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/prog_instruction.h"
76 #include "util/string_to_uint_map.h"
78 #include "link_varyings.h"
79 #include "ir_optimization.h"
80 #include "ir_rvalue_visitor.h"
81 #include "ir_uniform.h"
83 #include "main/shaderobj.h"
84 #include "main/enums.h"
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 array_resize_visitor
: public ir_hierarchical_visitor
{
186 unsigned num_vertices
;
187 gl_shader_program
*prog
;
188 gl_shader_stage stage
;
190 array_resize_visitor(unsigned num_vertices
,
191 gl_shader_program
*prog
,
192 gl_shader_stage stage
)
194 this->num_vertices
= num_vertices
;
199 virtual ~array_resize_visitor()
204 virtual ir_visitor_status
visit(ir_variable
*var
)
206 if (!var
->type
->is_array() || var
->data
.mode
!= ir_var_shader_in
||
208 return visit_continue
;
210 unsigned size
= var
->type
->length
;
212 if (stage
== MESA_SHADER_GEOMETRY
) {
213 /* Generate a link error if the shader has declared this array with
216 if (!var
->data
.implicit_sized_array
&&
217 size
&& size
!= this->num_vertices
) {
218 linker_error(this->prog
, "size of array %s declared as %u, "
219 "but number of input vertices is %u\n",
220 var
->name
, size
, this->num_vertices
);
221 return visit_continue
;
224 /* Generate a link error if the shader attempts to access an input
225 * array using an index too large for its actual size assigned at
228 if (var
->data
.max_array_access
>= (int)this->num_vertices
) {
229 linker_error(this->prog
, "%s shader accesses element %i of "
230 "%s, but only %i input vertices\n",
231 _mesa_shader_stage_to_string(this->stage
),
232 var
->data
.max_array_access
, var
->name
, this->num_vertices
);
233 return visit_continue
;
237 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
239 var
->data
.max_array_access
= this->num_vertices
- 1;
241 return visit_continue
;
244 /* Dereferences of input variables need to be updated so that their type
245 * matches the newly assigned type of the variable they are accessing. */
246 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
248 ir
->type
= ir
->var
->type
;
249 return visit_continue
;
252 /* Dereferences of 2D input arrays need to be updated so that their type
253 * matches the newly assigned type of the array they are accessing. */
254 virtual ir_visitor_status
visit_leave(ir_dereference_array
*ir
)
256 const glsl_type
*const vt
= ir
->array
->type
;
258 ir
->type
= vt
->fields
.array
;
259 return visit_continue
;
264 * Visitor that determines the highest stream id to which a (geometry) shader
265 * emits vertices. It also checks whether End{Stream}Primitive is ever called.
267 class find_emit_vertex_visitor
: public ir_hierarchical_visitor
{
269 find_emit_vertex_visitor(int max_allowed
)
270 : max_stream_allowed(max_allowed
),
271 invalid_stream_id(0),
272 invalid_stream_id_from_emit_vertex(false),
273 end_primitive_found(false),
274 uses_non_zero_stream(false)
279 virtual ir_visitor_status
visit_leave(ir_emit_vertex
*ir
)
281 int stream_id
= ir
->stream_id();
284 invalid_stream_id
= stream_id
;
285 invalid_stream_id_from_emit_vertex
= true;
289 if (stream_id
> max_stream_allowed
) {
290 invalid_stream_id
= stream_id
;
291 invalid_stream_id_from_emit_vertex
= true;
296 uses_non_zero_stream
= true;
298 return visit_continue
;
301 virtual ir_visitor_status
visit_leave(ir_end_primitive
*ir
)
303 end_primitive_found
= true;
305 int stream_id
= ir
->stream_id();
308 invalid_stream_id
= stream_id
;
309 invalid_stream_id_from_emit_vertex
= false;
313 if (stream_id
> max_stream_allowed
) {
314 invalid_stream_id
= stream_id
;
315 invalid_stream_id_from_emit_vertex
= false;
320 uses_non_zero_stream
= true;
322 return visit_continue
;
327 return invalid_stream_id
!= 0;
330 const char *error_func()
332 return invalid_stream_id_from_emit_vertex
?
333 "EmitStreamVertex" : "EndStreamPrimitive";
338 return invalid_stream_id
;
343 return uses_non_zero_stream
;
346 bool uses_end_primitive()
348 return end_primitive_found
;
352 int max_stream_allowed
;
353 int invalid_stream_id
;
354 bool invalid_stream_id_from_emit_vertex
;
355 bool end_primitive_found
;
356 bool uses_non_zero_stream
;
359 /* Class that finds array derefs and check if indexes are dynamic. */
360 class dynamic_sampler_array_indexing_visitor
: public ir_hierarchical_visitor
363 dynamic_sampler_array_indexing_visitor() :
364 dynamic_sampler_array_indexing(false)
368 ir_visitor_status
visit_enter(ir_dereference_array
*ir
)
370 if (!ir
->variable_referenced())
371 return visit_continue
;
373 if (!ir
->variable_referenced()->type
->contains_sampler())
374 return visit_continue
;
376 if (!ir
->array_index
->constant_expression_value()) {
377 dynamic_sampler_array_indexing
= true;
380 return visit_continue
;
383 bool uses_dynamic_sampler_array_indexing()
385 return dynamic_sampler_array_indexing
;
389 bool dynamic_sampler_array_indexing
;
392 } /* anonymous namespace */
395 linker_error(gl_shader_program
*prog
, const char *fmt
, ...)
399 ralloc_strcat(&prog
->InfoLog
, "error: ");
401 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
404 prog
->LinkStatus
= false;
409 linker_warning(gl_shader_program
*prog
, const char *fmt
, ...)
413 ralloc_strcat(&prog
->InfoLog
, "warning: ");
415 ralloc_vasprintf_append(&prog
->InfoLog
, fmt
, ap
);
422 * Given a string identifying a program resource, break it into a base name
423 * and an optional array index in square brackets.
425 * If an array index is present, \c out_base_name_end is set to point to the
426 * "[" that precedes the array index, and the array index itself is returned
429 * If no array index is present (or if the array index is negative or
430 * mal-formed), \c out_base_name_end, is set to point to the null terminator
431 * at the end of the input string, and -1 is returned.
433 * Only the final array index is parsed; if the string contains other array
434 * indices (or structure field accesses), they are left in the base name.
436 * No attempt is made to check that the base name is properly formed;
437 * typically the caller will look up the base name in a hash table, so
438 * ill-formed base names simply turn into hash table lookup failures.
441 parse_program_resource_name(const GLchar
*name
,
442 const GLchar
**out_base_name_end
)
444 /* Section 7.3.1 ("Program Interfaces") of the OpenGL 4.3 spec says:
446 * "When an integer array element or block instance number is part of
447 * the name string, it will be specified in decimal form without a "+"
448 * or "-" sign or any extra leading zeroes. Additionally, the name
449 * string will not include white space anywhere in the string."
452 const size_t len
= strlen(name
);
453 *out_base_name_end
= name
+ len
;
455 if (len
== 0 || name
[len
-1] != ']')
458 /* Walk backwards over the string looking for a non-digit character. This
459 * had better be the opening bracket for an array index.
461 * Initially, i specifies the location of the ']'. Since the string may
462 * contain only the ']' charcater, walk backwards very carefully.
465 for (i
= len
- 1; (i
> 0) && isdigit(name
[i
-1]); --i
)
468 if ((i
== 0) || name
[i
-1] != '[')
471 long array_index
= strtol(&name
[i
], NULL
, 10);
475 /* Check for leading zero */
476 if (name
[i
] == '0' && name
[i
+1] != ']')
479 *out_base_name_end
= name
+ (i
- 1);
485 link_invalidate_variable_locations(exec_list
*ir
)
487 foreach_in_list(ir_instruction
, node
, ir
) {
488 ir_variable
*const var
= node
->as_variable();
493 /* Only assign locations for variables that lack an explicit location.
494 * Explicit locations are set for all built-in variables, generic vertex
495 * shader inputs (via layout(location=...)), and generic fragment shader
496 * outputs (also via layout(location=...)).
498 if (!var
->data
.explicit_location
) {
499 var
->data
.location
= -1;
500 var
->data
.location_frac
= 0;
503 /* ir_variable::is_unmatched_generic_inout is used by the linker while
504 * connecting outputs from one stage to inputs of the next stage.
506 if (var
->data
.explicit_location
&&
507 var
->data
.location
< VARYING_SLOT_VAR0
) {
508 var
->data
.is_unmatched_generic_inout
= 0;
510 var
->data
.is_unmatched_generic_inout
= 1;
517 * Set clip_distance_array_size based and cull_distance_array_size on the given
520 * Also check for errors based on incorrect usage of gl_ClipVertex and
521 * gl_ClipDistance and gl_CullDistance.
522 * Additionally test whether the arrays gl_ClipDistance and gl_CullDistance
523 * exceed the maximum size defined by gl_MaxCombinedClipAndCullDistances.
525 * Return false if an error was reported.
528 analyze_clip_cull_usage(struct gl_shader_program
*prog
,
529 struct gl_linked_shader
*shader
,
530 struct gl_context
*ctx
,
531 GLuint
*clip_distance_array_size
,
532 GLuint
*cull_distance_array_size
)
534 *clip_distance_array_size
= 0;
535 *cull_distance_array_size
= 0;
537 if (prog
->Version
>= (prog
->IsES
? 300 : 130)) {
538 /* From section 7.1 (Vertex Shader Special Variables) of the
541 * "It is an error for a shader to statically write both
542 * gl_ClipVertex and gl_ClipDistance."
544 * This does not apply to GLSL ES shaders, since GLSL ES defines neither
545 * gl_ClipVertex nor gl_ClipDistance. However with
546 * GL_EXT_clip_cull_distance, this functionality is exposed in ES 3.0.
548 find_assignment_visitor
clip_distance("gl_ClipDistance");
549 find_assignment_visitor
cull_distance("gl_CullDistance");
551 clip_distance
.run(shader
->ir
);
552 cull_distance
.run(shader
->ir
);
554 /* From the ARB_cull_distance spec:
556 * It is a compile-time or link-time error for the set of shaders forming
557 * a program to statically read or write both gl_ClipVertex and either
558 * gl_ClipDistance or gl_CullDistance.
560 * This does not apply to GLSL ES shaders, since GLSL ES doesn't define
564 find_assignment_visitor
clip_vertex("gl_ClipVertex");
566 clip_vertex
.run(shader
->ir
);
568 if (clip_vertex
.variable_found() && clip_distance
.variable_found()) {
569 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
570 "and `gl_ClipDistance'\n",
571 _mesa_shader_stage_to_string(shader
->Stage
));
574 if (clip_vertex
.variable_found() && cull_distance
.variable_found()) {
575 linker_error(prog
, "%s shader writes to both `gl_ClipVertex' "
576 "and `gl_CullDistance'\n",
577 _mesa_shader_stage_to_string(shader
->Stage
));
582 if (clip_distance
.variable_found()) {
583 ir_variable
*clip_distance_var
=
584 shader
->symbols
->get_variable("gl_ClipDistance");
585 assert(clip_distance_var
);
586 *clip_distance_array_size
= clip_distance_var
->type
->length
;
588 if (cull_distance
.variable_found()) {
589 ir_variable
*cull_distance_var
=
590 shader
->symbols
->get_variable("gl_CullDistance");
591 assert(cull_distance_var
);
592 *cull_distance_array_size
= cull_distance_var
->type
->length
;
594 /* From the ARB_cull_distance spec:
596 * It is a compile-time or link-time error for the set of shaders forming
597 * a program to have the sum of the sizes of the gl_ClipDistance and
598 * gl_CullDistance arrays to be larger than
599 * gl_MaxCombinedClipAndCullDistances.
601 if ((*clip_distance_array_size
+ *cull_distance_array_size
) >
602 ctx
->Const
.MaxClipPlanes
) {
603 linker_error(prog
, "%s shader: the combined size of "
604 "'gl_ClipDistance' and 'gl_CullDistance' size cannot "
606 "gl_MaxCombinedClipAndCullDistances (%u)",
607 _mesa_shader_stage_to_string(shader
->Stage
),
608 ctx
->Const
.MaxClipPlanes
);
615 * Verify that a vertex shader executable meets all semantic requirements.
617 * Also sets prog->Vert.ClipDistanceArraySize and
618 * prog->Vert.CullDistanceArraySize as a side effect.
620 * \param shader Vertex shader executable to be verified
623 validate_vertex_shader_executable(struct gl_shader_program
*prog
,
624 struct gl_linked_shader
*shader
,
625 struct gl_context
*ctx
)
630 /* From the GLSL 1.10 spec, page 48:
632 * "The variable gl_Position is available only in the vertex
633 * language and is intended for writing the homogeneous vertex
634 * position. All executions of a well-formed vertex shader
635 * executable must write a value into this variable. [...] The
636 * variable gl_Position is available only in the vertex
637 * language and is intended for writing the homogeneous vertex
638 * position. All executions of a well-formed vertex shader
639 * executable must write a value into this variable."
641 * while in GLSL 1.40 this text is changed to:
643 * "The variable gl_Position is available only in the vertex
644 * language and is intended for writing the homogeneous vertex
645 * position. It can be written at any time during shader
646 * execution. It may also be read back by a vertex shader
647 * after being written. This value will be used by primitive
648 * assembly, clipping, culling, and other fixed functionality
649 * operations, if present, that operate on primitives after
650 * vertex processing has occurred. Its value is undefined if
651 * the vertex shader executable does not write gl_Position."
653 * All GLSL ES Versions are similar to GLSL 1.40--failing to write to
654 * gl_Position is not an error.
656 if (prog
->Version
< (prog
->IsES
? 300 : 140)) {
657 find_assignment_visitor
find("gl_Position");
658 find
.run(shader
->ir
);
659 if (!find
.variable_found()) {
662 "vertex shader does not write to `gl_Position'. "
663 "Its value is undefined. \n");
666 "vertex shader does not write to `gl_Position'. \n");
672 analyze_clip_cull_usage(prog
, shader
, ctx
,
673 &prog
->Vert
.ClipDistanceArraySize
,
674 &prog
->Vert
.CullDistanceArraySize
);
678 validate_tess_eval_shader_executable(struct gl_shader_program
*prog
,
679 struct gl_linked_shader
*shader
,
680 struct gl_context
*ctx
)
685 analyze_clip_cull_usage(prog
, shader
, ctx
,
686 &prog
->TessEval
.ClipDistanceArraySize
,
687 &prog
->TessEval
.CullDistanceArraySize
);
692 * Verify that a fragment shader executable meets all semantic requirements
694 * \param shader Fragment shader executable to be verified
697 validate_fragment_shader_executable(struct gl_shader_program
*prog
,
698 struct gl_linked_shader
*shader
)
703 find_assignment_visitor
frag_color("gl_FragColor");
704 find_assignment_visitor
frag_data("gl_FragData");
706 frag_color
.run(shader
->ir
);
707 frag_data
.run(shader
->ir
);
709 if (frag_color
.variable_found() && frag_data
.variable_found()) {
710 linker_error(prog
, "fragment shader writes to both "
711 "`gl_FragColor' and `gl_FragData'\n");
716 * Verify that a geometry shader executable meets all semantic requirements
718 * Also sets prog->Geom.VerticesIn, and prog->Geom.ClipDistanceArraySize and
719 * prog->Geom.CullDistanceArraySize as a side effect.
