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.
244 /* According to the OpenGL and OpenGLES GLSL specs, the centroid qualifier
245 * should match until OpenGL 4.3 and OpenGLES 3.1. The OpenGLES 3.0
246 * conformance test suite does not verify that the qualifiers must match.
247 * The deqp test suite expects the opposite (OpenGLES 3.1) behavior for
248 * OpenGLES 3.0 drivers, so we relax the checking in all cases.
250 if (false /* always skip the centroid check */ &&
251 prog
->Version
< (prog
->IsES
? 310 : 430) &&
252 input
->data
.centroid
!= output
->data
.centroid
) {
254 "%s shader output `%s' %s centroid qualifier, "
255 "but %s shader input %s centroid qualifier\n",
256 _mesa_shader_stage_to_string(producer_stage
),
258 (output
->data
.centroid
) ? "has" : "lacks",
259 _mesa_shader_stage_to_string(consumer_stage
),
260 (input
->data
.centroid
) ? "has" : "lacks");
264 if (input
->data
.sample
!= output
->data
.sample
) {
266 "%s shader output `%s' %s sample qualifier, "
267 "but %s shader input %s sample qualifier\n",
268 _mesa_shader_stage_to_string(producer_stage
),
270 (output
->data
.sample
) ? "has" : "lacks",
271 _mesa_shader_stage_to_string(consumer_stage
),
272 (input
->data
.sample
) ? "has" : "lacks");
276 if (input
->data
.patch
!= output
->data
.patch
) {
278 "%s shader output `%s' %s patch qualifier, "
279 "but %s shader input %s patch qualifier\n",
280 _mesa_shader_stage_to_string(producer_stage
),
282 (output
->data
.patch
) ? "has" : "lacks",
283 _mesa_shader_stage_to_string(consumer_stage
),
284 (input
->data
.patch
) ? "has" : "lacks");
288 if (!prog
->IsES
&& input
->data
.invariant
!= output
->data
.invariant
) {
290 "%s shader output `%s' %s invariant qualifier, "
291 "but %s shader input %s invariant qualifier\n",
292 _mesa_shader_stage_to_string(producer_stage
),
294 (output
->data
.invariant
) ? "has" : "lacks",
295 _mesa_shader_stage_to_string(consumer_stage
),
296 (input
->data
.invariant
) ? "has" : "lacks");
300 /* GLSL >= 4.40 removes text requiring interpolation qualifiers
301 * to match cross stage, they must only match within the same stage.
303 * From page 84 (page 90 of the PDF) of the GLSL 4.40 spec:
305 * "It is a link-time error if, within the same stage, the interpolation
306 * qualifiers of variables of the same name do not match.
309 if (input
->data
.interpolation
!= output
->data
.interpolation
&&
310 prog
->Version
< 440) {
312 "%s shader output `%s' specifies %s "
313 "interpolation qualifier, "
314 "but %s shader input specifies %s "
315 "interpolation qualifier\n",
316 _mesa_shader_stage_to_string(producer_stage
),
318 interpolation_string(output
->data
.interpolation
),
319 _mesa_shader_stage_to_string(consumer_stage
),
320 interpolation_string(input
->data
.interpolation
));
326 * Validate front and back color outputs against single color input
329 cross_validate_front_and_back_color(struct gl_shader_program
*prog
,
330 const ir_variable
*input
,
331 const ir_variable
*front_color
,
332 const ir_variable
*back_color
,
333 gl_shader_stage consumer_stage
,
334 gl_shader_stage producer_stage
)
336 if (front_color
!= NULL
&& front_color
->data
.assigned
)
337 cross_validate_types_and_qualifiers(prog
, input
, front_color
,
338 consumer_stage
, producer_stage
);
340 if (back_color
!= NULL
&& back_color
->data
.assigned
)
341 cross_validate_types_and_qualifiers(prog
, input
, back_color
,
342 consumer_stage
, producer_stage
);
346 * Validate that outputs from one stage match inputs of another
349 cross_validate_outputs_to_inputs(struct gl_shader_program
*prog
,
350 gl_shader
*producer
, gl_shader
*consumer
)
352 glsl_symbol_table parameters
;
353 ir_variable
*explicit_locations
[MAX_VARYING
] = { NULL
, };
355 /* Find all shader outputs in the "producer" stage.
357 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
358 ir_variable
*const var
= node
->as_variable();
360 if ((var
== NULL
) || (var
->data
.mode
!= ir_var_shader_out
))
363 if (!var
->data
.explicit_location
364 || var
->data
.location
< VARYING_SLOT_VAR0
)
365 parameters
.add_variable(var
);
367 /* User-defined varyings with explicit locations are handled
368 * differently because they do not need to have matching names.
370 const glsl_type
*type
= get_varying_type(var
, producer
->Stage
);
371 unsigned num_elements
= type
->count_attribute_slots(false);
372 unsigned idx
= var
->data
.location
- VARYING_SLOT_VAR0
;
373 unsigned slot_limit
= idx
+ num_elements
;
375 while (idx
< slot_limit
) {
376 if (explicit_locations
[idx
] != NULL
) {
378 "%s shader has multiple outputs explicitly "
379 "assigned to location %d\n",
380 _mesa_shader_stage_to_string(producer
->Stage
),
385 explicit_locations
[idx
] = var
;
392 /* Find all shader inputs in the "consumer" stage. Any variables that have
393 * matching outputs already in the symbol table must have the same type and
396 * Exception: if the consumer is the geometry shader, then the inputs
397 * should be arrays and the type of the array element should match the type
398 * of the corresponding producer output.
400 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
401 ir_variable
*const input
= node
->as_variable();
403 if ((input
== NULL
) || (input
->data
.mode
!= ir_var_shader_in
))
406 if (strcmp(input
->name
, "gl_Color") == 0 && input
->data
.used
) {
407 const ir_variable
*const front_color
=
408 parameters
.get_variable("gl_FrontColor");
410 const ir_variable
*const back_color
=
411 parameters
.get_variable("gl_BackColor");
413 cross_validate_front_and_back_color(prog
, input
,
414 front_color
, back_color
,
415 consumer
->Stage
, producer
->Stage
);
416 } else if (strcmp(input
->name
, "gl_SecondaryColor") == 0 && input
->data
.used
) {
417 const ir_variable
*const front_color
=
418 parameters
.get_variable("gl_FrontSecondaryColor");
420 const ir_variable
*const back_color
=
421 parameters
.get_variable("gl_BackSecondaryColor");
423 cross_validate_front_and_back_color(prog
, input
,
424 front_color
, back_color
,
425 consumer
->Stage
, producer
->Stage
);
427 /* The rules for connecting inputs and outputs change in the presence
428 * of explicit locations. In this case, we no longer care about the
429 * names of the variables. Instead, we care only about the
430 * explicitly assigned location.
432 ir_variable
*output
= NULL
;
433 if (input
->data
.explicit_location
434 && input
->data
.location
>= VARYING_SLOT_VAR0
) {
436 const glsl_type
*type
= get_varying_type(input
, consumer
->Stage
);
437 unsigned num_elements
= type
->count_attribute_slots(false);
438 unsigned idx
= input
->data
.location
- VARYING_SLOT_VAR0
;
439 unsigned slot_limit
= idx
+ num_elements
;
441 while (idx
< slot_limit
) {
442 output
= explicit_locations
[idx
];
444 if (output
== NULL
||
445 input
->data
.location
!= output
->data
.location
) {
447 "%s shader input `%s' with explicit location "
448 "has no matching output\n",
449 _mesa_shader_stage_to_string(consumer
->Stage
),
456 output
= parameters
.get_variable(input
->name
);
459 if (output
!= NULL
) {
460 /* Interface blocks have their own validation elsewhere so don't
461 * try validating them here.
463 if (!(input
->get_interface_type() &&
464 output
->get_interface_type()))
465 cross_validate_types_and_qualifiers(prog
, input
, output
,
469 /* Check for input vars with unmatched output vars in prev stage
470 * taking into account that interface blocks could have a matching
471 * output but with different name, so we ignore them.
473 assert(!input
->data
.assigned
);
474 if (input
->data
.used
&& !input
->get_interface_type() &&
475 !input
->data
.explicit_location
&& !prog
->SeparateShader
)
477 "%s shader input `%s' "
478 "has no matching output in the previous stage\n",
479 _mesa_shader_stage_to_string(consumer
->Stage
),
487 * Demote shader inputs and outputs that are not used in other stages, and
488 * remove them via dead code elimination.
