vertex
721 * language and is intended for writing the homogeneous vertex
722 * position. All executions of a well-formed vertex shader
723 * executable must write a value into this variable."
725 * while in GLSL 1.40 this text is changed to:
727 * "The variable gl_Position is available only in the vertex
728 * language and is intended for writing the homogeneous vertex
729 * position. It can be written at any time during shader
730 * execution. It may also be read back by a vertex shader
731 * after being written. This value will be used by primitive
732 * assembly, clipping, culling, and other fixed functionality
733 * operations, if present, that operate on primitives after
734 * vertex processing has occurred. Its value is undefined if
735 * the vertex shader executable does not write gl_Position."
737 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
738 * gl_Position is not an error.
740 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
741 find_assignment_visitor
find("gl_Position");
742 find
.run(shader
->ir
);
743 if (!find
.variable_found()) {
746 "vertex shader does not write to `gl_Position'."
747 "It's value is undefined. \n");
750 "vertex shader does not write to `gl_Position'. \n");
756 analyze_clip_usage(prog
, shader
, &prog
->Vert
.UsesClipDistance
,
757 &prog
->Vert
.ClipDistanceArraySize
);
761 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
762 struct gl_shader
*shader
)
767 analyze_clip_usage(prog
, shader
, &prog
->TessEval
.UsesClipDistance
,
768 &prog
->TessEval
.ClipDistanceArraySize
);
773 * Verify that a fragment shader executable meets all semantic requirements
775 * \param shader Fragment shader executable to be verified
778 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
779 struct gl_shader
*shader
)
784 find_assignment_visitor
frag_color("gl_FragColor");
785 find_assignment_visitor
frag_data("gl_FragData");
787 frag_color
.run(shader
->ir
);
788 frag_data
.run(shader
->ir
);
790 if (frag_color
.variable_found() && frag_data
.variable_found()) {
791 linker_error(prog
, "fragment shader writes to both "
792 "`gl_FragColor' and `gl_FragData'\n");
797 * Verify that a geometry shader executable meets all semantic requirements
799 * Also sets prog->Geom.VerticesIn, prog->Geom.UsesClipDistance, and
800 * prog->Geom.ClipDistanceArraySize as a side effect.
802 * \param shader Geometry shader executable to be verified
805 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
806 struct gl_shader
*shader
)
811 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
812 prog
->Geom
.VerticesIn
= num_vertices
;
814 analyze_clip_usage(prog
, shader
, &prog
->Geom
.UsesClipDistance
,
815 &prog
->Geom
.ClipDistanceArraySize
);
819 * Check if geometry shaders emit to non-zero streams and do corresponding
823 validate_geometry_shader_emissions(struct gl_context
*ctx
,
824 struct gl_shader_program
*prog
)
826 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
827 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
828 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
829 if (emit_vertex
.error()) {
830 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
831 "stream parameter are in the range [0, %d].\n",
832 emit_vertex
.error_func(),
833 emit_vertex
.error_stream(),
834 ctx
->Const
.MaxVertexStreams
- 1);
836 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
837 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
839 /* From the ARB_gpu_shader5 spec:
841 * "Multiple vertex streams are supported only if the output primitive
842 * type is declared to be "points". A program will fail to link if it
843 * contains a geometry shader calling EmitStreamVertex() or
844 * EndStreamPrimitive() if its output primitive type is not "points".
846 * However, in the same spec:
848 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
849 * with <stream> set to zero."
853 * "The function EndPrimitive() is equivalent to calling
854 * EndStreamPrimitive() with <stream> set to zero."
856 * Since we can call EmitVertex() and EndPrimitive() when we output
857 * primitives other than points, calling EmitStreamVertex(0) or
858 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
859 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
860 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
863 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
864 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
865 "with n>0 requires point output\n");
871 validate_intrastage_arrays(struct gl_shader_program
*prog
,
872 ir_variable
*const var
,
873 ir_variable
*const existing
)
875 /* Consider the types to be "the same" if both types are arrays
876 * of the same type and one of the arrays is implicitly sized.
877 * In addition, set the type of the linked variable to the
878 * explicitly sized array.
880 if (var
->type
->is_array() && existing
->type
->is_array() &&
881 (var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
882 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
883 if (var
->type
->length
!= 0) {
884 if (var
->type
->length
<= existing
->data
.max_array_access
) {
885 linker_error(prog
, "%s `%s' declared as type "
886 "`%s' but outermost dimension has an index"
889 var
->name
, var
->type
->name
,
890 existing
->data
.max_array_access
);
892 existing
->type
= var
->type
;
894 } else if (existing
->type
->length
!= 0) {
895 if(existing
->type
->length
<= var
->data
.max_array_access
) {
896 linker_error(prog
, "%s `%s' declared as type "
897 "`%s' but outermost dimension has an index"
900 var
->name
, existing
->type
->name
,
901 var
->data
.max_array_access
);
911 * Perform validation of global variables used across multiple shaders
914 cross_validate_globals(struct gl_shader_program
*prog
,
915 struct gl_shader
**shader_list
,
916 unsigned num_shaders
,
919 /* Examine all of the uniforms in all of the shaders and cross validate
922 glsl_symbol_table variables
;
923 for (unsigned i
= 0; i
< num_shaders
; i
++) {
924 if (shader_list
[i
] == NULL
)
927 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
928 ir_variable
*const var
= node
->as_variable();
933 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
936 /* don't cross validate subroutine uniforms */
937 if (var
->type
->contains_subroutine())
940 /* Don't cross validate temporaries that are at global scope. These
941 * will eventually get pulled into the shaders 'main'.
943 if (var
->data
.mode
== ir_var_temporary
)
946 /* If a global with this name has already been seen, verify that the
947 * new instance has the same type. In addition, if the globals have
948 * initializers, the values of the initializers must be the same.
950 ir_variable
*const existing
= variables
.get_variable(var
->name
);
951 if (existing
!= NULL
) {
952 /* Check if types match. Interface blocks have some special
953 * rules so we handle those elsewhere.
955 if (var
->type
!= existing
->type
&&
956 !var
->is_interface_instance()) {
957 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
958 if (var
->type
->is_record() && existing
->type
->is_record()
959 && existing
->type
->record_compare(var
->type
)) {
960 existing
->type
= var
->type
;
962 linker_error(prog
, "%s `%s' declared as type "
963 "`%s' and type `%s'\n",
965 var
->name
, var
->type
->name
,
966 existing
->type
->name
);
972 if (var
->data
.explicit_location
) {
973 if (existing
->data
.explicit_location
974 && (var
->data
.location
!= existing
->data
.location
)) {
975 linker_error(prog
, "explicit locations for %s "
976 "`%s' have differing values\n",
977 mode_string(var
), var
->name
);
981 existing
->data
.location
= var
->data
.location
;
982 existing
->data
.explicit_location
= true;
985 /* From the GLSL 4.20 specification:
986 * "A link error will result if two compilation units in a program
987 * specify different integer-constant bindings for the same
988 * opaque-uniform name. However, it is not an error to specify a
989 * binding on some but not all declarations for the same name"
991 if (var
->data
.explicit_binding
) {
992 if (existing
->data
.explicit_binding
&&
993 var
->data
.binding
!= existing
->data
.binding
) {
994 linker_error(prog
, "explicit bindings for %s "
995 "`%s' have differing values\n",
996 mode_string(var
), var
->name
);
1000 existing
->data
.binding
= var
->data
.binding
;
1001 existing
->data
.explicit_binding
= true;
1004 if (var
->type
->contains_atomic() &&
1005 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
1006 linker_error(prog
, "offset specifications for %s "
1007 "`%s' have differing values\n",
1008 mode_string(var
), var
->name
);
1012 /* Validate layout qualifiers for gl_FragDepth.
1014 * From the AMD/ARB_conservative_depth specs:
1016 * "If gl_FragDepth is redeclared in any fragment shader in a
1017 * program, it must be redeclared in all fragment shaders in
1018 * that program that have static assignments to
1019 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1020 * fragment shaders in a single program must have the same set
1023 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1024 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1025 bool layout_differs
=
1026 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1028 if (layout_declared
&& layout_differs
) {
1030 "All redeclarations of gl_FragDepth in all "
1031 "fragment shaders in a single program must have "
1032 "the same set of qualifiers.\n");
1035 if (var
->data
.used
&& layout_differs
) {
1037 "If gl_FragDepth is redeclared with a layout "
1038 "qualifier in any fragment shader, it must be "
1039 "redeclared with the same layout qualifier in "
1040 "all fragment shaders that have assignments to "
1045 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1047 * "If a shared global has multiple initializers, the
1048 * initializers must all be constant expressions, and they
1049 * must all have the same value. Otherwise, a link error will
1050 * result. (A shared global having only one initializer does
1051 * not require that initializer to be a constant expression.)"
1053 * Previous to 4.20 the GLSL spec simply said that initializers
1054 * must have the same value. In this case of non-constant
1055 * initializers, this was impossible to determine. As a result,
1056 * no vendor actually implemented that behavior. The 4.20
1057 * behavior matches the implemented behavior of at least one other
1058 * vendor, so we'll implement that for all GLSL versions.
1060 if (var
->constant_initializer
!= NULL
) {
1061 if (existing
->constant_initializer
!= NULL
) {
1062 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1063 linker_error(prog
, "initializers for %s "
1064 "`%s' have differing values\n",
1065 mode_string(var
), var
->name
);
1069 /* If the first-seen instance of a particular uniform did not
1070 * have an initializer but a later instance does, copy the
1071 * initializer to the version stored in the symbol table.