721 * \param shader Geometry shader executable to be verified
724 validate_geometry_shader_executable(struct gl_shader_program
*prog
,
725 struct gl_linked_shader
*shader
,
726 struct gl_context
*ctx
)
731 unsigned num_vertices
= vertices_per_prim(shader
->info
.Geom
.InputType
);
732 prog
->Geom
.VerticesIn
= num_vertices
;
734 analyze_clip_cull_usage(prog
, shader
, ctx
,
735 &prog
->Geom
.ClipDistanceArraySize
,
736 &prog
->Geom
.CullDistanceArraySize
);
740 * Check if geometry shaders emit to non-zero streams and do corresponding
744 validate_geometry_shader_emissions(struct gl_context
*ctx
,
745 struct gl_shader_program
*prog
)
747 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_GEOMETRY
];
750 find_emit_vertex_visitor
emit_vertex(ctx
->Const
.MaxVertexStreams
- 1);
751 emit_vertex
.run(sh
->ir
);
752 if (emit_vertex
.error()) {
753 linker_error(prog
, "Invalid call %s(%d). Accepted values for the "
754 "stream parameter are in the range [0, %d].\n",
755 emit_vertex
.error_func(),
756 emit_vertex
.error_stream(),
757 ctx
->Const
.MaxVertexStreams
- 1);
759 prog
->Geom
.UsesStreams
= emit_vertex
.uses_streams();
760 prog
->Geom
.UsesEndPrimitive
= emit_vertex
.uses_end_primitive();
762 /* From the ARB_gpu_shader5 spec:
764 * "Multiple vertex streams are supported only if the output primitive
765 * type is declared to be "points". A program will fail to link if it
766 * contains a geometry shader calling EmitStreamVertex() or
767 * EndStreamPrimitive() if its output primitive type is not "points".
769 * However, in the same spec:
771 * "The function EmitVertex() is equivalent to calling EmitStreamVertex()
772 * with <stream> set to zero."
776 * "The function EndPrimitive() is equivalent to calling
777 * EndStreamPrimitive() with <stream> set to zero."
779 * Since we can call EmitVertex() and EndPrimitive() when we output
780 * primitives other than points, calling EmitStreamVertex(0) or
781 * EmitEndPrimitive(0) should not produce errors. This it also what Nvidia
782 * does. Currently we only set prog->Geom.UsesStreams to TRUE when
783 * EmitStreamVertex() or EmitEndPrimitive() are called with a non-zero
786 if (prog
->Geom
.UsesStreams
&& sh
->info
.Geom
.OutputType
!= GL_POINTS
) {
787 linker_error(prog
, "EmitStreamVertex(n) and EndStreamPrimitive(n) "
788 "with n>0 requires point output\n");
794 validate_intrastage_arrays(struct gl_shader_program
*prog
,
795 ir_variable
*const var
,
796 ir_variable
*const existing
)
798 /* Consider the types to be "the same" if both types are arrays
799 * of the same type and one of the arrays is implicitly sized.
800 * In addition, set the type of the linked variable to the
801 * explicitly sized array.
803 if (var
->type
->is_array() && existing
->type
->is_array()) {
804 if ((var
->type
->fields
.array
== existing
->type
->fields
.array
) &&
805 ((var
->type
->length
== 0)|| (existing
->type
->length
== 0))) {
806 if (var
->type
->length
!= 0) {
807 if ((int)var
->type
->length
<= existing
->data
.max_array_access
) {
808 linker_error(prog
, "%s `%s' declared as type "
809 "`%s' but outermost dimension has an index"
812 var
->name
, var
->type
->name
,
813 existing
->data
.max_array_access
);
815 existing
->type
= var
->type
;
817 } else if (existing
->type
->length
!= 0) {
818 if((int)existing
->type
->length
<= var
->data
.max_array_access
&&
819 !existing
->data
.from_ssbo_unsized_array
) {
820 linker_error(prog
, "%s `%s' declared as type "
821 "`%s' but outermost dimension has an index"
824 var
->name
, existing
->type
->name
,
825 var
->data
.max_array_access
);
830 /* The arrays of structs could have different glsl_type pointers but
831 * they are actually the same type. Use record_compare() to check that.
833 if (existing
->type
->fields
.array
->is_record() &&
834 var
->type
->fields
.array
->is_record() &&
835 existing
->type
->fields
.array
->record_compare(var
->type
->fields
.array
))
844 * Perform validation of global variables used across multiple shaders
847 cross_validate_globals(struct gl_shader_program
*prog
,
848 struct exec_list
*ir
, glsl_symbol_table
*variables
,
851 foreach_in_list(ir_instruction
, node
, ir
) {
852 ir_variable
*const var
= node
->as_variable();
857 if (uniforms_only
&& (var
->data
.mode
!= ir_var_uniform
&& var
->data
.mode
!= ir_var_shader_storage
))
860 /* don't cross validate subroutine uniforms */
861 if (var
->type
->contains_subroutine())
864 /* Don't cross validate temporaries that are at global scope. These
865 * will eventually get pulled into the shaders 'main'.
867 if (var
->data
.mode
== ir_var_temporary
)
870 /* If a global with this name has already been seen, verify that the
871 * new instance has the same type. In addition, if the globals have
872 * initializers, the values of the initializers must be the same.
874 ir_variable
*const existing
= variables
->get_variable(var
->name
);
875 if (existing
!= NULL
) {
876 /* Check if types match. Interface blocks have some special
877 * rules so we handle those elsewhere.
879 if (var
->type
!= existing
->type
&&
880 !var
->is_interface_instance()) {
881 if (!validate_intrastage_arrays(prog
, var
, existing
)) {
882 if (var
->type
->is_record() && existing
->type
->is_record()
883 && existing
->type
->record_compare(var
->type
)) {
884 existing
->type
= var
->type
;
886 /* If it is an unsized array in a Shader Storage Block,
887 * two different shaders can access to different elements.
888 * Because of that, they might be converted to different
889 * sized arrays, then check that they are compatible but
890 * ignore the array size.
892 if (!(var
->data
.mode
== ir_var_shader_storage
&&
893 var
->data
.from_ssbo_unsized_array
&&
894 existing
->data
.mode
== ir_var_shader_storage
&&
895 existing
->data
.from_ssbo_unsized_array
&&
896 var
->type
->gl_type
== existing
->type
->gl_type
)) {
897 linker_error(prog
, "%s `%s' declared as type "
898 "`%s' and type `%s'\n",
900 var
->name
, var
->type
->name
,
901 existing
->type
->name
);
908 if (var
->data
.explicit_location
) {
909 if (existing
->data
.explicit_location
910 && (var
->data
.location
!= existing
->data
.location
)) {
911 linker_error(prog
, "explicit locations for %s "
912 "`%s' have differing values\n",
913 mode_string(var
), var
->name
);
917 if (var
->data
.location_frac
!= existing
->data
.location_frac
) {
918 linker_error(prog
, "explicit components for %s `%s' have "
919 "differing values\n", mode_string(var
), var
->name
);
923 existing
->data
.location
= var
->data
.location
;
924 existing
->data
.explicit_location
= true;
926 /* Check if uniform with implicit location was marked explicit
927 * by earlier shader stage. If so, mark it explicit in this stage
928 * too to make sure later processing does not treat it as
931 if (existing
->data
.explicit_location
) {
932 var
->data
.location
= existing
->data
.location
;
933 var
->data
.explicit_location
= true;
937 /* From the GLSL 4.20 specification:
938 * "A link error will result if two compilation units in a program
939 * specify different integer-constant bindings for the same
940 * opaque-uniform name. However, it is not an error to specify a
941 * binding on some but not all declarations for the same name"
943 if (var
->data
.explicit_binding
) {
944 if (existing
->data
.explicit_binding
&&
945 var
->data
.binding
!= existing
->data
.binding
) {
946 linker_error(prog
, "explicit bindings for %s "
947 "`%s' have differing values\n",
948 mode_string(var
), var
->name
);
952 existing
->data
.binding
= var
->data
.binding
;
953 existing
->data
.explicit_binding
= true;
956 if (var
->type
->contains_atomic() &&
957 var
->data
.offset
!= existing
->data
.offset
) {
958 linker_error(prog
, "offset specifications for %s "
959 "`%s' have differing values\n",
960 mode_string(var
), var
->name
);
964 /* Validate layout qualifiers for gl_FragDepth.
966 * From the AMD/ARB_conservative_depth specs:
968 * "If gl_FragDepth is redeclared in any fragment shader in a
969 * program, it must be redeclared in all fragment shaders in
970 * that program that have static assignments to
971 * gl_FragDepth. All redeclarations of gl_FragDepth in all
972 * fragment shaders in a single program must have the same set
975 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
976 bool layout_declared
= var
->data
.depth_layout
!= ir_depth_layout_none
;
977 bool layout_differs
=
978 var
->data
.depth_layout
!= existing
->data
.depth_layout
;
980 if (layout_declared
&& layout_differs
) {
982 "All redeclarations of gl_FragDepth in all "
983 "fragment shaders in a single program must have "
984 "the same set of qualifiers.\n");
987 if (var
->data
.used
&& layout_differs
) {
989 "If gl_FragDepth is redeclared with a layout "
990 "qualifier in any fragment shader, it must be "
991 "redeclared with the same layout qualifier in "
992 "all fragment shaders that have assignments to "
997 /* Page 35 (page 41 of the PDF) of the GLSL 4.20 spec says:
999 * "If a shared global has multiple initializers, the
1000 * initializers must all be constant expressions, and they
1001 * must all have the same value. Otherwise, a link error will
1002 * result. (A shared global having only one initializer does
1003 * not require that initializer to be a constant expression.)"
1005 * Previous to 4.20 the GLSL spec simply said that initializers
1006 * must have the same value. In this case of non-constant
1007 * initializers, this was impossible to determine. As a result,
1008 * no vendor actually implemented that behavior. The 4.20
1009 * behavior matches the implemented behavior of at least one other
1010 * vendor, so we'll implement that for all GLSL versions.
1012 if (var
->constant_initializer
!= NULL
) {
1013 if (existing
->constant_initializer
!= NULL
) {
1014 if (!var
->constant_initializer
->has_value(existing
->constant_initializer
)) {
1015 linker_error(prog
, "initializers for %s "
1016 "`%s' have differing values\n",
1017 mode_string(var
), var
->name
);
1021 /* If the first-seen instance of a particular uniform did
1022 * not have an initializer but a later instance does,
1023 * replace the former with the later.
1025 variables
->replace_variable(existing
->name
, var
);
1029 if (var
->data
.has_initializer
) {
1030 if (existing
->data
.has_initializer
1031 && (var
->constant_initializer
== NULL
1032 || existing
->constant_initializer
== NULL
)) {
1034 "shared global variable `%s' has multiple "
1035 "non-constant initializers.\n",
1041 if (existing
->data
.invariant
!= var
->data
.invariant
) {
1042 linker_error(prog
, "declarations for %s `%s' have "
1043 "mismatching invariant qualifiers\n",
1044 mode_string(var
), var
->name
);
1047 if (existing
->data
.centroid
!= var
->data
.centroid
) {
1048 linker_error(prog
, "declarations for %s `%s' have "
1049 "mismatching centroid qualifiers\n",
1050 mode_string(var
), var
->name
);
1053 if (existing
->data
.sample
!= var
->data
.sample
) {
1054 linker_error(prog
, "declarations for %s `%s` have "
1055 "mismatching sample qualifiers\n",
1056 mode_string(var
), var
->name
);
1059 if (existing
->data
.image_format
!= var
->data
.image_format
) {
1060 linker_error(prog
, "declarations for %s `%s` have "
1061 "mismatching image format qualifiers\n",
1062 mode_string(var
), var
->name
);
1066 if (prog
->IsES
&& existing
->data
.precision
!= var
->data
.precision
) {
1067 linker_error(prog
, "declarations for %s `%s` have "
1068 "mismatching precision qualifiers\n",
1069 mode_string(var
), var
->name
);
1073 variables
->add_variable(var
);
1079 * Perform validation of uniforms used across multiple shader stages
1082 cross_validate_uniforms(struct gl_shader_program
*prog
)
1084 glsl_symbol_table variables
;
1085 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1086 if (prog
->_LinkedShaders
[i
] == NULL
)
1089 cross_validate_globals(prog
, prog
->_LinkedShaders
[i
]->ir
, &variables
,
1095 * Accumulates the array of buffer blocks and checks that all definitions of
1096 * blocks agree on their contents.
1099 interstage_cross_validate_uniform_blocks(struct gl_shader_program
*prog
,
1102 int *InterfaceBlockStageIndex
[MESA_SHADER_STAGES
];
1103 struct gl_uniform_block
*blks
= NULL
;
1104 unsigned *num_blks
= validate_ssbo
? &prog
->NumShaderStorageBlocks
:
1105 &prog
->NumUniformBlocks
;
1107 unsigned max_num_buffer_blocks
= 0;
1108 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1109 if (prog
->_LinkedShaders
[i
]) {
1110 if (validate_ssbo
) {
1111 max_num_buffer_blocks
+=
1112 prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1114 max_num_buffer_blocks
+=
1115 prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1120 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1121 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1123 InterfaceBlockStageIndex
[i
] = new int[max_num_buffer_blocks
];
1124 for (unsigned int j
= 0; j
< max_num_buffer_blocks
; j
++)
1125 InterfaceBlockStageIndex
[i
][j
] = -1;
1130 unsigned sh_num_blocks
;
1131 struct gl_uniform_block
**sh_blks
;
1132 if (validate_ssbo
) {
1133 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumShaderStorageBlocks
;
1134 sh_blks
= sh
->ShaderStorageBlocks
;
1136 sh_num_blocks
= prog
->_LinkedShaders
[i
]->NumUniformBlocks
;
1137 sh_blks
= sh
->UniformBlocks
;
1140 for (unsigned int j
= 0; j
< sh_num_blocks
; j
++) {
1141 int index
= link_cross_validate_uniform_block(prog
, &blks
, num_blks
,
1145 linker_error(prog
, "buffer block `%s' has mismatching "
1146 "definitions\n", sh_blks
[j
]->Name
);
1148 for (unsigned k
= 0; k
<= i
; k
++) {
1149 delete[] InterfaceBlockStageIndex
[k
];
1154 InterfaceBlockStageIndex
[i
][index
] = j
;
1158 /* Update per stage block pointers to point to the program list.
1159 * FIXME: We should be able to free the per stage blocks here.
1161 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1162 for (unsigned j
= 0; j
< *num_blks
; j
++) {
1163 int stage_index
= InterfaceBlockStageIndex
[i
][j
];
1165 if (stage_index
!= -1) {
1166 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
1168 blks
[j
].stageref
|= (1 << i
);
1170 struct gl_uniform_block
**sh_blks
= validate_ssbo
?
1171 sh
->ShaderStorageBlocks
: sh
->UniformBlocks
;
1173 sh_blks
[stage_index
] = &blks
[j
];
1178 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
1179 delete[] InterfaceBlockStageIndex
[i
];
1183 prog
->ShaderStorageBlocks
= blks
;
1185 prog
->UniformBlocks
= blks
;
1192 * Populates a shaders symbol table with all global declarations
1195 populate_symbol_table(gl_linked_shader
*sh
)
1197 sh
->symbols
= new(sh
) glsl_symbol_table
;
1199 foreach_in_list(ir_instruction
, inst
, sh
->ir
) {
1203 if ((func
= inst
->as_function()) != NULL
) {
1204 sh
->symbols
->add_function(func
);
1205 } else if ((var
= inst
->as_variable()) != NULL
) {
1206 if (var
->data
.mode
!= ir_var_temporary
)
1207 sh
->symbols
->add_variable(var
);
1214 * Remap variables referenced in an instruction tree
1216 * This is used when instruction trees are cloned from one shader and placed in
1217 * another. These trees will contain references to \c ir_variable nodes that
1218 * do not exist in the target shader. This function finds these \c ir_variable
1219 * references and replaces the references with matching variables in the target
1222 * If there is no matching variable in the target shader, a clone of the
1223 * \c ir_variable is made and added to the target shader. The new variable is
1224 * added to \b both the instruction stream and the symbol table.
