nir: Delete all traces of nir_op_flog
[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 (!prog->IsES && 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 } else {
267 /* Check for input vars with unmatched output vars in prev stage
268 * taking into account that interface blocks could have a matching
269 * output but with different name, so we ignore them.
270 */
271 assert(!input->data.assigned);
272 if (input->data.used && !input->get_interface_type() &&
273 !input->data.explicit_location && !prog->SeparateShader)
274 linker_error(prog,
275 "%s shader input `%s' "
276 "has no matching output in the previous stage\n",
277 _mesa_shader_stage_to_string(consumer->Stage),
278 input->name);
279 }
280 }
281 }
282 }
283
284
285 /**
286 * Initialize this object based on a string that was passed to
287 * glTransformFeedbackVaryings.
288 *
289 * If the input is mal-formed, this call still succeeds, but it sets
290 * this->var_name to a mal-formed input, so tfeedback_decl::find_output_var()
291 * will fail to find any matching variable.
292 */
293 void
294 tfeedback_decl::init(struct gl_context *ctx, const void *mem_ctx,
295 const char *input)
296 {
297 /* We don't have to be pedantic about what is a valid GLSL variable name,
298 * because any variable with an invalid name can't exist in the IR anyway.
299 */
300
301 this->location = -1;
302 this->orig_name = input;
303 this->is_clip_distance_mesa = false;
304 this->skip_components = 0;
305 this->next_buffer_separator = false;
306 this->matched_candidate = NULL;
307 this->stream_id = 0;
308
309 if (ctx->Extensions.ARB_transform_feedback3) {
310 /* Parse gl_NextBuffer. */
311 if (strcmp(input, "gl_NextBuffer") == 0) {
312 this->next_buffer_separator = true;
313 return;
314 }
315
316 /* Parse gl_SkipComponents. */
317 if (strcmp(input, "gl_SkipComponents1") == 0)
318 this->skip_components = 1;
319 else if (strcmp(input, "gl_SkipComponents2") == 0)
320 this->skip_components = 2;
321 else if (strcmp(input, "gl_SkipComponents3") == 0)
322 this->skip_components = 3;
323 else if (strcmp(input, "gl_SkipComponents4") == 0)
324 this->skip_components = 4;
325
326 if (this->skip_components)
327 return;
328 }
329
330 /* Parse a declaration. */
331 const char *base_name_end;
332 long subscript = parse_program_resource_name(input, &base_name_end);
333 this->var_name = ralloc_strndup(mem_ctx, input, base_name_end - input);
334 if (this->var_name == NULL) {
335 _mesa_error_no_memory(__func__);
336 return;
337 }
338
339 if (subscript >= 0) {
340 this->array_subscript = subscript;
341 this->is_subscripted = true;
342 } else {
343 this->is_subscripted = false;
344 }
345
346 /* For drivers that lower gl_ClipDistance to gl_ClipDistanceMESA, this
347 * class must behave specially to account for the fact that gl_ClipDistance
348 * is converted from a float[8] to a vec4[2].
349 */
350 if (ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].LowerClipDistance &&
351 strcmp(this->var_name, "gl_ClipDistance") == 0) {
352 this->is_clip_distance_mesa = true;
353 }
354 }
355
356
357 /**
358 * Determine whether two tfeedback_decl objects refer to the same variable and
359 * array index (if applicable).
360 */
361 bool
362 tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y)
363 {
364 assert(x.is_varying() && y.is_varying());
365
366 if (strcmp(x.var_name, y.var_name) != 0)
367 return false;
368 if (x.is_subscripted != y.is_subscripted)
369 return false;
370 if (x.is_subscripted && x.array_subscript != y.array_subscript)
371 return false;
372 return true;
373 }
374
375
376 /**
377 * Assign a location and stream ID for this tfeedback_decl object based on the
378 * transform feedback candidate found by find_candidate.
379 *
380 * If an error occurs, the error is reported through linker_error() and false
381 * is returned.
382 */
383 bool
384 tfeedback_decl::assign_location(struct gl_context *ctx,
385 struct gl_shader_program *prog)
386 {
387 assert(this->is_varying());
388
389 unsigned fine_location
390 = this->matched_candidate->toplevel_var->data.location * 4
391 + this->matched_candidate->toplevel_var->data.location_frac
392 + this->matched_candidate->offset;
393
394 if (this->matched_candidate->type->is_array()) {
395 /* Array variable */
396 const unsigned matrix_cols =
397 this->matched_candidate->type->fields.array->matrix_columns;
398 const unsigned vector_elements =
399 this->matched_candidate->type->fields.array->vector_elements;
400 unsigned actual_array_size = this->is_clip_distance_mesa ?
401 prog->LastClipDistanceArraySize :
402 this->matched_candidate->type->array_size();
403
404 if (this->is_subscripted) {
405 /* Check array bounds. */
406 if (this->array_subscript >= actual_array_size) {
407 linker_error(prog, "Transform feedback varying %s has index "
408 "%i, but the array size is %u.",
409 this->orig_name, this->array_subscript,
410 actual_array_size);
411 return false;
412 }
413 unsigned array_elem_size = this->is_clip_distance_mesa ?
