2 * Copyright © 2012 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.
25 * \file link_varyings.cpp
27 * Linker functions related specifically to linking varyings between shader
32 #include "main/mtypes.h"
33 #include "glsl_symbol_table.h"
34 #include "glsl_parser_extras.h"
35 #include "ir_optimization.h"
37 #include "link_varyings.h"
38 #include "main/macros.h"
39 #include "program/hash_table.h"
44 * Get the varying type stripped of the outermost array if we're processing
45 * a stage whose varyings are arrays indexed by a vertex number (such as
46 * geometry shader inputs).
48 static const glsl_type
*
49 get_varying_type(const ir_variable
*var
, gl_shader_stage stage
)
51 const glsl_type
*type
= var
->type
;
53 if (!var
->data
.patch
&&
54 ((var
->data
.mode
== ir_var_shader_out
&&
55 stage
== MESA_SHADER_TESS_CTRL
) ||
56 (var
->data
.mode
== ir_var_shader_in
&&
57 (stage
== MESA_SHADER_TESS_CTRL
|| stage
== MESA_SHADER_TESS_EVAL
||
58 stage
== MESA_SHADER_GEOMETRY
)))) {
59 assert(type
->is_array());
60 type
= type
->fields
.array
;
67 create_xfb_varying_names(void *mem_ctx
, const glsl_type
*t
, char **name
,
68 size_t name_length
, unsigned *count
,
69 const char *ifc_member_name
,
70 const glsl_type
*ifc_member_t
, char ***varying_names
)
72 if (t
->is_interface()) {
73 size_t new_length
= name_length
;
75 assert(ifc_member_name
&& ifc_member_t
);
76 ralloc_asprintf_rewrite_tail(name
, &new_length
, ".%s", ifc_member_name
);
78 create_xfb_varying_names(mem_ctx
, ifc_member_t
, name
, new_length
, count
,
79 NULL
, NULL
, varying_names
);
80 } else if (t
->is_record()) {
81 for (unsigned i
= 0; i
< t
->length
; i
++) {
82 const char *field
= t
->fields
.structure
[i
].name
;
83 size_t new_length
= name_length
;
85 ralloc_asprintf_rewrite_tail(name
, &new_length
, ".%s", field
);
87 create_xfb_varying_names(mem_ctx
, t
->fields
.structure
[i
].type
, name
,
88 new_length
, count
, NULL
, NULL
,
91 } else if (t
->without_array()->is_record() ||
92 t
->without_array()->is_interface() ||
93 (t
->is_array() && t
->fields
.array
->is_array())) {
94 for (unsigned i
= 0; i
< t
->length
; i
++) {
95 size_t new_length
= name_length
;
97 /* Append the subscript to the current variable name */
98 ralloc_asprintf_rewrite_tail(name
, &new_length
, "[%u]", i
);
100 create_xfb_varying_names(mem_ctx
, t
->fields
.array
, name
, new_length
,
101 count
, ifc_member_name
, ifc_member_t
,
105 (*varying_names
)[(*count
)++] = ralloc_strdup(mem_ctx
, *name
);
110 process_xfb_layout_qualifiers(void *mem_ctx
, const gl_shader
*sh
,
111 unsigned *num_tfeedback_decls
,
112 char ***varying_names
)
114 bool has_xfb_qualifiers
= false;
116 /* We still need to enable transform feedback mode even if xfb_stride is
117 * only applied to a global out. Also we don't bother to propagate
118 * xfb_stride to interface block members so this will catch that case also.
120 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
121 if (sh
->TransformFeedback
.BufferStride
[j
]) {
122 has_xfb_qualifiers
= true;
126 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
127 ir_variable
*var
= node
->as_variable();
128 if (!var
|| var
->data
.mode
!= ir_var_shader_out
)
131 /* From the ARB_enhanced_layouts spec:
133 * "Any shader making any static use (after preprocessing) of any of
134 * these *xfb_* qualifiers will cause the shader to be in a
135 * transform feedback capturing mode and hence responsible for
136 * describing the transform feedback setup. This mode will capture
137 * any output selected by *xfb_offset*, directly or indirectly, to
138 * a transform feedback buffer."
140 if (var
->data
.explicit_xfb_buffer
|| var
->data
.explicit_xfb_stride
) {
141 has_xfb_qualifiers
= true;
144 if (var
->data
.explicit_xfb_offset
) {
145 *num_tfeedback_decls
+= var
->type
->varying_count();
146 has_xfb_qualifiers
= true;
150 if (*num_tfeedback_decls
== 0)
151 return has_xfb_qualifiers
;
154 *varying_names
= ralloc_array(mem_ctx
, char *, *num_tfeedback_decls
);
155 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
156 ir_variable
*var
= node
->as_variable();
157 if (!var
|| var
->data
.mode
!= ir_var_shader_out
)
160 if (var
->data
.explicit_xfb_offset
) {
162 const glsl_type
*type
, *member_type
;
164 if (var
->data
.from_named_ifc_block
) {
165 type
= var
->get_interface_type();
166 /* Find the member type before it was altered by lowering */
168 type
->fields
.structure
[type
->field_index(var
->name
)].type
;
169 name
= ralloc_strdup(NULL
, type
->without_array()->name
);
173 name
= ralloc_strdup(NULL
, var
->name
);
175 create_xfb_varying_names(mem_ctx
, type
, &name
, strlen(name
), &i
,
176 var
->name
, member_type
, varying_names
);
181 assert(i
== *num_tfeedback_decls
);
182 return has_xfb_qualifiers
;
186 * Validate the types and qualifiers of an output from one stage against the
187 * matching input to another stage.
190 cross_validate_types_and_qualifiers(struct gl_shader_program
*prog
,
191 const ir_variable
*input
,
192 const ir_variable
*output
,
193 gl_shader_stage consumer_stage
,
194 gl_shader_stage producer_stage
)
196 /* Check that the types match between stages.
198 const glsl_type
*type_to_match
= input
->type
;
200 /* VS -> GS, VS -> TCS, VS -> TES, TES -> GS */
201 const bool extra_array_level
= (producer_stage
== MESA_SHADER_VERTEX
&&
202 consumer_stage
!= MESA_SHADER_FRAGMENT
) ||
203 consumer_stage
== MESA_SHADER_GEOMETRY
;
204 if (extra_array_level
) {
205 assert(type_to_match
->is_array());
206 type_to_match
= type_to_match
->fields
.array
;
209 if (type_to_match
!= output
->type
) {
210 /* There is a bit of a special case for gl_TexCoord. This
211 * built-in is unsized by default. Applications that variable
212 * access it must redeclare it with a size. There is some
213 * language in the GLSL spec that implies the fragment shader
214 * and vertex shader do not have to agree on this size. Other
215 * driver behave this way, and one or two applications seem to
218 * Neither declaration needs to be modified here because the array
219 * sizes are fixed later when update_array_sizes is called.
221 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
223 * "Unlike user-defined varying variables, the built-in
224 * varying variables don't have a strict one-to-one
225 * correspondence between the vertex language and the
226 * fragment language."
228 if (!output
->type
->is_array() || !is_gl_identifier(output
->name
)) {
230 "%s shader output `%s' declared as type `%s', "
231 "but %s shader input declared as type `%s'\n",
232 _mesa_shader_stage_to_string(producer_stage
),
235 _mesa_shader_stage_to_string(consumer_stage
),
241 /* Check that all of the qualifiers match between stages.
243 if (input
->data
.centroid
!= output
->data
.centroid
) {
245 "%s shader output `%s' %s centroid qualifier, "
246 "but %s shader input %s centroid qualifier\n",
247 _mesa_shader_stage_to_string(producer_stage
),
249 (output
->data
.centroid
) ? "has" : "lacks",
250 _mesa_shader_stage_to_string(consumer_stage
),
251 (input
->data
.centroid
) ? "has" : "lacks");
255 if (input
->data
.sample
!= output
->data
.sample
) {
257 "%s shader output `%s' %s sample qualifier, "
258 "but %s shader input %s sample qualifier\n",
259 _mesa_shader_stage_to_string(producer_stage
),
261 (output
->data
.sample
) ? "has" : "lacks",
262 _mesa_shader_stage_to_string(consumer_stage
),
263 (input
->data
.sample
) ? "has" : "lacks");
267 if (input
->data
.patch
!= output
->data
.patch
) {
269 "%s shader output `%s' %s patch qualifier, "
270 "but %s shader input %s patch qualifier\n",
271 _mesa_shader_stage_to_string(producer_stage
),
273 (output
->data
.patch
) ? "has" : "lacks",
274 _mesa_shader_stage_to_string(consumer_stage
),
275 (input
->data
.patch
) ? "has" : "lacks");
279 if (!prog
->IsES
&& input
->data
.invariant
!= output
->data
.invariant
) {
281 "%s shader output `%s' %s invariant qualifier, "
282 "but %s shader input %s invariant qualifier\n",
283 _mesa_shader_stage_to_string(producer_stage
),
285 (output
->data
.invariant
) ? "has" : "lacks",
286 _mesa_shader_stage_to_string(consumer_stage
),
287 (input
->data
.invariant
) ? "has" : "lacks");
291 /* GLSL >= 4.40 removes text requiring interpolation qualifiers
292 * to match cross stage, they must only match within the same stage.