1226 * \param inst IR tree that is to be processed.
1227 * \param symbols Symbol table containing global scope symbols in the
1229 * \param instructions Instruction stream where new variable declarations
1233 remap_variables(ir_instruction
*inst
, struct gl_linked_shader
*target
,
1236 class remap_visitor
: public ir_hierarchical_visitor
{
1238 remap_visitor(struct gl_linked_shader
*target
, hash_table
*temps
)
1240 this->target
= target
;
1241 this->symbols
= target
->symbols
;
1242 this->instructions
= target
->ir
;
1243 this->temps
= temps
;
1246 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1248 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1249 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1250 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1252 assert(var
!= NULL
);
1254 return visit_continue
;
1257 ir_variable
*const existing
=
1258 this->symbols
->get_variable(ir
->var
->name
);
1259 if (existing
!= NULL
)
1262 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1264 this->symbols
->add_variable(copy
);
1265 this->instructions
->push_head(copy
);
1269 return visit_continue
;
1273 struct gl_linked_shader
*target
;
1274 glsl_symbol_table
*symbols
;
1275 exec_list
*instructions
;
1279 remap_visitor
v(target
, temps
);
1286 * Move non-declarations from one instruction stream to another
1288 * The intended usage pattern of this function is to pass the pointer to the
1289 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1290 * pointer) for \c last and \c false for \c make_copies on the first
1291 * call. Successive calls pass the return value of the previous call for
1292 * \c last and \c true for \c make_copies.
1294 * \param instructions Source instruction stream
1295 * \param last Instruction after which new instructions should be
1296 * inserted in the target instruction stream
1297 * \param make_copies Flag selecting whether instructions in \c instructions
1298 * should be copied (via \c ir_instruction::clone) into the
1299 * target list or moved.
1302 * The new "last" instruction in the target instruction stream. This pointer
1303 * is suitable for use as the \c last parameter of a later call to this
1307 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1308 bool make_copies
, gl_linked_shader
*target
)
1310 hash_table
*temps
= NULL
;
1313 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1314 _mesa_key_pointer_equal
);
1316 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1317 if (inst
->as_function())
1320 ir_variable
*var
= inst
->as_variable();
1321 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1324 assert(inst
->as_assignment()
1326 || inst
->as_if() /* for initializers with the ?: operator */
1327 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1330 inst
= inst
->clone(target
, NULL
);
1333 _mesa_hash_table_insert(temps
, var
, inst
);
1335 remap_variables(inst
, target
, temps
);
1340 last
->insert_after(inst
);
1345 _mesa_hash_table_destroy(temps
, NULL
);
1352 * This class is only used in link_intrastage_shaders() below but declaring
1353 * it inside that function leads to compiler warnings with some versions of
1356 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1358 array_sizing_visitor()
1359 : mem_ctx(ralloc_context(NULL
)),
1360 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1361 _mesa_key_pointer_equal
))
1365 ~array_sizing_visitor()
1367 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1368 ralloc_free(this->mem_ctx
);
1371 virtual ir_visitor_status
visit(ir_variable
*var
)
1373 const glsl_type
*type_without_array
;
1374 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1375 fixup_type(&var
->type
, var
->data
.max_array_access
,
1376 var
->data
.from_ssbo_unsized_array
,
1377 &implicit_sized_array
);
1378 var
->data
.implicit_sized_array
= implicit_sized_array
;
1379 type_without_array
= var
->type
->without_array();
1380 if (var
->type
->is_interface()) {
1381 if (interface_contains_unsized_arrays(var
->type
)) {
1382 const glsl_type
*new_type
=
1383 resize_interface_members(var
->type
,
1384 var
->get_max_ifc_array_access(),
1385 var
->is_in_shader_storage_block());
1386 var
->type
= new_type
;
1387 var
->change_interface_type(new_type
);
1389 } else if (type_without_array
->is_interface()) {
1390 if (interface_contains_unsized_arrays(type_without_array
)) {
1391 const glsl_type
*new_type
=
1392 resize_interface_members(type_without_array
,
1393 var
->get_max_ifc_array_access(),
1394 var
->is_in_shader_storage_block());
1395 var
->change_interface_type(new_type
);
1396 var
->type
= update_interface_members_array(var
->type
, new_type
);
1398 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1399 /* Store a pointer to the variable in the unnamed_interfaces
1403 _mesa_hash_table_search(this->unnamed_interfaces
,
1406 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1408 if (interface_vars
== NULL
) {
1409 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1411 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1414 unsigned index
= ifc_type
->field_index(var
->name
);
1415 assert(index
< ifc_type
->length
);
1416 assert(interface_vars
[index
] == NULL
);
1417 interface_vars
[index
] = var
;
1419 return visit_continue
;
1423 * For each unnamed interface block that was discovered while running the
1424 * visitor, adjust the interface type to reflect the newly assigned array
1425 * sizes, and fix up the ir_variable nodes to point to the new interface
1428 void fixup_unnamed_interface_types()
1430 hash_table_call_foreach(this->unnamed_interfaces
,
1431 fixup_unnamed_interface_type
, NULL
);
1436 * If the type pointed to by \c type represents an unsized array, replace
1437 * it with a sized array whose size is determined by max_array_access.
1439 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1440 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1442 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1443 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1444 max_array_access
+ 1);
1445 *implicit_sized
= true;
1446 assert(*type
!= NULL
);
1450 static const glsl_type
*
1451 update_interface_members_array(const glsl_type
*type
,
1452 const glsl_type
*new_interface_type
)
1454 const glsl_type
*element_type
= type
->fields
.array
;
1455 if (element_type
->is_array()) {
1456 const glsl_type
*new_array_type
=
1457 update_interface_members_array(element_type
, new_interface_type
);
1458 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1460 return glsl_type::get_array_instance(new_interface_type
,
1466 * Determine whether the given interface type contains unsized arrays (if
1467 * it doesn't, array_sizing_visitor doesn't need to process it).
1469 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1471 for (unsigned i
= 0; i
< type
->length
; i
++) {
1472 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1473 if (elem_type
->is_unsized_array())
1480 * Create a new interface type based on the given type, with unsized arrays
1481 * replaced by sized arrays whose size is determined by
1482 * max_ifc_array_access.
1484 static const glsl_type
*
1485 resize_interface_members(const glsl_type
*type
,
1486 const int *max_ifc_array_access
,
1489 unsigned num_fields
= type
->length
;
1490 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1491 memcpy(fields
, type
->fields
.structure
,
1492 num_fields
* sizeof(*fields
));
1493 for (unsigned i
= 0; i
< num_fields
; i
++) {
1494 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1495 /* If SSBO last member is unsized array, we don't replace it by a sized
1498 if (is_ssbo
&& i
== (num_fields
- 1))
1499 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1500 true, &implicit_sized_array
);
1502 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1503 false, &implicit_sized_array
);
1504 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1506 glsl_interface_packing packing
=
1507 (glsl_interface_packing
) type
->interface_packing
;
1508 const glsl_type
*new_ifc_type
=
1509 glsl_type::get_interface_instance(fields
, num_fields
,
1510 packing
, type
->name
);
1512 return new_ifc_type
;
1515 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1518 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1519 ir_variable
**interface_vars
= (ir_variable
**) data
;
1520 unsigned num_fields
= ifc_type
->length
;
1521 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1522 memcpy(fields
, ifc_type
->fields
.structure
,
1523 num_fields
* sizeof(*fields
));
1524 bool interface_type_changed
= false;
1525 for (unsigned i
= 0; i
< num_fields
; i
++) {
1526 if (interface_vars
[i
] != NULL
&&
1527 fields
[i
].type
!= interface_vars
[i
]->type
) {
1528 fields
[i
].type
= interface_vars
[i
]->type
;
1529 interface_type_changed
= true;
1532 if (!interface_type_changed
) {
1536 glsl_interface_packing packing
=
1537 (glsl_interface_packing
) ifc_type
->interface_packing
;
1538 const glsl_type
*new_ifc_type
=
1539 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1542 for (unsigned i
= 0; i
< num_fields
; i
++) {
1543 if (interface_vars
[i
] != NULL
)
1544 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1549 * Memory context used to allocate the data in \c unnamed_interfaces.
1554 * Hash table from const glsl_type * to an array of ir_variable *'s
1555 * pointing to the ir_variables constituting each unnamed interface block.
1557 hash_table
*unnamed_interfaces
;
1561 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1565 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1566 struct gl_shader_program
*prog
,
1567 struct gl_linked_shader
*linked_shader
,
1568 struct gl_shader
**shader_list
,
1569 unsigned num_shaders
)
1571 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1572 linked_shader
->info
.TransformFeedback
.BufferStride
[i
] = 0;
1575 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1576 struct gl_shader
*shader
= shader_list
[i
];
1578 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1579 if (shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1580 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1581 shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1582 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] !=
1583 shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1585 "intrastage shaders defined with conflicting "
1586 "xfb_stride for buffer %d (%d and %d)\n", j
,
1588 info
.TransformFeedback
.BufferStride
[j
],
1589 shader
->info
.TransformFeedback
.BufferStride
[j
]);
1593 if (shader
->info
.TransformFeedback
.BufferStride
[j
])
1594 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] =
1595 shader
->info
.TransformFeedback
.BufferStride
[j
];
1600 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1601 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1602 prog
->TransformFeedback
.BufferStride
[j
] =
1603 linked_shader
->info
.TransformFeedback
.BufferStride
[j
];
1605 /* We will validate doubles at a later stage */
1606 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1607 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1608 "multiple of 4 or if its applied to a type that is "
1609 "or contains a double a multiple of 8.",
1610 prog
->TransformFeedback
.BufferStride
[j
]);
1614 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1615 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1617 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1618 "limit has been exceeded.");
1626 * Performs the cross-validation of tessellation control shader vertices and
1627 * layout qualifiers for the attached tessellation control shaders,
1628 * and propagates them to the linked TCS and linked shader program.
1631 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1632 struct gl_linked_shader
*linked_shader
,
1633 struct gl_shader
**shader_list
,
1634 unsigned num_shaders
)
1636 linked_shader
->info
.TessCtrl
.VerticesOut
= 0;
1638 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1641 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1643 * "All tessellation control shader layout declarations in a program
1644 * must specify the same output patch vertex count. There must be at
1645 * least one layout qualifier specifying an output patch vertex count
1646 * in any program containing tessellation control shaders; however,
1647 * such a declaration is not required in all tessellation control
1651 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1652 struct gl_shader
*shader
= shader_list
[i
];
1654 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1655 if (linked_shader
->info
.TessCtrl
.VerticesOut
!= 0 &&
1656 linked_shader
->info
.TessCtrl
.VerticesOut
!=
1657 shader
->info
.TessCtrl
.VerticesOut
) {
1658 linker_error(prog
, "tessellation control shader defined with "
1659 "conflicting output vertex count (%d and %d)\n",
1660 linked_shader
->info
.TessCtrl
.VerticesOut
,
1661 shader
->info
.TessCtrl
.VerticesOut
);
1664 linked_shader
->info
.TessCtrl
.VerticesOut
=
1665 shader
->info
.TessCtrl
.VerticesOut
;
1669 /* Just do the intrastage -> interstage propagation right now,
1670 * since we already know we're in the right type of shader program
1673 if (linked_shader
->info
.TessCtrl
.VerticesOut
== 0) {
1674 linker_error(prog
, "tessellation control shader didn't declare "
1675 "vertices out layout qualifier\n");
1682 * Performs the cross-validation of tessellation evaluation shader
1683 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1684 * for the attached tessellation evaluation shaders, and propagates them
1685 * to the linked TES and linked shader program.
1688 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1689 struct gl_linked_shader
*linked_shader
,
1690 struct gl_shader
**shader_list
,
1691 unsigned num_shaders
)
1693 linked_shader
->info
.TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1694 linked_shader
->info
.TessEval
.Spacing
= 0;
1695 linked_shader
->info
.TessEval
.VertexOrder
= 0;
1696 linked_shader
->info
.TessEval
.PointMode
= -1;
1698 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1701 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1703 * "At least one tessellation evaluation shader (compilation unit) in
1704 * a program must declare a primitive mode in its input layout.
1705 * Declaration vertex spacing, ordering, and point mode identifiers is
1706 * optional. It is not required that all tessellation evaluation
1707 * shaders in a program declare a primitive mode. If spacing or
1708 * vertex ordering declarations are omitted, the tessellation
1709 * primitive generator will use equal spacing or counter-clockwise
1710 * vertex ordering, respectively. If a point mode declaration is
1711 * omitted, the tessellation primitive generator will produce lines or
1712 * triangles according to the primitive mode."
1715 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1716 struct gl_shader
*shader
= shader_list
[i
];
1718 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1719 if (linked_shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1720 linked_shader
->info
.TessEval
.PrimitiveMode
!=
1721 shader
->info
.TessEval
.PrimitiveMode
) {
1722 linker_error(prog
, "tessellation evaluation shader defined with "
1723 "conflicting input primitive modes.\n");
1726 linked_shader
->info
.TessEval
.PrimitiveMode
= shader
->info
.TessEval
.PrimitiveMode
;
1729 if (shader
->info
.TessEval
.Spacing
!= 0) {
1730 if (linked_shader
->info
.TessEval
.Spacing
!= 0 &&
1731 linked_shader
->info
.TessEval
.Spacing
!=
1732 shader
->info
.TessEval
.Spacing
) {
1733 linker_error(prog
, "tessellation evaluation shader defined with "
1734 "conflicting vertex spacing.\n");
1737 linked_shader
->info
.TessEval
.Spacing
= shader
->info
.TessEval
.Spacing
;
1740 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1741 if (linked_shader
->info
.TessEval
.VertexOrder
!= 0 &&
1742 linked_shader
->info
.TessEval
.VertexOrder
!=
1743 shader
->info
.TessEval
.VertexOrder
) {
1744 linker_error(prog
, "tessellation evaluation shader defined with "
1745 "conflicting ordering.\n");
1748 linked_shader
->info
.TessEval
.VertexOrder
=
1749 shader
->info
.TessEval
.VertexOrder
;
1752 if (shader
->info
.TessEval
.PointMode
!= -1) {
1753 if (linked_shader
->info
.TessEval
.PointMode
!= -1 &&
1754 linked_shader
->info
.TessEval
.PointMode
!=
1755 shader
->info
.TessEval
.PointMode
) {
1756 linker_error(prog
, "tessellation evaluation shader defined with "
1757 "conflicting point modes.\n");
1760 linked_shader
->info
.TessEval
.PointMode
=
1761 shader
->info
.TessEval
.PointMode
;
1766 /* Just do the intrastage -> interstage propagation right now,
1767 * since we already know we're in the right type of shader program
1770 if (linked_shader
->info
.TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1772 "tessellation evaluation shader didn't declare input "
1773 "primitive modes.\n");
1777 if (linked_shader
->info
.TessEval
.Spacing
== 0)
1778 linked_shader
->info
.TessEval
.Spacing
= GL_EQUAL
;
1780 if (linked_shader
->info
.TessEval
.VertexOrder
== 0)
1781 linked_shader
->info
.TessEval
.VertexOrder
= GL_CCW
;
1783 if (linked_shader
->info
.TessEval
.PointMode
== -1)
1784 linked_shader
->info
.TessEval
.PointMode
= GL_FALSE
;
1789 * Performs the cross-validation of layout qualifiers specified in
1790 * redeclaration of gl_FragCoord for the attached fragment shaders,
1791 * and propagates them to the linked FS and linked shader program.
