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 "util/strndup.h"
69 #include "main/core.h"
70 #include "glsl_symbol_table.h"
71 #include "glsl_parser_extras.h"
74 #include "program/hash_table.h"
76 #include "link_varyings.h"
77 #include "ir_optimization.h"
78 #include "ir_rvalue_visitor.h"
79 #include "ir_uniform.h"
81 #include "main/shaderobj.h"
82 #include "main/enums.h"
85 void linker_error(gl_shader_program
*, const char *, ...);
90 * Visitor that determines whether or not a variable is ever written.
92 class find_assignment_visitor
: public ir_hierarchical_visitor
{
94 find_assignment_visitor(const char *name
)
95 : name(name
), found(false)
100 virtual ir_visitor_status
visit_enter(ir_assignment
*ir
)
102 ir_variable
*const var
= ir
->lhs
->variable_referenced();
104 if (strcmp(name
, var
->name
) == 0) {
109 return visit_continue_with_parent
;
112 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
114 foreach_two_lists(formal_node
, &ir
->callee
->parameters
,
115 actual_node
, &ir
->actual_parameters
) {
116 ir_rvalue
*param_rval
= (ir_rvalue
*) actual_node
;
117 ir_variable
*sig_param
= (ir_variable
*) formal_node
;
119 if (sig_param
->data
.mode
== ir_var_function_out
||
120 sig_param
->data
.mode
== ir_var_function_inout
) {
121 ir_variable
*var
= param_rval
->variable_referenced();
122 if (var
&& strcmp(name
, var
->name
) == 0) {
129 if (ir
->return_deref
!= NULL
) {
130 ir_variable
*const var
= ir
->return_deref
->variable_referenced();
132 if (strcmp(name
, var
->name
) == 0) {
138 return visit_continue_with_parent
;
141 bool variable_found()
147 const char *name
; /**< Find writes to a variable with this name. */
148 bool found
; /**< Was a write to the variable found? */
153 * Visitor that determines whether or not a variable is ever read.
155 class find_deref_visitor
: public ir_hierarchical_visitor
{
157 find_deref_visitor(const char *name
)
158 : name(name
), found(false)
163 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
165 if (strcmp(this->name
, ir
->var
->name
) == 0) {
170 return visit_continue
;
173 bool variable_found() const
179 const char *name
; /**< Find writes to a variable with this name. */
180 bool found
; /**< Was a write to the variable found? */
184 class geom_array_resize_visitor
: public ir_hierarchical_visitor
{
186 unsigned num_vertices
;
187 gl_shader_program
*prog
;
189 geom_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
191 this->num_vertices
= num_vertices
;
195 virtual ~geom_array_resize_visitor()
200 virtual ir_visitor_status
visit(ir_variable
*var
)
202 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
)
203 return visit_continue
;
205 unsigned size
= var
->type
->length
;
207 /* Generate a link error if the shader has declared this array with an
210 if (size
&& size
!= this->num_vertices
) {
211 linker_error(this->prog
, "size of array %s declared as %u, "
212 "but number of input vertices is %u\n",
213 var
->name
, size
, this->num_vertices
);
214 return visit_continue
;
217 /* Generate a link error if the shader attempts to access an input
218 * array using an index too large for its actual size assigned at link
221 if (var
->data
.max_array_access
>= this->num_vertices
) {
222 linker_error(this->prog
, "geometry shader accesses element %i of "
223 "%s, but only %i input vertices\n",
224 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
225 return visit_continue
;
228 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
230 var
->data
.max_array_access
= this->num_vertices
- 1;
232 return visit_continue
;
235 /* Dereferences of input variables need to be updated so that their type
236 * matches the newly assigned type of the variable they are accessing. */
237 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
239 ir
->type
= ir
->var
->type
;
240 return visit_continue
;
243 /* Dereferences of 2D input arrays need to be updated so that their type
244 * matches the newly assigned type of the array they are accessing. */
245 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
247 const glsl_type
*const vt
= ir
->array
->type
;
249 ir
->type
= vt
->fields
.array
;
250 return visit_continue
;
254 class tess_eval_array_resize_visitor
: public ir_hierarchical_visitor
{
256 unsigned num_vertices
;
257 gl_shader_program
*prog
;
259 tess_eval_array_resize_visitor(unsigned num_vertices
, gl_shader_program
*prog
)
261 this->num_vertices
= num_vertices
;
265 virtual ~tess_eval_array_resize_visitor()
270 virtual ir_visitor_status
visit(ir_variable
*var
)
272 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
|| var
->data
.patch
)
273 return visit_continue
;
275 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
277 var
->data
.max_array_access
= this->num_vertices
- 1;
279 return visit_continue
;
282 /* Dereferences of input variables need to be updated so that their type
283 * matches the newly assigned type of the variable they are accessing. */
284 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
286 ir
->type
= ir
->var
->type
;
287 return visit_continue
;
290 /* Dereferences of 2D input arrays need to be updated so that their type
291 * matches the newly assigned type of the array they are accessing. */
292 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
294 const glsl_type
*const vt
= ir
->array
->type
;
296 ir
->type
= vt
->fields
.array
;
297 return visit_continue
;
301 class barrier_use_visitor
: public ir_hierarchical_visitor
{
303 barrier_use_visitor(gl_shader_program
*prog
)
304 : prog(prog
), in_main(false), after_return(false), control_flow(0)
308 virtual ~barrier_use_visitor()
313 virtual ir_visitor_status
visit_enter(ir_function
*ir
)
315 if (strcmp(ir
->name
, "main") == 0)
318 return visit_continue
;
321 virtual ir_visitor_status
visit_leave(ir_function
*)
324 after_return
= false;
325 return visit_continue
;
328 virtual ir_visitor_status
visit_leave(ir_return
*)
331 return visit_continue
;
334 virtual ir_visitor_status
visit_enter(ir_if
*)
337 return visit_continue
;
340 virtual ir_visitor_status
visit_leave(ir_if
*)
343 return visit_continue
;
346 virtual ir_visitor_status
visit_enter(ir_loop
*)
349 return visit_continue
;
352 virtual ir_visitor_status
visit_leave(ir_loop
*)
355 return visit_continue
;
358 /* FINISHME: `switch` is not expressed at the IR level -- it's already
359 * been lowered to a mess of `if`s. We'll correctly disallow any use of
360 * barrier() in a conditional path within the switch, but not in a path
361 * which is always hit.
364 virtual ir_visitor_status
visit_enter(ir_call
*ir
)
366 if (ir
->use_builtin
&& strcmp(ir
->callee_name(), "barrier") == 0) {
367 /* Use of barrier(); determine if it is legal: */
369 linker_error(prog
, "Builtin barrier() may only be used in main");
374 linker_error(prog
, "Builtin barrier() may not be used after return");
378 if (control_flow
!= 0) {
379 linker_error(prog
, "Builtin barrier() may not be used inside control flow");
383 return visit_continue
;
387 gl_shader_program
*prog
;
388 bool in_main
, after_return
;
393 * Visitor that determines the highest stream id to which a (geometry) shader
394 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
396 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
398 find_emit_vertex_visitor(int max_allowed
)
399 : max_stream_allowed(max_allowed
),
400 invalid_stream_id(0),
401 invalid_stream_id_from_emit_vertex(false),
402 end_primitive_found(false),
403 uses_non_zero_stream(false)
408 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
410 int stream_id
= ir
->stream_id();
413 invalid_stream_id
= stream_id
;
414 invalid_stream_id_from_emit_vertex
= true;
418 if (stream_id
> max_stream_allowed
) {
419 invalid_stream_id
= stream_id
;
420 invalid_stream_id_from_emit_vertex
= true;
425 uses_non_zero_stream
= true;
427 return visit_continue
;
430 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
432 end_primitive_found
= true;
434 int stream_id
= ir
->stream_id();
437 invalid_stream_id
= stream_id
;
438 invalid_stream_id_from_emit_vertex
= false;
442 if (stream_id
> max_stream_allowed
) {
443 invalid_stream_id
= stream_id
;
444 invalid_stream_id_from_emit_vertex
= false;
449 uses_non_zero_stream
= true;
451 return visit_continue
;
456 return invalid_stream_id
!= 0;
459 const char *error_func()
461 return invalid_stream_id_from_emit_vertex
?
462 "EmitStreamVertex" : "EndStreamPrimitive";
467 return invalid_stream_id
;
472 return uses_non_zero_stream
;
475 bool uses_end_primitive()
477 return end_primitive_found
;
481 int max_stream_allowed
;
482 int invalid_stream_id
;
483 bool invalid_stream_id_from_emit_vertex
;
484 bool end_primitive_found
;
485 bool uses_non_zero_stream
;
488 /* Class that finds array derefs and check if indexes are dynamic. */
489 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
492 dynamic_sampler_array_indexing_visitor() :
493 dynamic_sampler_array_indexing(false)
497 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
499 if (!ir
->variable_referenced())
500 return visit_continue
;
502 if (!ir
->variable_referenced()->type
->contains_sampler())
503 return visit_continue
;
505 if (!ir
->array_index
->constant_expression_value()) {
506 dynamic_sampler_array_indexing
= true;
509 return visit_continue
;
512 bool uses_dynamic_sampler_array_indexing()
514 return dynamic_sampler_array_indexing
;
518 bool dynamic_sampler_array_indexing
;
521 } /* anonymous namespace */
524 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
528 ralloc_strcat(&prog
->InfoLog
, "error: ");
530 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
533 prog
->LinkStatus
= false;
538 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
542 ralloc_strcat(&prog
->InfoLog
, "warning: ");
544 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
551 * Given a string identifying a program resource, break it into a base name
552 * and an optional array index in square brackets.
554 * If an array index is present, \c out_base_name_end is set to point to the
555 * "[" that precedes the array index, and the array index itself is returned
558 * If no array index is present (or if the array index is negative or
559 * mal-formed), \c out_base_name_end, is set to point to the null terminator
560 * at the end of the input string, and -1 is returned.
562 * Only the final array index is parsed; if the string contains other array
563 * indices (or structure field accesses), they are left in the base name.
565 * No attempt is made to check that the base name is properly formed;
566 * typically the caller will look up the base name in a hash table, so
567 * ill-formed base names simply turn into hash table lookup failures.
570 parse_program_resource_name(const GLchar
*name
,
571 const GLchar
**out_base_name_end
)
573 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
575 * "When an integer array element or block instance number is part of
576 * the name string, it will be specified in decimal form without a "+"
577 * or "-" sign or any extra leading zeroes. Additionally, the name
578 * string will not include white space anywhere in the string."
581 const size_t len
= strlen(name
);
582 *out_base_name_end
= name
+ len
;
584 if (len
== 0 || name
[len
-1] != ']')
587 /* Walk backwards over the string looking for a non-digit character. This
588 * had better be the opening bracket for an array index.
590 * Initially, i specifies the location of the ']'. Since the string may
591 * contain only the ']' charcater, walk backwards very carefully.
594 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
597 if ((i
== 0) || name
[i
-1] != '[')
600 long array_index
= strtol(&name
[i
], NULL
, 10);
604 /* Check for leading zero */
605 if (name
[i
] == '0' && name
[i
+1] != ']')
608 *out_base_name_end
= name
+ (i
- 1);
614 link_invalidate_variable_locations(exec_list
*ir
)
616 foreach_in_list(ir_instruction
, node
, ir
) {
617 ir_variable
*const var
= node
->as_variable();
622 /* Only assign locations for variables that lack an explicit location.
623 * Explicit locations are set for all built-in variables, generic vertex
624 * shader inputs (via layout(location=...)), and generic fragment shader
625 * outputs (also via layout(location=...)).
627 if (!var
->data
.explicit_location
) {
628 var
->data
.location
= -1;
629 var
->data
.location_frac
= 0;
632 /* ir_variable::is_unmatched_generic_inout is used by the linker while
633 * connecting outputs from one stage to inputs of the next stage.
635 * There are two implicit assumptions here. First, we assume that any
636 * built-in variable (i.e., non-generic in or out) will have
637 * explicit_location set. Second, we assume that any generic in or out
638 * will not have explicit_location set.
640 * This second assumption will only be valid until
641 * GL_ARB_separate_shader_objects is supported. When that extension is
642 * implemented, this function will need some modifications.
644 if (!var
->data
.explicit_location
) {
645 var
->data
.is_unmatched_generic_inout
= 1;
647 var
->data
.is_unmatched_generic_inout
= 0;
654 * Set clip_distance_array_size based on the given shader.
656 * Also check for errors based on incorrect usage of gl_ClipVertex and
659 * Return false if an error was reported.
662 analyze_clip_usage(struct gl_shader_program
*prog
,
663 struct gl_shader
*shader
,
664 GLuint
*clip_distance_array_size
)
666 *clip_distance_array_size
= 0;
668 if (!prog
->IsES
&& prog
->Version
>= 130) {
669 /* From section 7.1 (Vertex Shader Special Variables) of the
672 * "It is an error for a shader to statically write both
673 * gl_ClipVertex and gl_ClipDistance."
675 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
676 * gl_ClipVertex nor gl_ClipDistance.
678 find_assignment_visitor
clip_vertex("gl_ClipVertex");
679 find_assignment_visitor
clip_distance("gl_ClipDistance");
681 clip_vertex
.run(shader
->ir
);
682 clip_distance
.run(shader
->ir
);
683 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
684 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
685 "and `gl_ClipDistance'\n",
686 _mesa_shader_stage_to_string(shader
->Stage
));
690 if (clip_distance
.variable_found()) {
691 ir_variable
*clip_distance_var
=
692 shader
->symbols
->get_variable("gl_ClipDistance");
694 assert(clip_distance_var
);
695 *clip_distance_array_size
= clip_distance_var
->type
->length
;
702 * Verify that a vertex shader executable meets all semantic requirements.
704 * Also sets prog->Vert.ClipDistanceArraySize as a side effect.
706 * \param shader Vertex shader executable to be verified
709 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
710 struct gl_shader
*shader
)
715 /* From the GLSL 1.10 spec, page 48:
717 * "The variable gl_Position is available only in the vertex
718 * language and is intended for writing the homogeneous vertex
719 * position. All executions of a well-formed vertex shader
720 * executable must write a value into this variable. [...] The
721 * variable gl_Position is available only in the vertex
722 * language and is intended for writing the homogeneous vertex
723 * position. All executions of a well-formed vertex shader
724 * executable must write a value into this variable."
