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
,
1241 this->target
= target
;
1242 this->symbols
= target
->symbols
;
1243 this->instructions
= target
->ir
;
1244 this->temps
= temps
;
1247 virtual ir_visitor_status
visit(ir_dereference_variable
*ir
)
1249 if (ir
->var
->data
.mode
== ir_var_temporary
) {
1250 hash_entry
*entry
= _mesa_hash_table_search(temps
, ir
->var
);
1251 ir_variable
*var
= entry
? (ir_variable
*) entry
->data
: NULL
;
1253 assert(var
!= NULL
);
1255 return visit_continue
;
1258 ir_variable
*const existing
=
1259 this->symbols
->get_variable(ir
->var
->name
);
1260 if (existing
!= NULL
)
1263 ir_variable
*copy
= ir
->var
->clone(this->target
, NULL
);
1265 this->symbols
->add_variable(copy
);
1266 this->instructions
->push_head(copy
);
1270 return visit_continue
;
1274 struct gl_linked_shader
*target
;
1275 glsl_symbol_table
*symbols
;
1276 exec_list
*instructions
;
1280 remap_visitor
v(target
, temps
);
1287 * Move non-declarations from one instruction stream to another
1289 * The intended usage pattern of this function is to pass the pointer to the
1290 * head sentinel of a list (i.e., a pointer to the list cast to an \c exec_node
1291 * pointer) for \c last and \c false for \c make_copies on the first
1292 * call. Successive calls pass the return value of the previous call for
1293 * \c last and \c true for \c make_copies.
1295 * \param instructions Source instruction stream
1296 * \param last Instruction after which new instructions should be
1297 * inserted in the target instruction stream
1298 * \param make_copies Flag selecting whether instructions in \c instructions
1299 * should be copied (via \c ir_instruction::clone) into the
1300 * target list or moved.
1303 * The new "last" instruction in the target instruction stream. This pointer
1304 * is suitable for use as the \c last parameter of a later call to this
1308 move_non_declarations(exec_list
*instructions
, exec_node
*last
,
1309 bool make_copies
, gl_linked_shader
*target
)
1311 hash_table
*temps
= NULL
;
1314 temps
= _mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1315 _mesa_key_pointer_equal
);
1317 foreach_in_list_safe(ir_instruction
, inst
, instructions
) {
1318 if (inst
->as_function())
1321 ir_variable
*var
= inst
->as_variable();
1322 if ((var
!= NULL
) && (var
->data
.mode
!= ir_var_temporary
))
1325 assert(inst
->as_assignment()
1327 || inst
->as_if() /* for initializers with the ?: operator */
1328 || ((var
!= NULL
) && (var
->data
.mode
== ir_var_temporary
)));
1331 inst
= inst
->clone(target
, NULL
);
1334 _mesa_hash_table_insert(temps
, var
, inst
);
1336 remap_variables(inst
, target
, temps
);
1341 last
->insert_after(inst
);
1346 _mesa_hash_table_destroy(temps
, NULL
);
1353 * This class is only used in link_intrastage_shaders() below but declaring
1354 * it inside that function leads to compiler warnings with some versions of
1357 class array_sizing_visitor
: public ir_hierarchical_visitor
{
1359 array_sizing_visitor()
1360 : mem_ctx(ralloc_context(NULL
)),
1361 unnamed_interfaces(_mesa_hash_table_create(NULL
, _mesa_hash_pointer
,
1362 _mesa_key_pointer_equal
))
1366 ~array_sizing_visitor()
1368 _mesa_hash_table_destroy(this->unnamed_interfaces
, NULL
);
1369 ralloc_free(this->mem_ctx
);
1372 virtual ir_visitor_status
visit(ir_variable
*var
)
1374 const glsl_type
*type_without_array
;
1375 bool implicit_sized_array
= var
->data
.implicit_sized_array
;
1376 fixup_type(&var
->type
, var
->data
.max_array_access
,
1377 var
->data
.from_ssbo_unsized_array
,
1378 &implicit_sized_array
);
1379 var
->data
.implicit_sized_array
= implicit_sized_array
;
1380 type_without_array
= var
->type
->without_array();
1381 if (var
->type
->is_interface()) {
1382 if (interface_contains_unsized_arrays(var
->type
)) {
1383 const glsl_type
*new_type
=
1384 resize_interface_members(var
->type
,
1385 var
->get_max_ifc_array_access(),
1386 var
->is_in_shader_storage_block());
1387 var
->type
= new_type
;
1388 var
->change_interface_type(new_type
);
1390 } else if (type_without_array
->is_interface()) {
1391 if (interface_contains_unsized_arrays(type_without_array
)) {
1392 const glsl_type
*new_type
=
1393 resize_interface_members(type_without_array
,
1394 var
->get_max_ifc_array_access(),
1395 var
->is_in_shader_storage_block());
1396 var
->change_interface_type(new_type
);
1397 var
->type
= update_interface_members_array(var
->type
, new_type
);
1399 } else if (const glsl_type
*ifc_type
= var
->get_interface_type()) {
1400 /* Store a pointer to the variable in the unnamed_interfaces
1404 _mesa_hash_table_search(this->unnamed_interfaces
,
1407 ir_variable
**interface_vars
= entry
? (ir_variable
**) entry
->data
: NULL
;
1409 if (interface_vars
== NULL
) {
1410 interface_vars
= rzalloc_array(mem_ctx
, ir_variable
*,
1412 _mesa_hash_table_insert(this->unnamed_interfaces
, ifc_type
,
1415 unsigned index
= ifc_type
->field_index(var
->name
);
1416 assert(index
< ifc_type
->length
);
1417 assert(interface_vars
[index
] == NULL
);
1418 interface_vars
[index
] = var
;
1420 return visit_continue
;
1424 * For each unnamed interface block that was discovered while running the
1425 * visitor, adjust the interface type to reflect the newly assigned array
1426 * sizes, and fix up the ir_variable nodes to point to the new interface
1429 void fixup_unnamed_interface_types()
1431 hash_table_call_foreach(this->unnamed_interfaces
,
1432 fixup_unnamed_interface_type
, NULL
);
1437 * If the type pointed to by \c type represents an unsized array, replace
1438 * it with a sized array whose size is determined by max_array_access.
1440 static void fixup_type(const glsl_type
**type
, unsigned max_array_access
,
1441 bool from_ssbo_unsized_array
, bool *implicit_sized
)
1443 if (!from_ssbo_unsized_array
&& (*type
)->is_unsized_array()) {
1444 *type
= glsl_type::get_array_instance((*type
)->fields
.array
,
1445 max_array_access
+ 1);
1446 *implicit_sized
= true;
1447 assert(*type
!= NULL
);
1451 static const glsl_type
*
1452 update_interface_members_array(const glsl_type
*type
,
1453 const glsl_type
*new_interface_type
)
1455 const glsl_type
*element_type
= type
->fields
.array
;
1456 if (element_type
->is_array()) {
1457 const glsl_type
*new_array_type
=
1458 update_interface_members_array(element_type
, new_interface_type
);
1459 return glsl_type::get_array_instance(new_array_type
, type
->length
);
1461 return glsl_type::get_array_instance(new_interface_type
,
1467 * Determine whether the given interface type contains unsized arrays (if
1468 * it doesn't, array_sizing_visitor doesn't need to process it).
1470 static bool interface_contains_unsized_arrays(const glsl_type
*type
)
1472 for (unsigned i
= 0; i
< type
->length
; i
++) {
1473 const glsl_type
*elem_type
= type
->fields
.structure
[i
].type
;
1474 if (elem_type
->is_unsized_array())
1481 * Create a new interface type based on the given type, with unsized arrays
1482 * replaced by sized arrays whose size is determined by
1483 * max_ifc_array_access.
1485 static const glsl_type
*
1486 resize_interface_members(const glsl_type
*type
,
1487 const int *max_ifc_array_access
,
1490 unsigned num_fields
= type
->length
;
1491 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1492 memcpy(fields
, type
->fields
.structure
,
1493 num_fields
* sizeof(*fields
));
1494 for (unsigned i
= 0; i
< num_fields
; i
++) {
1495 bool implicit_sized_array
= fields
[i
].implicit_sized_array
;
1496 /* If SSBO last member is unsized array, we don't replace it by a sized
1499 if (is_ssbo
&& i
== (num_fields
- 1))
1500 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1501 true, &implicit_sized_array
);
1503 fixup_type(&fields
[i
].type
, max_ifc_array_access
[i
],
1504 false, &implicit_sized_array
);
1505 fields
[i
].implicit_sized_array
= implicit_sized_array
;
1507 glsl_interface_packing packing
=
1508 (glsl_interface_packing
) type
->interface_packing
;
1509 const glsl_type
*new_ifc_type
=
1510 glsl_type::get_interface_instance(fields
, num_fields
,
1511 packing
, type
->name
);
1513 return new_ifc_type
;
1516 static void fixup_unnamed_interface_type(const void *key
, void *data
,
1519 const glsl_type
*ifc_type
= (const glsl_type
*) key
;
1520 ir_variable
**interface_vars
= (ir_variable
**) data
;
1521 unsigned num_fields
= ifc_type
->length
;
1522 glsl_struct_field
*fields
= new glsl_struct_field
[num_fields
];
1523 memcpy(fields
, ifc_type
->fields
.structure
,
1524 num_fields
* sizeof(*fields
));
1525 bool interface_type_changed
= false;
1526 for (unsigned i
= 0; i
< num_fields
; i
++) {
1527 if (interface_vars
[i
] != NULL
&&
1528 fields
[i
].type
!= interface_vars
[i
]->type
) {
1529 fields
[i
].type
= interface_vars
[i
]->type
;
1530 interface_type_changed
= true;
1533 if (!interface_type_changed
) {
1537 glsl_interface_packing packing
=
1538 (glsl_interface_packing
) ifc_type
->interface_packing
;
1539 const glsl_type
*new_ifc_type
=
1540 glsl_type::get_interface_instance(fields
, num_fields
, packing
,
1543 for (unsigned i
= 0; i
< num_fields
; i
++) {
1544 if (interface_vars
[i
] != NULL
)
1545 interface_vars
[i
]->change_interface_type(new_ifc_type
);
1550 * Memory context used to allocate the data in \c unnamed_interfaces.
1555 * Hash table from const glsl_type * to an array of ir_variable *'s
1556 * pointing to the ir_variables constituting each unnamed interface block.
1558 hash_table
*unnamed_interfaces
;
1562 * Check for conflicting xfb_stride default qualifiers and store buffer stride
1566 link_xfb_stride_layout_qualifiers(struct gl_context
*ctx
,
1567 struct gl_shader_program
*prog
,
1568 struct gl_linked_shader
*linked_shader
,
1569 struct gl_shader
**shader_list
,
1570 unsigned num_shaders
)
1572 for (unsigned i
= 0; i
< MAX_FEEDBACK_BUFFERS
; i
++) {
1573 linked_shader
->info
.TransformFeedback
.BufferStride
[i
] = 0;
1576 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1577 struct gl_shader
*shader
= shader_list
[i
];
1579 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1580 if (shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1581 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1582 shader
->info
.TransformFeedback
.BufferStride
[j
] != 0 &&
1583 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] !=
1584 shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1586 "intrastage shaders defined with conflicting "
1587 "xfb_stride for buffer %d (%d and %d)\n", j
,
1589 info
.TransformFeedback
.BufferStride
[j
],
1590 shader
->info
.TransformFeedback
.BufferStride
[j
]);
1594 if (shader
->info
.TransformFeedback
.BufferStride
[j
])
1595 linked_shader
->info
.TransformFeedback
.BufferStride
[j
] =
1596 shader
->info
.TransformFeedback
.BufferStride
[j
];
1601 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1602 if (linked_shader
->info
.TransformFeedback
.BufferStride
[j
]) {
1603 prog
->TransformFeedback
.BufferStride
[j
] =
1604 linked_shader
->info
.TransformFeedback
.BufferStride
[j
];
1606 /* We will validate doubles at a later stage */
1607 if (prog
->TransformFeedback
.BufferStride
[j
] % 4) {
1608 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
1609 "multiple of 4 or if its applied to a type that is "
1610 "or contains a double a multiple of 8.",
1611 prog
->TransformFeedback
.BufferStride
[j
]);
1615 if (prog
->TransformFeedback
.BufferStride
[j
] / 4 >
1616 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
1618 "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
1619 "limit has been exceeded.");
1627 * Performs the cross-validation of tessellation control shader vertices and
1628 * layout qualifiers for the attached tessellation control shaders,
1629 * and propagates them to the linked TCS and linked shader program.
1632 link_tcs_out_layout_qualifiers(struct gl_shader_program
*prog
,
1633 struct gl_linked_shader
*linked_shader
,
1634 struct gl_shader
**shader_list
,
1635 unsigned num_shaders
)
1637 linked_shader
->info
.TessCtrl
.VerticesOut
= 0;
1639 if (linked_shader
->Stage
!= MESA_SHADER_TESS_CTRL
)
1642 /* From the GLSL 4.0 spec (chapter 4.3.8.2):
1644 * "All tessellation control shader layout declarations in a program
1645 * must specify the same output patch vertex count. There must be at
1646 * least one layout qualifier specifying an output patch vertex count
1647 * in any program containing tessellation control shaders; however,
1648 * such a declaration is not required in all tessellation control
1652 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1653 struct gl_shader
*shader
= shader_list
[i
];
1655 if (shader
->info
.TessCtrl
.VerticesOut
!= 0) {
1656 if (linked_shader
->info
.TessCtrl
.VerticesOut
!= 0 &&
1657 linked_shader
->info
.TessCtrl
.VerticesOut
!=
1658 shader
->info
.TessCtrl
.VerticesOut
) {
1659 linker_error(prog
, "tessellation control shader defined with "
1660 "conflicting output vertex count (%d and %d)\n",
1661 linked_shader
->info
.TessCtrl
.VerticesOut
,
1662 shader
->info
.TessCtrl
.VerticesOut
);
1665 linked_shader
->info
.TessCtrl
.VerticesOut
=
1666 shader
->info
.TessCtrl
.VerticesOut
;
1670 /* Just do the intrastage -> interstage propagation right now,
1671 * since we already know we're in the right type of shader program
1674 if (linked_shader
->info
.TessCtrl
.VerticesOut
== 0) {
1675 linker_error(prog
, "tessellation control shader didn't declare "
1676 "vertices out layout qualifier\n");
1683 * Performs the cross-validation of tessellation evaluation shader
1684 * primitive type, vertex spacing, ordering and point_mode layout qualifiers
1685 * for the attached tessellation evaluation shaders, and propagates them
1686 * to the linked TES and linked shader program.
1689 link_tes_in_layout_qualifiers(struct gl_shader_program
*prog
,
1690 struct gl_linked_shader
*linked_shader
,
1691 struct gl_shader
**shader_list
,
1692 unsigned num_shaders
)
1694 linked_shader
->info
.TessEval
.PrimitiveMode
= PRIM_UNKNOWN
;
1695 linked_shader
->info
.TessEval
.Spacing
= 0;
1696 linked_shader
->info
.TessEval
.VertexOrder
= 0;
1697 linked_shader
->info
.TessEval
.PointMode
= -1;
1699 if (linked_shader
->Stage
!= MESA_SHADER_TESS_EVAL
)
1702 /* From the GLSL 4.0 spec (chapter 4.3.8.1):
1704 * "At least one tessellation evaluation shader (compilation unit) in
1705 * a program must declare a primitive mode in its input layout.
1706 * Declaration vertex spacing, ordering, and point mode identifiers is
1707 * optional. It is not required that all tessellation evaluation
1708 * shaders in a program declare a primitive mode. If spacing or
1709 * vertex ordering declarations are omitted, the tessellation
1710 * primitive generator will use equal spacing or counter-clockwise
1711 * vertex ordering, respectively. If a point mode declaration is
1712 * omitted, the tessellation primitive generator will produce lines or
1713 * triangles according to the primitive mode."