414 1 : vector_elements * matrix_cols;
415 fine_location += array_elem_size * this->array_subscript;
416 this->size = 1;
417 } else {
418 this->size = actual_array_size;
419 }
420 this->vector_elements = vector_elements;
421 this->matrix_columns = matrix_cols;
422 if (this->is_clip_distance_mesa)
423 this->type = GL_FLOAT;
424 else
425 this->type = this->matched_candidate->type->fields.array->gl_type;
426 } else {
427 /* Regular variable (scalar, vector, or matrix) */
428 if (this->is_subscripted) {
429 linker_error(prog, "Transform feedback varying %s requested, "
430 "but %s is not an array.",
431 this->orig_name, this->var_name);
432 return false;
433 }
434 this->size = 1;
435 this->vector_elements = this->matched_candidate->type->vector_elements;
436 this->matrix_columns = this->matched_candidate->type->matrix_columns;
437 this->type = this->matched_candidate->type->gl_type;
438 }
439 this->location = fine_location / 4;
440 this->location_frac = fine_location % 4;
441
442 /* From GL_EXT_transform_feedback:
443 * A program will fail to link if:
444 *
445 * * the total number of components to capture in any varying
446 * variable in <varyings> is greater than the constant
447 * MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS_EXT and the
448 * buffer mode is SEPARATE_ATTRIBS_EXT;
449 */
450 if (prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS &&
451 this->num_components() >
452 ctx->Const.MaxTransformFeedbackSeparateComponents) {
453 linker_error(prog, "Transform feedback varying %s exceeds "
454 "MAX_TRANSFORM_FEEDBACK_SEPARATE_COMPONENTS.",
455 this->orig_name);
456 return false;
457 }
458
459 /* Only transform feedback varyings can be assigned to non-zero streams,
460 * so assign the stream id here.
461 */
462 this->stream_id = this->matched_candidate->toplevel_var->data.stream;
463
464 return true;
465 }
466
467
468 unsigned
469 tfeedback_decl::get_num_outputs() const
470 {
471 if (!this->is_varying()) {
472 return 0;
473 }
474
475 return (this->num_components() + this->location_frac + 3)/4;
476 }
477
478
479 /**
480 * Update gl_transform_feedback_info to reflect this tfeedback_decl.
481 *
482 * If an error occurs, the error is reported through linker_error() and false
483 * is returned.
484 */
485 bool
486 tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog,
487 struct gl_transform_feedback_info *info,
488 unsigned buffer, const unsigned max_outputs) const
489 {
490 assert(!this->next_buffer_separator);
491
492 /* Handle gl_SkipComponents. */
493 if (this->skip_components) {
494 info->BufferStride[buffer] += this->skip_components;
495 return true;
496 }
497
498 /* From GL_EXT_transform_feedback:
499 * A program will fail to link if:
500 *
501 * * the total number of components to capture is greater than
502 * the constant MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS_EXT
503 * and the buffer mode is INTERLEAVED_ATTRIBS_EXT.
504 */
505 if (prog->TransformFeedback.BufferMode == GL_INTERLEAVED_ATTRIBS &&
506 info->BufferStride[buffer] + this->num_components() >
507 ctx->Const.MaxTransformFeedbackInterleavedComponents) {
508 linker_error(prog, "The MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS "
509 "limit has been exceeded.");
510 return false;
511 }
512
513 unsigned location = this->location;
514 unsigned location_frac = this->location_frac;
515 unsigned num_components = this->num_components();
516 while (num_components > 0) {
517 unsigned output_size = MIN2(num_components, 4 - location_frac);
518 assert(info->NumOutputs < max_outputs);
519 info->Outputs[info->NumOutputs].ComponentOffset = location_frac;
520 info->Outputs[info->NumOutputs].OutputRegister = location;
521 info->Outputs[info->NumOutputs].NumComponents = output_size;
522 info->Outputs[info->NumOutputs].StreamId = stream_id;
523 info->Outputs[info->NumOutputs].OutputBuffer = buffer;
524 info->Outputs[info->NumOutputs].DstOffset = info->BufferStride[buffer];
525 ++info->NumOutputs;
526 info->BufferStride[buffer] += output_size;
527 num_components -= output_size;
528 location++;
529 location_frac = 0;
530 }
531
532 info->Varyings[info->NumVarying].Name = ralloc_strdup(prog, this->orig_name);
533 info->Varyings[info->NumVarying].Type = this->type;
534 info->Varyings[info->NumVarying].Size = this->size;
535 info->NumVarying++;
536
537 return true;
538 }
539
540
541 const tfeedback_candidate *
542 tfeedback_decl::find_candidate(gl_shader_program *prog,
543 hash_table *tfeedback_candidates)
544 {
545 const char *name = this->is_clip_distance_mesa
546 ? "gl_ClipDistanceMESA" : this->var_name;
547 this->matched_candidate = (const tfeedback_candidate *)
548 hash_table_find(tfeedback_candidates, name);
549 if (!this->matched_candidate) {
550 /* From GL_EXT_transform_feedback:
551 * A program will fail to link if:
552 *
553 * * any variable name specified in the <varyings> array is not
554 * declared as an output in the geometry shader (if present) or
555 * the vertex shader (if no geometry shader is present);
556 */
557 linker_error(prog, "Transform feedback varying %s undeclared.",
558 this->orig_name);
559 }
560 return this->matched_candidate;
561 }
562
563
564 /**
565 * Parse all the transform feedback declarations that were passed to
566 * glTransformFeedbackVaryings() and store them in tfeedback_decl objects.
567 *
568 * If an error occurs, the error is reported through linker_error() and false
569 * is returned.
570 */
571 bool
572 parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog,
573 const void *mem_ctx, unsigned num_names,
574 char **varying_names, tfeedback_decl *decls)
575 {
576 for (unsigned i = 0; i < num_names; ++i) {
577 decls[i].init(ctx, mem_ctx, varying_names[i]);
578
579 if (!decls[i].is_varying())
580 continue;
581
582 /* From GL_EXT_transform_feedback:
583 * A program will fail to link if:
584 *
585 * * any two entries in the <varyings> array specify the same varying
586 * variable;
587 *
588 * We interpret this to mean "any two entries in the <varyings> array
589 * specify the same varying variable and array index", since transform
590 * feedback of arrays would be useless otherwise.