726 * while in GLSL 1.40 this text is changed to:
728 * "The variable gl_Position is available only in the vertex
729 * language and is intended for writing the homogeneous vertex
730 * position. It can be written at any time during shader
731 * execution. It may also be read back by a vertex shader
732 * after being written. This value will be used by primitive
733 * assembly, clipping, culling, and other fixed functionality
734 * operations, if present, that operate on primitives after
735 * vertex processing has occurred. Its value is undefined if
736 * the vertex shader executable does not write gl_Position."
738 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
739 * gl_Position is not an error.
741 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
742 find_assignment_visitor
find("gl_Position");
743 find
.run(shader
->ir
);
744 if (!find
.variable_found()) {
747 "vertex shader does not write to `gl_Position'."
748 "It's value is undefined. \n");
751 "vertex shader does not write to `gl_Position'. \n");
757 analyze_clip_usage(prog
, shader
, &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
.ClipDistanceArraySize
);
772 * Verify that a fragment shader executable meets all semantic requirements
774 * \param shader Fragment shader executable to be verified
777 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
778 struct gl_shader
*shader
)
783 find_assignment_visitor
frag_color("gl_FragColor");
784 find_assignment_visitor
frag_data("gl_FragData");
786 frag_color
.run(shader
->ir
);
787 frag_data
.run(shader
->ir
);
789 if (frag_color
.variable_found() && frag_data
.variable_found()) {
790 linker_error(prog
, "fragment shader writes to both "
791 "`gl_FragColor' and `gl_FragData'\n");
796 * Verify that a geometry shader executable meets all semantic requirements
798 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize as
801 * \param shader Geometry shader executable to be verified
804 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
805 struct gl_shader
*shader
)
810 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
811 prog
->Geom
.VerticesIn
= num_vertices
;
813 analyze_clip_usage(prog
, shader
, &prog
->Geom
.ClipDistanceArraySize
);
817 * Check if geometry shaders emit to non-zero streams and do corresponding
821 validate_geometry_shader_emissions(struct gl_context
*ctx
,
822 struct gl_shader_program
*prog
)
824 if (prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
] != NULL
) {
825 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
826 emit_vertex
.run(prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
]->ir
);
827 if (emit_vertex
.error()) {
828 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
829 "stream parameter are in the range [0, %d].\n",
830 emit_vertex
.error_func(),
831 emit_vertex
.error_stream(),
832 ctx
->Const
.MaxVertexStreams
- 1);
834 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
835 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
837 /* From the ARB_gpu_shader5 spec:
839 * "Multiple vertex streams are supported only if the output primitive
840 * type is declared to be "points". A program will fail to link if it
841 * contains a geometry shader calling EmitStreamVertex() or
842 * EndStreamPrimitive() if its output primitive type is not "points".
844 * However, in the same spec:
846 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
847 * with <stream> set to zero."
851 * "The function EndPrimitive() is equivalent to calling
852 * EndStreamPrimitive() with <stream> set to zero."
854 * Since we can call EmitVertex() and EndPrimitive() when we output
855 * primitives other than points, calling EmitStreamVertex(0) or
856 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
857 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
858 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
861 if (prog
->Geom
.UsesStreams
&& prog
->Geom
.OutputType
!= GL_POINTS
) {
862 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
863 "with n>0 requires point output\n");
869 validate_intrastage_arrays(struct gl_shader_program
*prog
,
870 ir_variable
*const var
,
871 ir_variable
*const existing
)
873 /* Consider the types to be "the same" if both types are arrays
874 * of the same type and one of the arrays is implicitly sized.
875 * In addition, set the type of the linked variable to the
876 * explicitly sized array.
878 if (var
->type
->is_array() && existing
->type
->is_array()) {
879 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
880 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
881 if (var
->type
->length
!= 0) {
882 if (var
->type
->length
<= existing
->data
.max_array_access
) {
883 linker_error(prog
, "%s `%s' declared as type "
884 "`%s' but outermost dimension has an index"
887 var
->name
, var
->type
->name
,
888 existing
->data
.max_array_access
);
890 existing
->type
= var
->type
;
892 } else if (existing
->type
->length
!= 0) {
893 if(existing
->type
->length
<= var
->data
.max_array_access
&&
894 !existing
->data
.from_ssbo_unsized_array
) {
895 linker_error(prog
, "%s `%s' declared as type "
896 "`%s' but outermost dimension has an index"
899 var
->name
, existing
->type
->name
,
900 var
->data
.max_array_access
);
905 /* The arrays of structs could have different glsl_type pointers but
906 * they are actually the same type. Use record_compare() to check that.
908 if (existing
->type
->fields
.array
->is_record() &&
909 var
->type
->fields
.array
->is_record() &&
910 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
919 * Perform validation of global variables used across multiple shaders
922 cross_validate_globals(struct gl_shader_program
*prog
,
923 struct gl_shader
**shader_list
,
924 unsigned num_shaders
,
927 /* Examine all of the uniforms in all of the shaders and cross validate
930 glsl_symbol_table variables
;
931 for (unsigned i
= 0; i
< num_shaders
; i
++) {
932 if (shader_list
[i
] == NULL
)
935 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
936 ir_variable
*const var
= node
->as_variable();
941 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
944 /* don't cross validate subroutine uniforms */
945 if (var
->type
->contains_subroutine())
948 /* Don't cross validate temporaries that are at global scope. These
949 * will eventually get pulled into the shaders 'main'.
951 if (var
->data
.mode
== ir_var_temporary
)
954 /* If a global with this name has already been seen, verify that the
955 * new instance has the same type. In addition, if the globals have
956 * initializers, the values of the initializers must be the same.
958 ir_variable
*const existing
= variables
.get_variable(var
->name
);
959 if (existing
!= NULL
) {
960 /* Check if types match. Interface blocks have some special
961 * rules so we handle those elsewhere.
963 if (var
->type
!= existing
->type
&&
964 !var
->is_interface_instance()) {
965 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
966 if (var
->type
->is_record() && existing
->type
->is_record()
967 && existing
->type
->record_compare(var
->type
)) {
968 existing
->type
= var
->type
;
970 /* If it is an unsized array in a Shader Storage Block,
971 * two different shaders can access to different elements.
972 * Because of that, they might be converted to different
973 * sized arrays, then check that they are compatible but
974 * ignore the array size.
976 if (!(var
->data
.mode
== ir_var_shader_storage
&&
977 var
->data
.from_ssbo_unsized_array
&&
978 existing
->data
.mode
== ir_var_shader_storage
&&
979 existing
->data
.from_ssbo_unsized_array
&&
980 var
->type
->gl_type
== existing
->type
->gl_type
)) {
981 linker_error(prog
, "%s `%s' declared as type "
982 "`%s' and type `%s'\n",
984 var
->name
, var
->type
->name
,
985 existing
->type
->name
);
992 if (var
->data
.explicit_location
) {
993 if (existing
->data
.explicit_location
994 && (var
->data
.location
!= existing
->data
.location
)) {
995 linker_error(prog
, "explicit locations for %s "
996 "`%s' have differing values\n",
997 mode_string(var
), var
->name
);
1001 existing
->data
.location
= var
->data
.location
;
1002 existing
->data
.explicit_location
= true;
1005 /* From the GLSL 4.20 specification:
1006 * "A link error will result if two compilation units in a program
1007 * specify different integer-constant bindings for the same
1008 * opaque-uniform name. However, it is not an error to specify a
1009 * binding on some but not all declarations for the same name"
1011 if (var
->data
.explicit_binding
) {
1012 if (existing
->data
.explicit_binding
&&
1013 var
->data
.binding
!= existing
->data
.binding
) {
1014 linker_error(prog
, "explicit bindings for %s "
1015 "`%s' have differing values\n",
1016 mode_string(var
), var
->name
);
1020 existing
->data
.binding
= var
->data
.binding
;
1021 existing
->data
.explicit_binding
= true;
1024 if (var
->type
->contains_atomic() &&
1025 var
->data
.atomic
.offset
!= existing
->data
.atomic
.offset
) {
1026 linker_error(prog
, "offset specifications for %s "
1027 "`%s' have differing values\n",
1028 mode_string(var
), var
->name
);
1032 /* Validate layout qualifiers for gl_FragDepth.
1034 * From the AMD/ARB_conservative_depth specs:
1036 * "If gl_FragDepth is redeclared in any fragment shader in a
1037 * program, it must be redeclared in all fragment shaders in
1038 * that program that have static assignments to
1039 * gl_FragDepth. All redeclarations of gl_FragDepth in all
1040 * fragment shaders in a single program must have the same set
1043 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
1044 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
1045 bool layout_differs
=
1046 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
1048 if (layout_declared
&& layout_differs
) {
1050 "All redeclarations of gl_FragDepth in all "
1051 "fragment shaders in a single program must have "
1052 "the same set of qualifiers.\n");
1055 if (var
->data
.used
&& layout_differs
) {
1057 "If gl_FragDepth is redeclared with a layout "
1058 "qualifier in any fragment shader, it must be "
1059 "redeclared with the same layout qualifier in "
1060 "all fragment shaders that have assignments to "
1065 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
1067 * "If a shared global has multiple initializers, the
1068 * initializers must all be constant expressions, and they
1069 * must all have the same value. Otherwise, a link error will
1070 * result. (A shared global having only one initializer does
1071 * not require that initializer to be a constant expression.)"
1073 * Previous to 4.20 the GLSL spec simply said that initializers
1074 * must have the same value. In this case of non-constant
1075 * initializers, this was impossible to determine. As a result,
1076 * no vendor actually implemented that behavior. The 4.20
1077 * behavior matches the implemented behavior of at least one other
1078 * vendor, so we'll implement that for all GLSL versions.
1080 if (var
->constant_initializer
!= NULL
) {
1081 if (existing
->constant_initializer
!= NULL
) {
1082 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1083 linker_error(prog
, "initializers for %s "
1084 "`%s' have differing values\n",
1085 mode_string(var
), var
->name
);
1089 /* If the first-seen instance of a particular uniform did not
1090 * have an initializer but a later instance does, copy the
1091 * initializer to the version stored in the symbol table.
1093 /* FINISHME: This is wrong. The constant_value field should
1094 * FINISHME: not be modified! Imagine a case where a shader
1095 * FINISHME: without an initializer is linked in two different
1096 * FINISHME: programs with shaders that have differing
1097 * FINISHME: initializers. Linking with the first will
1098 * FINISHME: modify the shader, and linking with the second
1099 * FINISHME: will fail.
1101 existing
->constant_initializer
=
1102 var
->constant_initializer
->clone(ralloc_parent(existing
),
1107 if (var
->data
.has_initializer
) {
1108 if (existing
->data
.has_initializer
1109 && (var
->constant_initializer
== NULL
1110 || existing
->constant_initializer
== NULL
)) {
1112 "shared global variable `%s' has multiple "
1113 "non-constant initializers.\n",
1118 /* Some instance had an initializer, so keep track of that. In
1119 * this location, all sorts of initializers (constant or
1120 * otherwise) will propagate the existence to the variable
1121 * stored in the symbol table.
1123 existing
->data
.has_initializer
= true;
1126 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1127 linker_error(prog
, "declarations for %s `%s' have "
1128 "mismatching invariant qualifiers\n",
1129 mode_string(var
), var
->name
);
1132 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1133 linker_error(prog
, "declarations for %s `%s' have "
1134 "mismatching centroid qualifiers\n",
1135 mode_string(var
), var
->name
);
1138 if (existing
->data
.sample
!= var
->data
.sample
) {
1139 linker_error(prog
, "declarations for %s `%s` have "
1140 "mismatching sample qualifiers\n",
1141 mode_string(var
), var
->name
);
1145 variables
.add_variable(var
);
1152 * Perform validation of uniforms used across multiple shader stages
1155 cross_validate_uniforms(struct gl_shader_program
*prog
)
1157 cross_validate_globals(prog
, prog
->_LinkedShaders
,
1158 MESA_SHADER_STAGES
, true);
1162 * Accumulates the array of prog->BufferInterfaceBlocks and checks that all
1163 * definitons of blocks agree on their contents.
1166 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
)
1168 unsigned max_num_uniform_blocks
= 0;
1169 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1170 if (prog
->_LinkedShaders
[i
])
1171 max_num_uniform_blocks
+= prog
->_LinkedShaders
[i
]->NumBufferInterfaceBlocks
;
1174 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1175 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
1177 prog
->InterfaceBlockStageIndex
[i
] = ralloc_array(prog
, int,
1178 max_num_uniform_blocks
);
1179 for (unsigned int j
= 0; j
< max_num_uniform_blocks
; j
++)
1180 prog
->InterfaceBlockStageIndex
[i
][j
] = -1;
1185 for (unsigned int j
= 0; j
< sh
->NumBufferInterfaceBlocks
; j
++) {
1186 int index
= link_cross_validate_uniform_block(prog
,
1187 &prog
->BufferInterfaceBlocks
,
1188 &prog
->NumBufferInterfaceBlocks
,
1189 &sh
->BufferInterfaceBlocks
[j
]);
1192 linker_error(prog
, "uniform block `%s' has mismatching definitions\n",
1193 sh
->BufferInterfaceBlocks
[j
].Name
);
1197 prog
->InterfaceBlockStageIndex
[i
][index
] = j
;
1206 * Populates a shaders symbol table with all global declarations
1209 populate_symbol_table(gl_shader
*sh
)
1211 sh
->symbols
= new(sh
) glsl_symbol_table
;
1213 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1217 if ((func
= inst
->as_function()) != NULL
) {
1218 sh
->symbols
->add_function(func
);
1219 } else if ((var
= inst
->as_variable()) != NULL
) {
1220 if (var
->data
.mode
!= ir_var_temporary
)
1221 sh
->symbols
->add_variable(var
);
1228 * Remap variables referenced in an instruction tree
1230 * This is used when instruction trees are cloned from one shader and placed in
1231 * another. These trees will contain references to \c ir_variable nodes that
1232 * do not exist in the target shader. This function finds these \c ir_variable
1233 * references and replaces the references with matching variables in the target
1236 * If there is no matching variable in the target shader, a clone of the
1237 * \c ir_variable is made and added to the target shader. The new variable is
1238 * added to \b both the instruction stream and the symbol table.
1240 * \param inst IR tree that is to be processed.