1716 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1717 struct gl_shader
*shader
= shader_list
[i
];
1719 if (shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
) {
1720 if (linked_shader
->info
.TessEval
.PrimitiveMode
!= PRIM_UNKNOWN
&&
1721 linked_shader
->info
.TessEval
.PrimitiveMode
!=
1722 shader
->info
.TessEval
.PrimitiveMode
) {
1723 linker_error(prog
, "tessellation evaluation shader defined with "
1724 "conflicting input primitive modes.\n");
1727 linked_shader
->info
.TessEval
.PrimitiveMode
= shader
->info
.TessEval
.PrimitiveMode
;
1730 if (shader
->info
.TessEval
.Spacing
!= 0) {
1731 if (linked_shader
->info
.TessEval
.Spacing
!= 0 &&
1732 linked_shader
->info
.TessEval
.Spacing
!=
1733 shader
->info
.TessEval
.Spacing
) {
1734 linker_error(prog
, "tessellation evaluation shader defined with "
1735 "conflicting vertex spacing.\n");
1738 linked_shader
->info
.TessEval
.Spacing
= shader
->info
.TessEval
.Spacing
;
1741 if (shader
->info
.TessEval
.VertexOrder
!= 0) {
1742 if (linked_shader
->info
.TessEval
.VertexOrder
!= 0 &&
1743 linked_shader
->info
.TessEval
.VertexOrder
!=
1744 shader
->info
.TessEval
.VertexOrder
) {
1745 linker_error(prog
, "tessellation evaluation shader defined with "
1746 "conflicting ordering.\n");
1749 linked_shader
->info
.TessEval
.VertexOrder
=
1750 shader
->info
.TessEval
.VertexOrder
;
1753 if (shader
->info
.TessEval
.PointMode
!= -1) {
1754 if (linked_shader
->info
.TessEval
.PointMode
!= -1 &&
1755 linked_shader
->info
.TessEval
.PointMode
!=
1756 shader
->info
.TessEval
.PointMode
) {
1757 linker_error(prog
, "tessellation evaluation shader defined with "
1758 "conflicting point modes.\n");
1761 linked_shader
->info
.TessEval
.PointMode
=
1762 shader
->info
.TessEval
.PointMode
;
1767 /* Just do the intrastage -> interstage propagation right now,
1768 * since we already know we're in the right type of shader program
1771 if (linked_shader
->info
.TessEval
.PrimitiveMode
== PRIM_UNKNOWN
) {
1773 "tessellation evaluation shader didn't declare input "
1774 "primitive modes.\n");
1778 if (linked_shader
->info
.TessEval
.Spacing
== 0)
1779 linked_shader
->info
.TessEval
.Spacing
= GL_EQUAL
;
1781 if (linked_shader
->info
.TessEval
.VertexOrder
== 0)
1782 linked_shader
->info
.TessEval
.VertexOrder
= GL_CCW
;
1784 if (linked_shader
->info
.TessEval
.PointMode
== -1)
1785 linked_shader
->info
.TessEval
.PointMode
= GL_FALSE
;
1790 * Performs the cross-validation of layout qualifiers specified in
1791 * redeclaration of gl_FragCoord for the attached fragment shaders,
1792 * and propagates them to the linked FS and linked shader program.
1795 link_fs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1796 struct gl_linked_shader
*linked_shader
,
1797 struct gl_shader
**shader_list
,
1798 unsigned num_shaders
)
1800 linked_shader
->info
.redeclares_gl_fragcoord
= false;
1801 linked_shader
->info
.uses_gl_fragcoord
= false;
1802 linked_shader
->info
.origin_upper_left
= false;
1803 linked_shader
->info
.pixel_center_integer
= false;
1804 linked_shader
->info
.BlendSupport
= 0;
1806 if (linked_shader
->Stage
!= MESA_SHADER_FRAGMENT
||
1807 (prog
->Version
< 150 && !prog
->ARB_fragment_coord_conventions_enable
))
1810 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1811 struct gl_shader
*shader
= shader_list
[i
];
1812 /* From the GLSL 1.50 spec, page 39:
1814 * "If gl_FragCoord is redeclared in any fragment shader in a program,
1815 * it must be redeclared in all the fragment shaders in that program
1816 * that have a static use gl_FragCoord."
1818 if ((linked_shader
->info
.redeclares_gl_fragcoord
1819 && !shader
->info
.redeclares_gl_fragcoord
1820 && shader
->info
.uses_gl_fragcoord
)
1821 || (shader
->info
.redeclares_gl_fragcoord
1822 && !linked_shader
->info
.redeclares_gl_fragcoord
1823 && linked_shader
->info
.uses_gl_fragcoord
)) {
1824 linker_error(prog
, "fragment shader defined with conflicting "
1825 "layout qualifiers for gl_FragCoord\n");
1828 /* From the GLSL 1.50 spec, page 39:
1830 * "All redeclarations of gl_FragCoord in all fragment shaders in a
1831 * single program must have the same set of qualifiers."
1833 if (linked_shader
->info
.redeclares_gl_fragcoord
&&
1834 shader
->info
.redeclares_gl_fragcoord
&&
1835 (shader
->info
.origin_upper_left
!=
1836 linked_shader
->info
.origin_upper_left
||
1837 shader
->info
.pixel_center_integer
!=
1838 linked_shader
->info
.pixel_center_integer
)) {
1839 linker_error(prog
, "fragment shader defined with conflicting "
1840 "layout qualifiers for gl_FragCoord\n");
1843 /* Update the linked shader state. Note that uses_gl_fragcoord should
1844 * accumulate the results. The other values should replace. If there
1845 * are multiple redeclarations, all the fields except uses_gl_fragcoord
1846 * are already known to be the same.
1848 if (shader
->info
.redeclares_gl_fragcoord
||
1849 shader
->info
.uses_gl_fragcoord
) {
1850 linked_shader
->info
.redeclares_gl_fragcoord
=
1851 shader
->info
.redeclares_gl_fragcoord
;
1852 linked_shader
->info
.uses_gl_fragcoord
=
1853 linked_shader
->info
.uses_gl_fragcoord
||
1854 shader
->info
.uses_gl_fragcoord
;
1855 linked_shader
->info
.origin_upper_left
=
1856 shader
->info
.origin_upper_left
;
1857 linked_shader
->info
.pixel_center_integer
=
1858 shader
->info
.pixel_center_integer
;
1861 linked_shader
->info
.EarlyFragmentTests
|=
1862 shader
->info
.EarlyFragmentTests
;
1863 linked_shader
->info
.BlendSupport
|= shader
->info
.BlendSupport
;
1868 * Performs the cross-validation of geometry shader max_vertices and
1869 * primitive type layout qualifiers for the attached geometry shaders,
1870 * and propagates them to the linked GS and linked shader program.
1873 link_gs_inout_layout_qualifiers(struct gl_shader_program
*prog
,
1874 struct gl_linked_shader
*linked_shader
,
1875 struct gl_shader
**shader_list
,
1876 unsigned num_shaders
)
1878 linked_shader
->info
.Geom
.VerticesOut
= -1;
1879 linked_shader
->info
.Geom
.Invocations
= 0;
1880 linked_shader
->info
.Geom
.InputType
= PRIM_UNKNOWN
;
1881 linked_shader
->info
.Geom
.OutputType
= PRIM_UNKNOWN
;
1883 /* No in/out qualifiers defined for anything but GLSL 1.50+
1884 * geometry shaders so far.
1886 if (linked_shader
->Stage
!= MESA_SHADER_GEOMETRY
|| prog
->Version
< 150)
1889 /* From the GLSL 1.50 spec, page 46:
1891 * "All geometry shader output layout declarations in a program
1892 * must declare the same layout and same value for
1893 * max_vertices. There must be at least one geometry output
1894 * layout declaration somewhere in a program, but not all
1895 * geometry shaders (compilation units) are required to
1899 for (unsigned i
= 0; i
< num_shaders
; i
++) {
1900 struct gl_shader
*shader
= shader_list
[i
];
1902 if (shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
) {
1903 if (linked_shader
->info
.Geom
.InputType
!= PRIM_UNKNOWN
&&
1904 linked_shader
->info
.Geom
.InputType
!=
1905 shader
->info
.Geom
.InputType
) {
1906 linker_error(prog
, "geometry shader defined with conflicting "
1910 linked_shader
->info
.Geom
.InputType
= shader
->info
.Geom
.InputType
;
1913 if (shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
) {
1914 if (linked_shader
->info
.Geom
.OutputType
!= PRIM_UNKNOWN
&&
1915 linked_shader
->info
.Geom
.OutputType
!=
1916 shader
->info
.Geom
.OutputType
) {
1917 linker_error(prog
, "geometry shader defined with conflicting "
1921 linked_shader
->info
.Geom
.OutputType
= shader
->info
.Geom
.OutputType
;
1924 if (shader
->info
.Geom
.VerticesOut
!= -1) {
1925 if (linked_shader
->info
.Geom
.VerticesOut
!= -1 &&
1926 linked_shader
->info
.Geom
.VerticesOut
!=
1927 shader
->info
.Geom
.VerticesOut
) {
1928 linker_error(prog
, "geometry shader defined with conflicting "
1929 "output vertex count (%d and %d)\n",
1930 linked_shader
->info
.Geom
.VerticesOut
,
1931 shader
->info
.Geom
.VerticesOut
);
1934 linked_shader
->info
.Geom
.VerticesOut
= shader
->info
.Geom
.VerticesOut
;
1937 if (shader
->info
.Geom
.Invocations
!= 0) {
1938 if (linked_shader
->info
.Geom
.Invocations
!= 0 &&
1939 linked_shader
->info
.Geom
.Invocations
!=
1940 shader
->info
.Geom
.Invocations
) {
1941 linker_error(prog
, "geometry shader defined with conflicting "
1942 "invocation count (%d and %d)\n",
1943 linked_shader
->info
.Geom
.Invocations
,
1944 shader
->info
.Geom
.Invocations
);
1947 linked_shader
->info
.Geom
.Invocations
= shader
->info
.Geom
.Invocations
;
1951 /* Just do the intrastage -> interstage propagation right now,
1952 * since we already know we're in the right type of shader program
1955 if (linked_shader
->info
.Geom
.InputType
== PRIM_UNKNOWN
) {
1957 "geometry shader didn't declare primitive input type\n");
1961 if (linked_shader
->info
.Geom
.OutputType
== PRIM_UNKNOWN
) {
1963 "geometry shader didn't declare primitive output type\n");
1967 if (linked_shader
->info
.Geom
.VerticesOut
== -1) {
1969 "geometry shader didn't declare max_vertices\n");
1973 if (linked_shader
->info
.Geom
.Invocations
== 0)
1974 linked_shader
->info
.Geom
.Invocations
= 1;
1979 * Perform cross-validation of compute shader local_size_{x,y,z} layout
1980 * qualifiers for the attached compute shaders, and propagate them to the
1981 * linked CS and linked shader program.
1984 link_cs_input_layout_qualifiers(struct gl_shader_program
*prog
,
1985 struct gl_linked_shader
*linked_shader
,
1986 struct gl_shader
**shader_list
,
1987 unsigned num_shaders
)
1989 for (int i
= 0; i
< 3; i
++)
1990 linked_shader
->info
.Comp
.LocalSize
[i
] = 0;
1992 /* This function is called for all shader stages, but it only has an effect
1993 * for compute shaders.
1995 if (linked_shader
->Stage
!= MESA_SHADER_COMPUTE
)
1998 /* From the ARB_compute_shader spec, in the section describing local size
2001 * If multiple compute shaders attached to a single program object
2002 * declare local work-group size, the declarations must be identical;
2003 * otherwise a link-time error results. Furthermore, if a program
2004 * object contains any compute shaders, at least one must contain an
2005 * input layout qualifier specifying the local work sizes of the
2006 * program, or a link-time error will occur.
2008 for (unsigned sh
= 0; sh
< num_shaders
; sh
++) {
2009 struct gl_shader
*shader
= shader_list
[sh
];
2011 if (shader
->info
.Comp
.LocalSize
[0] != 0) {
2012 if (linked_shader
->info
.Comp
.LocalSize
[0] != 0) {
2013 for (int i
= 0; i
< 3; i
++) {
2014 if (linked_shader
->info
.Comp
.LocalSize
[i
] !=
2015 shader
->info
.Comp
.LocalSize
[i
]) {
2016 linker_error(prog
, "compute shader defined with conflicting "
2022 for (int i
= 0; i
< 3; i
++) {
2023 linked_shader
->info
.Comp
.LocalSize
[i
] =
2024 shader
->info
.Comp
.LocalSize
[i
];
2029 /* Just do the intrastage -> interstage propagation right now,
2030 * since we already know we're in the right type of shader program
2033 if (linked_shader
->info
.Comp
.LocalSize
[0] == 0) {
2034 linker_error(prog
, "compute shader didn't declare local size\n");
2037 for (int i
= 0; i
< 3; i
++)
2038 prog
->Comp
.LocalSize
[i
] = linked_shader
->info
.Comp
.LocalSize
[i
];
2043 * Combine a group of shaders for a single stage to generate a linked shader
2046 * If this function is supplied a single shader, it is cloned, and the new
2047 * shader is returned.
2049 static struct gl_linked_shader
*
2050 link_intrastage_shaders(void *mem_ctx
,
2051 struct gl_context
*ctx
,
2052 struct gl_shader_program
*prog
,
2053 struct gl_shader
**shader_list
,
2054 unsigned num_shaders
)
2056 struct gl_uniform_block
*ubo_blocks
= NULL
;
2057 struct gl_uniform_block
*ssbo_blocks
= NULL
;
2058 unsigned num_ubo_blocks
= 0;
2059 unsigned num_ssbo_blocks
= 0;
2061 /* Check that global variables defined in multiple shaders are consistent.
2063 glsl_symbol_table variables
;
2064 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2065 if (shader_list
[i
] == NULL
)
2067 cross_validate_globals(prog
, shader_list
[i
]->ir
, &variables
, false);
2070 if (!prog
->LinkStatus
)
2073 /* Check that interface blocks defined in multiple shaders are consistent.
2075 validate_intrastage_interface_blocks(prog
, (const gl_shader
**)shader_list
,
2077 if (!prog
->LinkStatus
)
2080 /* Check that there is only a single definition of each function signature
2081 * across all shaders.
2083 for (unsigned i
= 0; i
< (num_shaders
- 1); i
++) {
2084 foreach_in_list(ir_instruction
, node
, shader_list
[i
]->ir
) {
2085 ir_function
*const f
= node
->as_function();
2090 for (unsigned j
= i
+ 1; j
< num_shaders
; j
++) {
2091 ir_function
*const other
=
2092 shader_list
[j
]->symbols
->get_function(f
->name
);
2094 /* If the other shader has no function (and therefore no function
2095 * signatures) with the same name, skip to the next shader.
2100 foreach_in_list(ir_function_signature
, sig
, &f
->signatures
) {
2101 if (!sig
->is_defined
)
2104 ir_function_signature
*other_sig
=
2105 other
->exact_matching_signature(NULL
, &sig
->parameters
);
2107 if (other_sig
!= NULL
&& other_sig
->is_defined
) {
2108 linker_error(prog
, "function `%s' is multiply defined\n",
2117 /* Find the shader that defines main, and make a clone of it.
2119 * Starting with the clone, search for undefined references. If one is
2120 * found, find the shader that defines it. Clone the reference and add
2121 * it to the shader. Repeat until there are no undefined references or
2122 * until a reference cannot be resolved.
2124 gl_shader
*main
= NULL
;
2125 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2126 if (_mesa_get_main_function_signature(shader_list
[i
]->symbols
)) {
2127 main
= shader_list
[i
];
2133 linker_error(prog
, "%s shader lacks `main'\n",
2134 _mesa_shader_stage_to_string(shader_list
[0]->Stage
));
2138 gl_linked_shader
*linked
= ctx
->Driver
.NewShader(shader_list
[0]->Stage
);
2139 linked
->ir
= new(linked
) exec_list
;
2140 clone_ir_list(mem_ctx
, linked
->ir
, main
->ir
);
2142 link_fs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2143 link_tcs_out_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2144 link_tes_in_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2145 link_gs_inout_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2146 link_cs_input_layout_qualifiers(prog
, linked
, shader_list
, num_shaders
);
2147 link_xfb_stride_layout_qualifiers(ctx
, prog
, linked
, shader_list
,
2150 populate_symbol_table(linked
);
2152 /* The pointer to the main function in the final linked shader (i.e., the
2153 * copy of the original shader that contained the main function).