591 */
592 for (unsigned j = 0; j < i; ++j) {
593 if (!decls[j].is_varying())
594 continue;
595
596 if (tfeedback_decl::is_same(decls[i], decls[j])) {
597 linker_error(prog, "Transform feedback varying %s specified "
598 "more than once.", varying_names[i]);
599 return false;
600 }
601 }
602 }
603 return true;
604 }
605
606
607 /**
608 * Store transform feedback location assignments into
609 * prog->LinkedTransformFeedback based on the data stored in tfeedback_decls.
610 *
611 * If an error occurs, the error is reported through linker_error() and false
612 * is returned.
613 */
614 bool
615 store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog,
616 unsigned num_tfeedback_decls,
617 tfeedback_decl *tfeedback_decls)
618 {
619 bool separate_attribs_mode =
620 prog->TransformFeedback.BufferMode == GL_SEPARATE_ATTRIBS;
621
622 ralloc_free(prog->LinkedTransformFeedback.Varyings);
623 ralloc_free(prog->LinkedTransformFeedback.Outputs);
624
625 memset(&prog->LinkedTransformFeedback, 0,
626 sizeof(prog->LinkedTransformFeedback));
627
628 prog->LinkedTransformFeedback.Varyings =
629 rzalloc_array(prog,
630 struct gl_transform_feedback_varying_info,
631 num_tfeedback_decls);
632
633 unsigned num_outputs = 0;
634 for (unsigned i = 0; i < num_tfeedback_decls; ++i)
635 num_outputs += tfeedback_decls[i].get_num_outputs();
636
637 prog->LinkedTransformFeedback.Outputs =
638 rzalloc_array(prog,
639 struct gl_transform_feedback_output,
640 num_outputs);
641
642 unsigned num_buffers = 0;
643
644 if (separate_attribs_mode) {
645 /* GL_SEPARATE_ATTRIBS */
646 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
647 if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback,
648 num_buffers, num_outputs))
649 return false;
650
651 num_buffers++;
652 }
653 }
654 else {
655 /* GL_INVERLEAVED_ATTRIBS */
656 int buffer_stream_id = -1;
657 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
658 if (tfeedback_decls[i].is_next_buffer_separator()) {
659 num_buffers++;
660 buffer_stream_id = -1;
661 continue;
662 } else if (buffer_stream_id == -1) {
663 /* First varying writing to this buffer: remember its stream */
664 buffer_stream_id = (int) tfeedback_decls[i].get_stream_id();
665 } else if (buffer_stream_id !=
666 (int) tfeedback_decls[i].get_stream_id()) {
667 /* Varying writes to the same buffer from a different stream */
668 linker_error(prog,
669 "Transform feedback can't capture varyings belonging "
670 "to different vertex streams in a single buffer. "
671 "Varying %s writes to buffer from stream %u, other "
672 "varyings in the same buffer write from stream %u.",
673 tfeedback_decls[i].name(),
674 tfeedback_decls[i].get_stream_id(),
675 buffer_stream_id);
676 return false;
677 }
678
679 if (!tfeedback_decls[i].store(ctx, prog,
680 &prog->LinkedTransformFeedback,
681 num_buffers, num_outputs))
682 return false;
683 }
684 num_buffers++;
685 }
686
687 assert(prog->LinkedTransformFeedback.NumOutputs == num_outputs);
688
689 prog->LinkedTransformFeedback.NumBuffers = num_buffers;
690 return true;
691 }
692
693 namespace {
694
695 /**
696 * Data structure recording the relationship between outputs of one shader
697 * stage (the "producer") and inputs of another (the "consumer").
698 */
699 class varying_matches
700 {
701 public:
702 varying_matches(bool disable_varying_packing, bool consumer_is_fs);
703 ~varying_matches();
704 void record(ir_variable *producer_var, ir_variable *consumer_var);
705 unsigned assign_locations();
706 void store_locations() const;
707
708 private:
709 /**
710 * If true, this driver disables varying packing, so all varyings need to
711 * be aligned on slot boundaries, and take up a number of slots equal to
712 * their number of matrix columns times their array size.
713 */
714 const bool disable_varying_packing;
715
716 /**
717 * Enum representing the order in which varyings are packed within a
718 * packing class.
719 *
720 * Currently we pack vec4's first, then vec2's, then scalar values, then
721 * vec3's. This order ensures that the only vectors that are at risk of
722 * having to be "double parked" (split between two adjacent varying slots)
723 * are the vec3's.
724 */
725 enum packing_order_enum {
726 PACKING_ORDER_VEC4,
727 PACKING_ORDER_VEC2,
728 PACKING_ORDER_SCALAR,
729 PACKING_ORDER_VEC3,
730 };
731
732 static unsigned compute_packing_class(const ir_variable *var);
733 static packing_order_enum compute_packing_order(const ir_variable *var);
734 static int match_comparator(const void *x_generic, const void *y_generic);
735
736 /**
737 * Structure recording the relationship between a single producer output
738 * and a single consumer input.
739 */
740 struct match {
741 /**
742 * Packing class for this varying, computed by compute_packing_class().
743 */
744 unsigned packing_class;
745
746 /**
747 * Packing order for this varying, computed by compute_packing_order().
748 */
749 packing_order_enum packing_order;
750 unsigned num_components;
751
752 /**
753 * The output variable in the producer stage.
754 */
755 ir_variable *producer_var;
756
757 /**
758 * The input variable in the consumer stage.
759 */
760 ir_variable *consumer_var;
761
762 /**
763 * The location which has been assigned for this varying. This is
764 * expressed in multiples of a float, with the first generic varying
765 * (i.e. the one referred to by VARYING_SLOT_VAR0) represented by the
766 * value 0.
767 */
768 unsigned generic_location;
769 } *matches;
770
771 /**
772 * The number of elements in the \c matches array that are currently in
773 * use.
774 */
775 unsigned num_matches;
776
777 /**
778 * The number of elements that were set aside for the \c matches array when
779 * it was allocated.