1241 * \param symbols Symbol table containing global scope symbols in the
1243 * \param instructions Instruction stream where new variable declarations
1247 remap_variables(ir_instruction
*inst
, struct gl_shader
*target
,
1250 class remap_visitor
: public ir_hierarchical_visitor
{
1252 remap_visitor(struct gl_shader
*target
,
1255 this->target
= target
;
1256 this->symbols
= target
->symbols
;
1257 this->instructions
= target
->ir
;
1258 this->temps
= temps
;
1261 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1263 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1264 ir_variable
*var
= (ir_variable
*) hash_table_find(temps
, ir
->var
);
1266 assert(var
!= NULL
);
1268 return visit_continue
;
1271 ir_variable
*const existing
=
1272 this->symbols
->get_variable(ir
->var
->name
);
1273 if (existing
!= NULL
)
1276 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1278 this->symbols
->add_variable(copy
);
1279 this->instructions
->push_head(copy
);
1283 return visit_continue
;
1287 struct gl_shader
*target
;
1288 glsl_symbol_table
*symbols
;
1289 exec_list
*instructions
;
1293 remap_visitor
v(target
, temps
);
1300 * Move non-declarations from one instruction stream to another
1302 * The intended usage pattern of this function is to pass the pointer to the
1303 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1304 * pointer) for \c last and \c false for \c make_copies on the first
1305 * call. Successive calls pass the return value of the previous call for
1306 * \c last and \c true for \c make_copies.
1308 * \param instructions Source instruction stream
1309 * \param last Instruction after which new instructions should be
1310 * inserted in the target instruction stream
1311 * \param make_copies Flag selecting whether instructions in \c instructions
1312 * should be copied (via \c ir_instruction::clone) into the
1313 * target list or moved.
1316 * The new "last" instruction in the target instruction stream. This pointer
1317 * is suitable for use as the \c last parameter of a later call to this
1321 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1322 bool make_copies
, gl_shader
*target
)
1324 hash_table
*temps
= NULL
;
1327 temps
= hash_table_ctor(0, hash_table_pointer_hash
,
1328 hash_table_pointer_compare
);
1330 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1331 if (inst
->as_function())
1334 ir_variable
*var
= inst
->as_variable();
1335 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1338 assert(inst
->as_assignment()
1340 || inst
->as_if() /* for initializers with the ?: operator */
1341 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1344 inst
= inst
->clone(target
, NULL
);
1347 hash_table_insert(temps
, inst
, var
);
1349 remap_variables(inst
, target
, temps
);
1354 last
->insert_after(inst
);
1359 hash_table_dtor(temps
);
1366 * This class is only used in link_intrastage_shaders() below but declaring
1367 * it inside that function leads to compiler warnings with some versions of
1370 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1372 array_sizing_visitor()
1373 : mem_ctx(ralloc_context(NULL
)),
1374 unnamed_interfaces(hash_table_ctor(0, hash_table_pointer_hash
,
1375 hash_table_pointer_compare
))
1379 ~array_sizing_visitor()
1381 hash_table_dtor(this->unnamed_interfaces
);
1382 ralloc_free(this->mem_ctx
);
1385 virtual ir_visitor_status
visit(ir_variable
*var
)
1387 const glsl_type
*type_without_array
;
1388 fixup_type(&var
->type
, var
->data
.max_array_access
,
1389 var
->data
.from_ssbo_unsized_array
);
1390 type_without_array
= var
->type
->without_array();
1391 if (var
->type
->is_interface()) {
1392 if (interface_contains_unsized_arrays(var
->type
)) {
1393 const glsl_type
*new_type
=
1394 resize_interface_members(var
->type
,
1395 var
->get_max_ifc_array_access(),
1396 var
->is_in_shader_storage_block());
1397 var
->type
= new_type
;
1398 var
->change_interface_type(new_type
);
1400 } else if (type_without_array
->is_interface()) {
1401 if (interface_contains_unsized_arrays(type_without_array
)) {
1402 const glsl_type
*new_type
=
1403 resize_interface_members(type_without_array
,
1404 var
->get_max_ifc_array_access(),
1405 var
->is_in_shader_storage_block());
1406 var
->change_interface_type(new_type
);
1407 var
->type
= update_interface_members_array(var
->type
, new_type
);
1409 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1410 /* Store a pointer to the variable in the unnamed_interfaces
1413 ir_variable
**interface_vars
= (ir_variable
**)
1414 hash_table_find(this->unnamed_interfaces
, ifc_type
);
1415 if (interface_vars
== NULL
) {
1416 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1418 hash_table_insert(this->unnamed_interfaces
, interface_vars
,
1421 unsigned index
= ifc_type
->field_index(var
->name
);
1422 assert(index
< ifc_type
->length
);
1423 assert(interface_vars
[index
] == NULL
);
1424 interface_vars
[index
] = var
;
1426 return visit_continue
;
1430 * For each unnamed interface block that was discovered while running the
1431 * visitor, adjust the interface type to reflect the newly assigned array
1432 * sizes, and fix up the ir_variable nodes to point to the new interface
1435 void fixup_unnamed_interface_types()
1437 hash_table_call_foreach(this->unnamed_interfaces
,
1438 fixup_unnamed_interface_type
, NULL
);
1443 * If the type pointed to by \c type represents an unsized array, replace
1444 * it with a sized array whose size is determined by max_array_access.
1446 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1447 bool from_ssbo_unsized_array
)
1449 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1450 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1451 max_array_access
+ 1);
1452 assert(*type
!= NULL
);
1456 static const glsl_type
*
1457 update_interface_members_array(const glsl_type
*type
,
1458 const glsl_type
*new_interface_type
)
1460 const glsl_type
*element_type
= type
->fields
.array
;
1461 if (element_type
->is_array()) {
1462 const glsl_type
*new_array_type
=
1463 update_interface_members_array(element_type
, new_interface_type
);
1464 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1466 return glsl_type::get_array_instance(new_interface_type
,
1472 * Determine whether the given interface type contains unsized arrays (if
1473 * it doesn't, array_sizing_visitor doesn't need to process it).
1475 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1477 for (unsigned i
= 0; i
< type
->length
; i
++) {
1478 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1479 if (elem_type
->is_unsized_array())
1486 * Create a new interface type based on the given type, with unsized arrays
1487 * replaced by sized arrays whose size is determined by
1488 * max_ifc_array_access.
1490 static const glsl_type
*
1491 resize_interface_members(const glsl_type
*type
,
1492 const unsigned *max_ifc_array_access
,
1495 unsigned num_fields
= type
->length
;
1496 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1497 memcpy(fields
, type
->fields
.structure
,
1498 num_fields
* sizeof(*fields
));
1499 for (unsigned i
= 0; i
< num_fields
; i
++) {
1500 /* If SSBO last member is unsized array, we don't replace it by a sized
1503 if (is_ssbo
&& i
== (num_fields
- 1))
1504 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1507 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1510 glsl_interface_packing packing
=
1511 (glsl_interface_packing
) type
->interface_packing
;
1512 const glsl_type
*new_ifc_type
=
1513 glsl_type::get_interface_instance(fields
, num_fields
,
1514 packing
, type
->name
);
1516 return new_ifc_type
;
1519 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1522 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1523 ir_variable
**interface_vars
= (ir_variable
**) data
;
1524 unsigned num_fields
= ifc_type
->length
;
1525 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1526 memcpy(fields
, ifc_type
->fields
.structure
,
1527 num_fields
* sizeof(*fields
));
1528 bool interface_type_changed
= false;
1529 for (unsigned i
= 0; i
< num_fields
; i
++) {
1530 if (interface_vars
[i
] != NULL
&&
1531 fields
[i
].type
!= interface_vars
[i
]->type
) {
1532 fields
[i
].type
= interface_vars
[i
]->type
;
1533 interface_type_changed
= true;
1536 if (!interface_type_changed
) {
1540 glsl_interface_packing packing
=
1541 (glsl_interface_packing
) ifc_type
->interface_packing
;
1542 const glsl_type
*new_ifc_type
=
1543 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1546 for (unsigned i
= 0; i
< num_fields
; i
++) {
1547 if (interface_vars
[i
] != NULL
)
1548 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1553 * Memory context used to allocate the data in \c unnamed_interfaces.
1558 * Hash table from const glsl_type * to an array of ir_variable *'s
1559 * pointing to the ir_variables constituting each unnamed interface block.
1561 hash_table
*unnamed_interfaces
;
1566 * Performs the cross-validation of tessellation control shader vertices and
1567 * layout qualifiers for the attached tessellation control shaders,
1568 * and propagates them to the linked TCS and linked shader program.
1571 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1572 struct gl_shader
*linked_shader
,
1573 struct gl_shader
**shader_list
,
1574 unsigned num_shaders
)
1576 linked_shader
->TessCtrl
.VerticesOut
= 0;
1578 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1581 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1583 * "All tessellation control shader layout declarations in a program
1584 * must specify the same output patch vertex count. There must be at
1585 * least one layout qualifier specifying an output patch vertex count
1586 * in any program containing tessellation control shaders; however,
1587 * such a declaration is not required in all tessellation control
1591 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1592 struct gl_shader
*shader
= shader_list
[i
];
1594 if (shader
->TessCtrl
.VerticesOut
!= 0) {
1595 if (linked_shader
->TessCtrl
.VerticesOut
!= 0 &&
1596 linked_shader
->TessCtrl
.VerticesOut
!= shader
->TessCtrl
.VerticesOut
) {
1597 linker_error(prog
, "tessellation control shader defined with "
1598 "conflicting output vertex count (%d and %d)\n",
1599 linked_shader
->TessCtrl
.VerticesOut
,
1600 shader
->TessCtrl
.VerticesOut
);
1603 linked_shader
->TessCtrl
.VerticesOut
= shader
->TessCtrl
.VerticesOut
;
1607 /* Just do the intrastage -> interstage propagation right now,
1608 * since we already know we're in the right type of shader program
1611 if (linked_shader
->TessCtrl
.VerticesOut
== 0) {
1612 linker_error(prog
, "tessellation control shader didn't declare "
1613 "vertices out layout qualifier\n");
1616 prog
->TessCtrl
.VerticesOut
= linked_shader
->TessCtrl
.VerticesOut
;
1621 * Performs the cross-validation of tessellation evaluation shader
1622 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1623 * for the attached tessellation evaluation shaders, and propagates them
1624 * to the linked TES and linked shader program.
1627 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1628 struct gl_shader
*linked_shader
,
1629 struct gl_shader
**shader_list
,
1630 unsigned num_shaders
)
1632 linked_shader
->TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1633 linked_shader
->TessEval
.Spacing
= 0;
1634 linked_shader
->TessEval
.VertexOrder
= 0;
1635 linked_shader
->TessEval
.PointMode
= -1;
1637 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1640 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1642 * "At least one tessellation evaluation shader (compilation unit) in
1643 * a program must declare a primitive mode in its input layout.
1644 * Declaration vertex spacing, ordering, and point mode identifiers is
1645 * optional. It is not required that all tessellation evaluation
1646 * shaders in a program declare a primitive mode. If spacing or
1647 * vertex ordering declarations are omitted, the tessellation
1648 * primitive generator will use equal spacing or counter-clockwise
1649 * vertex ordering, respectively. If a point mode declaration is
1650 * omitted, the tessellation primitive generator will produce lines or
1651 * triangles according to the primitive mode."
1654 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1655 struct gl_shader
*shader
= shader_list
[i
];
1657 if (shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1658 if (linked_shader
->TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1659 linked_shader
->TessEval
.PrimitiveMode
!= shader
->TessEval
.PrimitiveMode
) {
1660 linker_error(prog
, "tessellation evaluation shader defined with "
1661 "conflicting input primitive modes.\n");
1664 linked_shader
->TessEval
.PrimitiveMode
= shader
->TessEval
.PrimitiveMode
;
1667 if (shader
->TessEval
.Spacing
!= 0) {
1668 if (linked_shader
->TessEval
.Spacing
!= 0 &&
1669 linked_shader
->TessEval
.Spacing
!= shader
->TessEval
.Spacing
) {
1670 linker_error(prog
, "tessellation evaluation shader defined with "
1671 "conflicting vertex spacing.\n");
1674 linked_shader
->TessEval
.Spacing
= shader
->TessEval
.Spacing
;
1677 if (shader
->TessEval
.VertexOrder
!= 0) {
1678 if (linked_shader
->TessEval
.VertexOrder
!= 0 &&
1679 linked_shader
->TessEval
.VertexOrder
!= shader
->TessEval
.VertexOrder
) {
1680 linker_error(prog
, "tessellation evaluation shader defined with "
1681 "conflicting ordering.\n");
1684 linked_shader
->TessEval
.VertexOrder
= shader
->TessEval
.VertexOrder
;
1687 if (shader
->TessEval
.PointMode
!= -1) {
1688 if (linked_shader
->TessEval
.PointMode
!= -1 &&
1689 linked_shader
->TessEval
.PointMode
!= shader
->TessEval
.PointMode
) {
1690 linker_error(prog
, "tessellation evaluation shader defined with "
1691 "conflicting point modes.\n");
1694 linked_shader
->TessEval
.PointMode
= shader
->TessEval
.PointMode
;
1699 /* Just do the intrastage -> interstage propagation right now,
1700 * since we already know we're in the right type of shader program
1703 if (linked_shader
->TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1705 "tessellation evaluation shader didn't declare input "
1706 "primitive modes.\n");
1709 prog
->TessEval
.PrimitiveMode
= linked_shader
->TessEval
.PrimitiveMode
;
1711 if (linked_shader
->TessEval
.Spacing
== 0)
1712 linked_shader
->TessEval
.Spacing
= GL_EQUAL
;
1713 prog
->TessEval
.Spacing
= linked_shader
->TessEval
.Spacing
;
1715 if (linked_shader
->TessEval
.VertexOrder
== 0)
1716 linked_shader
->TessEval
.VertexOrder
= GL_CCW
;
1717 prog
->TessEval
.VertexOrder
= linked_shader
->TessEval
.VertexOrder
;
1719 if (linked_shader
->TessEval
.PointMode
== -1)
1720 linked_shader
->TessEval
.PointMode
= GL_FALSE
;
1721 prog
->TessEval
.PointMode
= linked_shader
->TessEval
.PointMode
;
1726 * Performs the cross-validation of layout qualifiers specified in
1727 * redeclaration of gl_FragCoord for the attached fragment shaders,
1728 * and propagates them to the linked FS and linked shader program.
1731 link_fs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1732 struct gl_shader
*linked_shader
,
1733 struct gl_shader
**shader_list
,
1734 unsigned num_shaders
)
1736 linked_shader
->redeclares_gl_fragcoord
= false;
1737 linked_shader
->uses_gl_fragcoord
= false;
1738 linked_shader
->origin_upper_left
= false;
1739 linked_shader
->pixel_center_integer
= false;
1741 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1742 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1745 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1746 struct gl_shader
*shader
= shader_list
[i
];
1747 /* From the GLSL 1.50 spec, page 39:
1749 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1750 * it must be redeclared in all the fragment shaders in that program
1751 * that have a static use gl_FragCoord."