2155 ir_function_signature
*const main_sig
=
2156 _mesa_get_main_function_signature(linked
->symbols
);
2158 /* Move any instructions other than variable declarations or function
2159 * declarations into main.
2161 exec_node
*insertion_point
=
2162 move_non_declarations(linked
->ir
, (exec_node
*) &main_sig
->body
, false,
2165 for (unsigned i
= 0; i
< num_shaders
; i
++) {
2166 if (shader_list
[i
] == main
)
2169 insertion_point
= move_non_declarations(shader_list
[i
]->ir
,
2170 insertion_point
, true, linked
);
2173 if (!link_function_calls(prog
, linked
, shader_list
, num_shaders
)) {
2174 _mesa_delete_linked_shader(ctx
, linked
);
2178 /* Make a pass over all variable declarations to ensure that arrays with
2179 * unspecified sizes have a size specified. The size is inferred from the
2180 * max_array_access field.
2182 array_sizing_visitor v
;
2184 v
.fixup_unnamed_interface_types();
2186 /* Link up uniform blocks defined within this stage. */
2187 link_uniform_blocks(mem_ctx
, ctx
, prog
, linked
, &ubo_blocks
,
2188 &num_ubo_blocks
, &ssbo_blocks
, &num_ssbo_blocks
);
2190 if (!prog
->LinkStatus
) {
2191 _mesa_delete_linked_shader(ctx
, linked
);
2195 /* Copy ubo blocks to linked shader list */
2196 linked
->UniformBlocks
=
2197 ralloc_array(linked
, gl_uniform_block
*, num_ubo_blocks
);
2198 ralloc_steal(linked
, ubo_blocks
);
2199 for (unsigned i
= 0; i
< num_ubo_blocks
; i
++) {
2200 linked
->UniformBlocks
[i
] = &ubo_blocks
[i
];
2202 linked
->NumUniformBlocks
= num_ubo_blocks
;
2204 /* Copy ssbo blocks to linked shader list */
2205 linked
->ShaderStorageBlocks
=
2206 ralloc_array(linked
, gl_uniform_block
*, num_ssbo_blocks
);
2207 ralloc_steal(linked
, ssbo_blocks
);
2208 for (unsigned i
= 0; i
< num_ssbo_blocks
; i
++) {
2209 linked
->ShaderStorageBlocks
[i
] = &ssbo_blocks
[i
];
2211 linked
->NumShaderStorageBlocks
= num_ssbo_blocks
;
2213 /* At this point linked should contain all of the linked IR, so
2214 * validate it to make sure nothing went wrong.
2216 validate_ir_tree(linked
->ir
);
2218 /* Set the size of geometry shader input arrays */
2219 if (linked
->Stage
== MESA_SHADER_GEOMETRY
) {
2220 unsigned num_vertices
= vertices_per_prim(linked
->info
.Geom
.InputType
);
2221 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2222 MESA_SHADER_GEOMETRY
);
2223 foreach_in_list(ir_instruction
, ir
, linked
->ir
) {
2224 ir
->accept(&input_resize_visitor
);
2228 if (ctx
->Const
.VertexID_is_zero_based
)
2229 lower_vertex_id(linked
);
2235 * Update the sizes of linked shader uniform arrays to the maximum
2238 * From page 81 (page 95 of the PDF) of the OpenGL 2.1 spec:
2240 * If one or more elements of an array are active,
2241 * GetActiveUniform will return the name of the array in name,
2242 * subject to the restrictions listed above. The type of the array
2243 * is returned in type. The size parameter contains the highest
2244 * array element index used, plus one. The compiler or linker
2245 * determines the highest index used. There will be only one
2246 * active uniform reported by the GL per uniform array.
2250 update_array_sizes(struct gl_shader_program
*prog
)
2252 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2253 if (prog
->_LinkedShaders
[i
] == NULL
)
2256 foreach_in_list(ir_instruction
, node
, prog
->_LinkedShaders
[i
]->ir
) {
2257 ir_variable
*const var
= node
->as_variable();
2259 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_uniform
) ||
2260 !var
->type
->is_array())
2263 /* GL_ARB_uniform_buffer_object says that std140 uniforms
2264 * will not be eliminated. Since we always do std140, just
2265 * don't resize arrays in UBOs.
2267 * Atomic counters are supposed to get deterministic
2268 * locations assigned based on the declaration ordering and
2269 * sizes, array compaction would mess that up.
2271 * Subroutine uniforms are not removed.
2273 if (var
->is_in_buffer_block() || var
->type
->contains_atomic() ||
2274 var
->type
->contains_subroutine() || var
->constant_initializer
)
2277 int size
= var
->data
.max_array_access
;
2278 for (unsigned j
= 0; j
< MESA_SHADER_STAGES
; j
++) {
2279 if (prog
->_LinkedShaders
[j
] == NULL
)
2282 foreach_in_list(ir_instruction
, node2
, prog
->_LinkedShaders
[j
]->ir
) {
2283 ir_variable
*other_var
= node2
->as_variable();
2287 if (strcmp(var
->name
, other_var
->name
) == 0 &&
2288 other_var
->data
.max_array_access
> size
) {
2289 size
= other_var
->data
.max_array_access
;
2294 if (size
+ 1 != (int)var
->type
->length
) {
2295 /* If this is a built-in uniform (i.e., it's backed by some
2296 * fixed-function state), adjust the number of state slots to
2297 * match the new array size. The number of slots per array entry
2298 * is not known. It seems safe to assume that the total number of
2299 * slots is an integer multiple of the number of array elements.
2300 * Determine the number of slots per array element by dividing by
2301 * the old (total) size.
2303 const unsigned num_slots
= var
->get_num_state_slots();
2304 if (num_slots
> 0) {
2305 var
->set_num_state_slots((size
+ 1)
2306 * (num_slots
/ var
->type
->length
));
2309 var
->type
= glsl_type::get_array_instance(var
->type
->fields
.array
,
2311 /* FINISHME: We should update the types of array
2312 * dereferences of this variable now.
2320 * Resize tessellation evaluation per-vertex inputs to the size of
2321 * tessellation control per-vertex outputs.
2324 resize_tes_inputs(struct gl_context
*ctx
,
2325 struct gl_shader_program
*prog
)
2327 if (prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
] == NULL
)
2330 gl_linked_shader
*const tcs
= prog
->_LinkedShaders
[MESA_SHADER_TESS_CTRL
];
2331 gl_linked_shader
*const tes
= prog
->_LinkedShaders
[MESA_SHADER_TESS_EVAL
];
2333 /* If no control shader is present, then the TES inputs are statically
2334 * sized to MaxPatchVertices; the actual size of the arrays won't be
2335 * known until draw time.
2337 const int num_vertices
= tcs
2338 ? tcs
->info
.TessCtrl
.VerticesOut
2339 : ctx
->Const
.MaxPatchVertices
;
2341 array_resize_visitor
input_resize_visitor(num_vertices
, prog
,
2342 MESA_SHADER_TESS_EVAL
);
2343 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2344 ir
->accept(&input_resize_visitor
);
2347 if (tcs
|| ctx
->Const
.LowerTESPatchVerticesIn
) {
2348 /* Convert the gl_PatchVerticesIn system value into a constant, since
2349 * the value is known at this point.
2351 foreach_in_list(ir_instruction
, ir
, tes
->ir
) {
2352 ir_variable
*var
= ir
->as_variable();
2353 if (var
&& var
->data
.mode
== ir_var_system_value
&&
2354 var
->data
.location
== SYSTEM_VALUE_VERTICES_IN
) {
2355 void *mem_ctx
= ralloc_parent(var
);
2356 var
->data
.location
= 0;
2357 var
->data
.explicit_location
= false;
2359 var
->data
.mode
= ir_var_auto
;
2360 var
->constant_value
= new(mem_ctx
) ir_constant(num_vertices
);
2362 var
->data
.mode
= ir_var_uniform
;
2363 var
->data
.how_declared
= ir_var_hidden
;
2364 var
->allocate_state_slots(1);
2365 ir_state_slot
*slot0
= &var
->get_state_slots()[0];
2366 slot0
->swizzle
= SWIZZLE_XXXX
;
2367 slot0
->tokens
[0] = STATE_INTERNAL
;
2368 slot0
->tokens
[1] = STATE_TES_PATCH_VERTICES_IN
;
2369 for (int i
= 2; i
< STATE_LENGTH
; i
++)
2370 slot0
->tokens
[i
] = 0;
2378 * Find a contiguous set of available bits in a bitmask.
2380 * \param used_mask Bits representing used (1) and unused (0) locations
2381 * \param needed_count Number of contiguous bits needed.
2384 * Base location of the available bits on success or -1 on failure.
2387 find_available_slots(unsigned used_mask
, unsigned needed_count
)
2389 unsigned needed_mask
= (1 << needed_count
) - 1;
2390 const int max_bit_to_test
= (8 * sizeof(used_mask
)) - needed_count
;
2392 /* The comparison to 32 is redundant, but without it GCC emits "warning:
2393 * cannot optimize possibly infinite loops" for the loop below.
2395 if ((needed_count
== 0) || (max_bit_to_test
< 0) || (max_bit_to_test
> 32))
2398 for (int i
= 0; i
<= max_bit_to_test
; i
++) {
2399 if ((needed_mask
& ~used_mask
) == needed_mask
)
2410 * Assign locations for either VS inputs or FS outputs
2412 * \param mem_ctx Temporary ralloc context used for linking
2413 * \param prog Shader program whose variables need locations assigned
2414 * \param constants Driver specific constant values for the program.
2415 * \param target_index Selector for the program target to receive location
2416 * assignmnets. Must be either \c MESA_SHADER_VERTEX or
2417 * \c MESA_SHADER_FRAGMENT.
2420 * If locations are successfully assigned, true is returned. Otherwise an
2421 * error is emitted to the shader link log and false is returned.
2424 assign_attribute_or_color_locations(void *mem_ctx
,
2425 gl_shader_program
*prog
,
2426 struct gl_constants
*constants
,
2427 unsigned target_index
)
2429 /* Maximum number of generic locations. This corresponds to either the
2430 * maximum number of draw buffers or the maximum number of generic
2433 unsigned max_index
= (target_index
== MESA_SHADER_VERTEX
) ?
2434 constants
->Program
[target_index
].MaxAttribs
:
2435 MAX2(constants
->MaxDrawBuffers
, constants
->MaxDualSourceDrawBuffers
);
2437 /* Mark invalid locations as being used.
2439 unsigned used_locations
= (max_index
>= 32)
2440 ? ~0 : ~((1 << max_index
) - 1);
2441 unsigned double_storage_locations
= 0;
2443 assert((target_index
== MESA_SHADER_VERTEX
)
2444 || (target_index
== MESA_SHADER_FRAGMENT
));
2446 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[target_index
];
2450 /* Operate in a total of four passes.
2452 * 1. Invalidate the location assignments for all vertex shader inputs.
2454 * 2. Assign locations for inputs that have user-defined (via
2455 * glBindVertexAttribLocation) locations and outputs that have
2456 * user-defined locations (via glBindFragDataLocation).
2458 * 3. Sort the attributes without assigned locations by number of slots
2459 * required in decreasing order. Fragmentation caused by attribute
2460 * locations assigned by the application may prevent large attributes
2461 * from having enough contiguous space.
2463 * 4. Assign locations to any inputs without assigned locations.
2466 const int generic_base
= (target_index
== MESA_SHADER_VERTEX
)
2467 ? (int) VERT_ATTRIB_GENERIC0
: (int) FRAG_RESULT_DATA0
;
2469 const enum ir_variable_mode direction
=
2470 (target_index
== MESA_SHADER_VERTEX
)
2471 ? ir_var_shader_in
: ir_var_shader_out
;
2474 /* Temporary storage for the set of attributes that need locations assigned.
2480 /* Used below in the call to qsort. */
2481 static int compare(const void *a
, const void *b
)
2483 const temp_attr
*const l
= (const temp_attr
*) a
;
2484 const temp_attr
*const r
= (const temp_attr
*) b
;
2486 /* Reversed because we want a descending order sort below. */
2487 return r
->slots
- l
->slots
;
2490 assert(max_index
<= 32);
2492 /* Temporary array for the set of attributes that have locations assigned.
2494 ir_variable
*assigned
[16];
2496 unsigned num_attr
= 0;
2497 unsigned assigned_attr
= 0;
2499 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
2500 ir_variable
*const var
= node
->as_variable();
2502 if ((var
== NULL
) || (var
->data
.mode
!= (unsigned) direction
))
2505 if (var
->data
.explicit_location
) {
2506 var
->data
.is_unmatched_generic_inout
= 0;
2507 if ((var
->data
.location
>= (int)(max_index
+ generic_base
))
2508 || (var
->data
.location
< 0)) {
2510 "invalid explicit location %d specified for `%s'\n",
2511 (var
->data
.location
< 0)
2512 ? var
->data
.location
2513 : var
->data
.location
- generic_base
,
2517 } else if (target_index
== MESA_SHADER_VERTEX
) {
2520 if (prog
->AttributeBindings
->get(binding
, var
->name
)) {
2521 assert(binding
>= VERT_ATTRIB_GENERIC0
);
2522 var
->data
.location
= binding
;
2523 var
->data
.is_unmatched_generic_inout
= 0;
2525 } else if (target_index
== MESA_SHADER_FRAGMENT
) {
2528 const char *name
= var
->name
;
2529 const glsl_type
*type
= var
->type
;
2532 /* Check if there's a binding for the variable name */
2533 if (prog
->FragDataBindings
->get(binding
, name
)) {
2534 assert(binding
>= FRAG_RESULT_DATA0
);
2535 var
->data
.location
= binding
;
2536 var
->data
.is_unmatched_generic_inout
= 0;
2538 if (prog
->FragDataIndexBindings
->get(index
, name
)) {
2539 var
->data
.index
= index
;
2544 /* If not, but it's an array type, look for name[0] */
2545 if (type
->is_array()) {
2546 name
= ralloc_asprintf(mem_ctx
, "%s[0]", name
);
2547 type
= type
->fields
.array
;
2555 if (strcmp(var
->name
, "gl_LastFragData") == 0)
2558 /* From GL4.5 core spec, section 15.2 (Shader Execution):
2560 * "Output binding assignments will cause LinkProgram to fail:
2562 * If the program has an active output assigned to a location greater
2563 * than or equal to the value of MAX_DUAL_SOURCE_DRAW_BUFFERS and has
2564 * an active output assigned an index greater than or equal to one;"
2566 if (target_index
== MESA_SHADER_FRAGMENT
&& var
->data
.index
>= 1 &&
2567 var
->data
.location
- generic_base
>=
2568 (int) constants
->MaxDualSourceDrawBuffers
) {
2570 "output location %d >= GL_MAX_DUAL_SOURCE_DRAW_BUFFERS "
2571 "with index %u for %s\n",
2572 var
->data
.location
- generic_base
, var
->data
.index
,
2577 const unsigned slots
= var
->type
->count_attribute_slots(target_index
== MESA_SHADER_VERTEX
);
2579 /* If the variable is not a built-in and has a location statically
2580 * assigned in the shader (presumably via a layout qualifier), make sure
2581 * that it doesn't collide with other assigned locations. Otherwise,
2582 * add it to the list of variables that need linker-assigned locations.
2584 if (var
->data
.location
!= -1) {
2585 if (var
->data
.location
>= generic_base
&& var
->data
.index
< 1) {
2586 /* From page 61 of the OpenGL 4.0 spec:
2588 * "LinkProgram will fail if the attribute bindings assigned
2589 * by BindAttribLocation do not leave not enough space to
2590 * assign a location for an active matrix attribute or an
2591 * active attribute array, both of which require multiple
2592 * contiguous generic attributes."