491 remove_unused_shader_inputs_and_outputs(bool is_separate_shader_object
,
493 enum ir_variable_mode mode
)
495 if (is_separate_shader_object
)
498 foreach_in_list(ir_instruction
, node
, sh
->ir
) {
499 ir_variable
*const var
= node
->as_variable();
501 if ((var
== NULL
) || (var
->data
.mode
!= int(mode
)))
504 /* A shader 'in' or 'out' variable is only really an input or output if
505 * its value is used by other shader stages. This will cause the
506 * variable to have a location assigned.
508 if (var
->data
.is_unmatched_generic_inout
&& !var
->data
.is_xfb_only
) {
509 assert(var
->data
.mode
!= ir_var_temporary
);
510 var
->data
.mode
= ir_var_auto
;
514 /* Eliminate code that is now dead due to unused inputs/outputs being
517 while (do_dead_code(sh
->ir
, false))
523 * Initialize this object based on a string that was passed to
524 * glTransformFeedbackVaryings.
526 * If the input is mal-formed, this call still succeeds, but it sets
527 * this->var_name to a mal-formed input, so tfeedback_decl::find_output_var()
528 * will fail to find any matching variable.
531 tfeedback_decl::init(struct gl_context
*ctx
, const void *mem_ctx
,
534 /* We don't have to be pedantic about what is a valid GLSL variable name,
535 * because any variable with an invalid name can't exist in the IR anyway.
539 this->orig_name
= input
;
540 this->lowered_builtin_array_variable
= none
;
541 this->skip_components
= 0;
542 this->next_buffer_separator
= false;
543 this->matched_candidate
= NULL
;
548 if (ctx
->Extensions
.ARB_transform_feedback3
) {
549 /* Parse gl_NextBuffer. */
550 if (strcmp(input
, "gl_NextBuffer") == 0) {
551 this->next_buffer_separator
= true;
555 /* Parse gl_SkipComponents. */
556 if (strcmp(input
, "gl_SkipComponents1") == 0)
557 this->skip_components
= 1;
558 else if (strcmp(input
, "gl_SkipComponents2") == 0)
559 this->skip_components
= 2;
560 else if (strcmp(input
, "gl_SkipComponents3") == 0)
561 this->skip_components
= 3;
562 else if (strcmp(input
, "gl_SkipComponents4") == 0)
563 this->skip_components
= 4;
565 if (this->skip_components
)
569 /* Parse a declaration. */
570 const char *base_name_end
;
571 long subscript
= parse_program_resource_name(input
, &base_name_end
);
572 this->var_name
= ralloc_strndup(mem_ctx
, input
, base_name_end
- input
);
573 if (this->var_name
== NULL
) {
574 _mesa_error_no_memory(__func__
);
578 if (subscript
>= 0) {
579 this->array_subscript
= subscript
;
580 this->is_subscripted
= true;
582 this->is_subscripted
= false;
585 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this
586 * class must behave specially to account for the fact that gl_ClipDistance
587 * is converted from a float[8] to a vec4[2].
589 if (ctx
->Const
.ShaderCompilerOptions
[MESA_SHADER_VERTEX
].LowerClipDistance
&&
590 strcmp(this->var_name
, "gl_ClipDistance") == 0) {
591 this->lowered_builtin_array_variable
= clip_distance
;
594 if (ctx
->Const
.LowerTessLevel
&&
595 (strcmp(this->var_name
, "gl_TessLevelOuter") == 0))
596 this->lowered_builtin_array_variable
= tess_level_outer
;
597 if (ctx
->Const
.LowerTessLevel
&&
598 (strcmp(this->var_name
, "gl_TessLevelInner") == 0))
599 this->lowered_builtin_array_variable
= tess_level_inner
;
604 * Determine whether two tfeedback_decl objects refer to the same variable and
605 * array index (if applicable).
608 tfeedback_decl::is_same(const tfeedback_decl
&x
, const tfeedback_decl
&y
)
610 assert(x
.is_varying() && y
.is_varying());
612 if (strcmp(x
.var_name
, y
.var_name
) != 0)
614 if (x
.is_subscripted
!= y
.is_subscripted
)
616 if (x
.is_subscripted
&& x
.array_subscript
!= y
.array_subscript
)
623 * Assign a location and stream ID for this tfeedback_decl object based on the
624 * transform feedback candidate found by find_candidate.
626 * If an error occurs, the error is reported through linker_error() and false
630 tfeedback_decl::assign_location(struct gl_context
*ctx
,
631 struct gl_shader_program
*prog
)
633 assert(this->is_varying());
635 unsigned fine_location
636 = this->matched_candidate
->toplevel_var
->data
.location
* 4
637 + this->matched_candidate
->toplevel_var
->data
.location_frac
638 + this->matched_candidate
->offset
;
639 const unsigned dmul
=
640 this->matched_candidate
->type
->without_array()->is_double() ? 2 : 1;
642 if (this->matched_candidate
->type
->is_array()) {
644 const unsigned matrix_cols
=
645 this->matched_candidate
->type
->fields
.array
->matrix_columns
;
646 const unsigned vector_elements
=
647 this->matched_candidate
->type
->fields
.array
->vector_elements
;
648 unsigned actual_array_size
;
649 switch (this->lowered_builtin_array_variable
) {
651 actual_array_size
= prog
->LastClipDistanceArraySize
;
653 case tess_level_outer
:
654 actual_array_size
= 4;
656 case tess_level_inner
:
657 actual_array_size
= 2;
661 actual_array_size
= this->matched_candidate
->type
->array_size();
665 if (this->is_subscripted
) {
666 /* Check array bounds. */
667 if (this->array_subscript
>= actual_array_size
) {
668 linker_error(prog
, "Transform feedback varying %s has index "
669 "%i, but the array size is %u.",
670 this->orig_name
, this->array_subscript
,
674 unsigned array_elem_size
= this->lowered_builtin_array_variable
?
675 1 : vector_elements
* matrix_cols
* dmul
;
676 fine_location
+= array_elem_size
* this->array_subscript
;
679 this->size
= actual_array_size
;
681 this->vector_elements
= vector_elements
;
682 this->matrix_columns
= matrix_cols
;
683 if (this->lowered_builtin_array_variable
)
684 this->type
= GL_FLOAT
;
686 this->type
= this->matched_candidate
->type
->fields
.array
->gl_type
;
688 /* Regular variable (scalar, vector, or matrix) */
689 if (this->is_subscripted
) {
690 linker_error(prog
, "Transform feedback varying %s requested, "
691 "but %s is not an array.",
692 this->orig_name
, this->var_name
);
696 this->vector_elements
= this->matched_candidate
->type
->vector_elements
;
697 this->matrix_columns
= this->matched_candidate
->type
->matrix_columns
;
698 this->type
= this->matched_candidate
->type
->gl_type
;
700 this->location
= fine_location
/ 4;
701 this->location_frac
= fine_location
% 4;
703 /* From GL_EXT_transform_feedback:
704 * A program will fail to link if:
706 * * the total number of components to capture in any varying
707 * variable in <varyings> is greater than the constant
708 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
709 * buffer mode is SEPARATE_ATTRIBS_EXT;
711 if (prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
&&
712 this->num_components() >
713 ctx
->Const
.MaxTransformFeedbackSeparateComponents
) {
714 linker_error(prog
, "Transform feedback varying %s exceeds "
715 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
720 /* Only transform feedback varyings can be assigned to non-zero streams,
721 * so assign the stream id here.