1073 /* FINISHME: This is wrong. The constant_value field should
1074 * FINISHME: not be modified! Imagine a case where a shader
1075 * FINISHME: without an initializer is linked in two different
1076 * FINISHME: programs with shaders that have differing
1077 * FINISHME: initializers. Linking with the first will
1078 * FINISHME: modify the shader, and linking with the second
1079 * FINISHME: will fail.
1081 existing
->constant_initializer
=
1082 var
->constant_initializer
->clone(ralloc_parent(existing
),
1087 if (var
->data
.has_initializer
) {
1088 if (existing
->data
.has_initializer
1089 && (var
->constant_initializer
== NULL
1090 || existing
->constant_initializer
== NULL
)) {
1092 "shared global variable `%s' has multiple "
1093 "non-constant initializers.\n",
1098 /* Some instance had an initializer, so keep track of that. In
1099 * this location, all sorts of initializers (constant or
1100 * otherwise) will propagate the existence to the variable
1101 * stored in the symbol table.
1103 existing
->data
.has_initializer
= true;
1106 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1107 linker_error(prog
, "declarations for %s `%s' have "
1108 "mismatching invariant qualifiers\n",
1109 mode_string(var
), var
->name
);
1112 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1113 linker_error(prog
, "declarations for %s `%s' have "
1114 "mismatching centroid qualifiers\n",
1115 mode_string(var
), var
->name
);
1118 if (existing
->data
.sample
!= var
->data
.sample
) {
1119 linker_error(prog
, "declarations for %s `%s` have "
1120 "mismatching sample qualifiers\n",
1121 mode_string(var
), var
->name
);
1125 variables
.add_variable(var
);
1132 * Perform validation of uniforms used across multiple shader stages
1135 cross_validate_uniforms(struct gl_shader_program
*prog
)
1137 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1138 MESA_SHADER_STAGES
, true);
1142 * Accumulates the array of prog->UniformBlocks and checks that all
1143 * definitons of blocks agree on their contents.
1146 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1148 unsigned max_num_uniform_blocks
= 0;
1149 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1150 if (prog
->_LinkedShaders
[i
])
1151 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1154 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1155 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1157 prog
->UniformBlockStageIndex
[i
] = ralloc_array(prog
, int,
1158 max_num_uniform_blocks
);
1159 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1160 prog
->UniformBlockStageIndex
[i
][j
] = -1;
1165 for (unsigned int j
= 0; j
< sh
->NumUniformBlocks
; j
++) {
1166 int index
= link_cross_validate_uniform_block(prog
,
1167 &prog
->UniformBlocks
,
1168 &prog
->NumUniformBlocks
,
1169 &sh
->UniformBlocks
[j
]);
1172 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1173 sh
->UniformBlocks
[j
].Name
);
1177 prog
->UniformBlockStageIndex
[i
][index
] = j
;
1186 * Populates a shaders symbol table with all global declarations
1189 populate_symbol_table(gl_shader
*sh
)
1191 sh
->symbols
= new(sh
) glsl_symbol_table
;
1193 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1197 if ((func
= inst
->as_function()) != NULL
) {
1198 sh
->symbols
->add_function(func
);
1199 } else if ((var
= inst
->as_variable()) != NULL
) {
1200 if (var
->data
.mode
!= ir_var_temporary
)
1201 sh
->symbols
->add_variable(var
);
1208 * Remap variables referenced in an instruction tree
1210 * This is used when instruction trees are cloned from one shader and placed in
1211 * another. These trees will contain references to \c ir_variable nodes that
1212 * do not exist in the target shader. This function finds these \c ir_variable
1213 * references and replaces the references with matching variables in the target
1216 * If there is no matching variable in the target shader, a clone of the
1217 * \c ir_variable is made and added to the target shader. The new variable is
1218 * added to \b both the instruction stream and the symbol table.
1220 * \param inst IR tree that is to be processed.
1221 * \param symbols Symbol table containing global scope symbols in the
1223 * \param instructions Instruction stream where new variable declarations
1227 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1230 class remap_visitor
: public ir_hierarchical_visitor
{
1232 remap_visitor(struct gl_shader
*target
,
1235 this->target
= target
;
1236 this->symbols
= target
->symbols
;
1237 this->instructions
= target
->ir
;
1238 this->temps
= temps
;
1241 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1243 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1244 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1246 assert(var
!= NULL
);
1248 return visit_continue
;
1251 ir_variable
*const existing
=
1252 this->symbols
->get_variable(ir
->var
->name
);
1253 if (existing
!= NULL
)
1256 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1258 this->symbols
->add_variable(copy
);
1259 this->instructions
->push_head(copy
);
1263 return visit_continue
;
1267 struct gl_shader
*target
;
1268 glsl_symbol_table
*symbols
;
1269 exec_list
*instructions
;
1273 remap_visitor
v(target
, temps
);
1280 * Move non-declarations from one instruction stream to another
1282 * The intended usage pattern of this function is to pass the pointer to the
1283 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1284 * pointer) for \c last and \c false for \c make_copies on the first
1285 * call. Successive calls pass the return value of the previous call for
1286 * \c last and \c true for \c make_copies.
1288 * \param instructions Source instruction stream
1289 * \param last Instruction after which new instructions should be
1290 * inserted in the target instruction stream
1291 * \param make_copies Flag selecting whether instructions in \c instructions
1292 * should be copied (via \c ir_instruction::clone) into the
1293 * target list or moved.
1296 * The new "last" instruction in the target instruction stream. This pointer
1297 * is suitable for use as the \c last parameter of a later call to this
1301 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1302 bool make_copies
, gl_shader
*target
)
1304 hash_table
*temps
= NULL
;
1307 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1308 hash_table_pointer_compare
);
1310 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1311 if (inst
->as_function())
1314 ir_variable
*var
= inst
->as_variable();
1315 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1318 assert(inst
->as_assignment()
1320 || inst
->as_if() /* for initializers with the ?: operator */
1321 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1324 inst
= inst
->clone(target
, NULL
);
1327 hash_table_insert(temps
, inst
, var
);
1329 remap_variables(inst
, target
, temps
);
1334 last
->insert_after(inst
);
1339 hash_table_dtor(temps
);
1346 * This class is only used in link_intrastage_shaders() below but declaring
1347 * it inside that function leads to compiler warnings with some versions of
1350 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1352 array_sizing_visitor()
1353 : mem_ctx(ralloc_context(NULL
)),
1354 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1355 hash_table_pointer_compare
))
1359 ~array_sizing_visitor()
1361 hash_table_dtor(this->unnamed_interfaces
);
1362 ralloc_free(this->mem_ctx
);
1365 virtual ir_visitor_status
visit(ir_variable
*var
)
1367 fixup_type(&var
->type
, var
->data
.max_array_access
);
1368 if (var
->type
->is_interface()) {
1369 if (interface_contains_unsized_arrays(var
->type
)) {
1370 const glsl_type
*new_type
=
1371 resize_interface_members(var
->type
,
1372 var
->get_max_ifc_array_access());
1373 var
->type
= new_type
;
1374 var
->change_interface_type(new_type
);
1376 } else if (var
->type
->is_array() &&
1377 var
->type
->fields
.array
->is_interface()) {
1378 if (interface_contains_unsized_arrays(var
->type
->fields
.array
)) {
1379 const glsl_type
*new_type
=
1380 resize_interface_members(var
->type
->fields
.array
,
1381 var
->get_max_ifc_array_access());
1382 var
->change_interface_type(new_type
);
1383 var
->type
= update_interface_members_array(var
->type
, new_type
);
1385 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1386 /* Store a pointer to the variable in the unnamed_interfaces
1389 ir_variable
**interface_vars
= (ir_variable
**)
1390 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1391 if (interface_vars
== NULL
) {
1392 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1394 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1397 unsigned index
= ifc_type
->field_index(var
->name
);
1398 assert(index
< ifc_type
->length
);
1399 assert(interface_vars
[index
] == NULL
);
1400 interface_vars
[index
] = var
;
1402 return visit_continue
;
1406 * For each unnamed interface block that was discovered while running the
1407 * visitor, adjust the interface type to reflect the newly assigned array
1408 * sizes, and fix up the ir_variable nodes to point to the new interface
1411 void fixup_unnamed_interface_types()
1413 hash_table_call_foreach(this->unnamed_interfaces
,
1414 fixup_unnamed_interface_type
, NULL
);
1419 * If the type pointed to by \c type represents an unsized array, replace
1420 * it with a sized array whose size is determined by max_array_access.
1422 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
)
1424 if ((*type
)->is_unsized_array()) {
1425 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1426 max_array_access
+ 1);
1427 assert(*type
!= NULL
);
1431 static const glsl_type
*
1432 update_interface_members_array(const glsl_type
*type
,
1433 const glsl_type
*new_interface_type
)
1435 const glsl_type
*element_type
= type
->fields
.array
;
1436 if (element_type
->is_array()) {
1437 const glsl_type
*new_array_type
=
1438 update_interface_members_array(element_type
, new_interface_type
);
1439 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1441 return glsl_type::get_array_instance(new_interface_type
,
1447 * Determine whether the given interface type contains unsized arrays (if
1448 * it doesn't, array_sizing_visitor doesn't need to process it).
1450 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1452 for (unsigned i
= 0; i
< type
->length
; i
++) {
1453 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1454 if (elem_type
->is_unsized_array())
1461 * Create a new interface type based on the given type, with unsized arrays
1462 * replaced by sized arrays whose size is determined by
1463 * max_ifc_array_access.