2594 * I think above text prohibits the aliasing of explicit and
2595 * automatic assignments. But, aliasing is allowed in manual
2596 * assignments of attribute locations. See below comments for
2599 * From OpenGL 4.0 spec, page 61:
2601 * "It is possible for an application to bind more than one
2602 * attribute name to the same location. This is referred to as
2603 * aliasing. This will only work if only one of the aliased
2604 * attributes is active in the executable program, or if no
2605 * path through the shader consumes more than one attribute of
2606 * a set of attributes aliased to the same location. A link
2607 * error can occur if the linker determines that every path
2608 * through the shader consumes multiple aliased attributes,
2609 * but implementations are not required to generate an error
2612 * From GLSL 4.30 spec, page 54:
2614 * "A program will fail to link if any two non-vertex shader
2615 * input variables are assigned to the same location. For
2616 * vertex shaders, multiple input variables may be assigned
2617 * to the same location using either layout qualifiers or via
2618 * the OpenGL API. However, such aliasing is intended only to
2619 * support vertex shaders where each execution path accesses
2620 * at most one input per each location. Implementations are
2621 * permitted, but not required, to generate link-time errors
2622 * if they detect that every path through the vertex shader
2623 * executable accesses multiple inputs assigned to any single
2624 * location. For all shader types, a program will fail to link
2625 * if explicit location assignments leave the linker unable
2626 * to find space for other variables without explicit
2629 * From OpenGL ES 3.0 spec, page 56:
2631 * "Binding more than one attribute name to the same location
2632 * is referred to as aliasing, and is not permitted in OpenGL
2633 * ES Shading Language 3.00 vertex shaders. LinkProgram will
2634 * fail when this condition exists. However, aliasing is
2635 * possible in OpenGL ES Shading Language 1.00 vertex shaders.
2636 * This will only work if only one of the aliased attributes
2637 * is active in the executable program, or if no path through
2638 * the shader consumes more than one attribute of a set of
2639 * attributes aliased to the same location. A link error can
2640 * occur if the linker determines that every path through the
2641 * shader consumes multiple aliased attributes, but implemen-
2642 * tations are not required to generate an error in this case."
2644 * After looking at above references from OpenGL, OpenGL ES and
2645 * GLSL specifications, we allow aliasing of vertex input variables
2646 * in: OpenGL 2.0 (and above) and OpenGL ES 2.0.
2648 * NOTE: This is not required by the spec but its worth mentioning
2649 * here that we're not doing anything to make sure that no path
2650 * through the vertex shader executable accesses multiple inputs
2651 * assigned to any single location.
2654 /* Mask representing the contiguous slots that will be used by
2657 const unsigned attr
= var
->data
.location
- generic_base
;
2658 const unsigned use_mask
= (1 << slots
) - 1;
2659 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2660 ? "vertex shader input" : "fragment shader output";
2662 /* Generate a link error if the requested locations for this
2663 * attribute exceed the maximum allowed attribute location.
2665 if (attr
+ slots
> max_index
) {
2667 "insufficient contiguous locations "
2668 "available for %s `%s' %d %d %d\n", string
,
2669 var
->name
, used_locations
, use_mask
, attr
);
2673 /* Generate a link error if the set of bits requested for this
2674 * attribute overlaps any previously allocated bits.
2676 if ((~(use_mask
<< attr
) & used_locations
) != used_locations
) {
2677 if (target_index
== MESA_SHADER_FRAGMENT
&& !prog
->IsES
) {
2678 /* From section 4.4.2 (Output Layout Qualifiers) of the GLSL
2681 * "Additionally, for fragment shader outputs, if two
2682 * variables are placed within the same location, they
2683 * must have the same underlying type (floating-point or
2684 * integer). No component aliasing of output variables or
2685 * members is allowed.
2687 for (unsigned i
= 0; i
< assigned_attr
; i
++) {
2688 unsigned assigned_slots
=
2689 assigned
[i
]->type
->count_attribute_slots(false);
2690 unsigned assig_attr
=
2691 assigned
[i
]->data
.location
- generic_base
;
2692 unsigned assigned_use_mask
= (1 << assigned_slots
) - 1;
2694 if ((assigned_use_mask
<< assig_attr
) &
2695 (use_mask
<< attr
)) {
2697 const glsl_type
*assigned_type
=
2698 assigned
[i
]->type
->without_array();
2699 const glsl_type
*type
= var
->type
->without_array();
2700 if (assigned_type
->base_type
!= type
->base_type
) {
2701 linker_error(prog
, "types do not match for aliased"
2702 " %ss %s and %s\n", string
,
2703 assigned
[i
]->name
, var
->name
);
2707 unsigned assigned_component_mask
=
2708 ((1 << assigned_type
->vector_elements
) - 1) <<
2709 assigned
[i
]->data
.location_frac
;
2710 unsigned component_mask
=
2711 ((1 << type
->vector_elements
) - 1) <<
2712 var
->data
.location_frac
;
2713 if (assigned_component_mask
& component_mask
) {
2714 linker_error(prog
, "overlapping component is "
2715 "assigned to %ss %s and %s "
2717 string
, assigned
[i
]->name
, var
->name
,
2718 var
->data
.location_frac
);
2723 } else if (target_index
== MESA_SHADER_FRAGMENT
||
2724 (prog
->IsES
&& prog
->Version
>= 300)) {
2725 linker_error(prog
, "overlapping location is assigned "
2726 "to %s `%s' %d %d %d\n", string
, var
->name
,
2727 used_locations
, use_mask
, attr
);
2730 linker_warning(prog
, "overlapping location is assigned "
2731 "to %s `%s' %d %d %d\n", string
, var
->name
,
2732 used_locations
, use_mask
, attr
);
2736 used_locations
|= (use_mask
<< attr
);
2738 /* From the GL 4.5 core spec, section 11.1.1 (Vertex Attributes):
2740 * "A program with more than the value of MAX_VERTEX_ATTRIBS
2741 * active attribute variables may fail to link, unless
2742 * device-dependent optimizations are able to make the program
2743 * fit within available hardware resources. For the purposes
2744 * of this test, attribute variables of the type dvec3, dvec4,
2745 * dmat2x3, dmat2x4, dmat3, dmat3x4, dmat4x3, and dmat4 may
2746 * count as consuming twice as many attributes as equivalent
2747 * single-precision types. While these types use the same number
2748 * of generic attributes as their single-precision equivalents,
2749 * implementations are permitted to consume two single-precision
2750 * vectors of internal storage for each three- or four-component
2751 * double-precision vector."
2753 * Mark this attribute slot as taking up twice as much space
2754 * so we can count it properly against limits. According to
2755 * issue (3) of the GL_ARB_vertex_attrib_64bit behavior, this
2756 * is optional behavior, but it seems preferable.
2758 if (var
->type
->without_array()->is_dual_slot())
2759 double_storage_locations
|= (use_mask
<< attr
);
2762 assigned
[assigned_attr
] = var
;
2768 if (num_attr
>= max_index
) {
2769 linker_error(prog
, "too many %s (max %u)",
2770 target_index
== MESA_SHADER_VERTEX
?
2771 "vertex shader inputs" : "fragment shader outputs",
2775 to_assign
[num_attr
].slots
= slots
;
2776 to_assign
[num_attr
].var
= var
;
2780 if (target_index
== MESA_SHADER_VERTEX
) {
2781 unsigned total_attribs_size
=
2782 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2783 _mesa_bitcount(double_storage_locations
);
2784 if (total_attribs_size
> max_index
) {
2786 "attempt to use %d vertex attribute slots only %d available ",
2787 total_attribs_size
, max_index
);
2792 /* If all of the attributes were assigned locations by the application (or
2793 * are built-in attributes with fixed locations), return early. This should
2794 * be the common case.
2799 qsort(to_assign
, num_attr
, sizeof(to_assign
[0]), temp_attr::compare
);
2801 if (target_index
== MESA_SHADER_VERTEX
) {
2802 /* VERT_ATTRIB_GENERIC0 is a pseudo-alias for VERT_ATTRIB_POS. It can
2803 * only be explicitly assigned by via glBindAttribLocation. Mark it as
2804 * reserved to prevent it from being automatically allocated below.
2806 find_deref_visitor
find("gl_Vertex");
2808 if (find
.variable_found())
2809 used_locations
|= (1 << 0);
2812 for (unsigned i
= 0; i
< num_attr
; i
++) {
2813 /* Mask representing the contiguous slots that will be used by this
2816 const unsigned use_mask
= (1 << to_assign
[i
].slots
) - 1;
2818 int location
= find_available_slots(used_locations
, to_assign
[i
].slots
);
2821 const char *const string
= (target_index
== MESA_SHADER_VERTEX
)
2822 ? "vertex shader input" : "fragment shader output";
2825 "insufficient contiguous locations "
2826 "available for %s `%s'\n",
2827 string
, to_assign
[i
].var
->name
);
2831 to_assign
[i
].var
->data
.location
= generic_base
+ location
;
2832 to_assign
[i
].var
->data
.is_unmatched_generic_inout
= 0;
2833 used_locations
|= (use_mask
<< location
);
2835 if (to_assign
[i
].var
->type
->without_array()->is_dual_slot())
2836 double_storage_locations
|= (use_mask
<< location
);
2839 /* Now that we have all the locations, from the GL 4.5 core spec, section
2840 * 11.1.1 (Vertex Attributes), dvec3, dvec4, dmat2x3, dmat2x4, dmat3,
2841 * dmat3x4, dmat4x3, and dmat4 count as consuming twice as many attributes
2842 * as equivalent single-precision types.
2844 if (target_index
== MESA_SHADER_VERTEX
) {
2845 unsigned total_attribs_size
=
2846 _mesa_bitcount(used_locations
& ((1 << max_index
) - 1)) +
2847 _mesa_bitcount(double_storage_locations
);
2848 if (total_attribs_size
> max_index
) {
2850 "attempt to use %d vertex attribute slots only %d available ",
2851 total_attribs_size
, max_index
);
2860 * Match explicit locations of outputs to inputs and deactivate the
2861 * unmatch flag if found so we don't optimise them away.
2864 match_explicit_outputs_to_inputs(gl_linked_shader
*producer
,
2865 gl_linked_shader
*consumer
)
2867 glsl_symbol_table parameters
;
2868 ir_variable
*explicit_locations
[MAX_VARYINGS_INCL_PATCH
][4] =
2871 /* Find all shader outputs in the "producer" stage.
2873 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2874 ir_variable
*const var
= node
->as_variable();
2876 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
2879 if (var
->data
.explicit_location
&&
2880 var
->data
.location
>= VARYING_SLOT_VAR0
) {
2881 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
2882 if (explicit_locations
[idx
][var
->data
.location_frac
] == NULL
)
2883 explicit_locations
[idx
][var
->data
.location_frac
] = var
;
2887 /* Match inputs to outputs */
2888 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2889 ir_variable
*const input
= node
->as_variable();
2891 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
2894 ir_variable
*output
= NULL
;
2895 if (input
->data
.explicit_location
2896 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
2897 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
]
2898 [input
->data
.location_frac
];
2900 if (output
!= NULL
){
2901 input
->data
.is_unmatched_generic_inout
= 0;
2902 output
->data
.is_unmatched_generic_inout
= 0;
2909 * Store the gl_FragDepth layout in the gl_shader_program struct.
2912 store_fragdepth_layout(struct gl_shader_program
*prog
)
2914 if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
2918 struct exec_list
*ir
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
]->ir
;
2920 /* We don't look up the gl_FragDepth symbol directly because if
2921 * gl_FragDepth is not used in the shader, it's removed from the IR.
2922 * However, the symbol won't be removed from the symbol table.
2924 * We're only interested in the cases where the variable is NOT removed
2927 foreach_in_list(ir_instruction
, node
, ir
) {
2928 ir_variable
*const var
= node
->as_variable();
2930 if (var
== NULL
|| var
->data
.mode
!= ir_var_shader_out
) {
2934 if (strcmp(var
->name
, "gl_FragDepth") == 0) {
2935 switch (var
->data
.depth_layout
) {
2936 case ir_depth_layout_none
:
2937 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_NONE
;
2939 case ir_depth_layout_any
:
2940 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_ANY
;
2942 case ir_depth_layout_greater
:
2943 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_GREATER
;
2945 case ir_depth_layout_less
:
2946 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_LESS
;
2948 case ir_depth_layout_unchanged
:
2949 prog
->FragDepthLayout
= FRAG_DEPTH_LAYOUT_UNCHANGED
;
2960 * Validate the resources used by a program versus the implementation limits
2963 check_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
2965 unsigned total_uniform_blocks
= 0;
2966 unsigned total_shader_storage_blocks
= 0;
2968 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
2969 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
2974 if (sh
->num_samplers
> ctx
->Const
.Program
[i
].MaxTextureImageUnits
) {
2975 linker_error(prog
, "Too many %s shader texture samplers\n",
2976 _mesa_shader_stage_to_string(i
));
2979 if (sh
->num_uniform_components
>
2980 ctx
->Const
.Program
[i
].MaxUniformComponents
) {
2981 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2982 linker_warning(prog
, "Too many %s shader default uniform block "
2983 "components, but the driver will try to optimize "
2984 "them out; this is non-portable out-of-spec "
2986 _mesa_shader_stage_to_string(i
));
2988 linker_error(prog
, "Too many %s shader default uniform block "
2990 _mesa_shader_stage_to_string(i
));
2994 if (sh
->num_combined_uniform_components
>
2995 ctx
->Const
.Program
[i
].MaxCombinedUniformComponents
) {
2996 if (ctx
->Const
.GLSLSkipStrictMaxUniformLimitCheck
) {
2997 linker_warning(prog
, "Too many %s shader uniform components, "
2998 "but the driver will try to optimize them out; "
2999 "this is non-portable out-of-spec behavior\n",
3000 _mesa_shader_stage_to_string(i
));
3002 linker_error(prog
, "Too many %s shader uniform components\n",
3003 _mesa_shader_stage_to_string(i
));
3007 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3008 total_uniform_blocks
+= sh
->NumUniformBlocks
;
3010 const unsigned max_uniform_blocks
=
3011 ctx
->Const
.Program
[i
].MaxUniformBlocks
;
3012 if (max_uniform_blocks
< sh
->NumUniformBlocks
) {
3013 linker_error(prog
, "Too many %s uniform blocks (%d/%d)\n",
3014 _mesa_shader_stage_to_string(i
), sh
->NumUniformBlocks
,
3015 max_uniform_blocks
);
3018 const unsigned max_shader_storage_blocks
=
3019 ctx
->Const
.Program
[i
].MaxShaderStorageBlocks
;
3020 if (max_shader_storage_blocks
< sh
->NumShaderStorageBlocks
) {
3021 linker_error(prog
, "Too many %s shader storage blocks (%d/%d)\n",
3022 _mesa_shader_stage_to_string(i
),
3023 sh
->NumShaderStorageBlocks
, max_shader_storage_blocks
);
3027 if (total_uniform_blocks
> ctx
->Const
.MaxCombinedUniformBlocks
) {
3028 linker_error(prog
, "Too many combined uniform blocks (%d/%d)\n",
3029 total_uniform_blocks
, ctx
->Const
.MaxCombinedUniformBlocks
);
3032 if (total_shader_storage_blocks
> ctx
->Const
.MaxCombinedShaderStorageBlocks
) {
3033 linker_error(prog
, "Too many combined shader storage blocks (%d/%d)\n",
3034 total_shader_storage_blocks
,
3035 ctx
->Const
.MaxCombinedShaderStorageBlocks
);
3038 for (unsigned i
= 0; i
< prog
->NumUniformBlocks
; i
++) {
3039 if (prog
->UniformBlocks
[i
].UniformBufferSize
>
3040 ctx
->Const
.MaxUniformBlockSize
) {
3041 linker_error(prog
, "Uniform block %s too big (%d/%d)\n",
3042 prog
->UniformBlocks
[i
].Name
,
3043 prog
->UniformBlocks
[i
].UniformBufferSize
,
3044 ctx
->Const
.MaxUniformBlockSize
);
3048 for (unsigned i
= 0; i
< prog
->NumShaderStorageBlocks
; i
++) {
3049 if (prog
->ShaderStorageBlocks
[i
].UniformBufferSize
>
3050 ctx
->Const
.MaxShaderStorageBlockSize
) {
3051 linker_error(prog
, "Shader storage block %s too big (%d/%d)\n",
3052 prog
->ShaderStorageBlocks
[i
].Name
,
3053 prog
->ShaderStorageBlocks
[i
].UniformBufferSize
,
3054 ctx
->Const
.MaxShaderStorageBlockSize
);
3060 link_calculate_subroutine_compat(struct gl_shader_program
*prog
)
3062 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3063 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3068 for (unsigned j
= 0; j
< sh
->NumSubroutineUniformRemapTable
; j
++) {
3069 if (sh
->SubroutineUniformRemapTable
[j
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
)
3072 struct gl_uniform_storage
*uni
= sh
->SubroutineUniformRemapTable
[j
];
3077 sh
->NumSubroutineUniforms
++;
3079 if (sh
->NumSubroutineFunctions
== 0) {
3080 linker_error(prog
, "subroutine uniform %s defined but no valid functions found\n", uni
->type
->name
);
3083 for (unsigned f
= 0; f
< sh
->NumSubroutineFunctions
; f
++) {
3084 struct gl_subroutine_function
*fn
= &sh
->SubroutineFunctions
[f
];
3085 for (int k
= 0; k
< fn
->num_compat_types
; k
++) {
3086 if (fn
->types
[k
] == uni
->type
) {
3092 uni
->num_compatible_subroutines
= count
;
3098 check_subroutine_resources(struct gl_shader_program
*prog
)
3100 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3101 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3104 if (sh
->NumSubroutineUniformRemapTable
> MAX_SUBROUTINE_UNIFORM_LOCATIONS
)
3105 linker_error(prog
, "Too many %s shader subroutine uniforms\n",
3106 _mesa_shader_stage_to_string(i
));
3111 * Validate shader image resources.