1753 if ((linked_shader
->redeclares_gl_fragcoord
1754 && !shader
->redeclares_gl_fragcoord
1755 && shader
->uses_gl_fragcoord
)
1756 || (shader
->redeclares_gl_fragcoord
1757 && !linked_shader
->redeclares_gl_fragcoord
1758 && linked_shader
->uses_gl_fragcoord
)) {
1759 linker_error(prog
, "fragment shader defined with conflicting "
1760 "layout qualifiers for gl_FragCoord\n");
1763 /* From the GLSL 1.50 spec, page 39:
1765 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1766 * single program must have the same set of qualifiers."
1768 if (linked_shader
->redeclares_gl_fragcoord
&& shader
->redeclares_gl_fragcoord
1769 && (shader
->origin_upper_left
!= linked_shader
->origin_upper_left
1770 || shader
->pixel_center_integer
!= linked_shader
->pixel_center_integer
)) {
1771 linker_error(prog
, "fragment shader defined with conflicting "
1772 "layout qualifiers for gl_FragCoord\n");
1775 /* Update the linked shader state. Note that uses_gl_fragcoord should
1776 * accumulate the results. The other values should replace. If there
1777 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1778 * are already known to be the same.
1780 if (shader
->redeclares_gl_fragcoord
|| shader
->uses_gl_fragcoord
) {
1781 linked_shader
->redeclares_gl_fragcoord
=
1782 shader
->redeclares_gl_fragcoord
;
1783 linked_shader
->uses_gl_fragcoord
= linked_shader
->uses_gl_fragcoord
1784 || shader
->uses_gl_fragcoord
;
1785 linked_shader
->origin_upper_left
= shader
->origin_upper_left
;
1786 linked_shader
->pixel_center_integer
= shader
->pixel_center_integer
;
1789 linked_shader
->EarlyFragmentTests
|= shader
->EarlyFragmentTests
;
1794 * Performs the cross-validation of geometry shader max_vertices and
1795 * primitive type layout qualifiers for the attached geometry shaders,
1796 * and propagates them to the linked GS and linked shader program.
1799 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1800 struct gl_shader
*linked_shader
,
1801 struct gl_shader
**shader_list
,
1802 unsigned num_shaders
)
1804 linked_shader
->Geom
.VerticesOut
= 0;
1805 linked_shader
->Geom
.Invocations
= 0;
1806 linked_shader
->Geom
.InputType
= PRIM_UNKNOWN
;
1807 linked_shader
->Geom
.OutputType
= PRIM_UNKNOWN
;
1809 /* No in/out qualifiers defined for anything but GLSL 1.50+
1810 * geometry shaders so far.
1812 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1815 /* From the GLSL 1.50 spec, page 46:
1817 * "All geometry shader output layout declarations in a program
1818 * must declare the same layout and same value for
1819 * max_vertices. There must be at least one geometry output
1820 * layout declaration somewhere in a program, but not all
1821 * geometry shaders (compilation units) are required to
1825 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1826 struct gl_shader
*shader
= shader_list
[i
];
1828 if (shader
->Geom
.InputType
!= PRIM_UNKNOWN
) {
1829 if (linked_shader
->Geom
.InputType
!= PRIM_UNKNOWN
&&
1830 linked_shader
->Geom
.InputType
!= shader
->Geom
.InputType
) {
1831 linker_error(prog
, "geometry shader defined with conflicting "
1835 linked_shader
->Geom
.InputType
= shader
->Geom
.InputType
;
1838 if (shader
->Geom
.OutputType
!= PRIM_UNKNOWN
) {
1839 if (linked_shader
->Geom
.OutputType
!= PRIM_UNKNOWN
&&
1840 linked_shader
->Geom
.OutputType
!= shader
->Geom
.OutputType
) {
1841 linker_error(prog
, "geometry shader defined with conflicting "
1845 linked_shader
->Geom
.OutputType
= shader
->Geom
.OutputType
;
1848 if (shader
->Geom
.VerticesOut
!= 0) {
1849 if (linked_shader
->Geom
.VerticesOut
!= 0 &&
1850 linked_shader
->Geom
.VerticesOut
!= shader
->Geom
.VerticesOut
) {
1851 linker_error(prog
, "geometry shader defined with conflicting "
1852 "output vertex count (%d and %d)\n",
1853 linked_shader
->Geom
.VerticesOut
,
1854 shader
->Geom
.VerticesOut
);
1857 linked_shader
->Geom
.VerticesOut
= shader
->Geom
.VerticesOut
;
1860 if (shader
->Geom
.Invocations
!= 0) {
1861 if (linked_shader
->Geom
.Invocations
!= 0 &&
1862 linked_shader
->Geom
.Invocations
!= shader
->Geom
.Invocations
) {
1863 linker_error(prog
, "geometry shader defined with conflicting "
1864 "invocation count (%d and %d)\n",
1865 linked_shader
->Geom
.Invocations
,
1866 shader
->Geom
.Invocations
);
1869 linked_shader
->Geom
.Invocations
= shader
->Geom
.Invocations
;
1873 /* Just do the intrastage -> interstage propagation right now,
1874 * since we already know we're in the right type of shader program
1877 if (linked_shader
->Geom
.InputType
== PRIM_UNKNOWN
) {
1879 "geometry shader didn't declare primitive input type\n");
1882 prog
->Geom
.InputType
= linked_shader
->Geom
.InputType
;
1884 if (linked_shader
->Geom
.OutputType
== PRIM_UNKNOWN
) {
1886 "geometry shader didn't declare primitive output type\n");
1889 prog
->Geom
.OutputType
= linked_shader
->Geom
.OutputType
;
1891 if (linked_shader
->Geom
.VerticesOut
== 0) {
1893 "geometry shader didn't declare max_vertices\n");
1896 prog
->Geom
.VerticesOut
= linked_shader
->Geom
.VerticesOut
;
1898 if (linked_shader
->Geom
.Invocations
== 0)
1899 linked_shader
->Geom
.Invocations
= 1;
1901 prog
->Geom
.Invocations
= linked_shader
->Geom
.Invocations
;
1906 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1907 * qualifiers for the attached compute shaders, and propagate them to the
1908 * linked CS and linked shader program.
1911 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1912 struct gl_shader
*linked_shader
,
1913 struct gl_shader
**shader_list
,
1914 unsigned num_shaders
)
1916 for (int i
= 0; i
< 3; i
++)
1917 linked_shader
->Comp
.LocalSize
[i
] = 0;
1919 /* This function is called for all shader stages, but it only has an effect
1920 * for compute shaders.
1922 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1925 /* From the ARB_compute_shader spec, in the section describing local size
1928 * If multiple compute shaders attached to a single program object
1929 * declare local work-group size, the declarations must be identical;
1930 * otherwise a link-time error results. Furthermore, if a program
1931 * object contains any compute shaders, at least one must contain an
1932 * input layout qualifier specifying the local work sizes of the
1933 * program, or a link-time error will occur.
1935 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
1936 struct gl_shader
*shader
= shader_list
[sh
];
1938 if (shader
->Comp
.LocalSize
[0] != 0) {
1939 if (linked_shader
->Comp
.LocalSize
[0] != 0) {
1940 for (int i
= 0; i
< 3; i
++) {
1941 if (linked_shader
->Comp
.LocalSize
[i
] !=
1942 shader
->Comp
.LocalSize
[i
]) {
1943 linker_error(prog
, "compute shader defined with conflicting "
1949 for (int i
= 0; i
< 3; i
++)
1950 linked_shader
->Comp
.LocalSize
[i
] = shader
->Comp
.LocalSize
[i
];
1954 /* Just do the intrastage -> interstage propagation right now,
1955 * since we already know we're in the right type of shader program
1958 if (linked_shader
->Comp
.LocalSize
[0] == 0) {
1959 linker_error(prog
, "compute shader didn't declare local size\n");
1962 for (int i
= 0; i
< 3; i
++)
1963 prog
->Comp
.LocalSize
[i
] = linked_shader
->Comp
.LocalSize
[i
];
1968 * Combine a group of shaders for a single stage to generate a linked shader
1971 * If this function is supplied a single shader, it is cloned, and the new
1972 * shader is returned.
1974 static struct gl_shader
*
1975 link_intrastage_shaders(void *mem_ctx
,
1976 struct gl_context
*ctx
,
1977 struct gl_shader_program
*prog
,
1978 struct gl_shader
**shader_list
,
1979 unsigned num_shaders
)
1981 struct gl_uniform_block
*uniform_blocks
= NULL
;
1983 /* Check that global variables defined in multiple shaders are consistent.
1985 cross_validate_globals(prog
, shader_list
, num_shaders
, false);
1986 if (!prog
->LinkStatus
)
1989 /* Check that interface blocks defined in multiple shaders are consistent.
1991 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
1993 if (!prog
->LinkStatus
)
1996 /* Link up uniform blocks defined within this stage. */
1997 const unsigned num_uniform_blocks
=
1998 link_uniform_blocks(mem_ctx
, ctx
, prog
, shader_list
, num_shaders
,
2000 if (!prog
->LinkStatus
)
2003 /* Check that there is only a single definition of each function signature
2004 * across all shaders.
2006 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2007 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2008 ir_function
*const f
= node
->as_function();
2013 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2014 ir_function
*const other
=
2015 shader_list
[j
]->symbols
->get_function(f
->name
);
2017 /* If the other shader has no function (and therefore no function
2018 * signatures) with the same name, skip to the next shader.
2023 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2024 if (!sig
->is_defined
|| sig
->is_builtin())
2027 ir_function_signature
*other_sig
=
2028 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2030 if ((other_sig
!= NULL
) && other_sig
->is_defined
2031 && !other_sig
->is_builtin()) {
2032 linker_error(prog
, "function `%s' is multiply defined\n",
2041 /* Find the shader that defines main, and make a clone of it.
2043 * Starting with the clone, search for undefined references. If one is
2044 * found, find the shader that defines it. Clone the reference and add
2045 * it to the shader. Repeat until there are no undefined references or
2046 * until a reference cannot be resolved.
2048 gl_shader
*main
= NULL
;
2049 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2050 if (_mesa_get_main_function_signature(shader_list
[i
]) != NULL
) {
2051 main
= shader_list
[i
];
2057 linker_error(prog
, "%s shader lacks `main'\n",
2058 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2062 gl_shader
*linked
= ctx
->Driver
.NewShader(NULL
, 0, main
->Type
);
2063 linked
->ir
= new(linked
) exec_list
;
2064 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2066 linked
->BufferInterfaceBlocks
= uniform_blocks
;
2067 linked
->NumBufferInterfaceBlocks
= num_uniform_blocks
;
2068 ralloc_steal(linked
, linked
->BufferInterfaceBlocks
);
2070 link_fs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2071 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2072 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2073 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2074 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2076 populate_symbol_table(linked
);
2078 /* The pointer to the main function in the final linked shader (i.e., the
2079 * copy of the original shader that contained the main function).
2081 ir_function_signature
*const main_sig
=
2082 _mesa_get_main_function_signature(linked
);
2084 /* Move any instructions other than variable declarations or function
2085 * declarations into main.
2087 exec_node
*insertion_point
=
2088 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2091 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2092 if (shader_list
[i
] == main
)
2095 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2096 insertion_point
, true, linked
);
2099 /* Check if any shader needs built-in functions. */
2100 bool need_builtins
= false;
2101 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2102 if (shader_list
[i
]->uses_builtin_functions
) {
2103 need_builtins
= true;
2109 if (need_builtins
) {
2110 /* Make a temporary array one larger than shader_list, which will hold
2111 * the built-in function shader as well.
2113 gl_shader
**linking_shaders
= (gl_shader
**)
2114 calloc(num_shaders
+ 1, sizeof(gl_shader
*));
2116 ok
= linking_shaders
!= NULL
;
2119 memcpy(linking_shaders
, shader_list
, num_shaders
* sizeof(gl_shader
*));
2120 linking_shaders
[num_shaders
] = _mesa_glsl_get_builtin_function_shader();
2122 ok
= link_function_calls(prog
, linked
, linking_shaders
, num_shaders
+ 1);
2124 free(linking_shaders
);
2126 _mesa_error_no_memory(__func__
);
2129 ok
= link_function_calls(prog
, linked
, shader_list
, num_shaders
);
2134 _mesa_delete_shader(ctx
, linked
);
2138 /* At this point linked should contain all of the linked IR, so
2139 * validate it to make sure nothing went wrong.
2141 validate_ir_tree(linked
->ir
);
2143 /* Set the size of geometry shader input arrays */
2144 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2145 unsigned num_vertices
= vertices_per_prim(prog
->Geom
.InputType
);
2146 geom_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2147 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2148 ir
->accept(&input_resize_visitor
);
2152 if (ctx
->Const
.VertexID_is_zero_based
)
2153 lower_vertex_id(linked
);
2155 /* Validate correct usage of barrier() in the tess control shader */
2156 if (linked
->Stage
== MESA_SHADER_TESS_CTRL
) {
2157 barrier_use_visitor
visitor(prog
);
2158 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2159 ir
->accept(&visitor
);
2163 /* Make a pass over all variable declarations to ensure that arrays with
2164 * unspecified sizes have a size specified. The size is inferred from the
2165 * max_array_access field.
2167 array_sizing_visitor v
;
2169 v
.fixup_unnamed_interface_types();
2175 * Update the sizes of linked shader uniform arrays to the maximum
2178 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2180 * If one or more elements of an array are active,
2181 * GetActiveUniform will return the name of the array in name,
2182 * subject to the restrictions listed above. The type of the array
2183 * is returned in type. The size parameter contains the highest
2184 * array element index used, plus one. The compiler or linker
2185 * determines the highest index used. There will be only one
2186 * active uniform reported by the GL per uniform array.
2190 update_array_sizes(struct gl_shader_program
*prog
)
2192 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2193 if (prog
->_LinkedShaders
[i
] == NULL
)
2196 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2197 ir_variable
*const var
= node
->as_variable();
2199 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2200 !var
->type
->is_array())
2203 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2204 * will not be eliminated. Since we always do std140, just
2205 * don't resize arrays in UBOs.
2207 * Atomic counters are supposed to get deterministic
2208 * locations assigned based on the declaration ordering and
2209 * sizes, array compaction would mess that up.
2211 * Subroutine uniforms are not removed.