723 this->stream_id
= this->matched_candidate
->toplevel_var
->data
.stream
;
725 unsigned array_offset
= this->array_subscript
* 4 * dmul
;
726 unsigned struct_offset
= this->matched_candidate
->offset
* 4 * dmul
;
727 this->buffer
= this->matched_candidate
->toplevel_var
->data
.xfb_buffer
;
728 this->offset
= this->matched_candidate
->toplevel_var
->data
.offset
+
729 array_offset
+ struct_offset
;
736 tfeedback_decl::get_num_outputs() const
738 if (!this->is_varying()) {
741 return (this->num_components() + this->location_frac
+ 3)/4;
746 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
748 * If an error occurs, the error is reported through linker_error() and false
752 tfeedback_decl::store(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
753 struct gl_transform_feedback_info
*info
,
754 unsigned buffer
, unsigned buffer_index
,
755 const unsigned max_outputs
, bool *explicit_stride
,
756 bool has_xfb_qualifiers
) const
758 assert(!this->next_buffer_separator
);
760 /* Handle gl_SkipComponents. */
761 if (this->skip_components
) {
762 info
->Buffers
[buffer
].Stride
+= this->skip_components
;
766 unsigned xfb_offset
= 0;
767 if (has_xfb_qualifiers
) {
768 xfb_offset
= this->offset
/ 4;
770 xfb_offset
= info
->Buffers
[buffer
].Stride
;
772 info
->Varyings
[info
->NumVarying
].Offset
= xfb_offset
* 4;
774 unsigned location
= this->location
;
775 unsigned location_frac
= this->location_frac
;
776 unsigned num_components
= this->num_components();
777 while (num_components
> 0) {
778 unsigned output_size
= MIN2(num_components
, 4 - location_frac
);
779 assert((info
->NumOutputs
== 0 && max_outputs
== 0) ||
780 info
->NumOutputs
< max_outputs
);
782 /* From the ARB_enhanced_layouts spec:
784 * "If such a block member or variable is not written during a shader
785 * invocation, the buffer contents at the assigned offset will be
786 * undefined. Even if there are no static writes to a variable or
787 * member that is assigned a transform feedback offset, the space is
788 * still allocated in the buffer and still affects the stride."
790 if (this->is_varying_written()) {
791 info
->Outputs
[info
->NumOutputs
].ComponentOffset
= location_frac
;
792 info
->Outputs
[info
->NumOutputs
].OutputRegister
= location
;
793 info
->Outputs
[info
->NumOutputs
].NumComponents
= output_size
;
794 info
->Outputs
[info
->NumOutputs
].StreamId
= stream_id
;
795 info
->Outputs
[info
->NumOutputs
].OutputBuffer
= buffer
;
796 info
->Outputs
[info
->NumOutputs
].DstOffset
= xfb_offset
;
799 info
->Buffers
[buffer
].Stream
= this->stream_id
;
800 xfb_offset
+= output_size
;
802 num_components
-= output_size
;
807 if (explicit_stride
&& explicit_stride
[buffer
]) {
808 if (this->is_double() && info
->Buffers
[buffer
].Stride
% 2) {
809 linker_error(prog
, "invalid qualifier xfb_stride=%d must be a "
810 "multiple of 8 as its applied to a type that is or "
811 "contains a double.",
812 info
->Buffers
[buffer
].Stride
* 4);
816 if ((this->offset
/ 4) / info
->Buffers
[buffer
].Stride
!=
817 (xfb_offset
- 1) / info
->Buffers
[buffer
].Stride
) {
818 linker_error(prog
, "xfb_offset (%d) overflows xfb_stride (%d) for "
819 "buffer (%d)", xfb_offset
* 4,
820 info
->Buffers
[buffer
].Stride
* 4, buffer
);
824 info
->Buffers
[buffer
].Stride
= xfb_offset
;
827 /* From GL_EXT_transform_feedback:
828 * A program will fail to link if:
830 * * the total number of components to capture is greater than
831 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
832 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
834 * From GL_ARB_enhanced_layouts:
836 * "The resulting stride (implicit or explicit) must be less than or
837 * equal to the implementation-dependent constant
838 * gl_MaxTransformFeedbackInterleavedComponents."
840 if ((prog
->TransformFeedback
.BufferMode
== GL_INTERLEAVED_ATTRIBS
||
841 has_xfb_qualifiers
) &&
842 info
->Buffers
[buffer
].Stride
>
843 ctx
->Const
.MaxTransformFeedbackInterleavedComponents
) {
844 linker_error(prog
, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
845 "limit has been exceeded.");
849 info
->Varyings
[info
->NumVarying
].Name
= ralloc_strdup(prog
,
851 info
->Varyings
[info
->NumVarying
].Type
= this->type
;
852 info
->Varyings
[info
->NumVarying
].Size
= this->size
;
853 info
->Varyings
[info
->NumVarying
].BufferIndex
= buffer_index
;
855 info
->Buffers
[buffer
].NumVaryings
++;
861 const tfeedback_candidate
*
862 tfeedback_decl::find_candidate(gl_shader_program
*prog
,
863 hash_table
*tfeedback_candidates
)
865 const char *name
= this->var_name
;
866 switch (this->lowered_builtin_array_variable
) {
868 name
= this->var_name
;
871 name
= "gl_ClipDistanceMESA";
873 case tess_level_outer
:
874 name
= "gl_TessLevelOuterMESA";
876 case tess_level_inner
:
877 name
= "gl_TessLevelInnerMESA";
880 this->matched_candidate
= (const tfeedback_candidate
*)
881 hash_table_find(tfeedback_candidates
, name
);
882 if (!this->matched_candidate
) {
883 /* From GL_EXT_transform_feedback:
884 * A program will fail to link if:
886 * * any variable name specified in the <varyings> array is not
887 * declared as an output in the geometry shader (if present) or
888 * the vertex shader (if no geometry shader is present);
890 linker_error(prog
, "Transform feedback varying %s undeclared.",
893 return this->matched_candidate
;
898 * Parse all the transform feedback declarations that were passed to
899 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
901 * If an error occurs, the error is reported through linker_error() and false
905 parse_tfeedback_decls(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
906 const void *mem_ctx
, unsigned num_names
,
907 char **varying_names
, tfeedback_decl
*decls
)
909 for (unsigned i
= 0; i
< num_names
; ++i
) {
910 decls
[i
].init(ctx
, mem_ctx
, varying_names
[i
]);
912 if (!decls
[i
].is_varying())
915 /* From GL_EXT_transform_feedback:
916 * A program will fail to link if:
918 * * any two entries in the <varyings> array specify the same varying
921 * We interpret this to mean "any two entries in the <varyings> array
922 * specify the same varying variable and array index", since transform
923 * feedback of arrays would be useless otherwise.
925 for (unsigned j
= 0; j
< i
; ++j
) {
926 if (!decls
[j
].is_varying())
929 if (tfeedback_decl::is_same(decls
[i
], decls
[j
])) {
930 linker_error(prog
, "Transform feedback varying %s specified "
931 "more than once.", varying_names
[i
]);
941 cmp_xfb_offset(const void * x_generic
, const void * y_generic
)
943 tfeedback_decl
*x
= (tfeedback_decl
*) x_generic
;
944 tfeedback_decl
*y
= (tfeedback_decl
*) y_generic
;
946 if (x
->get_buffer() != y
->get_buffer())
947 return x
->get_buffer() - y
->get_buffer();
948 return x
->get_offset() - y
->get_offset();
952 * Store transform feedback location assignments into
953 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
955 * If an error occurs, the error is reported through linker_error() and false
959 store_tfeedback_info(struct gl_context
*ctx
, struct gl_shader_program
*prog
,
960 unsigned num_tfeedback_decls
,
961 tfeedback_decl
*tfeedback_decls
, bool has_xfb_qualifiers
)
963 /* Make sure MaxTransformFeedbackBuffers is less than 32 so the bitmask for
964 * tracking the number of buffers doesn't overflow.
966 assert(ctx
->Const
.MaxTransformFeedbackBuffers
< 32);
968 bool separate_attribs_mode
=
969 prog
->TransformFeedback
.BufferMode
== GL_SEPARATE_ATTRIBS
;
971 ralloc_free(prog
->LinkedTransformFeedback
.Varyings
);
972 ralloc_free(prog
->LinkedTransformFeedback
.Outputs
);
974 memset(&prog
->LinkedTransformFeedback
, 0,
975 sizeof(prog
->LinkedTransformFeedback
));
977 /* The xfb_offset qualifier does not have to be used in increasing order
978 * however some drivers expect to receive the list of transform feedback
979 * declarations in order so sort it now for convenience.