1465 static const glsl_type
*
1466 resize_interface_members(const glsl_type
*type
,
1467 const unsigned *max_ifc_array_access
)
1469 unsigned num_fields
= type
->length
;
1470 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1471 memcpy(fields
, type
->fields
.structure
,
1472 num_fields
* sizeof(*fields
));
1473 for (unsigned i
= 0; i
< num_fields
; i
++) {
1474 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
]);
1476 glsl_interface_packing packing
=
1477 (glsl_interface_packing
) type
->interface_packing
;
1478 const glsl_type
*new_ifc_type
=
1479 glsl_type::get_interface_instance(fields
, num_fields
,
1480 packing
, type
->name
);
1482 return new_ifc_type
;
1485 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1488 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1489 ir_variable
**interface_vars
= (ir_variable
**) data
;
1490 unsigned num_fields
= ifc_type
->length
;
1491 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1492 memcpy(fields
, ifc_type
->fields
.structure
,
1493 num_fields
* sizeof(*fields
));
1494 bool interface_type_changed
= false;
1495 for (unsigned i
= 0; i
< num_fields
; i
++) {
1496 if (interface_vars
[i
] != NULL
&&
1497 fields
[i
].type
!= interface_vars
[i
]->type
) {
1498 fields
[i
].type
= interface_vars
[i
]->type
;
1499 interface_type_changed
= true;
1502 if (!interface_type_changed
) {
1506 glsl_interface_packing packing
=
1507 (glsl_interface_packing
) ifc_type
->interface_packing
;
1508 const glsl_type
*new_ifc_type
=
1509 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1512 for (unsigned i
= 0; i
< num_fields
; i
++) {
1513 if (interface_vars
[i
] != NULL
)
1514 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1519 * Memory context used to allocate the data in \c unnamed_interfaces.
1524 * Hash table from const glsl_type * to an array of ir_variable *'s
1525 * pointing to the ir_variables constituting each unnamed interface block.
1527 hash_table
*unnamed_interfaces
;
1532 * Performs the cross-validation of tessellation control shader vertices and
1533 * layout qualifiers for the attached tessellation control shaders,
1534 * and propagates them to the linked TCS and linked shader program.
1537 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1538 struct gl_shader
*linked_shader
,
1539 struct gl_shader
**shader_list
,
1540 unsigned num_shaders
)
1542 linked_shader
->TessCtrl
.VerticesOut
= 0;
1544 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1547 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1549 * "All tessellation control shader layout declarations in a program
1550 * must specify the same output patch vertex count. There must be at
1551 * least one layout qualifier specifying an output patch vertex count
1552 * in any program containing tessellation control shaders; however,
1553 * such a declaration is not required in all tessellation control
1557 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1558 struct gl_shader
*shader
= shader_list
[i
];
1560 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1561 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1562 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1563 linker_error(prog
, "tessellation control shader defined with "
1564 "conflicting output vertex count (%d and %d)\n",
1565 linked_shader
->TessCtrl
.VerticesOut
,
1566 shader
->TessCtrl
.VerticesOut
);
1569 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1573 /* Just do the intrastage -> interstage propagation right now,
1574 * since we already know we're in the right type of shader program
1577 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1578 linker_error(prog
, "tessellation control shader didn't declare "
1579 "vertices out layout qualifier\n");
1582 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1587 * Performs the cross-validation of tessellation evaluation shader
1588 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1589 * for the attached tessellation evaluation shaders, and propagates them
1590 * to the linked TES and linked shader program.
1593 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1594 struct gl_shader
*linked_shader
,
1595 struct gl_shader
**shader_list
,
1596 unsigned num_shaders
)
1598 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1599 linked_shader
->TessEval
.Spacing
= 0;
1600 linked_shader
->TessEval
.VertexOrder
= 0;
1601 linked_shader
->TessEval
.PointMode
= -1;
1603 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1606 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1608 * "At least one tessellation evaluation shader (compilation unit) in
1609 * a program must declare a primitive mode in its input layout.
1610 * Declaration vertex spacing, ordering, and point mode identifiers is
1611 * optional. It is not required that all tessellation evaluation
1612 * shaders in a program declare a primitive mode. If spacing or
1613 * vertex ordering declarations are omitted, the tessellation
1614 * primitive generator will use equal spacing or counter-clockwise
1615 * vertex ordering, respectively. If a point mode declaration is
1616 * omitted, the tessellation primitive generator will produce lines or
1617 * triangles according to the primitive mode."
1620 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1621 struct gl_shader
*shader
= shader_list
[i
];
1623 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1624 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1625 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1626 linker_error(prog
, "tessellation evaluation shader defined with "
1627 "conflicting input primitive modes.\n");
1630 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1633 if (shader
->TessEval
.Spacing
!= 0) {
1634 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1635 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1636 linker_error(prog
, "tessellation evaluation shader defined with "
1637 "conflicting vertex spacing.\n");
1640 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1643 if (shader
->TessEval
.VertexOrder
!= 0) {
1644 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1645 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1646 linker_error(prog
, "tessellation evaluation shader defined with "
1647 "conflicting ordering.\n");
1650 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1653 if (shader
->TessEval
.PointMode
!= -1) {
1654 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1655 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1656 linker_error(prog
, "tessellation evaluation shader defined with "
1657 "conflicting point modes.\n");
1660 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1665 /* Just do the intrastage -> interstage propagation right now,
1666 * since we already know we're in the right type of shader program
1669 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1671 "tessellation evaluation shader didn't declare input "
1672 "primitive modes.\n");
1675 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1677 if (linked_shader
->TessEval
.Spacing
== 0)
1678 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1679 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1681 if (linked_shader
->TessEval
.VertexOrder
== 0)
1682 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1683 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1685 if (linked_shader
->TessEval
.PointMode
== -1)
1686 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1687 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1692 * Performs the cross-validation of layout qualifiers specified in
1693 * redeclaration of gl_FragCoord for the attached fragment shaders,
1694 * and propagates them to the linked FS and linked shader program.
1697 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1698 struct gl_shader
*linked_shader
,
1699 struct gl_shader
**shader_list
,
1700 unsigned num_shaders
)
1702 linked_shader
->redeclares_gl_fragcoord
= false;
1703 linked_shader
->uses_gl_fragcoord
= false;
1704 linked_shader
->origin_upper_left
= false;
1705 linked_shader
->pixel_center_integer
= false;
1707 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1708 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1711 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1712 struct gl_shader
*shader
= shader_list
[i
];
1713 /* From the GLSL 1.50 spec, page 39:
1715 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1716 * it must be redeclared in all the fragment shaders in that program
1717 * that have a static use gl_FragCoord."
1719 if ((linked_shader
->redeclares_gl_fragcoord
1720 && !shader
->redeclares_gl_fragcoord
1721 && shader
->uses_gl_fragcoord
)
1722 || (shader
->redeclares_gl_fragcoord
1723 && !linked_shader
->redeclares_gl_fragcoord
1724 && linked_shader
->uses_gl_fragcoord
)) {
1725 linker_error(prog
, "fragment shader defined with conflicting "
1726 "layout qualifiers for gl_FragCoord\n");
1729 /* From the GLSL 1.50 spec, page 39:
1731 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1732 * single program must have the same set of qualifiers."
1734 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1735 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1736 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1737 linker_error(prog
, "fragment shader defined with conflicting "
1738 "layout qualifiers for gl_FragCoord\n");
1741 /* Update the linked shader state. Note that uses_gl_fragcoord should
1742 * accumulate the results. The other values should replace. If there
1743 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1744 * are already known to be the same.
1746 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1747 linked_shader
->redeclares_gl_fragcoord
=
1748 shader
->redeclares_gl_fragcoord
;
1749 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1750 || shader
->uses_gl_fragcoord
;
1751 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1752 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1755 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1760 * Performs the cross-validation of geometry shader max_vertices and
1761 * primitive type layout qualifiers for the attached geometry shaders,
1762 * and propagates them to the linked GS and linked shader program.
1765 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1766 struct gl_shader
*linked_shader
,
1767 struct gl_shader
**shader_list
,
1768 unsigned num_shaders
)
1770 linked_shader
->Geom
.VerticesOut
= 0;
1771 linked_shader
->Geom
.Invocations
= 0;
1772 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1773 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1775 /* No in/out qualifiers defined for anything but GLSL 1.50+
1776 * geometry shaders so far.
1778 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1781 /* From the GLSL 1.50 spec, page 46:
1783 * "All geometry shader output layout declarations in a program
1784 * must declare the same layout and same value for
1785 * max_vertices. There must be at least one geometry output
1786 * layout declaration somewhere in a program, but not all
1787 * geometry shaders (compilation units) are required to
1791 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1792 struct gl_shader
*shader
= shader_list
[i
];
1794 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1795 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1796 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1797 linker_error(prog
, "geometry shader defined with conflicting "
1801 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1804 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1805 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1806 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1807 linker_error(prog
, "geometry shader defined with conflicting "
1811 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1814 if (shader
->Geom
.VerticesOut
!= 0) {
1815 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1816 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1817 linker_error(prog
, "geometry shader defined with conflicting "
1818 "output vertex count (%d and %d)\n",
1819 linked_shader
->Geom
.VerticesOut
,
1820 shader
->Geom
.VerticesOut
);
1823 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1826 if (shader
->Geom
.Invocations
!= 0) {
1827 if (linked_shader
->Geom
.Invocations
!= 0 &&
1828 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1829 linker_error(prog
, "geometry shader defined with conflicting "
1830 "invocation count (%d and %d)\n",
1831 linked_shader
->Geom
.Invocations
,
1832 shader
->Geom
.Invocations
);
1835 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1839 /* Just do the intrastage -> interstage propagation right now,
1840 * since we already know we're in the right type of shader program
1843 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1845 "geometry shader didn't declare primitive input type\n");
1848 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1850 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1852 "geometry shader didn't declare primitive output type\n");
1855 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1857 if (linked_shader
->Geom
.VerticesOut
== 0) {
1859 "geometry shader didn't declare max_vertices\n");
1862 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1864 if (linked_shader
->Geom
.Invocations
== 0)
1865 linked_shader
->Geom
.Invocations
= 1;
1867 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1872 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1873 * qualifiers for the attached compute shaders, and propagate them to the
1874 * linked CS and linked shader program.