1794 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1795 struct gl_linked_shader
*linked_shader
,
1796 struct gl_shader
**shader_list
,
1797 unsigned num_shaders
)
1799 linked_shader
->info
.redeclares_gl_fragcoord
= false;
1800 linked_shader
->info
.uses_gl_fragcoord
= false;
1801 linked_shader
->info
.origin_upper_left
= false;
1802 linked_shader
->info
.pixel_center_integer
= false;
1803 linked_shader
->info
.BlendSupport
= 0;
1805 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1806 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1809 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1810 struct gl_shader
*shader
= shader_list
[i
];
1811 /* From the GLSL 1.50 spec, page 39:
1813 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1814 * it must be redeclared in all the fragment shaders in that program
1815 * that have a static use gl_FragCoord."
1817 if ((linked_shader
->info
.redeclares_gl_fragcoord
1818 && !shader
->info
.redeclares_gl_fragcoord
1819 && shader
->info
.uses_gl_fragcoord
)
1820 || (shader
->info
.redeclares_gl_fragcoord
1821 && !linked_shader
->info
.redeclares_gl_fragcoord
1822 && linked_shader
->info
.uses_gl_fragcoord
)) {
1823 linker_error(prog
, "fragment shader defined with conflicting "
1824 "layout qualifiers for gl_FragCoord\n");
1827 /* From the GLSL 1.50 spec, page 39:
1829 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1830 * single program must have the same set of qualifiers."
1832 if (linked_shader
->info
.redeclares_gl_fragcoord
&&
1833 shader
->info
.redeclares_gl_fragcoord
&&
1834 (shader
->info
.origin_upper_left
!=
1835 linked_shader
->info
.origin_upper_left
||
1836 shader
->info
.pixel_center_integer
!=
1837 linked_shader
->info
.pixel_center_integer
)) {
1838 linker_error(prog
, "fragment shader defined with conflicting "
1839 "layout qualifiers for gl_FragCoord\n");
1842 /* Update the linked shader state. Note that uses_gl_fragcoord should
1843 * accumulate the results. The other values should replace. If there
1844 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1845 * are already known to be the same.
1847 if (shader
->info
.redeclares_gl_fragcoord
||
1848 shader
->info
.uses_gl_fragcoord
) {
1849 linked_shader
->info
.redeclares_gl_fragcoord
=
1850 shader
->info
.redeclares_gl_fragcoord
;
1851 linked_shader
->info
.uses_gl_fragcoord
=
1852 linked_shader
->info
.uses_gl_fragcoord
||
1853 shader
->info
.uses_gl_fragcoord
;
1854 linked_shader
->info
.origin_upper_left
=
1855 shader
->info
.origin_upper_left
;
1856 linked_shader
->info
.pixel_center_integer
=
1857 shader
->info
.pixel_center_integer
;
1860 linked_shader
->info
.EarlyFragmentTests
|=
1861 shader
->info
.EarlyFragmentTests
;
1862 linked_shader
->info
.BlendSupport
|= shader
->info
.BlendSupport
;
1867 * Performs the cross-validation of geometry shader max_vertices and
1868 * primitive type layout qualifiers for the attached geometry shaders,
1869 * and propagates them to the linked GS and linked shader program.
1872 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1873 struct gl_linked_shader
*linked_shader
,
1874 struct gl_shader
**shader_list
,
1875 unsigned num_shaders
)
1877 linked_shader
->info
.Geom
.VerticesOut
= -1;
1878 linked_shader
->info
.Geom
.Invocations
= 0;
1879 linked_shader
->info
.Geom
.InputType
= PRIM_UNKNOWN
;
1880 linked_shader
->info
.Geom
.OutputType
= PRIM_UNKNOWN
;
1882 /* No in/out qualifiers defined for anything but GLSL 1.50+
1883 * geometry shaders so far.
1885 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1888 /* From the GLSL 1.50 spec, page 46:
1890 * "All geometry shader output layout declarations in a program
1891 * must declare the same layout and same value for
1892 * max_vertices. There must be at least one geometry output
1893 * layout declaration somewhere in a program, but not all
1894 * geometry shaders (compilation units) are required to
1898 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1899 struct gl_shader
*shader
= shader_list
[i
];
1901 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1902 if (linked_shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
&&
1903 linked_shader
->info
.Geom
.InputType
!=
1904 shader
->info
.Geom
.InputType
) {
1905 linker_error(prog
, "geometry shader defined with conflicting "
1909 linked_shader
->info
.Geom
.InputType
= shader
->info
.Geom
.InputType
;
1912 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1913 if (linked_shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
&&
1914 linked_shader
->info
.Geom
.OutputType
!=
1915 shader
->info
.Geom
.OutputType
) {
1916 linker_error(prog
, "geometry shader defined with conflicting "
1920 linked_shader
->info
.Geom
.OutputType
= shader
->info
.Geom
.OutputType
;
1923 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1924 if (linked_shader
->info
.Geom
.VerticesOut
!= -1 &&
1925 linked_shader
->info
.Geom
.VerticesOut
!=
1926 shader
->info
.Geom
.VerticesOut
) {
1927 linker_error(prog
, "geometry shader defined with conflicting "
1928 "output vertex count (%d and %d)\n",
1929 linked_shader
->info
.Geom
.VerticesOut
,
1930 shader
->info
.Geom
.VerticesOut
);
1933 linked_shader
->info
.Geom
.VerticesOut
= shader
->info
.Geom
.VerticesOut
;
1936 if (shader
->info
.Geom
.Invocations
!= 0) {
1937 if (linked_shader
->info
.Geom
.Invocations
!= 0 &&
1938 linked_shader
->info
.Geom
.Invocations
!=
1939 shader
->info
.Geom
.Invocations
) {
1940 linker_error(prog
, "geometry shader defined with conflicting "
1941 "invocation count (%d and %d)\n",
1942 linked_shader
->info
.Geom
.Invocations
,
1943 shader
->info
.Geom
.Invocations
);
1946 linked_shader
->info
.Geom
.Invocations
= shader
->info
.Geom
.Invocations
;
1950 /* Just do the intrastage -> interstage propagation right now,
1951 * since we already know we're in the right type of shader program
1954 if (linked_shader
->info
.Geom
.InputType
== PRIM_UNKNOWN
) {
1956 "geometry shader didn't declare primitive input type\n");
1960 if (linked_shader
->info
.Geom
.OutputType
== PRIM_UNKNOWN
) {
1962 "geometry shader didn't declare primitive output type\n");
1966 if (linked_shader
->info
.Geom
.VerticesOut
== -1) {
1968 "geometry shader didn't declare max_vertices\n");
1972 if (linked_shader
->info
.Geom
.Invocations
== 0)
1973 linked_shader
->info
.Geom
.Invocations
= 1;
1978 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1979 * qualifiers for the attached compute shaders, and propagate them to the
1980 * linked CS and linked shader program.
1983 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1984 struct gl_linked_shader
*linked_shader
,
1985 struct gl_shader
**shader_list
,
1986 unsigned num_shaders
)
1988 for (int i
= 0; i
< 3; i
++)
1989 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
1991 /* This function is called for all shader stages, but it only has an effect
1992 * for compute shaders.
1994 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1997 /* From the ARB_compute_shader spec, in the section describing local size
2000 * If multiple compute shaders attached to a single program object
2001 * declare local work-group size, the declarations must be identical;
2002 * otherwise a link-time error results. Furthermore, if a program
2003 * object contains any compute shaders, at least one must contain an
2004 * input layout qualifier specifying the local work sizes of the
2005 * program, or a link-time error will occur.
2007 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2008 struct gl_shader
*shader
= shader_list
[sh
];
2010 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2011 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2012 for (int i
= 0; i
< 3; i
++) {
2013 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2014 shader
->info
.Comp
.LocalSize
[i
]) {
2015 linker_error(prog
, "compute shader defined with conflicting "
2021 for (int i
= 0; i
< 3; i
++) {
2022 linked_shader
->info
.Comp
.LocalSize
[i
] =
2023 shader
->info
.Comp
.LocalSize
[i
];
2028 /* Just do the intrastage -> interstage propagation right now,
2029 * since we already know we're in the right type of shader program
2032 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0) {
2033 linker_error(prog
, "compute shader didn't declare local size\n");
2036 for (int i
= 0; i
< 3; i
++)
2037 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2042 * Combine a group of shaders for a single stage to generate a linked shader
2045 * If this function is supplied a single shader, it is cloned, and the new
2046 * shader is returned.
2048 static struct gl_linked_shader
*
2049 link_intrastage_shaders(void *mem_ctx
,
2050 struct gl_context
*ctx
,
2051 struct gl_shader_program
*prog
,
2052 struct gl_shader
**shader_list
,
2053 unsigned num_shaders
)
2055 struct gl_uniform_block
*ubo_blocks
= NULL
;
2056 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2057 unsigned num_ubo_blocks
= 0;
2058 unsigned num_ssbo_blocks
= 0;
2060 /* Check that global variables defined in multiple shaders are consistent.
2062 glsl_symbol_table variables
;
2063 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2064 if (shader_list
[i
] == NULL
)
2066 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2069 if (!prog
->LinkStatus
)
2072 /* Check that interface blocks defined in multiple shaders are consistent.
2074 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2076 if (!prog
->LinkStatus
)
2079 /* Check that there is only a single definition of each function signature
2080 * across all shaders.
2082 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2083 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2084 ir_function
*const f
= node
->as_function();
2089 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2090 ir_function
*const other
=
2091 shader_list
[j
]->symbols
->get_function(f
->name
);
2093 /* If the other shader has no function (and therefore no function
2094 * signatures) with the same name, skip to the next shader.
2099 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2100 if (!sig
->is_defined
)
2103 ir_function_signature
*other_sig
=
2104 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2106 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2107 linker_error(prog
, "function `%s' is multiply defined\n",
2116 /* Find the shader that defines main, and make a clone of it.
2118 * Starting with the clone, search for undefined references. If one is
2119 * found, find the shader that defines it. Clone the reference and add
2120 * it to the shader. Repeat until there are no undefined references or
2121 * until a reference cannot be resolved.
2123 gl_shader
*main
= NULL
;
2124 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2125 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2126 main
= shader_list
[i
];
2132 linker_error(prog
, "%s shader lacks `main'\n",
2133 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2137 gl_linked_shader
*linked
= ctx
->Driver
.NewShader(shader_list
[0]->Stage
);
2138 linked
->ir
= new(linked
) exec_list
;
2139 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2141 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2142 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2143 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2144 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2145 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2146 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2149 populate_symbol_table(linked
);
2151 /* The pointer to the main function in the final linked shader (i.e., the
2152 * copy of the original shader that contained the main function).
2154 ir_function_signature
*const main_sig
=
2155 _mesa_get_main_function_signature(linked
->symbols
);
2157 /* Move any instructions other than variable declarations or function
2158 * declarations into main.
2160 exec_node
*insertion_point
=
2161 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2164 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2165 if (shader_list
[i
] == main
)
2168 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2169 insertion_point
, true, linked
);
2172 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2173 _mesa_delete_linked_shader(ctx
, linked
);
2177 /* Make a pass over all variable declarations to ensure that arrays with
2178 * unspecified sizes have a size specified. The size is inferred from the
2179 * max_array_access field.
2181 array_sizing_visitor v
;
2183 v
.fixup_unnamed_interface_types();
2185 /* Link up uniform blocks defined within this stage. */
2186 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2187 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2189 if (!prog
->LinkStatus
) {
2190 _mesa_delete_linked_shader(ctx
, linked
);
2194 /* Copy ubo blocks to linked shader list */
2195 linked
->UniformBlocks
=
2196 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2197 ralloc_steal(linked
, ubo_blocks
);
2198 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2199 linked
->UniformBlocks
[i
] = &ubo_blocks
[i
];
2201 linked
->NumUniformBlocks
= num_ubo_blocks
;
2203 /* Copy ssbo blocks to linked shader list */
2204 linked
->ShaderStorageBlocks
=
2205 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2206 ralloc_steal(linked
, ssbo_blocks
);
2207 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2208 linked
->ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2210 linked
->NumShaderStorageBlocks
= num_ssbo_blocks
;
2212 /* At this point linked should contain all of the linked IR, so
2213 * validate it to make sure nothing went wrong.
2215 validate_ir_tree(linked
->ir
);
2217 /* Set the size of geometry shader input arrays */
2218 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2219 unsigned num_vertices
= vertices_per_prim(linked
->info
.Geom
.InputType
);
2220 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2221 MESA_SHADER_GEOMETRY
);
2222 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2223 ir
->accept(&input_resize_visitor
);
2227 if (ctx
->Const
.VertexID_is_zero_based
)
2228 lower_vertex_id(linked
);
2234 * Update the sizes of linked shader uniform arrays to the maximum
2237 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2239 * If one or more elements of an array are active,
2240 * GetActiveUniform will return the name of the array in name,
2241 * subject to the restrictions listed above. The type of the array
2242 * is returned in type. The size parameter contains the highest
2243 * array element index used, plus one. The compiler or linker
2244 * determines the highest index used. There will be only one
2245 * active uniform reported by the GL per uniform array.
2249 update_array_sizes(struct gl_shader_program
*prog
)
2251 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2252 if (prog
->_LinkedShaders
[i
] == NULL
)
2255 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2256 ir_variable
*const var
= node
->as_variable();
2258 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2259 !var
->type
->is_array())
2262 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2263 * will not be eliminated. Since we always do std140, just
2264 * don't resize arrays in UBOs.
2266 * Atomic counters are supposed to get deterministic
2267 * locations assigned based on the declaration ordering and
2268 * sizes, array compaction would mess that up.
2270 * Subroutine uniforms are not removed.
2272 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2273 var
->type
->contains_subroutine() || var
->constant_initializer
)
2276 int size
= var
->data
.max_array_access
;
2277 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2278 if (prog
->_LinkedShaders
[j
] == NULL
)
2281 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2282 ir_variable
*other_var
= node2
->as_variable();
2286 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2287 other_var
->data
.max_array_access
> size
) {
2288 size
= other_var
->data
.max_array_access
;
2293 if (size
+ 1 != (int)var
->type
->length
) {
2294 /* If this is a built-in uniform (i.e., it's backed by some
2295 * fixed-function state), adjust the number of state slots to
2296 * match the new array size. The number of slots per array entry
2297 * is not known. It seems safe to assume that the total number of
2298 * slots is an integer multiple of the number of array elements.
2299 * Determine the number of slots per array element by dividing by
2300 * the old (total) size.
2302 const unsigned num_slots
= var
->get_num_state_slots();
2303 if (num_slots
> 0) {
2304 var
->set_num_state_slots((size
+ 1)
2305 * (num_slots
/ var
->type
->length
));
2308 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2310 /* FINISHME: We should update the types of array
2311 * dereferences of this variable now.
2319 * Resize tessellation evaluation per-vertex inputs to the size of
2320 * tessellation control per-vertex outputs.
2323 resize_tes_inputs(struct gl_context
*ctx
,
2324 struct gl_shader_program
*prog
)
2326 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2329 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2330 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2332 /* If no control shader is present, then the TES inputs are statically
2333 * sized to MaxPatchVertices; the actual size of the arrays won't be
2334 * known until draw time.