981 if (has_xfb_qualifiers
)
982 qsort(tfeedback_decls
, num_tfeedback_decls
, sizeof(*tfeedback_decls
),
985 prog
->LinkedTransformFeedback
.Varyings
=
987 struct gl_transform_feedback_varying_info
,
988 num_tfeedback_decls
);
990 unsigned num_outputs
= 0;
991 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
992 if (tfeedback_decls
[i
].is_varying_written())
993 num_outputs
+= tfeedback_decls
[i
].get_num_outputs();
996 prog
->LinkedTransformFeedback
.Outputs
=
998 struct gl_transform_feedback_output
,
1001 unsigned num_buffers
= 0;
1002 unsigned buffers
= 0;
1004 if (!has_xfb_qualifiers
&& separate_attribs_mode
) {
1005 /* GL_SEPARATE_ATTRIBS */
1006 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
1007 if (!tfeedback_decls
[i
].store(ctx
, prog
, &prog
->LinkedTransformFeedback
,
1008 num_buffers
, num_buffers
, num_outputs
,
1009 NULL
, has_xfb_qualifiers
))
1012 buffers
|= 1 << num_buffers
;
1017 /* GL_INVERLEAVED_ATTRIBS */
1018 int buffer_stream_id
= -1;
1020 num_tfeedback_decls
? tfeedback_decls
[0].get_buffer() : 0;
1021 bool explicit_stride
[MAX_FEEDBACK_BUFFERS
] = { false };
1023 /* Apply any xfb_stride global qualifiers */
1024 if (has_xfb_qualifiers
) {
1025 for (unsigned j
= 0; j
< MAX_FEEDBACK_BUFFERS
; j
++) {
1026 if (prog
->TransformFeedback
.BufferStride
[j
]) {
1028 explicit_stride
[j
] = true;
1029 prog
->LinkedTransformFeedback
.Buffers
[j
].Stride
=
1030 prog
->TransformFeedback
.BufferStride
[j
] / 4;
1035 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
1036 if (has_xfb_qualifiers
&&
1037 buffer
!= tfeedback_decls
[i
].get_buffer()) {
1038 /* we have moved to the next buffer so reset stream id */
1039 buffer_stream_id
= -1;
1043 if (tfeedback_decls
[i
].is_next_buffer_separator()) {
1045 buffer_stream_id
= -1;
1047 } else if (buffer_stream_id
== -1) {
1048 /* First varying writing to this buffer: remember its stream */
1049 buffer_stream_id
= (int) tfeedback_decls
[i
].get_stream_id();
1050 } else if (buffer_stream_id
!=
1051 (int) tfeedback_decls
[i
].get_stream_id()) {
1052 /* Varying writes to the same buffer from a different stream */
1054 "Transform feedback can't capture varyings belonging "
1055 "to different vertex streams in a single buffer. "
1056 "Varying %s writes to buffer from stream %u, other "
1057 "varyings in the same buffer write from stream %u.",
1058 tfeedback_decls
[i
].name(),
1059 tfeedback_decls
[i
].get_stream_id(),
1064 if (has_xfb_qualifiers
) {
1065 buffer
= tfeedback_decls
[i
].get_buffer();
1067 buffer
= num_buffers
;
1069 buffers
|= 1 << buffer
;
1071 if (!tfeedback_decls
[i
].store(ctx
, prog
,
1072 &prog
->LinkedTransformFeedback
,
1073 buffer
, num_buffers
, num_outputs
,
1074 explicit_stride
, has_xfb_qualifiers
))
1079 assert(prog
->LinkedTransformFeedback
.NumOutputs
== num_outputs
);
1081 prog
->LinkedTransformFeedback
.ActiveBuffers
= buffers
;
1088 * Data structure recording the relationship between outputs of one shader
1089 * stage (the "producer") and inputs of another (the "consumer").
1091 class varying_matches
1094 varying_matches(bool disable_varying_packing
, bool xfb_enabled
,
1095 gl_shader_stage producer_stage
,
1096 gl_shader_stage consumer_stage
);
1098 void record(ir_variable
*producer_var
, ir_variable
*consumer_var
);
1099 unsigned assign_locations(struct gl_shader_program
*prog
,
1100 uint64_t reserved_slots
, bool separate_shader
);
1101 void store_locations() const;
1104 bool is_varying_packing_safe(const glsl_type
*type
,
1105 const ir_variable
*var
);
1108 * If true, this driver disables varying packing, so all varyings need to
1109 * be aligned on slot boundaries, and take up a number of slots equal to
1110 * their number of matrix columns times their array size.
1112 * Packing may also be disabled because our current packing method is not
1113 * safe in SSO or versions of OpenGL where interpolation qualifiers are not
1114 * guaranteed to match across stages.
1116 const bool disable_varying_packing
;
1119 * If true, this driver has transform feedback enabled. The transform
1120 * feedback code requires at least some packing be done even when varying
1121 * packing is disabled, fortunately where transform feedback requires
1122 * packing it's safe to override the disabled setting. See
1123 * is_varying_packing_safe().
1125 const bool xfb_enabled
;
1128 * Enum representing the order in which varyings are packed within a
1131 * Currently we pack vec4's first, then vec2's, then scalar values, then
1132 * vec3's. This order ensures that the only vectors that are at risk of
1133 * having to be "double parked" (split between two adjacent varying slots)
1136 enum packing_order_enum
{
1139 PACKING_ORDER_SCALAR
,
1143 static unsigned compute_packing_class(const ir_variable
*var
);
1144 static packing_order_enum
compute_packing_order(const ir_variable
*var
);
1145 static int match_comparator(const void *x_generic
, const void *y_generic
);
1146 static int xfb_comparator(const void *x_generic
, const void *y_generic
);
1149 * Structure recording the relationship between a single producer output
1150 * and a single consumer input.
1154 * Packing class for this varying, computed by compute_packing_class().
1156 unsigned packing_class
;
1159 * Packing order for this varying, computed by compute_packing_order().
1161 packing_order_enum packing_order
;
1162 unsigned num_components
;
1165 * The output variable in the producer stage.
1167 ir_variable
*producer_var
;
1170 * The input variable in the consumer stage.
1172 ir_variable
*consumer_var
;
1175 * The location which has been assigned for this varying. This is
1176 * expressed in multiples of a float, with the first generic varying
1177 * (i.e. the one referred to by VARYING_SLOT_VAR0) represented by the
1180 unsigned generic_location
;
1184 * The number of elements in the \c matches array that are currently in
1187 unsigned num_matches
;
1190 * The number of elements that were set aside for the \c matches array when
1193 unsigned matches_capacity
;
1195 gl_shader_stage producer_stage
;
1196 gl_shader_stage consumer_stage
;
1199 } /* anonymous namespace */
1201 varying_matches::varying_matches(bool disable_varying_packing
,
1203 gl_shader_stage producer_stage
,
1204 gl_shader_stage consumer_stage
)
1205 : disable_varying_packing(disable_varying_packing
),
1206 xfb_enabled(xfb_enabled
),
1207 producer_stage(producer_stage
),
1208 consumer_stage(consumer_stage
)
1210 /* Note: this initial capacity is rather arbitrarily chosen to be large
1211 * enough for many cases without wasting an unreasonable amount of space.
1212 * varying_matches::record() will resize the array if there are more than
1213 * this number of varyings.
1215 this->matches_capacity
= 8;
1216 this->matches
= (match
*)
1217 malloc(sizeof(*this->matches
) * this->matches_capacity
);
1218 this->num_matches
= 0;
1222 varying_matches::~varying_matches()
1224 free(this->matches
);
1229 * Packing is always safe on individual arrays, structure and matices. It is
1230 * also safe if the varying is only used for transform feedback.
1233 varying_matches::is_varying_packing_safe(const glsl_type
*type
,
1234 const ir_variable
*var
)
1236 if (consumer_stage
== MESA_SHADER_TESS_EVAL
||
1237 consumer_stage
== MESA_SHADER_TESS_CTRL
||
1238 producer_stage
== MESA_SHADER_TESS_CTRL
)
1241 return xfb_enabled
&& (type
->is_array() || type
->is_record() ||
1242 type
->is_matrix() || var
->data
.is_xfb_only
);
1247 * Record the given producer/consumer variable pair in the list of variables
1248 * that should later be assigned locations.
1250 * It is permissible for \c consumer_var to be NULL (this happens if a
1251 * variable is output by the producer and consumed by transform feedback, but
1252 * not consumed by the consumer).
1254 * If \c producer_var has already been paired up with a consumer_var, or
1255 * producer_var is part of fixed pipeline functionality (and hence already has
1256 * a location assigned), this function has no effect.
1258 * Note: as a side effect this function may change the interpolation type of
1259 * \c producer_var, but only when the change couldn't possibly affect
1263 varying_matches::record(ir_variable
*producer_var
, ir_variable
*consumer_var
)
1265 assert(producer_var
!= NULL
|| consumer_var
!= NULL
);
1267 if ((producer_var
&& (!producer_var
->data
.is_unmatched_generic_inout
||
1268 producer_var
->data
.explicit_location
)) ||
1269 (consumer_var
&& (!consumer_var
->data
.is_unmatched_generic_inout
||
1270 consumer_var
->data
.explicit_location
))) {
1271 /* Either a location already exists for this variable (since it is part
1272 * of fixed functionality), or it has already been recorded as part of a
1278 bool needs_flat_qualifier
= consumer_var
== NULL
&&
1279 (producer_var
->type
->contains_integer() ||
1280 producer_var
->type
->contains_double());
1282 if (needs_flat_qualifier
||
1283 (consumer_stage
!= -1 && consumer_stage
!= MESA_SHADER_FRAGMENT
)) {
1284 /* Since this varying is not being consumed by the fragment shader, its
1285 * interpolation type varying cannot possibly affect rendering.