3114 check_image_resources(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
3116 unsigned total_image_units
= 0;
3117 unsigned fragment_outputs
= 0;
3118 unsigned total_shader_storage_blocks
= 0;
3120 if (!ctx
->Extensions
.ARB_shader_image_load_store
)
3123 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3124 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3127 if (sh
->NumImages
> ctx
->Const
.Program
[i
].MaxImageUniforms
)
3128 linker_error(prog
, "Too many %s shader image uniforms (%u > %u)\n",
3129 _mesa_shader_stage_to_string(i
), sh
->NumImages
,
3130 ctx
->Const
.Program
[i
].MaxImageUniforms
);
3132 total_image_units
+= sh
->NumImages
;
3133 total_shader_storage_blocks
+= sh
->NumShaderStorageBlocks
;
3135 if (i
== MESA_SHADER_FRAGMENT
) {
3136 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3137 ir_variable
*var
= node
->as_variable();
3138 if (var
&& var
->data
.mode
== ir_var_shader_out
)
3139 /* since there are no double fs outputs - pass false */
3140 fragment_outputs
+= var
->type
->count_attribute_slots(false);
3146 if (total_image_units
> ctx
->Const
.MaxCombinedImageUniforms
)
3147 linker_error(prog
, "Too many combined image uniforms\n");
3149 if (total_image_units
+ fragment_outputs
+ total_shader_storage_blocks
>
3150 ctx
->Const
.MaxCombinedShaderOutputResources
)
3151 linker_error(prog
, "Too many combined image uniforms, shader storage "
3152 " buffers and fragment outputs\n");
3157 * Initializes explicit location slots to INACTIVE_UNIFORM_EXPLICIT_LOCATION
3158 * for a variable, checks for overlaps between other uniforms using explicit
3162 reserve_explicit_locations(struct gl_shader_program
*prog
,
3163 string_to_uint_map
*map
, ir_variable
*var
)
3165 unsigned slots
= var
->type
->uniform_locations();
3166 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3167 unsigned return_value
= slots
;
3169 /* Resize remap table if locations do not fit in the current one. */
3170 if (max_loc
+ 1 > prog
->NumUniformRemapTable
) {
3171 prog
->UniformRemapTable
=
3172 reralloc(prog
, prog
->UniformRemapTable
,
3173 gl_uniform_storage
*,
3176 if (!prog
->UniformRemapTable
) {
3177 linker_error(prog
, "Out of memory during linking.\n");
3181 /* Initialize allocated space. */
3182 for (unsigned i
= prog
->NumUniformRemapTable
; i
< max_loc
+ 1; i
++)
3183 prog
->UniformRemapTable
[i
] = NULL
;
3185 prog
->NumUniformRemapTable
= max_loc
+ 1;
3188 for (unsigned i
= 0; i
< slots
; i
++) {
3189 unsigned loc
= var
->data
.location
+ i
;
3191 /* Check if location is already used. */
3192 if (prog
->UniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3194 /* Possibly same uniform from a different stage, this is ok. */
3196 if (map
->get(hash_loc
, var
->name
) && hash_loc
== loc
- i
) {
3201 /* ARB_explicit_uniform_location specification states:
3203 * "No two default-block uniform variables in the program can have
3204 * the same location, even if they are unused, otherwise a compiler
3205 * or linker error will be generated."
3208 "location qualifier for uniform %s overlaps "
3209 "previously used location\n",
3214 /* Initialize location as inactive before optimization
3215 * rounds and location assignment.
3217 prog
->UniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3220 /* Note, base location used for arrays. */
3221 map
->put(var
->data
.location
, var
->name
);
3223 return return_value
;
3227 reserve_subroutine_explicit_locations(struct gl_shader_program
*prog
,
3228 struct gl_linked_shader
*sh
,
3231 unsigned slots
= var
->type
->uniform_locations();
3232 unsigned max_loc
= var
->data
.location
+ slots
- 1;
3234 /* Resize remap table if locations do not fit in the current one. */
3235 if (max_loc
+ 1 > sh
->NumSubroutineUniformRemapTable
) {
3236 sh
->SubroutineUniformRemapTable
=
3237 reralloc(sh
, sh
->SubroutineUniformRemapTable
,
3238 gl_uniform_storage
*,
3241 if (!sh
->SubroutineUniformRemapTable
) {
3242 linker_error(prog
, "Out of memory during linking.\n");
3246 /* Initialize allocated space. */
3247 for (unsigned i
= sh
->NumSubroutineUniformRemapTable
; i
< max_loc
+ 1; i
++)
3248 sh
->SubroutineUniformRemapTable
[i
] = NULL
;
3250 sh
->NumSubroutineUniformRemapTable
= max_loc
+ 1;
3253 for (unsigned i
= 0; i
< slots
; i
++) {
3254 unsigned loc
= var
->data
.location
+ i
;
3256 /* Check if location is already used. */
3257 if (sh
->SubroutineUniformRemapTable
[loc
] == INACTIVE_UNIFORM_EXPLICIT_LOCATION
) {
3259 /* ARB_explicit_uniform_location specification states:
3260 * "No two subroutine uniform variables can have the same location
3261 * in the same shader stage, otherwise a compiler or linker error
3262 * will be generated."
3265 "location qualifier for uniform %s overlaps "
3266 "previously used location\n",
3271 /* Initialize location as inactive before optimization
3272 * rounds and location assignment.
3274 sh
->SubroutineUniformRemapTable
[loc
] = INACTIVE_UNIFORM_EXPLICIT_LOCATION
;
3280 * Check and reserve all explicit uniform locations, called before
3281 * any optimizations happen to handle also inactive uniforms and
3282 * inactive array elements that may get trimmed away.
3285 check_explicit_uniform_locations(struct gl_context
*ctx
,
3286 struct gl_shader_program
*prog
)
3288 if (!ctx
->Extensions
.ARB_explicit_uniform_location
)
3291 /* This map is used to detect if overlapping explicit locations
3292 * occur with the same uniform (from different stage) or a different one.
3294 string_to_uint_map
*uniform_map
= new string_to_uint_map
;
3297 linker_error(prog
, "Out of memory during linking.\n");
3301 unsigned entries_total
= 0;
3302 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3303 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
3308 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3309 ir_variable
*var
= node
->as_variable();
3310 if (!var
|| var
->data
.mode
!= ir_var_uniform
)
3313 if (var
->data
.explicit_location
) {
3315 if (var
->type
->without_array()->is_subroutine())
3316 ret
= reserve_subroutine_explicit_locations(prog
, sh
, var
);
3318 int slots
= reserve_explicit_locations(prog
, uniform_map
,
3322 entries_total
+= slots
;
3333 struct empty_uniform_block
*current_block
= NULL
;
3335 for (unsigned i
= 0; i
< prog
->NumUniformRemapTable
; i
++) {
3336 /* We found empty space in UniformRemapTable. */
3337 if (prog
->UniformRemapTable
[i
] == NULL
) {
3338 /* We've found the beginning of a new continous block of empty slots */
3339 if (!current_block
|| current_block
->start
+ current_block
->slots
!= i
) {
3340 current_block
= rzalloc(prog
, struct empty_uniform_block
);
3341 current_block
->start
= i
;
3342 exec_list_push_tail(&prog
->EmptyUniformLocations
,
3343 ¤t_block
->link
);
3346 /* The current block continues, so we simply increment its slots */
3347 current_block
->slots
++;
3352 return entries_total
;
3356 should_add_buffer_variable(struct gl_shader_program
*shProg
,
3357 GLenum type
, const char *name
)
3359 bool found_interface
= false;
3360 unsigned block_name_len
= 0;
3361 const char *block_name_dot
= strchr(name
, '.');
3363 /* These rules only apply to buffer variables. So we return
3364 * true for the rest of types.
3366 if (type
!= GL_BUFFER_VARIABLE
)
3369 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
3370 const char *block_name
= shProg
->ShaderStorageBlocks
[i
].Name
;
3371 block_name_len
= strlen(block_name
);
3373 const char *block_square_bracket
= strchr(block_name
, '[');
3374 if (block_square_bracket
) {
3375 /* The block is part of an array of named interfaces,
3376 * for the name comparison we ignore the "[x]" part.
3378 block_name_len
-= strlen(block_square_bracket
);
3381 if (block_name_dot
) {
3382 /* Check if the variable name starts with the interface
3383 * name. The interface name (if present) should have the
3384 * length than the interface block name we are comparing to.
3386 unsigned len
= strlen(name
) - strlen(block_name_dot
);
3387 if (len
!= block_name_len
)
3391 if (strncmp(block_name
, name
, block_name_len
) == 0) {
3392 found_interface
= true;
3397 /* We remove the interface name from the buffer variable name,
3398 * including the dot that follows it.
3400 if (found_interface
)
3401 name
= name
+ block_name_len
+ 1;
3403 /* The ARB_program_interface_query spec says:
3405 * "For an active shader storage block member declared as an array, an
3406 * entry will be generated only for the first array element, regardless
3407 * of its type. For arrays of aggregate types, the enumeration rules
3408 * are applied recursively for the single enumerated array element."
3410 const char *struct_first_dot
= strchr(name
, '.');
3411 const char *first_square_bracket
= strchr(name
, '[');
3413 /* The buffer variable is on top level and it is not an array */
3414 if (!first_square_bracket
) {
3416 /* The shader storage block member is a struct, then generate the entry */
3417 } else if (struct_first_dot
&& struct_first_dot
< first_square_bracket
) {
3420 /* Shader storage block member is an array, only generate an entry for the
3421 * first array element.
3423 if (strncmp(first_square_bracket
, "[0]", 3) == 0)
3431 add_program_resource(struct gl_shader_program
*prog
,
3432 struct set
*resource_set
,
3433 GLenum type
, const void *data
, uint8_t stages
)
3437 /* If resource already exists, do not add it again. */
3438 if (_mesa_set_search(resource_set
, data
))
3441 prog
->ProgramResourceList
=
3443 prog
->ProgramResourceList
,
3444 gl_program_resource
,
3445 prog
->NumProgramResourceList
+ 1);
3447 if (!prog
->ProgramResourceList
) {
3448 linker_error(prog
, "Out of memory during linking.\n");
3452 struct gl_program_resource
*res
=
3453 &prog
->ProgramResourceList
[prog
->NumProgramResourceList
];
3457 res
->StageReferences
= stages
;
3459 prog
->NumProgramResourceList
++;
3461 _mesa_set_add(resource_set
, data
);
3466 /* Function checks if a variable var is a packed varying and
3467 * if given name is part of packed varying's list.
3469 * If a variable is a packed varying, it has a name like
3470 * 'packed:a,b,c' where a, b and c are separate variables.
3473 included_in_packed_varying(ir_variable
*var
, const char *name
)
3475 if (strncmp(var
->name
, "packed:", 7) != 0)
3478 char *list
= strdup(var
->name
+ 7);
3483 char *token
= strtok_r(list
, ",", &saveptr
);
3485 if (strcmp(token
, name
) == 0) {
3489 token
= strtok_r(NULL
, ",", &saveptr
);
3496 * Function builds a stage reference bitmask from variable name.
3499 build_stageref(struct gl_shader_program
*shProg
, const char *name
,
3504 /* Note, that we assume MAX 8 stages, if there will be more stages, type
3505 * used for reference mask in gl_program_resource will need to be changed.
3507 assert(MESA_SHADER_STAGES
< 8);
3509 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
3510 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
3514 /* Shader symbol table may contain variables that have
3515 * been optimized away. Search IR for the variable instead.
3517 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
3518 ir_variable
*var
= node
->as_variable();
3520 unsigned baselen
= strlen(var
->name
);
3522 if (included_in_packed_varying(var
, name
)) {
3527 /* Type needs to match if specified, otherwise we might
3528 * pick a variable with same name but different interface.
3530 if (var
->data
.mode
!= mode
)
3533 if (strncmp(var
->name
, name
, baselen
) == 0) {
3534 /* Check for exact name matches but also check for arrays and
3537 if (name
[baselen
] == '\0' ||
3538 name
[baselen
] == '[' ||
3539 name
[baselen
] == '.') {
3551 * Create gl_shader_variable from ir_variable class.
3553 static gl_shader_variable
*
3554 create_shader_variable(struct gl_shader_program
*shProg
,
3555 const ir_variable
*in
,
3556 const char *name
, const glsl_type
*type
,
3557 bool use_implicit_location
, int location
,
3558 const glsl_type
*outermost_struct_type
)
3560 gl_shader_variable
*out
= ralloc(shProg
, struct gl_shader_variable
);
3564 /* Since gl_VertexID may be lowered to gl_VertexIDMESA, but applications
3565 * expect to see gl_VertexID in the program resource list. Pretend.