2213 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2214 var
->type
->contains_subroutine())
2217 unsigned int size
= var
->data
.max_array_access
;
2218 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2219 if (prog
->_LinkedShaders
[j
] == NULL
)
2222 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2223 ir_variable
*other_var
= node2
->as_variable();
2227 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2228 other_var
->data
.max_array_access
> size
) {
2229 size
= other_var
->data
.max_array_access
;
2234 if (size
+ 1 != var
->type
->length
) {
2235 /* If this is a built-in uniform (i.e., it's backed by some
2236 * fixed-function state), adjust the number of state slots to
2237 * match the new array size. The number of slots per array entry
2238 * is not known. It seems safe to assume that the total number of
2239 * slots is an integer multiple of the number of array elements.
2240 * Determine the number of slots per array element by dividing by
2241 * the old (total) size.
2243 const unsigned num_slots
= var
->get_num_state_slots();
2244 if (num_slots
> 0) {
2245 var
->set_num_state_slots((size
+ 1)
2246 * (num_slots
/ var
->type
->length
));
2249 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2251 /* FINISHME: We should update the types of array
2252 * dereferences of this variable now.
2260 * Resize tessellation evaluation per-vertex inputs to the size of
2261 * tessellation control per-vertex outputs.
2264 resize_tes_inputs(struct gl_context
*ctx
,
2265 struct gl_shader_program
*prog
)
2267 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2270 gl_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2271 gl_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2273 /* If no control shader is present, then the TES inputs are statically
2274 * sized to MaxPatchVertices; the actual size of the arrays won't be
2275 * known until draw time.
2277 const int num_vertices
= tcs
2278 ? tcs
->TessCtrl
.VerticesOut
2279 : ctx
->Const
.MaxPatchVertices
;
2281 tess_eval_array_resize_visitor
input_resize_visitor(num_vertices
, prog
);
2282 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2283 ir
->accept(&input_resize_visitor
);
2287 /* Convert the gl_PatchVerticesIn system value into a constant, since
2288 * the value is known at this point.
2290 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2291 ir_variable
*var
= ir
->as_variable();
2292 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2293 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2294 void *mem_ctx
= ralloc_parent(var
);
2295 var
->data
.mode
= ir_var_auto
;
2296 var
->data
.location
= 0;
2297 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2304 * Find a contiguous set of available bits in a bitmask.
2306 * \param used_mask Bits representing used (1) and unused (0) locations
2307 * \param needed_count Number of contiguous bits needed.
2310 * Base location of the available bits on success or -1 on failure.
2313 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2315 unsigned needed_mask
= (1 << needed_count
) - 1;
2316 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2318 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2319 * cannot optimize possibly infinite loops" for the loop below.
2321 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2324 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2325 if ((needed_mask
& ~used_mask
) == needed_mask
)
2336 * Assign locations for either VS inputs or FS outputs
2338 * \param prog Shader program whose variables need locations assigned
2339 * \param constants Driver specific constant values for the program.
2340 * \param target_index Selector for the program target to receive location
2341 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2342 * \c MESA_SHADER_FRAGMENT.
2345 * If locations are successfully assigned, true is returned. Otherwise an
2346 * error is emitted to the shader link log and false is returned.
2349 assign_attribute_or_color_locations(gl_shader_program
*prog
,
2350 struct gl_constants
*constants
,
2351 unsigned target_index
)
2353 /* Maximum number of generic locations. This corresponds to either the
2354 * maximum number of draw buffers or the maximum number of generic
2357 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2358 constants
->Program
[target_index
].MaxAttribs
:
2359 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2361 /* Mark invalid locations as being used.
2363 unsigned used_locations
= (max_index
>= 32)
2364 ? ~0 : ~((1 << max_index
) - 1);
2365 unsigned double_storage_locations
= 0;
2367 assert((target_index
== MESA_SHADER_VERTEX
)
2368 || (target_index
== MESA_SHADER_FRAGMENT
));
2370 gl_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2374 /* Operate in a total of four passes.
2376 * 1. Invalidate the location assignments for all vertex shader inputs.
2378 * 2. Assign locations for inputs that have user-defined (via
2379 * glBindVertexAttribLocation) locations and outputs that have
2380 * user-defined locations (via glBindFragDataLocation).
2382 * 3. Sort the attributes without assigned locations by number of slots
2383 * required in decreasing order. Fragmentation caused by attribute
2384 * locations assigned by the application may prevent large attributes
2385 * from having enough contiguous space.
2387 * 4. Assign locations to any inputs without assigned locations.
2390 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2391 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2393 const enum ir_variable_mode direction
=
2394 (target_index
== MESA_SHADER_VERTEX
)
2395 ? ir_var_shader_in
: ir_var_shader_out
;
2398 /* Temporary storage for the set of attributes that need locations assigned.
2404 /* Used below in the call to qsort. */
2405 static int compare(const void *a
, const void *b
)
2407 const temp_attr
*const l
= (const temp_attr
*) a
;
2408 const temp_attr
*const r
= (const temp_attr
*) b
;
2410 /* Reversed because we want a descending order sort below. */
2411 return r
->slots
- l
->slots
;
2415 unsigned num_attr
= 0;
2417 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2418 ir_variable
*const var
= node
->as_variable();
2420 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2423 if (var
->data
.explicit_location
) {
2424 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2425 || (var
->data
.location
< 0)) {
2427 "invalid explicit location %d specified for `%s'\n",
2428 (var
->data
.location
< 0)
2429 ? var
->data
.location
2430 : var
->data
.location
- generic_base
,
2434 } else if (target_index
== MESA_SHADER_VERTEX
) {
2437 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2438 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2439 var
->data
.location
= binding
;
2440 var
->data
.is_unmatched_generic_inout
= 0;
2442 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2446 if (prog
->FragDataBindings
->get(binding
, var
->name
)) {
2447 assert(binding
>= FRAG_RESULT_DATA0
);
2448 var
->data
.location
= binding
;
2449 var
->data
.is_unmatched_generic_inout
= 0;
2451 if (prog
->FragDataIndexBindings
->get(index
, var
->name
)) {
2452 var
->data
.index
= index
;
2457 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2459 * "Output binding assignments will cause LinkProgram to fail:
2461 * If the program has an active output assigned to a location greater
2462 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2463 * an active output assigned an index greater than or equal to one;"
2465 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2466 var
->data
.location
- generic_base
>=
2467 (int) constants
->MaxDualSourceDrawBuffers
) {
2469 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2470 "with index %u for %s\n",
2471 var
->data
.location
- generic_base
, var
->data
.index
,
2476 const unsigned slots
= var
->type
->count_attribute_slots();
2478 /* If the variable is not a built-in and has a location statically
2479 * assigned in the shader (presumably via a layout qualifier), make sure
2480 * that it doesn't collide with other assigned locations. Otherwise,
2481 * add it to the list of variables that need linker-assigned locations.
2483 if (var
->data
.location
!= -1) {
2484 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2485 /* From page 61 of the OpenGL 4.0 spec:
2487 * "LinkProgram will fail if the attribute bindings assigned
2488 * by BindAttribLocation do not leave not enough space to
2489 * assign a location for an active matrix attribute or an
2490 * active attribute array, both of which require multiple
2491 * contiguous generic attributes."
2493 * I think above text prohibits the aliasing of explicit and
2494 * automatic assignments. But, aliasing is allowed in manual
2495 * assignments of attribute locations. See below comments for
2498 * From OpenGL 4.0 spec, page 61:
2500 * "It is possible for an application to bind more than one
2501 * attribute name to the same location. This is referred to as
2502 * aliasing. This will only work if only one of the aliased
2503 * attributes is active in the executable program, or if no
2504 * path through the shader consumes more than one attribute of
2505 * a set of attributes aliased to the same location. A link
2506 * error can occur if the linker determines that every path
2507 * through the shader consumes multiple aliased attributes,
2508 * but implementations are not required to generate an error
2511 * From GLSL 4.30 spec, page 54:
2513 * "A program will fail to link if any two non-vertex shader
2514 * input variables are assigned to the same location. For
2515 * vertex shaders, multiple input variables may be assigned
2516 * to the same location using either layout qualifiers or via
2517 * the OpenGL API. However, such aliasing is intended only to
2518 * support vertex shaders where each execution path accesses
2519 * at most one input per each location. Implementations are
2520 * permitted, but not required, to generate link-time errors
2521 * if they detect that every path through the vertex shader
2522 * executable accesses multiple inputs assigned to any single
2523 * location. For all shader types, a program will fail to link
2524 * if explicit location assignments leave the linker unable
2525 * to find space for other variables without explicit
2528 * From OpenGL ES 3.0 spec, page 56:
2530 * "Binding more than one attribute name to the same location
2531 * is referred to as aliasing, and is not permitted in OpenGL
2532 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2533 * fail when this condition exists. However, aliasing is
2534 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2535 * This will only work if only one of the aliased attributes
2536 * is active in the executable program, or if no path through
2537 * the shader consumes more than one attribute of a set of
2538 * attributes aliased to the same location. A link error can
2539 * occur if the linker determines that every path through the
2540 * shader consumes multiple aliased attributes, but implemen-
2541 * tations are not required to generate an error in this case."
2543 * After looking at above references from OpenGL, OpenGL ES and
2544 * GLSL specifications, we allow aliasing of vertex input variables
2545 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2547 * NOTE: This is not required by the spec but its worth mentioning
2548 * here that we're not doing anything to make sure that no path
2549 * through the vertex shader executable accesses multiple inputs
2550 * assigned to any single location.
2553 /* Mask representing the contiguous slots that will be used by
2556 const unsigned attr
= var
->data
.location
- generic_base
;
2557 const unsigned use_mask
= (1 << slots
) - 1;
2558 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2559 ? "vertex shader input" : "fragment shader output";
2561 /* Generate a link error if the requested locations for this
2562 * attribute exceed the maximum allowed attribute location.
2564 if (attr
+ slots
> max_index
) {
2566 "insufficient contiguous locations "
2567 "available for %s `%s' %d %d %d\n", string
,
2568 var
->name
, used_locations
, use_mask
, attr
);
2572 /* Generate a link error if the set of bits requested for this
2573 * attribute overlaps any previously allocated bits.
2575 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2576 if (target_index
== MESA_SHADER_FRAGMENT
||
2577 (prog
->IsES
&& prog
->Version
>= 300)) {
2579 "overlapping location is assigned "
2580 "to %s `%s' %d %d %d\n", string
,
2581 var
->name
, used_locations
, use_mask
, attr
);
2584 linker_warning(prog
,
2585 "overlapping location is assigned "
2586 "to %s `%s' %d %d %d\n", string
,
2587 var
->name
, used_locations
, use_mask
, attr
);
2591 used_locations
|= (use_mask
<< attr
);
2593 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2595 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2596 * active attribute variables may fail to link, unless
2597 * device-dependent optimizations are able to make the program
2598 * fit within available hardware resources. For the purposes
2599 * of this test, attribute variables of the type dvec3, dvec4,
2600 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2601 * count as consuming twice as many attributes as equivalent
2602 * single-precision types. While these types use the same number
2603 * of generic attributes as their single-precision equivalents,
2604 * implementations are permitted to consume two single-precision
2605 * vectors of internal storage for each three- or four-component
2606 * double-precision vector."
2608 * Mark this attribute slot as taking up twice as much space
2609 * so we can count it properly against limits. According to
2610 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2611 * is optional behavior, but it seems preferable.
2613 const glsl_type
*type
= var
->type
->without_array();
2614 if (type
== glsl_type::dvec3_type
||
2615 type
== glsl_type::dvec4_type
||
2616 type
== glsl_type::dmat2x3_type
||
2617 type
== glsl_type::dmat2x4_type
||
2618 type
== glsl_type::dmat3_type
||
2619 type
== glsl_type::dmat3x4_type
||
2620 type
== glsl_type::dmat4x3_type
||
2621 type
== glsl_type::dmat4_type
) {
2622 double_storage_locations
|= (use_mask
<< attr
);
2629 to_assign
[num_attr
].slots
= slots
;
2630 to_assign
[num_attr
].var
= var
;
2634 if (target_index
== MESA_SHADER_VERTEX
) {
2635 unsigned total_attribs_size
=
2636 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2637 _mesa_bitcount(double_storage_locations
);
2638 if (total_attribs_size
> max_index
) {
2640 "attempt to use %d vertex attribute slots only %d available ",
2641 total_attribs_size
, max_index
);
2646 /* If all of the attributes were assigned locations by the application (or
2647 * are built-in attributes with fixed locations), return early. This should
2648 * be the common case.
2653 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2655 if (target_index
== MESA_SHADER_VERTEX
) {
2656 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2657 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2658 * reserved to prevent it from being automatically allocated below.
2660 find_deref_visitor
find("gl_Vertex");
2662 if (find
.variable_found())
2663 used_locations
|= (1 << 0);
2666 for (unsigned i
= 0; i
< num_attr
; i
++) {
2667 /* Mask representing the contiguous slots that will be used by this
2670 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2672 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2675 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2676 ? "vertex shader input" : "fragment shader output";
2679 "insufficient contiguous locations "
2680 "available for %s `%s'\n",
2681 string
, to_assign
[i
].var
->name
);
2685 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2686 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2687 used_locations
|= (use_mask
<< location
);
2695 * Demote shader inputs and outputs that are not used in other stages
2698 demote_shader_inputs_and_outputs(gl_shader
*sh
, enum ir_variable_mode mode
)
2700 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2701 ir_variable
*const var
= node
->as_variable();
2703 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
2706 /* A shader 'in' or 'out' variable is only really an input or output if
2707 * its value is used by other shader stages. This will cause the variable
2708 * to have a location assigned.
2710 if (var
->data
.is_unmatched_generic_inout
) {
2711 assert(var
->data
.mode
!= ir_var_temporary
);
2712 var
->data
.mode
= ir_var_auto
;
2719 * Store the gl_FragDepth layout in the gl_shader_program struct.
2722 store_fragdepth_layout(struct gl_shader_program
*prog
)
2724 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2728 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2730 /* We don't look up the gl_FragDepth symbol directly because if
2731 * gl_FragDepth is not used in the shader, it's removed from the IR.
2732 * However, the symbol won't be removed from the symbol table.
