glsl/glcpp: Rename one test to avoid a duplicate test number
[mesa.git] / src / glsl / link_varyings.cpp
1 /*
2 * Copyright © 2012 Intel Corporation
3 *
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:
10 *
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
13 * Software.
14 *
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.
22 */
23
24 /**
25 * \file link_varyings.cpp
26 *
27 * Linker functions related specifically to linking varyings between shader
28 * stages.
29 */
30
31
32 #include "main/mtypes.h"
33 #include "glsl_symbol_table.h"
34 #include "glsl_parser_extras.h"
35 #include "ir_optimization.h"
36 #include "linker.h"
37 #include "link_varyings.h"
38 #include "main/macros.h"
39 #include "program/hash_table.h"
40 #include "program.h"
41
42
43 /**
44 * Validate the types and qualifiers of an output from one stage against the
45 * matching input to another stage.
46 */
47 static void
48 cross_validate_types_and_qualifiers(struct gl_shader_program *prog,
49 const ir_variable *input,
50 const ir_variable *output,
51 gl_shader_stage consumer_stage,
52 gl_shader_stage producer_stage)
53 {
54 /* Check that the types match between stages.
55 */
56 const glsl_type *type_to_match = input->type;
57 if (consumer_stage == MESA_SHADER_GEOMETRY) {
58 assert(type_to_match->is_array()); /* Enforced by ast_to_hir */
59 type_to_match = type_to_match->element_type();
60 }
61 if (type_to_match != output->type) {
62 /* There is a bit of a special case for gl_TexCoord. This
63 * built-in is unsized by default. Applications that variable
64 * access it must redeclare it with a size. There is some
65 * language in the GLSL spec that implies the fragment shader
66 * and vertex shader do not have to agree on this size. Other
67 * driver behave this way, and one or two applications seem to
68 * rely on it.
69 *
70 * Neither declaration needs to be modified here because the array
71 * sizes are fixed later when update_array_sizes is called.
72 *
73 * From page 48 (page 54 of the PDF) of the GLSL 1.10 spec:
74 *
75 * "Unlike user-defined varying variables, the built-in
76 * varying variables don't have a strict one-to-one
77 * correspondence between the vertex language and the
78 * fragment language."
79 */
80 if (!output->type->is_array() || !is_gl_identifier(output->name)) {
81 linker_error(prog,
82 "%s shader output `%s' declared as type `%s', "
83 "but %s shader input declared as type `%s'\n",
84 _mesa_shader_stage_to_string(producer_stage),
85 output->name,
86 output->type->name,
87 _mesa_shader_stage_to_string(consumer_stage),
88 input->type->name);
89 return;
90 }
91 }
92
93 /* Check that all of the qualifiers match between stages.
94 */
95 if (input->data.centroid != output->data.centroid) {
96 linker_error(prog,
97 "%s shader output `%s' %s centroid qualifier, "
98 "but %s shader input %s centroid qualifier\n",
99 _mesa_shader_stage_to_string(producer_stage),
100 output->name,
101 (output->data.centroid) ? "has" : "lacks",
102 _mesa_shader_stage_to_string(consumer_stage),
103 (input->data.centroid) ? "has" : "lacks");
104 return;
105 }
106
107 if (input->data.sample != output->data.sample) {
108 linker_error(prog,
109 "%s shader output `%s' %s sample qualifier, "
110 "but %s shader input %s sample qualifier\n",
111 _mesa_shader_stage_to_string(producer_stage),
112 output->name,
113 (output->data.sample) ? "has" : "lacks",
114 _mesa_shader_stage_to_string(consumer_stage),
115 (input->data.sample) ? "has" : "lacks");
116 return;
117 }
118
119 if (input->data.invariant != output->data.invariant) {
120 linker_error(prog,
121 "%s shader output `%s' %s invariant qualifier, "
122 "but %s shader input %s invariant qualifier\n",
123 _mesa_shader_stage_to_string(producer_stage),
124 output->name,
125 (output->data.invariant) ? "has" : "lacks",
126 _mesa_shader_stage_to_string(consumer_stage),
127 (input->data.invariant) ? "has" : "lacks");
128 return;
129 }
130
131 if (input->data.interpolation != output->data.interpolation) {
132 linker_error(prog,
133 "%s shader output `%s' specifies %s "
134 "interpolation qualifier, "
135 "but %s shader input specifies %s "
136 "interpolation qualifier\n",
137 _mesa_shader_stage_to_string(producer_stage),
138 output->name,
139 interpolation_string(output->data.interpolation),
140 _mesa_shader_stage_to_string(consumer_stage),
141 interpolation_string(input->data.interpolation));
142 return;
143 }
144 }
145
146 /**
147 * Validate front and back color outputs against single color input
148 */
149 static void
150 cross_validate_front_and_back_color(struct gl_shader_program *prog,
151 const ir_variable *input,
152 const ir_variable *front_color,
153 const ir_variable *back_color,
154 gl_shader_stage consumer_stage,
155 gl_shader_stage producer_stage)
156 {
157 if (front_color != NULL && front_color->data.assigned)
158 cross_validate_types_and_qualifiers(prog, input, front_color,
159 consumer_stage, producer_stage);
160
161 if (back_color != NULL && back_color->data.assigned)
162 cross_validate_types_and_qualifiers(prog, input, back_color,
163 consumer_stage, producer_stage);
164 }
165
166 /**
167 * Validate that outputs from one stage match inputs of another
168 */
169 void
170 cross_validate_outputs_to_inputs(struct gl_shader_program *prog,
171 gl_shader *producer, gl_shader *consumer)
172 {
173 glsl_symbol_table parameters;
174 ir_variable *explicit_locations[MAX_VARYING] = { NULL, };
175
176 /* Find all shader outputs in the "producer" stage.
177 */
178 foreach_in_list(ir_instruction, node, producer->ir) {
179 ir_variable *const var = node->as_variable();
180
181 if ((var == NULL) || (var->data.mode != ir_var_shader_out))
182 continue;
183
184 if (!var->data.explicit_location
185 || var->data.location < VARYING_SLOT_VAR0)
186 parameters.add_variable(var);
187 else {
188 /* User-defined varyings with explicit locations are handled
189 * differently because they do not need to have matching names.
190 */
191 const unsigned idx = var->data.location - VARYING_SLOT_VAR0;
192
193 if (explicit_locations[idx] != NULL) {
194 linker_error(prog,
195 "%s shader has multiple outputs explicitly "
196 "assigned to location %d\n",
197 _mesa_shader_stage_to_string(producer->Stage),
198 idx);
199 return;
200 }
201
202 explicit_locations[idx] = var;
203 }
204 }
205
206
207 /* Find all shader inputs in the "consumer" stage. Any variables that have
208 * matching outputs already in the symbol table must have the same type and
209 * qualifiers.