294 * From page 84 (page 90 of the PDF) of the GLSL 4.40 spec:
296 * "It is a link-time error if, within the same stage, the interpolation
297 * qualifiers of variables of the same name do not match.
300 if (input
->data
.interpolation
!= output
->data
.interpolation
&&
301 prog
->Version
< 440) {
303 "%s shader output `%s' specifies %s "
304 "interpolation qualifier, "
305 "but %s shader input specifies %s "
306 "interpolation qualifier\n",
307 _mesa_shader_stage_to_string(producer_stage
),
309 interpolation_string(output
->data
.interpolation
),
310 _mesa_shader_stage_to_string(consumer_stage
),
311 interpolation_string(input
->data
.interpolation
));
317 * Validate front and back color outputs against single color input
320 cross_validate_front_and_back_color(struct gl_shader_program
*prog
,
321 const ir_variable
*input
,
322 const ir_variable
*front_color
,
323 const ir_variable
*back_color
,
324 gl_shader_stage consumer_stage
,
325 gl_shader_stage producer_stage
)
327 if (front_color
!= NULL
&& front_color
->data
.assigned
)
328 cross_validate_types_and_qualifiers(prog
, input
, front_color
,
329 consumer_stage
, producer_stage
);
331 if (back_color
!= NULL
&& back_color
->data
.assigned
)
332 cross_validate_types_and_qualifiers(prog
, input
, back_color
,
333 consumer_stage
, producer_stage
);
337 * Validate that outputs from one stage match inputs of another
340 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
341 gl_shader
*producer
, gl_shader
*consumer
)
343 glsl_symbol_table parameters
;
344 ir_variable
*explicit_locations
[MAX_VARYING
] = { NULL
, };
346 /* Find all shader outputs in the "producer" stage.
348 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
349 ir_variable
*const var
= node
->as_variable();
351 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
354 if (!var
->data
.explicit_location
355 || var
->data
.location
< VARYING_SLOT_VAR0
)
356 parameters
.add_variable(var
);
358 /* User-defined varyings with explicit locations are handled
359 * differently because they do not need to have matching names.
361 const unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
363 if (explicit_locations
[idx
] != NULL
) {
365 "%s shader has multiple outputs explicitly "
366 "assigned to location %d\n",
367 _mesa_shader_stage_to_string(producer
->Stage
),
372 explicit_locations
[idx
] = var
;
377 /* Find all shader inputs in the "consumer" stage. Any variables that have
378 * matching outputs already in the symbol table must have the same type and
381 * Exception: if the consumer is the geometry shader, then the inputs
382 * should be arrays and the type of the array element should match the type
383 * of the corresponding producer output.
385 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
386 ir_variable
*const input
= node
->as_variable();
388 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
391 if (strcmp(input
->name
, "gl_Color") == 0 && input
->data
.used
) {
392 const ir_variable
*const front_color
=
393 parameters
.get_variable("gl_FrontColor");
395 const ir_variable
*const back_color
=
396 parameters
.get_variable("gl_BackColor");
398 cross_validate_front_and_back_color(prog
, input
,
399 front_color
, back_color
,
400 consumer
->Stage
, producer
->Stage
);
401 } else if (strcmp(input
->name
, "gl_SecondaryColor") == 0 && input
->data
.used
) {
402 const ir_variable
*const front_color
=
403 parameters
.get_variable("gl_FrontSecondaryColor");
405 const ir_variable
*const back_color
=
406 parameters
.get_variable("gl_BackSecondaryColor");
408 cross_validate_front_and_back_color(prog
, input
,
409 front_color
, back_color
,
410 consumer
->Stage
, producer
->Stage
);
412 /* The rules for connecting inputs and outputs change in the presence
413 * of explicit locations. In this case, we no longer care about the
414 * names of the variables. Instead, we care only about the
415 * explicitly assigned location.
417 ir_variable
*output
= NULL
;
418 if (input
->data
.explicit_location
419 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
420 output
= explicit_locations
[input
->data
.location
- VARYING_SLOT_VAR0
];
422 if (output
== NULL
) {
424 "%s shader input `%s' with explicit location "
425 "has no matching output\n",
426 _mesa_shader_stage_to_string(consumer
->Stage
),
430 output
= parameters
.get_variable(input
->name
);
433 if (output
!= NULL
) {
434 /* Interface blocks have their own validation elsewhere so don't
435 * try validating them here.
437 if (!(input
->get_interface_type() &&
438 output
->get_interface_type()))
439 cross_validate_types_and_qualifiers(prog
, input
, output
,
443 /* Check for input vars with unmatched output vars in prev stage
444 * taking into account that interface blocks could have a matching
445 * output but with different name, so we ignore them.
447 assert(!input
->data
.assigned
);
448 if (input
->data
.used
&& !input
->get_interface_type() &&
449 !input
->data
.explicit_location
&& !prog
->SeparateShader
)
451 "%s shader input `%s' "
452 "has no matching output in the previous stage\n",
453 _mesa_shader_stage_to_string(consumer
->Stage
),
461 * Demote shader inputs and outputs that are not used in other stages, and
462 * remove them via dead code elimination.
465 remove_unused_shader_inputs_and_outputs(bool is_separate_shader_object
,
467 enum ir_variable_mode mode
)
469 if (is_separate_shader_object
)
472 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
473 ir_variable
*const var
= node
->as_variable();
475 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
478 /* A shader 'in' or 'out' variable is only really an input or output if
479 * its value is used by other shader stages. This will cause the
480 * variable to have a location assigned.
482 if (var
->data
.is_unmatched_generic_inout
) {
483 assert(var
->data
.mode
!= ir_var_temporary
);
484 var
->data
.mode
= ir_var_auto
;
488 /* Eliminate code that is now dead due to unused inputs/outputs being
491 while (do_dead_code(sh
->ir
, false))
497 * Initialize this object based on a string that was passed to
498 * glTransformFeedbackVaryings.
500 * If the input is mal-formed, this call still succeeds, but it sets
501 * this->var_name to a mal-formed input, so tfeedback_decl::find_output_var()
502 * will fail to find any matching variable.
505 tfeedback_decl::init(struct gl_context
*ctx
, const void *mem_ctx
,
508 /* We don't have to be pedantic about what is a valid GLSL variable name,
509 * because any variable with an invalid name can't exist in the IR anyway.
513 this->orig_name
= input
;
514 this->lowered_builtin_array_variable
= none
;
515 this->skip_components
= 0;
516 this->next_buffer_separator
= false;
517 this->matched_candidate
= NULL
;
522 if (ctx
->Extensions
.ARB_transform_feedback3
) {
523 /* Parse gl_NextBuffer. */
524 if (strcmp(input
, "gl_NextBuffer") == 0) {
525 this->next_buffer_separator
= true;
529 /* Parse gl_SkipComponents. */
530 if (strcmp(input
, "gl_SkipComponents1") == 0)
531 this->skip_components
= 1;
532 else if (strcmp(input
, "gl_SkipComponents2") == 0)
533 this->skip_components
= 2;
534 else if (strcmp(input
, "gl_SkipComponents3") == 0)
535 this->skip_components
= 3;
536 else if (strcmp(input
, "gl_SkipComponents4") == 0)
537 this->skip_components
= 4;
539 if (this->skip_components
)
543 /* Parse a declaration. */
544 const char *base_name_end
;
545 long subscript
= parse_program_resource_name(input
, &base_name_end
);
546 this->var_name
= ralloc_strndup(mem_ctx
, input
, base_name_end
- input
);
547 if (this->var_name
== NULL
) {
548 _mesa_error_no_memory(__func__
);
552 if (subscript
>= 0) {
553 this->array_subscript
= subscript
;
554 this->is_subscripted
= true;
556 this->is_subscripted
= false;
559 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this
560 * class must behave specially to account for the fact that gl_ClipDistance
561 * is converted from a float[8] to a vec4[2].
563 if (ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].LowerClipDistance
&&
564 strcmp(this->var_name
, "gl_ClipDistance") == 0) {
565 this->lowered_builtin_array_variable
= clip_distance
;
568 if (ctx
->Const
.LowerTessLevel
&&
569 (strcmp(this->var_name
, "gl_TessLevelOuter") == 0))
570 this->lowered_builtin_array_variable
= tess_level_outer
;
571 if (ctx
->Const
.LowerTessLevel
&&
572 (strcmp(this->var_name
, "gl_TessLevelInner") == 0))
573 this->lowered_builtin_array_variable
= tess_level_inner
;
578 * Determine whether two tfeedback_decl objects refer to the same variable and
579 * array index (if applicable).
