2 * Copyright © 2011 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
25 * \file lower_distance.cpp
27 * This pass accounts for the difference between the way
28 * gl_ClipDistance is declared in standard GLSL (as an array of
29 * floats), and the way it is frequently implemented in hardware (as
30 * a pair of vec4s, with four clip distances packed into each).
32 * The declaration of gl_ClipDistance is replaced with a declaration
33 * of gl_ClipDistanceMESA, and any references to gl_ClipDistance are
34 * translated to refer to gl_ClipDistanceMESA with the appropriate
35 * swizzling of array indices. For instance:
41 * gl_ClipDistanceMESA[i>>2][i&3]
43 * Since some hardware may not internally represent gl_ClipDistance as a pair
44 * of vec4's, this lowering pass is optional. To enable it, set the
45 * LowerCombinedClipCullDistance flag in gl_shader_compiler_options to true.
48 #include "main/macros.h"
49 #include "glsl_symbol_table.h"
50 #include "ir_rvalue_visitor.h"
52 #include "program/prog_instruction.h" /* For WRITEMASK_* */
54 #define GLSL_CLIP_VAR_NAME "gl_ClipDistanceMESA"
58 class lower_distance_visitor
: public ir_rvalue_visitor
{
60 explicit lower_distance_visitor(gl_shader_stage shader_stage
,
61 const char *in_name
, int total_size
,
63 : progress(false), old_distance_out_var(NULL
),
64 old_distance_in_var(NULL
), new_distance_out_var(NULL
),
65 new_distance_in_var(NULL
), shader_stage(shader_stage
),
66 in_name(in_name
), total_size(total_size
), offset(offset
)
70 explicit lower_distance_visitor(gl_shader_stage shader_stage
,
72 const lower_distance_visitor
*orig
,
75 old_distance_out_var(NULL
),
76 old_distance_in_var(NULL
),
77 new_distance_out_var(orig
->new_distance_out_var
),
78 new_distance_in_var(orig
->new_distance_in_var
),
79 shader_stage(shader_stage
),
81 total_size(orig
->total_size
),
86 virtual ir_visitor_status
visit(ir_variable
*);
87 void create_indices(ir_rvalue
*, ir_rvalue
*&, ir_rvalue
*&);
88 bool is_distance_vec8(ir_rvalue
*ir
);
89 ir_rvalue
*lower_distance_vec8(ir_rvalue
*ir
);
90 virtual ir_visitor_status
visit_leave(ir_assignment
*);
91 void visit_new_assignment(ir_assignment
*ir
);
92 virtual ir_visitor_status
visit_leave(ir_call
*);
94 virtual void handle_rvalue(ir_rvalue
**rvalue
);
96 void fix_lhs(ir_assignment
*);
101 * Pointer to the declaration of gl_ClipDistance, if found.
105 * - the in_var is for geometry and both tessellation shader inputs only.
107 * - since gl_ClipDistance is available in tessellation control,
108 * tessellation evaluation and geometry shaders as both an input
109 * and an output, it's possible for both old_distance_out_var
110 * and old_distance_in_var to be non-null.
112 ir_variable
*old_distance_out_var
;
113 ir_variable
*old_distance_in_var
;
116 * Pointer to the newly-created gl_ClipDistanceMESA variable.
118 ir_variable
*new_distance_out_var
;
119 ir_variable
*new_distance_in_var
;
122 * Type of shader we are compiling (e.g. MESA_SHADER_VERTEX)
124 const gl_shader_stage shader_stage
;
130 } /* anonymous namespace */
133 * Replace any declaration of 'in_name' as an array of floats with a
134 * declaration of gl_ClipDistanceMESA as an array of vec4's.