780 */
781 unsigned matches_capacity;
782
783 const bool consumer_is_fs;
784 };
785
786 } /* anonymous namespace */
787
788 varying_matches::varying_matches(bool disable_varying_packing,
789 bool consumer_is_fs)
790 : disable_varying_packing(disable_varying_packing),
791 consumer_is_fs(consumer_is_fs)
792 {
793 /* Note: this initial capacity is rather arbitrarily chosen to be large
794 * enough for many cases without wasting an unreasonable amount of space.
795 * varying_matches::record() will resize the array if there are more than
796 * this number of varyings.
797 */
798 this->matches_capacity = 8;
799 this->matches = (match *)
800 malloc(sizeof(*this->matches) * this->matches_capacity);
801 this->num_matches = 0;
802 }
803
804
805 varying_matches::~varying_matches()
806 {
807 free(this->matches);
808 }
809
810
811 /**
812 * Record the given producer/consumer variable pair in the list of variables
813 * that should later be assigned locations.
814 *
815 * It is permissible for \c consumer_var to be NULL (this happens if a
816 * variable is output by the producer and consumed by transform feedback, but
817 * not consumed by the consumer).
818 *
819 * If \c producer_var has already been paired up with a consumer_var, or
820 * producer_var is part of fixed pipeline functionality (and hence already has
821 * a location assigned), this function has no effect.
822 *
823 * Note: as a side effect this function may change the interpolation type of
824 * \c producer_var, but only when the change couldn't possibly affect
825 * rendering.
826 */
827 void
828 varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var)
829 {
830 assert(producer_var != NULL || consumer_var != NULL);
831
832 if ((producer_var && !producer_var->data.is_unmatched_generic_inout)
833 || (consumer_var && !consumer_var->data.is_unmatched_generic_inout)) {
834 /* Either a location already exists for this variable (since it is part
835 * of fixed functionality), or it has already been recorded as part of a
836 * previous match.
837 */
838 return;
839 }
840
841 if ((consumer_var == NULL && producer_var->type->contains_integer()) ||
842 !consumer_is_fs) {
843 /* Since this varying is not being consumed by the fragment shader, its
844 * interpolation type varying cannot possibly affect rendering. Also,
845 * this variable is non-flat and is (or contains) an integer.
846 *
847 * lower_packed_varyings requires all integer varyings to flat,
848 * regardless of where they appear. We can trivially satisfy that
849 * requirement by changing the interpolation type to flat here.
850 */
851 if (producer_var) {
852 producer_var->data.centroid = false;
853 producer_var->data.sample = false;
854 producer_var->data.interpolation = INTERP_QUALIFIER_FLAT;
855 }
856
857 if (consumer_var) {
858 consumer_var->data.centroid = false;
859 consumer_var->data.sample = false;
860 consumer_var->data.interpolation = INTERP_QUALIFIER_FLAT;
861 }
862 }
863
864 if (this->num_matches == this->matches_capacity) {
865 this->matches_capacity *= 2;
866 this->matches = (match *)
867 realloc(this->matches,
868 sizeof(*this->matches) * this->matches_capacity);
869 }
870
871 const ir_variable *const var = (producer_var != NULL)
872 ? producer_var : consumer_var;
873
874 this->matches[this->num_matches].packing_class
875 = this->compute_packing_class(var);
876 this->matches[this->num_matches].packing_order
877 = this->compute_packing_order(var);
878 if (this->disable_varying_packing) {
879 unsigned slots = var->type->is_array()
880 ? (var->type->length * var->type->fields.array->matrix_columns)
881 : var->type->matrix_columns;
882 this->matches[this->num_matches].num_components = 4 * slots;
883 } else {
884 this->matches[this->num_matches].num_components
885 = var->type->component_slots();
886 }
887 this->matches[this->num_matches].producer_var = producer_var;
888 this->matches[this->num_matches].consumer_var = consumer_var;
889 this->num_matches++;
890 if (producer_var)
891 producer_var->data.is_unmatched_generic_inout = 0;
892 if (consumer_var)
893 consumer_var->data.is_unmatched_generic_inout = 0;
894 }
895
896
897 /**
898 * Choose locations for all of the variable matches that were previously
899 * passed to varying_matches::record().
900 */
901 unsigned
902 varying_matches::assign_locations()
903 {
904 /* Sort varying matches into an order that makes them easy to pack. */
905 qsort(this->matches, this->num_matches, sizeof(*this->matches),
906 &varying_matches::match_comparator);
907
908 unsigned generic_location = 0;
909
910 for (unsigned i = 0; i < this->num_matches; i++) {
911 /* Advance to the next slot if this varying has a different packing
912 * class than the previous one, and we're not already on a slot
913 * boundary.
914 */
915 if (i > 0 &&
916 this->matches[i - 1].packing_class
917 != this->matches[i].packing_class) {
918 generic_location = ALIGN(generic_location, 4);
919 }
920
921 this->matches[i].generic_location = generic_location;
922
923 generic_location += this->matches[i].num_components;
924 }
925
926 return (generic_location + 3) / 4;
927 }
928
929
930 /**
931 * Update the producer and consumer shaders to reflect the locations
932 * assignments that were made by varying_matches::assign_locations().
933 */
934 void
935 varying_matches::store_locations() const
936 {
937 for (unsigned i = 0; i < this->num_matches; i++) {
938 ir_variable *producer_var = this->matches[i].producer_var;
939 ir_variable *consumer_var = this->matches[i].consumer_var;
940 unsigned generic_location = this->matches[i].generic_location;
941 unsigned slot = generic_location / 4;
942 unsigned offset = generic_location % 4;
943
944 if (producer_var) {
945 producer_var->data.location = VARYING_SLOT_VAR0 + slot;
946 producer_var->data.location_frac = offset;
947 }
948
949 if (consumer_var) {
950 assert(consumer_var->data.location == -1);
951 consumer_var->data.location = VARYING_SLOT_VAR0 + slot;
952 consumer_var->data.location_frac = offset;
953 }
954 }
955 }
956
957
958 /**
959 * Compute the "packing class" of the given varying. This is an unsigned
960 * integer with the property that two variables in the same packing class can
961 * be safely backed into the same vec4.