582 tfeedback_decl::is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
)
584 assert(x
.is_varying() && y
.is_varying());
586 if (strcmp(x
.var_name
, y
.var_name
) != 0)
588 if (x
.is_subscripted
!= y
.is_subscripted
)
590 if (x
.is_subscripted
&& x
.array_subscript
!= y
.array_subscript
)
597 * Assign a location and stream ID for this tfeedback_decl object based on the
598 * transform feedback candidate found by find_candidate.
600 * If an error occurs, the error is reported through linker_error() and false
604 tfeedback_decl::assign_location(struct gl_context
*ctx
,
605 struct gl_shader_program
*prog
)
607 assert(this->is_varying());
609 unsigned fine_location
610 = this->matched_candidate
->toplevel_var
->data
.location
* 4
611 + this->matched_candidate
->toplevel_var
->data
.location_frac
612 + this->matched_candidate
->offset
;
613 const unsigned dmul
=
614 this->matched_candidate
->type
->without_array()->is_double() ? 2 : 1;
616 if (this->matched_candidate
->type
->is_array()) {
618 const unsigned matrix_cols
=
619 this->matched_candidate
->type
->fields
.array
->matrix_columns
;
620 const unsigned vector_elements
=
621 this->matched_candidate
->type
->fields
.array
->vector_elements
;
622 unsigned actual_array_size
;
623 switch (this->lowered_builtin_array_variable
) {
625 actual_array_size
= prog
->LastClipDistanceArraySize
;
627 case tess_level_outer
:
628 actual_array_size
= 4;
630 case tess_level_inner
:
631 actual_array_size
= 2;
635 actual_array_size
= this->matched_candidate
->type
->array_size();
639 if (this->is_subscripted
) {
640 /* Check array bounds. */
641 if (this->array_subscript
>= actual_array_size
) {
642 linker_error(prog
, "Transform feedback varying %s has index "
643 "%i, but the array size is %u.",
644 this->orig_name
, this->array_subscript
,
648 unsigned array_elem_size
= this->lowered_builtin_array_variable
?
649 1 : vector_elements
* matrix_cols
* dmul
;
650 fine_location
+= array_elem_size
* this->array_subscript
;
653 this->size
= actual_array_size
;
655 this->vector_elements
= vector_elements
;
656 this->matrix_columns
= matrix_cols
;
657 if (this->lowered_builtin_array_variable
)
658 this->type
= GL_FLOAT
;
660 this->type
= this->matched_candidate
->type
->fields
.array
->gl_type
;
662 /* Regular variable (scalar, vector, or matrix) */
663 if (this->is_subscripted
) {
664 linker_error(prog
, "Transform feedback varying %s requested, "
665 "but %s is not an array.",
666 this->orig_name
, this->var_name
);
670 this->vector_elements
= this->matched_candidate
->type
->vector_elements
;
671 this->matrix_columns
= this->matched_candidate
->type
->matrix_columns
;
672 this->type
= this->matched_candidate
->type
->gl_type
;
674 this->location
= fine_location
/ 4;
675 this->location_frac
= fine_location
% 4;
677 /* From GL_EXT_transform_feedback:
678 * A program will fail to link if:
680 * * the total number of components to capture in any varying
681 * variable in <varyings> is greater than the constant
682 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
683 * buffer mode is SEPARATE_ATTRIBS_EXT;
685 if (prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
&&
686 this->num_components() >
687 ctx
->Const
.MaxTransformFeedbackSeparateComponents
) {
688 linker_error(prog
, "Transform feedback varying %s exceeds "
689 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
694 /* Only transform feedback varyings can be assigned to non-zero streams,
695 * so assign the stream id here.
697 this->stream_id
= this->matched_candidate
->toplevel_var
->data
.stream
;
699 unsigned array_offset
= this->array_subscript
* 4 * dmul
;
700 unsigned struct_offset
= this->matched_candidate
->offset
* 4 * dmul
;
701 this->buffer
= this->matched_candidate
->toplevel_var
->data
.xfb_buffer
;
702 this->offset
= this->matched_candidate
->toplevel_var
->data
.offset
+
703 array_offset
+ struct_offset
;
710 tfeedback_decl::get_num_outputs() const
712 if (!this->is_varying()) {
715 return (this->num_components() + this->location_frac
+ 3)/4;
720 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
722 * If an error occurs, the error is reported through linker_error() and false
726 tfeedback_decl::store(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
727 struct gl_transform_feedback_info
*info
,
728 unsigned buffer
, unsigned buffer_index
,
729 const unsigned max_outputs
, bool *explicit_stride
,
730 bool has_xfb_qualifiers
) const
732 assert(!this->next_buffer_separator
);
734 /* Handle gl_SkipComponents. */
735 if (this->skip_components
) {
736 info
->Buffers
[buffer
].Stride
+= this->skip_components
;
740 unsigned xfb_offset
= 0;
741 if (has_xfb_qualifiers
) {
742 xfb_offset
= this->offset
/ 4;
744 xfb_offset
= info
->Buffers
[buffer
].Stride
;
746 info
->Varyings
[info
->NumVarying
].Offset
= xfb_offset
* 4;
748 unsigned location
= this->location
;
749 unsigned location_frac
= this->location_frac
;
750 unsigned num_components
= this->num_components();
751 while (num_components
> 0) {
752 unsigned output_size
= MIN2(num_components
, 4 - location_frac
);
753 assert((info
->NumOutputs
== 0 && max_outputs
== 0) ||
754 info
->NumOutputs
< max_outputs
);
756 /* From the ARB_enhanced_layouts spec:
758 * "If such a block member or variable is not written during a shader
759 * invocation, the buffer contents at the assigned offset will be
760 * undefined. Even if there are no static writes to a variable or
761 * member that is assigned a transform feedback offset, the space is
762 * still allocated in the buffer and still affects the stride."
764 if (this->is_varying_written()) {
765 info
->Outputs
[info
->NumOutputs
].ComponentOffset
= location_frac
;
766 info
->Outputs
[info
->NumOutputs
].OutputRegister
= location
;
767 info
->Outputs
[info
->NumOutputs
].NumComponents
= output_size
;
768 info
->Outputs
[info
->NumOutputs
].StreamId
= stream_id
;
769 info
->Outputs
[info
->NumOutputs
].OutputBuffer
= buffer
;
770 info
->Outputs
[info
->NumOutputs
].DstOffset
= xfb_offset
;
773 info
->Buffers
[buffer
].Stream
= this->stream_id
;
774 xfb_offset
+= output_size
;
776 num_components
-= output_size
;
781 if (explicit_stride
&& explicit_stride
[buffer
]) {
782 if (this->is_double() && info
->Buffers
[buffer
].Stride
% 2) {
783 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
784 "multiple of 8 as its applied to a type that is or "
785 "contains a double.",
786 info
->Buffers
[buffer
].Stride
* 4);
790 if ((this->offset
/ 4) / info
->Buffers
[buffer
].Stride
!=
791 (xfb_offset
- 1) / info
->Buffers
[buffer
].Stride
) {
792 linker_error(prog
, "xfb_offset (%d) overflows xfb_stride (%d) for "
793 "buffer (%d)", xfb_offset
* 4,
794 info
->Buffers
[buffer
].Stride
* 4, buffer
);
798 info
->Buffers
[buffer
].Stride
= xfb_offset
;
801 /* From GL_EXT_transform_feedback:
802 * A program will fail to link if:
804 * * the total number of components to capture is greater than
805 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
806 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
808 * From GL_ARB_enhanced_layouts:
810 * "The resulting stride (implicit or explicit) must be less than or
811 * equal to the implementation-dependent constant
812 * gl_MaxTransformFeedbackInterleavedComponents."
814 if ((prog
->TransformFeedback
.BufferMode
== GL_INTERLEAVED_ATTRIBS
||
815 has_xfb_qualifiers
) &&
816 info
->Buffers
[buffer
].Stride
>
817 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
818 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
819 "limit has been exceeded.");
823 info
->Varyings
[info
->NumVarying
].Name
= ralloc_strdup(prog
,
825 info
->Varyings
[info
->NumVarying
].Type
= this->type
;
826 info
->Varyings
[info
->NumVarying
].Size
= this->size
;
827 info
->Varyings
[info
->NumVarying
].BufferIndex
= buffer_index
;
829 info
->Buffers
[buffer
].NumVaryings
++;
835 const tfeedback_candidate
*
836 tfeedback_decl::find_candidate(gl_shader_program
*prog
,
837 hash_table
*tfeedback_candidates
)
839 const char *name
= this->var_name
;
840 switch (this->lowered_builtin_array_variable
) {
842 name
= this->var_name
;
845 name
= "gl_ClipDistanceMESA";
847 case tess_level_outer
:
848 name
= "gl_TessLevelOuterMESA";
850 case tess_level_inner
:
851 name
= "gl_TessLevelInnerMESA";
854 this->matched_candidate
= (const tfeedback_candidate
*)
855 hash_table_find(tfeedback_candidates
, name
);
856 if (!this->matched_candidate
) {
857 /* From GL_EXT_transform_feedback:
858 * A program will fail to link if:
860 * * any variable name specified in the <varyings> array is not
861 * declared as an output in the geometry shader (if present) or
862 * the vertex shader (if no geometry shader is present);
864 linker_error(prog
, "Transform feedback varying %s undeclared.",
867 return this->matched_candidate
;
872 * Parse all the transform feedback declarations that were passed to
873 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
875 * If an error occurs, the error is reported through linker_error() and false
879 parse_tfeedback_decls(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
880 const void *mem_ctx
, unsigned num_names
,
881 char **varying_names
, tfeedback_decl
*decls
)
883 for (unsigned i
= 0; i
< num_names
; ++i
) {
884 decls
[i
].init(ctx
, mem_ctx
, varying_names
[i
]);
886 if (!decls
[i
].is_varying())
889 /* From GL_EXT_transform_feedback:
890 * A program will fail to link if:
892 * * any two entries in the <varyings> array specify the same varying
895 * We interpret this to mean "any two entries in the <varyings> array
896 * specify the same varying variable and array index", since transform
897 * feedback of arrays would be useless otherwise.