137 lower_distance_visitor::visit(ir_variable
*ir
)
139 ir_variable
**old_var
;
140 ir_variable
**new_var
;
142 if (!ir
->name
|| strcmp(ir
->name
, in_name
) != 0)
143 return visit_continue
;
144 assert (ir
->type
->is_array());
146 if (ir
->data
.mode
== ir_var_shader_out
) {
147 if (this->old_distance_out_var
)
148 return visit_continue
;
149 old_var
= &old_distance_out_var
;
150 new_var
= &new_distance_out_var
;
151 } else if (ir
->data
.mode
== ir_var_shader_in
) {
152 if (this->old_distance_in_var
)
153 return visit_continue
;
154 old_var
= &old_distance_in_var
;
155 new_var
= &new_distance_in_var
;
157 unreachable("not reached");
160 this->progress
= true;
165 unsigned new_size
= (total_size
+ 3) / 4;
167 /* Clone the old var so that we inherit all of its properties */
168 *new_var
= ir
->clone(ralloc_parent(ir
), NULL
);
169 (*new_var
)->name
= ralloc_strdup(*new_var
, GLSL_CLIP_VAR_NAME
);
170 (*new_var
)->data
.location
= VARYING_SLOT_CLIP_DIST0
;
172 if (!ir
->type
->fields
.array
->is_array()) {
173 /* gl_ClipDistance (used for vertex, tessellation evaluation and
174 * geometry output, and fragment input).
176 assert((ir
->data
.mode
== ir_var_shader_in
&&
177 this->shader_stage
== MESA_SHADER_FRAGMENT
) ||
178 (ir
->data
.mode
== ir_var_shader_out
&&
179 (this->shader_stage
== MESA_SHADER_VERTEX
||
180 this->shader_stage
== MESA_SHADER_TESS_EVAL
||
181 this->shader_stage
== MESA_SHADER_GEOMETRY
)));
183 assert (ir
->type
->fields
.array
== glsl_type::float_type
);
184 (*new_var
)->data
.max_array_access
= new_size
- 1;
186 /* And change the properties that we need to change */
187 (*new_var
)->type
= glsl_type::get_array_instance(glsl_type::vec4_type
,
190 /* 2D gl_ClipDistance (used for tessellation control, tessellation
191 * evaluation and geometry input, and tessellation control output).
193 assert((ir
->data
.mode
== ir_var_shader_in
&&
194 (this->shader_stage
== MESA_SHADER_GEOMETRY
||
195 this->shader_stage
== MESA_SHADER_TESS_EVAL
)) ||
196 this->shader_stage
== MESA_SHADER_TESS_CTRL
);
198 assert (ir
->type
->fields
.array
->fields
.array
== glsl_type::float_type
);
200 /* And change the properties that we need to change */
201 (*new_var
)->type
= glsl_type::get_array_instance(
202 glsl_type::get_array_instance(glsl_type::vec4_type
,
204 ir
->type
->array_size());
206 ir
->replace_with(*new_var
);
211 return visit_continue
;
216 * Create the necessary GLSL rvalues to index into gl_ClipDistanceMESA based
217 * on the rvalue previously used to index into gl_ClipDistance.
219 * \param array_index Selects one of the vec4's in gl_ClipDistanceMESA
220 * \param swizzle_index Selects a component within the vec4 selected by
224 lower_distance_visitor::create_indices(ir_rvalue
*old_index
,
225 ir_rvalue
*&array_index
,
226 ir_rvalue
*&swizzle_index
)
228 void *ctx
= ralloc_parent(old_index
);
230 /* Make sure old_index is a signed int so that the bitwise "shift" and
231 * "and" operations below type check properly.
233 if (old_index
->type
!= glsl_type::int_type
) {
234 assert (old_index
->type
== glsl_type::uint_type
);
235 old_index
= new(ctx
) ir_expression(ir_unop_u2i
, old_index
);
238 ir_constant
*old_index_constant
=
239 old_index
->constant_expression_value(ctx
);
240 if (old_index_constant
) {
241 /* gl_ClipDistance is being accessed via a constant index. Don't bother
242 * creating expressions to calculate the lowered indices. Just create
245 int const_val
= old_index_constant
->get_int_component(0) + offset
;
246 array_index
= new(ctx
) ir_constant(const_val
/ 4);
247 swizzle_index
= new(ctx
) ir_constant(const_val
% 4);
249 /* Create a variable to hold the value of old_index (so that we
250 * don't compute it twice).