1877 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1878 struct gl_shader
*linked_shader
,
1879 struct gl_shader
**shader_list
,
1880 unsigned num_shaders
)
1882 for (int i
= 0; i
< 3; i
++)
1883 linked_shader
->Comp
.LocalSize
[i
] = 0;
1885 /* This function is called for all shader stages, but it only has an effect
1886 * for compute shaders.
1888 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1891 /* From the ARB_compute_shader spec, in the section describing local size
1894 * If multiple compute shaders attached to a single program object
1895 * declare local work-group size, the declarations must be identical;
1896 * otherwise a link-time error results. Furthermore, if a program
1897 * object contains any compute shaders, at least one must contain an
1898 * input layout qualifier specifying the local work sizes of the
1899 * program, or a link-time error will occur.
1901 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1902 struct gl_shader
*shader
= shader_list
[sh
];
1904 if (shader
->Comp
.LocalSize
[0] != 0) {
1905 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1906 for (int i
= 0; i
< 3; i
++) {
1907 if (linked_shader
->Comp
.LocalSize
[i
] !=
1908 shader
->Comp
.LocalSize
[i
]) {
1909 linker_error(prog
, "compute shader defined with conflicting "
1915 for (int i
= 0; i
< 3; i
++)
1916 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1920 /* Just do the intrastage -> interstage propagation right now,
1921 * since we already know we're in the right type of shader program
1924 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1925 linker_error(prog
, "compute shader didn't declare local size\n");
1928 for (int i
= 0; i
< 3; i
++)
1929 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1934 * Combine a group of shaders for a single stage to generate a linked shader
1937 * If this function is supplied a single shader, it is cloned, and the new
1938 * shader is returned.
1940 static struct gl_shader
*
1941 link_intrastage_shaders(void *mem_ctx
,
1942 struct gl_context
*ctx
,
1943 struct gl_shader_program
*prog
,
1944 struct gl_shader
**shader_list
,
1945 unsigned num_shaders
)
1947 struct gl_uniform_block
*uniform_blocks
= NULL
;
1949 /* Check that global variables defined in multiple shaders are consistent.
1951 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1952 if (!prog
->LinkStatus
)
1955 /* Check that interface blocks defined in multiple shaders are consistent.
1957 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1959 if (!prog
->LinkStatus
)
1962 /* Link up uniform blocks defined within this stage. */
1963 const unsigned num_uniform_blocks
=
1964 link_uniform_blocks(mem_ctx
, prog
, shader_list
, num_shaders
,
1966 if (!prog
->LinkStatus
)
1969 /* Check that there is only a single definition of each function signature
1970 * across all shaders.
1972 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
1973 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
1974 ir_function
*const f
= node
->as_function();
1979 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
1980 ir_function
*const other
=
1981 shader_list
[j
]->symbols
->get_function(f
->name
);
1983 /* If the other shader has no function (and therefore no function
1984 * signatures) with the same name, skip to the next shader.
1989 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
1990 if (!sig
->is_defined
|| sig
->is_builtin())
1993 ir_function_signature
*other_sig
=
1994 other
->exact_matching_signature(NULL
, &sig
->parameters
);
1996 if ((other_sig
!= NULL
) && other_sig
->is_defined
1997 && !other_sig
->is_builtin()) {
1998 linker_error(prog
, "function `%s' is multiply defined\n",
2007 /* Find the shader that defines main, and make a clone of it.
2009 * Starting with the clone, search for undefined references. If one is
2010 * found, find the shader that defines it. Clone the reference and add
2011 * it to the shader. Repeat until there are no undefined references or
2012 * until a reference cannot be resolved.
2014 gl_shader
*main
= NULL
;
2015 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2016 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2017 main
= shader_list
[i
];
2023 linker_error(prog
, "%s shader lacks `main'\n",
2024 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2028 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2029 linked
->ir
= new(linked
) exec_list
;
2030 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2032 linked
->UniformBlocks
= uniform_blocks
;
2033 linked
->NumUniformBlocks
= num_uniform_blocks
;
2034 ralloc_steal(linked
, linked
->UniformBlocks
);
2036 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2037 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2038 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2039 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2040 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2042 populate_symbol_table(linked
);
2044 /* The pointer to the main function in the final linked shader (i.e., the
2045 * copy of the original shader that contained the main function).
2047 ir_function_signature
*const main_sig
=
2048 _mesa_get_main_function_signature(linked
);
2050 /* Move any instructions other than variable declarations or function
2051 * declarations into main.
2053 exec_node
*insertion_point
=
2054 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2057 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2058 if (shader_list
[i
] == main
)
2061 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2062 insertion_point
, true, linked
);
2065 /* Check if any shader needs built-in functions. */
2066 bool need_builtins
= false;
2067 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2068 if (shader_list
[i
]->uses_builtin_functions
) {
2069 need_builtins
= true;
2075 if (need_builtins
) {
2076 /* Make a temporary array one larger than shader_list, which will hold
2077 * the built-in function shader as well.
2079 gl_shader
**linking_shaders
= (gl_shader
**)
2080 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2082 ok
= linking_shaders
!= NULL
;
2085 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2086 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2088 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2090 free(linking_shaders
);
2092 _mesa_error_no_memory(__func__
);
2095 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2100 ctx
->Driver
.DeleteShader(ctx
, linked
);
2104 /* At this point linked should contain all of the linked IR, so
2105 * validate it to make sure nothing went wrong.
2107 validate_ir_tree(linked
->ir
);
2109 /* Set the size of geometry shader input arrays */
2110 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2111 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2112 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2113 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2114 ir
->accept(&input_resize_visitor
);
2118 if (ctx
->Const
.VertexID_is_zero_based
)
2119 lower_vertex_id(linked
);
2121 /* Validate correct usage of barrier() in the tess control shader */
2122 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2123 barrier_use_visitor
visitor(prog
);
2124 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2125 ir
->accept(&visitor
);
2129 /* Make a pass over all variable declarations to ensure that arrays with
2130 * unspecified sizes have a size specified. The size is inferred from the
2131 * max_array_access field.
2133 array_sizing_visitor v
;
2135 v
.fixup_unnamed_interface_types();
2141 * Update the sizes of linked shader uniform arrays to the maximum
2144 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2146 * If one or more elements of an array are active,
2147 * GetActiveUniform will return the name of the array in name,
2148 * subject to the restrictions listed above. The type of the array
2149 * is returned in type. The size parameter contains the highest
2150 * array element index used, plus one. The compiler or linker
2151 * determines the highest index used. There will be only one
2152 * active uniform reported by the GL per uniform array.
2156 update_array_sizes(struct gl_shader_program
*prog
)
2158 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2159 if (prog
->_LinkedShaders
[i
] == NULL
)
2162 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2163 ir_variable
*const var
= node
->as_variable();
2165 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2166 !var
->type
->is_array())
2169 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2170 * will not be eliminated. Since we always do std140, just
2171 * don't resize arrays in UBOs.
2173 * Atomic counters are supposed to get deterministic
2174 * locations assigned based on the declaration ordering and
2175 * sizes, array compaction would mess that up.
2177 * Subroutine uniforms are not removed.
2179 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2180 var
->type
->contains_subroutine())
2183 unsigned int size
= var
->data
.max_array_access
;
2184 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2185 if (prog
->_LinkedShaders
[j
] == NULL
)
2188 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2189 ir_variable
*other_var
= node2
->as_variable();
2193 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2194 other_var
->data
.max_array_access
> size
) {
2195 size
= other_var
->data
.max_array_access
;
2200 if (size
+ 1 != var
->type
->length
) {
2201 /* If this is a built-in uniform (i.e., it's backed by some
2202 * fixed-function state), adjust the number of state slots to
2203 * match the new array size. The number of slots per array entry
2204 * is not known. It seems safe to assume that the total number of
2205 * slots is an integer multiple of the number of array elements.
2206 * Determine the number of slots per array element by dividing by
2207 * the old (total) size.
2209 const unsigned num_slots
= var
->get_num_state_slots();
2210 if (num_slots
> 0) {
2211 var
->set_num_state_slots((size
+ 1)
2212 * (num_slots
/ var
->type
->length
));
2215 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2217 /* FINISHME: We should update the types of array
2218 * dereferences of this variable now.
2226 * Resize tessellation evaluation per-vertex inputs to the size of
2227 * tessellation control per-vertex outputs.
2230 resize_tes_inputs(struct gl_context
*ctx
,
2231 struct gl_shader_program
*prog
)
2233 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2236 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2237 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2239 /* If no control shader is present, then the TES inputs are statically
2240 * sized to MaxPatchVertices; the actual size of the arrays won't be
2241 * known until draw time.
2243 const int num_vertices
= tcs
2244 ? tcs
->TessCtrl
.VerticesOut
2245 : ctx
->Const
.MaxPatchVertices
;
2247 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2248 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2249 ir
->accept(&input_resize_visitor
);
2254 * Find a contiguous set of available bits in a bitmask.
2256 * \param used_mask Bits representing used (1) and unused (0) locations
2257 * \param needed_count Number of contiguous bits needed.
2260 * Base location of the available bits on success or -1 on failure.
2263 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2265 unsigned needed_mask
= (1 << needed_count
) - 1;
2266 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2268 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2269 * cannot optimize possibly infinite loops" for the loop below.