1286 * Also, this variable is non-flat and is (or contains) an integer
1288 * If the consumer stage is unknown, don't modify the interpolation
1289 * type as it could affect rendering later with separate shaders.
1291 * lower_packed_varyings requires all integer varyings to flat,
1292 * regardless of where they appear. We can trivially satisfy that
1293 * requirement by changing the interpolation type to flat here.
1296 producer_var
->data
.centroid
= false;
1297 producer_var
->data
.sample
= false;
1298 producer_var
->data
.interpolation
= INTERP_QUALIFIER_FLAT
;
1302 consumer_var
->data
.centroid
= false;
1303 consumer_var
->data
.sample
= false;
1304 consumer_var
->data
.interpolation
= INTERP_QUALIFIER_FLAT
;
1308 if (this->num_matches
== this->matches_capacity
) {
1309 this->matches_capacity
*= 2;
1310 this->matches
= (match
*)
1311 realloc(this->matches
,
1312 sizeof(*this->matches
) * this->matches_capacity
);
1315 const ir_variable
*const var
= (producer_var
!= NULL
)
1316 ? producer_var
: consumer_var
;
1317 const gl_shader_stage stage
= (producer_var
!= NULL
)
1318 ? producer_stage
: consumer_stage
;
1319 const glsl_type
*type
= get_varying_type(var
, stage
);
1321 this->matches
[this->num_matches
].packing_class
1322 = this->compute_packing_class(var
);
1323 this->matches
[this->num_matches
].packing_order
1324 = this->compute_packing_order(var
);
1325 if (this->disable_varying_packing
&& !is_varying_packing_safe(type
, var
)) {
1326 unsigned slots
= type
->count_attribute_slots(false);
1327 this->matches
[this->num_matches
].num_components
= slots
* 4;
1329 this->matches
[this->num_matches
].num_components
1330 = type
->component_slots();
1332 this->matches
[this->num_matches
].producer_var
= producer_var
;
1333 this->matches
[this->num_matches
].consumer_var
= consumer_var
;
1334 this->num_matches
++;
1336 producer_var
->data
.is_unmatched_generic_inout
= 0;
1338 consumer_var
->data
.is_unmatched_generic_inout
= 0;
1343 * Choose locations for all of the variable matches that were previously
1344 * passed to varying_matches::record().
1347 varying_matches::assign_locations(struct gl_shader_program
*prog
,
1348 uint64_t reserved_slots
,
1349 bool separate_shader
)
1351 /* If packing has been disabled then we cannot safely sort the varyings by
1352 * class as it may mean we are using a version of OpenGL where
1353 * interpolation qualifiers are not guaranteed to be matching across
1354 * shaders, sorting in this case could result in mismatching shader
1356 * When packing is disabled the sort orders varyings used by transform
1357 * feedback first, but also depends on *undefined behaviour* of qsort to
1358 * reverse the order of the varyings. See: xfb_comparator().
1360 if (!this->disable_varying_packing
) {
1361 /* Sort varying matches into an order that makes them easy to pack. */
1362 qsort(this->matches
, this->num_matches
, sizeof(*this->matches
),
1363 &varying_matches::match_comparator
);
1365 /* Only sort varyings that are only used by transform feedback. */
1366 qsort(this->matches
, this->num_matches
, sizeof(*this->matches
),
1367 &varying_matches::xfb_comparator
);
1370 unsigned generic_location
= 0;
1371 unsigned generic_patch_location
= MAX_VARYING
*4;
1372 bool previous_var_xfb_only
= false;
1374 for (unsigned i
= 0; i
< this->num_matches
; i
++) {
1375 unsigned *location
= &generic_location
;
1377 const ir_variable
*var
;
1378 const glsl_type
*type
;
1379 bool is_vertex_input
= false;
1380 if (matches
[i
].consumer_var
) {
1381 var
= matches
[i
].consumer_var
;
1382 type
= get_varying_type(var
, consumer_stage
);
1383 if (consumer_stage
== MESA_SHADER_VERTEX
)
1384 is_vertex_input
= true;
1386 var
= matches
[i
].producer_var
;
1387 type
= get_varying_type(var
, producer_stage
);
1390 if (var
->data
.patch
)
1391 location
= &generic_patch_location
;
1393 /* Advance to the next slot if this varying has a different packing
1394 * class than the previous one, and we're not already on a slot
1397 * Also advance to the next slot if packing is disabled. This makes sure
1398 * we don't assign varyings the same locations which is possible
1399 * because we still pack individual arrays, records and matrices even
1400 * when packing is disabled. Note we don't advance to the next slot if
1401 * we can pack varyings together that are only used for transform
1404 if ((this->disable_varying_packing
&&
1405 !(previous_var_xfb_only
&& var
->data
.is_xfb_only
)) ||
1406 (i
> 0 && this->matches
[i
- 1].packing_class
1407 != this->matches
[i
].packing_class
)) {
1408 *location
= ALIGN(*location
, 4);
1411 previous_var_xfb_only
= var
->data
.is_xfb_only
;
1413 unsigned num_elements
= type
->count_attribute_slots(is_vertex_input
);
1415 if (this->disable_varying_packing
&&
1416 !is_varying_packing_safe(type
, var
))
1419 slot_end
= type
->without_array()->vector_elements
;
1420 slot_end
+= *location
- 1;
1422 /* FIXME: We could be smarter in the below code and loop back over
1423 * trying to fill any locations that we skipped because we couldn't pack
1424 * the varying between an explicit location. For now just let the user
1425 * hit the linking error if we run out of room and suggest they use
1426 * explicit locations.
1428 for (unsigned j
= 0; j
< num_elements
; j
++) {
1429 while ((slot_end
< MAX_VARYING
* 4u) &&
1430 ((reserved_slots
& (UINT64_C(1) << *location
/ 4u) ||
1431 (reserved_slots
& (UINT64_C(1) << slot_end
/ 4u))))) {
1433 *location
= ALIGN(*location
+ 1, 4);
1434 slot_end
= *location
;
1436 /* reset the counter and try again */
1440 /* Increase the slot to make sure there is enough room for next
1443 if (this->disable_varying_packing
&&
1444 !is_varying_packing_safe(type
, var
))
1447 slot_end
+= type
->without_array()->vector_elements
;
1450 if (!var
->data
.patch
&& *location
>= MAX_VARYING
* 4u) {
1451 linker_error(prog
, "insufficient contiguous locations available for "
1452 "%s it is possible an array or struct could not be "
1453 "packed between varyings with explicit locations. Try "
1454 "using an explicit location for arrays and structs.",
1458 this->matches
[i
].generic_location
= *location
;
1460 *location
+= this->matches
[i
].num_components
;
1463 return (generic_location
+ 3) / 4;
1468 * Update the producer and consumer shaders to reflect the locations
1469 * assignments that were made by varying_matches::assign_locations().
1472 varying_matches::store_locations() const
1474 for (unsigned i
= 0; i
< this->num_matches
; i
++) {
1475 ir_variable
*producer_var
= this->matches
[i
].producer_var
;
1476 ir_variable
*consumer_var
= this->matches
[i
].consumer_var
;
1477 unsigned generic_location
= this->matches
[i
].generic_location
;
1478 unsigned slot
= generic_location
/ 4;
1479 unsigned offset
= generic_location
% 4;
1482 producer_var
->data
.location
= VARYING_SLOT_VAR0
+ slot
;
1483 producer_var
->data
.location_frac
= offset
;
1487 assert(consumer_var
->data
.location
== -1);
1488 consumer_var
->data
.location
= VARYING_SLOT_VAR0
+ slot
;
1489 consumer_var
->data
.location_frac
= offset
;
1496 * Compute the "packing class" of the given varying. This is an unsigned
1497 * integer with the property that two variables in the same packing class can
1498 * be safely backed into the same vec4.