962 */
963 unsigned
964 varying_matches::compute_packing_class(const ir_variable *var)
965 {
966 /* Without help from the back-end, there is no way to pack together
967 * variables with different interpolation types, because
968 * lower_packed_varyings must choose exactly one interpolation type for
969 * each packed varying it creates.
970 *
971 * However, we can safely pack together floats, ints, and uints, because:
972 *
973 * - varyings of base type "int" and "uint" must use the "flat"
974 * interpolation type, which can only occur in GLSL 1.30 and above.
975 *
976 * - On platforms that support GLSL 1.30 and above, lower_packed_varyings
977 * can store flat floats as ints without losing any information (using
978 * the ir_unop_bitcast_* opcodes).
979 *
980 * Therefore, the packing class depends only on the interpolation type.
981 */
982 unsigned packing_class = var->data.centroid | (var->data.sample << 1);
983 packing_class *= 4;
984 packing_class += var->data.interpolation;
985 return packing_class;
986 }
987
988
989 /**
990 * Compute the "packing order" of the given varying. This is a sort key we
991 * use to determine when to attempt to pack the given varying relative to
992 * other varyings in the same packing class.
993 */
994 varying_matches::packing_order_enum
995 varying_matches::compute_packing_order(const ir_variable *var)
996 {
997 const glsl_type *element_type = var->type;
998
999 while (element_type->base_type == GLSL_TYPE_ARRAY) {
1000 element_type = element_type->fields.array;
1001 }
1002
1003 switch (element_type->component_slots() % 4) {
1004 case 1: return PACKING_ORDER_SCALAR;
1005 case 2: return PACKING_ORDER_VEC2;
1006 case 3: return PACKING_ORDER_VEC3;
1007 case 0: return PACKING_ORDER_VEC4;
1008 default:
1009 assert(!"Unexpected value of vector_elements");
1010 return PACKING_ORDER_VEC4;
1011 }
1012 }
1013
1014
1015 /**
1016 * Comparison function passed to qsort() to sort varyings by packing_class and
1017 * then by packing_order.
1018 */
1019 int
1020 varying_matches::match_comparator(const void *x_generic, const void *y_generic)
1021 {
1022 const match *x = (const match *) x_generic;
1023 const match *y = (const match *) y_generic;
1024
1025 if (x->packing_class != y->packing_class)
1026 return x->packing_class - y->packing_class;
1027 return x->packing_order - y->packing_order;
1028 }
1029
1030
1031 /**
1032 * Is the given variable a varying variable to be counted against the
1033 * limit in ctx->Const.MaxVarying?
1034 * This includes variables such as texcoords, colors and generic
1035 * varyings, but excludes variables such as gl_FrontFacing and gl_FragCoord.
1036 */
1037 static bool
1038 var_counts_against_varying_limit(gl_shader_stage stage, const ir_variable *var)
1039 {
1040 /* Only fragment shaders will take a varying variable as an input */
1041 if (stage == MESA_SHADER_FRAGMENT &&
1042 var->data.mode == ir_var_shader_in) {
1043 switch (var->data.location) {
1044 case VARYING_SLOT_POS:
1045 case VARYING_SLOT_FACE:
1046 case VARYING_SLOT_PNTC:
1047 return false;
1048 default:
1049 return true;
1050 }
1051 }
1052 return false;
1053 }
1054
1055
1056 /**
1057 * Visitor class that generates tfeedback_candidate structs describing all
1058 * possible targets of transform feedback.
1059 *
1060 * tfeedback_candidate structs are stored in the hash table
1061 * tfeedback_candidates, which is passed to the constructor. This hash table
1062 * maps varying names to instances of the tfeedback_candidate struct.
1063 */
1064 class tfeedback_candidate_generator : public program_resource_visitor
1065 {
1066 public:
1067 tfeedback_candidate_generator(void *mem_ctx,
1068 hash_table *tfeedback_candidates)
1069 : mem_ctx(mem_ctx),
1070 tfeedback_candidates(tfeedback_candidates),
1071 toplevel_var(NULL),
1072 varying_floats(0)
1073 {
1074 }
1075
1076 void process(ir_variable *var)
1077 {
1078 this->toplevel_var = var;
1079 this->varying_floats = 0;
1080 if (var->is_interface_instance())
1081 program_resource_visitor::process(var->get_interface_type(),
1082 var->get_interface_type()->name);
1083 else
1084 program_resource_visitor::process(var);
1085 }
1086
1087 private:
1088 virtual void visit_field(const glsl_type *type, const char *name,
1089 bool row_major)
1090 {
1091 assert(!type->without_array()->is_record());
1092 assert(!type->without_array()->is_interface());
1093
1094 (void) row_major;
1095
1096 tfeedback_candidate *candidate
1097 = rzalloc(this->mem_ctx, tfeedback_candidate);
1098 candidate->toplevel_var = this->toplevel_var;
1099 candidate->type = type;
1100 candidate->offset = this->varying_floats;
1101 hash_table_insert(this->tfeedback_candidates, candidate,
1102 ralloc_strdup(this->mem_ctx, name));
1103 this->varying_floats += type->component_slots();
1104 }
1105
1106 /**
1107 * Memory context used to allocate hash table keys and values.
1108 */
1109 void * const mem_ctx;
1110
1111 /**
1112 * Hash table in which tfeedback_candidate objects should be stored.
1113 */
1114 hash_table * const tfeedback_candidates;
1115
1116 /**
1117 * Pointer to the toplevel variable that is being traversed.