899 for (unsigned j
= 0; j
< i
; ++j
) {
900 if (!decls
[j
].is_varying())
903 if (tfeedback_decl::is_same(decls
[i
], decls
[j
])) {
904 linker_error(prog
, "Transform feedback varying %s specified "
905 "more than once.", varying_names
[i
]);
915 cmp_xfb_offset(const void * x_generic
, const void * y_generic
)
917 tfeedback_decl
*x
= (tfeedback_decl
*) x_generic
;
918 tfeedback_decl
*y
= (tfeedback_decl
*) y_generic
;
920 if (x
->get_buffer() != y
->get_buffer())
921 return x
->get_buffer() - y
->get_buffer();
922 return x
->get_offset() - y
->get_offset();
926 * Store transform feedback location assignments into
927 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
929 * If an error occurs, the error is reported through linker_error() and false
933 store_tfeedback_info(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
934 unsigned num_tfeedback_decls
,
935 tfeedback_decl
*tfeedback_decls
, bool has_xfb_qualifiers
)
937 /* Make sure MaxTransformFeedbackBuffers is less than 32 so the bitmask for
938 * tracking the number of buffers doesn't overflow.
940 assert(ctx
->Const
.MaxTransformFeedbackBuffers
< 32);
942 bool separate_attribs_mode
=
943 prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
;
945 ralloc_free(prog
->LinkedTransformFeedback
.Varyings
);
946 ralloc_free(prog
->LinkedTransformFeedback
.Outputs
);
948 memset(&prog
->LinkedTransformFeedback
, 0,
949 sizeof(prog
->LinkedTransformFeedback
));
951 /* The xfb_offset qualifier does not have to be used in increasing order
952 * however some drivers expect to receive the list of transform feedback
953 * declarations in order so sort it now for convenience.
955 if (has_xfb_qualifiers
)
956 qsort(tfeedback_decls
, num_tfeedback_decls
, sizeof(*tfeedback_decls
),
959 prog
->LinkedTransformFeedback
.Varyings
=
961 struct gl_transform_feedback_varying_info
,
962 num_tfeedback_decls
);
964 unsigned num_outputs
= 0;
965 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
966 if (tfeedback_decls
[i
].is_varying_written())
967 num_outputs
+= tfeedback_decls
[i
].get_num_outputs();
970 prog
->LinkedTransformFeedback
.Outputs
=
972 struct gl_transform_feedback_output
,
975 unsigned num_buffers
= 0;
976 unsigned buffers
= 0;
978 if (!has_xfb_qualifiers
&& separate_attribs_mode
) {
979 /* GL_SEPARATE_ATTRIBS */
980 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
981 if (!tfeedback_decls
[i
].store(ctx
, prog
, &prog
->LinkedTransformFeedback
,
982 num_buffers
, num_buffers
, num_outputs
,
983 NULL
, has_xfb_qualifiers
))
986 buffers
|= 1 << num_buffers
;
991 /* GL_INVERLEAVED_ATTRIBS */
992 int buffer_stream_id
= -1;
994 num_tfeedback_decls
? tfeedback_decls
[0].get_buffer() : 0;
995 bool explicit_stride
[MAX_FEEDBACK_BUFFERS
] = { false };
997 /* Apply any xfb_stride global qualifiers */
998 if (has_xfb_qualifiers
) {
999 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1000 if (prog
->TransformFeedback
.BufferStride
[j
]) {
1002 explicit_stride
[j
] = true;
1003 prog
->LinkedTransformFeedback
.Buffers
[j
].Stride
=
1004 prog
->TransformFeedback
.BufferStride
[j
] / 4;
1009 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
1010 if (has_xfb_qualifiers
&&
1011 buffer
!= tfeedback_decls
[i
].get_buffer()) {
1012 /* we have moved to the next buffer so reset stream id */
1013 buffer_stream_id
= -1;
1017 if (tfeedback_decls
[i
].is_next_buffer_separator()) {
1019 buffer_stream_id
= -1;
1021 } else if (buffer_stream_id
== -1) {
1022 /* First varying writing to this buffer: remember its stream */
1023 buffer_stream_id
= (int) tfeedback_decls
[i
].get_stream_id();
1024 } else if (buffer_stream_id
!=
1025 (int) tfeedback_decls
[i
].get_stream_id()) {
1026 /* Varying writes to the same buffer from a different stream */
1028 "Transform feedback can't capture varyings belonging "
1029 "to different vertex streams in a single buffer. "
1030 "Varying %s writes to buffer from stream %u, other "
1031 "varyings in the same buffer write from stream %u.",
1032 tfeedback_decls
[i
].name(),
1033 tfeedback_decls
[i
].get_stream_id(),
1038 if (has_xfb_qualifiers
) {
1039 buffer
= tfeedback_decls
[i
].get_buffer();
1041 buffer
= num_buffers
;
1043 buffers
|= 1 << buffer
;
1045 if (!tfeedback_decls
[i
].store(ctx
, prog
,
1046 &prog
->LinkedTransformFeedback
,
1047 buffer
, num_buffers
, num_outputs
,
1048 explicit_stride
, has_xfb_qualifiers
))
1053 assert(prog
->LinkedTransformFeedback
.NumOutputs
== num_outputs
);
1055 prog
->LinkedTransformFeedback
.ActiveBuffers
= buffers
;
1062 * Data structure recording the relationship between outputs of one shader
1063 * stage (the "producer") and inputs of another (the "consumer").
1065 class varying_matches
1068 varying_matches(bool disable_varying_packing
, bool xfb_enabled
,
1069 gl_shader_stage producer_stage
,
1070 gl_shader_stage consumer_stage
);
1072 void record(ir_variable
*producer_var
, ir_variable
*consumer_var
);
1073 unsigned assign_locations(struct gl_shader_program
*prog
,
1074 uint64_t reserved_slots
, bool separate_shader
);
1075 void store_locations() const;
1078 bool is_varying_packing_safe(const glsl_type
*type
,
1079 const ir_variable
*var
);
1082 * If true, this driver disables varying packing, so all varyings need to
1083 * be aligned on slot boundaries, and take up a number of slots equal to
1084 * their number of matrix columns times their array size.
1086 * Packing may also be disabled because our current packing method is not
1087 * safe in SSO or versions of OpenGL where interpolation qualifiers are not
1088 * guaranteed to match across stages.
1090 const bool disable_varying_packing
;
1093 * If true, this driver has transform feedback enabled. The transform
1094 * feedback code requires at least some packing be done even when varying
1095 * packing is disabled, fortunately where transform feedback requires
1096 * packing it's safe to override the disabled setting. See
1097 * is_varying_packing_safe().
1099 const bool xfb_enabled
;
1102 * Enum representing the order in which varyings are packed within a
1105 * Currently we pack vec4's first, then vec2's, then scalar values, then
1106 * vec3's. This order ensures that the only vectors that are at risk of
1107 * having to be "double parked" (split between two adjacent varying slots)
1110 enum packing_order_enum
{
1113 PACKING_ORDER_SCALAR
,
1117 static unsigned compute_packing_class(const ir_variable
*var
);
1118 static packing_order_enum
compute_packing_order(const ir_variable
*var
);
1119 static int match_comparator(const void *x_generic
, const void *y_generic
);
1120 static int xfb_comparator(const void *x_generic
, const void *y_generic
);
1123 * Structure recording the relationship between a single producer output
1124 * and a single consumer input.
1128 * Packing class for this varying, computed by compute_packing_class().
1130 unsigned packing_class
;
1133 * Packing order for this varying, computed by compute_packing_order().
1135 packing_order_enum packing_order
;
1136 unsigned num_components
;
1139 * The output variable in the producer stage.
1141 ir_variable
*producer_var
;
1144 * The input variable in the consumer stage.