2734 * We're only interested in the cases where the variable is NOT removed
2737 foreach_in_list(ir_instruction
, node
, ir
) {
2738 ir_variable
*const var
= node
->as_variable();
2740 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2744 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2745 switch (var
->data
.depth_layout
) {
2746 case ir_depth_layout_none
:
2747 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2749 case ir_depth_layout_any
:
2750 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2752 case ir_depth_layout_greater
:
2753 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2755 case ir_depth_layout_less
:
2756 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2758 case ir_depth_layout_unchanged
:
2759 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2770 * Validate the resources used by a program versus the implementation limits
2773 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2775 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2776 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2781 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2782 linker_error(prog
, "Too many %s shader texture samplers\n",
2783 _mesa_shader_stage_to_string(i
));
2786 if (sh
->num_uniform_components
>
2787 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2788 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2789 linker_warning(prog
, "Too many %s shader default uniform block "
2790 "components, but the driver will try to optimize "
2791 "them out; this is non-portable out-of-spec "
2793 _mesa_shader_stage_to_string(i
));
2795 linker_error(prog
, "Too many %s shader default uniform block "
2797 _mesa_shader_stage_to_string(i
));
2801 if (sh
->num_combined_uniform_components
>
2802 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2803 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2804 linker_warning(prog
, "Too many %s shader uniform components, "
2805 "but the driver will try to optimize them out; "
2806 "this is non-portable out-of-spec behavior\n",
2807 _mesa_shader_stage_to_string(i
));
2809 linker_error(prog
, "Too many %s shader uniform components\n",
2810 _mesa_shader_stage_to_string(i
));
2815 unsigned blocks
[MESA_SHADER_STAGES
] = {0};
2816 unsigned total_uniform_blocks
= 0;
2817 unsigned shader_blocks
[MESA_SHADER_STAGES
] = {0};
2818 unsigned total_shader_storage_blocks
= 0;
2820 for (unsigned i
= 0; i
< prog
->NumBufferInterfaceBlocks
; i
++) {
2821 /* Don't check SSBOs for Uniform Block Size */
2822 if (!prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2823 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxUniformBlockSize
) {
2824 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
2825 prog
->BufferInterfaceBlocks
[i
].Name
,
2826 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2827 ctx
->Const
.MaxUniformBlockSize
);
2830 if (prog
->BufferInterfaceBlocks
[i
].IsShaderStorage
&&
2831 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
> ctx
->Const
.MaxShaderStorageBlockSize
) {
2832 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
2833 prog
->BufferInterfaceBlocks
[i
].Name
,
2834 prog
->BufferInterfaceBlocks
[i
].UniformBufferSize
,
2835 ctx
->Const
.MaxShaderStorageBlockSize
);
2838 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2839 if (prog
->InterfaceBlockStageIndex
[j
][i
] != -1) {
2840 struct gl_shader
*sh
= prog
->_LinkedShaders
[j
];
2841 int stage_index
= prog
->InterfaceBlockStageIndex
[j
][i
];
2842 if (sh
&& sh
->BufferInterfaceBlocks
[stage_index
].IsShaderStorage
) {
2844 total_shader_storage_blocks
++;
2847 total_uniform_blocks
++;
2852 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
2853 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
2854 total_uniform_blocks
,
2855 ctx
->Const
.MaxCombinedUniformBlocks
);
2857 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2858 const unsigned max_uniform_blocks
=
2859 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
2860 if (blocks
[i
] > max_uniform_blocks
) {
2861 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
2862 _mesa_shader_stage_to_string(i
),
2864 max_uniform_blocks
);
2870 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
2871 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
2872 total_shader_storage_blocks
,
2873 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
2875 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2876 const unsigned max_shader_storage_blocks
=
2877 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
2878 if (shader_blocks
[i
] > max_shader_storage_blocks
) {
2879 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
2880 _mesa_shader_stage_to_string(i
),
2882 max_shader_storage_blocks
);
2891 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
2893 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2894 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2899 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
2900 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
2906 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
2907 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
2908 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
2909 if (fn
->types
[k
] == uni
->type
) {
2915 uni
->num_compatible_subroutines
= count
;
2921 check_subroutine_resources(struct gl_shader_program
*prog
)
2923 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2924 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2927 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
2928 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
2929 _mesa_shader_stage_to_string(i
));
2934 * Validate shader image resources.
2937 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2939 unsigned total_image_units
= 0;
2940 unsigned fragment_outputs
= 0;
2941 unsigned total_shader_storage_blocks
= 0;
2943 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
2946 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2947 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
2950 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
2951 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
2952 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
2953 ctx
->Const
.Program
[i
].MaxImageUniforms
);
2955 total_image_units
+= sh
->NumImages
;
2957 for (unsigned j
= 0; j
< prog
->NumBufferInterfaceBlocks
; j
++) {
2958 int stage_index
= prog
->InterfaceBlockStageIndex
[i
][j
];
2959 if (stage_index
!= -1 && sh
->BufferInterfaceBlocks
[stage_index
].IsShaderStorage
)
2960 total_shader_storage_blocks
++;
2963 if (i
== MESA_SHADER_FRAGMENT
) {
2964 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2965 ir_variable
*var
= node
->as_variable();
2966 if (var
&& var
->data
.mode
== ir_var_shader_out
)
2967 fragment_outputs
+= var
->type
->count_attribute_slots();
2973 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
2974 linker_error(prog
, "Too many combined image uniforms\n");
2976 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
2977 ctx
->Const
.MaxCombinedShaderOutputResources
)
2978 linker_error(prog
, "Too many combined image uniforms, shader storage "
2979 " buffers and fragment outputs\n");
2984 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
2985 * for a variable, checks for overlaps between other uniforms using explicit
2989 reserve_explicit_locations(struct gl_shader_program
*prog
,
2990 string_to_uint_map
*map
, ir_variable
*var
)
2992 unsigned slots
= var
->type
->uniform_locations();
2993 unsigned max_loc
= var
->data
.location
+ slots
- 1;
2995 /* Resize remap table if locations do not fit in the current one. */
2996 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
2997 prog
->UniformRemapTable
=
2998 reralloc(prog
, prog
->UniformRemapTable
,
2999 gl_uniform_storage
*,
3002 if (!prog
->UniformRemapTable
) {
3003 linker_error(prog
, "Out of memory during linking.\n");
3007 /* Initialize allocated space. */
3008 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3009 prog
->UniformRemapTable
[i
] = NULL
;
3011 prog
->NumUniformRemapTable
= max_loc
+ 1;
3014 for (unsigned i
= 0; i
< slots
; i
++) {
3015 unsigned loc
= var
->data
.location
+ i
;
3017 /* Check if location is already used. */
3018 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3020 /* Possibly same uniform from a different stage, this is ok. */
3022 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
)
3025 /* ARB_explicit_uniform_location specification states:
3027 * "No two default-block uniform variables in the program can have
3028 * the same location, even if they are unused, otherwise a compiler
3029 * or linker error will be generated."
3032 "location qualifier for uniform %s overlaps "
3033 "previously used location\n",
3038 /* Initialize location as inactive before optimization
3039 * rounds and location assignment.
3041 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3044 /* Note, base location used for arrays. */
3045 map
->put(var
->data
.location
, var
->name
);
3051 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3052 struct gl_shader
*sh
,
3055 unsigned slots
= var
->type
->uniform_locations();
3056 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3058 /* Resize remap table if locations do not fit in the current one. */
3059 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3060 sh
->SubroutineUniformRemapTable
=
3061 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3062 gl_uniform_storage
*,
3065 if (!sh
->SubroutineUniformRemapTable
) {
3066 linker_error(prog
, "Out of memory during linking.\n");
3070 /* Initialize allocated space. */
3071 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3072 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3074 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3077 for (unsigned i
= 0; i
< slots
; i
++) {
3078 unsigned loc
= var
->data
.location
+ i
;
3080 /* Check if location is already used. */
3081 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3083 /* ARB_explicit_uniform_location specification states:
3084 * "No two subroutine uniform variables can have the same location
3085 * in the same shader stage, otherwise a compiler or linker error
3086 * will be generated."
3089 "location qualifier for uniform %s overlaps "
3090 "previously used location\n",
3095 /* Initialize location as inactive before optimization
3096 * rounds and location assignment.
3098 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3104 * Check and reserve all explicit uniform locations, called before
3105 * any optimizations happen to handle also inactive uniforms and
3106 * inactive array elements that may get trimmed away.
3109 check_explicit_uniform_locations(struct gl_context
*ctx
,
3110 struct gl_shader_program
*prog
)
3112 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3115 /* This map is used to detect if overlapping explicit locations
3116 * occur with the same uniform (from different stage) or a different one.
3118 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3121 linker_error(prog
, "Out of memory during linking.\n");
3125 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3126 struct gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3131 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3132 ir_variable
*var
= node
->as_variable();
3133 if (var
&& (var
->data
.mode
== ir_var_uniform
&&
3134 var
->data
.explicit_location
)) {
3136 if (var
->type
->is_subroutine())
3137 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3139 ret
= reserve_explicit_locations(prog
, uniform_map
, var
);
3152 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3153 GLenum type
, const char *name
)
3155 bool found_interface
= false;
3156 unsigned block_name_len
= 0;
3157 const char *block_name_dot
= strchr(name
, '.');
3159 /* These rules only apply to buffer variables. So we return
3160 * true for the rest of types.
3162 if (type
!= GL_BUFFER_VARIABLE
)
3165 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3166 const char *block_name
= shProg
->BufferInterfaceBlocks
[i
].Name
;
3167 block_name_len
= strlen(block_name
);
3169 const char *block_square_bracket
= strchr(block_name
, '[');
3170 if (block_square_bracket
) {
3171 /* The block is part of an array of named interfaces,
3172 * for the name comparison we ignore the "[x]" part.
3174 block_name_len
-= strlen(block_square_bracket
);
3177 if (block_name_dot
) {
3178 /* Check if the variable name starts with the interface
3179 * name. The interface name (if present) should have the
3180 * length than the interface block name we are comparing to.
3182 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3183 if (len
!= block_name_len
)
3187 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3188 found_interface
= true;
3193 /* We remove the interface name from the buffer variable name,
3194 * including the dot that follows it.
3196 if (found_interface
)
3197 name
= name
+ block_name_len
+ 1;
3199 /* From: ARB_program_interface_query extension:
3201 * "For an active shader storage block member declared as an array, an
3202 * entry will be generated only for the first array element, regardless
3203 * of its type. For arrays of aggregate types, the enumeration rules are
3204 * applied recursively for the single enumerated array element.
3206 const char *struct_first_dot
= strchr(name
, '.');
3207 const char *first_square_bracket
= strchr(name
, '[');
3209 /* The buffer variable is on top level and it is not an array */
3210 if (!first_square_bracket
) {
3212 /* The shader storage block member is a struct, then generate the entry */
3213 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3216 /* Shader storage block member is an array, only generate an entry for the
3217 * first array element.
3219 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3227 add_program_resource(struct gl_shader_program
*prog
, GLenum type
,
3228 const void *data
, uint8_t stages
)
3232 /* If resource already exists, do not add it again. */
3233 for (unsigned i
= 0; i
< prog
->NumProgramResourceList
; i
++)
3234 if (prog
->ProgramResourceList
[i
].Data
== data
)
3237 prog
->ProgramResourceList
=
3239 prog
->ProgramResourceList
,
3240 gl_program_resource
,
3241 prog
->NumProgramResourceList
+ 1);
3243 if (!prog
->ProgramResourceList
) {
3244 linker_error(prog
, "Out of memory during linking.\n");
3248 struct gl_program_resource
*res
=
3249 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3253 res
->StageReferences
= stages
;
3255 prog
->NumProgramResourceList
++;
3260 /* Function checks if a variable var is a packed varying and
3261 * if given name is part of packed varying's list.
3263 * If a variable is a packed varying, it has a name like
3264 * 'packed:a,b,c' where a, b and c are separate variables.
3267 included_in_packed_varying(ir_variable
*var
, const char *name
)
3269 if (strncmp(var
->name
, "packed:", 7) != 0)
3272 char *list
= strdup(var
->name
+ 7);
3277 char *token
= strtok_r(list
, ",", &saveptr
);
3279 if (strcmp(token
, name
) == 0) {
3283 token
= strtok_r(NULL
, ",", &saveptr
);
3290 * Function builds a stage reference bitmask from variable name.
3293 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3298 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3299 * used for reference mask in gl_program_resource will need to be changed.
3301 assert(MESA_SHADER_STAGES
< 8);
3303 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3304 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3308 /* Shader symbol table may contain variables that have
3309 * been optimized away. Search IR for the variable instead.
3311 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3312 ir_variable
*var
= node
->as_variable();
3314 unsigned baselen
= strlen(var
->name
);
3316 if (included_in_packed_varying(var
, name
)) {
3321 /* Type needs to match if specified, otherwise we might
3322 * pick a variable with same name but different interface.
3324 if (var
->data
.mode
!= mode
)
3327 if (strncmp(var
->name
, name
, baselen
) == 0) {
3328 /* Check for exact name matches but also check for arrays and
3331 if (name
[baselen
] == '\0' ||
3332 name
[baselen
] == '[' ||
3333 name
[baselen
] == '.') {
3345 add_interface_variables(struct gl_shader_program
*shProg
,
3346 exec_list
*ir
, GLenum programInterface
)
3348 foreach_in_list(ir_instruction
, node
, ir
) {
3349 ir_variable
*var
= node
->as_variable();
3355 switch (var
->data
.mode
) {
3356 /* From GL 4.3 core spec, section 11.1.1 (Vertex Attributes):
3357 * "For GetActiveAttrib, all active vertex shader input variables
3358 * are enumerated, including the special built-in inputs gl_VertexID
3359 * and gl_InstanceID."
3361 case ir_var_system_value
:
3362 if (var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID
&&
3363 var
->data
.location
!= SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
&&
3364 var
->data
.location
!= SYSTEM_VALUE_INSTANCE_ID
)
3366 /* Mark special built-in inputs referenced by the vertex stage so
3367 * that they are considered active by the shader queries.
3369 mask
= (1 << (MESA_SHADER_VERTEX
));
3371 case ir_var_shader_in
:
3372 if (programInterface
!= GL_PROGRAM_INPUT
)
3375 case ir_var_shader_out
:
3376 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3383 /* Skip packed varyings, packed varyings are handled separately
3384 * by add_packed_varyings.
3386 if (strncmp(var
->name
, "packed:", 7) == 0)
3389 /* Skip fragdata arrays, these are handled separately
3390 * by add_fragdata_arrays.