252 ir_variable
*old_index_var
= new(ctx
) ir_variable(
253 glsl_type::int_type
, "distance_index", ir_var_temporary
);
254 this->base_ir
->insert_before(old_index_var
);
255 this->base_ir
->insert_before(new(ctx
) ir_assignment(
256 new(ctx
) ir_dereference_variable(old_index_var
), old_index
));
258 /* Create the expression distance_index / 4. Do this as a bit
259 * shift because that's likely to be more efficient.
261 array_index
= new(ctx
) ir_expression(
263 new(ctx
) ir_expression(ir_binop_add
,
264 new(ctx
) ir_dereference_variable(old_index_var
),
265 new(ctx
) ir_constant(offset
)),
266 new(ctx
) ir_constant(2));
268 /* Create the expression distance_index % 4. Do this as a bitwise
269 * AND because that's likely to be more efficient.
271 swizzle_index
= new(ctx
) ir_expression(
273 new(ctx
) ir_expression(ir_binop_add
,
274 new(ctx
) ir_dereference_variable(old_index_var
),
275 new(ctx
) ir_constant(offset
)),
276 new(ctx
) ir_constant(3));
282 * Determine whether the given rvalue describes an array of 8 floats that
283 * needs to be lowered to an array of 2 vec4's; that is, determine whether it
284 * matches one of the following patterns:
286 * - gl_ClipDistance (if gl_ClipDistance is 1D)
287 * - gl_ClipDistance[i] (if gl_ClipDistance is 2D)
290 lower_distance_visitor::is_distance_vec8(ir_rvalue
*ir
)
292 /* Note that geometry shaders contain gl_ClipDistance both as an input
293 * (which is a 2D array) and an output (which is a 1D array), so it's
294 * possible for both this->old_distance_out_var and
295 * this->old_distance_in_var to be non-NULL in the same shader.
298 if (!ir
->type
->is_array())
300 if (ir
->type
->fields
.array
!= glsl_type::float_type
)
303 if (this->old_distance_out_var
) {
304 if (ir
->variable_referenced() == this->old_distance_out_var
)
307 if (this->old_distance_in_var
) {
308 assert(this->shader_stage
== MESA_SHADER_TESS_CTRL
||
309 this->shader_stage
== MESA_SHADER_TESS_EVAL
||
310 this->shader_stage
== MESA_SHADER_GEOMETRY
||
311 this->shader_stage
== MESA_SHADER_FRAGMENT
);
313 if (ir
->variable_referenced() == this->old_distance_in_var
)
321 * If the given ir satisfies is_distance_vec8(), return new ir
322 * representing its lowered equivalent. That is, map:
324 * - gl_ClipDistance => gl_ClipDistanceMESA (if gl_ClipDistance is 1D)
325 * - gl_ClipDistance[i] => gl_ClipDistanceMESA[i] (if gl_ClipDistance is 2D)
327 * Otherwise return NULL.
330 lower_distance_visitor::lower_distance_vec8(ir_rvalue
*ir
)
332 if (!ir
->type
->is_array())
334 if (ir
->type
->fields
.array
!= glsl_type::float_type
)
337 ir_variable
**new_var
= NULL
;
338 if (this->old_distance_out_var
) {
339 if (ir
->variable_referenced() == this->old_distance_out_var
)
340 new_var
= &this->new_distance_out_var
;
342 if (this->old_distance_in_var
) {
343 if (ir
->variable_referenced() == this->old_distance_in_var
)
344 new_var
= &this->new_distance_in_var
;
349 if (ir
->as_dereference_variable()) {
350 return new(ralloc_parent(ir
)) ir_dereference_variable(*new_var
);
352 ir_dereference_array
*array_ref
= ir
->as_dereference_array();
354 assert(array_ref
->array
->as_dereference_variable());
356 return new(ralloc_parent(ir
))
357 ir_dereference_array(*new_var
, array_ref
->array_index
);
363 lower_distance_visitor::handle_rvalue(ir_rvalue
**rv
)
368 ir_dereference_array
*const array_deref
= (*rv
)->as_dereference_array();
369 if (array_deref
== NULL
)
372 /* Replace any expression that indexes one of the floats in gl_ClipDistance
373 * with an expression that indexes into one of the vec4's in
374 * gl_ClipDistanceMESA and accesses the appropriate component.