1146 ir_variable
*consumer_var
;
1149 * The location which has been assigned for this varying. This is
1150 * expressed in multiples of a float, with the first generic varying
1151 * (i.e. the one referred to by VARYING_SLOT_VAR0) represented by the
1154 unsigned generic_location
;
1158 * The number of elements in the \c matches array that are currently in
1161 unsigned num_matches
;
1164 * The number of elements that were set aside for the \c matches array when
1167 unsigned matches_capacity
;
1169 gl_shader_stage producer_stage
;
1170 gl_shader_stage consumer_stage
;
1173 } /* anonymous namespace */
1175 varying_matches::varying_matches(bool disable_varying_packing
,
1177 gl_shader_stage producer_stage
,
1178 gl_shader_stage consumer_stage
)
1179 : disable_varying_packing(disable_varying_packing
),
1180 xfb_enabled(xfb_enabled
),
1181 producer_stage(producer_stage
),
1182 consumer_stage(consumer_stage
)
1184 /* Note: this initial capacity is rather arbitrarily chosen to be large
1185 * enough for many cases without wasting an unreasonable amount of space.
1186 * varying_matches::record() will resize the array if there are more than
1187 * this number of varyings.
1189 this->matches_capacity
= 8;
1190 this->matches
= (match
*)
1191 malloc(sizeof(*this->matches
) * this->matches_capacity
);
1192 this->num_matches
= 0;
1196 varying_matches::~varying_matches()
1198 free(this->matches
);
1203 * Packing is always safe on individual arrays, structure and matices. It is
1204 * also safe if the varying is only used for transform feedback.
1207 varying_matches::is_varying_packing_safe(const glsl_type
*type
,
1208 const ir_variable
*var
)
1210 if (consumer_stage
== MESA_SHADER_TESS_EVAL
||
1211 consumer_stage
== MESA_SHADER_TESS_CTRL
||
1212 producer_stage
== MESA_SHADER_TESS_CTRL
)
1215 return xfb_enabled
&& (type
->is_array() || type
->is_record() ||
1216 type
->is_matrix() || var
->data
.is_xfb_only
);
1221 * Record the given producer/consumer variable pair in the list of variables
1222 * that should later be assigned locations.
1224 * It is permissible for \c consumer_var to be NULL (this happens if a
1225 * variable is output by the producer and consumed by transform feedback, but
1226 * not consumed by the consumer).
1228 * If \c producer_var has already been paired up with a consumer_var, or
1229 * producer_var is part of fixed pipeline functionality (and hence already has
1230 * a location assigned), this function has no effect.
1232 * Note: as a side effect this function may change the interpolation type of
1233 * \c producer_var, but only when the change couldn't possibly affect
1237 varying_matches::record(ir_variable
*producer_var
, ir_variable
*consumer_var
)
1239 assert(producer_var
!= NULL
|| consumer_var
!= NULL
);
1241 if ((producer_var
&& (!producer_var
->data
.is_unmatched_generic_inout
||
1242 producer_var
->data
.explicit_location
)) ||
1243 (consumer_var
&& (!consumer_var
->data
.is_unmatched_generic_inout
||
1244 consumer_var
->data
.explicit_location
))) {
1245 /* Either a location already exists for this variable (since it is part
1246 * of fixed functionality), or it has already been recorded as part of a
1252 bool needs_flat_qualifier
= consumer_var
== NULL
&&
1253 (producer_var
->type
->contains_integer() ||
1254 producer_var
->type
->contains_double());
1256 if (needs_flat_qualifier
||
1257 (consumer_stage
!= -1 && consumer_stage
!= MESA_SHADER_FRAGMENT
)) {
1258 /* Since this varying is not being consumed by the fragment shader, its
1259 * interpolation type varying cannot possibly affect rendering.
1260 * Also, this variable is non-flat and is (or contains) an integer
1262 * If the consumer stage is unknown, don't modify the interpolation
1263 * type as it could affect rendering later with separate shaders.
1265 * lower_packed_varyings requires all integer varyings to flat,
1266 * regardless of where they appear. We can trivially satisfy that
1267 * requirement by changing the interpolation type to flat here.
1270 producer_var
->data
.centroid
= false;
1271 producer_var
->data
.sample
= false;
1272 producer_var
->data
.interpolation
= INTERP_QUALIFIER_FLAT
;
1276 consumer_var
->data
.centroid
= false;
1277 consumer_var
->data
.sample
= false;
1278 consumer_var
->data
.interpolation
= INTERP_QUALIFIER_FLAT
;
1282 if (this->num_matches
== this->matches_capacity
) {
1283 this->matches_capacity
*= 2;
1284 this->matches
= (match
*)
1285 realloc(this->matches
,
1286 sizeof(*this->matches
) * this->matches_capacity
);
1289 const ir_variable
*const var
= (producer_var
!= NULL
)
1290 ? producer_var
: consumer_var
;
1291 const gl_shader_stage stage
= (producer_var
!= NULL
)
1292 ? producer_stage
: consumer_stage
;
1293 const glsl_type
*type
= get_varying_type(var
, stage
);
1295 this->matches
[this->num_matches
].packing_class
1296 = this->compute_packing_class(var
);
1297 this->matches
[this->num_matches
].packing_order
1298 = this->compute_packing_order(var
);
1299 if (this->disable_varying_packing
&& !is_varying_packing_safe(type
, var
)) {
1300 unsigned slots
= type
->count_attribute_slots(false);
1301 this->matches
[this->num_matches
].num_components
= slots
* 4;
1303 this->matches
[this->num_matches
].num_components
1304 = type
->component_slots();
1306 this->matches
[this->num_matches
].producer_var
= producer_var
;
1307 this->matches
[this->num_matches
].consumer_var
= consumer_var
;
1308 this->num_matches
++;
1310 producer_var
->data
.is_unmatched_generic_inout
= 0;
1312 consumer_var
->data
.is_unmatched_generic_inout
= 0;
1317 * Choose locations for all of the variable matches that were previously
1318 * passed to varying_matches::record().
1321 varying_matches::assign_locations(struct gl_shader_program
*prog
,
1322 uint64_t reserved_slots
,
1323 bool separate_shader
)
1325 /* If packing has been disabled then we cannot safely sort the varyings by
1326 * class as it may mean we are using a version of OpenGL where
1327 * interpolation qualifiers are not guaranteed to be matching across
1328 * shaders, sorting in this case could result in mismatching shader
1330 * When packing is disabled the sort orders varyings used by transform
1331 * feedback first, but also depends on *undefined behaviour* of qsort to
1332 * reverse the order of the varyings. See: xfb_comparator().
1334 if (!this->disable_varying_packing
) {
1335 /* Sort varying matches into an order that makes them easy to pack. */
1336 qsort(this->matches
, this->num_matches
, sizeof(*this->matches
),
1337 &varying_matches::match_comparator
);
1339 /* Only sort varyings that are only used by transform feedback. */
1340 qsort(this->matches
, this->num_matches
, sizeof(*this->matches
),
1341 &varying_matches::xfb_comparator
);
1344 unsigned generic_location
= 0;
1345 unsigned generic_patch_location
= MAX_VARYING
*4;
1346 bool previous_var_xfb_only
= false;
1348 for (unsigned i
= 0; i
< this->num_matches
; i
++) {
1349 unsigned *location
= &generic_location
;
1351 const ir_variable
*var
;
1352 const glsl_type
*type
;
1353 bool is_vertex_input
= false;
1354 if (matches
[i
].consumer_var
) {
1355 var
= matches
[i
].consumer_var
;
1356 type
= get_varying_type(var
, consumer_stage
);
1357 if (consumer_stage
== MESA_SHADER_VERTEX
)
1358 is_vertex_input
= true;
1360 var
= matches
[i
].producer_var
;
1361 type
= get_varying_type(var
, producer_stage
);
1364 if (var
->data
.patch
)
1365 location
= &generic_patch_location
;
1367 /* Advance to the next slot if this varying has a different packing
1368 * class than the previous one, and we're not already on a slot
1371 * Also advance to the next slot if packing is disabled. This makes sure
1372 * we don't assign varyings the same locations which is possible
1373 * because we still pack individual arrays, records and matrices even
1374 * when packing is disabled. Note we don't advance to the next slot if
1375 * we can pack varyings together that are only used for transform
1378 if ((this->disable_varying_packing
&&
1379 !(previous_var_xfb_only
&& var
->data
.is_xfb_only
)) ||
1380 (i
> 0 && this->matches
[i
- 1].packing_class
1381 != this->matches
[i
].packing_class
)) {
1382 *location
= ALIGN(*location
, 4);
1385 previous_var_xfb_only
= var
->data
.is_xfb_only
;
1387 unsigned num_elements
= type
->count_attribute_slots(is_vertex_input
);
1389 if (this->disable_varying_packing
&&
1390 !is_varying_packing_safe(type
, var
))
1393 slot_end
= type
->without_array()->vector_elements
;
1394 slot_end
+= *location
- 1;
1396 /* FIXME: We could be smarter in the below code and loop back over
1397 * trying to fill any locations that we skipped because we couldn't pack
1398 * the varying between an explicit location. For now just let the user
1399 * hit the linking error if we run out of room and suggest they use
1400 * explicit locations.