3392 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3395 if (!add_program_resource(shProg
, programInterface
, var
,
3396 build_stageref(shProg
, var
->name
,
3397 var
->data
.mode
) | mask
))
3404 add_packed_varyings(struct gl_shader_program
*shProg
, int stage
)
3406 struct gl_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3409 if (!sh
|| !sh
->packed_varyings
)
3412 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3413 ir_variable
*var
= node
->as_variable();
3415 switch (var
->data
.mode
) {
3416 case ir_var_shader_in
:
3417 iface
= GL_PROGRAM_INPUT
;
3419 case ir_var_shader_out
:
3420 iface
= GL_PROGRAM_OUTPUT
;
3423 unreachable("unexpected type");
3425 if (!add_program_resource(shProg
, iface
, var
,
3426 build_stageref(shProg
, var
->name
,
3435 add_fragdata_arrays(struct gl_shader_program
*shProg
)
3437 struct gl_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3439 if (!sh
|| !sh
->fragdata_arrays
)
3442 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3443 ir_variable
*var
= node
->as_variable();
3445 assert(var
->data
.mode
== ir_var_shader_out
);
3446 if (!add_program_resource(shProg
, GL_PROGRAM_OUTPUT
, var
,
3447 1 << MESA_SHADER_FRAGMENT
))
3455 get_top_level_name(const char *name
)
3457 const char *first_dot
= strchr(name
, '.');
3458 const char *first_square_bracket
= strchr(name
, '[');
3460 /* From ARB_program_interface_query spec:
3462 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying the
3463 * number of active array elements of the top-level shader storage block
3464 * member containing to the active variable is written to <params>. If the
3465 * top-level block member is not declared as an array, the value one is
3466 * written to <params>. If the top-level block member is an array with no
3467 * declared size, the value zero is written to <params>.
3470 /* The buffer variable is on top level.*/
3471 if (!first_square_bracket
&& !first_dot
)
3472 name_size
= strlen(name
);
3473 else if ((!first_square_bracket
||
3474 (first_dot
&& first_dot
< first_square_bracket
)))
3475 name_size
= first_dot
- name
;
3477 name_size
= first_square_bracket
- name
;
3479 return strndup(name
, name_size
);
3483 get_var_name(const char *name
)
3485 const char *first_dot
= strchr(name
, '.');
3488 return strdup(name
);
3490 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3494 is_top_level_shader_storage_block_member(const char* name
,
3495 const char* interface_name
,
3496 const char* field_name
)
3498 bool result
= false;
3500 /* If the given variable is already a top-level shader storage
3501 * block member, then return array_size = 1.
3502 * We could have two possibilities: if we have an instanced
3503 * shader storage block or not instanced.
3505 * For the first, we check create a name as it was in top level and
3506 * compare it with the real name. If they are the same, then
3507 * the variable is already at top-level.
3509 * Full instanced name is: interface name + '.' + var name +
3512 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3513 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3514 if (!full_instanced_name
) {
3515 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3519 snprintf(full_instanced_name
, name_length
, "%s.%s",
3520 interface_name
, field_name
);
3522 /* Check if its top-level shader storage block member of an
3523 * instanced interface block, or of a unnamed interface block.
3525 if (strcmp(name
, full_instanced_name
) == 0 ||
3526 strcmp(name
, field_name
) == 0)
3529 free(full_instanced_name
);
3534 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3535 char *interface_name
, char *var_name
)
3537 /* From GL_ARB_program_interface_query spec:
3539 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer
3540 * identifying the number of active array elements of the top-level
3541 * shader storage block member containing to the active variable is
3542 * written to <params>. If the top-level block member is not
3543 * declared as an array, the value one is written to <params>. If
3544 * the top-level block member is an array with no declared size,
3545 * the value zero is written to <params>.
3547 if (is_top_level_shader_storage_block_member(uni
->name
,
3551 else if (field
->type
->is_unsized_array())
3553 else if (field
->type
->is_array())
3554 return field
->type
->length
;
3560 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3561 const glsl_struct_field
*field
, char *interface_name
,
3564 /* From GL_ARB_program_interface_query:
3566 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3567 * identifying the stride between array elements of the top-level
3568 * shader storage block member containing the active variable is
3569 * written to <params>. For top-level block members declared as
3570 * arrays, the value written is the difference, in basic machine
3571 * units, between the offsets of the active variable for
3572 * consecutive elements in the top-level array. For top-level
3573 * block members not declared as an array, zero is written to
3576 if (field
->type
->is_array()) {
3577 const enum glsl_matrix_layout matrix_layout
=
3578 glsl_matrix_layout(field
->matrix_layout
);
3579 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3580 const glsl_type
*array_type
= field
->type
->fields
.array
;
3582 if (is_top_level_shader_storage_block_member(uni
->name
,
3587 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3588 if (array_type
->is_record() || array_type
->is_array())
3589 return glsl_align(array_type
->std140_size(row_major
), 16);
3591 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3593 return array_type
->std430_array_stride(row_major
);
3600 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3601 struct gl_uniform_storage
*uni
)
3603 int block_index
= uni
->block_index
;
3604 int array_size
= -1;
3605 int array_stride
= -1;
3606 char *var_name
= get_top_level_name(uni
->name
);
3607 char *interface_name
=
3608 get_top_level_name(shProg
->BufferInterfaceBlocks
[block_index
].Name
);
3610 if (strcmp(var_name
, interface_name
) == 0) {
3611 /* Deal with instanced array of SSBOs */
3612 char *temp_name
= get_var_name(uni
->name
);
3614 linker_error(shProg
, "Out of memory during linking.\n");
3615 goto write_top_level_array_size_and_stride
;
3618 var_name
= get_top_level_name(temp_name
);
3621 linker_error(shProg
, "Out of memory during linking.\n");
3622 goto write_top_level_array_size_and_stride
;
3626 for (unsigned i
= 0; i
< shProg
->NumShaders
; i
++) {
3627 if (shProg
->Shaders
[i
] == NULL
)
3630 const gl_shader
*stage
= shProg
->Shaders
[i
];
3631 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
3632 ir_variable
*var
= node
->as_variable();
3633 if (!var
|| !var
->get_interface_type() ||
3634 var
->data
.mode
!= ir_var_shader_storage
)
3637 const glsl_type
*interface
= var
->get_interface_type();
3639 if (strcmp(interface_name
, interface
->name
) != 0)
3642 for (unsigned i
= 0; i
< interface
->length
; i
++) {
3643 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
3644 if (strcmp(field
->name
, var_name
) != 0)
3647 array_stride
= get_array_stride(uni
, interface
, field
,
3648 interface_name
, var_name
);
3649 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
3650 goto write_top_level_array_size_and_stride
;
3654 write_top_level_array_size_and_stride
:
3655 free(interface_name
);
3657 uni
->top_level_array_stride
= array_stride
;
3658 uni
->top_level_array_size
= array_size
;
3662 * Builds up a list of program resources that point to existing
3666 build_program_resource_list(struct gl_shader_program
*shProg
)
3668 /* Rebuild resource list. */
3669 if (shProg
->ProgramResourceList
) {
3670 ralloc_free(shProg
->ProgramResourceList
);
3671 shProg
->ProgramResourceList
= NULL
;
3672 shProg
->NumProgramResourceList
= 0;
3675 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
3677 /* Determine first input and final output stage. These are used to
3678 * detect which variables should be enumerated in the resource list
3679 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
3681 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3682 if (!shProg
->_LinkedShaders
[i
])
3684 if (input_stage
== MESA_SHADER_STAGES
)
3689 /* Empty shader, no resources. */
3690 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
3693 /* Program interface needs to expose varyings in case of SSO. */
3694 if (shProg
->SeparateShader
) {
3695 if (!add_packed_varyings(shProg
, input_stage
))
3697 if (!add_packed_varyings(shProg
, output_stage
))
3701 if (!add_fragdata_arrays(shProg
))
3704 /* Add inputs and outputs to the resource list. */
3705 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[input_stage
]->ir
,
3709 if (!add_interface_variables(shProg
, shProg
->_LinkedShaders
[output_stage
]->ir
,
3713 /* Add transform feedback varyings. */
3714 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
3715 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
3716 if (!add_program_resource(shProg
, GL_TRANSFORM_FEEDBACK_VARYING
,
3717 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
3723 /* Add uniforms from uniform storage. */
3724 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3725 /* Do not add uniforms internally used by Mesa. */
3726 if (shProg
->UniformStorage
[i
].hidden
)
3730 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
3733 /* Add stagereferences for uniforms in a uniform block. */
3734 int block_index
= shProg
->UniformStorage
[i
].block_index
;
3735 if (block_index
!= -1) {
3736 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
3737 if (shProg
->InterfaceBlockStageIndex
[j
][block_index
] != -1)
3738 stageref
|= (1 << j
);
3742 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
3743 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
3744 if (!should_add_buffer_variable(shProg
, type
,
3745 shProg
->UniformStorage
[i
].name
))
3748 if (is_shader_storage
) {
3749 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
3752 if (!add_program_resource(shProg
, type
,
3753 &shProg
->UniformStorage
[i
], stageref
))
3757 /* Add program uniform blocks and shader storage blocks. */
3758 for (unsigned i
= 0; i
< shProg
->NumBufferInterfaceBlocks
; i
++) {
3759 bool is_shader_storage
= shProg
->BufferInterfaceBlocks
[i
].IsShaderStorage
;
3760 GLenum type
= is_shader_storage
? GL_SHADER_STORAGE_BLOCK
: GL_UNIFORM_BLOCK
;
3761 if (!add_program_resource(shProg
, type
,
3762 &shProg
->BufferInterfaceBlocks
[i
], 0))
3766 /* Add atomic counter buffers. */
3767 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
3768 if (!add_program_resource(shProg
, GL_ATOMIC_COUNTER_BUFFER
,
3769 &shProg
->AtomicBuffers
[i
], 0))
3773 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
3775 if (!shProg
->UniformStorage
[i
].hidden
)
3778 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
3779 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
3780 !shProg
->UniformStorage
[i
].type
->is_subroutine())
3783 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
3784 /* add shader subroutines */
3785 if (!add_program_resource(shProg
, type
, &shProg
->UniformStorage
[i
], 0))
3790 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3791 struct gl_shader
*sh
= shProg
->_LinkedShaders
[i
];
3797 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
3798 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3799 if (!add_program_resource(shProg
, type
, &sh
->SubroutineFunctions
[j
], 0))
3806 * This check is done to make sure we allow only constant expression
3807 * indexing and "constant-index-expression" (indexing with an expression
3808 * that includes loop induction variable).
3811 validate_sampler_array_indexing(struct gl_context
*ctx
,
3812 struct gl_shader_program
*prog
)
3814 dynamic_sampler_array_indexing_visitor v
;
3815 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3816 if (prog
->_LinkedShaders
[i
] == NULL
)
3819 bool no_dynamic_indexing
=
3820 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
3822 /* Search for array derefs in shader. */
3823 v
.run(prog
->_LinkedShaders
[i
]->ir
);
3824 if (v
.uses_dynamic_sampler_array_indexing()) {
3825 const char *msg
= "sampler arrays indexed with non-constant "
3826 "expressions is forbidden in GLSL %s %u";
3827 /* Backend has indicated that it has no dynamic indexing support. */
3828 if (no_dynamic_indexing
) {
3829 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3832 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
3840 link_assign_subroutine_types(struct gl_shader_program
*prog
)
3842 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3843 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
3848 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3849 ir_function
*fn
= node
->as_function();
3853 if (fn
->is_subroutine
)
3854 sh
->NumSubroutineUniformTypes
++;
3856 if (!fn
->num_subroutine_types
)
3859 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
3860 struct gl_subroutine_function
,
3861 sh
->NumSubroutineFunctions
+ 1);
3862 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
3863 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
3864 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
3865 ralloc_array(sh
, const struct glsl_type
*,
3866 fn
->num_subroutine_types
);
3868 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
3871 * "Each subroutine with an index qualifier in the shader must be
3872 * given a unique index, otherwise a compile or link error will be
3875 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3876 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
3877 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
3878 linker_error(prog
, "each subroutine index qualifier in the "
3879 "shader must be unique\n");
3883 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
3884 fn
->subroutine_index
;
3886 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
3887 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
3888 sh
->NumSubroutineFunctions
++;
3891 /* Assign index for subroutines without an explicit index*/
3893 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
3894 while (sh
->SubroutineFunctions
[j
].index
== -1) {
3895 for (unsigned k
= 0; k
< sh
->NumSubroutineFunctions
; k
++) {
3896 if (sh
->SubroutineFunctions
[k
].index
== index
)
3898 else if (k
== sh
->NumSubroutineFunctions
- 1)
3899 sh
->SubroutineFunctions
[j
].index
= index
;
3908 split_ubos_and_ssbos(void *mem_ctx
,
3909 struct gl_uniform_block
*blocks
,
3910 unsigned num_blocks
,
3911 struct gl_uniform_block
***ubos
,
3913 struct gl_uniform_block
***ssbos
,
3914 unsigned *num_ssbos
)
3916 unsigned num_ubo_blocks
= 0;
3917 unsigned num_ssbo_blocks
= 0;
3919 for (unsigned i
= 0; i
< num_blocks
; i
++) {
3920 if (blocks
[i
].IsShaderStorage
)
3926 *ubos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ubo_blocks
);
3929 *ssbos
= ralloc_array(mem_ctx
, gl_uniform_block
*, num_ssbo_blocks
);
3932 for (unsigned i
= 0; i
< num_blocks
; i
++) {
3933 if (blocks
[i
].IsShaderStorage
) {
3934 (*ssbos
)[(*num_ssbos
)++] = &blocks
[i
];
3936 (*ubos
)[(*num_ubos
)++] = &blocks
[i
];
3940 assert(*num_ubos
+ *num_ssbos
== num_blocks
);
3944 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3946 tfeedback_decl
*tfeedback_decls
= NULL
;
3947 unsigned num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
3949 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
3951 prog
->LinkStatus
= true; /* All error paths will set this to false */
3952 prog
->Validated
= false;
3953 prog
->_Used
= false;
3955 prog
->ARB_fragment_coord_conventions_enable
= false;
3957 /* Separate the shaders into groups based on their type.
3959 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
3960 unsigned num_shaders
[MESA_SHADER_STAGES
];
3962 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3963 shader_list
[i
] = (struct gl_shader
**)
3964 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
3968 unsigned min_version
= UINT_MAX
;
3969 unsigned max_version
= 0;
3970 const bool is_es_prog
=
3971 (prog
->NumShaders
> 0 && prog
->Shaders
[0]->IsES
) ? true : false;
3972 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
3973 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
3974 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
3976 if (prog
->Shaders
[i
]->IsES
!= is_es_prog
) {
3977 linker_error(prog
, "all shaders must use same shading "
3978 "language version\n");
3982 if (prog
->Shaders
[i
]->ARB_fragment_coord_conventions_enable
) {
3983 prog
->ARB_fragment_coord_conventions_enable
= true;
3986 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
3987 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
3988 num_shaders
[shader_type
]++;
3991 /* In desktop GLSL, different shader versions may be linked together. In
3992 * GLSL ES, all shader versions must be the same.