1501 varying_matches::compute_packing_class(const ir_variable
*var
)
1503 /* Without help from the back-end, there is no way to pack together
1504 * variables with different interpolation types, because
1505 * lower_packed_varyings must choose exactly one interpolation type for
1506 * each packed varying it creates.
1508 * However, we can safely pack together floats, ints, and uints, because:
1510 * - varyings of base type "int" and "uint" must use the "flat"
1511 * interpolation type, which can only occur in GLSL 1.30 and above.
1513 * - On platforms that support GLSL 1.30 and above, lower_packed_varyings
1514 * can store flat floats as ints without losing any information (using
1515 * the ir_unop_bitcast_* opcodes).
1517 * Therefore, the packing class depends only on the interpolation type.
1519 unsigned packing_class
= var
->data
.centroid
| (var
->data
.sample
<< 1) |
1520 (var
->data
.patch
<< 2);
1522 packing_class
+= var
->data
.interpolation
;
1523 return packing_class
;
1528 * Compute the "packing order" of the given varying. This is a sort key we
1529 * use to determine when to attempt to pack the given varying relative to
1530 * other varyings in the same packing class.
1532 varying_matches::packing_order_enum
1533 varying_matches::compute_packing_order(const ir_variable
*var
)
1535 const glsl_type
*element_type
= var
->type
;
1537 while (element_type
->base_type
== GLSL_TYPE_ARRAY
) {
1538 element_type
= element_type
->fields
.array
;
1541 switch (element_type
->component_slots() % 4) {
1542 case 1: return PACKING_ORDER_SCALAR
;
1543 case 2: return PACKING_ORDER_VEC2
;
1544 case 3: return PACKING_ORDER_VEC3
;
1545 case 0: return PACKING_ORDER_VEC4
;
1547 assert(!"Unexpected value of vector_elements");
1548 return PACKING_ORDER_VEC4
;
1554 * Comparison function passed to qsort() to sort varyings by packing_class and
1555 * then by packing_order.
1558 varying_matches::match_comparator(const void *x_generic
, const void *y_generic
)
1560 const match
*x
= (const match
*) x_generic
;
1561 const match
*y
= (const match
*) y_generic
;
1563 if (x
->packing_class
!= y
->packing_class
)
1564 return x
->packing_class
- y
->packing_class
;
1565 return x
->packing_order
- y
->packing_order
;
1570 * Comparison function passed to qsort() to sort varyings used only by
1571 * transform feedback when packing of other varyings is disabled.
1574 varying_matches::xfb_comparator(const void *x_generic
, const void *y_generic
)
1576 const match
*x
= (const match
*) x_generic
;
1578 if (x
->producer_var
!= NULL
&& x
->producer_var
->data
.is_xfb_only
)
1579 return match_comparator(x_generic
, y_generic
);
1581 /* FIXME: When the comparator returns 0 it means the elements being
1582 * compared are equivalent. However the qsort documentation says:
1584 * "The order of equivalent elements is undefined."
1586 * In practice the sort ends up reversing the order of the varyings which
1587 * means locations are also assigned in this reversed order and happens to
1588 * be what we want. This is also whats happening in
1589 * varying_matches::match_comparator().
1596 * Is the given variable a varying variable to be counted against the
1597 * limit in ctx->Const.MaxVarying?
1598 * This includes variables such as texcoords, colors and generic
1599 * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord.
1602 var_counts_against_varying_limit(gl_shader_stage stage
, const ir_variable
*var
)
1604 /* Only fragment shaders will take a varying variable as an input */
1605 if (stage
== MESA_SHADER_FRAGMENT
&&
1606 var
->data
.mode
== ir_var_shader_in
) {
1607 switch (var
->data
.location
) {
1608 case VARYING_SLOT_POS
:
1609 case VARYING_SLOT_FACE
:
1610 case VARYING_SLOT_PNTC
:
1621 * Visitor class that generates tfeedback_candidate structs describing all
1622 * possible targets of transform feedback.
1624 * tfeedback_candidate structs are stored in the hash table
1625 * tfeedback_candidates, which is passed to the constructor. This hash table
1626 * maps varying names to instances of the tfeedback_candidate struct.
1628 class tfeedback_candidate_generator
: public program_resource_visitor
1631 tfeedback_candidate_generator(void *mem_ctx
,
1632 hash_table
*tfeedback_candidates
)
1634 tfeedback_candidates(tfeedback_candidates
),
1640 void process(ir_variable
*var
)
1642 /* All named varying interface blocks should be flattened by now */
1643 assert(!var
->is_interface_instance());
1645 this->toplevel_var
= var
;
1646 this->varying_floats
= 0;
1647 program_resource_visitor::process(var
);
1651 virtual void visit_field(const glsl_type
*type
, const char *name
,
1654 assert(!type
->without_array()->is_record());
1655 assert(!type
->without_array()->is_interface());
1659 tfeedback_candidate
*candidate
1660 = rzalloc(this->mem_ctx
, tfeedback_candidate
);
1661 candidate
->toplevel_var
= this->toplevel_var
;
1662 candidate
->type
= type
;
1663 candidate
->offset
= this->varying_floats
;
1664 hash_table_insert(this->tfeedback_candidates
, candidate
,
1665 ralloc_strdup(this->mem_ctx
, name
));
1666 this->varying_floats
+= type
->component_slots();
1670 * Memory context used to allocate hash table keys and values.
1672 void * const mem_ctx
;
1675 * Hash table in which tfeedback_candidate objects should be stored.
1677 hash_table
* const tfeedback_candidates
;
1680 * Pointer to the toplevel variable that is being traversed.
1682 ir_variable
*toplevel_var
;
1685 * Total number of varying floats that have been visited so far. This is
1686 * used to determine the offset to each varying within the toplevel
1689 unsigned varying_floats
;
1696 populate_consumer_input_sets(void *mem_ctx
, exec_list
*ir
,
1697 hash_table
*consumer_inputs
,
1698 hash_table
*consumer_interface_inputs
,
1699 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
])
1701 memset(consumer_inputs_with_locations
,
1703 sizeof(consumer_inputs_with_locations
[0]) * VARYING_SLOT_TESS_MAX
);
1705 foreach_in_list(ir_instruction
, node
, ir
) {
1706 ir_variable
*const input_var
= node
->as_variable();
1708 if ((input_var
!= NULL
) && (input_var
->data
.mode
== ir_var_shader_in
)) {
1709 /* All interface blocks should have been lowered by this point */
1710 assert(!input_var
->type
->is_interface());
1712 if (input_var
->data
.explicit_location
) {
1713 /* assign_varying_locations only cares about finding the
1714 * ir_variable at the start of a contiguous location block.
1716 * - For !producer, consumer_inputs_with_locations isn't used.
1718 * - For !consumer, consumer_inputs_with_locations is empty.
1720 * For consumer && producer, if you were trying to set some
1721 * ir_variable to the middle of a location block on the other side
1722 * of producer/consumer, cross_validate_outputs_to_inputs() should
1723 * be link-erroring due to either type mismatch or location
1724 * overlaps. If the variables do match up, then they've got a
1725 * matching data.location and you only looked at
1726 * consumer_inputs_with_locations[var->data.location], not any
1727 * following entries for the array/structure.
1729 consumer_inputs_with_locations
[input_var
->data
.location
] =
1731 } else if (input_var
->get_interface_type() != NULL
) {
1732 char *const iface_field_name
=
1733 ralloc_asprintf(mem_ctx
, "%s.%s",
1734 input_var
->get_interface_type()->without_array()->name
,
1736 hash_table_insert(consumer_interface_inputs
, input_var
,
1739 hash_table_insert(consumer_inputs
, input_var
,
1740 ralloc_strdup(mem_ctx
, input_var
->name
));
1747 * Find a variable from the consumer that "matches" the specified variable
1749 * This function only finds inputs with names that match. There is no
1750 * validation (here) that the types, etc. are compatible.