1402 for (unsigned j
= 0; j
< num_elements
; j
++) {
1403 while ((slot_end
< MAX_VARYING
* 4u) &&
1404 ((reserved_slots
& (UINT64_C(1) << *location
/ 4u) ||
1405 (reserved_slots
& (UINT64_C(1) << slot_end
/ 4u))))) {
1407 *location
= ALIGN(*location
+ 1, 4);
1408 slot_end
= *location
;
1410 /* reset the counter and try again */
1414 /* Increase the slot to make sure there is enough room for next
1417 if (this->disable_varying_packing
&&
1418 !is_varying_packing_safe(type
, var
))
1421 slot_end
+= type
->without_array()->vector_elements
;
1424 if (!var
->data
.patch
&& *location
>= MAX_VARYING
* 4u) {
1425 linker_error(prog
, "insufficient contiguous locations available for "
1426 "%s it is possible an array or struct could not be "
1427 "packed between varyings with explicit locations. Try "
1428 "using an explicit location for arrays and structs.",
1432 this->matches
[i
].generic_location
= *location
;
1434 *location
+= this->matches
[i
].num_components
;
1437 return (generic_location
+ 3) / 4;
1442 * Update the producer and consumer shaders to reflect the locations
1443 * assignments that were made by varying_matches::assign_locations().
1446 varying_matches::store_locations() const
1448 for (unsigned i
= 0; i
< this->num_matches
; i
++) {
1449 ir_variable
*producer_var
= this->matches
[i
].producer_var
;
1450 ir_variable
*consumer_var
= this->matches
[i
].consumer_var
;
1451 unsigned generic_location
= this->matches
[i
].generic_location
;
1452 unsigned slot
= generic_location
/ 4;
1453 unsigned offset
= generic_location
% 4;
1456 producer_var
->data
.location
= VARYING_SLOT_VAR0
+ slot
;
1457 producer_var
->data
.location_frac
= offset
;
1461 assert(consumer_var
->data
.location
== -1);
1462 consumer_var
->data
.location
= VARYING_SLOT_VAR0
+ slot
;
1463 consumer_var
->data
.location_frac
= offset
;
1470 * Compute the "packing class" of the given varying. This is an unsigned
1471 * integer with the property that two variables in the same packing class can
1472 * be safely backed into the same vec4.
1475 varying_matches::compute_packing_class(const ir_variable
*var
)
1477 /* Without help from the back-end, there is no way to pack together
1478 * variables with different interpolation types, because
1479 * lower_packed_varyings must choose exactly one interpolation type for
1480 * each packed varying it creates.
1482 * However, we can safely pack together floats, ints, and uints, because:
1484 * - varyings of base type "int" and "uint" must use the "flat"
1485 * interpolation type, which can only occur in GLSL 1.30 and above.
1487 * - On platforms that support GLSL 1.30 and above, lower_packed_varyings
1488 * can store flat floats as ints without losing any information (using
1489 * the ir_unop_bitcast_* opcodes).
1491 * Therefore, the packing class depends only on the interpolation type.
1493 unsigned packing_class
= var
->data
.centroid
| (var
->data
.sample
<< 1) |
1494 (var
->data
.patch
<< 2);
1496 packing_class
+= var
->data
.interpolation
;
1497 return packing_class
;
1502 * Compute the "packing order" of the given varying. This is a sort key we
1503 * use to determine when to attempt to pack the given varying relative to
1504 * other varyings in the same packing class.
1506 varying_matches::packing_order_enum
1507 varying_matches::compute_packing_order(const ir_variable
*var
)
1509 const glsl_type
*element_type
= var
->type
;
1511 while (element_type
->base_type
== GLSL_TYPE_ARRAY
) {
1512 element_type
= element_type
->fields
.array
;
1515 switch (element_type
->component_slots() % 4) {
1516 case 1: return PACKING_ORDER_SCALAR
;
1517 case 2: return PACKING_ORDER_VEC2
;
1518 case 3: return PACKING_ORDER_VEC3
;
1519 case 0: return PACKING_ORDER_VEC4
;
1521 assert(!"Unexpected value of vector_elements");
1522 return PACKING_ORDER_VEC4
;
1528 * Comparison function passed to qsort() to sort varyings by packing_class and
1529 * then by packing_order.
1532 varying_matches::match_comparator(const void *x_generic
, const void *y_generic
)
1534 const match
*x
= (const match
*) x_generic
;
1535 const match
*y
= (const match
*) y_generic
;
1537 if (x
->packing_class
!= y
->packing_class
)
1538 return x
->packing_class
- y
->packing_class
;
1539 return x
->packing_order
- y
->packing_order
;
1544 * Comparison function passed to qsort() to sort varyings used only by
1545 * transform feedback when packing of other varyings is disabled.
1548 varying_matches::xfb_comparator(const void *x_generic
, const void *y_generic
)
1550 const match
*x
= (const match
*) x_generic
;
1552 if (x
->producer_var
!= NULL
&& x
->producer_var
->data
.is_xfb_only
)
1553 return match_comparator(x_generic
, y_generic
);
1555 /* FIXME: When the comparator returns 0 it means the elements being
1556 * compared are equivalent. However the qsort documentation says:
1558 * "The order of equivalent elements is undefined."
1560 * In practice the sort ends up reversing the order of the varyings which
1561 * means locations are also assigned in this reversed order and happens to
1562 * be what we want. This is also whats happening in
1563 * varying_matches::match_comparator().
1570 * Is the given variable a varying variable to be counted against the
1571 * limit in ctx->Const.MaxVarying?
1572 * This includes variables such as texcoords, colors and generic
1573 * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord.
1576 var_counts_against_varying_limit(gl_shader_stage stage
, const ir_variable
*var
)
1578 /* Only fragment shaders will take a varying variable as an input */
1579 if (stage
== MESA_SHADER_FRAGMENT
&&
1580 var
->data
.mode
== ir_var_shader_in
) {
1581 switch (var
->data
.location
) {
1582 case VARYING_SLOT_POS
:
1583 case VARYING_SLOT_FACE
:
1584 case VARYING_SLOT_PNTC
:
1595 * Visitor class that generates tfeedback_candidate structs describing all
1596 * possible targets of transform feedback.
1598 * tfeedback_candidate structs are stored in the hash table
1599 * tfeedback_candidates, which is passed to the constructor. This hash table
1600 * maps varying names to instances of the tfeedback_candidate struct.
1602 class tfeedback_candidate_generator
: public program_resource_visitor
1605 tfeedback_candidate_generator(void *mem_ctx
,
1606 hash_table
*tfeedback_candidates
)
1608 tfeedback_candidates(tfeedback_candidates
),
1614 void process(ir_variable
*var
)
1616 /* All named varying interface blocks should be flattened by now */
1617 assert(!var
->is_interface_instance());
1619 this->toplevel_var
= var
;
1620 this->varying_floats
= 0;
1621 program_resource_visitor::process(var
);
1625 virtual void visit_field(const glsl_type
*type
, const char *name
,
1628 assert(!type
->without_array()->is_record());
1629 assert(!type
->without_array()->is_interface());
1633 tfeedback_candidate
*candidate
1634 = rzalloc(this->mem_ctx
, tfeedback_candidate
);
1635 candidate
->toplevel_var
= this->toplevel_var
;
1636 candidate
->type
= type
;
1637 candidate
->offset
= this->varying_floats
;
1638 hash_table_insert(this->tfeedback_candidates
, candidate
,
1639 ralloc_strdup(this->mem_ctx
, name
));
1640 this->varying_floats
+= type
->component_slots();
1644 * Memory context used to allocate hash table keys and values.
1646 void * const mem_ctx
;
1649 * Hash table in which tfeedback_candidate objects should be stored.
1651 hash_table
* const tfeedback_candidates
;
1654 * Pointer to the toplevel variable that is being traversed.
1656 ir_variable
*toplevel_var
;
1659 * Total number of varying floats that have been visited so far. This is
1660 * used to determine the offset to each varying within the toplevel
1663 unsigned varying_floats
;
1670 populate_consumer_input_sets(void *mem_ctx
, exec_list
*ir
,
1671 hash_table
*consumer_inputs
,
1672 hash_table
*consumer_interface_inputs
,
1673 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
])
1675 memset(consumer_inputs_with_locations
,
1677 sizeof(consumer_inputs_with_locations
[0]) * VARYING_SLOT_TESS_MAX
);
1679 foreach_in_list(ir_instruction
, node
, ir
) {
1680 ir_variable
*const input_var
= node
->as_variable();
1682 if ((input_var
!= NULL
) && (input_var
->data
.mode
== ir_var_shader_in
)) {
1683 /* All interface blocks should have been lowered by this point */
1684 assert(!input_var
->type
->is_interface());
1686 if (input_var
->data
.explicit_location
) {
1687 /* assign_varying_locations only cares about finding the
1688 * ir_variable at the start of a contiguous location block.
1690 * - For !producer, consumer_inputs_with_locations isn't used.
1692 * - For !consumer, consumer_inputs_with_locations is empty.