2336 const int num_vertices
= tcs
2337 ? tcs
->info
.TessCtrl
.VerticesOut
2338 : ctx
->Const
.MaxPatchVertices
;
2340 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2341 MESA_SHADER_TESS_EVAL
);
2342 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2343 ir
->accept(&input_resize_visitor
);
2346 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2347 /* Convert the gl_PatchVerticesIn system value into a constant, since
2348 * the value is known at this point.
2350 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2351 ir_variable
*var
= ir
->as_variable();
2352 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2353 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2354 void *mem_ctx
= ralloc_parent(var
);
2355 var
->data
.location
= 0;
2356 var
->data
.explicit_location
= false;
2358 var
->data
.mode
= ir_var_auto
;
2359 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2361 var
->data
.mode
= ir_var_uniform
;
2362 var
->data
.how_declared
= ir_var_hidden
;
2363 var
->allocate_state_slots(1);
2364 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2365 slot0
->swizzle
= SWIZZLE_XXXX
;
2366 slot0
->tokens
[0] = STATE_INTERNAL
;
2367 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2368 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2369 slot0
->tokens
[i
] = 0;
2377 * Find a contiguous set of available bits in a bitmask.
2379 * \param used_mask Bits representing used (1) and unused (0) locations
2380 * \param needed_count Number of contiguous bits needed.
2383 * Base location of the available bits on success or -1 on failure.
2386 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2388 unsigned needed_mask
= (1 << needed_count
) - 1;
2389 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2391 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2392 * cannot optimize possibly infinite loops" for the loop below.
2394 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2397 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2398 if ((needed_mask
& ~used_mask
) == needed_mask
)
2409 * Assign locations for either VS inputs or FS outputs
2411 * \param mem_ctx Temporary ralloc context used for linking
2412 * \param prog Shader program whose variables need locations assigned
2413 * \param constants Driver specific constant values for the program.
2414 * \param target_index Selector for the program target to receive location
2415 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2416 * \c MESA_SHADER_FRAGMENT.
2419 * If locations are successfully assigned, true is returned. Otherwise an
2420 * error is emitted to the shader link log and false is returned.
2423 assign_attribute_or_color_locations(void *mem_ctx
,
2424 gl_shader_program
*prog
,
2425 struct gl_constants
*constants
,
2426 unsigned target_index
)
2428 /* Maximum number of generic locations. This corresponds to either the
2429 * maximum number of draw buffers or the maximum number of generic
2432 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2433 constants
->Program
[target_index
].MaxAttribs
:
2434 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2436 /* Mark invalid locations as being used.
2438 unsigned used_locations
= (max_index
>= 32)
2439 ? ~0 : ~((1 << max_index
) - 1);
2440 unsigned double_storage_locations
= 0;
2442 assert((target_index
== MESA_SHADER_VERTEX
)
2443 || (target_index
== MESA_SHADER_FRAGMENT
));
2445 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2449 /* Operate in a total of four passes.
2451 * 1. Invalidate the location assignments for all vertex shader inputs.
2453 * 2. Assign locations for inputs that have user-defined (via
2454 * glBindVertexAttribLocation) locations and outputs that have
2455 * user-defined locations (via glBindFragDataLocation).
2457 * 3. Sort the attributes without assigned locations by number of slots
2458 * required in decreasing order. Fragmentation caused by attribute
2459 * locations assigned by the application may prevent large attributes
2460 * from having enough contiguous space.
2462 * 4. Assign locations to any inputs without assigned locations.
2465 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2466 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2468 const enum ir_variable_mode direction
=
2469 (target_index
== MESA_SHADER_VERTEX
)
2470 ? ir_var_shader_in
: ir_var_shader_out
;
2473 /* Temporary storage for the set of attributes that need locations assigned.
2479 /* Used below in the call to qsort. */
2480 static int compare(const void *a
, const void *b
)
2482 const temp_attr
*const l
= (const temp_attr
*) a
;
2483 const temp_attr
*const r
= (const temp_attr
*) b
;
2485 /* Reversed because we want a descending order sort below. */
2486 return r
->slots
- l
->slots
;
2489 assert(max_index
<= 32);
2491 /* Temporary array for the set of attributes that have locations assigned.
2493 ir_variable
*assigned
[16];
2495 unsigned num_attr
= 0;
2496 unsigned assigned_attr
= 0;
2498 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2499 ir_variable
*const var
= node
->as_variable();
2501 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2504 if (var
->data
.explicit_location
) {
2505 var
->data
.is_unmatched_generic_inout
= 0;
2506 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2507 || (var
->data
.location
< 0)) {
2509 "invalid explicit location %d specified for `%s'\n",
2510 (var
->data
.location
< 0)
2511 ? var
->data
.location
2512 : var
->data
.location
- generic_base
,
2516 } else if (target_index
== MESA_SHADER_VERTEX
) {
2519 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2520 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2521 var
->data
.location
= binding
;
2522 var
->data
.is_unmatched_generic_inout
= 0;
2524 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2527 const char *name
= var
->name
;
2528 const glsl_type
*type
= var
->type
;
2531 /* Check if there's a binding for the variable name */
2532 if (prog
->FragDataBindings
->get(binding
, name
)) {
2533 assert(binding
>= FRAG_RESULT_DATA0
);
2534 var
->data
.location
= binding
;
2535 var
->data
.is_unmatched_generic_inout
= 0;
2537 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2538 var
->data
.index
= index
;
2543 /* If not, but it's an array type, look for name[0] */
2544 if (type
->is_array()) {
2545 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2546 type
= type
->fields
.array
;
2554 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2557 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2559 * "Output binding assignments will cause LinkProgram to fail:
2561 * If the program has an active output assigned to a location greater
2562 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2563 * an active output assigned an index greater than or equal to one;"
2565 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2566 var
->data
.location
- generic_base
>=
2567 (int) constants
->MaxDualSourceDrawBuffers
) {
2569 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2570 "with index %u for %s\n",
2571 var
->data
.location
- generic_base
, var
->data
.index
,
2576 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2578 /* If the variable is not a built-in and has a location statically
2579 * assigned in the shader (presumably via a layout qualifier), make sure
2580 * that it doesn't collide with other assigned locations. Otherwise,
2581 * add it to the list of variables that need linker-assigned locations.
2583 if (var
->data
.location
!= -1) {
2584 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2585 /* From page 61 of the OpenGL 4.0 spec:
2587 * "LinkProgram will fail if the attribute bindings assigned
2588 * by BindAttribLocation do not leave not enough space to
2589 * assign a location for an active matrix attribute or an
2590 * active attribute array, both of which require multiple
2591 * contiguous generic attributes."
2593 * I think above text prohibits the aliasing of explicit and
2594 * automatic assignments. But, aliasing is allowed in manual
2595 * assignments of attribute locations. See below comments for
2598 * From OpenGL 4.0 spec, page 61:
2600 * "It is possible for an application to bind more than one
2601 * attribute name to the same location. This is referred to as
2602 * aliasing. This will only work if only one of the aliased
2603 * attributes is active in the executable program, or if no
2604 * path through the shader consumes more than one attribute of
2605 * a set of attributes aliased to the same location. A link
2606 * error can occur if the linker determines that every path
2607 * through the shader consumes multiple aliased attributes,
2608 * but implementations are not required to generate an error
2611 * From GLSL 4.30 spec, page 54:
2613 * "A program will fail to link if any two non-vertex shader
2614 * input variables are assigned to the same location. For
2615 * vertex shaders, multiple input variables may be assigned
2616 * to the same location using either layout qualifiers or via
2617 * the OpenGL API. However, such aliasing is intended only to
2618 * support vertex shaders where each execution path accesses
2619 * at most one input per each location. Implementations are
2620 * permitted, but not required, to generate link-time errors
2621 * if they detect that every path through the vertex shader
2622 * executable accesses multiple inputs assigned to any single
2623 * location. For all shader types, a program will fail to link
2624 * if explicit location assignments leave the linker unable
2625 * to find space for other variables without explicit
2628 * From OpenGL ES 3.0 spec, page 56:
2630 * "Binding more than one attribute name to the same location
2631 * is referred to as aliasing, and is not permitted in OpenGL
2632 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2633 * fail when this condition exists. However, aliasing is
2634 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2635 * This will only work if only one of the aliased attributes
2636 * is active in the executable program, or if no path through
2637 * the shader consumes more than one attribute of a set of
2638 * attributes aliased to the same location. A link error can
2639 * occur if the linker determines that every path through the
2640 * shader consumes multiple aliased attributes, but implemen-
2641 * tations are not required to generate an error in this case."
2643 * After looking at above references from OpenGL, OpenGL ES and
2644 * GLSL specifications, we allow aliasing of vertex input variables
2645 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2647 * NOTE: This is not required by the spec but its worth mentioning
2648 * here that we're not doing anything to make sure that no path
2649 * through the vertex shader executable accesses multiple inputs
2650 * assigned to any single location.
2653 /* Mask representing the contiguous slots that will be used by
2656 const unsigned attr
= var
->data
.location
- generic_base
;
2657 const unsigned use_mask
= (1 << slots
) - 1;
2658 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2659 ? "vertex shader input" : "fragment shader output";
2661 /* Generate a link error if the requested locations for this
2662 * attribute exceed the maximum allowed attribute location.
2664 if (attr
+ slots
> max_index
) {
2666 "insufficient contiguous locations "
2667 "available for %s `%s' %d %d %d\n", string
,
2668 var
->name
, used_locations
, use_mask
, attr
);
2672 /* Generate a link error if the set of bits requested for this
2673 * attribute overlaps any previously allocated bits.
2675 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2676 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2677 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2680 * "Additionally, for fragment shader outputs, if two
2681 * variables are placed within the same location, they
2682 * must have the same underlying type (floating-point or
2683 * integer). No component aliasing of output variables or
2684 * members is allowed.
2686 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2687 unsigned assigned_slots
=
2688 assigned
[i
]->type
->count_attribute_slots(false);
2689 unsigned assig_attr
=
2690 assigned
[i
]->data
.location
- generic_base
;
2691 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2693 if ((assigned_use_mask
<< assig_attr
) &
2694 (use_mask
<< attr
)) {
2696 const glsl_type
*assigned_type
=
2697 assigned
[i
]->type
->without_array();
2698 const glsl_type
*type
= var
->type
->without_array();
2699 if (assigned_type
->base_type
!= type
->base_type
) {
2700 linker_error(prog
, "types do not match for aliased"
2701 " %ss %s and %s\n", string
,
2702 assigned
[i
]->name
, var
->name
);
2706 unsigned assigned_component_mask
=
2707 ((1 << assigned_type
->vector_elements
) - 1) <<
2708 assigned
[i
]->data
.location_frac
;
2709 unsigned component_mask
=
2710 ((1 << type
->vector_elements
) - 1) <<
2711 var
->data
.location_frac
;
2712 if (assigned_component_mask
& component_mask
) {
2713 linker_error(prog
, "overlapping component is "
2714 "assigned to %ss %s and %s "
2716 string
, assigned
[i
]->name
, var
->name
,
2717 var
->data
.location_frac
);
2722 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2723 (prog
->IsES
&& prog
->Version
>= 300)) {
2724 linker_error(prog
, "overlapping location is assigned "
2725 "to %s `%s' %d %d %d\n", string
, var
->name
,
2726 used_locations
, use_mask
, attr
);
2729 linker_warning(prog
, "overlapping location is assigned "
2730 "to %s `%s' %d %d %d\n", string
, var
->name
,
2731 used_locations
, use_mask
, attr
);
2735 used_locations
|= (use_mask
<< attr
);
2737 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2739 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2740 * active attribute variables may fail to link, unless
2741 * device-dependent optimizations are able to make the program
2742 * fit within available hardware resources. For the purposes
2743 * of this test, attribute variables of the type dvec3, dvec4,
2744 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2745 * count as consuming twice as many attributes as equivalent
2746 * single-precision types. While these types use the same number
2747 * of generic attributes as their single-precision equivalents,
2748 * implementations are permitted to consume two single-precision
2749 * vectors of internal storage for each three- or four-component
2750 * double-precision vector."
2752 * Mark this attribute slot as taking up twice as much space
2753 * so we can count it properly against limits. According to
2754 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2755 * is optional behavior, but it seems preferable.
2757 if (var
->type
->without_array()->is_dual_slot())
2758 double_storage_locations
|= (use_mask
<< attr
);
2761 assigned
[assigned_attr
] = var
;
2767 if (num_attr
>= max_index
) {
2768 linker_error(prog
, "too many %s (max %u)",
2769 target_index
== MESA_SHADER_VERTEX
?
2770 "vertex shader inputs" : "fragment shader outputs",
2774 to_assign
[num_attr
].slots
= slots
;
2775 to_assign
[num_attr
].var
= var
;
2779 if (target_index
== MESA_SHADER_VERTEX
) {
2780 unsigned total_attribs_size
=
2781 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2782 _mesa_bitcount(double_storage_locations
);
2783 if (total_attribs_size
> max_index
) {
2785 "attempt to use %d vertex attribute slots only %d available ",
2786 total_attribs_size
, max_index
);
2791 /* If all of the attributes were assigned locations by the application (or
2792 * are built-in attributes with fixed locations), return early. This should
2793 * be the common case.
2798 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2800 if (target_index
== MESA_SHADER_VERTEX
) {
2801 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2802 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2803 * reserved to prevent it from being automatically allocated below.
2805 find_deref_visitor
find("gl_Vertex");
2807 if (find
.variable_found())
2808 used_locations
|= (1 << 0);
2811 for (unsigned i
= 0; i
< num_attr
; i
++) {
2812 /* Mask representing the contiguous slots that will be used by this
2815 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2817 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2820 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2821 ? "vertex shader input" : "fragment shader output";
2824 "insufficient contiguous locations "
2825 "available for %s `%s'\n",
2826 string
, to_assign
[i
].var
->name
);
2830 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2831 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2832 used_locations
|= (use_mask
<< location
);
2834 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2835 double_storage_locations
|= (use_mask
<< location
);
2838 /* Now that we have all the locations, from the GL 4.5 core spec, section
2839 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2840 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2841 * as equivalent single-precision types.
2843 if (target_index
== MESA_SHADER_VERTEX
) {
2844 unsigned total_attribs_size
=
2845 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2846 _mesa_bitcount(double_storage_locations
);
2847 if (total_attribs_size
> max_index
) {
2849 "attempt to use %d vertex attribute slots only %d available ",
2850 total_attribs_size
, max_index
);
2859 * Match explicit locations of outputs to inputs and deactivate the
2860 * unmatch flag if found so we don't optimise them away.
2863 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2864 gl_linked_shader
*consumer
)
2866 glsl_symbol_table parameters
;
2867 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2870 /* Find all shader outputs in the "producer" stage.
2872 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2873 ir_variable
*const var
= node
->as_variable();
2875 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2878 if (var
->data
.explicit_location
&&
2879 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2880 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2881 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2882 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2886 /* Match inputs to outputs */
2887 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2888 ir_variable
*const input
= node
->as_variable();
2890 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2893 ir_variable
*output
= NULL
;
2894 if (input
->data
.explicit_location
2895 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2896 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2897 [input
->data
.location_frac
];
2899 if (output
!= NULL
){
2900 input
->data
.is_unmatched_generic_inout
= 0;
2901 output
->data
.is_unmatched_generic_inout
= 0;
2908 * Store the gl_FragDepth layout in the gl_shader_program struct.
2911 store_fragdepth_layout(struct gl_shader_program
*prog
)
2913 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2917 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2919 /* We don't look up the gl_FragDepth symbol directly because if
2920 * gl_FragDepth is not used in the shader, it's removed from the IR.