1118 */
1119 ir_variable *toplevel_var;
1120
1121 /**
1122 * Total number of varying floats that have been visited so far. This is
1123 * used to determine the offset to each varying within the toplevel
1124 * variable.
1125 */
1126 unsigned varying_floats;
1127 };
1128
1129
1130 namespace linker {
1131
1132 bool
1133 populate_consumer_input_sets(void *mem_ctx, exec_list *ir,
1134 hash_table *consumer_inputs,
1135 hash_table *consumer_interface_inputs,
1136 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX])
1137 {
1138 memset(consumer_inputs_with_locations,
1139 0,
1140 sizeof(consumer_inputs_with_locations[0]) * VARYING_SLOT_MAX);
1141
1142 foreach_in_list(ir_instruction, node, ir) {
1143 ir_variable *const input_var = node->as_variable();
1144
1145 if ((input_var != NULL) && (input_var->data.mode == ir_var_shader_in)) {
1146 if (input_var->type->is_interface())
1147 return false;
1148
1149 if (input_var->data.explicit_location) {
1150 /* assign_varying_locations only cares about finding the
1151 * ir_variable at the start of a contiguous location block.
1152 *
1153 * - For !producer, consumer_inputs_with_locations isn't used.
1154 *
1155 * - For !consumer, consumer_inputs_with_locations is empty.
1156 *
1157 * For consumer && producer, if you were trying to set some
1158 * ir_variable to the middle of a location block on the other side
1159 * of producer/consumer, cross_validate_outputs_to_inputs() should
1160 * be link-erroring due to either type mismatch or location
1161 * overlaps. If the variables do match up, then they've got a
1162 * matching data.location and you only looked at
1163 * consumer_inputs_with_locations[var->data.location], not any
1164 * following entries for the array/structure.
1165 */
1166 consumer_inputs_with_locations[input_var->data.location] =
1167 input_var;
1168 } else if (input_var->get_interface_type() != NULL) {
1169 char *const iface_field_name =
1170 ralloc_asprintf(mem_ctx, "%s.%s",
1171 input_var->get_interface_type()->name,
1172 input_var->name);
1173 hash_table_insert(consumer_interface_inputs, input_var,
1174 iface_field_name);
1175 } else {
1176 hash_table_insert(consumer_inputs, input_var,
1177 ralloc_strdup(mem_ctx, input_var->name));
1178 }
1179 }
1180 }
1181
1182 return true;
1183 }
1184
1185 /**
1186 * Find a variable from the consumer that "matches" the specified variable
1187 *
1188 * This function only finds inputs with names that match. There is no
1189 * validation (here) that the types, etc. are compatible.
1190 */
1191 ir_variable *
1192 get_matching_input(void *mem_ctx,
1193 const ir_variable *output_var,
1194 hash_table *consumer_inputs,
1195 hash_table *consumer_interface_inputs,
1196 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX])
1197 {
1198 ir_variable *input_var;
1199
1200 if (output_var->data.explicit_location) {
1201 input_var = consumer_inputs_with_locations[output_var->data.location];
1202 } else if (output_var->get_interface_type() != NULL) {
1203 char *const iface_field_name =
1204 ralloc_asprintf(mem_ctx, "%s.%s",
1205 output_var->get_interface_type()->name,
1206 output_var->name);
1207 input_var =
1208 (ir_variable *) hash_table_find(consumer_interface_inputs,
1209 iface_field_name);
1210 } else {
1211 input_var =
1212 (ir_variable *) hash_table_find(consumer_inputs, output_var->name);
1213 }
1214
1215 return (input_var == NULL || input_var->data.mode != ir_var_shader_in)
1216 ? NULL : input_var;
1217 }
1218
1219 }
1220
1221 static int
1222 io_variable_cmp(const void *_a, const void *_b)
1223 {
1224 const ir_variable *const a = *(const ir_variable **) _a;
1225 const ir_variable *const b = *(const ir_variable **) _b;
1226
1227 if (a->data.explicit_location && b->data.explicit_location)
1228 return b->data.location - a->data.location;
1229
1230 if (a->data.explicit_location && !b->data.explicit_location)
1231 return 1;
1232
1233 if (!a->data.explicit_location && b->data.explicit_location)
1234 return -1;
1235
1236 return -strcmp(a->name, b->name);
1237 }
1238
1239 /**
1240 * Sort the shader IO variables into canonical order
1241 */
1242 static void
1243 canonicalize_shader_io(exec_list *ir, enum ir_variable_mode io_mode)
1244 {
1245 ir_variable *var_table[MAX_PROGRAM_OUTPUTS * 4];
1246 unsigned num_variables = 0;
1247
1248 foreach_in_list(ir_instruction, node, ir) {
1249 ir_variable *const var = node->as_variable();
1250
1251 if (var == NULL || var->data.mode != io_mode)
1252 continue;
1253
1254 /* If we have already encountered more I/O variables that could
1255 * successfully link, bail.
1256 */
1257 if (num_variables == ARRAY_SIZE(var_table))
1258 return;
1259
1260 var_table[num_variables++] = var;
1261 }
1262
1263 if (num_variables == 0)
1264 return;
1265
1266 /* Sort the list in reverse order (io_variable_cmp handles this). Later
1267 * we're going to push the variables on to the IR list as a stack, so we
1268 * want the last variable (in canonical order) to be first in the list.
1269 */
1270 qsort(var_table, num_variables, sizeof(var_table[0]), io_variable_cmp);
1271
1272 /* Remove the variable from it's current location in the IR, and put it at
1273 * the front.
1274 */
1275 for (unsigned i = 0; i < num_variables; i++) {
1276 var_table[i]->remove();
1277 ir->push_head(var_table[i]);
1278 }
1279 }
1280
1281 /**
1282 * Assign locations for all variables that are produced in one pipeline stage
1283 * (the "producer") and consumed in the next stage (the "consumer").