1694 * For consumer && producer, if you were trying to set some
1695 * ir_variable to the middle of a location block on the other side
1696 * of producer/consumer, cross_validate_outputs_to_inputs() should
1697 * be link-erroring due to either type mismatch or location
1698 * overlaps. If the variables do match up, then they've got a
1699 * matching data.location and you only looked at
1700 * consumer_inputs_with_locations[var->data.location], not any
1701 * following entries for the array/structure.
1703 consumer_inputs_with_locations
[input_var
->data
.location
] =
1705 } else if (input_var
->get_interface_type() != NULL
) {
1706 char *const iface_field_name
=
1707 ralloc_asprintf(mem_ctx
, "%s.%s",
1708 input_var
->get_interface_type()->without_array()->name
,
1710 hash_table_insert(consumer_interface_inputs
, input_var
,
1713 hash_table_insert(consumer_inputs
, input_var
,
1714 ralloc_strdup(mem_ctx
, input_var
->name
));
1721 * Find a variable from the consumer that "matches" the specified variable
1723 * This function only finds inputs with names that match. There is no
1724 * validation (here) that the types, etc. are compatible.
1727 get_matching_input(void *mem_ctx
,
1728 const ir_variable
*output_var
,
1729 hash_table
*consumer_inputs
,
1730 hash_table
*consumer_interface_inputs
,
1731 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
])
1733 ir_variable
*input_var
;
1735 if (output_var
->data
.explicit_location
) {
1736 input_var
= consumer_inputs_with_locations
[output_var
->data
.location
];
1737 } else if (output_var
->get_interface_type() != NULL
) {
1738 char *const iface_field_name
=
1739 ralloc_asprintf(mem_ctx
, "%s.%s",
1740 output_var
->get_interface_type()->without_array()->name
,
1743 (ir_variable
*) hash_table_find(consumer_interface_inputs
,
1747 (ir_variable
*) hash_table_find(consumer_inputs
, output_var
->name
);
1750 return (input_var
== NULL
|| input_var
->data
.mode
!= ir_var_shader_in
)
1757 io_variable_cmp(const void *_a
, const void *_b
)
1759 const ir_variable
*const a
= *(const ir_variable
**) _a
;
1760 const ir_variable
*const b
= *(const ir_variable
**) _b
;
1762 if (a
->data
.explicit_location
&& b
->data
.explicit_location
)
1763 return b
->data
.location
- a
->data
.location
;
1765 if (a
->data
.explicit_location
&& !b
->data
.explicit_location
)
1768 if (!a
->data
.explicit_location
&& b
->data
.explicit_location
)
1771 return -strcmp(a
->name
, b
->name
);
1775 * Sort the shader IO variables into canonical order
1778 canonicalize_shader_io(exec_list
*ir
, enum ir_variable_mode io_mode
)
1780 ir_variable
*var_table
[MAX_PROGRAM_OUTPUTS
* 4];
1781 unsigned num_variables
= 0;
1783 foreach_in_list(ir_instruction
, node
, ir
) {
1784 ir_variable
*const var
= node
->as_variable();
1786 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
1789 /* If we have already encountered more I/O variables that could
1790 * successfully link, bail.
1792 if (num_variables
== ARRAY_SIZE(var_table
))
1795 var_table
[num_variables
++] = var
;
1798 if (num_variables
== 0)
1801 /* Sort the list in reverse order (io_variable_cmp handles this). Later
1802 * we're going to push the variables on to the IR list as a stack, so we
1803 * want the last variable (in canonical order) to be first in the list.
1805 qsort(var_table
, num_variables
, sizeof(var_table
[0]), io_variable_cmp
);
1807 /* Remove the variable from it's current location in the IR, and put it at
1810 for (unsigned i
= 0; i
< num_variables
; i
++) {
1811 var_table
[i
]->remove();
1812 ir
->push_head(var_table
[i
]);
1817 * Generate a bitfield map of the explicit locations for shader varyings.
1819 * In theory a 32 bits value will be enough but a 64 bits value is future proof.
1822 reserved_varying_slot(struct gl_shader
*stage
, ir_variable_mode io_mode
)
1824 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
1825 assert(MAX_VARYING
<= 64); /* avoid an overflow of the returned value */
1833 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
1834 ir_variable
*const var
= node
->as_variable();
1836 if (var
== NULL
|| var
->data
.mode
!= io_mode
||
1837 !var
->data
.explicit_location
||
1838 var
->data
.location
< VARYING_SLOT_VAR0
)
1841 var_slot
= var
->data
.location
- VARYING_SLOT_VAR0
;
1843 unsigned num_elements
= get_varying_type(var
, stage
->Stage
)
1844 ->count_attribute_slots(stage
->Stage
== MESA_SHADER_VERTEX
);
1845 for (unsigned i
= 0; i
< num_elements
; i
++) {
1846 if (var_slot
>= 0 && var_slot
< MAX_VARYING
)
1847 slots
|= UINT64_C(1) << var_slot
;
1857 * Assign locations for all variables that are produced in one pipeline stage
1858 * (the "producer") and consumed in the next stage (the "consumer").
1860 * Variables produced by the producer may also be consumed by transform
1863 * \param num_tfeedback_decls is the number of declarations indicating
1864 * variables that may be consumed by transform feedback.
1866 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
1867 * representing the result of parsing the strings passed to
1868 * glTransformFeedbackVaryings(). assign_location() will be called for
1869 * each of these objects that matches one of the outputs of the
1872 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
1873 * be NULL. In this case, varying locations are assigned solely based on the
1874 * requirements of transform feedback.
1877 assign_varying_locations(struct gl_context
*ctx
,
1879 struct gl_shader_program
*prog
,
1880 gl_shader
*producer
, gl_shader
*consumer
,
1881 unsigned num_tfeedback_decls
,
1882 tfeedback_decl
*tfeedback_decls
)
1884 /* Tessellation shaders treat inputs and outputs as shared memory and can
1885 * access inputs and outputs of other invocations.
1886 * Therefore, they can't be lowered to temps easily (and definitely not
1889 bool unpackable_tess
=
1890 (consumer
&& consumer
->Stage
== MESA_SHADER_TESS_EVAL
) ||
1891 (consumer
&& consumer
->Stage
== MESA_SHADER_TESS_CTRL
) ||
1892 (producer
&& producer
->Stage
== MESA_SHADER_TESS_CTRL
);
1894 /* Transform feedback code assumes varying arrays are packed, so if the
1895 * driver has disabled varying packing, make sure to at least enable
1896 * packing required by transform feedback.
1899 ctx
->Extensions
.EXT_transform_feedback
&& !unpackable_tess
;
1901 /* Disable varying packing for GL 4.4+ as there is no guarantee
1902 * that interpolation qualifiers will match between shaders in these
1903 * versions. We also disable packing on outerward facing interfaces for
1904 * SSO because in ES we need to retain the unpacked varying information
1905 * for draw time validation. For desktop GL we could allow packing for
1906 * versions < 4.4 but its just safer not to do packing.
1908 * Packing is still enabled on individual arrays, structs, and matrices as
1909 * these are required by the transform feedback code and it is still safe
1910 * to do so. We also enable packing when a varying is only used for
1911 * transform feedback and its not a SSO.
1913 * Varying packing currently only packs together varyings with matching
1914 * interpolation qualifiers as the backends assume all packed components
1915 * are to be processed in the same way. Therefore we cannot do packing in
1916 * these versions of GL without the risk of mismatching interfaces.
1918 * From Section 4.5 (Interpolation Qualifiers) of the GLSL 4.30 spec:
1920 * "The type and presence of interpolation qualifiers of variables with
1921 * the same name declared in all linked shaders for the same cross-stage
1922 * interface must match, otherwise the link command will fail.
1924 * When comparing an output from one stage to an input of a subsequent
1925 * stage, the input and output don't match if their interpolation
1926 * qualifiers (or lack thereof) are not the same."
1928 * This text was also in at least revison 7 of the 4.40 spec but is no
1929 * longer in revision 9 and not in the 4.50 spec.
1931 bool disable_varying_packing
=
1932 ctx
->Const
.DisableVaryingPacking
|| unpackable_tess
;
1933 if ((ctx
->API
== API_OPENGL_CORE
&& ctx
->Version
>= 44) ||
1934 (prog
->SeparateShader
&& (producer
== NULL
|| consumer
== NULL
)))
1935 disable_varying_packing
= true;
1937 varying_matches
matches(disable_varying_packing
, xfb_enabled
,
1938 producer
? producer
->Stage
: (gl_shader_stage
)-1,
1939 consumer
? consumer
->Stage
: (gl_shader_stage
)-1);
1940 hash_table
*tfeedback_candidates
1941 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1942 hash_table
*consumer_inputs
1943 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1944 hash_table
*consumer_interface_inputs
1945 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1946 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
] = {
1950 unsigned consumer_vertices
= 0;
1951 if (consumer
&& consumer
->Stage
== MESA_SHADER_GEOMETRY
)
1952 consumer_vertices
= prog
->Geom
.VerticesIn
;
1954 /* Operate in a total of four passes.