376 ir_rvalue
*lowered_vec8
=
377 this->lower_distance_vec8(array_deref
->array
);
378 if (lowered_vec8
!= NULL
) {
379 this->progress
= true;
380 ir_rvalue
*array_index
;
381 ir_rvalue
*swizzle_index
;
382 this->create_indices(array_deref
->array_index
, array_index
, swizzle_index
);
383 void *mem_ctx
= ralloc_parent(array_deref
);
385 ir_dereference_array
*const new_array_deref
=
386 new(mem_ctx
) ir_dereference_array(lowered_vec8
, array_index
);
388 ir_expression
*const expr
=
389 new(mem_ctx
) ir_expression(ir_binop_vector_extract
,
398 lower_distance_visitor::fix_lhs(ir_assignment
*ir
)
400 if (ir
->lhs
->ir_type
== ir_type_expression
) {
401 void *mem_ctx
= ralloc_parent(ir
);
402 ir_expression
*const expr
= (ir_expression
*) ir
->lhs
;
404 /* The expression must be of the form:
406 * (vector_extract gl_ClipDistanceMESA[i], j).
408 assert(expr
->operation
== ir_binop_vector_extract
);
409 assert(expr
->operands
[0]->ir_type
== ir_type_dereference_array
);
410 assert(expr
->operands
[0]->type
== glsl_type::vec4_type
);
412 ir_dereference
*const new_lhs
= (ir_dereference
*) expr
->operands
[0];
413 ir
->rhs
= new(mem_ctx
) ir_expression(ir_triop_vector_insert
,
414 glsl_type::vec4_type
,
415 new_lhs
->clone(mem_ctx
, NULL
),
418 ir
->set_lhs(new_lhs
);
419 ir
->write_mask
= WRITEMASK_XYZW
;
424 * Replace any assignment having the 1D gl_ClipDistance (undereferenced) as
425 * its LHS or RHS with a sequence of assignments, one for each component of
426 * the array. Each of these assignments is lowered to refer to
427 * gl_ClipDistanceMESA as appropriate.
429 * We need to do a similar replacement for 2D gl_ClipDistance, however since
430 * it's an input, the only case we need to address is where a 1D slice of it
431 * is the entire RHS of an assignment, e.g.:
433 * foo = gl_in[i].gl_ClipDistance
436 lower_distance_visitor::visit_leave(ir_assignment
*ir
)
438 /* First invoke the base class visitor. This causes handle_rvalue() to be
439 * called on ir->rhs and ir->condition.
441 ir_rvalue_visitor::visit_leave(ir
);
443 if (this->is_distance_vec8(ir
->lhs
) ||
444 this->is_distance_vec8(ir
->rhs
)) {
445 /* LHS or RHS of the assignment is the entire 1D gl_ClipDistance array
446 * (or a 1D slice of a 2D gl_ClipDistance input array). Since we are
447 * reshaping gl_ClipDistance from an array of floats to an array of
448 * vec4's, this isn't going to work as a bulk assignment anymore, so
449 * unroll it to element-by-element assignments and lower each of them.
451 * Note: to unroll into element-by-element assignments, we need to make
452 * clones of the LHS and RHS. This is safe because expressions and
453 * l-values are side-effect free.
455 void *ctx
= ralloc_parent(ir
);
456 int array_size
= ir
->lhs
->type
->array_size();
457 for (int i
= 0; i
< array_size
; ++i
) {
458 ir_dereference_array
*new_lhs
= new(ctx
) ir_dereference_array(
459 ir
->lhs
->clone(ctx
, NULL
), new(ctx
) ir_constant(i
));
460 ir_dereference_array
*new_rhs
= new(ctx
) ir_dereference_array(
461 ir
->rhs
->clone(ctx
, NULL
), new(ctx
) ir_constant(i
));
462 this->handle_rvalue((ir_rvalue
**) &new_rhs
);
464 /* Handle the LHS after creating the new assignment. This must
465 * happen in this order because handle_rvalue may replace the old LHS
466 * with an ir_expression of ir_binop_vector_extract. Since this is
467 * not a valide l-value, this will cause an assertion in the
468 * ir_assignment constructor to fail.