2271 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2274 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2275 if ((needed_mask
& ~used_mask
) == needed_mask
)
2286 * Assign locations for either VS inputs or FS outputs
2288 * \param prog Shader program whose variables need locations assigned
2289 * \param constants Driver specific constant values for the program.
2290 * \param target_index Selector for the program target to receive location
2291 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2292 * \c MESA_SHADER_FRAGMENT.
2295 * If locations are successfully assigned, true is returned. Otherwise an
2296 * error is emitted to the shader link log and false is returned.
2299 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2300 struct gl_constants
*constants
,
2301 unsigned target_index
)
2303 /* Maximum number of generic locations. This corresponds to either the
2304 * maximum number of draw buffers or the maximum number of generic
2307 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2308 constants
->Program
[target_index
].MaxAttribs
:
2309 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2311 /* Mark invalid locations as being used.
2313 unsigned used_locations
= (max_index
>= 32)
2314 ? ~0 : ~((1 << max_index
) - 1);
2315 unsigned double_storage_locations
= 0;
2317 assert((target_index
== MESA_SHADER_VERTEX
)
2318 || (target_index
== MESA_SHADER_FRAGMENT
));
2320 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2324 /* Operate in a total of four passes.
2326 * 1. Invalidate the location assignments for all vertex shader inputs.
2328 * 2. Assign locations for inputs that have user-defined (via
2329 * glBindVertexAttribLocation) locations and outputs that have
2330 * user-defined locations (via glBindFragDataLocation).
2332 * 3. Sort the attributes without assigned locations by number of slots
2333 * required in decreasing order. Fragmentation caused by attribute
2334 * locations assigned by the application may prevent large attributes
2335 * from having enough contiguous space.
2337 * 4. Assign locations to any inputs without assigned locations.
2340 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2341 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2343 const enum ir_variable_mode direction
=
2344 (target_index
== MESA_SHADER_VERTEX
)
2345 ? ir_var_shader_in
: ir_var_shader_out
;
2348 /* Temporary storage for the set of attributes that need locations assigned.
2354 /* Used below in the call to qsort. */
2355 static int compare(const void *a
, const void *b
)
2357 const temp_attr
*const l
= (const temp_attr
*) a
;
2358 const temp_attr
*const r
= (const temp_attr
*) b
;
2360 /* Reversed because we want a descending order sort below. */
2361 return r
->slots
- l
->slots
;
2365 unsigned num_attr
= 0;
2367 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2368 ir_variable
*const var
= node
->as_variable();
2370 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2373 if (var
->data
.explicit_location
) {
2374 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2375 || (var
->data
.location
< 0)) {
2377 "invalid explicit location %d specified for `%s'\n",
2378 (var
->data
.location
< 0)
2379 ? var
->data
.location
2380 : var
->data
.location
- generic_base
,
2384 } else if (target_index
== MESA_SHADER_VERTEX
) {
2387 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2388 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2389 var
->data
.location
= binding
;
2390 var
->data
.is_unmatched_generic_inout
= 0;
2392 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2396 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2397 assert(binding
>= FRAG_RESULT_DATA0
);
2398 var
->data
.location
= binding
;
2399 var
->data
.is_unmatched_generic_inout
= 0;
2401 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2402 var
->data
.index
= index
;
2407 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2409 * "Output binding assignments will cause LinkProgram to fail:
2411 * If the program has an active output assigned to a location greater
2412 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2413 * an active output assigned an index greater than or equal to one;"
2415 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2416 var
->data
.location
- generic_base
>=
2417 (int) constants
->MaxDualSourceDrawBuffers
) {
2419 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2420 "with index %u for %s\n",
2421 var
->data
.location
- generic_base
, var
->data
.index
,
2426 const unsigned slots
= var
->type
->count_attribute_slots();
2428 /* If the variable is not a built-in and has a location statically
2429 * assigned in the shader (presumably via a layout qualifier), make sure
2430 * that it doesn't collide with other assigned locations. Otherwise,
2431 * add it to the list of variables that need linker-assigned locations.
2433 if (var
->data
.location
!= -1) {
2434 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2435 /* From page 61 of the OpenGL 4.0 spec:
2437 * "LinkProgram will fail if the attribute bindings assigned
2438 * by BindAttribLocation do not leave not enough space to
2439 * assign a location for an active matrix attribute or an
2440 * active attribute array, both of which require multiple
2441 * contiguous generic attributes."
2443 * I think above text prohibits the aliasing of explicit and
2444 * automatic assignments. But, aliasing is allowed in manual
2445 * assignments of attribute locations. See below comments for
2448 * From OpenGL 4.0 spec, page 61:
2450 * "It is possible for an application to bind more than one
2451 * attribute name to the same location. This is referred to as
2452 * aliasing. This will only work if only one of the aliased
2453 * attributes is active in the executable program, or if no
2454 * path through the shader consumes more than one attribute of
2455 * a set of attributes aliased to the same location. A link
2456 * error can occur if the linker determines that every path
2457 * through the shader consumes multiple aliased attributes,
2458 * but implementations are not required to generate an error
2461 * From GLSL 4.30 spec, page 54:
2463 * "A program will fail to link if any two non-vertex shader
2464 * input variables are assigned to the same location. For
2465 * vertex shaders, multiple input variables may be assigned
2466 * to the same location using either layout qualifiers or via
2467 * the OpenGL API. However, such aliasing is intended only to
2468 * support vertex shaders where each execution path accesses
2469 * at most one input per each location. Implementations are
2470 * permitted, but not required, to generate link-time errors
2471 * if they detect that every path through the vertex shader
2472 * executable accesses multiple inputs assigned to any single
2473 * location. For all shader types, a program will fail to link
2474 * if explicit location assignments leave the linker unable
2475 * to find space for other variables without explicit
2478 * From OpenGL ES 3.0 spec, page 56:
2480 * "Binding more than one attribute name to the same location
2481 * is referred to as aliasing, and is not permitted in OpenGL
2482 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2483 * fail when this condition exists. However, aliasing is
2484 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2485 * This will only work if only one of the aliased attributes
2486 * is active in the executable program, or if no path through
2487 * the shader consumes more than one attribute of a set of
2488 * attributes aliased to the same location. A link error can
2489 * occur if the linker determines that every path through the
2490 * shader consumes multiple aliased attributes, but implemen-
2491 * tations are not required to generate an error in this case."
2493 * After looking at above references from OpenGL, OpenGL ES and
2494 * GLSL specifications, we allow aliasing of vertex input variables
2495 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2497 * NOTE: This is not required by the spec but its worth mentioning
2498 * here that we're not doing anything to make sure that no path
2499 * through the vertex shader executable accesses multiple inputs
2500 * assigned to any single location.
2503 /* Mask representing the contiguous slots that will be used by
2506 const unsigned attr
= var
->data
.location
- generic_base
;
2507 const unsigned use_mask
= (1 << slots
) - 1;
2508 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2509 ? "vertex shader input" : "fragment shader output";
2511 /* Generate a link error if the requested locations for this
2512 * attribute exceed the maximum allowed attribute location.
2514 if (attr
+ slots
> max_index
) {
2516 "insufficient contiguous locations "
2517 "available for %s `%s' %d %d %d\n", string
,
2518 var
->name
, used_locations
, use_mask
, attr
);
2522 /* Generate a link error if the set of bits requested for this
2523 * attribute overlaps any previously allocated bits.
2525 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2526 if (target_index
== MESA_SHADER_FRAGMENT
||
2527 (prog
->IsES
&& prog
->Version
>= 300)) {
2529 "overlapping location is assigned "
2530 "to %s `%s' %d %d %d\n", string
,
2531 var
->name
, used_locations
, use_mask
, attr
);
2534 linker_warning(prog
,
2535 "overlapping location is assigned "
2536 "to %s `%s' %d %d %d\n", string
,
2537 var
->name
, used_locations
, use_mask
, attr
);
2541 used_locations
|= (use_mask
<< attr
);
2543 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2545 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2546 * active attribute variables may fail to link, unless
2547 * device-dependent optimizations are able to make the program
2548 * fit within available hardware resources. For the purposes
2549 * of this test, attribute variables of the type dvec3, dvec4,
2550 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2551 * count as consuming twice as many attributes as equivalent
2552 * single-precision types. While these types use the same number
2553 * of generic attributes as their single-precision equivalents,
2554 * implementations are permitted to consume two single-precision
2555 * vectors of internal storage for each three- or four-component
2556 * double-precision vector."
2558 * Mark this attribute slot as taking up twice as much space
2559 * so we can count it properly against limits. According to
2560 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2561 * is optional behavior, but it seems preferable.
2563 const glsl_type
*type
= var
->type
->without_array();
2564 if (type
== glsl_type::dvec3_type
||
2565 type
== glsl_type::dvec4_type
||
2566 type
== glsl_type::dmat2x3_type
||
2567 type
== glsl_type::dmat2x4_type
||
2568 type
== glsl_type::dmat3_type
||
2569 type
== glsl_type::dmat3x4_type
||
2570 type
== glsl_type::dmat4x3_type
||
2571 type
== glsl_type::dmat4_type
) {
2572 double_storage_locations
|= (use_mask
<< attr
);
2579 to_assign
[num_attr
].slots
= slots
;
2580 to_assign
[num_attr
].var
= var
;
2584 if (target_index
== MESA_SHADER_VERTEX
) {
2585 unsigned total_attribs_size
=
2586 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2587 _mesa_bitcount(double_storage_locations
);
2588 if (total_attribs_size
> max_index
) {
2590 "attempt to use %d vertex attribute slots only %d available ",
2591 total_attribs_size
, max_index
);
2596 /* If all of the attributes were assigned locations by the application (or
2597 * are built-in attributes with fixed locations), return early. This should
2598 * be the common case.