1753 get_matching_input(void *mem_ctx
,
1754 const ir_variable
*output_var
,
1755 hash_table
*consumer_inputs
,
1756 hash_table
*consumer_interface_inputs
,
1757 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
])
1759 ir_variable
*input_var
;
1761 if (output_var
->data
.explicit_location
) {
1762 input_var
= consumer_inputs_with_locations
[output_var
->data
.location
];
1763 } else if (output_var
->get_interface_type() != NULL
) {
1764 char *const iface_field_name
=
1765 ralloc_asprintf(mem_ctx
, "%s.%s",
1766 output_var
->get_interface_type()->without_array()->name
,
1769 (ir_variable
*) hash_table_find(consumer_interface_inputs
,
1773 (ir_variable
*) hash_table_find(consumer_inputs
, output_var
->name
);
1776 return (input_var
== NULL
|| input_var
->data
.mode
!= ir_var_shader_in
)
1783 io_variable_cmp(const void *_a
, const void *_b
)
1785 const ir_variable
*const a
= *(const ir_variable
**) _a
;
1786 const ir_variable
*const b
= *(const ir_variable
**) _b
;
1788 if (a
->data
.explicit_location
&& b
->data
.explicit_location
)
1789 return b
->data
.location
- a
->data
.location
;
1791 if (a
->data
.explicit_location
&& !b
->data
.explicit_location
)
1794 if (!a
->data
.explicit_location
&& b
->data
.explicit_location
)
1797 return -strcmp(a
->name
, b
->name
);
1801 * Sort the shader IO variables into canonical order
1804 canonicalize_shader_io(exec_list
*ir
, enum ir_variable_mode io_mode
)
1806 ir_variable
*var_table
[MAX_PROGRAM_OUTPUTS
* 4];
1807 unsigned num_variables
= 0;
1809 foreach_in_list(ir_instruction
, node
, ir
) {
1810 ir_variable
*const var
= node
->as_variable();
1812 if (var
== NULL
|| var
->data
.mode
!= io_mode
)
1815 /* If we have already encountered more I/O variables that could
1816 * successfully link, bail.
1818 if (num_variables
== ARRAY_SIZE(var_table
))
1821 var_table
[num_variables
++] = var
;
1824 if (num_variables
== 0)
1827 /* Sort the list in reverse order (io_variable_cmp handles this). Later
1828 * we're going to push the variables on to the IR list as a stack, so we
1829 * want the last variable (in canonical order) to be first in the list.
1831 qsort(var_table
, num_variables
, sizeof(var_table
[0]), io_variable_cmp
);
1833 /* Remove the variable from it's current location in the IR, and put it at
1836 for (unsigned i
= 0; i
< num_variables
; i
++) {
1837 var_table
[i
]->remove();
1838 ir
->push_head(var_table
[i
]);
1843 * Generate a bitfield map of the explicit locations for shader varyings.
1845 * In theory a 32 bits value will be enough but a 64 bits value is future proof.
1848 reserved_varying_slot(struct gl_shader
*stage
, ir_variable_mode io_mode
)
1850 assert(io_mode
== ir_var_shader_in
|| io_mode
== ir_var_shader_out
);
1851 assert(MAX_VARYING
<= 64); /* avoid an overflow of the returned value */
1859 foreach_in_list(ir_instruction
, node
, stage
->ir
) {
1860 ir_variable
*const var
= node
->as_variable();
1862 if (var
== NULL
|| var
->data
.mode
!= io_mode
||
1863 !var
->data
.explicit_location
||
1864 var
->data
.location
< VARYING_SLOT_VAR0
)
1867 var_slot
= var
->data
.location
- VARYING_SLOT_VAR0
;
1869 unsigned num_elements
= get_varying_type(var
, stage
->Stage
)
1870 ->count_attribute_slots(stage
->Stage
== MESA_SHADER_VERTEX
);
1871 for (unsigned i
= 0; i
< num_elements
; i
++) {
1872 if (var_slot
>= 0 && var_slot
< MAX_VARYING
)
1873 slots
|= UINT64_C(1) << var_slot
;
1883 * Assign locations for all variables that are produced in one pipeline stage
1884 * (the "producer") and consumed in the next stage (the "consumer").
1886 * Variables produced by the producer may also be consumed by transform
1889 * \param num_tfeedback_decls is the number of declarations indicating
1890 * variables that may be consumed by transform feedback.
1892 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
1893 * representing the result of parsing the strings passed to
1894 * glTransformFeedbackVaryings(). assign_location() will be called for
1895 * each of these objects that matches one of the outputs of the
1898 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
1899 * be NULL. In this case, varying locations are assigned solely based on the
1900 * requirements of transform feedback.
1903 assign_varying_locations(struct gl_context
*ctx
,
1905 struct gl_shader_program
*prog
,
1906 gl_shader
*producer
, gl_shader
*consumer
,
1907 unsigned num_tfeedback_decls
,
1908 tfeedback_decl
*tfeedback_decls
)
1910 /* Tessellation shaders treat inputs and outputs as shared memory and can
1911 * access inputs and outputs of other invocations.
1912 * Therefore, they can't be lowered to temps easily (and definitely not
1915 bool unpackable_tess
=
1916 (consumer
&& consumer
->Stage
== MESA_SHADER_TESS_EVAL
) ||
1917 (consumer
&& consumer
->Stage
== MESA_SHADER_TESS_CTRL
) ||
1918 (producer
&& producer
->Stage
== MESA_SHADER_TESS_CTRL
);
1920 /* Transform feedback code assumes varying arrays are packed, so if the
1921 * driver has disabled varying packing, make sure to at least enable
1922 * packing required by transform feedback.
1925 ctx
->Extensions
.EXT_transform_feedback
&& !unpackable_tess
;
1927 /* Disable varying packing for GL 4.4+ as there is no guarantee
1928 * that interpolation qualifiers will match between shaders in these
1929 * versions. We also disable packing on outerward facing interfaces for
1930 * SSO because in ES we need to retain the unpacked varying information
1931 * for draw time validation. For desktop GL we could allow packing for
1932 * versions < 4.4 but its just safer not to do packing.
1934 * Packing is still enabled on individual arrays, structs, and matrices as
1935 * these are required by the transform feedback code and it is still safe
1936 * to do so. We also enable packing when a varying is only used for
1937 * transform feedback and its not a SSO.
1939 * Varying packing currently only packs together varyings with matching
1940 * interpolation qualifiers as the backends assume all packed components
1941 * are to be processed in the same way. Therefore we cannot do packing in
1942 * these versions of GL without the risk of mismatching interfaces.
1944 * From Section 4.5 (Interpolation Qualifiers) of the GLSL 4.30 spec:
1946 * "The type and presence of interpolation qualifiers of variables with
1947 * the same name declared in all linked shaders for the same cross-stage
1948 * interface must match, otherwise the link command will fail.
1950 * When comparing an output from one stage to an input of a subsequent
1951 * stage, the input and output don't match if their interpolation
1952 * qualifiers (or lack thereof) are not the same."
1954 * This text was also in at least revison 7 of the 4.40 spec but is no
1955 * longer in revision 9 and not in the 4.50 spec.
1957 bool disable_varying_packing
=
1958 ctx
->Const
.DisableVaryingPacking
|| unpackable_tess
;
1959 if ((ctx
->API
== API_OPENGL_CORE
&& ctx
->Version
>= 44) ||
1960 (prog
->SeparateShader
&& (producer
== NULL
|| consumer
== NULL
)))
1961 disable_varying_packing
= true;
1963 varying_matches
matches(disable_varying_packing
, xfb_enabled
,
1964 producer
? producer
->Stage
: (gl_shader_stage
)-1,
1965 consumer
? consumer
->Stage
: (gl_shader_stage
)-1);
1966 hash_table
*tfeedback_candidates
1967 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1968 hash_table
*consumer_inputs
1969 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1970 hash_table
*consumer_interface_inputs
1971 = hash_table_ctor(0, hash_table_string_hash
, hash_table_string_compare
);
1972 ir_variable
*consumer_inputs_with_locations
[VARYING_SLOT_TESS_MAX
] = {
1976 unsigned consumer_vertices
= 0;
1977 if (consumer
&& consumer
->Stage
== MESA_SHADER_GEOMETRY
)
1978 consumer_vertices
= prog
->Geom
.VerticesIn
;
1980 /* Operate in a total of four passes.
1982 * 1. Sort inputs / outputs into a canonical order. This is necessary so
1983 * that inputs / outputs of separable shaders will be assigned
1984 * predictable locations regardless of the order in which declarations
1985 * appeared in the shader source.
1987 * 2. Assign locations for any matching inputs and outputs.
1989 * 3. Mark output variables in the producer that do not have locations as
1990 * not being outputs. This lets the optimizer eliminate them.
1992 * 4. Mark input variables in the consumer that do not have locations as
1993 * not being inputs. This lets the optimizer eliminate them.