210 *
211 * Exception: if the consumer is the geometry shader, then the inputs
212 * should be arrays and the type of the array element should match the type
213 * of the corresponding producer output.
214 */
215 foreach_in_list(ir_instruction, node, consumer->ir) {
216 ir_variable *const input = node->as_variable();
217
218 if ((input == NULL) || (input->data.mode != ir_var_shader_in))
219 continue;
220
221 if (strcmp(input->name, "gl_Color") == 0 && input->data.used) {
222 const ir_variable *const front_color =
223 parameters.get_variable("gl_FrontColor");
224
225 const ir_variable *const back_color =
226 parameters.get_variable("gl_BackColor");
227
228 cross_validate_front_and_back_color(prog, input,
229 front_color, back_color,
230 consumer->Stage, producer->Stage);
231 } else if (strcmp(input->name, "gl_SecondaryColor") == 0 && input->data.used) {
232 const ir_variable *const front_color =
233 parameters.get_variable("gl_FrontSecondaryColor");
234
235 const ir_variable *const back_color =
236 parameters.get_variable("gl_BackSecondaryColor");
237
238 cross_validate_front_and_back_color(prog, input,
239 front_color, back_color,
240 consumer->Stage, producer->Stage);
241 } else {
242 /* The rules for connecting inputs and outputs change in the presence
243 * of explicit locations. In this case, we no longer care about the
244 * names of the variables. Instead, we care only about the
245 * explicitly assigned location.
246 */
247 ir_variable *output = NULL;
248 if (input->data.explicit_location
249 && input->data.location >= VARYING_SLOT_VAR0) {
250 output = explicit_locations[input->data.location - VARYING_SLOT_VAR0];
251
252 if (output == NULL) {
253 linker_error(prog,
254 "%s shader input `%s' with explicit location "
255 "has no matching output\n",
256 _mesa_shader_stage_to_string(consumer->Stage),
257 input->name);
258 }
259 } else {
260 output = parameters.get_variable(input->name);
261 }
262
263 if (output != NULL) {
264 cross_validate_types_and_qualifiers(prog, input, output,
265 consumer->Stage, producer->Stage);
266 }
267 }
268 }
269 }
270
271
272 /**
273 * Initialize this object based on a string that was passed to
274 * glTransformFeedbackVaryings.
275 *
276 * If the input is mal-formed, this call still succeeds, but it sets
277 * this->var_name to a mal-formed input, so tfeedback_decl::find_output_var()
278 * will fail to find any matching variable.
279 */
280 void
281 tfeedback_decl::init(struct gl_context *ctx, const void *mem_ctx,
282 const char *input)
283 {
284 /* We don't have to be pedantic about what is a valid GLSL variable name,
285 * because any variable with an invalid name can't exist in the IR anyway.
286 */
287
288 this->location = -1;
289 this->orig_name = input;
290 this->is_clip_distance_mesa = false;
291 this->skip_components = 0;
292 this->next_buffer_separator = false;
293 this->matched_candidate = NULL;
294 this->stream_id = 0;
295
296 if (ctx->Extensions.ARB_transform_feedback3) {
297 /* Parse gl_NextBuffer. */
298 if (strcmp(input, "gl_NextBuffer") == 0) {
299 this->next_buffer_separator = true;
300 return;
301 }
302
303 /* Parse gl_SkipComponents. */
304 if (strcmp(input, "gl_SkipComponents1") == 0)
305 this->skip_components = 1;
306 else if (strcmp(input, "gl_SkipComponents2") == 0)
307 this->skip_components = 2;
308 else if (strcmp(input, "gl_SkipComponents3") == 0)
309 this->skip_components = 3;
310 else if (strcmp(input, "gl_SkipComponents4") == 0)
311 this->skip_components = 4;
312
313 if (this->skip_components)
314 return;
315 }
316
317 /* Parse a declaration. */
318 const char *base_name_end;
319 long subscript = parse_program_resource_name(input, &base_name_end);
320 this->var_name = ralloc_strndup(mem_ctx, input, base_name_end - input);
321 if (subscript >= 0) {
322 this->array_subscript = subscript;
323 this->is_subscripted = true;
324 } else {
325 this->is_subscripted = false;
326 }
327
328 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this
329 * class must behave specially to account for the fact that gl_ClipDistance
330 * is converted from a float[8] to a vec4[2].
331 */
332 if (ctx->ShaderCompilerOptions[MESA_SHADER_VERTEX].LowerClipDistance &&
333 strcmp(this->var_name, "gl_ClipDistance") == 0) {
334 this->is_clip_distance_mesa = true;
335 }
336 }
337
338
339 /**
340 * Determine whether two tfeedback_decl objects refer to the same variable and
341 * array index (if applicable).
342 */
343 bool
344 tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y)
345 {
346 assert(x.is_varying() && y.is_varying());
347
348 if (strcmp(x.var_name, y.var_name) != 0)
349 return false;
350 if (x.is_subscripted != y.is_subscripted)
351 return false;
352 if (x.is_subscripted && x.array_subscript != y.array_subscript)
353 return false;
354 return true;
355 }
356
357
358 /**
359 * Assign a location and stream ID for this tfeedback_decl object based on the
360 * transform feedback candidate found by find_candidate.
361 *
362 * If an error occurs, the error is reported through linker_error() and false
363 * is returned.
364 */
365 bool
366 tfeedback_decl::assign_location(struct gl_context *ctx,
367 struct gl_shader_program *prog)
368 {
369 assert(this->is_varying());
370
371 unsigned fine_location
372 = this->matched_candidate->toplevel_var->data.location * 4
373 + this->matched_candidate->toplevel_var->data.location_frac
374 + this->matched_candidate->offset;
375
376 if (this->matched_candidate->type->is_array()) {
377 /* Array variable */
378 const unsigned matrix_cols =
379 this->matched_candidate->type->fields.array->matrix_columns;
380 const unsigned vector_elements =
381 this->matched_candidate->type->fields.array->vector_elements;
382 unsigned actual_array_size = this->is_clip_distance_mesa ?
383 prog->LastClipDistanceArraySize :
384 this->matched_candidate->type->array_size();
385
386 if (this->is_subscripted) {
387 /* Check array bounds. */
388 if (this->array_subscript >= actual_array_size) {
389 linker_error(prog, "Transform feedback varying %s has index "
390 "%i, but the array size is %u.",
391 this->orig_name, this->array_subscript,
392 actual_array_size);
393 return false;
394 }
395 unsigned array_elem_size = this->is_clip_distance_mesa ?