2603 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2605 if (target_index
== MESA_SHADER_VERTEX
) {
2606 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2607 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2608 * reserved to prevent it from being automatically allocated below.
2610 find_deref_visitor
find("gl_Vertex");
2612 if (find
.variable_found())
2613 used_locations
|= (1 << 0);
2616 for (unsigned i
= 0; i
< num_attr
; i
++) {
2617 /* Mask representing the contiguous slots that will be used by this
2620 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2622 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2625 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2626 ? "vertex shader input" : "fragment shader output";
2629 "insufficient contiguous locations "
2630 "available for %s `%s'\n",
2631 string
, to_assign
[i
].var
->name
);
2635 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2636 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2637 used_locations
|= (use_mask
<< location
);
2645 * Demote shader inputs and outputs that are not used in other stages
2648 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2650 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2651 ir_variable
*const var
= node
->as_variable();
2653 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2656 /* A shader 'in' or 'out' variable is only really an input or output if
2657 * its value is used by other shader stages. This will cause the variable
2658 * to have a location assigned.
2660 if (var
->data
.is_unmatched_generic_inout
) {
2661 assert(var
->data
.mode
!= ir_var_temporary
);
2662 var
->data
.mode
= ir_var_auto
;
2669 * Store the gl_FragDepth layout in the gl_shader_program struct.
2672 store_fragdepth_layout(struct gl_shader_program
*prog
)
2674 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2678 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2680 /* We don't look up the gl_FragDepth symbol directly because if
2681 * gl_FragDepth is not used in the shader, it's removed from the IR.
2682 * However, the symbol won't be removed from the symbol table.
2684 * We're only interested in the cases where the variable is NOT removed
2687 foreach_in_list(ir_instruction
, node
, ir
) {
2688 ir_variable
*const var
= node
->as_variable();
2690 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2694 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2695 switch (var
->data
.depth_layout
) {
2696 case ir_depth_layout_none
:
2697 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2699 case ir_depth_layout_any
:
2700 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2702 case ir_depth_layout_greater
:
2703 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2705 case ir_depth_layout_less
:
2706 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2708 case ir_depth_layout_unchanged
:
2709 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2720 * Validate the resources used by a program versus the implementation limits
2723 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2725 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2726 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2731 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2732 linker_error(prog
, "Too many %s shader texture samplers\n",
2733 _mesa_shader_stage_to_string(i
));
2736 if (sh
->num_uniform_components
>
2737 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2738 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2739 linker_warning(prog
, "Too many %s shader default uniform block "
2740 "components, but the driver will try to optimize "
2741 "them out; this is non-portable out-of-spec "
2743 _mesa_shader_stage_to_string(i
));
2745 linker_error(prog
, "Too many %s shader default uniform block "
2747 _mesa_shader_stage_to_string(i
));
2751 if (sh
->num_combined_uniform_components
>
2752 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2753 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2754 linker_warning(prog
, "Too many %s shader uniform components, "
2755 "but the driver will try to optimize them out; "
2756 "this is non-portable out-of-spec behavior\n",
2757 _mesa_shader_stage_to_string(i
));
2759 linker_error(prog
, "Too many %s shader uniform components\n",
2760 _mesa_shader_stage_to_string(i
));
2765 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2766 unsigned total_uniform_blocks
= 0;
2768 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
2769 if (prog
->UniformBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2770 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2771 prog
->UniformBlocks
[i
].Name
,
2772 prog
->UniformBlocks
[i
].UniformBufferSize
,
2773 ctx
->Const
.MaxUniformBlockSize
);
2776 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2777 if (prog
->UniformBlockStageIndex
[j
][i
] != -1) {
2779 total_uniform_blocks
++;
2783 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2784 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2785 prog
->NumUniformBlocks
,
2786 ctx
->Const
.MaxCombinedUniformBlocks
);
2788 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2789 const unsigned max_uniform_blocks
=
2790 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2791 if (blocks
[i
] > max_uniform_blocks
) {
2792 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2793 _mesa_shader_stage_to_string(i
),
2795 max_uniform_blocks
);
2804 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
2806 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2807 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2812 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
2813 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
2819 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
2820 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
2821 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
2822 if (fn
->types
[k
] == uni
->type
) {
2828 uni
->num_compatible_subroutines
= count
;
2834 check_subroutine_resources(struct gl_shader_program
*prog
)
2836 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2837 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2840 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
2841 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
2842 _mesa_shader_stage_to_string(i
));
2847 * Validate shader image resources.
2850 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2852 unsigned total_image_units
= 0;
2853 unsigned fragment_outputs
= 0;
2855 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2858 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2859 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2862 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2863 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
2864 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
2865 ctx
->Const
.Program
[i
].MaxImageUniforms
);
2867 total_image_units
+= sh
->NumImages
;
2869 if (i
== MESA_SHADER_FRAGMENT
) {
2870 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2871 ir_variable
*var
= node
->as_variable();
2872 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2873 fragment_outputs
+= var
->type
->count_attribute_slots();
2879 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2880 linker_error(prog
, "Too many combined image uniforms\n");
2882 if (total_image_units
+ fragment_outputs
>
2883 ctx
->Const
.MaxCombinedShaderOutputResources
)
2884 linker_error(prog
, "Too many combined image uniforms and fragment outputs\n");
2889 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2890 * for a variable, checks for overlaps between other uniforms using explicit
2894 reserve_explicit_locations(struct gl_shader_program
*prog
,
2895 string_to_uint_map
*map
, ir_variable
*var
)
2897 unsigned slots
= var
->type
->uniform_locations();
2898 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2900 /* Resize remap table if locations do not fit in the current one. */
2901 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2902 prog
->UniformRemapTable
=
2903 reralloc(prog
, prog
->UniformRemapTable
,
2904 gl_uniform_storage
*,
2907 if (!prog
->UniformRemapTable
) {
2908 linker_error(prog
, "Out of memory during linking.\n");
2912 /* Initialize allocated space. */
2913 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
2914 prog
->UniformRemapTable
[i
] = NULL
;
2916 prog
->NumUniformRemapTable
= max_loc
+ 1;
2919 for (unsigned i
= 0; i
< slots
; i
++) {
2920 unsigned loc
= var
->data
.location
+ i
;
2922 /* Check if location is already used. */
2923 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2925 /* Possibly same uniform from a different stage, this is ok. */
2927 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
2930 /* ARB_explicit_uniform_location specification states:
2932 * "No two default-block uniform variables in the program can have
2933 * the same location, even if they are unused, otherwise a compiler
2934 * or linker error will be generated."
2937 "location qualifier for uniform %s overlaps "
2938 "previously used location\n",
2943 /* Initialize location as inactive before optimization
2944 * rounds and location assignment.
2946 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
2949 /* Note, base location used for arrays. */
2950 map
->put(var
->data
.location
, var
->name
);
2956 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
2957 struct gl_shader
*sh
,
2960 unsigned slots
= var
->type
->uniform_locations();
2961 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2963 /* Resize remap table if locations do not fit in the current one. */
2964 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
2965 sh
->SubroutineUniformRemapTable
=
2966 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
2967 gl_uniform_storage
*,
2970 if (!sh
->SubroutineUniformRemapTable
) {
2971 linker_error(prog
, "Out of memory during linking.\n");
2975 /* Initialize allocated space. */
2976 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
2977 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
2979 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
2982 for (unsigned i
= 0; i
< slots
; i
++) {
2983 unsigned loc
= var
->data
.location
+ i
;
2985 /* Check if location is already used. */
2986 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
2988 /* ARB_explicit_uniform_location specification states:
2989 * "No two subroutine uniform variables can have the same location
2990 * in the same shader stage, otherwise a compiler or linker error
2991 * will be generated."
2994 "location qualifier for uniform %s overlaps "
2995 "previously used location\n",
3000 /* Initialize location as inactive before optimization
3001 * rounds and location assignment.
3003 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3009 * Check and reserve all explicit uniform locations, called before
3010 * any optimizations happen to handle also inactive uniforms and
3011 * inactive array elements that may get trimmed away.
3014 check_explicit_uniform_locations(struct gl_context
*ctx
,
3015 struct gl_shader_program
*prog
)
3017 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3020 /* This map is used to detect if overlapping explicit locations
3021 * occur with the same uniform (from different stage) or a different one.
3023 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3026 linker_error(prog
, "Out of memory during linking.\n");
3030 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3031 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3036 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3037 ir_variable
*var
= node
->as_variable();
3038 if (var
&& (var
->data
.mode
== ir_var_uniform
|| var
->data
.mode
== ir_var_shader_storage
) &&
3039 var
->data
.explicit_location
) {
3041 if (var
->type
->is_subroutine())
3042 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3044 ret
= reserve_explicit_locations(prog
, uniform_map
, var
);
3057 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3058 const void *data
, uint8_t stages
)
3062 /* If resource already exists, do not add it again. */
3063 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3064 if (prog
->ProgramResourceList
[i
].Data
== data
)
3067 prog
->ProgramResourceList
=
3069 prog
->ProgramResourceList
,
3070 gl_program_resource
,
3071 prog
->NumProgramResourceList
+ 1);
3073 if (!prog
->ProgramResourceList
) {
3074 linker_error(prog
, "Out of memory during linking.\n");
3078 struct gl_program_resource
*res
=
3079 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3083 res
->StageReferences
= stages
;
3085 prog
->NumProgramResourceList
++;
3091 * Function builds a stage reference bitmask from variable name.