470 * If this occurs, replace the mangled LHS with a dereference of the
471 * vector, and replace the RHS with an ir_triop_vector_insert.
473 ir_assignment
*const assign
= new(ctx
) ir_assignment(new_lhs
, new_rhs
);
474 this->handle_rvalue((ir_rvalue
**) &assign
->lhs
);
475 this->fix_lhs(assign
);
477 this->base_ir
->insert_before(assign
);
481 return visit_continue
;
484 /* Handle the LHS as if it were an r-value. Normally
485 * rvalue_visit(ir_assignment *) only visits the RHS, but we need to lower
486 * expressions in the LHS as well.
488 * This may cause the LHS to get replaced with an ir_expression of
489 * ir_binop_vector_extract. If this occurs, replace it with a dereference
490 * of the vector, and replace the RHS with an ir_triop_vector_insert.
492 handle_rvalue((ir_rvalue
**)&ir
->lhs
);
495 return rvalue_visit(ir
);
500 * Set up base_ir properly and call visit_leave() on a newly created
501 * ir_assignment node. This is used in cases where we have to insert an
502 * ir_assignment in a place where we know the hierarchical visitor won't see
506 lower_distance_visitor::visit_new_assignment(ir_assignment
*ir
)
508 ir_instruction
*old_base_ir
= this->base_ir
;
511 this->base_ir
= old_base_ir
;
516 * If a 1D gl_ClipDistance variable appears as an argument in an ir_call
517 * expression, replace it with a temporary variable, and make sure the ir_call
518 * is preceded and/or followed by assignments that copy the contents of the
519 * temporary variable to and/or from gl_ClipDistance. Each of these
520 * assignments is then lowered to refer to gl_ClipDistanceMESA.
522 * We need to do a similar replacement for 2D gl_ClipDistance, however since
523 * it's an input, the only case we need to address is where a 1D slice of it
524 * is passed as an "in" parameter to an ir_call, e.g.:
526 * foo(gl_in[i].gl_ClipDistance)
529 lower_distance_visitor::visit_leave(ir_call
*ir
)
531 void *ctx
= ralloc_parent(ir
);
533 const exec_node
*formal_param_node
= ir
->callee
->parameters
.get_head_raw();
534 const exec_node
*actual_param_node
= ir
->actual_parameters
.get_head_raw();
535 while (!actual_param_node
->is_tail_sentinel()) {
536 ir_variable
*formal_param
= (ir_variable
*) formal_param_node
;
537 ir_rvalue
*actual_param
= (ir_rvalue
*) actual_param_node
;
539 /* Advance formal_param_node and actual_param_node now so that we can
540 * safely replace actual_param with another node, if necessary, below.
542 formal_param_node
= formal_param_node
->next
;
543 actual_param_node
= actual_param_node
->next
;
545 if (this->is_distance_vec8(actual_param
)) {
546 /* User is trying to pass the whole 1D gl_ClipDistance array (or a 1D
547 * slice of a 2D gl_ClipDistance array) to a function call. Since we
548 * are reshaping gl_ClipDistance from an array of floats to an array
549 * of vec4's, this isn't going to work anymore, so use a temporary
552 ir_variable
*temp_clip_distance
= new(ctx
) ir_variable(
553 actual_param
->type
, "temp_clip_distance", ir_var_temporary
);
554 this->base_ir
->insert_before(temp_clip_distance
);
555 actual_param
->replace_with(
556 new(ctx
) ir_dereference_variable(temp_clip_distance
));
557 if (formal_param
->data
.mode
== ir_var_function_in
558 || formal_param
->data
.mode
== ir_var_function_inout
) {
559 /* Copy from gl_ClipDistance to the temporary before the call.