396 1 : vector_elements * matrix_cols;
397 fine_location += array_elem_size * this->array_subscript;
398 this->size = 1;
399 } else {
400 this->size = actual_array_size;
401 }
402 this->vector_elements = vector_elements;
403 this->matrix_columns = matrix_cols;
404 if (this->is_clip_distance_mesa)
405 this->type = GL_FLOAT;
406 else
407 this->type = this->matched_candidate->type->fields.array->gl_type;
408 } else {
409 /* Regular variable (scalar, vector, or matrix) */
410 if (this->is_subscripted) {
411 linker_error(prog, "Transform feedback varying %s requested, "
412 "but %s is not an array.",
413 this->orig_name, this->var_name);
414 return false;
415 }
416 this->size = 1;
417 this->vector_elements = this->matched_candidate->type->vector_elements;
418 this->matrix_columns = this->matched_candidate->type->matrix_columns;
419 this->type = this->matched_candidate->type->gl_type;
420 }
421 this->location = fine_location / 4;
422 this->location_frac = fine_location % 4;
423
424 /* From GL_EXT_transform_feedback:
425 * A program will fail to link if:
426 *
427 * * the total number of components to capture in any varying
428 * variable in <varyings> is greater than the constant
429 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
430 * buffer mode is SEPARATE_ATTRIBS_EXT;
431 */
432 if (prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS &&
433 this->num_components() >
434 ctx->Const.MaxTransformFeedbackSeparateComponents) {
435 linker_error(prog, "Transform feedback varying %s exceeds "
436 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
437 this->orig_name);
438 return false;
439 }
440
441 /* Only transform feedback varyings can be assigned to non-zero streams,
442 * so assign the stream id here.
443 */
444 this->stream_id = this->matched_candidate->toplevel_var->data.stream;
445
446 return true;
447 }
448
449
450 unsigned
451 tfeedback_decl::get_num_outputs() const
452 {
453 if (!this->is_varying()) {
454 return 0;
455 }
456
457 return (this->num_components() + this->location_frac + 3)/4;
458 }
459
460
461 /**
462 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
463 *
464 * If an error occurs, the error is reported through linker_error() and false
465 * is returned.
466 */
467 bool
468 tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog,
469 struct gl_transform_feedback_info *info,
470 unsigned buffer, const unsigned max_outputs) const
471 {
472 assert(!this->next_buffer_separator);
473
474 /* Handle gl_SkipComponents. */
475 if (this->skip_components) {
476 info->BufferStride[buffer] += this->skip_components;
477 return true;
478 }
479
480 /* From GL_EXT_transform_feedback:
481 * A program will fail to link if:
482 *
483 * * the total number of components to capture is greater than
484 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
485 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
486 */
487 if (prog->TransformFeedback.BufferMode == GL_INTERLEAVED_ATTRIBS &&
488 info->BufferStride[buffer] + this->num_components() >
489 ctx->Const.MaxTransformFeedbackInterleavedComponents) {
490 linker_error(prog, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
491 "limit has been exceeded.");
492 return false;
493 }
494
495 unsigned location = this->location;
496 unsigned location_frac = this->location_frac;
497 unsigned num_components = this->num_components();
498 while (num_components > 0) {
499 unsigned output_size = MIN2(num_components, 4 - location_frac);
500 assert(info->NumOutputs < max_outputs);
501 info->Outputs[info->NumOutputs].ComponentOffset = location_frac;
502 info->Outputs[info->NumOutputs].OutputRegister = location;
503 info->Outputs[info->NumOutputs].NumComponents = output_size;
504 info->Outputs[info->NumOutputs].StreamId = stream_id;
505 info->Outputs[info->NumOutputs].OutputBuffer = buffer;
506 info->Outputs[info->NumOutputs].DstOffset = info->BufferStride[buffer];
507 ++info->NumOutputs;
508 info->BufferStride[buffer] += output_size;
509 num_components -= output_size;
510 location++;
511 location_frac = 0;
512 }
513
514 info->Varyings[info->NumVarying].Name = ralloc_strdup(prog, this->orig_name);
515 info->Varyings[info->NumVarying].Type = this->type;
516 info->Varyings[info->NumVarying].Size = this->size;
517 info->NumVarying++;
518
519 return true;
520 }
521
522
523 const tfeedback_candidate *
524 tfeedback_decl::find_candidate(gl_shader_program *prog,
525 hash_table *tfeedback_candidates)
526 {
527 const char *name = this->is_clip_distance_mesa
528 ? "gl_ClipDistanceMESA" : this->var_name;
529 this->matched_candidate = (const tfeedback_candidate *)
530 hash_table_find(tfeedback_candidates, name);
531 if (!this->matched_candidate) {
532 /* From GL_EXT_transform_feedback:
533 * A program will fail to link if:
534 *
535 * * any variable name specified in the <varyings> array is not
536 * declared as an output in the geometry shader (if present) or
537 * the vertex shader (if no geometry shader is present);
538 */
539 linker_error(prog, "Transform feedback varying %s undeclared.",
540 this->orig_name);
541 }
542 return this->matched_candidate;
543 }
544
545
546 /**
547 * Parse all the transform feedback declarations that were passed to
548 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
549 *
550 * If an error occurs, the error is reported through linker_error() and false
551 * is returned.
552 */
553 bool
554 parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog,
555 const void *mem_ctx, unsigned num_names,
556 char **varying_names, tfeedback_decl *decls)
557 {
558 for (unsigned i = 0; i < num_names; ++i) {
559 decls[i].init(ctx, mem_ctx, varying_names[i]);
560
561 if (!decls[i].is_varying())
562 continue;
563
564 /* From GL_EXT_transform_feedback:
565 * A program will fail to link if:
566 *
567 * * any two entries in the <varyings> array specify the same varying
568 * variable;
569 *
570 * We interpret this to mean "any two entries in the <varyings> array
571 * specify the same varying variable and array index", since transform
572 * feedback of arrays would be useless otherwise.
573 */
574 for (unsigned j = 0; j < i; ++j) {
575 if (!decls[j].is_varying())
576 continue;
577
578 if (tfeedback_decl::is_same(decls[i], decls[j])) {
579 linker_error(prog, "Transform feedback varying %s specified "
580 "more than once.", varying_names[i]);
581 return false;
582 }
583 }
584 }
585 return true;
586 }
587
588
589 /**
590 * Store transform feedback location assignments into
591 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
592 *
593 * If an error occurs, the error is reported through linker_error() and false
594 * is returned.