2921 * However, the symbol won't be removed from the symbol table.
2923 * We're only interested in the cases where the variable is NOT removed
2926 foreach_in_list(ir_instruction
, node
, ir
) {
2927 ir_variable
*const var
= node
->as_variable();
2929 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2933 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2934 switch (var
->data
.depth_layout
) {
2935 case ir_depth_layout_none
:
2936 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2938 case ir_depth_layout_any
:
2939 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2941 case ir_depth_layout_greater
:
2942 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2944 case ir_depth_layout_less
:
2945 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2947 case ir_depth_layout_unchanged
:
2948 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2959 * Validate the resources used by a program versus the implementation limits
2962 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2964 unsigned total_uniform_blocks
= 0;
2965 unsigned total_shader_storage_blocks
= 0;
2967 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2968 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
2973 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2974 linker_error(prog
, "Too many %s shader texture samplers\n",
2975 _mesa_shader_stage_to_string(i
));
2978 if (sh
->num_uniform_components
>
2979 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2980 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2981 linker_warning(prog
, "Too many %s shader default uniform block "
2982 "components, but the driver will try to optimize "
2983 "them out; this is non-portable out-of-spec "
2985 _mesa_shader_stage_to_string(i
));
2987 linker_error(prog
, "Too many %s shader default uniform block "
2989 _mesa_shader_stage_to_string(i
));
2993 if (sh
->num_combined_uniform_components
>
2994 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2995 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2996 linker_warning(prog
, "Too many %s shader uniform components, "
2997 "but the driver will try to optimize them out; "
2998 "this is non-portable out-of-spec behavior\n",
2999 _mesa_shader_stage_to_string(i
));
3001 linker_error(prog
, "Too many %s shader uniform components\n",
3002 _mesa_shader_stage_to_string(i
));
3006 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3007 total_uniform_blocks
+= sh
->NumUniformBlocks
;
3009 const unsigned max_uniform_blocks
=
3010 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3011 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
3012 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3013 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
3014 max_uniform_blocks
);
3017 const unsigned max_shader_storage_blocks
=
3018 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3019 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
3020 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3021 _mesa_shader_stage_to_string(i
),
3022 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
3026 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3027 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3028 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3031 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3032 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3033 total_shader_storage_blocks
,
3034 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3037 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
3038 if (prog
->UniformBlocks
[i
].UniformBufferSize
>
3039 ctx
->Const
.MaxUniformBlockSize
) {
3040 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3041 prog
->UniformBlocks
[i
].Name
,
3042 prog
->UniformBlocks
[i
].UniformBufferSize
,
3043 ctx
->Const
.MaxUniformBlockSize
);
3047 for (unsigned i
= 0; i
< prog
->NumShaderStorageBlocks
; i
++) {
3048 if (prog
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3049 ctx
->Const
.MaxShaderStorageBlockSize
) {
3050 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3051 prog
->ShaderStorageBlocks
[i
].Name
,
3052 prog
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3053 ctx
->Const
.MaxShaderStorageBlockSize
);
3059 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3061 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3062 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3067 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3068 if (sh
->SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3071 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3076 sh
->NumSubroutineUniforms
++;
3078 if (sh
->NumSubroutineFunctions
== 0) {
3079 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3082 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3083 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3084 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3085 if (fn
->types
[k
] == uni
->type
) {
3091 uni
->num_compatible_subroutines
= count
;
3097 check_subroutine_resources(struct gl_shader_program
*prog
)
3099 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3100 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3103 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3104 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3105 _mesa_shader_stage_to_string(i
));
3110 * Validate shader image resources.
3113 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3115 unsigned total_image_units
= 0;
3116 unsigned fragment_outputs
= 0;
3117 unsigned total_shader_storage_blocks
= 0;
3119 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3122 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3123 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3126 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3127 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3128 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3129 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3131 total_image_units
+= sh
->NumImages
;
3132 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3134 if (i
== MESA_SHADER_FRAGMENT
) {
3135 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3136 ir_variable
*var
= node
->as_variable();
3137 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3138 /* since there are no double fs outputs - pass false */
3139 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3145 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3146 linker_error(prog
, "Too many combined image uniforms\n");
3148 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3149 ctx
->Const
.MaxCombinedShaderOutputResources
)
3150 linker_error(prog
, "Too many combined image uniforms, shader storage "
3151 " buffers and fragment outputs\n");
3156 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3157 * for a variable, checks for overlaps between other uniforms using explicit
3161 reserve_explicit_locations(struct gl_shader_program
*prog
,
3162 string_to_uint_map
*map
, ir_variable
*var
)
3164 unsigned slots
= var
->type
->uniform_locations();
3165 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3166 unsigned return_value
= slots
;
3168 /* Resize remap table if locations do not fit in the current one. */
3169 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3170 prog
->UniformRemapTable
=
3171 reralloc(prog
, prog
->UniformRemapTable
,
3172 gl_uniform_storage
*,
3175 if (!prog
->UniformRemapTable
) {
3176 linker_error(prog
, "Out of memory during linking.\n");
3180 /* Initialize allocated space. */
3181 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3182 prog
->UniformRemapTable
[i
] = NULL
;
3184 prog
->NumUniformRemapTable
= max_loc
+ 1;
3187 for (unsigned i
= 0; i
< slots
; i
++) {
3188 unsigned loc
= var
->data
.location
+ i
;
3190 /* Check if location is already used. */
3191 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3193 /* Possibly same uniform from a different stage, this is ok. */
3195 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3200 /* ARB_explicit_uniform_location specification states:
3202 * "No two default-block uniform variables in the program can have
3203 * the same location, even if they are unused, otherwise a compiler
3204 * or linker error will be generated."
3207 "location qualifier for uniform %s overlaps "
3208 "previously used location\n",
3213 /* Initialize location as inactive before optimization
3214 * rounds and location assignment.
3216 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3219 /* Note, base location used for arrays. */
3220 map
->put(var
->data
.location
, var
->name
);
3222 return return_value
;
3226 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3227 struct gl_linked_shader
*sh
,
3230 unsigned slots
= var
->type
->uniform_locations();
3231 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3233 /* Resize remap table if locations do not fit in the current one. */
3234 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3235 sh
->SubroutineUniformRemapTable
=
3236 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3237 gl_uniform_storage
*,
3240 if (!sh
->SubroutineUniformRemapTable
) {
3241 linker_error(prog
, "Out of memory during linking.\n");
3245 /* Initialize allocated space. */
3246 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3247 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3249 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3252 for (unsigned i
= 0; i
< slots
; i
++) {
3253 unsigned loc
= var
->data
.location
+ i
;
3255 /* Check if location is already used. */
3256 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3258 /* ARB_explicit_uniform_location specification states:
3259 * "No two subroutine uniform variables can have the same location
3260 * in the same shader stage, otherwise a compiler or linker error
3261 * will be generated."
3264 "location qualifier for uniform %s overlaps "
3265 "previously used location\n",
3270 /* Initialize location as inactive before optimization
3271 * rounds and location assignment.
3273 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3279 * Check and reserve all explicit uniform locations, called before
3280 * any optimizations happen to handle also inactive uniforms and
3281 * inactive array elements that may get trimmed away.
3284 check_explicit_uniform_locations(struct gl_context
*ctx
,
3285 struct gl_shader_program
*prog
)
3287 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3290 /* This map is used to detect if overlapping explicit locations
3291 * occur with the same uniform (from different stage) or a different one.
3293 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3296 linker_error(prog
, "Out of memory during linking.\n");
3300 unsigned entries_total
= 0;
3301 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3302 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3307 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3308 ir_variable
*var
= node
->as_variable();
3309 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3312 if (var
->data
.explicit_location
) {
3314 if (var
->type
->without_array()->is_subroutine())
3315 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3317 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3321 entries_total
+= slots
;
3332 struct empty_uniform_block
*current_block
= NULL
;
3334 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3335 /* We found empty space in UniformRemapTable. */
3336 if (prog
->UniformRemapTable
[i
] == NULL
) {
3337 /* We've found the beginning of a new continous block of empty slots */
3338 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3339 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3340 current_block
->start
= i
;
3341 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3342 ¤t_block
->link
);
3345 /* The current block continues, so we simply increment its slots */
3346 current_block
->slots
++;
3351 return entries_total
;
3355 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3356 GLenum type
, const char *name
)
3358 bool found_interface
= false;
3359 unsigned block_name_len
= 0;
3360 const char *block_name_dot
= strchr(name
, '.');
3362 /* These rules only apply to buffer variables. So we return
3363 * true for the rest of types.
3365 if (type
!= GL_BUFFER_VARIABLE
)
3368 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
3369 const char *block_name
= shProg
->ShaderStorageBlocks
[i
].Name
;
3370 block_name_len
= strlen(block_name
);
3372 const char *block_square_bracket
= strchr(block_name
, '[');
3373 if (block_square_bracket
) {
3374 /* The block is part of an array of named interfaces,
3375 * for the name comparison we ignore the "[x]" part.
3377 block_name_len
-= strlen(block_square_bracket
);
3380 if (block_name_dot
) {
3381 /* Check if the variable name starts with the interface
3382 * name. The interface name (if present) should have the
3383 * length than the interface block name we are comparing to.
3385 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3386 if (len
!= block_name_len
)
3390 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3391 found_interface
= true;
3396 /* We remove the interface name from the buffer variable name,
3397 * including the dot that follows it.
3399 if (found_interface
)
3400 name
= name
+ block_name_len
+ 1;
3402 /* The ARB_program_interface_query spec says:
3404 * "For an active shader storage block member declared as an array, an
3405 * entry will be generated only for the first array element, regardless
3406 * of its type. For arrays of aggregate types, the enumeration rules
3407 * are applied recursively for the single enumerated array element."
3409 const char *struct_first_dot
= strchr(name
, '.');
3410 const char *first_square_bracket
= strchr(name
, '[');
3412 /* The buffer variable is on top level and it is not an array */
3413 if (!first_square_bracket
) {
3415 /* The shader storage block member is a struct, then generate the entry */
3416 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3419 /* Shader storage block member is an array, only generate an entry for the
3420 * first array element.
3422 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3430 add_program_resource(struct gl_shader_program
*prog
,
3431 struct set
*resource_set
,
3432 GLenum type
, const void *data
, uint8_t stages
)
3436 /* If resource already exists, do not add it again. */
3437 if (_mesa_set_search(resource_set
, data
))
3440 prog
->ProgramResourceList
=
3442 prog
->ProgramResourceList
,
3443 gl_program_resource
,
3444 prog
->NumProgramResourceList
+ 1);
3446 if (!prog
->ProgramResourceList
) {
3447 linker_error(prog
, "Out of memory during linking.\n");
3451 struct gl_program_resource
*res
=
3452 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3456 res
->StageReferences
= stages
;
3458 prog
->NumProgramResourceList
++;
3460 _mesa_set_add(resource_set
, data
);
3465 /* Function checks if a variable var is a packed varying and
3466 * if given name is part of packed varying's list.
3468 * If a variable is a packed varying, it has a name like
3469 * 'packed:a,b,c' where a, b and c are separate variables.
3472 included_in_packed_varying(ir_variable
*var
, const char *name
)
3474 if (strncmp(var
->name
, "packed:", 7) != 0)
3477 char *list
= strdup(var
->name
+ 7);
3482 char *token
= strtok_r(list
, ",", &saveptr
);
3484 if (strcmp(token
, name
) == 0) {
3488 token
= strtok_r(NULL
, ",", &saveptr
);
3495 * Function builds a stage reference bitmask from variable name.
3498 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3503 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3504 * used for reference mask in gl_program_resource will need to be changed.
3506 assert(MESA_SHADER_STAGES
< 8);
3508 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3509 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3513 /* Shader symbol table may contain variables that have
3514 * been optimized away. Search IR for the variable instead.
3516 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3517 ir_variable
*var
= node
->as_variable();
3519 unsigned baselen
= strlen(var
->name
);
3521 if (included_in_packed_varying(var
, name
)) {
3526 /* Type needs to match if specified, otherwise we might
3527 * pick a variable with same name but different interface.
3529 if (var
->data
.mode
!= mode
)
3532 if (strncmp(var
->name
, name
, baselen
) == 0) {
3533 /* Check for exact name matches but also check for arrays and
3536 if (name
[baselen
] == '\0' ||
3537 name
[baselen
] == '[' ||
3538 name
[baselen
] == '.') {
3550 * Create gl_shader_variable from ir_variable class.
3552 static gl_shader_variable
*
3553 create_shader_variable(struct gl_shader_program
*shProg
,
3554 const ir_variable
*in
,
3555 const char *name
, const glsl_type
*type
,
3556 bool use_implicit_location
, int location
,
3557 const glsl_type
*outermost_struct_type
)
3559 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3563 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3564 * expect to see gl_VertexID in the program resource list. Pretend.
3566 if (in
->data
.mode
== ir_var_system_value
&&
3567 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3568 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3569 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3570 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3571 (in
->data
.mode
== ir_var_system_value
&&
3572 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3573 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3574 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3575 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3576 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3577 (in
->data
.mode
== ir_var_system_value
&&
3578 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3579 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3580 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3582 out
->name
= ralloc_strdup(shProg
, name
);
3588 /* The ARB_program_interface_query spec says:
3590 * "Not all active variables are assigned valid locations; the
3591 * following variables will have an effective location of -1:
3593 * * uniforms declared as atomic counters;
3595 * * members of a uniform block;
3597 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3599 * * inputs or outputs not declared with a "location" layout
3600 * qualifier, except for vertex shader inputs and fragment shader
3603 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3604 is_gl_identifier(in
->name
) ||
3605 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3608 out
->location
= location
;
3612 out
->outermost_struct_type
= outermost_struct_type
;
3613 out
->interface_type
= in
->get_interface_type();
3614 out
->component
= in
->data
.location_frac
;
3615 out
->index
= in
->data
.index
;
3616 out
->patch
= in
->data
.patch
;
3617 out
->mode
= in
->data
.mode
;
3618 out
->interpolation
= in
->data
.interpolation
;
3619 out
->explicit_location
= in
->data
.explicit_location
;
3620 out
->precision
= in
->data
.precision
;
3626 add_shader_variable(struct gl_shader_program
*shProg
, struct set
*resource_set
,
3627 unsigned stage_mask
,
3628 GLenum programInterface
, ir_variable
*var
,
3629 const char *name
, const glsl_type
*type
,
3630 bool use_implicit_location
, int location
,
3631 const glsl_type
*outermost_struct_type
= NULL
)
3633 const bool is_vertex_input
=
3634 programInterface
== GL_PROGRAM_INPUT
&&
3635 stage_mask
== MESA_SHADER_VERTEX
;
3637 switch (type
->base_type
) {
3638 case GLSL_TYPE_STRUCT
: {
3639 /* The ARB_program_interface_query spec says:
3641 * "For an active variable declared as a structure, a separate entry
3642 * will be generated for each active structure member. The name of
3643 * each entry is formed by concatenating the name of the structure,
3644 * the "." character, and the name of the structure member. If a
3645 * structure member to enumerate is itself a structure or array,
3646 * these enumeration rules are applied recursively."
3648 if (outermost_struct_type
== NULL
)
3649 outermost_struct_type
= type
;
3651 unsigned field_location
= location
;
3652 for (unsigned i
= 0; i
< type
->length
; i
++) {
3653 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3654 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3655 if (!add_shader_variable(shProg
, resource_set
,
3656 stage_mask
, programInterface
,
3657 var
, field_name
, field
->type
,
3658 use_implicit_location
, field_location
,
3659 outermost_struct_type
))
3663 field
->type
->count_attribute_slots(is_vertex_input
);
3669 /* Issue #16 of the ARB_program_interface_query spec says:
3671 * "* If a variable is a member of an interface block without an
3672 * instance name, it is enumerated using just the variable name.