560 * Since we are going to insert this copy before the current
561 * instruction, we need to visit it afterwards to make sure it
564 ir_assignment
*new_assignment
= new(ctx
) ir_assignment(
565 new(ctx
) ir_dereference_variable(temp_clip_distance
),
566 actual_param
->clone(ctx
, NULL
));
567 this->base_ir
->insert_before(new_assignment
);
568 this->visit_new_assignment(new_assignment
);
570 if (formal_param
->data
.mode
== ir_var_function_out
571 || formal_param
->data
.mode
== ir_var_function_inout
) {
572 /* Copy from the temporary to gl_ClipDistance after the call.
573 * Since visit_list_elements() has already decided which
574 * instruction it's going to visit next, we need to visit
575 * afterwards to make sure it gets lowered.
577 ir_assignment
*new_assignment
= new(ctx
) ir_assignment(
578 actual_param
->clone(ctx
, NULL
),
579 new(ctx
) ir_dereference_variable(temp_clip_distance
));
580 this->base_ir
->insert_after(new_assignment
);
581 this->visit_new_assignment(new_assignment
);
586 return rvalue_visit(ir
);
590 class lower_distance_visitor_counter
: public ir_rvalue_visitor
{
592 explicit lower_distance_visitor_counter(void)
593 : in_clip_size(0), in_cull_size(0),
594 out_clip_size(0), out_cull_size(0)
598 virtual ir_visitor_status
visit(ir_variable
*);
599 virtual void handle_rvalue(ir_rvalue
**rvalue
);
609 * Count gl_ClipDistance and gl_CullDistance sizes.
612 lower_distance_visitor_counter::visit(ir_variable
*ir
)
614 int *clip_size
, *cull_size
;
617 return visit_continue
;
619 if (ir
->data
.mode
== ir_var_shader_out
) {
620 clip_size
= &out_clip_size
;
621 cull_size
= &out_cull_size
;
622 } else if (ir
->data
.mode
== ir_var_shader_in
) {
623 clip_size
= &in_clip_size
;
624 cull_size
= &in_cull_size
;
626 return visit_continue
;
628 if (ir
->type
->is_unsized_array())
629 return visit_continue
;
631 if (*clip_size
== 0) {
632 if (!strcmp(ir
->name
, "gl_ClipDistance")) {
633 if (!ir
->type
->fields
.array
->is_array())
634 *clip_size
= ir
->type
->array_size();
636 *clip_size
= ir
->type
->fields
.array
->array_size();
640 if (*cull_size
== 0) {
641 if (!strcmp(ir
->name
, "gl_CullDistance")) {
642 if (!ir
->type
->fields
.array
->is_array())
643 *cull_size
= ir
->type
->array_size();
645 *cull_size
= ir
->type
->fields
.array
->array_size();
648 return visit_continue
;
652 lower_distance_visitor_counter::handle_rvalue(ir_rvalue
**)
658 lower_clip_cull_distance(struct gl_shader_program
*prog
,
659 struct gl_linked_shader
*shader
)
661 int clip_size
, cull_size
;
663 lower_distance_visitor_counter count
;
664 visit_list_elements(&count
, shader
->ir
);
666 clip_size
= MAX2(count
.in_clip_size
, count
.out_clip_size
);
667 cull_size
= MAX2(count
.in_cull_size
, count
.out_cull_size
);
669 if (clip_size
== 0 && cull_size
== 0)
672 lower_distance_visitor
v(shader
->Stage
, "gl_ClipDistance", clip_size
+ cull_size
, 0);
673 visit_list_elements(&v
, shader
->ir
);
675 lower_distance_visitor
v2(shader
->Stage
, "gl_CullDistance", &v
, clip_size
);
676 visit_list_elements(&v2
, shader
->ir
);
678 if (v2
.new_distance_out_var
)
679 shader
->symbols
->add_variable(v2
.new_distance_out_var
);
680 if (v2
.new_distance_in_var
)
681 shader
->symbols
->add_variable(v2
.new_distance_in_var
);