1956 * 1. Sort inputs / outputs into a canonical order. This is necessary so
1957 * that inputs / outputs of separable shaders will be assigned
1958 * predictable locations regardless of the order in which declarations
1959 * appeared in the shader source.
1961 * 2. Assign locations for any matching inputs and outputs.
1963 * 3. Mark output variables in the producer that do not have locations as
1964 * not being outputs. This lets the optimizer eliminate them.
1966 * 4. Mark input variables in the consumer that do not have locations as
1967 * not being inputs. This lets the optimizer eliminate them.
1970 canonicalize_shader_io(consumer
->ir
, ir_var_shader_in
);
1973 canonicalize_shader_io(producer
->ir
, ir_var_shader_out
);
1976 linker::populate_consumer_input_sets(mem_ctx
, consumer
->ir
,
1978 consumer_interface_inputs
,
1979 consumer_inputs_with_locations
);
1982 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
1983 ir_variable
*const output_var
= node
->as_variable();
1985 if ((output_var
== NULL
) ||
1986 (output_var
->data
.mode
!= ir_var_shader_out
))
1989 /* Only geometry shaders can use non-zero streams */
1990 assert(output_var
->data
.stream
== 0 ||
1991 (output_var
->data
.stream
< MAX_VERTEX_STREAMS
&&
1992 producer
->Stage
== MESA_SHADER_GEOMETRY
));
1994 if (num_tfeedback_decls
> 0) {
1995 tfeedback_candidate_generator
g(mem_ctx
, tfeedback_candidates
);
1996 g
.process(output_var
);
1999 ir_variable
*const input_var
=
2000 linker::get_matching_input(mem_ctx
, output_var
, consumer_inputs
,
2001 consumer_interface_inputs
,
2002 consumer_inputs_with_locations
);
2004 /* If a matching input variable was found, add this ouptut (and the
2005 * input) to the set. If this is a separable program and there is no
2006 * consumer stage, add the output.
2008 * Always add TCS outputs. They are shared by all invocations
2009 * within a patch and can be used as shared memory.
2011 if (input_var
|| (prog
->SeparateShader
&& consumer
== NULL
) ||
2012 producer
->Type
== GL_TESS_CONTROL_SHADER
) {
2013 matches
.record(output_var
, input_var
);
2016 /* Only stream 0 outputs can be consumed in the next stage */
2017 if (input_var
&& output_var
->data
.stream
!= 0) {
2018 linker_error(prog
, "output %s is assigned to stream=%d but "
2019 "is linked to an input, which requires stream=0",
2020 output_var
->name
, output_var
->data
.stream
);
2025 /* If there's no producer stage, then this must be a separable program.
2026 * For example, we may have a program that has just a fragment shader.
2027 * Later this program will be used with some arbitrary vertex (or
2028 * geometry) shader program. This means that locations must be assigned
2029 * for all the inputs.
2031 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2032 ir_variable
*const input_var
= node
->as_variable();
2034 if ((input_var
== NULL
) ||
2035 (input_var
->data
.mode
!= ir_var_shader_in
))
2038 matches
.record(NULL
, input_var
);
2042 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2043 if (!tfeedback_decls
[i
].is_varying())
2046 const tfeedback_candidate
*matched_candidate
2047 = tfeedback_decls
[i
].find_candidate(prog
, tfeedback_candidates
);
2049 if (matched_candidate
== NULL
) {
2050 hash_table_dtor(tfeedback_candidates
);
2051 hash_table_dtor(consumer_inputs
);
2052 hash_table_dtor(consumer_interface_inputs
);
2056 if (matched_candidate
->toplevel_var
->data
.is_unmatched_generic_inout
) {
2057 matched_candidate
->toplevel_var
->data
.is_xfb_only
= 1;
2058 matches
.record(matched_candidate
->toplevel_var
, NULL
);
2062 const uint64_t reserved_slots
=
2063 reserved_varying_slot(producer
, ir_var_shader_out
) |
2064 reserved_varying_slot(consumer
, ir_var_shader_in
);
2066 const unsigned slots_used
= matches
.assign_locations(prog
, reserved_slots
,
2067 prog
->SeparateShader
);
2068 matches
.store_locations();
2070 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2071 if (!tfeedback_decls
[i
].is_varying())
2074 if (!tfeedback_decls
[i
].assign_location(ctx
, prog
)) {
2075 hash_table_dtor(tfeedback_candidates
);
2076 hash_table_dtor(consumer_inputs
);
2077 hash_table_dtor(consumer_interface_inputs
);
2082 hash_table_dtor(tfeedback_candidates
);
2083 hash_table_dtor(consumer_inputs
);
2084 hash_table_dtor(consumer_interface_inputs
);
2086 if (consumer
&& producer
) {
2087 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2088 ir_variable
*const var
= node
->as_variable();
2090 if (var
&& var
->data
.mode
== ir_var_shader_in
&&
2091 var
->data
.is_unmatched_generic_inout
) {
2092 if (!prog
->IsES
&& prog
->Version
<= 120) {
2093 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
2095 * Only those varying variables used (i.e. read) in
2096 * the fragment shader executable must be written to
2097 * by the vertex shader executable; declaring
2098 * superfluous varying variables in a vertex shader is
2101 * We interpret this text as meaning that the VS must
2102 * write the variable for the FS to read it. See
2103 * "glsl1-varying read but not written" in piglit.
2105 linker_error(prog
, "%s shader varying %s not written "
2107 _mesa_shader_stage_to_string(consumer
->Stage
),
2109 _mesa_shader_stage_to_string(producer
->Stage
));
2111 linker_warning(prog
, "%s shader varying %s not written "
2113 _mesa_shader_stage_to_string(consumer
->Stage
),
2115 _mesa_shader_stage_to_string(producer
->Stage
));
2120 /* Now that validation is done its safe to remove unused varyings. As
2121 * we have both a producer and consumer its safe to remove unused
2122 * varyings even if the program is a SSO because the stages are being
2123 * linked together i.e. we have a multi-stage SSO.
2125 remove_unused_shader_inputs_and_outputs(false, producer
,
2127 remove_unused_shader_inputs_and_outputs(false, consumer
,
2132 lower_packed_varyings(mem_ctx
, slots_used
, ir_var_shader_out
,
2133 0, producer
, disable_varying_packing
,
2138 lower_packed_varyings(mem_ctx
, slots_used
, ir_var_shader_in
,
2139 consumer_vertices
, consumer
,
2140 disable_varying_packing
, xfb_enabled
);
2147 check_against_output_limit(struct gl_context
*ctx
,
2148 struct gl_shader_program
*prog
,
2149 gl_shader
*producer
)
2151 unsigned output_vectors
= 0;
2153 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2154 ir_variable
*const var
= node
->as_variable();
2156 if (var
&& var
->data
.mode
== ir_var_shader_out
&&
2157 var_counts_against_varying_limit(producer
->Stage
, var
)) {
2158 /* outputs for fragment shader can't be doubles */
2159 output_vectors
+= var
->type
->count_attribute_slots(false);
2163 assert(producer
->Stage
!= MESA_SHADER_FRAGMENT
);
2164 unsigned max_output_components
=
2165 ctx
->Const
.Program
[producer
->Stage
].MaxOutputComponents
;
2167 const unsigned output_components
= output_vectors
* 4;
2168 if (output_components
> max_output_components
) {
2169 if (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)
2170 linker_error(prog
, "%s shader uses too many output vectors "
2172 _mesa_shader_stage_to_string(producer
->Stage
),
2174 max_output_components
/ 4);
2176 linker_error(prog
, "%s shader uses too many output components "
2178 _mesa_shader_stage_to_string(producer
->Stage
),
2180 max_output_components
);
2189 check_against_input_limit(struct gl_context
*ctx
,
2190 struct gl_shader_program
*prog
,
2191 gl_shader
*consumer
)
2193 unsigned input_vectors
= 0;
2195 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2196 ir_variable
*const var
= node
->as_variable();
2198 if (var
&& var
->data
.mode
== ir_var_shader_in
&&
2199 var_counts_against_varying_limit(consumer
->Stage
, var
)) {
2200 /* vertex inputs aren't varying counted */
2201 input_vectors
+= var
->type
->count_attribute_slots(false);
2205 assert(consumer
->Stage
!= MESA_SHADER_VERTEX
);
2206 unsigned max_input_components
=
2207 ctx
->Const
.Program
[consumer
->Stage
].MaxInputComponents
;
2209 const unsigned input_components
= input_vectors
* 4;
2210 if (input_components
> max_input_components
) {
2211 if (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)
2212 linker_error(prog
, "%s shader uses too many input vectors "
2214 _mesa_shader_stage_to_string(consumer
->Stage
),
2216 max_input_components
/ 4);
2218 linker_error(prog
, "%s shader uses too many input components "
2220 _mesa_shader_stage_to_string(consumer
->Stage
),
2222 max_input_components
);