3567 if (in
->data
.mode
== ir_var_system_value
&&
3568 in
->data
.location
== SYSTEM_VALUE_VERTEX_ID_ZERO_BASE
) {
3569 out
->name
= ralloc_strdup(shProg
, "gl_VertexID");
3570 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3571 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_OUTER
) ||
3572 (in
->data
.mode
== ir_var_system_value
&&
3573 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_OUTER
)) {
3574 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelOuter");
3575 type
= glsl_type::get_array_instance(glsl_type::float_type
, 4);
3576 } else if ((in
->data
.mode
== ir_var_shader_out
&&
3577 in
->data
.location
== VARYING_SLOT_TESS_LEVEL_INNER
) ||
3578 (in
->data
.mode
== ir_var_system_value
&&
3579 in
->data
.location
== SYSTEM_VALUE_TESS_LEVEL_INNER
)) {
3580 out
->name
= ralloc_strdup(shProg
, "gl_TessLevelInner");
3581 type
= glsl_type::get_array_instance(glsl_type::float_type
, 2);
3583 out
->name
= ralloc_strdup(shProg
, name
);
3589 /* The ARB_program_interface_query spec says:
3591 * "Not all active variables are assigned valid locations; the
3592 * following variables will have an effective location of -1:
3594 * * uniforms declared as atomic counters;
3596 * * members of a uniform block;
3598 * * built-in inputs, outputs, and uniforms (starting with "gl_"); and
3600 * * inputs or outputs not declared with a "location" layout
3601 * qualifier, except for vertex shader inputs and fragment shader
3604 if (in
->type
->base_type
== GLSL_TYPE_ATOMIC_UINT
||
3605 is_gl_identifier(in
->name
) ||
3606 !(in
->data
.explicit_location
|| use_implicit_location
)) {
3609 out
->location
= location
;
3613 out
->outermost_struct_type
= outermost_struct_type
;
3614 out
->interface_type
= in
->get_interface_type();
3615 out
->component
= in
->data
.location_frac
;
3616 out
->index
= in
->data
.index
;
3617 out
->patch
= in
->data
.patch
;
3618 out
->mode
= in
->data
.mode
;
3619 out
->interpolation
= in
->data
.interpolation
;
3620 out
->explicit_location
= in
->data
.explicit_location
;
3621 out
->precision
= in
->data
.precision
;
3627 add_shader_variable(struct gl_shader_program
*shProg
, struct set
*resource_set
,
3628 unsigned stage_mask
,
3629 GLenum programInterface
, ir_variable
*var
,
3630 const char *name
, const glsl_type
*type
,
3631 bool use_implicit_location
, int location
,
3632 const glsl_type
*outermost_struct_type
= NULL
)
3634 const bool is_vertex_input
=
3635 programInterface
== GL_PROGRAM_INPUT
&&
3636 stage_mask
== MESA_SHADER_VERTEX
;
3638 switch (type
->base_type
) {
3639 case GLSL_TYPE_STRUCT
: {
3640 /* The ARB_program_interface_query spec says:
3642 * "For an active variable declared as a structure, a separate entry
3643 * will be generated for each active structure member. The name of
3644 * each entry is formed by concatenating the name of the structure,
3645 * the "." character, and the name of the structure member. If a
3646 * structure member to enumerate is itself a structure or array,
3647 * these enumeration rules are applied recursively."
3649 if (outermost_struct_type
== NULL
)
3650 outermost_struct_type
= type
;
3652 unsigned field_location
= location
;
3653 for (unsigned i
= 0; i
< type
->length
; i
++) {
3654 const struct glsl_struct_field
*field
= &type
->fields
.structure
[i
];
3655 char *field_name
= ralloc_asprintf(shProg
, "%s.%s", name
, field
->name
);
3656 if (!add_shader_variable(shProg
, resource_set
,
3657 stage_mask
, programInterface
,
3658 var
, field_name
, field
->type
,
3659 use_implicit_location
, field_location
,
3660 outermost_struct_type
))
3664 field
->type
->count_attribute_slots(is_vertex_input
);
3670 /* Issue #16 of the ARB_program_interface_query spec says:
3672 * "* If a variable is a member of an interface block without an
3673 * instance name, it is enumerated using just the variable name.
3675 * * If a variable is a member of an interface block with an instance
3676 * name, it is enumerated as "BlockName.Member", where "BlockName" is
3677 * the name of the interface block (not the instance name) and
3678 * "Member" is the name of the variable."
3680 const char *prefixed_name
= (var
->data
.from_named_ifc_block
&&
3681 !is_gl_identifier(var
->name
))
3682 ? ralloc_asprintf(shProg
, "%s.%s", var
->get_interface_type()->name
,
3686 /* The ARB_program_interface_query spec says:
3688 * "For an active variable declared as a single instance of a basic
3689 * type, a single entry will be generated, using the variable name
3690 * from the shader source."
3692 gl_shader_variable
*sha_v
=
3693 create_shader_variable(shProg
, var
, prefixed_name
, type
,
3694 use_implicit_location
, location
,
3695 outermost_struct_type
);
3699 return add_program_resource(shProg
, resource_set
,
3700 programInterface
, sha_v
, stage_mask
);
3706 add_interface_variables(struct gl_shader_program
*shProg
,
3707 struct set
*resource_set
,
3708 unsigned stage
, GLenum programInterface
)
3710 exec_list
*ir
= shProg
->_LinkedShaders
[stage
]->ir
;
3712 foreach_in_list(ir_instruction
, node
, ir
) {
3713 ir_variable
*var
= node
->as_variable();
3715 if (!var
|| var
->data
.how_declared
== ir_var_hidden
)
3720 switch (var
->data
.mode
) {
3721 case ir_var_system_value
:
3722 case ir_var_shader_in
:
3723 if (programInterface
!= GL_PROGRAM_INPUT
)
3725 loc_bias
= (stage
== MESA_SHADER_VERTEX
) ? int(VERT_ATTRIB_GENERIC0
)
3726 : int(VARYING_SLOT_VAR0
);
3728 case ir_var_shader_out
:
3729 if (programInterface
!= GL_PROGRAM_OUTPUT
)
3731 loc_bias
= (stage
== MESA_SHADER_FRAGMENT
) ? int(FRAG_RESULT_DATA0
)
3732 : int(VARYING_SLOT_VAR0
);
3738 if (var
->data
.patch
)
3739 loc_bias
= int(VARYING_SLOT_PATCH0
);
3741 /* Skip packed varyings, packed varyings are handled separately
3742 * by add_packed_varyings.
3744 if (strncmp(var
->name
, "packed:", 7) == 0)
3747 /* Skip fragdata arrays, these are handled separately
3748 * by add_fragdata_arrays.
3750 if (strncmp(var
->name
, "gl_out_FragData", 15) == 0)
3753 const bool vs_input_or_fs_output
=
3754 (stage
== MESA_SHADER_VERTEX
&& var
->data
.mode
== ir_var_shader_in
) ||
3755 (stage
== MESA_SHADER_FRAGMENT
&& var
->data
.mode
== ir_var_shader_out
);
3757 if (!add_shader_variable(shProg
, resource_set
,
3758 1 << stage
, programInterface
,
3759 var
, var
->name
, var
->type
, vs_input_or_fs_output
,
3760 var
->data
.location
- loc_bias
))
3767 add_packed_varyings(struct gl_shader_program
*shProg
, struct set
*resource_set
,
3768 int stage
, GLenum type
)
3770 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[stage
];
3773 if (!sh
|| !sh
->packed_varyings
)
3776 foreach_in_list(ir_instruction
, node
, sh
->packed_varyings
) {
3777 ir_variable
*var
= node
->as_variable();
3779 switch (var
->data
.mode
) {
3780 case ir_var_shader_in
:
3781 iface
= GL_PROGRAM_INPUT
;
3783 case ir_var_shader_out
:
3784 iface
= GL_PROGRAM_OUTPUT
;
3787 unreachable("unexpected type");
3790 if (type
== iface
) {
3791 const int stage_mask
=
3792 build_stageref(shProg
, var
->name
, var
->data
.mode
);
3793 if (!add_shader_variable(shProg
, resource_set
,
3795 iface
, var
, var
->name
, var
->type
, false,
3796 var
->data
.location
- VARYING_SLOT_VAR0
))
3805 add_fragdata_arrays(struct gl_shader_program
*shProg
, struct set
*resource_set
)
3807 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
3809 if (!sh
|| !sh
->fragdata_arrays
)
3812 foreach_in_list(ir_instruction
, node
, sh
->fragdata_arrays
) {
3813 ir_variable
*var
= node
->as_variable();
3815 assert(var
->data
.mode
== ir_var_shader_out
);
3817 if (!add_shader_variable(shProg
, resource_set
,
3818 1 << MESA_SHADER_FRAGMENT
,
3819 GL_PROGRAM_OUTPUT
, var
, var
->name
, var
->type
,
3820 true, var
->data
.location
- FRAG_RESULT_DATA0
))
3828 get_top_level_name(const char *name
)
3830 const char *first_dot
= strchr(name
, '.');
3831 const char *first_square_bracket
= strchr(name
, '[');
3834 /* The ARB_program_interface_query spec says:
3836 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3837 * the number of active array elements of the top-level shader storage
3838 * block member containing to the active variable is written to
3839 * <params>. If the top-level block member is not declared as an
3840 * array, the value one is written to <params>. If the top-level block
3841 * member is an array with no declared size, the value zero is written
3845 /* The buffer variable is on top level.*/
3846 if (!first_square_bracket
&& !first_dot
)
3847 name_size
= strlen(name
);
3848 else if ((!first_square_bracket
||
3849 (first_dot
&& first_dot
< first_square_bracket
)))
3850 name_size
= first_dot
- name
;
3852 name_size
= first_square_bracket
- name
;
3854 return strndup(name
, name_size
);
3858 get_var_name(const char *name
)
3860 const char *first_dot
= strchr(name
, '.');
3863 return strdup(name
);
3865 return strndup(first_dot
+1, strlen(first_dot
) - 1);
3869 is_top_level_shader_storage_block_member(const char* name
,
3870 const char* interface_name
,
3871 const char* field_name
)
3873 bool result
= false;
3875 /* If the given variable is already a top-level shader storage
3876 * block member, then return array_size = 1.
3877 * We could have two possibilities: if we have an instanced
3878 * shader storage block or not instanced.
3880 * For the first, we check create a name as it was in top level and
3881 * compare it with the real name. If they are the same, then
3882 * the variable is already at top-level.
3884 * Full instanced name is: interface name + '.' + var name +
3887 int name_length
= strlen(interface_name
) + 1 + strlen(field_name
) + 1;
3888 char *full_instanced_name
= (char *) calloc(name_length
, sizeof(char));
3889 if (!full_instanced_name
) {
3890 fprintf(stderr
, "%s: Cannot allocate space for name\n", __func__
);
3894 snprintf(full_instanced_name
, name_length
, "%s.%s",
3895 interface_name
, field_name
);
3897 /* Check if its top-level shader storage block member of an
3898 * instanced interface block, or of a unnamed interface block.
3900 if (strcmp(name
, full_instanced_name
) == 0 ||
3901 strcmp(name
, field_name
) == 0)
3904 free(full_instanced_name
);
3909 get_array_size(struct gl_uniform_storage
*uni
, const glsl_struct_field
*field
,
3910 char *interface_name
, char *var_name
)
3912 /* The ARB_program_interface_query spec says:
3914 * "For the property TOP_LEVEL_ARRAY_SIZE, a single integer identifying
3915 * the number of active array elements of the top-level shader storage
3916 * block member containing to the active variable is written to
3917 * <params>. If the top-level block member is not declared as an
3918 * array, the value one is written to <params>. If the top-level block
3919 * member is an array with no declared size, the value zero is written
3922 if (is_top_level_shader_storage_block_member(uni
->name
,
3926 else if (field
->type
->is_unsized_array())
3928 else if (field
->type
->is_array())
3929 return field
->type
->length
;
3935 get_array_stride(struct gl_uniform_storage
*uni
, const glsl_type
*interface
,
3936 const glsl_struct_field
*field
, char *interface_name
,
3939 /* The ARB_program_interface_query spec says:
3941 * "For the property TOP_LEVEL_ARRAY_STRIDE, a single integer
3942 * identifying the stride between array elements of the top-level
3943 * shader storage block member containing the active variable is
3944 * written to <params>. For top-level block members declared as
3945 * arrays, the value written is the difference, in basic machine units,
3946 * between the offsets of the active variable for consecutive elements
3947 * in the top-level array. For top-level block members not declared as
3948 * an array, zero is written to <params>."
3950 if (field
->type
->is_array()) {
3951 const enum glsl_matrix_layout matrix_layout
=
3952 glsl_matrix_layout(field
->matrix_layout
);
3953 bool row_major
= matrix_layout
== GLSL_MATRIX_LAYOUT_ROW_MAJOR
;
3954 const glsl_type
*array_type
= field
->type
->fields
.array
;
3956 if (is_top_level_shader_storage_block_member(uni
->name
,
3961 if (interface
->interface_packing
!= GLSL_INTERFACE_PACKING_STD430
) {
3962 if (array_type
->is_record() || array_type
->is_array())
3963 return glsl_align(array_type
->std140_size(row_major
), 16);
3965 return MAX2(array_type
->std140_base_alignment(row_major
), 16);
3967 return array_type
->std430_array_stride(row_major
);
3974 calculate_array_size_and_stride(struct gl_shader_program
*shProg
,
3975 struct gl_uniform_storage
*uni
)
3977 int block_index
= uni
->block_index
;
3978 int array_size
= -1;
3979 int array_stride
= -1;
3980 char *var_name
= get_top_level_name(uni
->name
);
3981 char *interface_name
=
3982 get_top_level_name(uni
->is_shader_storage
?
3983 shProg
->ShaderStorageBlocks
[block_index
].Name
:
3984 shProg
->UniformBlocks
[block_index
].Name
);
3986 if (strcmp(var_name
, interface_name
) == 0) {
3987 /* Deal with instanced array of SSBOs */
3988 char *temp_name
= get_var_name(uni
->name
);
3990 linker_error(shProg
, "Out of memory during linking.\n");
3991 goto write_top_level_array_size_and_stride
;
3994 var_name
= get_top_level_name(temp_name
);
3997 linker_error(shProg
, "Out of memory during linking.\n");
3998 goto write_top_level_array_size_and_stride
;
4002 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4003 const gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4007 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4008 ir_variable
*var
= node
->as_variable();
4009 if (!var
|| !var
->get_interface_type() ||
4010 var
->data
.mode
!= ir_var_shader_storage
)
4013 const glsl_type
*interface
= var
->get_interface_type();
4015 if (strcmp(interface_name
, interface
->name
) != 0)
4018 for (unsigned i
= 0; i
< interface
->length
; i
++) {
4019 const glsl_struct_field
*field
= &interface
->fields
.structure
[i
];
4020 if (strcmp(field
->name
, var_name
) != 0)
4023 array_stride
= get_array_stride(uni
, interface
, field
,
4024 interface_name
, var_name
);
4025 array_size
= get_array_size(uni
, field
, interface_name
, var_name
);
4026 goto write_top_level_array_size_and_stride
;
4030 write_top_level_array_size_and_stride
:
4031 free(interface_name
);
4033 uni
->top_level_array_stride
= array_stride
;
4034 uni
->top_level_array_size
= array_size
;
4038 * Builds up a list of program resources that point to existing
4042 build_program_resource_list(struct gl_context
*ctx
,
4043 struct gl_shader_program
*shProg
)
4045 /* Rebuild resource list. */
4046 if (shProg
->ProgramResourceList
) {
4047 ralloc_free(shProg
->ProgramResourceList
);
4048 shProg
->ProgramResourceList
= NULL
;
4049 shProg
->NumProgramResourceList
= 0;
4052 int input_stage
= MESA_SHADER_STAGES
, output_stage
= 0;
4054 /* Determine first input and final output stage. These are used to
4055 * detect which variables should be enumerated in the resource list
4056 * for GL_PROGRAM_INPUT and GL_PROGRAM_OUTPUT.