1284 *
1285 * Variables produced by the producer may also be consumed by transform
1286 * feedback.
1287 *
1288 * \param num_tfeedback_decls is the number of declarations indicating
1289 * variables that may be consumed by transform feedback.
1290 *
1291 * \param tfeedback_decls is a pointer to an array of tfeedback_decl objects
1292 * representing the result of parsing the strings passed to
1293 * glTransformFeedbackVaryings(). assign_location() will be called for
1294 * each of these objects that matches one of the outputs of the
1295 * producer.
1296 *
1297 * \param gs_input_vertices: if \c consumer is a geometry shader, this is the
1298 * number of input vertices it accepts. Otherwise zero.
1299 *
1300 * When num_tfeedback_decls is nonzero, it is permissible for the consumer to
1301 * be NULL. In this case, varying locations are assigned solely based on the
1302 * requirements of transform feedback.
1303 */
1304 bool
1305 assign_varying_locations(struct gl_context *ctx,
1306 void *mem_ctx,
1307 struct gl_shader_program *prog,
1308 gl_shader *producer, gl_shader *consumer,
1309 unsigned num_tfeedback_decls,
1310 tfeedback_decl *tfeedback_decls,
1311 unsigned gs_input_vertices)
1312 {
1313 varying_matches matches(ctx->Const.DisableVaryingPacking,
1314 consumer && consumer->Stage == MESA_SHADER_FRAGMENT);
1315 hash_table *tfeedback_candidates
1316 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1317 hash_table *consumer_inputs
1318 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1319 hash_table *consumer_interface_inputs
1320 = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
1321 ir_variable *consumer_inputs_with_locations[VARYING_SLOT_MAX] = {
1322 NULL,
1323 };
1324
1325 /* Operate in a total of four passes.
1326 *
1327 * 1. Sort inputs / outputs into a canonical order. This is necessary so
1328 * that inputs / outputs of separable shaders will be assigned
1329 * predictable locations regardless of the order in which declarations
1330 * appeared in the shader source.
1331 *
1332 * 2. Assign locations for any matching inputs and outputs.
1333 *
1334 * 3. Mark output variables in the producer that do not have locations as
1335 * not being outputs. This lets the optimizer eliminate them.
1336 *
1337 * 4. Mark input variables in the consumer that do not have locations as
1338 * not being inputs. This lets the optimizer eliminate them.
1339 */
1340 if (consumer)
1341 canonicalize_shader_io(consumer->ir, ir_var_shader_in);
1342
1343 if (producer)
1344 canonicalize_shader_io(producer->ir, ir_var_shader_out);
1345
1346 if (consumer
1347 && !linker::populate_consumer_input_sets(mem_ctx,
1348 consumer->ir,
1349 consumer_inputs,
1350 consumer_interface_inputs,
1351 consumer_inputs_with_locations)) {
1352 assert(!"populate_consumer_input_sets failed");
1353 hash_table_dtor(tfeedback_candidates);
1354 hash_table_dtor(consumer_inputs);
1355 hash_table_dtor(consumer_interface_inputs);
1356 return false;
1357 }
1358
1359 if (producer) {
1360 foreach_in_list(ir_instruction, node, producer->ir) {
1361 ir_variable *const output_var = node->as_variable();
1362
1363 if ((output_var == NULL) ||
1364 (output_var->data.mode != ir_var_shader_out))
1365 continue;
1366
1367 /* Only geometry shaders can use non-zero streams */
1368 assert(output_var->data.stream == 0 ||
1369 (output_var->data.stream < MAX_VERTEX_STREAMS &&
1370 producer->Stage == MESA_SHADER_GEOMETRY));
1371
1372 tfeedback_candidate_generator g(mem_ctx, tfeedback_candidates);
1373 g.process(output_var);
1374
1375 ir_variable *const input_var =
1376 linker::get_matching_input(mem_ctx, output_var, consumer_inputs,
1377 consumer_interface_inputs,
1378 consumer_inputs_with_locations);
1379
1380 /* If a matching input variable was found, add this ouptut (and the
1381 * input) to the set. If this is a separable program and there is no
1382 * consumer stage, add the output.
1383 */
1384 if (input_var || (prog->SeparateShader && consumer == NULL)) {
1385 matches.record(output_var, input_var);
1386 }
1387
1388 /* Only stream 0 outputs can be consumed in the next stage */
1389 if (input_var && output_var->data.stream != 0) {
1390 linker_error(prog, "output %s is assigned to stream=%d but "
1391 "is linked to an input, which requires stream=0",
1392 output_var->name, output_var->data.stream);
1393 return false;
1394 }
1395 }
1396 } else {
1397 /* If there's no producer stage, then this must be a separable program.
1398 * For example, we may have a program that has just a fragment shader.
1399 * Later this program will be used with some arbitrary vertex (or
1400 * geometry) shader program. This means that locations must be assigned
1401 * for all the inputs.
1402 */
1403 foreach_in_list(ir_instruction, node, consumer->ir) {
1404 ir_variable *const input_var = node->as_variable();
1405
1406 if ((input_var == NULL) ||
1407 (input_var->data.mode != ir_var_shader_in))
1408 continue;
1409
1410 matches.record(NULL, input_var);
1411 }
1412 }
1413
1414 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
1415 if (!tfeedback_decls[i].is_varying())
1416 continue;
1417
1418 const tfeedback_candidate *matched_candidate
1419 = tfeedback_decls[i].find_candidate(prog, tfeedback_candidates);
1420
1421 if (matched_candidate == NULL) {
1422 hash_table_dtor(tfeedback_candidates);
1423 hash_table_dtor(consumer_inputs);
1424 hash_table_dtor(consumer_interface_inputs);
1425 return false;
1426 }
1427
1428 if (matched_candidate->toplevel_var->data.is_unmatched_generic_inout)
1429 matches.record(matched_candidate->toplevel_var, NULL);
1430 }
1431
1432 const unsigned slots_used = matches.assign_locations();
1433 matches.store_locations();
1434
1435 for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
1436 if (!tfeedback_decls[i].is_varying())
1437 continue;
1438
1439 if (!tfeedback_decls[i].assign_location(ctx, prog)) {
1440 hash_table_dtor(tfeedback_candidates);
1441 hash_table_dtor(consumer_inputs);
1442 hash_table_dtor(consumer_interface_inputs);
1443 return false;
1444 }
1445 }
1446
1447 hash_table_dtor(tfeedback_candidates);
1448 hash_table_dtor(consumer_inputs);
1449 hash_table_dtor(consumer_interface_inputs);
1450
1451 if (ctx->Const.DisableVaryingPacking) {
1452 /* Transform feedback code assumes varyings are packed, so if the driver
1453 * has disabled varying packing, make sure it does not support transform
1454 * feedback.