1996 canonicalize_shader_io(consumer
->ir
, ir_var_shader_in
);
1999 canonicalize_shader_io(producer
->ir
, ir_var_shader_out
);
2002 linker::populate_consumer_input_sets(mem_ctx
, consumer
->ir
,
2004 consumer_interface_inputs
,
2005 consumer_inputs_with_locations
);
2008 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2009 ir_variable
*const output_var
= node
->as_variable();
2011 if ((output_var
== NULL
) ||
2012 (output_var
->data
.mode
!= ir_var_shader_out
))
2015 /* Only geometry shaders can use non-zero streams */
2016 assert(output_var
->data
.stream
== 0 ||
2017 (output_var
->data
.stream
< MAX_VERTEX_STREAMS
&&
2018 producer
->Stage
== MESA_SHADER_GEOMETRY
));
2020 if (num_tfeedback_decls
> 0) {
2021 tfeedback_candidate_generator
g(mem_ctx
, tfeedback_candidates
);
2022 g
.process(output_var
);
2025 ir_variable
*const input_var
=
2026 linker::get_matching_input(mem_ctx
, output_var
, consumer_inputs
,
2027 consumer_interface_inputs
,
2028 consumer_inputs_with_locations
);
2030 /* If a matching input variable was found, add this ouptut (and the
2031 * input) to the set. If this is a separable program and there is no
2032 * consumer stage, add the output.
2034 * Always add TCS outputs. They are shared by all invocations
2035 * within a patch and can be used as shared memory.
2037 if (input_var
|| (prog
->SeparateShader
&& consumer
== NULL
) ||
2038 producer
->Type
== GL_TESS_CONTROL_SHADER
) {
2039 matches
.record(output_var
, input_var
);
2042 /* Only stream 0 outputs can be consumed in the next stage */
2043 if (input_var
&& output_var
->data
.stream
!= 0) {
2044 linker_error(prog
, "output %s is assigned to stream=%d but "
2045 "is linked to an input, which requires stream=0",
2046 output_var
->name
, output_var
->data
.stream
);
2051 /* If there's no producer stage, then this must be a separable program.
2052 * For example, we may have a program that has just a fragment shader.
2053 * Later this program will be used with some arbitrary vertex (or
2054 * geometry) shader program. This means that locations must be assigned
2055 * for all the inputs.
2057 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2058 ir_variable
*const input_var
= node
->as_variable();
2060 if ((input_var
== NULL
) ||
2061 (input_var
->data
.mode
!= ir_var_shader_in
))
2064 matches
.record(NULL
, input_var
);
2068 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2069 if (!tfeedback_decls
[i
].is_varying())
2072 const tfeedback_candidate
*matched_candidate
2073 = tfeedback_decls
[i
].find_candidate(prog
, tfeedback_candidates
);
2075 if (matched_candidate
== NULL
) {
2076 hash_table_dtor(tfeedback_candidates
);
2077 hash_table_dtor(consumer_inputs
);
2078 hash_table_dtor(consumer_interface_inputs
);
2082 if (matched_candidate
->toplevel_var
->data
.is_unmatched_generic_inout
) {
2083 matched_candidate
->toplevel_var
->data
.is_xfb_only
= 1;
2084 matches
.record(matched_candidate
->toplevel_var
, NULL
);
2088 const uint64_t reserved_slots
=
2089 reserved_varying_slot(producer
, ir_var_shader_out
) |
2090 reserved_varying_slot(consumer
, ir_var_shader_in
);
2092 const unsigned slots_used
= matches
.assign_locations(prog
, reserved_slots
,
2093 prog
->SeparateShader
);
2094 matches
.store_locations();
2096 for (unsigned i
= 0; i
< num_tfeedback_decls
; ++i
) {
2097 if (!tfeedback_decls
[i
].is_varying())
2100 if (!tfeedback_decls
[i
].assign_location(ctx
, prog
)) {
2101 hash_table_dtor(tfeedback_candidates
);
2102 hash_table_dtor(consumer_inputs
);
2103 hash_table_dtor(consumer_interface_inputs
);
2108 hash_table_dtor(tfeedback_candidates
);
2109 hash_table_dtor(consumer_inputs
);
2110 hash_table_dtor(consumer_interface_inputs
);
2112 if (consumer
&& producer
) {
2113 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2114 ir_variable
*const var
= node
->as_variable();
2116 if (var
&& var
->data
.mode
== ir_var_shader_in
&&
2117 var
->data
.is_unmatched_generic_inout
) {
2118 if (!prog
->IsES
&& prog
->Version
<= 120) {
2119 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
2121 * Only those varying variables used (i.e. read) in
2122 * the fragment shader executable must be written to
2123 * by the vertex shader executable; declaring
2124 * superfluous varying variables in a vertex shader is
2127 * We interpret this text as meaning that the VS must
2128 * write the variable for the FS to read it. See
2129 * "glsl1-varying read but not written" in piglit.
2131 linker_error(prog
, "%s shader varying %s not written "
2133 _mesa_shader_stage_to_string(consumer
->Stage
),
2135 _mesa_shader_stage_to_string(producer
->Stage
));
2137 linker_warning(prog
, "%s shader varying %s not written "
2139 _mesa_shader_stage_to_string(consumer
->Stage
),
2141 _mesa_shader_stage_to_string(producer
->Stage
));
2146 /* Now that validation is done its safe to remove unused varyings. As
2147 * we have both a producer and consumer its safe to remove unused
2148 * varyings even if the program is a SSO because the stages are being
2149 * linked together i.e. we have a multi-stage SSO.
2151 remove_unused_shader_inputs_and_outputs(false, producer
,
2153 remove_unused_shader_inputs_and_outputs(false, consumer
,
2158 lower_packed_varyings(mem_ctx
, slots_used
, ir_var_shader_out
,
2159 0, producer
, disable_varying_packing
,
2164 lower_packed_varyings(mem_ctx
, slots_used
, ir_var_shader_in
,
2165 consumer_vertices
, consumer
,
2166 disable_varying_packing
, xfb_enabled
);
2173 check_against_output_limit(struct gl_context
*ctx
,
2174 struct gl_shader_program
*prog
,
2175 gl_shader
*producer
)
2177 unsigned output_vectors
= 0;
2179 foreach_in_list(ir_instruction
, node
, producer
->ir
) {
2180 ir_variable
*const var
= node
->as_variable();
2182 if (var
&& var
->data
.mode
== ir_var_shader_out
&&
2183 var_counts_against_varying_limit(producer
->Stage
, var
)) {
2184 /* outputs for fragment shader can't be doubles */
2185 output_vectors
+= var
->type
->count_attribute_slots(false);
2189 assert(producer
->Stage
!= MESA_SHADER_FRAGMENT
);
2190 unsigned max_output_components
=
2191 ctx
->Const
.Program
[producer
->Stage
].MaxOutputComponents
;
2193 const unsigned output_components
= output_vectors
* 4;
2194 if (output_components
> max_output_components
) {
2195 if (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)
2196 linker_error(prog
, "%s shader uses too many output vectors "
2198 _mesa_shader_stage_to_string(producer
->Stage
),
2200 max_output_components
/ 4);
2202 linker_error(prog
, "%s shader uses too many output components "
2204 _mesa_shader_stage_to_string(producer
->Stage
),
2206 max_output_components
);
2215 check_against_input_limit(struct gl_context
*ctx
,
2216 struct gl_shader_program
*prog
,
2217 gl_shader
*consumer
)
2219 unsigned input_vectors
= 0;
2221 foreach_in_list(ir_instruction
, node
, consumer
->ir
) {
2222 ir_variable
*const var
= node
->as_variable();
2224 if (var
&& var
->data
.mode
== ir_var_shader_in
&&
2225 var_counts_against_varying_limit(consumer
->Stage
, var
)) {
2226 /* vertex inputs aren't varying counted */
2227 input_vectors
+= var
->type
->count_attribute_slots(false);
2231 assert(consumer
->Stage
!= MESA_SHADER_VERTEX
);
2232 unsigned max_input_components
=
2233 ctx
->Const
.Program
[consumer
->Stage
].MaxInputComponents
;
2235 const unsigned input_components
= input_vectors
* 4;
2236 if (input_components
> max_input_components
) {
2237 if (ctx
->API
== API_OPENGLES2
|| prog
->IsES
)
2238 linker_error(prog
, "%s shader uses too many input vectors "
2240 _mesa_shader_stage_to_string(consumer
->Stage
),
2242 max_input_components
/ 4);
2244 linker_error(prog
, "%s shader uses too many input components "
2246 _mesa_shader_stage_to_string(consumer
->Stage
),
2248 max_input_components
);