595 */
596 bool
597 store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog,
598 unsigned num_tfeedback_decls,
599 tfeedback_decl *tfeedback_decls)
600 {
601 bool separate_attribs_mode =
602 prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS;
603
604 ralloc_free(prog->LinkedTransformFeedback.Varyings);
605 ralloc_free(prog->LinkedTransformFeedback.Outputs);
606
607 memset(&prog->LinkedTransformFeedback, 0,
608 sizeof(prog->LinkedTransformFeedback));
609
610 prog->LinkedTransformFeedback.Varyings =
611 rzalloc_array(prog,
612 struct gl_transform_feedback_varying_info,
613 num_tfeedback_decls);
614
615 unsigned num_outputs = 0;
616 for (unsigned i = 0; i < num_tfeedback_decls; ++i)
617 num_outputs += tfeedback_decls[i].get_num_outputs();
618
619 prog->LinkedTransformFeedback.Outputs =
620 rzalloc_array(prog,
621 struct gl_transform_feedback_output,
622 num_outputs);
623
624 unsigned num_buffers = 0;
625
626 if (separate_attribs_mode) {
627 /* GL_SEPARATE_ATTRIBS */
628 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
629 if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback,
630 num_buffers, num_outputs))
631 return false;
632
633 num_buffers++;
634 }
635 }
636 else {
637 /* GL_INVERLEAVED_ATTRIBS */
638 int buffer_stream_id = -1;
639 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
640 if (tfeedback_decls[i].is_next_buffer_separator()) {
641 num_buffers++;
642 buffer_stream_id = -1;
643 continue;
644 } else if (buffer_stream_id == -1) {
645 /* First varying writing to this buffer: remember its stream */
646 buffer_stream_id = (int) tfeedback_decls[i].get_stream_id();
647 } else if (buffer_stream_id !=
648 (int) tfeedback_decls[i].get_stream_id()) {
649 /* Varying writes to the same buffer from a different stream */
650 linker_error(prog,
651 "Transform feedback can't capture varyings belonging "
652 "to different vertex streams in a single buffer. "
653 "Varying %s writes to buffer from stream %u, other "
654 "varyings in the same buffer write from stream %u.",
655 tfeedback_decls[i].name(),
656 tfeedback_decls[i].get_stream_id(),
657 buffer_stream_id);
658 return false;
659 }
660
661 if (!tfeedback_decls[i].store(ctx, prog,
662 &prog->LinkedTransformFeedback,
663 num_buffers, num_outputs))
664 return false;
665 }
666 num_buffers++;
667 }
668
669 assert(prog->LinkedTransformFeedback.NumOutputs == num_outputs);
670
671 prog->LinkedTransformFeedback.NumBuffers = num_buffers;
672 return true;
673 }
674
675 namespace {
676
677 /**
678 * Data structure recording the relationship between outputs of one shader
679 * stage (the "producer") and inputs of another (the "consumer").
680 */
681 class varying_matches
682 {
683 public:
684 varying_matches(bool disable_varying_packing, bool consumer_is_fs);
685 ~varying_matches();
686 void record(ir_variable *producer_var, ir_variable *consumer_var);
687 unsigned assign_locations();
688 void store_locations() const;
689
690 private:
691 /**
692 * If true, this driver disables varying packing, so all varyings need to
693 * be aligned on slot boundaries, and take up a number of slots equal to
694 * their number of matrix columns times their array size.
695 */
696 const bool disable_varying_packing;
697
698 /**
699 * Enum representing the order in which varyings are packed within a
700 * packing class.
701 *
702 * Currently we pack vec4's first, then vec2's, then scalar values, then
703 * vec3's. This order ensures that the only vectors that are at risk of
704 * having to be "double parked" (split between two adjacent varying slots)
705 * are the vec3's.
706 */
707 enum packing_order_enum {
708 PACKING_ORDER_VEC4,
709 PACKING_ORDER_VEC2,
710 PACKING_ORDER_SCALAR,
711 PACKING_ORDER_VEC3,
712 };
713
714 static unsigned compute_packing_class(const ir_variable *var);
715 static packing_order_enum compute_packing_order(const ir_variable *var);
716 static int match_comparator(const void *x_generic, const void *y_generic);
717
718 /**
719 * Structure recording the relationship between a single producer output
720 * and a single consumer input.
721 */
722 struct match {
723 /**
724 * Packing class for this varying, computed by compute_packing_class().
725 */
726 unsigned packing_class;
727
728 /**
729 * Packing order for this varying, computed by compute_packing_order().
730 */
731 packing_order_enum packing_order;
732 unsigned num_components;
733
734 /**
735 * The output variable in the producer stage.
736 */
737 ir_variable *producer_var;
738
739 /**
740 * The input variable in the consumer stage.
741 */
742 ir_variable *consumer_var;
743
744 /**
745 * The location which has been assigned for this varying. This is
746 * expressed in multiples of a float, with the first generic varying
747 * (i.e. the one referred to by VARYING_SLOT_VAR0) represented by the
748 * value 0.
749 */
750 unsigned generic_location;
751 } *matches;
752
753 /**
754 * The number of elements in the \c matches array that are currently in
755 * use.
756 */
757 unsigned num_matches;
758
759 /**
760 * The number of elements that were set aside for the \c matches array when
761 * it was allocated.
762 */
763 unsigned matches_capacity;
764
765 const bool consumer_is_fs;
766 };
767
768 } /* anonymous namespace */
769
770 varying_matches::varying_matches(bool disable_varying_packing,
771 bool consumer_is_fs)
772 : disable_varying_packing(disable_varying_packing),
773 consumer_is_fs(consumer_is_fs)
774 {
775 /* Note: this initial capacity is rather arbitrarily chosen to be large
776 * enough for many cases without wasting an unreasonable amount of space.
777 * varying_matches::record() will resize the array if there are more than
778 * this number of varyings.
779 */
780 this->matches_capacity = 8;
781 this->matches = (match *)
782 malloc(sizeof(*this->matches) * this->matches_capacity);
783 this->num_matches = 0;
784 }
785
786
787 varying_matches::~varying_matches()
788 {
789 free(this->matches);
790 }
791
792
793 /**
794 * Record the given producer/consumer variable pair in the list of variables
795 * that should later be assigned locations.
796 *
797 * It is permissible for \c consumer_var to be NULL (this happens if a
798 * variable is output by the producer and consumed by transform feedback, but
799 * not consumed by the consumer).
800 *
801 * If \c producer_var has already been paired up with a consumer_var, or
802 * producer_var is part of fixed pipeline functionality (and hence already has
803 * a location assigned), this function has no effect.
804 *
805 * Note: as a side effect this function may change the interpolation type of
806 * \c producer_var, but only when the change couldn't possibly affect
807 * rendering.