3094 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3099 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3100 * used for reference mask in gl_program_resource will need to be changed.
3102 assert(MESA_SHADER_STAGES
< 8);
3104 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3105 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3109 /* Shader symbol table may contain variables that have
3110 * been optimized away. Search IR for the variable instead.
3112 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3113 ir_variable
*var
= node
->as_variable();
3115 unsigned baselen
= strlen(var
->name
);
3117 /* Type needs to match if specified, otherwise we might
3118 * pick a variable with same name but different interface.
3120 if (var
->data
.mode
!= mode
)
3123 if (strncmp(var
->name
, name
, baselen
) == 0) {
3124 /* Check for exact name matches but also check for arrays and
3127 if (name
[baselen
] == '\0' ||
3128 name
[baselen
] == '[' ||
3129 name
[baselen
] == '.') {
3141 add_interface_variables(struct gl_shader_program
*shProg
,
3142 struct gl_shader
*sh
, GLenum programInterface
)
3144 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3145 ir_variable
*var
= node
->as_variable();
3151 switch (var
->data
.mode
) {
3152 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
3153 * "For GetActiveAttrib, all active vertex shader input variables
3154 * are enumerated, including the special built-in inputs gl_VertexID
3155 * and gl_InstanceID."
3157 case ir_var_system_value
:
3158 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
3159 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
3160 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
3162 /* Mark special built-in inputs referenced by the vertex stage so
3163 * that they are considered active by the shader queries.
3165 mask
= (1 << (MESA_SHADER_VERTEX
));
3167 case ir_var_shader_in
:
3168 if (programInterface
!= GL_PROGRAM_INPUT
)
3171 case ir_var_shader_out
:
3172 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3179 if (!add_program_resource(shProg
, programInterface
, var
,
3180 build_stageref(shProg
, var
->name
,
3181 var
->data
.mode
) | mask
))
3188 * Builds up a list of program resources that point to existing
3192 build_program_resource_list(struct gl_shader_program
*shProg
)
3194 /* Rebuild resource list. */
3195 if (shProg
->ProgramResourceList
) {
3196 ralloc_free(shProg
->ProgramResourceList
);
3197 shProg
->ProgramResourceList
= NULL
;
3198 shProg
->NumProgramResourceList
= 0;
3201 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
3203 /* Determine first input and final output stage. These are used to
3204 * detect which variables should be enumerated in the resource list
3205 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
3207 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3208 if (!shProg
->_LinkedShaders
[i
])
3210 if (input_stage
== MESA_SHADER_STAGES
)
3215 /* Empty shader, no resources. */
3216 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
3219 /* Add inputs and outputs to the resource list. */
3220 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
],
3224 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
],
3228 /* Add transform feedback varyings. */
3229 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
3230 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
3231 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
3232 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
3238 /* Add uniforms from uniform storage. */
3239 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3240 /* Do not add uniforms internally used by Mesa. */
3241 if (shProg
->UniformStorage
[i
].hidden
)
3245 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
3248 /* Add stagereferences for uniforms in a uniform block. */
3249 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3250 if (block_index
!= -1) {
3251 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
3252 if (shProg
->UniformBlockStageIndex
[j
][block_index
] != -1)
3253 stageref
|= (1 << j
);
3257 if (!add_program_resource(shProg
, GL_UNIFORM
,
3258 &shProg
->UniformStorage
[i
], stageref
))
3262 /* Add program uniform blocks. */
3263 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
3264 if (!add_program_resource(shProg
, GL_UNIFORM_BLOCK
,
3265 &shProg
->UniformBlocks
[i
], 0))
3269 /* Add atomic counter buffers. */
3270 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3271 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3272 &shProg
->AtomicBuffers
[i
], 0))
3276 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3278 if (!shProg
->UniformStorage
[i
].hidden
)
3281 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
3282 if (!shProg
->UniformStorage
[i
].subroutine
[j
].active
)
3285 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
3286 /* add shader subroutines */
3287 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
3292 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3293 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3299 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
3300 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3301 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
3306 /* TODO - following extensions will require more resource types:
3308 * GL_ARB_shader_storage_buffer_object
3313 * This check is done to make sure we allow only constant expression
3314 * indexing and "constant-index-expression" (indexing with an expression
3315 * that includes loop induction variable).
3318 validate_sampler_array_indexing(struct gl_context
*ctx
,
3319 struct gl_shader_program
*prog
)
3321 dynamic_sampler_array_indexing_visitor v
;
3322 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3323 if (prog
->_LinkedShaders
[i
] == NULL
)
3326 bool no_dynamic_indexing
=
3327 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
3329 /* Search for array derefs in shader. */
3330 v
.run(prog
->_LinkedShaders
[i
]->ir
);
3331 if (v
.uses_dynamic_sampler_array_indexing()) {
3332 const char *msg
= "sampler arrays indexed with non-constant "
3333 "expressions is forbidden in GLSL %s %u";
3334 /* Backend has indicated that it has no dynamic indexing support. */
3335 if (no_dynamic_indexing
) {
3336 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3339 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3347 link_assign_subroutine_types(struct gl_shader_program
*prog
)
3349 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3350 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3355 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3356 ir_function
*fn
= node
->as_function();
3360 if (fn
->is_subroutine
)
3361 sh
->NumSubroutineUniformTypes
++;
3363 if (!fn
->num_subroutine_types
)
3366 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
3367 struct gl_subroutine_function
,
3368 sh
->NumSubroutineFunctions
+ 1);
3369 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
3370 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
3371 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
3372 ralloc_array(sh
, const struct glsl_type
*,
3373 fn
->num_subroutine_types
);
3374 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
3375 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
3376 sh
->NumSubroutineFunctions
++;
3382 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3384 tfeedback_decl
*tfeedback_decls
= NULL
;
3385 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
3387 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
3389 prog
->LinkStatus
= true; /* All error paths will set this to false */
3390 prog
->Validated
= false;
3391 prog
->_Used
= false;
3393 prog
->ARB_fragment_coord_conventions_enable
= false;
3395 /* Separate the shaders into groups based on their type.
3397 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
3398 unsigned num_shaders
[MESA_SHADER_STAGES
];
3400 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3401 shader_list
[i
] = (struct gl_shader
**)
3402 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
3406 unsigned min_version
= UINT_MAX
;
3407 unsigned max_version
= 0;
3408 const bool is_es_prog
=
3409 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
3410 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
3411 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
3412 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
3414 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
3415 linker_error(prog
, "all shaders must use same shading "
3416 "language version\n");
3420 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
3421 prog
->ARB_fragment_coord_conventions_enable
= true;
3424 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
3425 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
3426 num_shaders
[shader_type
]++;
3429 /* In desktop GLSL, different shader versions may be linked together. In
3430 * GLSL ES, all shader versions must be the same.
3432 if (is_es_prog
&& min_version
!= max_version
) {
3433 linker_error(prog
, "all shaders must use same shading "
3434 "language version\n");
3438 prog
->Version
= max_version
;
3439 prog
->IsES
= is_es_prog
;
3441 /* From OpenGL 4.5 Core specification (7.3 Program Objects):
3442 * "Linking can fail for a variety of reasons as specified in the OpenGL
3443 * Shading Language Specification, as well as any of the following
3446 * * No shader objects are attached to program.
3450 * Same rule applies for OpenGL ES >= 3.1.
3453 if (prog
->NumShaders
== 0 &&
3454 ((ctx
->API
== API_OPENGL_CORE
&& ctx
->Version
>= 45) ||
3455 (ctx
->API
== API_OPENGLES2
&& ctx
->Version
>= 31))) {
3456 linker_error(prog
, "No shader objects are attached to program.\n");
3460 /* Some shaders have to be linked with some other shaders present.
3462 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
3463 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3464 !prog
->SeparateShader
) {
3465 linker_error(prog
, "Geometry shader must be linked with "
3469 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
3470 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3471 !prog
->SeparateShader
) {
3472 linker_error(prog
, "Tessellation evaluation shader must be linked with "
3476 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3477 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
3478 !prog
->SeparateShader
) {
3479 linker_error(prog
, "Tessellation control shader must be linked with "
3484 /* The spec is self-contradictory here. It allows linking without a tess
3485 * eval shader, but that can only be used with transform feedback and
3486 * rasterization disabled. However, transform feedback isn't allowed
3487 * with GL_PATCHES, so it can't be used.
3489 * More investigation showed that the idea of transform feedback after
3490 * a tess control shader was dropped, because some hw vendors couldn't
3491 * support tessellation without a tess eval shader, but the linker section
3492 * wasn't updated to reflect that.
3494 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
3497 * Do what's reasonable and always require a tess eval shader if a tess
3498 * control shader is present.
3500 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
3501 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0 &&
3502 !prog
->SeparateShader
) {
3503 linker_error(prog
, "Tessellation control shader must be linked with "
3504 "tessellation evaluation shader\n");
3508 /* Compute shaders have additional restrictions. */
3509 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
3510 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
3511 linker_error(prog
, "Compute shaders may not be linked with any other "
3512 "type of shader\n");
3515 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3516 if (prog
->_LinkedShaders
[i
] != NULL
)
3517 ctx
->Driver
.DeleteShader(ctx
, prog
->_LinkedShaders
[i
]);
3519 prog
->_LinkedShaders
[i
] = NULL
;
3522 /* Link all shaders for a particular stage and validate the result.