3994 if (is_es_prog
&& min_version
!= max_version
) {
3995 linker_error(prog
, "all shaders must use same shading "
3996 "language version\n");
4000 prog
->Version
= max_version
;
4001 prog
->IsES
= is_es_prog
;
4003 /* From OpenGL 4.5 Core specification (7.3 Program Objects):
4004 * "Linking can fail for a variety of reasons as specified in the OpenGL
4005 * Shading Language Specification, as well as any of the following
4008 * * No shader objects are attached to program.
4012 * Same rule applies for OpenGL ES >= 3.1.
4015 if (prog
->NumShaders
== 0 &&
4016 ((ctx
->API
== API_OPENGL_CORE
&& ctx
->Version
>= 45) ||
4017 (ctx
->API
== API_OPENGLES2
&& ctx
->Version
>= 31))) {
4018 linker_error(prog
, "No shader objects are attached to program.\n");
4022 /* Some shaders have to be linked with some other shaders present.
4024 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4025 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
4026 !prog
->SeparateShader
) {
4027 linker_error(prog
, "Geometry shader must be linked with "
4031 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4032 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
4033 !prog
->SeparateShader
) {
4034 linker_error(prog
, "Tessellation evaluation shader must be linked with "
4038 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4039 num_shaders
[MESA_SHADER_VERTEX
] == 0 &&
4040 !prog
->SeparateShader
) {
4041 linker_error(prog
, "Tessellation control shader must be linked with "
4046 /* The spec is self-contradictory here. It allows linking without a tess
4047 * eval shader, but that can only be used with transform feedback and
4048 * rasterization disabled. However, transform feedback isn't allowed
4049 * with GL_PATCHES, so it can't be used.
4051 * More investigation showed that the idea of transform feedback after
4052 * a tess control shader was dropped, because some hw vendors couldn't
4053 * support tessellation without a tess eval shader, but the linker section
4054 * wasn't updated to reflect that.
4056 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4059 * Do what's reasonable and always require a tess eval shader if a tess
4060 * control shader is present.
4062 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4063 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0 &&
4064 !prog
->SeparateShader
) {
4065 linker_error(prog
, "Tessellation control shader must be linked with "
4066 "tessellation evaluation shader\n");
4070 /* Compute shaders have additional restrictions. */
4071 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4072 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4073 linker_error(prog
, "Compute shaders may not be linked with any other "
4074 "type of shader\n");
4077 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4078 if (prog
->_LinkedShaders
[i
] != NULL
)
4079 _mesa_delete_shader(ctx
, prog
->_LinkedShaders
[i
]);
4081 prog
->_LinkedShaders
[i
] = NULL
;
4084 /* Link all shaders for a particular stage and validate the result.
4086 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4087 if (num_shaders
[stage
] > 0) {
4088 gl_shader
*const sh
=
4089 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4090 num_shaders
[stage
]);
4092 if (!prog
->LinkStatus
) {
4094 _mesa_delete_shader(ctx
, sh
);
4099 case MESA_SHADER_VERTEX
:
4100 validate_vertex_shader_executable(prog
, sh
);
4102 case MESA_SHADER_TESS_CTRL
:
4103 /* nothing to be done */
4105 case MESA_SHADER_TESS_EVAL
:
4106 validate_tess_eval_shader_executable(prog
, sh
);
4108 case MESA_SHADER_GEOMETRY
:
4109 validate_geometry_shader_executable(prog
, sh
);
4111 case MESA_SHADER_FRAGMENT
:
4112 validate_fragment_shader_executable(prog
, sh
);
4115 if (!prog
->LinkStatus
) {
4117 _mesa_delete_shader(ctx
, sh
);
4121 _mesa_reference_shader(ctx
, &prog
->_LinkedShaders
[stage
], sh
);
4125 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0)
4126 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4127 else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0)
4128 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4129 else if (num_shaders
[MESA_SHADER_VERTEX
] > 0)
4130 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4132 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4134 /* Here begins the inter-stage linking phase. Some initial validation is
4135 * performed, then locations are assigned for uniforms, attributes, and
4138 cross_validate_uniforms(prog
);
4139 if (!prog
->LinkStatus
)
4144 for (prev
= 0; prev
<= MESA_SHADER_FRAGMENT
; prev
++) {
4145 if (prog
->_LinkedShaders
[prev
] != NULL
)
4149 check_explicit_uniform_locations(ctx
, prog
);
4150 link_assign_subroutine_types(prog
);
4152 if (!prog
->LinkStatus
)
4155 resize_tes_inputs(ctx
, prog
);
4157 /* Validate the inputs of each stage with the output of the preceding
4160 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4161 if (prog
->_LinkedShaders
[i
] == NULL
)
4164 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4165 prog
->_LinkedShaders
[i
]);
4166 if (!prog
->LinkStatus
)
4169 cross_validate_outputs_to_inputs(prog
,
4170 prog
->_LinkedShaders
[prev
],
4171 prog
->_LinkedShaders
[i
]);
4172 if (!prog
->LinkStatus
)
4178 /* Cross-validate uniform blocks between shader stages */
4179 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
,
4180 MESA_SHADER_STAGES
);
4181 if (!prog
->LinkStatus
)
4184 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4185 if (prog
->_LinkedShaders
[i
] != NULL
)
4186 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4189 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4190 * it before optimization because we want most of the checks to get
4191 * dropped thanks to constant propagation.
4193 * This rule also applies to GLSL ES 3.00.
4195 if (max_version
>= (is_es_prog
? 300 : 130)) {
4196 struct gl_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4198 lower_discard_flow(sh
->ir
);
4202 if (!interstage_cross_validate_uniform_blocks(prog
))
4205 /* Do common optimization before assigning storage for attributes,
4206 * uniforms, and varyings. Later optimization could possibly make
4207 * some of that unused.
4209 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4210 if (prog
->_LinkedShaders
[i
] == NULL
)
4213 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4214 if (!prog
->LinkStatus
)
4217 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerClipDistance
) {
4218 lower_clip_distance(prog
->_LinkedShaders
[i
]);
4221 if (ctx
->Const
.LowerTessLevel
) {
4222 lower_tess_level(prog
->_LinkedShaders
[i
]);
4225 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4226 &ctx
->Const
.ShaderCompilerOptions
[i
],
4227 ctx
->Const
.NativeIntegers
))
4230 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
);
4233 /* Validation for special cases where we allow sampler array indexing
4234 * with loop induction variable. This check emits a warning or error
4235 * depending if backend can handle dynamic indexing.
4237 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4238 (prog
->IsES
&& prog
->Version
< 300)) {
4239 if (!validate_sampler_array_indexing(ctx
, prog
))
4243 /* Check and validate stream emissions in geometry shaders */
4244 validate_geometry_shader_emissions(ctx
, prog
);
4246 /* Mark all generic shader inputs and outputs as unpaired. */
4247 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4248 if (prog
->_LinkedShaders
[i
] != NULL
) {
4249 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4253 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4254 MESA_SHADER_VERTEX
)) {
4258 if (!assign_attribute_or_color_locations(prog
, &ctx
->Const
,
4259 MESA_SHADER_FRAGMENT
)) {
4263 unsigned first
, last
;
4265 first
= MESA_SHADER_STAGES
;
4268 /* Determine first and last stage. */
4269 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4270 if (!prog
->_LinkedShaders
[i
])
4272 if (first
== MESA_SHADER_STAGES
)
4277 if (num_tfeedback_decls
!= 0) {
4278 /* From GL_EXT_transform_feedback:
4279 * A program will fail to link if:
4281 * * the <count> specified by TransformFeedbackVaryingsEXT is
4282 * non-zero, but the program object has no vertex or geometry
4285 if (first
== MESA_SHADER_FRAGMENT
) {
4286 linker_error(prog
, "Transform feedback varyings specified, but "
4287 "no vertex or geometry shader is present.\n");
4291 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4292 prog
->TransformFeedback
.NumVarying
);
4293 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4294 prog
->TransformFeedback
.VaryingNames
,
4299 /* Linking the stages in the opposite order (from fragment to vertex)
4300 * ensures that inter-shader outputs written to in an earlier stage are
4301 * eliminated if they are (transitively) not used in a later stage.
4305 if (first
< MESA_SHADER_FRAGMENT
) {
4306 gl_shader
*const sh
= prog
->_LinkedShaders
[last
];
4308 if (first
== MESA_SHADER_GEOMETRY
) {
4309 /* There was no vertex shader, but we still have to assign varying
4310 * locations for use by geometry shader inputs in SSO.
4312 * If the shader is not separable (i.e., prog->SeparateShader is
4313 * false), linking will have already failed when first is
4314 * MESA_SHADER_GEOMETRY.
4316 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4317 NULL
, prog
->_LinkedShaders
[first
],
4318 num_tfeedback_decls
, tfeedback_decls
))
4322 if (last
!= MESA_SHADER_FRAGMENT
&&
4323 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4324 /* There was no fragment shader, but we still have to assign varying
4325 * locations for use by transform feedback.
4327 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4329 num_tfeedback_decls
, tfeedback_decls
))
4333 do_dead_builtin_varyings(ctx
, sh
, NULL
,
4334 num_tfeedback_decls
, tfeedback_decls
);
4336 if (!prog
->SeparateShader
)
4337 demote_shader_inputs_and_outputs(sh
, ir_var_shader_out
);
4339 /* Eliminate code that is now dead due to unused outputs being demoted.
4341 while (do_dead_code(sh
->ir
, false))
4344 else if (first
== MESA_SHADER_FRAGMENT
) {
4345 /* If the program only contains a fragment shader...
4347 gl_shader
*const sh
= prog
->_LinkedShaders
[first
];
4349 do_dead_builtin_varyings(ctx
, NULL
, sh
,
4350 num_tfeedback_decls
, tfeedback_decls
);
4352 if (prog
->SeparateShader
) {
4353 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4354 NULL
/* producer */,
4356 0 /* num_tfeedback_decls */,
4357 NULL
/* tfeedback_decls */))
4360 demote_shader_inputs_and_outputs(sh
, ir_var_shader_in
);
4362 while (do_dead_code(sh
->ir
, false))
4367 for (int i
= next
- 1; i
>= 0; i
--) {
4368 if (prog
->_LinkedShaders
[i
] == NULL
)
4371 gl_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4372 gl_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4374 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4375 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4379 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4380 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4383 demote_shader_inputs_and_outputs(sh_i
, ir_var_shader_out
);
4384 demote_shader_inputs_and_outputs(sh_next
, ir_var_shader_in
);
4386 /* Eliminate code that is now dead due to unused outputs being demoted.
4388 while (do_dead_code(sh_i
->ir
, false))
4390 while (do_dead_code(sh_next
->ir
, false))
4393 /* This must be done after all dead varyings are eliminated. */
4394 if (!check_against_output_limit(ctx
, prog
, sh_i
))
4396 if (!check_against_input_limit(ctx
, prog
, sh_next
))
4402 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
))
4405 update_array_sizes(prog
);
4406 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
);
4407 link_assign_atomic_counter_resources(ctx
, prog
);
4408 store_fragdepth_layout(prog
);
4410 link_calculate_subroutine_compat(prog
);
4411 check_resources(ctx
, prog
);
4412 check_subroutine_resources(prog
);
4413 check_image_resources(ctx
, prog
);
4414 link_check_atomic_counter_resources(ctx
, prog
);
4416 if (!prog
->LinkStatus
)
4419 /* OpenGL ES requires that a vertex shader and a fragment shader both be
4420 * present in a linked program. GL_ARB_ES2_compatibility doesn't say
4421 * anything about shader linking when one of the shaders (vertex or
4422 * fragment shader) is absent. So, the extension shouldn't change the
4423 * behavior specified in GLSL specification.
4425 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
) {
4426 /* With ES < 3.1 one needs to have always vertex + fragment shader. */
4427 if (ctx
->Version
< 31) {
4428 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4429 linker_error(prog
, "program lacks a vertex shader\n");
4430 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4431 linker_error(prog
, "program lacks a fragment shader\n");
4434 /* From OpenGL ES 3.1 specification (7.3 Program Objects):
4435 * "Linking can fail for a variety of reasons as specified in the
4436 * OpenGL ES Shading Language Specification, as well as any of the
4437 * following reasons:
4441 * * program contains objects to form either a vertex shader or
4442 * fragment shader, and program is not separable, and does not
4443 * contain objects to form both a vertex shader and fragment
4446 if (!!prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] ^
4447 !!prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]) {
4448 linker_error(prog
, "Program needs to contain both vertex and "
4449 "fragment shaders.\n");
4454 /* Split BufferInterfaceBlocks into UniformBlocks and ShaderStorageBlocks
4455 * for gl_shader_program and gl_shader, so that drivers that need separate
4456 * index spaces for each set can have that.
4458 for (unsigned i
= MESA_SHADER_VERTEX
; i
< MESA_SHADER_STAGES
; i
++) {
4459 if (prog
->_LinkedShaders
[i
] != NULL
) {
4460 gl_shader
*sh
= prog
->_LinkedShaders
[i
];
4461 split_ubos_and_ssbos(sh
,
4462 sh
->BufferInterfaceBlocks
,
4463 sh
->NumBufferInterfaceBlocks
,
4465 &sh
->NumUniformBlocks
,
4466 &sh
->ShaderStorageBlocks
,
4467 &sh
->NumShaderStorageBlocks
);
4471 split_ubos_and_ssbos(prog
,
4472 prog
->BufferInterfaceBlocks
,
4473 prog
->NumBufferInterfaceBlocks
,
4474 &prog
->UniformBlocks
,
4475 &prog
->NumUniformBlocks
,
4476 &prog
->ShaderStorageBlocks
,
4477 &prog
->NumShaderStorageBlocks
);
4479 /* FINISHME: Assign fragment shader output locations. */
4481 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4482 if (prog
->_LinkedShaders
[i
] == NULL
)
4485 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerBufferInterfaceBlocks
)
4486 lower_ubo_reference(prog
->_LinkedShaders
[i
]);
4488 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4492 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4493 free(shader_list
[i
]);
4494 if (prog
->_LinkedShaders
[i
] == NULL
)
4497 /* Do a final validation step to make sure that the IR wasn't
4498 * invalidated by any modifications performed after intrastage linking.
4500 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4502 /* Retain any live IR, but trash the rest. */
4503 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4505 /* The symbol table in the linked shaders may contain references to
4506 * variables that were removed (e.g., unused uniforms). Since it may
4507 * contain junk, there is no possible valid use. Delete it and set the
4510 delete prog
->_LinkedShaders
[i
]->symbols
;
4511 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4514 ralloc_free(mem_ctx
);