808 */
809 void
810 varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var)
811 {
812 assert(producer_var != NULL || consumer_var != NULL);
813
814 if ((producer_var && !producer_var->data.is_unmatched_generic_inout)
815 || (consumer_var && !consumer_var->data.is_unmatched_generic_inout)) {
816 /* Either a location already exists for this variable (since it is part
817 * of fixed functionality), or it has already been recorded as part of a
818 * previous match.
819 */
820 return;
821 }
822
823 if ((consumer_var == NULL && producer_var->type->contains_integer()) ||
824 !consumer_is_fs) {
825 /* Since this varying is not being consumed by the fragment shader, its
826 * interpolation type varying cannot possibly affect rendering. Also,
827 * this variable is non-flat and is (or contains) an integer.
828 *
829 * lower_packed_varyings requires all integer varyings to flat,
830 * regardless of where they appear. We can trivially satisfy that
831 * requirement by changing the interpolation type to flat here.
832 */
833 producer_var->data.centroid = false;
834 producer_var->data.sample = false;
835 producer_var->data.interpolation = INTERP_QUALIFIER_FLAT;
836
837 if (consumer_var) {
838 consumer_var->data.centroid = false;
839 consumer_var->data.sample = false;
840 consumer_var->data.interpolation = INTERP_QUALIFIER_FLAT;
841 }
842 }
843
844 if (this->num_matches == this->matches_capacity) {
845 this->matches_capacity *= 2;
846 this->matches = (match *)
847 realloc(this->matches,
848 sizeof(*this->matches) * this->matches_capacity);
849 }
850
851 const ir_variable *const var = (producer_var != NULL)
852 ? producer_var : consumer_var;
853
854 this->matches[this->num_matches].packing_class
855 = this->compute_packing_class(var);
856 this->matches[this->num_matches].packing_order
857 = this->compute_packing_order(var);
858 if (this->disable_varying_packing) {
859 unsigned slots = var->type->is_array()
860 ? (var->type->length * var->type->fields.array->matrix_columns)
861 : var->type->matrix_columns;
862 this->matches[this->num_matches].num_components = 4 * slots;
863 } else {
864 this->matches[this->num_matches].num_components
865 = var->type->component_slots();
866 }
867 this->matches[this->num_matches].producer_var = producer_var;
868 this->matches[this->num_matches].consumer_var = consumer_var;
869 this->num_matches++;
870 if (producer_var)
871 producer_var->data.is_unmatched_generic_inout = 0;
872 if (consumer_var)
873 consumer_var->data.is_unmatched_generic_inout = 0;
874 }
875
876
877 /**
878 * Choose locations for all of the variable matches that were previously
879 * passed to varying_matches::record().
880 */
881 unsigned
882 varying_matches::assign_locations()
883 {
884 /* Sort varying matches into an order that makes them easy to pack. */
885 qsort(this->matches, this->num_matches, sizeof(*this->matches),
886 &varying_matches::match_comparator);
887
888 unsigned generic_location = 0;
889
890 for (unsigned i = 0; i < this->num_matches; i++) {
891 /* Advance to the next slot if this varying has a different packing
892 * class than the previous one, and we're not already on a slot
893 * boundary.
894 */
895 if (i > 0 &&
896 this->matches[i - 1].packing_class
897 != this->matches[i].packing_class) {
898 generic_location = ALIGN(generic_location, 4);
899 }
900
901 this->matches[i].generic_location = generic_location;
902
903 generic_location += this->matches[i].num_components;
904 }
905
906 return (generic_location + 3) / 4;
907 }
908
909
910 /**
911 * Update the producer and consumer shaders to reflect the locations
912 * assignments that were made by varying_matches::assign_locations().
913 */
914 void
915 varying_matches::store_locations() const
916 {
917 for (unsigned i = 0; i < this->num_matches; i++) {
918 ir_variable *producer_var = this->matches[i].producer_var;
919 ir_variable *consumer_var = this->matches[i].consumer_var;
920 unsigned generic_location = this->matches[i].generic_location;
921 unsigned slot = generic_location / 4;
922 unsigned offset = generic_location % 4;
923
924 if (producer_var) {
925 producer_var->data.location = VARYING_SLOT_VAR0 + slot;
926 producer_var->data.location_frac = offset;
927 }
928
929 if (consumer_var) {
930 assert(consumer_var->data.location == -1);
931 consumer_var->data.location = VARYING_SLOT_VAR0 + slot;
932 consumer_var->data.location_frac = offset;
933 }
934 }
935 }
936
937
938 /**
939 * Compute the "packing class" of the given varying. This is an unsigned
940 * integer with the property that two variables in the same packing class can
941 * be safely backed into the same vec4.
942 */
943 unsigned
944 varying_matches::compute_packing_class(const ir_variable *var)
945 {
946 /* Without help from the back-end, there is no way to pack together
947 * variables with different interpolation types, because
948 * lower_packed_varyings must choose exactly one interpolation type for
949 * each packed varying it creates.
950 *
951 * However, we can safely pack together floats, ints, and uints, because:
952 *
953 * - varyings of base type "int" and "uint" must use the "flat"
954 * interpolation type, which can only occur in GLSL 1.30 and above.
955 *
956 * - On platforms that support GLSL 1.30 and above, lower_packed_varyings
957 * can store flat floats as ints without losing any information (using
958 * the ir_unop_bitcast_* opcodes).
959 *
960 * Therefore, the packing class depends only on the interpolation type.
961 */
962 unsigned packing_class = var->data.centroid | (var->data.sample << 1);
963 packing_class *= 4;
964 packing_class += var->data.interpolation;
965 return packing_class;
966 }
967
968
969 /**
970 * Compute the "packing order" of the given varying. This is a sort key we
971 * use to determine when to attempt to pack the given varying relative to
972 * other varyings in the same packing class.
973 */
974 varying_matches::packing_order_enum
975 varying_matches::compute_packing_order(const ir_variable *var)
976 {
977 const glsl_type *element_type = var->type;
978
979 while (element_type->base_type == GLSL_TYPE_ARRAY) {
980 element_type = element_type->fields.array;
981 }
982
983 switch (element_type->component_slots() % 4) {
984 case 1: return PACKING_ORDER_SCALAR;
985 case 2: return PACKING_ORDER_VEC2;
986 case 3: return PACKING_ORDER_VEC3;
987 case 0: return PACKING_ORDER_VEC4;
988 default:
989 assert(!"Unexpected value of vector_elements");
990 return PACKING_ORDER_VEC4;
991 }
992 }
993
994
995 /**
996 * Comparison function passed to qsort() to sort varyings by packing_class and
997 * then by packing_order.