4058 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4059 if (!shProg
->_LinkedShaders
[i
])
4061 if (input_stage
== MESA_SHADER_STAGES
)
4066 /* Empty shader, no resources. */
4067 if (input_stage
== MESA_SHADER_STAGES
&& output_stage
== 0)
4070 struct set
*resource_set
= _mesa_set_create(NULL
,
4072 _mesa_key_pointer_equal
);
4074 /* Program interface needs to expose varyings in case of SSO. */
4075 if (shProg
->SeparateShader
) {
4076 if (!add_packed_varyings(shProg
, resource_set
,
4077 input_stage
, GL_PROGRAM_INPUT
))
4080 if (!add_packed_varyings(shProg
, resource_set
,
4081 output_stage
, GL_PROGRAM_OUTPUT
))
4085 if (!add_fragdata_arrays(shProg
, resource_set
))
4088 /* Add inputs and outputs to the resource list. */
4089 if (!add_interface_variables(shProg
, resource_set
,
4090 input_stage
, GL_PROGRAM_INPUT
))
4093 if (!add_interface_variables(shProg
, resource_set
,
4094 output_stage
, GL_PROGRAM_OUTPUT
))
4097 /* Add transform feedback varyings. */
4098 if (shProg
->LinkedTransformFeedback
.NumVarying
> 0) {
4099 for (int i
= 0; i
< shProg
->LinkedTransformFeedback
.NumVarying
; i
++) {
4100 if (!add_program_resource(shProg
, resource_set
,
4101 GL_TRANSFORM_FEEDBACK_VARYING
,
4102 &shProg
->LinkedTransformFeedback
.Varyings
[i
],
4108 /* Add transform feedback buffers. */
4109 for (unsigned i
= 0; i
< ctx
->Const
.MaxTransformFeedbackBuffers
; i
++) {
4110 if ((shProg
->LinkedTransformFeedback
.ActiveBuffers
>> i
) & 1) {
4111 shProg
->LinkedTransformFeedback
.Buffers
[i
].Binding
= i
;
4112 if (!add_program_resource(shProg
, resource_set
,
4113 GL_TRANSFORM_FEEDBACK_BUFFER
,
4114 &shProg
->LinkedTransformFeedback
.Buffers
[i
],
4120 /* Add uniforms from uniform storage. */
4121 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4122 /* Do not add uniforms internally used by Mesa. */
4123 if (shProg
->UniformStorage
[i
].hidden
)
4127 build_stageref(shProg
, shProg
->UniformStorage
[i
].name
,
4130 /* Add stagereferences for uniforms in a uniform block. */
4131 bool is_shader_storage
= shProg
->UniformStorage
[i
].is_shader_storage
;
4132 int block_index
= shProg
->UniformStorage
[i
].block_index
;
4133 if (block_index
!= -1) {
4134 stageref
|= is_shader_storage
?
4135 shProg
->ShaderStorageBlocks
[block_index
].stageref
:
4136 shProg
->UniformBlocks
[block_index
].stageref
;
4139 GLenum type
= is_shader_storage
? GL_BUFFER_VARIABLE
: GL_UNIFORM
;
4140 if (!should_add_buffer_variable(shProg
, type
,
4141 shProg
->UniformStorage
[i
].name
))
4144 if (is_shader_storage
) {
4145 calculate_array_size_and_stride(shProg
, &shProg
->UniformStorage
[i
]);
4148 if (!add_program_resource(shProg
, resource_set
, type
,
4149 &shProg
->UniformStorage
[i
], stageref
))
4153 /* Add program uniform blocks. */
4154 for (unsigned i
= 0; i
< shProg
->NumUniformBlocks
; i
++) {
4155 if (!add_program_resource(shProg
, resource_set
, GL_UNIFORM_BLOCK
,
4156 &shProg
->UniformBlocks
[i
], 0))
4160 /* Add program shader storage blocks. */
4161 for (unsigned i
= 0; i
< shProg
->NumShaderStorageBlocks
; i
++) {
4162 if (!add_program_resource(shProg
, resource_set
, GL_SHADER_STORAGE_BLOCK
,
4163 &shProg
->ShaderStorageBlocks
[i
], 0))
4167 /* Add atomic counter buffers. */
4168 for (unsigned i
= 0; i
< shProg
->NumAtomicBuffers
; i
++) {
4169 if (!add_program_resource(shProg
, resource_set
, GL_ATOMIC_COUNTER_BUFFER
,
4170 &shProg
->AtomicBuffers
[i
], 0))
4174 for (unsigned i
= 0; i
< shProg
->NumUniformStorage
; i
++) {
4176 if (!shProg
->UniformStorage
[i
].hidden
)
4179 for (int j
= MESA_SHADER_VERTEX
; j
< MESA_SHADER_STAGES
; j
++) {
4180 if (!shProg
->UniformStorage
[i
].opaque
[j
].active
||
4181 !shProg
->UniformStorage
[i
].type
->is_subroutine())
4184 type
= _mesa_shader_stage_to_subroutine_uniform((gl_shader_stage
)j
);
4185 /* add shader subroutines */
4186 if (!add_program_resource(shProg
, resource_set
,
4187 type
, &shProg
->UniformStorage
[i
], 0))
4192 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4193 struct gl_linked_shader
*sh
= shProg
->_LinkedShaders
[i
];
4199 type
= _mesa_shader_stage_to_subroutine((gl_shader_stage
)i
);
4200 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4201 if (!add_program_resource(shProg
, resource_set
,
4202 type
, &sh
->SubroutineFunctions
[j
], 0))
4207 _mesa_set_destroy(resource_set
, NULL
);
4211 * This check is done to make sure we allow only constant expression
4212 * indexing and "constant-index-expression" (indexing with an expression
4213 * that includes loop induction variable).
4216 validate_sampler_array_indexing(struct gl_context
*ctx
,
4217 struct gl_shader_program
*prog
)
4219 dynamic_sampler_array_indexing_visitor v
;
4220 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4221 if (prog
->_LinkedShaders
[i
] == NULL
)
4224 bool no_dynamic_indexing
=
4225 ctx
->Const
.ShaderCompilerOptions
[i
].EmitNoIndirectSampler
;
4227 /* Search for array derefs in shader. */
4228 v
.run(prog
->_LinkedShaders
[i
]->ir
);
4229 if (v
.uses_dynamic_sampler_array_indexing()) {
4230 const char *msg
= "sampler arrays indexed with non-constant "
4231 "expressions is forbidden in GLSL %s %u";
4232 /* Backend has indicated that it has no dynamic indexing support. */
4233 if (no_dynamic_indexing
) {
4234 linker_error(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4237 linker_warning(prog
, msg
, prog
->IsES
? "ES" : "", prog
->Version
);
4245 link_assign_subroutine_types(struct gl_shader_program
*prog
)
4247 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4248 gl_linked_shader
*sh
= prog
->_LinkedShaders
[i
];
4253 sh
->MaxSubroutineFunctionIndex
= 0;
4254 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
4255 ir_function
*fn
= node
->as_function();
4259 if (fn
->is_subroutine
)
4260 sh
->NumSubroutineUniformTypes
++;
4262 if (!fn
->num_subroutine_types
)
4265 /* these should have been calculated earlier. */
4266 assert(fn
->subroutine_index
!= -1);
4267 if (sh
->NumSubroutineFunctions
+ 1 > MAX_SUBROUTINES
) {
4268 linker_error(prog
, "Too many subroutine functions declared.\n");
4271 sh
->SubroutineFunctions
= reralloc(sh
, sh
->SubroutineFunctions
,
4272 struct gl_subroutine_function
,
4273 sh
->NumSubroutineFunctions
+ 1);
4274 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].name
= ralloc_strdup(sh
, fn
->name
);
4275 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].num_compat_types
= fn
->num_subroutine_types
;
4276 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
=
4277 ralloc_array(sh
, const struct glsl_type
*,
4278 fn
->num_subroutine_types
);
4280 /* From Section 4.4.4(Subroutine Function Layout Qualifiers) of the
4283 * "Each subroutine with an index qualifier in the shader must be
4284 * given a unique index, otherwise a compile or link error will be
4287 for (unsigned j
= 0; j
< sh
->NumSubroutineFunctions
; j
++) {
4288 if (sh
->SubroutineFunctions
[j
].index
!= -1 &&
4289 sh
->SubroutineFunctions
[j
].index
== fn
->subroutine_index
) {
4290 linker_error(prog
, "each subroutine index qualifier in the "
4291 "shader must be unique\n");
4295 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].index
=
4296 fn
->subroutine_index
;
4298 if (fn
->subroutine_index
> (int)sh
->MaxSubroutineFunctionIndex
)
4299 sh
->MaxSubroutineFunctionIndex
= fn
->subroutine_index
;
4301 for (int j
= 0; j
< fn
->num_subroutine_types
; j
++)
4302 sh
->SubroutineFunctions
[sh
->NumSubroutineFunctions
].types
[j
] = fn
->subroutine_types
[j
];
4303 sh
->NumSubroutineFunctions
++;
4309 set_always_active_io(exec_list
*ir
, ir_variable_mode io_mode
)
4311 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
4313 foreach_in_list(ir_instruction
, node
, ir
) {
4314 ir_variable
*const var
= node
->as_variable();
4316 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
4319 /* Don't set always active on builtins that haven't been redeclared */
4320 if (var
->data
.how_declared
== ir_var_declared_implicitly
)
4323 var
->data
.always_active_io
= true;
4328 * When separate shader programs are enabled, only input/outputs between
4329 * the stages of a multi-stage separate program can be safely removed
4330 * from the shader interface. Other inputs/outputs must remain active.
4333 disable_varying_optimizations_for_sso(struct gl_shader_program
*prog
)
4335 unsigned first
, last
;
4336 assert(prog
->SeparateShader
);
4338 first
= MESA_SHADER_STAGES
;
4341 /* Determine first and last stage. Excluding the compute stage */
4342 for (unsigned i
= 0; i
< MESA_SHADER_COMPUTE
; i
++) {
4343 if (!prog
->_LinkedShaders
[i
])
4345 if (first
== MESA_SHADER_STAGES
)
4350 if (first
== MESA_SHADER_STAGES
)
4353 for (unsigned stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4354 gl_linked_shader
*sh
= prog
->_LinkedShaders
[stage
];
4358 if (first
== last
) {
4359 /* For a single shader program only allow inputs to the vertex shader
4360 * and outputs from the fragment shader to be removed.
4362 if (stage
!= MESA_SHADER_VERTEX
)
4363 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4364 if (stage
!= MESA_SHADER_FRAGMENT
)
4365 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4367 /* For multi-stage separate shader programs only allow inputs and
4368 * outputs between the shader stages to be removed as well as inputs
4369 * to the vertex shader and outputs from the fragment shader.
4371 if (stage
== first
&& stage
!= MESA_SHADER_VERTEX
)
4372 set_always_active_io(sh
->ir
, ir_var_shader_in
);
4373 else if (stage
== last
&& stage
!= MESA_SHADER_FRAGMENT
)
4374 set_always_active_io(sh
->ir
, ir_var_shader_out
);
4380 link_varyings_and_uniforms(unsigned first
, unsigned last
,
4381 unsigned num_explicit_uniform_locs
,
4382 struct gl_context
*ctx
,
4383 struct gl_shader_program
*prog
, void *mem_ctx
)
4385 bool has_xfb_qualifiers
= false;
4386 unsigned num_tfeedback_decls
= 0;
4387 char **varying_names
= NULL
;
4388 tfeedback_decl
*tfeedback_decls
= NULL
;
4390 /* Mark all generic shader inputs and outputs as unpaired. */
4391 for (unsigned i
= MESA_SHADER_VERTEX
; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4392 if (prog
->_LinkedShaders
[i
] != NULL
) {
4393 link_invalidate_variable_locations(prog
->_LinkedShaders
[i
]->ir
);
4397 unsigned prev
= first
;
4398 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4399 if (prog
->_LinkedShaders
[i
] == NULL
)
4402 match_explicit_outputs_to_inputs(prog
->_LinkedShaders
[prev
],
4403 prog
->_LinkedShaders
[i
]);
4407 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4408 MESA_SHADER_VERTEX
)) {
4412 if (!assign_attribute_or_color_locations(mem_ctx
, prog
, &ctx
->Const
,
4413 MESA_SHADER_FRAGMENT
)) {
4417 /* From the ARB_enhanced_layouts spec:
4419 * "If the shader used to record output variables for transform feedback
4420 * varyings uses the "xfb_buffer", "xfb_offset", or "xfb_stride" layout
4421 * qualifiers, the values specified by TransformFeedbackVaryings are
4422 * ignored, and the set of variables captured for transform feedback is
4423 * instead derived from the specified layout qualifiers."
4425 for (int i
= MESA_SHADER_FRAGMENT
- 1; i
>= 0; i
--) {
4426 /* Find last stage before fragment shader */
4427 if (prog
->_LinkedShaders
[i
]) {
4428 has_xfb_qualifiers
=
4429 process_xfb_layout_qualifiers(mem_ctx
, prog
->_LinkedShaders
[i
],
4430 &num_tfeedback_decls
,
4436 if (!has_xfb_qualifiers
) {
4437 num_tfeedback_decls
= prog
->TransformFeedback
.NumVarying
;
4438 varying_names
= prog
->TransformFeedback
.VaryingNames
;
4441 if (num_tfeedback_decls
!= 0) {
4442 /* From GL_EXT_transform_feedback:
4443 * A program will fail to link if:
4445 * * the <count> specified by TransformFeedbackVaryingsEXT is
4446 * non-zero, but the program object has no vertex or geometry
4449 if (first
>= MESA_SHADER_FRAGMENT
) {
4450 linker_error(prog
, "Transform feedback varyings specified, but "
4451 "no vertex, tessellation, or geometry shader is "
4456 tfeedback_decls
= ralloc_array(mem_ctx
, tfeedback_decl
,
4457 num_tfeedback_decls
);
4458 if (!parse_tfeedback_decls(ctx
, prog
, mem_ctx
, num_tfeedback_decls
,
4459 varying_names
, tfeedback_decls
))
4463 /* If there is no fragment shader we need to set transform feedback.
4465 * For SSO we also need to assign output locations. We assign them here
4466 * because we need to do it for both single stage programs and multi stage
4469 if (last
< MESA_SHADER_FRAGMENT
&&
4470 (num_tfeedback_decls
!= 0 || prog
->SeparateShader
)) {
4471 const uint64_t reserved_out_slots
=
4472 reserved_varying_slot(prog
->_LinkedShaders
[last
], ir_var_shader_out
);
4473 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4474 prog
->_LinkedShaders
[last
], NULL
,
4475 num_tfeedback_decls
, tfeedback_decls
,
4476 reserved_out_slots
))
4480 if (last
<= MESA_SHADER_FRAGMENT
) {
4481 /* Remove unused varyings from the first/last stage unless SSO */
4482 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4483 prog
->_LinkedShaders
[first
],
4485 remove_unused_shader_inputs_and_outputs(prog
->SeparateShader
,
4486 prog
->_LinkedShaders
[last
],
4489 /* If the program is made up of only a single stage */
4490 if (first
== last
) {
4491 gl_linked_shader
*const sh
= prog
->_LinkedShaders
[last
];
4493 do_dead_builtin_varyings(ctx
, NULL
, sh
, 0, NULL
);
4494 do_dead_builtin_varyings(ctx
, sh
, NULL
, num_tfeedback_decls
,
4497 if (prog
->SeparateShader
) {
4498 const uint64_t reserved_slots
=
4499 reserved_varying_slot(sh
, ir_var_shader_in
);
4501 /* Assign input locations for SSO, output locations are already
4504 if (!assign_varying_locations(ctx
, mem_ctx
, prog
,
4505 NULL
/* producer */,
4507 0 /* num_tfeedback_decls */,
4508 NULL
/* tfeedback_decls */,
4513 /* Linking the stages in the opposite order (from fragment to vertex)
4514 * ensures that inter-shader outputs written to in an earlier stage
4515 * are eliminated if they are (transitively) not used in a later
4519 for (int i
= next
- 1; i
>= 0; i
--) {
4520 if (prog
->_LinkedShaders
[i
] == NULL
&& i
!= 0)
4523 gl_linked_shader
*const sh_i
= prog
->_LinkedShaders
[i
];
4524 gl_linked_shader
*const sh_next
= prog
->_LinkedShaders
[next
];
4526 const uint64_t reserved_out_slots
=
4527 reserved_varying_slot(sh_i
, ir_var_shader_out
);
4528 const uint64_t reserved_in_slots
=
4529 reserved_varying_slot(sh_next
, ir_var_shader_in
);
4531 do_dead_builtin_varyings(ctx
, sh_i
, sh_next
,
4532 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4535 if (!assign_varying_locations(ctx
, mem_ctx
, prog
, sh_i
, sh_next
,
4536 next
== MESA_SHADER_FRAGMENT
? num_tfeedback_decls
: 0,
4538 reserved_out_slots
| reserved_in_slots
))
4541 /* This must be done after all dead varyings are eliminated. */
4543 unsigned slots_used
= _mesa_bitcount_64(reserved_out_slots
);
4544 if (!check_against_output_limit(ctx
, prog
, sh_i
, slots_used
)) {
4549 unsigned slots_used
= _mesa_bitcount_64(reserved_in_slots
);
4550 if (!check_against_input_limit(ctx
, prog
, sh_next
, slots_used
))
4558 if (!store_tfeedback_info(ctx
, prog
, num_tfeedback_decls
, tfeedback_decls
,
4559 has_xfb_qualifiers
))
4562 update_array_sizes(prog
);
4563 link_assign_uniform_locations(prog
, ctx
->Const
.UniformBooleanTrue
,
4564 num_explicit_uniform_locs
,
4565 ctx
->Const
.MaxUserAssignableUniformLocations
);
4566 link_assign_atomic_counter_resources(ctx
, prog
);
4568 link_calculate_subroutine_compat(prog
);
4569 check_resources(ctx
, prog
);
4570 check_subroutine_resources(prog
);
4571 check_image_resources(ctx
, prog
);
4572 link_check_atomic_counter_resources(ctx
, prog
);
4574 if (!prog
->LinkStatus
)
4577 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4578 if (prog
->_LinkedShaders
[i
] == NULL
)
4581 const struct gl_shader_compiler_options
*options
=
4582 &ctx
->Const
.ShaderCompilerOptions
[i
];
4584 if (options
->LowerBufferInterfaceBlocks
)
4585 lower_ubo_reference(prog
->_LinkedShaders
[i
],
4586 options
->ClampBlockIndicesToArrayBounds
);
4588 if (options
->LowerShaderSharedVariables
)
4589 lower_shared_reference(prog
->_LinkedShaders
[i
],
4590 &prog
->Comp
.SharedSize
);
4592 lower_vector_derefs(prog
->_LinkedShaders
[i
]);
4593 do_vec_index_to_swizzle(prog
->_LinkedShaders
[i
]->ir
);
4600 link_shaders(struct gl_context
*ctx
, struct gl_shader_program
*prog
)
4602 prog
->LinkStatus
= true; /* All error paths will set this to false */
4603 prog
->Validated
= false;
4604 prog
->_Used
= false;
4606 /* Section 7.3 (Program Objects) of the OpenGL 4.5 Core Profile spec says:
4608 * "Linking can fail for a variety of reasons as specified in the
4609 * OpenGL Shading Language Specification, as well as any of the
4610 * following reasons:
4612 * - No shader objects are attached to program."