3674 * * If a variable is a member of an interface block with an instance
3675 * name, it is enumerated as "BlockName.Member", where "BlockName" is
3676 * the name of the interface block (not the instance name) and
3677 * "Member" is the name of the variable."
3679 const char *prefixed_name
= (var
->data
.from_named_ifc_block
&&
3680 !is_gl_identifier(var
->name
))
3681 ? ralloc_asprintf(shProg
, "%s.%s", var
->get_interface_type()->name
,
3685 /* The ARB_program_interface_query spec says:
3687 * "For an active variable declared as a single instance of a basic
3688 * type, a single entry will be generated, using the variable name
3689 * from the shader source."
3691 gl_shader_variable
*sha_v
=
3692 create_shader_variable(shProg
, var
, prefixed_name
, type
,
3693 use_implicit_location
, location
,
3694 outermost_struct_type
);
3698 return add_program_resource(shProg
, resource_set
,
3699 programInterface
, sha_v
, stage_mask
);
3705 add_interface_variables(struct gl_shader_program
*shProg
,
3706 struct set
*resource_set
,
3707 unsigned stage
, GLenum programInterface
)
3709 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3711 foreach_in_list(ir_instruction
, node
, ir
) {
3712 ir_variable
*var
= node
->as_variable();
3714 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3719 switch (var
->data
.mode
) {
3720 case ir_var_system_value
:
3721 case ir_var_shader_in
:
3722 if (programInterface
!= GL_PROGRAM_INPUT
)
3724 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3725 : int(VARYING_SLOT_VAR0
);
3727 case ir_var_shader_out
:
3728 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3730 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3731 : int(VARYING_SLOT_VAR0
);
3737 if (var
->data
.patch
)
3738 loc_bias
= int(VARYING_SLOT_PATCH0
);
3740 /* Skip packed varyings, packed varyings are handled separately
3741 * by add_packed_varyings.
3743 if (strncmp(var
->name
, "packed:", 7) == 0)
3746 /* Skip fragdata arrays, these are handled separately
3747 * by add_fragdata_arrays.
3749 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3752 const bool vs_input_or_fs_output
=
3753 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3754 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3756 if (!add_shader_variable(shProg
, resource_set
,
3757 1 << stage
, programInterface
,
3758 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3759 var
->data
.location
- loc_bias
))
3766 add_packed_varyings(struct gl_shader_program
*shProg
, struct set
*resource_set
,
3767 int stage
, GLenum type
)
3769 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3772 if (!sh
|| !sh
->packed_varyings
)
3775 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3776 ir_variable
*var
= node
->as_variable();
3778 switch (var
->data
.mode
) {
3779 case ir_var_shader_in
:
3780 iface
= GL_PROGRAM_INPUT
;
3782 case ir_var_shader_out
:
3783 iface
= GL_PROGRAM_OUTPUT
;
3786 unreachable("unexpected type");
3789 if (type
== iface
) {
3790 const int stage_mask
=
3791 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3792 if (!add_shader_variable(shProg
, resource_set
,
3794 iface
, var
, var
->name
, var
->type
, false,
3795 var
->data
.location
- VARYING_SLOT_VAR0
))
3804 add_fragdata_arrays(struct gl_shader_program
*shProg
, struct set
*resource_set
)
3806 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3808 if (!sh
|| !sh
->fragdata_arrays
)
3811 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3812 ir_variable
*var
= node
->as_variable();
3814 assert(var
->data
.mode
== ir_var_shader_out
);
3816 if (!add_shader_variable(shProg
, resource_set
,
3817 1 << MESA_SHADER_FRAGMENT
,
3818 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3819 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3827 get_top_level_name(const char *name
)
3829 const char *first_dot
= strchr(name
, '.');
3830 const char *first_square_bracket
= strchr(name
, '[');
3833 /* The ARB_program_interface_query spec says:
3835 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3836 * the number of active array elements of the top-level shader storage
3837 * block member containing to the active variable is written to
3838 * <params>. If the top-level block member is not declared as an
3839 * array, the value one is written to <params>. If the top-level block
3840 * member is an array with no declared size, the value zero is written
3844 /* The buffer variable is on top level.*/
3845 if (!first_square_bracket
&& !first_dot
)
3846 name_size
= strlen(name
);
3847 else if ((!first_square_bracket
||
3848 (first_dot
&& first_dot
< first_square_bracket
)))
3849 name_size
= first_dot
- name
;
3851 name_size
= first_square_bracket
- name
;
3853 return strndup(name
, name_size
);
3857 get_var_name(const char *name
)
3859 const char *first_dot
= strchr(name
, '.');
3862 return strdup(name
);
3864 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3868 is_top_level_shader_storage_block_member(const char* name
,
3869 const char* interface_name
,
3870 const char* field_name
)
3872 bool result
= false;
3874 /* If the given variable is already a top-level shader storage
3875 * block member, then return array_size = 1.
3876 * We could have two possibilities: if we have an instanced
3877 * shader storage block or not instanced.
3879 * For the first, we check create a name as it was in top level and
3880 * compare it with the real name. If they are the same, then
3881 * the variable is already at top-level.
3883 * Full instanced name is: interface name + '.' + var name +
3886 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3887 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3888 if (!full_instanced_name
) {
3889 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3893 snprintf(full_instanced_name
, name_length
, "%s.%s",
3894 interface_name
, field_name
);
3896 /* Check if its top-level shader storage block member of an
3897 * instanced interface block, or of a unnamed interface block.
3899 if (strcmp(name
, full_instanced_name
) == 0 ||
3900 strcmp(name
, field_name
) == 0)
3903 free(full_instanced_name
);
3908 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3909 char *interface_name
, char *var_name
)
3911 /* The ARB_program_interface_query spec says:
3913 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3914 * the number of active array elements of the top-level shader storage
3915 * block member containing to the active variable is written to
3916 * <params>. If the top-level block member is not declared as an
3917 * array, the value one is written to <params>. If the top-level block
3918 * member is an array with no declared size, the value zero is written
3921 if (is_top_level_shader_storage_block_member(uni
->name
,
3925 else if (field
->type
->is_unsized_array())
3927 else if (field
->type
->is_array())
3928 return field
->type
->length
;
3934 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3935 const glsl_struct_field
*field
, char *interface_name
,
3938 /* The ARB_program_interface_query spec says:
3940 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3941 * identifying the stride between array elements of the top-level
3942 * shader storage block member containing the active variable is
3943 * written to <params>. For top-level block members declared as
3944 * arrays, the value written is the difference, in basic machine units,
3945 * between the offsets of the active variable for consecutive elements
3946 * in the top-level array. For top-level block members not declared as
3947 * an array, zero is written to <params>."
3949 if (field
->type
->is_array()) {
3950 const enum glsl_matrix_layout matrix_layout
=
3951 glsl_matrix_layout(field
->matrix_layout
);
3952 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3953 const glsl_type
*array_type
= field
->type
->fields
.array
;
3955 if (is_top_level_shader_storage_block_member(uni
->name
,
3960 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3961 if (array_type
->is_record() || array_type
->is_array())
3962 return glsl_align(array_type
->std140_size(row_major
), 16);
3964 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3966 return array_type
->std430_array_stride(row_major
);
3973 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3974 struct gl_uniform_storage
*uni
)
3976 int block_index
= uni
->block_index
;
3977 int array_size
= -1;
3978 int array_stride
= -1;
3979 char *var_name
= get_top_level_name(uni
->name
);
3980 char *interface_name
=
3981 get_top_level_name(uni
->is_shader_storage
?
3982 shProg
->ShaderStorageBlocks
[block_index
].Name
:
3983 shProg
->UniformBlocks
[block_index
].Name
);
3985 if (strcmp(var_name
, interface_name
) == 0) {
3986 /* Deal with instanced array of SSBOs */
3987 char *temp_name
= get_var_name(uni
->name
);
3989 linker_error(shProg
, "Out of memory during linking.\n");
3990 goto write_top_level_array_size_and_stride
;
3993 var_name
= get_top_level_name(temp_name
);
3996 linker_error(shProg
, "Out of memory during linking.\n");
3997 goto write_top_level_array_size_and_stride
;
4001 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4002 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4006 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4007 ir_variable
*var
= node
->as_variable();
4008 if (!var
|| !var
->get_interface_type() ||
4009 var
->data
.mode
!= ir_var_shader_storage
)
4012 const glsl_type
*interface
= var
->get_interface_type();
4014 if (strcmp(interface_name
, interface
->name
) != 0)
4017 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4018 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4019 if (strcmp(field
->name
, var_name
) != 0)
4022 array_stride
= get_array_stride(uni
, interface
, field
,
4023 interface_name
, var_name
);
4024 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4025 goto write_top_level_array_size_and_stride
;
4029 write_top_level_array_size_and_stride
:
4030 free(interface_name
);
4032 uni
->top_level_array_stride
= array_stride
;
4033 uni
->top_level_array_size
= array_size
;
4037 * Builds up a list of program resources that point to existing
4041 build_program_resource_list(struct gl_context
*ctx
,
4042 struct gl_shader_program
*shProg
)
4044 /* Rebuild resource list. */
4045 if (shProg
->ProgramResourceList
) {
4046 ralloc_free(shProg
->ProgramResourceList
);
4047 shProg
->ProgramResourceList
= NULL
;
4048 shProg
->NumProgramResourceList
= 0;
4051 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4053 /* Determine first input and final output stage. These are used to
4054 * detect which variables should be enumerated in the resource list
4055 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4057 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4058 if (!shProg
->_LinkedShaders
[i
])
4060 if (input_stage
== MESA_SHADER_STAGES
)
4065 /* Empty shader, no resources. */
4066 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4069 struct set
*resource_set
= _mesa_set_create(NULL
,
4071 _mesa_key_pointer_equal
);
4073 /* Program interface needs to expose varyings in case of SSO. */
4074 if (shProg
->SeparateShader
) {
4075 if (!add_packed_varyings(shProg
, resource_set
,
4076 input_stage
, GL_PROGRAM_INPUT
))
4079 if (!add_packed_varyings(shProg
, resource_set
,
4080 output_stage
, GL_PROGRAM_OUTPUT
))
4084 if (!add_fragdata_arrays(shProg
, resource_set
))
4087 /* Add inputs and outputs to the resource list. */
4088 if (!add_interface_variables(shProg
, resource_set
,
4089 input_stage
, GL_PROGRAM_INPUT
))
4092 if (!add_interface_variables(shProg
, resource_set
,
4093 output_stage
, GL_PROGRAM_OUTPUT
))
4096 /* Add transform feedback varyings. */
4097 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
4098 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
4099 if (!add_program_resource(shProg
, resource_set
,
4100 GL_TRANSFORM_FEEDBACK_VARYING
,
4101 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
4107 /* Add transform feedback buffers. */
4108 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4109 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
4110 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
4111 if (!add_program_resource(shProg
, resource_set
,
4112 GL_TRANSFORM_FEEDBACK_BUFFER
,
4113 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
4119 /* Add uniforms from uniform storage. */
4120 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4121 /* Do not add uniforms internally used by Mesa. */
4122 if (shProg
->UniformStorage
[i
].hidden
)
4126 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
4129 /* Add stagereferences for uniforms in a uniform block. */
4130 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
4131 int block_index
= shProg
->UniformStorage
[i
].block_index
;
4132 if (block_index
!= -1) {
4133 stageref
|= is_shader_storage
?
4134 shProg
->ShaderStorageBlocks
[block_index
].stageref
:
4135 shProg
->UniformBlocks
[block_index
].stageref
;
4138 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4139 if (!should_add_buffer_variable(shProg
, type
,
4140 shProg
->UniformStorage
[i
].name
))
4143 if (is_shader_storage
) {
4144 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
4147 if (!add_program_resource(shProg
, resource_set
, type
,
4148 &shProg
->UniformStorage
[i
], stageref
))
4152 /* Add program uniform blocks. */
4153 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
4154 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4155 &shProg
->UniformBlocks
[i
], 0))
4159 /* Add program shader storage blocks. */
4160 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
4161 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4162 &shProg
->ShaderStorageBlocks
[i
], 0))
4166 /* Add atomic counter buffers. */
4167 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
4168 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4169 &shProg
->AtomicBuffers
[i
], 0))
4173 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4175 if (!shProg
->UniformStorage
[i
].hidden
)
4178 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4179 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
4180 !shProg
->UniformStorage
[i
].type
->is_subroutine())
4183 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4184 /* add shader subroutines */
4185 if (!add_program_resource(shProg
, resource_set
,
4186 type
, &shProg
->UniformStorage
[i
], 0))
4191 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4192 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4198 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4199 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4200 if (!add_program_resource(shProg
, resource_set
,
4201 type
, &sh
->SubroutineFunctions
[j
], 0))
4206 _mesa_set_destroy(resource_set
, NULL
);
4210 * This check is done to make sure we allow only constant expression
4211 * indexing and "constant-index-expression" (indexing with an expression
4212 * that includes loop induction variable).
4215 validate_sampler_array_indexing(struct gl_context
*ctx
,
4216 struct gl_shader_program
*prog
)
4218 dynamic_sampler_array_indexing_visitor v
;
4219 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4220 if (prog
->_LinkedShaders
[i
] == NULL
)
4223 bool no_dynamic_indexing
=
4224 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4226 /* Search for array derefs in shader. */
4227 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4228 if (v
.uses_dynamic_sampler_array_indexing()) {
4229 const char *msg
= "sampler arrays indexed with non-constant "
4230 "expressions is forbidden in GLSL %s %u";
4231 /* Backend has indicated that it has no dynamic indexing support. */
4232 if (no_dynamic_indexing
) {
4233 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4236 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4244 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4246 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4247 gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
4252 sh
->MaxSubroutineFunctionIndex
= 0;
4253 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4254 ir_function
*fn
= node
->as_function();
4258 if (fn
->is_subroutine
)
4259 sh
->NumSubroutineUniformTypes
++;
4261 if (!fn
->num_subroutine_types
)
4264 /* these should have been calculated earlier. */
4265 assert(fn
->subroutine_index
!= -1);
4266 if (sh
->NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4267 linker_error(prog
, "Too many subroutine functions declared.\n");
4270 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4271 struct gl_subroutine_function
,
4272 sh
->NumSubroutineFunctions
+ 1);
4273 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4274 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4275 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4276 ralloc_array(sh
, const struct glsl_type
*,
4277 fn
->num_subroutine_types
);
4279 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4282 * "Each subroutine with an index qualifier in the shader must be
4283 * given a unique index, otherwise a compile or link error will be
4286 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4287 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4288 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4289 linker_error(prog
, "each subroutine index qualifier in the "
4290 "shader must be unique\n");
4294 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4295 fn
->subroutine_index
;
4297 if (fn
->subroutine_index
> (int)sh
->MaxSubroutineFunctionIndex
)
4298 sh
->MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4300 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4301 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4302 sh
->NumSubroutineFunctions
++;
4308 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4310 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4312 foreach_in_list(ir_instruction
, node
, ir
) {
4313 ir_variable
*const var
= node
->as_variable();
4315 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4318 /* Don't set always active on builtins that haven't been redeclared */
4319 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4322 var
->data
.always_active_io
= true;
4327 * When separate shader programs are enabled, only input/outputs between
4328 * the stages of a multi-stage separate program can be safely removed
4329 * from the shader interface. Other inputs/outputs must remain active.