998 */
999 int
1000 varying_matches::match_comparator(const void *x_generic, const void *y_generic)
1001 {
1002 const match *x = (const match *) x_generic;
1003 const match *y = (const match *) y_generic;
1004
1005 if (x->packing_class != y->packing_class)
1006 return x->packing_class - y->packing_class;
1007 return x->packing_order - y->packing_order;
1008 }
1009
1010
1011 /**
1012 * Is the given variable a varying variable to be counted against the
1013 * limit in ctx->Const.MaxVarying?
1014 * This includes variables such as texcoords, colors and generic
1015 * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord.
1016 */
1017 static bool
1018 var_counts_against_varying_limit(gl_shader_stage stage, const ir_variable *var)
1019 {
1020 /* Only fragment shaders will take a varying variable as an input */
1021 if (stage == MESA_SHADER_FRAGMENT &&
1022 var->data.mode == ir_var_shader_in) {
1023 switch (var->data.location) {
1024 case VARYING_SLOT_POS:
1025 case VARYING_SLOT_FACE:
1026 case VARYING_SLOT_PNTC:
1027 return false;
1028 default:
1029 return true;
1030 }
1031 }
1032 return false;
1033 }
1034
1035
1036 /**
1037 * Visitor class that generates tfeedback_candidate structs describing all
1038 * possible targets of transform feedback.
1039 *
1040 * tfeedback_candidate structs are stored in the hash table
1041 * tfeedback_candidates, which is passed to the constructor. This hash table
1042 * maps varying names to instances of the tfeedback_candidate struct.
1043 */
1044 class tfeedback_candidate_generator : public program_resource_visitor
1045 {
1046 public:
1047 tfeedback_candidate_generator(void *mem_ctx,
1048 hash_table *tfeedback_candidates)
1049 : mem_ctx(mem_ctx),
1050 tfeedback_candidates(tfeedback_candidates),
1051 toplevel_var(NULL),
1052 varying_floats(0)
1053 {
1054 }
1055
1056 void process(ir_variable *var)
1057 {
1058 this->toplevel_var = var;
1059 this->varying_floats = 0;
1060 if (var->is_interface_instance())
1061 program_resource_visitor::process(var->get_interface_type(),
1062 var->get_interface_type()->name);
1063 else
1064 program_resource_visitor::process(var);
1065 }
1066
1067 private:
1068 virtual void visit_field(const glsl_type *type, const char *name,
1069 bool row_major)
1070 {
1071 assert(!type->without_array()->is_record());
1072 assert(!type->without_array()->is_interface());
1073
1074 (void) row_major;
1075
1076 tfeedback_candidate *candidate
1077 = rzalloc(this->mem_ctx, tfeedback_candidate);
1078 candidate->toplevel_var = this->toplevel_var;
1079 candidate->type = type;
1080 candidate->offset = this->varying_floats;
1081 hash_table_insert(this->tfeedback_candidates, candidate,
1082 ralloc_strdup(this->mem_ctx, name));
1083 this->varying_floats += type->component_slots();
1084 }
1085
1086 /**
1087 * Memory context used to allocate hash table keys and values.
1088 */
1089 void * const mem_ctx;
1090
1091 /**
1092 * Hash table in which tfeedback_candidate objects should be stored.
1093 */
1094 hash_table * const tfeedback_candidates;
1095
1096 /**
1097 * Pointer to the toplevel variable that is being traversed.
1098 */
1099 ir_variable *toplevel_var;
1100
1101 /**
1102 * Total number of varying floats that have been visited so far. This is
1103 * used to determine the offset to each varying within the toplevel
1104 * variable.
1105 */
1106 unsigned varying_floats;
1107 };
1108
1109
1110 namespace linker {
1111
1112 bool
1113 populate_consumer_input_sets(void *mem_ctx, exec_list *ir,
1114 hash_table *consumer_inputs,
1115 hash_table *consumer_interface_inputs,
1116 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX])
1117 {
1118 memset(consumer_inputs_with_locations,
1119 0,
1120 sizeof(consumer_inputs_with_locations[0]) * VARYING_SLOT_MAX);
1121
1122 foreach_in_list(ir_instruction, node, ir) {
1123 ir_variable *const input_var = node->as_variable();
1124
1125 if ((input_var != NULL) && (input_var->data.mode == ir_var_shader_in)) {
1126 if (input_var->type->is_interface())
1127 return false;
1128
1129 if (input_var->data.explicit_location) {
1130 /* assign_varying_locations only cares about finding the
1131 * ir_variable at the start of a contiguous location block.
1132 *
1133 * - For !producer, consumer_inputs_with_locations isn't used.
1134 *
1135 * - For !consumer, consumer_inputs_with_locations is empty.
1136 *
1137 * For consumer && producer, if you were trying to set some
1138 * ir_variable to the middle of a location block on the other side
1139 * of producer/consumer, cross_validate_outputs_to_inputs() should
1140 * be link-erroring due to either type mismatch or location
1141 * overlaps. If the variables do match up, then they've got a
1142 * matching data.location and you only looked at
1143 * consumer_inputs_with_locations[var->data.location], not any
1144 * following entries for the array/structure.
1145 */
1146 consumer_inputs_with_locations[input_var->data.location] =
1147 input_var;
1148 } else if (input_var->get_interface_type() != NULL) {
1149 char *const iface_field_name =
1150 ralloc_asprintf(mem_ctx, "%s.%s",
1151 input_var->get_interface_type()->name,
1152 input_var->name);
1153 hash_table_insert(consumer_interface_inputs, input_var,
1154 iface_field_name);
1155 } else {
1156 hash_table_insert(consumer_inputs, input_var,
1157 ralloc_strdup(mem_ctx, input_var->name));
1158 }
1159 }
1160 }
1161
1162 return true;
1163 }
1164
1165 /**
1166 * Find a variable from the consumer that "matches" the specified variable
1167 *
1168 * This function only finds inputs with names that match. There is no
1169 * validation (here) that the types, etc. are compatible.