3524 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
3525 if (num_shaders
[stage
] > 0) {
3526 gl_shader
*const sh
=
3527 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
3528 num_shaders
[stage
]);
3530 if (!prog
->LinkStatus
) {
3532 ctx
->Driver
.DeleteShader(ctx
, sh
);
3537 case MESA_SHADER_VERTEX
:
3538 validate_vertex_shader_executable(prog
, sh
);
3540 case MESA_SHADER_TESS_CTRL
:
3541 /* nothing to be done */
3543 case MESA_SHADER_TESS_EVAL
:
3544 validate_tess_eval_shader_executable(prog
, sh
);
3546 case MESA_SHADER_GEOMETRY
:
3547 validate_geometry_shader_executable(prog
, sh
);
3549 case MESA_SHADER_FRAGMENT
:
3550 validate_fragment_shader_executable(prog
, sh
);
3553 if (!prog
->LinkStatus
) {
3555 ctx
->Driver
.DeleteShader(ctx
, sh
);
3559 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
3563 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
3564 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
3565 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
3566 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
3567 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
3568 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
3570 prog
->LastClipDistanceArraySize
= 0; /* Not used */
3572 /* Here begins the inter-stage linking phase. Some initial validation is
3573 * performed, then locations are assigned for uniforms, attributes, and
3576 cross_validate_uniforms(prog
);
3577 if (!prog
->LinkStatus
)
3582 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
3583 if (prog
->_LinkedShaders
[prev
] != NULL
)
3587 check_explicit_uniform_locations(ctx
, prog
);
3588 link_assign_subroutine_types(prog
);
3590 if (!prog
->LinkStatus
)
3593 resize_tes_inputs(ctx
, prog
);
3595 /* Validate the inputs of each stage with the output of the preceding
3598 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3599 if (prog
->_LinkedShaders
[i
] == NULL
)
3602 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
3603 prog
->_LinkedShaders
[i
]);
3604 if (!prog
->LinkStatus
)
3607 cross_validate_outputs_to_inputs(prog
,
3608 prog
->_LinkedShaders
[prev
],
3609 prog
->_LinkedShaders
[i
]);
3610 if (!prog
->LinkStatus
)
3616 /* Cross-validate uniform blocks between shader stages */
3617 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
3618 MESA_SHADER_STAGES
);
3619 if (!prog
->LinkStatus
)
3622 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3623 if (prog
->_LinkedShaders
[i
] != NULL
)
3624 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
3627 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
3628 * it before optimization because we want most of the checks to get
3629 * dropped thanks to constant propagation.
3631 * This rule also applies to GLSL ES 3.00.
3633 if (max_version
>= (is_es_prog
? 300 : 130)) {
3634 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3636 lower_discard_flow(sh
->ir
);
3640 if (!interstage_cross_validate_uniform_blocks(prog
))
3643 /* Do common optimization before assigning storage for attributes,
3644 * uniforms, and varyings. Later optimization could possibly make
3645 * some of that unused.
3647 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3648 if (prog
->_LinkedShaders
[i
] == NULL
)
3651 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
3652 if (!prog
->LinkStatus
)
3655 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
3656 lower_clip_distance(prog
->_LinkedShaders
[i
]);
3659 if (ctx
->Const
.LowerTessLevel
) {
3660 lower_tess_level(prog
->_LinkedShaders
[i
]);
3663 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
3664 &ctx
->Const
.ShaderCompilerOptions
[i
],
3665 ctx
->Const
.NativeIntegers
))
3668 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
3671 /* Validation for special cases where we allow sampler array indexing
3672 * with loop induction variable. This check emits a warning or error
3673 * depending if backend can handle dynamic indexing.
3675 if ((!prog
->IsES
&& prog
->Version
< 130) ||
3676 (prog
->IsES
&& prog
->Version
< 300)) {
3677 if (!validate_sampler_array_indexing(ctx
, prog
))
3681 /* Check and validate stream emissions in geometry shaders */
3682 validate_geometry_shader_emissions(ctx
, prog
);
3684 /* Mark all generic shader inputs and outputs as unpaired. */
3685 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
3686 if (prog
->_LinkedShaders
[i
] != NULL
) {
3687 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
3691 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
3692 MESA_SHADER_VERTEX
)) {
3696 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
3697 MESA_SHADER_FRAGMENT
)) {
3701 unsigned first
, last
;
3703 first
= MESA_SHADER_STAGES
;
3706 /* Determine first and last stage. */
3707 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3708 if (!prog
->_LinkedShaders
[i
])
3710 if (first
== MESA_SHADER_STAGES
)
3715 if (num_tfeedback_decls
!= 0) {
3716 /* From GL_EXT_transform_feedback:
3717 * A program will fail to link if:
3719 * * the <count> specified by TransformFeedbackVaryingsEXT is
3720 * non-zero, but the program object has no vertex or geometry
3723 if (first
== MESA_SHADER_FRAGMENT
) {
3724 linker_error(prog
, "Transform feedback varyings specified, but "
3725 "no vertex or geometry shader is present.\n");
3729 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
3730 prog
->TransformFeedback
.NumVarying
);
3731 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
3732 prog
->TransformFeedback
.VaryingNames
,
3737 /* Linking the stages in the opposite order (from fragment to vertex)
3738 * ensures that inter-shader outputs written to in an earlier stage are
3739 * eliminated if they are (transitively) not used in a later stage.
3743 if (first
< MESA_SHADER_FRAGMENT
) {
3744 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
3746 if (first
== MESA_SHADER_GEOMETRY
) {
3747 /* There was no vertex shader, but we still have to assign varying
3748 * locations for use by geometry shader inputs in SSO.
3750 * If the shader is not separable (i.e., prog->SeparateShader is
3751 * false), linking will have already failed when first is
3752 * MESA_SHADER_GEOMETRY.
3754 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3755 NULL
, prog
->_LinkedShaders
[first
],
3756 num_tfeedback_decls
, tfeedback_decls
))
3760 if (last
!= MESA_SHADER_FRAGMENT
&&
3761 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
3762 /* There was no fragment shader, but we still have to assign varying
3763 * locations for use by transform feedback.
3765 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3767 num_tfeedback_decls
, tfeedback_decls
))
3771 do_dead_builtin_varyings(ctx
, sh
, NULL
,
3772 num_tfeedback_decls
, tfeedback_decls
);
3774 if (!prog
->SeparateShader
)
3775 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
3777 /* Eliminate code that is now dead due to unused outputs being demoted.
3779 while (do_dead_code(sh
->ir
, false))
3782 else if (first
== MESA_SHADER_FRAGMENT
) {
3783 /* If the program only contains a fragment shader...
3785 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
3787 do_dead_builtin_varyings(ctx
, NULL
, sh
,
3788 num_tfeedback_decls
, tfeedback_decls
);
3790 if (prog
->SeparateShader
) {
3791 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
3792 NULL
/* producer */,
3794 0 /* num_tfeedback_decls */,
3795 NULL
/* tfeedback_decls */))
3798 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
3800 while (do_dead_code(sh
->ir
, false))
3805 for (int i
= next
- 1; i
>= 0; i
--) {
3806 if (prog
->_LinkedShaders
[i
] == NULL
)
3809 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
3810 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
3812 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
3813 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3817 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
3818 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
3821 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
3822 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
3824 /* Eliminate code that is now dead due to unused outputs being demoted.
3826 while (do_dead_code(sh_i
->ir
, false))
3828 while (do_dead_code(sh_next
->ir
, false))
3831 /* This must be done after all dead varyings are eliminated. */
3832 if (!check_against_output_limit(ctx
, prog
, sh_i
))
3834 if (!check_against_input_limit(ctx
, prog
, sh_next
))
3840 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
3843 update_array_sizes(prog
);
3844 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
3845 link_assign_atomic_counter_resources(ctx
, prog
);
3846 store_fragdepth_layout(prog
);
3848 link_calculate_subroutine_compat(prog
);
3849 check_resources(ctx
, prog
);
3850 check_subroutine_resources(prog
);
3851 check_image_resources(ctx
, prog
);
3852 link_check_atomic_counter_resources(ctx
, prog
);
3854 if (!prog
->LinkStatus
)
3857 /* OpenGL ES requires that a vertex shader and a fragment shader both be
3858 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
3859 * anything about shader linking when one of the shaders (vertex or
3860 * fragment shader) is absent. So, the extension shouldn't change the
3861 * behavior specified in GLSL specification.
3863 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
3864 /* With ES < 3.1 one needs to have always vertex + fragment shader. */
3865 if (ctx
->Version
< 31) {
3866 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
3867 linker_error(prog
, "program lacks a vertex shader\n");
3868 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
3869 linker_error(prog
, "program lacks a fragment shader\n");
3872 /* From OpenGL ES 3.1 specification (7.3 Program Objects):
3873 * "Linking can fail for a variety of reasons as specified in the
3874 * OpenGL ES Shading Language Specification, as well as any of the
3875 * following reasons:
3879 * * program contains objects to form either a vertex shader or
3880 * fragment shader, and program is not separable, and does not
3881 * contain objects to form both a vertex shader and fragment
3884 if (!!prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] ^
3885 !!prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]) {
3886 linker_error(prog
, "Program needs to contain both vertex and "
3887 "fragment shaders.\n");
3892 /* FINISHME: Assign fragment shader output locations. */
3895 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3896 free(shader_list
[i
]);
3897 if (prog
->_LinkedShaders
[i
] == NULL
)
3900 /* Do a final validation step to make sure that the IR wasn't
3901 * invalidated by any modifications performed after intrastage linking.
3903 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
3905 /* Retain any live IR, but trash the rest. */
3906 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
3908 /* The symbol table in the linked shaders may contain references to
3909 * variables that were removed (e.g., unused uniforms). Since it may
3910 * contain junk, there is no possible valid use. Delete it and set the
3913 delete prog
->_LinkedShaders
[i
]->symbols
;
3914 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
3917 ralloc_free(mem_ctx
);