4332 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4334 unsigned first
, last
;
4335 assert(prog
->SeparateShader
);
4337 first
= MESA_SHADER_STAGES
;
4340 /* Determine first and last stage. Excluding the compute stage */
4341 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4342 if (!prog
->_LinkedShaders
[i
])
4344 if (first
== MESA_SHADER_STAGES
)
4349 if (first
== MESA_SHADER_STAGES
)
4352 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4353 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4357 if (first
== last
) {
4358 /* For a single shader program only allow inputs to the vertex shader
4359 * and outputs from the fragment shader to be removed.
4361 if (stage
!= MESA_SHADER_VERTEX
)
4362 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4363 if (stage
!= MESA_SHADER_FRAGMENT
)
4364 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4366 /* For multi-stage separate shader programs only allow inputs and
4367 * outputs between the shader stages to be removed as well as inputs
4368 * to the vertex shader and outputs from the fragment shader.
4370 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4371 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4372 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4373 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4379 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4380 unsigned num_explicit_uniform_locs
,
4381 struct gl_context
*ctx
,
4382 struct gl_shader_program
*prog
, void *mem_ctx
)
4384 bool has_xfb_qualifiers
= false;
4385 unsigned num_tfeedback_decls
= 0;
4386 char **varying_names
= NULL
;
4387 tfeedback_decl
*tfeedback_decls
= NULL
;
4389 /* Mark all generic shader inputs and outputs as unpaired. */
4390 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4391 if (prog
->_LinkedShaders
[i
] != NULL
) {
4392 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4396 unsigned prev
= first
;
4397 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4398 if (prog
->_LinkedShaders
[i
] == NULL
)
4401 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4402 prog
->_LinkedShaders
[i
]);
4406 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4407 MESA_SHADER_VERTEX
)) {
4411 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4412 MESA_SHADER_FRAGMENT
)) {
4416 /* From the ARB_enhanced_layouts spec:
4418 * "If the shader used to record output variables for transform feedback
4419 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4420 * qualifiers, the values specified by TransformFeedbackVaryings are
4421 * ignored, and the set of variables captured for transform feedback is
4422 * instead derived from the specified layout qualifiers."
4424 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4425 /* Find last stage before fragment shader */
4426 if (prog
->_LinkedShaders
[i
]) {
4427 has_xfb_qualifiers
=
4428 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4429 &num_tfeedback_decls
,
4435 if (!has_xfb_qualifiers
) {
4436 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4437 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4440 if (num_tfeedback_decls
!= 0) {
4441 /* From GL_EXT_transform_feedback:
4442 * A program will fail to link if:
4444 * * the <count> specified by TransformFeedbackVaryingsEXT is
4445 * non-zero, but the program object has no vertex or geometry
4448 if (first
>= MESA_SHADER_FRAGMENT
) {
4449 linker_error(prog
, "Transform feedback varyings specified, but "
4450 "no vertex, tessellation, or geometry shader is "
4455 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4456 num_tfeedback_decls
);
4457 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4458 varying_names
, tfeedback_decls
))
4462 /* If there is no fragment shader we need to set transform feedback.
4464 * For SSO we also need to assign output locations. We assign them here
4465 * because we need to do it for both single stage programs and multi stage
4468 if (last
< MESA_SHADER_FRAGMENT
&&
4469 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4470 const uint64_t reserved_out_slots
=
4471 reserved_varying_slot(prog
->_LinkedShaders
[last
], ir_var_shader_out
);
4472 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4473 prog
->_LinkedShaders
[last
], NULL
,
4474 num_tfeedback_decls
, tfeedback_decls
,
4475 reserved_out_slots
))
4479 if (last
<= MESA_SHADER_FRAGMENT
) {
4480 /* Remove unused varyings from the first/last stage unless SSO */
4481 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4482 prog
->_LinkedShaders
[first
],
4484 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4485 prog
->_LinkedShaders
[last
],
4488 /* If the program is made up of only a single stage */
4489 if (first
== last
) {
4490 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[last
];
4492 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4493 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4496 if (prog
->SeparateShader
) {
4497 const uint64_t reserved_slots
=
4498 reserved_varying_slot(sh
, ir_var_shader_in
);
4500 /* Assign input locations for SSO, output locations are already
4503 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4504 NULL
/* producer */,
4506 0 /* num_tfeedback_decls */,
4507 NULL
/* tfeedback_decls */,
4512 /* Linking the stages in the opposite order (from fragment to vertex)
4513 * ensures that inter-shader outputs written to in an earlier stage
4514 * are eliminated if they are (transitively) not used in a later
4518 for (int i
= next
- 1; i
>= 0; i
--) {
4519 if (prog
->_LinkedShaders
[i
] == NULL
&& i
!= 0)
4522 gl_linked_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4523 gl_linked_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4525 const uint64_t reserved_out_slots
=
4526 reserved_varying_slot(sh_i
, ir_var_shader_out
);
4527 const uint64_t reserved_in_slots
=
4528 reserved_varying_slot(sh_next
, ir_var_shader_in
);
4530 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4531 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4534 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4535 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4537 reserved_out_slots
| reserved_in_slots
))
4540 /* This must be done after all dead varyings are eliminated. */
4542 unsigned slots_used
= _mesa_bitcount_64(reserved_out_slots
);
4543 if (!check_against_output_limit(ctx
, prog
, sh_i
, slots_used
)) {
4548 unsigned slots_used
= _mesa_bitcount_64(reserved_in_slots
);
4549 if (!check_against_input_limit(ctx
, prog
, sh_next
, slots_used
))
4557 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4558 has_xfb_qualifiers
))
4561 update_array_sizes(prog
);
4562 link_assign_uniform_locations(prog
, ctx
, num_explicit_uniform_locs
);
4563 link_assign_atomic_counter_resources(ctx
, prog
);
4565 link_calculate_subroutine_compat(prog
);
4566 check_resources(ctx
, prog
);
4567 check_subroutine_resources(prog
);
4568 check_image_resources(ctx
, prog
);
4569 link_check_atomic_counter_resources(ctx
, prog
);
4571 if (!prog
->LinkStatus
)
4574 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4575 if (prog
->_LinkedShaders
[i
] == NULL
)
4578 const struct gl_shader_compiler_options
*options
=
4579 &ctx
->Const
.ShaderCompilerOptions
[i
];
4581 if (options
->LowerBufferInterfaceBlocks
)
4582 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4583 options
->ClampBlockIndicesToArrayBounds
);
4585 if (options
->LowerShaderSharedVariables
)
4586 lower_shared_reference(prog
->_LinkedShaders
[i
],
4587 &prog
->Comp
.SharedSize
);
4589 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4590 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4597 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4599 prog
->LinkStatus
= true; /* All error paths will set this to false */
4600 prog
->Validated
= false;
4601 prog
->_Used
= false;
4603 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4605 * "Linking can fail for a variety of reasons as specified in the
4606 * OpenGL Shading Language Specification, as well as any of the
4607 * following reasons:
4609 * - No shader objects are attached to program."
4611 * The Compatibility Profile specification does not list the error. In
4612 * Compatibility Profile missing shader stages are replaced by
4613 * fixed-function. This applies to the case where all stages are
4616 if (prog
->NumShaders
== 0) {
4617 if (ctx
->API
!= API_OPENGL_COMPAT
)
4618 linker_error(prog
, "no shaders attached to the program\n");
4622 unsigned int num_explicit_uniform_locs
= 0;
4624 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4626 prog
->ARB_fragment_coord_conventions_enable
= false;
4628 /* Separate the shaders into groups based on their type.
4630 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4631 unsigned num_shaders
[MESA_SHADER_STAGES
];
4633 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4634 shader_list
[i
] = (struct gl_shader
**)
4635 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4639 unsigned min_version
= UINT_MAX
;
4640 unsigned max_version
= 0;
4641 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4642 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4643 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4645 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4646 linker_error(prog
, "all shaders must use same shading "
4647 "language version\n");
4651 if (prog
->Shaders
[i
]->info
.ARB_fragment_coord_conventions_enable
) {
4652 prog
->ARB_fragment_coord_conventions_enable
= true;
4655 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4656 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4657 num_shaders
[shader_type
]++;
4660 /* In desktop GLSL, different shader versions may be linked together. In
4661 * GLSL ES, all shader versions must be the same.
4663 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4664 linker_error(prog
, "all shaders must use same shading "
4665 "language version\n");
4669 prog
->Version
= max_version
;
4670 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4672 /* Some shaders have to be linked with some other shaders present.
4674 if (!prog
->SeparateShader
) {
4675 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4676 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4677 linker_error(prog
, "Geometry shader must be linked with "
4681 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4682 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4683 linker_error(prog
, "Tessellation evaluation shader must be linked "
4684 "with vertex shader\n");
4687 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4688 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4689 linker_error(prog
, "Tessellation control shader must be linked with "
4694 /* The spec is self-contradictory here. It allows linking without a tess
4695 * eval shader, but that can only be used with transform feedback and
4696 * rasterization disabled. However, transform feedback isn't allowed
4697 * with GL_PATCHES, so it can't be used.
4699 * More investigation showed that the idea of transform feedback after
4700 * a tess control shader was dropped, because some hw vendors couldn't
4701 * support tessellation without a tess eval shader, but the linker
4702 * section wasn't updated to reflect that.
4704 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4707 * Do what's reasonable and always require a tess eval shader if a tess
4708 * control shader is present.
4710 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4711 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4712 linker_error(prog
, "Tessellation control shader must be linked with "
4713 "tessellation evaluation shader\n");
4718 /* Compute shaders have additional restrictions. */
4719 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4720 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4721 linker_error(prog
, "Compute shaders may not be linked with any other "
4722 "type of shader\n");
4725 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4726 if (prog
->_LinkedShaders
[i
] != NULL
) {
4727 _mesa_delete_linked_shader(ctx
, prog
->_LinkedShaders
[i
]);
4730 prog
->_LinkedShaders
[i
] = NULL
;
4733 /* Link all shaders for a particular stage and validate the result.
4735 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4736 if (num_shaders
[stage
] > 0) {
4737 gl_linked_shader
*const sh
=
4738 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4739 num_shaders
[stage
]);
4741 if (!prog
->LinkStatus
) {
4743 _mesa_delete_linked_shader(ctx
, sh
);
4748 case MESA_SHADER_VERTEX
:
4749 validate_vertex_shader_executable(prog
, sh
, ctx
);
4751 case MESA_SHADER_TESS_CTRL
:
4752 /* nothing to be done */
4754 case MESA_SHADER_TESS_EVAL
:
4755 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4757 case MESA_SHADER_GEOMETRY
:
4758 validate_geometry_shader_executable(prog
, sh
, ctx
);
4760 case MESA_SHADER_FRAGMENT
:
4761 validate_fragment_shader_executable(prog
, sh
);
4764 if (!prog
->LinkStatus
) {
4766 _mesa_delete_linked_shader(ctx
, sh
);
4770 prog
->_LinkedShaders
[stage
] = sh
;
4774 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4775 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4776 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4777 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4778 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4779 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4780 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4781 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4782 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4784 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4785 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4788 /* Here begins the inter-stage linking phase. Some initial validation is
4789 * performed, then locations are assigned for uniforms, attributes, and
4792 cross_validate_uniforms(prog
);
4793 if (!prog
->LinkStatus
)
4796 unsigned first
, last
, prev
;
4798 first
= MESA_SHADER_STAGES
;
4801 /* Determine first and last stage. */
4802 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4803 if (!prog
->_LinkedShaders
[i
])
4805 if (first
== MESA_SHADER_STAGES
)
4810 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4811 link_assign_subroutine_types(prog
);
4813 if (!prog
->LinkStatus
)
4816 resize_tes_inputs(ctx
, prog
);
4818 /* Validate the inputs of each stage with the output of the preceding
4822 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4823 if (prog
->_LinkedShaders
[i
] == NULL
)
4826 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4827 prog
->_LinkedShaders
[i
]);
4828 if (!prog
->LinkStatus
)
4831 cross_validate_outputs_to_inputs(prog
,
4832 prog
->_LinkedShaders
[prev
],
4833 prog
->_LinkedShaders
[i
]);
4834 if (!prog
->LinkStatus
)
4840 /* Cross-validate uniform blocks between shader stages */
4841 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4842 if (!prog
->LinkStatus
)
4845 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4846 if (prog
->_LinkedShaders
[i
] != NULL
)
4847 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4850 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4851 * it before optimization because we want most of the checks to get
4852 * dropped thanks to constant propagation.
4854 * This rule also applies to GLSL ES 3.00.
4856 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4857 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4859 lower_discard_flow(sh
->ir
);
4863 if (prog
->SeparateShader
)
4864 disable_varying_optimizations_for_sso(prog
);
4867 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4871 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4874 /* Do common optimization before assigning storage for attributes,
4875 * uniforms, and varyings. Later optimization could possibly make
4876 * some of that unused.
4878 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4879 if (prog
->_LinkedShaders
[i
] == NULL
)
4882 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4883 if (!prog
->LinkStatus
)
4886 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4887 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4890 if (ctx
->Const
.LowerTessLevel
) {
4891 lower_tess_level(prog
->_LinkedShaders
[i
]);
4894 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4895 &ctx
->Const
.ShaderCompilerOptions
[i
],
4896 ctx
->Const
.NativeIntegers
))
4899 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
4900 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4903 /* Validation for special cases where we allow sampler array indexing
4904 * with loop induction variable. This check emits a warning or error
4905 * depending if backend can handle dynamic indexing.
4907 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4908 (prog
->IsES
&& prog
->Version
< 300)) {
4909 if (!validate_sampler_array_indexing(ctx
, prog
))
4913 /* Check and validate stream emissions in geometry shaders */
4914 validate_geometry_shader_emissions(ctx
, prog
);
4916 store_fragdepth_layout(prog
);
4918 if(!link_varyings_and_uniforms(first
, last
, num_explicit_uniform_locs
, ctx
,
4922 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4923 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4924 * anything about shader linking when one of the shaders (vertex or
4925 * fragment shader) is absent. So, the extension shouldn't change the
4926 * behavior specified in GLSL specification.
4928 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4929 * "Linking can fail for a variety of reasons as specified in the
4930 * OpenGL ES Shading Language Specification, as well as any of the
4931 * following reasons:
4935 * * program contains objects to form either a vertex shader or
4936 * fragment shader, and program is not separable, and does not
4937 * contain objects to form both a vertex shader and fragment
4940 * However, the only scenario in 3.1+ where we don't require them both is
4941 * when we have a compute shader. For example:
4943 * - No shaders is a link error.
4944 * - Geom or Tess without a Vertex shader is a link error which means we
4945 * always require a Vertex shader and hence a Fragment shader.
4946 * - Finally a Compute shader linked with any other stage is a link error.
4948 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4949 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4950 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4951 linker_error(prog
, "program lacks a vertex shader\n");
4952 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4953 linker_error(prog
, "program lacks a fragment shader\n");
4958 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4959 free(shader_list
[i
]);
4960 if (prog
->_LinkedShaders
[i
] == NULL
)
4963 /* Do a final validation step to make sure that the IR wasn't
4964 * invalidated by any modifications performed after intrastage linking.
4966 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4968 /* Retain any live IR, but trash the rest. */
4969 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4971 /* The symbol table in the linked shaders may contain references to
4972 * variables that were removed (e.g., unused uniforms). Since it may
4973 * contain junk, there is no possible valid use. Delete it and set the
4976 delete prog
->_LinkedShaders
[i
]->symbols
;
4977 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4980 ralloc_free(mem_ctx
);