1170 */
1171 ir_variable *
1172 get_matching_input(void *mem_ctx,
1173 const ir_variable *output_var,
1174 hash_table *consumer_inputs,
1175 hash_table *consumer_interface_inputs,
1176 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX])
1177 {
1178 ir_variable *input_var;
1179
1180 if (output_var->data.explicit_location) {
1181 input_var = consumer_inputs_with_locations[output_var->data.location];
1182 } else if (output_var->get_interface_type() != NULL) {
1183 char *const iface_field_name =
1184 ralloc_asprintf(mem_ctx, "%s.%s",
1185 output_var->get_interface_type()->name,
1186 output_var->name);
1187 input_var =
1188 (ir_variable *) hash_table_find(consumer_interface_inputs,
1189 iface_field_name);
1190 } else {
1191 input_var =
1192 (ir_variable *) hash_table_find(consumer_inputs, output_var->name);
1193 }
1194
1195 return (input_var == NULL || input_var->data.mode != ir_var_shader_in)
1196 ? NULL : input_var;
1197 }
1198
1199 }
1200
1201 static int
1202 io_variable_cmp(const void *_a, const void *_b)
1203 {
1204 const ir_variable *const a = *(const ir_variable **) _a;
1205 const ir_variable *const b = *(const ir_variable **) _b;
1206
1207 if (a->data.explicit_location && b->data.explicit_location)
1208 return b->data.location - a->data.location;
1209
1210 if (a->data.explicit_location && !b->data.explicit_location)
1211 return 1;
1212
1213 if (!a->data.explicit_location && b->data.explicit_location)
1214 return -1;
1215
1216 return -strcmp(a->name, b->name);
1217 }
1218
1219 /**
1220 * Sort the shader IO variables into canonical order
1221 */
1222 static void
1223 canonicalize_shader_io(exec_list *ir, enum ir_variable_mode io_mode)
1224 {
1225 ir_variable *var_table[MAX_PROGRAM_OUTPUTS * 4];
1226 unsigned num_variables = 0;
1227
1228 foreach_in_list(ir_instruction, node, ir) {
1229 ir_variable *const var = node->as_variable();
1230
1231 if (var == NULL || var->data.mode != io_mode)
1232 continue;
1233
1234 /* If we have already encountered more I/O variables that could
1235 * successfully link, bail.
1236 */
1237 if (num_variables == ARRAY_SIZE(var_table))
1238 return;
1239
1240 var_table[num_variables++] = var;
1241 }
1242
1243 if (num_variables == 0)
1244 return;
1245
1246 /* Sort the list in reverse order (io_variable_cmp handles this). Later
1247 * we're going to push the variables on to the IR list as a stack, so we
1248 * want the last variable (in canonical order) to be first in the list.
1249 */
1250 qsort(var_table, num_variables, sizeof(var_table[0]), io_variable_cmp);
1251
1252 /* Remove the variable from it's current location in the IR, and put it at
1253 * the front.
1254 */
1255 for (unsigned i = 0; i < num_variables; i++) {
1256 var_table[i]->remove();
1257 ir->push_head(var_table[i]);
1258 }
1259 }
1260
1261 /**
1262 * Assign locations for all variables that are produced in one pipeline stage
1263 * (the "producer") and consumed in the next stage (the "consumer").
1264 *
1265 * Variables produced by the producer may also be consumed by transform
1266 * feedback.
1267 *
1268 * \param num_tfeedback_decls is the number of declarations indicating
1269 * variables that may be consumed by transform feedback.
1270 *
1271 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
1272 * representing the result of parsing the strings passed to
1273 * glTransformFeedbackVaryings(). assign_location() will be called for
1274 * each of these objects that matches one of the outputs of the
1275 * producer.
1276 *
1277 * \param gs_input_vertices: if \c consumer is a geometry shader, this is the
1278 * number of input vertices it accepts. Otherwise zero.
1279 *
1280 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
1281 * be NULL. In this case, varying locations are assigned solely based on the
1282 * requirements of transform feedback.
1283 */
1284 bool
1285 assign_varying_locations(struct gl_context *ctx,
1286 void *mem_ctx,
1287 struct gl_shader_program *prog,
1288 gl_shader *producer, gl_shader *consumer,
1289 unsigned num_tfeedback_decls,
1290 tfeedback_decl *tfeedback_decls,
1291 unsigned gs_input_vertices)
1292 {
1293 varying_matches matches(ctx->Const.DisableVaryingPacking,
1294 consumer && consumer->Stage == MESA_SHADER_FRAGMENT);
1295 hash_table *tfeedback_candidates
1296 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1297 hash_table *consumer_inputs
1298 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1299 hash_table *consumer_interface_inputs
1300 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1301 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX] = {
1302 NULL,
1303 };
1304
1305 /* Operate in a total of four passes.
1306 *
1307 * 1. Sort inputs / outputs into a canonical order. This is necessary so
1308 * that inputs / outputs of separable shaders will be assigned
1309 * predictable locations regardless of the order in which declarations
1310 * appeared in the shader source.
1311 *
1312 * 2. Assign locations for any matching inputs and outputs.
1313 *
1314 * 3. Mark output variables in the producer that do not have locations as
1315 * not being outputs. This lets the optimizer eliminate them.
1316 *
1317 * 4. Mark input variables in the consumer that do not have locations as
1318 * not being inputs. This lets the optimizer eliminate them.
1319 */
1320 if (consumer)
1321 canonicalize_shader_io(consumer->ir, ir_var_shader_in);
1322
1323 if (producer)
1324 canonicalize_shader_io(producer->ir, ir_var_shader_out);
1325
1326 if (consumer
1327 && !linker::populate_consumer_input_sets(mem_ctx,
1328 consumer->ir,
1329 consumer_inputs,
1330 consumer_interface_inputs,
1331 consumer_inputs_with_locations)) {
1332 assert(!"populate_consumer_input_sets failed");
1333 hash_table_dtor(tfeedback_candidates);
1334 hash_table_dtor(consumer_inputs);
1335 hash_table_dtor(consumer_interface_inputs);
1336 return false;
1337 }
1338
1339 if (producer) {
1340 foreach_in_list(ir_instruction, node, producer->ir) {
1341 ir_variable *const output_var = node->as_variable();
1342
1343 if ((output_var == NULL) ||
1344 (output_var->data.mode != ir_var_shader_out))
1345 continue;
1346
1347 /* Only geometry shaders can use non-zero streams */
1348 assert(output_var->data.stream == 0 ||
1349 (output_var->data.stream < MAX_VERTEX_STREAMS &&
1350 producer->Stage == MESA_SHADER_GEOMETRY));
1351
1352 tfeedback_candidate_generator g(mem_ctx, tfeedback_candidates);
1353 g.process(output_var);
1354
1355 ir_variable *const input_var =
1356 linker::get_matching_input(mem_ctx, output_var, consumer_inputs,
1357 consumer_interface_inputs,
1358 consumer_inputs_with_locations);
1359
1360 /* If a matching input variable was found, add this ouptut (and the
1361 * input) to the set. If this is a separable program and there is no
1362 * consumer stage, add the output.
1363 */
1364 if (input_var || (prog->SeparateShader && consumer == NULL)) {
1365 matches.record(output_var, input_var);
1366 }
1367
1368 /* Only stream 0 outputs can be consumed in the next stage */
1369 if (input_var && output_var->data.stream != 0) {
1370 linker_error(prog, "output %s is assigned to stream=%d but "
1371 "is linked to an input, which requires stream=0",
1372 output_var->name, output_var->data.stream);
1373 return false;
1374 }
1375 }
1376 } else {
1377 /* If there's no producer stage, then this must be a separable program.