1455 */
1456 assert(!ctx->Extensions.EXT_transform_feedback);
1457 } else {
1458 if (producer) {
1459 lower_packed_varyings(mem_ctx, slots_used, ir_var_shader_out,
1460 0, producer);
1461 }
1462 if (consumer) {
1463 lower_packed_varyings(mem_ctx, slots_used, ir_var_shader_in,
1464 gs_input_vertices, consumer);
1465 }
1466 }
1467
1468 if (consumer && producer) {
1469 foreach_in_list(ir_instruction, node, consumer->ir) {
1470 ir_variable *const var = node->as_variable();
1471
1472 if (var && var->data.mode == ir_var_shader_in &&
1473 var->data.is_unmatched_generic_inout) {
1474 if (prog->IsES) {
1475 /*
1476 * On Page 91 (Page 97 of the PDF) of the GLSL ES 1.0 spec:
1477 *
1478 * If the vertex shader declares but doesn't write to a
1479 * varying and the fragment shader declares and reads it,
1480 * is this an error?
1481 *
1482 * RESOLUTION: No.
1483 */
1484 linker_warning(prog, "%s shader varying %s not written "
1485 "by %s shader\n.",
1486 _mesa_shader_stage_to_string(consumer->Stage),
1487 var->name,
1488 _mesa_shader_stage_to_string(producer->Stage));
1489 } else if (prog->Version <= 120) {
1490 /* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
1491 *
1492 * Only those varying variables used (i.e. read) in
1493 * the fragment shader executable must be written to
1494 * by the vertex shader executable; declaring
1495 * superfluous varying variables in a vertex shader is
1496 * permissible.
1497 *
1498 * We interpret this text as meaning that the VS must
1499 * write the variable for the FS to read it. See
1500 * "glsl1-varying read but not written" in piglit.
1501 */
1502 linker_error(prog, "%s shader varying %s not written "
1503 "by %s shader\n.",
1504 _mesa_shader_stage_to_string(consumer->Stage),
1505 var->name,
1506 _mesa_shader_stage_to_string(producer->Stage));
1507 }
1508
1509 /* An 'in' variable is only really a shader input if its
1510 * value is written by the previous stage.
1511 */
1512 var->data.mode = ir_var_auto;
1513 }
1514 }
1515 }
1516
1517 return true;
1518 }
1519
1520 bool
1521 check_against_output_limit(struct gl_context *ctx,
1522 struct gl_shader_program *prog,
1523 gl_shader *producer)
1524 {
1525 unsigned output_vectors = 0;
1526
1527 foreach_in_list(ir_instruction, node, producer->ir) {
1528 ir_variable *const var = node->as_variable();
1529
1530 if (var && var->data.mode == ir_var_shader_out &&
1531 var_counts_against_varying_limit(producer->Stage, var)) {
1532 output_vectors += var->type->count_attribute_slots();
1533 }
1534 }
1535
1536 assert(producer->Stage != MESA_SHADER_FRAGMENT);
1537 unsigned max_output_components =
1538 ctx->Const.Program[producer->Stage].MaxOutputComponents;
1539
1540 const unsigned output_components = output_vectors * 4;
1541 if (output_components > max_output_components) {
1542 if (ctx->API == API_OPENGLES2 || prog->IsES)
1543 linker_error(prog, "shader uses too many output vectors "
1544 "(%u > %u)\n",
1545 output_vectors,
1546 max_output_components / 4);
1547 else
1548 linker_error(prog, "shader uses too many output components "
1549 "(%u > %u)\n",
1550 output_components,
1551 max_output_components);
1552
1553 return false;
1554 }
1555
1556 return true;
1557 }
1558
1559 bool
1560 check_against_input_limit(struct gl_context *ctx,
1561 struct gl_shader_program *prog,
1562 gl_shader *consumer)
1563 {
1564 unsigned input_vectors = 0;
1565
1566 foreach_in_list(ir_instruction, node, consumer->ir) {
1567 ir_variable *const var = node->as_variable();
1568
1569 if (var && var->data.mode == ir_var_shader_in &&
1570 var_counts_against_varying_limit(consumer->Stage, var)) {
1571 input_vectors += var->type->count_attribute_slots();
1572 }
1573 }
1574
1575 assert(consumer->Stage != MESA_SHADER_VERTEX);
1576 unsigned max_input_components =
1577 ctx->Const.Program[consumer->Stage].MaxInputComponents;
1578
1579 const unsigned input_components = input_vectors * 4;
1580 if (input_components > max_input_components) {
1581 if (ctx->API == API_OPENGLES2 || prog->IsES)
1582 linker_error(prog, "shader uses too many input vectors "
1583 "(%u > %u)\n",
1584 input_vectors,
1585 max_input_components / 4);
1586 else
1587 linker_error(prog, "shader uses too many input components "
1588 "(%u > %u)\n",
1589 input_components,
1590 max_input_components);
1591
1592 return false;
1593 }
1594
1595 return true;
1596 }