4614 * The Compatibility Profile specification does not list the error. In
4615 * Compatibility Profile missing shader stages are replaced by
4616 * fixed-function. This applies to the case where all stages are
4619 if (prog
->NumShaders
== 0) {
4620 if (ctx
->API
!= API_OPENGL_COMPAT
)
4621 linker_error(prog
, "no shaders attached to the program\n");
4625 unsigned int num_explicit_uniform_locs
= 0;
4627 void *mem_ctx
= ralloc_context(NULL
); // temporary linker context
4629 prog
->ARB_fragment_coord_conventions_enable
= false;
4631 /* Separate the shaders into groups based on their type.
4633 struct gl_shader
**shader_list
[MESA_SHADER_STAGES
];
4634 unsigned num_shaders
[MESA_SHADER_STAGES
];
4636 for (int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4637 shader_list
[i
] = (struct gl_shader
**)
4638 calloc(prog
->NumShaders
, sizeof(struct gl_shader
*));
4642 unsigned min_version
= UINT_MAX
;
4643 unsigned max_version
= 0;
4644 for (unsigned i
= 0; i
< prog
->NumShaders
; i
++) {
4645 min_version
= MIN2(min_version
, prog
->Shaders
[i
]->Version
);
4646 max_version
= MAX2(max_version
, prog
->Shaders
[i
]->Version
);
4648 if (prog
->Shaders
[i
]->IsES
!= prog
->Shaders
[0]->IsES
) {
4649 linker_error(prog
, "all shaders must use same shading "
4650 "language version\n");
4654 if (prog
->Shaders
[i
]->info
.ARB_fragment_coord_conventions_enable
) {
4655 prog
->ARB_fragment_coord_conventions_enable
= true;
4658 gl_shader_stage shader_type
= prog
->Shaders
[i
]->Stage
;
4659 shader_list
[shader_type
][num_shaders
[shader_type
]] = prog
->Shaders
[i
];
4660 num_shaders
[shader_type
]++;
4663 /* In desktop GLSL, different shader versions may be linked together. In
4664 * GLSL ES, all shader versions must be the same.
4666 if (prog
->Shaders
[0]->IsES
&& min_version
!= max_version
) {
4667 linker_error(prog
, "all shaders must use same shading "
4668 "language version\n");
4672 prog
->Version
= max_version
;
4673 prog
->IsES
= prog
->Shaders
[0]->IsES
;
4675 /* Some shaders have to be linked with some other shaders present.
4677 if (!prog
->SeparateShader
) {
4678 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0 &&
4679 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4680 linker_error(prog
, "Geometry shader must be linked with "
4684 if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0 &&
4685 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4686 linker_error(prog
, "Tessellation evaluation shader must be linked "
4687 "with vertex shader\n");
4690 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4691 num_shaders
[MESA_SHADER_VERTEX
] == 0) {
4692 linker_error(prog
, "Tessellation control shader must be linked with "
4697 /* The spec is self-contradictory here. It allows linking without a tess
4698 * eval shader, but that can only be used with transform feedback and
4699 * rasterization disabled. However, transform feedback isn't allowed
4700 * with GL_PATCHES, so it can't be used.
4702 * More investigation showed that the idea of transform feedback after
4703 * a tess control shader was dropped, because some hw vendors couldn't
4704 * support tessellation without a tess eval shader, but the linker
4705 * section wasn't updated to reflect that.
4707 * All specifications (ARB_tessellation_shader, GL 4.0-4.5) have this
4710 * Do what's reasonable and always require a tess eval shader if a tess
4711 * control shader is present.
4713 if (num_shaders
[MESA_SHADER_TESS_CTRL
] > 0 &&
4714 num_shaders
[MESA_SHADER_TESS_EVAL
] == 0) {
4715 linker_error(prog
, "Tessellation control shader must be linked with "
4716 "tessellation evaluation shader\n");
4721 /* Compute shaders have additional restrictions. */
4722 if (num_shaders
[MESA_SHADER_COMPUTE
] > 0 &&
4723 num_shaders
[MESA_SHADER_COMPUTE
] != prog
->NumShaders
) {
4724 linker_error(prog
, "Compute shaders may not be linked with any other "
4725 "type of shader\n");
4728 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4729 if (prog
->_LinkedShaders
[i
] != NULL
) {
4730 _mesa_delete_linked_shader(ctx
, prog
->_LinkedShaders
[i
]);
4733 prog
->_LinkedShaders
[i
] = NULL
;
4736 /* Link all shaders for a particular stage and validate the result.
4738 for (int stage
= 0; stage
< MESA_SHADER_STAGES
; stage
++) {
4739 if (num_shaders
[stage
] > 0) {
4740 gl_linked_shader
*const sh
=
4741 link_intrastage_shaders(mem_ctx
, ctx
, prog
, shader_list
[stage
],
4742 num_shaders
[stage
]);
4744 if (!prog
->LinkStatus
) {
4746 _mesa_delete_linked_shader(ctx
, sh
);
4751 case MESA_SHADER_VERTEX
:
4752 validate_vertex_shader_executable(prog
, sh
, ctx
);
4754 case MESA_SHADER_TESS_CTRL
:
4755 /* nothing to be done */
4757 case MESA_SHADER_TESS_EVAL
:
4758 validate_tess_eval_shader_executable(prog
, sh
, ctx
);
4760 case MESA_SHADER_GEOMETRY
:
4761 validate_geometry_shader_executable(prog
, sh
, ctx
);
4763 case MESA_SHADER_FRAGMENT
:
4764 validate_fragment_shader_executable(prog
, sh
);
4767 if (!prog
->LinkStatus
) {
4769 _mesa_delete_linked_shader(ctx
, sh
);
4773 prog
->_LinkedShaders
[stage
] = sh
;
4777 if (num_shaders
[MESA_SHADER_GEOMETRY
] > 0) {
4778 prog
->LastClipDistanceArraySize
= prog
->Geom
.ClipDistanceArraySize
;
4779 prog
->LastCullDistanceArraySize
= prog
->Geom
.CullDistanceArraySize
;
4780 } else if (num_shaders
[MESA_SHADER_TESS_EVAL
] > 0) {
4781 prog
->LastClipDistanceArraySize
= prog
->TessEval
.ClipDistanceArraySize
;
4782 prog
->LastCullDistanceArraySize
= prog
->TessEval
.CullDistanceArraySize
;
4783 } else if (num_shaders
[MESA_SHADER_VERTEX
] > 0) {
4784 prog
->LastClipDistanceArraySize
= prog
->Vert
.ClipDistanceArraySize
;
4785 prog
->LastCullDistanceArraySize
= prog
->Vert
.CullDistanceArraySize
;
4787 prog
->LastClipDistanceArraySize
= 0; /* Not used */
4788 prog
->LastCullDistanceArraySize
= 0; /* Not used */
4791 /* Here begins the inter-stage linking phase. Some initial validation is
4792 * performed, then locations are assigned for uniforms, attributes, and
4795 cross_validate_uniforms(prog
);
4796 if (!prog
->LinkStatus
)
4799 unsigned first
, last
, prev
;
4801 first
= MESA_SHADER_STAGES
;
4804 /* Determine first and last stage. */
4805 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4806 if (!prog
->_LinkedShaders
[i
])
4808 if (first
== MESA_SHADER_STAGES
)
4813 num_explicit_uniform_locs
= check_explicit_uniform_locations(ctx
, prog
);
4814 link_assign_subroutine_types(prog
);
4816 if (!prog
->LinkStatus
)
4819 resize_tes_inputs(ctx
, prog
);
4821 /* Validate the inputs of each stage with the output of the preceding
4825 for (unsigned i
= prev
+ 1; i
<= MESA_SHADER_FRAGMENT
; i
++) {
4826 if (prog
->_LinkedShaders
[i
] == NULL
)
4829 validate_interstage_inout_blocks(prog
, prog
->_LinkedShaders
[prev
],
4830 prog
->_LinkedShaders
[i
]);
4831 if (!prog
->LinkStatus
)
4834 cross_validate_outputs_to_inputs(prog
,
4835 prog
->_LinkedShaders
[prev
],
4836 prog
->_LinkedShaders
[i
]);
4837 if (!prog
->LinkStatus
)
4843 /* Cross-validate uniform blocks between shader stages */
4844 validate_interstage_uniform_blocks(prog
, prog
->_LinkedShaders
);
4845 if (!prog
->LinkStatus
)
4848 for (unsigned int i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4849 if (prog
->_LinkedShaders
[i
] != NULL
)
4850 lower_named_interface_blocks(mem_ctx
, prog
->_LinkedShaders
[i
]);
4853 /* Implement the GLSL 1.30+ rule for discard vs infinite loops Do
4854 * it before optimization because we want most of the checks to get
4855 * dropped thanks to constant propagation.
4857 * This rule also applies to GLSL ES 3.00.
4859 if (max_version
>= (prog
->IsES
? 300 : 130)) {
4860 struct gl_linked_shader
*sh
= prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
];
4862 lower_discard_flow(sh
->ir
);
4866 if (prog
->SeparateShader
)
4867 disable_varying_optimizations_for_sso(prog
);
4870 if (!interstage_cross_validate_uniform_blocks(prog
, false))
4874 if (!interstage_cross_validate_uniform_blocks(prog
, true))
4877 /* Do common optimization before assigning storage for attributes,
4878 * uniforms, and varyings. Later optimization could possibly make
4879 * some of that unused.
4881 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4882 if (prog
->_LinkedShaders
[i
] == NULL
)
4885 detect_recursion_linked(prog
, prog
->_LinkedShaders
[i
]->ir
);
4886 if (!prog
->LinkStatus
)
4889 if (ctx
->Const
.ShaderCompilerOptions
[i
].LowerCombinedClipCullDistance
) {
4890 lower_clip_cull_distance(prog
, prog
->_LinkedShaders
[i
]);
4893 if (ctx
->Const
.LowerTessLevel
) {
4894 lower_tess_level(prog
->_LinkedShaders
[i
]);
4897 while (do_common_optimization(prog
->_LinkedShaders
[i
]->ir
, true, false,
4898 &ctx
->Const
.ShaderCompilerOptions
[i
],
4899 ctx
->Const
.NativeIntegers
))
4902 lower_const_arrays_to_uniforms(prog
->_LinkedShaders
[i
]->ir
, i
);
4903 propagate_invariance(prog
->_LinkedShaders
[i
]->ir
);
4906 /* Validation for special cases where we allow sampler array indexing
4907 * with loop induction variable. This check emits a warning or error
4908 * depending if backend can handle dynamic indexing.
4910 if ((!prog
->IsES
&& prog
->Version
< 130) ||
4911 (prog
->IsES
&& prog
->Version
< 300)) {
4912 if (!validate_sampler_array_indexing(ctx
, prog
))
4916 /* Check and validate stream emissions in geometry shaders */
4917 validate_geometry_shader_emissions(ctx
, prog
);
4919 store_fragdepth_layout(prog
);
4921 if(!link_varyings_and_uniforms(first
, last
, num_explicit_uniform_locs
, ctx
,
4925 /* OpenGL ES < 3.1 requires that a vertex shader and a fragment shader both
4926 * be present in a linked program. GL_ARB_ES2_compatibility doesn't say
4927 * anything about shader linking when one of the shaders (vertex or
4928 * fragment shader) is absent. So, the extension shouldn't change the
4929 * behavior specified in GLSL specification.
4931 * From OpenGL ES 3.1 specification (7.3 Program Objects):
4932 * "Linking can fail for a variety of reasons as specified in the
4933 * OpenGL ES Shading Language Specification, as well as any of the
4934 * following reasons:
4938 * * program contains objects to form either a vertex shader or
4939 * fragment shader, and program is not separable, and does not
4940 * contain objects to form both a vertex shader and fragment
4943 * However, the only scenario in 3.1+ where we don't require them both is
4944 * when we have a compute shader. For example:
4946 * - No shaders is a link error.
4947 * - Geom or Tess without a Vertex shader is a link error which means we
4948 * always require a Vertex shader and hence a Fragment shader.
4949 * - Finally a Compute shader linked with any other stage is a link error.
4951 if (!prog
->SeparateShader
&& ctx
->API
== API_OPENGLES2
&&
4952 num_shaders
[MESA_SHADER_COMPUTE
] == 0) {
4953 if (prog
->_LinkedShaders
[MESA_SHADER_VERTEX
] == NULL
) {
4954 linker_error(prog
, "program lacks a vertex shader\n");
4955 } else if (prog
->_LinkedShaders
[MESA_SHADER_FRAGMENT
] == NULL
) {
4956 linker_error(prog
, "program lacks a fragment shader\n");
4961 for (unsigned i
= 0; i
< MESA_SHADER_STAGES
; i
++) {
4962 free(shader_list
[i
]);
4963 if (prog
->_LinkedShaders
[i
] == NULL
)
4966 /* Do a final validation step to make sure that the IR wasn't
4967 * invalidated by any modifications performed after intrastage linking.
4969 validate_ir_tree(prog
->_LinkedShaders
[i
]->ir
);
4971 /* Retain any live IR, but trash the rest. */
4972 reparent_ir(prog
->_LinkedShaders
[i
]->ir
, prog
->_LinkedShaders
[i
]->ir
);
4974 /* The symbol table in the linked shaders may contain references to
4975 * variables that were removed (e.g., unused uniforms). Since it may
4976 * contain junk, there is no possible valid use. Delete it and set the
4979 delete prog
->_LinkedShaders
[i
]->symbols
;
4980 prog
->_LinkedShaders
[i
]->symbols
= NULL
;
4983 ralloc_free(mem_ctx
);