1378 * For example, we may have a program that has just a fragment shader.
1379 * Later this program will be used with some arbitrary vertex (or
1380 * geometry) shader program. This means that locations must be assigned
1381 * for all the inputs.
1382 */
1383 foreach_in_list(ir_instruction, node, consumer->ir) {
1384 ir_variable *const input_var = node->as_variable();
1385
1386 if ((input_var == NULL) ||
1387 (input_var->data.mode != ir_var_shader_in))
1388 continue;
1389
1390 matches.record(NULL, input_var);
1391 }
1392 }
1393
1394 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
1395 if (!tfeedback_decls[i].is_varying())
1396 continue;
1397
1398 const tfeedback_candidate *matched_candidate
1399 = tfeedback_decls[i].find_candidate(prog, tfeedback_candidates);
1400
1401 if (matched_candidate == NULL) {
1402 hash_table_dtor(tfeedback_candidates);
1403 hash_table_dtor(consumer_inputs);
1404 hash_table_dtor(consumer_interface_inputs);
1405 return false;
1406 }
1407
1408 if (matched_candidate->toplevel_var->data.is_unmatched_generic_inout)
1409 matches.record(matched_candidate->toplevel_var, NULL);
1410 }
1411
1412 const unsigned slots_used = matches.assign_locations();
1413 matches.store_locations();
1414
1415 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
1416 if (!tfeedback_decls[i].is_varying())
1417 continue;
1418
1419 if (!tfeedback_decls[i].assign_location(ctx, prog)) {
1420 hash_table_dtor(tfeedback_candidates);
1421 hash_table_dtor(consumer_inputs);
1422 hash_table_dtor(consumer_interface_inputs);
1423 return false;
1424 }
1425 }
1426
1427 hash_table_dtor(tfeedback_candidates);
1428 hash_table_dtor(consumer_inputs);
1429 hash_table_dtor(consumer_interface_inputs);
1430
1431 if (ctx->Const.DisableVaryingPacking) {
1432 /* Transform feedback code assumes varyings are packed, so if the driver
1433 * has disabled varying packing, make sure it does not support transform
1434 * feedback.
1435 */
1436 assert(!ctx->Extensions.EXT_transform_feedback);
1437 } else {
1438 if (producer) {
1439 lower_packed_varyings(mem_ctx, slots_used, ir_var_shader_out,
1440 0, producer);
1441 }
1442 if (consumer) {
1443 lower_packed_varyings(mem_ctx, slots_used, ir_var_shader_in,
1444 gs_input_vertices, consumer);
1445 }
1446 }
1447
1448 if (consumer && producer) {
1449 foreach_in_list(ir_instruction, node, consumer->ir) {
1450 ir_variable *const var = node->as_variable();
1451
1452 if (var && var->data.mode == ir_var_shader_in &&
1453 var->data.is_unmatched_generic_inout) {
1454 if (prog->Version <= 120) {
1455 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1456 *
1457 * Only those varying variables used (i.e. read) in
1458 * the fragment shader executable must be written to
1459 * by the vertex shader executable; declaring
1460 * superfluous varying variables in a vertex shader is
1461 * permissible.
1462 *
1463 * We interpret this text as meaning that the VS must
1464 * write the variable for the FS to read it. See
1465 * "glsl1-varying read but not written" in piglit.
1466 */
1467
1468 linker_error(prog, "%s shader varying %s not written "
1469 "by %s shader\n.",
1470 _mesa_shader_stage_to_string(consumer->Stage),
1471 var->name,
1472 _mesa_shader_stage_to_string(producer->Stage));
1473 }
1474
1475 /* An 'in' variable is only really a shader input if its
1476 * value is written by the previous stage.
1477 */
1478 var->data.mode = ir_var_auto;
1479 }
1480 }
1481 }
1482
1483 return true;
1484 }
1485
1486 bool
1487 check_against_output_limit(struct gl_context *ctx,
1488 struct gl_shader_program *prog,
1489 gl_shader *producer)
1490 {
1491 unsigned output_vectors = 0;
1492
1493 foreach_in_list(ir_instruction, node, producer->ir) {
1494 ir_variable *const var = node->as_variable();
1495
1496 if (var && var->data.mode == ir_var_shader_out &&
1497 var_counts_against_varying_limit(producer->Stage, var)) {
1498 output_vectors += var->type->count_attribute_slots();
1499 }
1500 }
1501
1502 assert(producer->Stage != MESA_SHADER_FRAGMENT);
1503 unsigned max_output_components =
1504 ctx->Const.Program[producer->Stage].MaxOutputComponents;
1505
1506 const unsigned output_components = output_vectors * 4;
1507 if (output_components > max_output_components) {
1508 if (ctx->API == API_OPENGLES2 || prog->IsES)
1509 linker_error(prog, "shader uses too many output vectors "
1510 "(%u > %u)\n",
1511 output_vectors,
1512 max_output_components / 4);
1513 else
1514 linker_error(prog, "shader uses too many output components "
1515 "(%u > %u)\n",
1516 output_components,
1517 max_output_components);
1518
1519 return false;
1520 }
1521
1522 return true;
1523 }
1524
1525 bool
1526 check_against_input_limit(struct gl_context *ctx,
1527 struct gl_shader_program *prog,
1528 gl_shader *consumer)
1529 {
1530 unsigned input_vectors = 0;
1531
1532 foreach_in_list(ir_instruction, node, consumer->ir) {
1533 ir_variable *const var = node->as_variable();
1534
1535 if (var && var->data.mode == ir_var_shader_in &&
1536 var_counts_against_varying_limit(consumer->Stage, var)) {
1537 input_vectors += var->type->count_attribute_slots();
1538 }
1539 }
1540
1541 assert(consumer->Stage != MESA_SHADER_VERTEX);
1542 unsigned max_input_components =
1543 ctx->Const.Program[consumer->Stage].MaxInputComponents;
1544
1545 const unsigned input_components = input_vectors * 4;
1546 if (input_components > max_input_components) {
1547 if (ctx->API == API_OPENGLES2 || prog->IsES)
1548 linker_error(prog, "shader uses too many input vectors "
1549 "(%u > %u)\n",
1550 input_vectors,
1551 max_input_components / 4);
1552 else
1553 linker_error(prog, "shader uses too many input components "
1554 "(%u > %u)\n",
1555 input_components,
1556 max_input_components);
1557
1558 return false;
1559 }
1560
1561 return true;
1562 }