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_clip_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 * LowerClipDistance flag in gl_shader_compiler_options to true.
48 #include "glsl_symbol_table.h"
49 #include "ir_rvalue_visitor.h"
51 #include "program/prog_instruction.h" /* For WRITEMASK_* */
53 class lower_clip_distance_visitor
: public ir_rvalue_visitor
{
55 lower_clip_distance_visitor()
56 : progress(false), old_clip_distance_var(NULL
),
57 new_clip_distance_var(NULL
)
61 virtual ir_visitor_status
visit(ir_variable
*);
62 void create_indices(ir_rvalue
*, ir_rvalue
*&, ir_rvalue
*&);
63 virtual ir_visitor_status
visit_leave(ir_assignment
*);
64 void visit_new_assignment(ir_assignment
*ir
);
65 virtual ir_visitor_status
visit_leave(ir_call
*);
67 virtual void handle_rvalue(ir_rvalue
**rvalue
);
69 void fix_lhs(ir_assignment
*);
74 * Pointer to the declaration of gl_ClipDistance, if found.
76 ir_variable
*old_clip_distance_var
;
79 * Pointer to the newly-created gl_ClipDistanceMESA variable.
81 ir_variable
*new_clip_distance_var
;
86 * Replace any declaration of gl_ClipDistance as an array of floats with a
87 * declaration of gl_ClipDistanceMESA as an array of vec4's.
90 lower_clip_distance_visitor::visit(ir_variable
*ir
)
92 /* No point in looking for the declaration of gl_ClipDistance if
93 * we've already found it.
95 if (this->old_clip_distance_var
)
96 return visit_continue
;
98 if (ir
->name
&& strcmp(ir
->name
, "gl_ClipDistance") == 0) {
99 this->progress
= true;
100 this->old_clip_distance_var
= ir
;
101 assert (ir
->type
->is_array());
102 assert (ir
->type
->element_type() == glsl_type::float_type
);
103 unsigned new_size
= (ir
->type
->array_size() + 3) / 4;
105 /* Clone the old var so that we inherit all of its properties */
106 this->new_clip_distance_var
= ir
->clone(ralloc_parent(ir
), NULL
);
108 /* And change the properties that we need to change */
109 this->new_clip_distance_var
->name
110 = ralloc_strdup(this->new_clip_distance_var
, "gl_ClipDistanceMESA");
111 this->new_clip_distance_var
->type
112 = glsl_type::get_array_instance(glsl_type::vec4_type
, new_size
);
113 this->new_clip_distance_var
->max_array_access
= ir
->max_array_access
/ 4;
115 ir
->replace_with(this->new_clip_distance_var
);
117 return visit_continue
;
122 * Create the necessary GLSL rvalues to index into gl_ClipDistanceMESA based
123 * on the rvalue previously used to index into gl_ClipDistance.
125 * \param array_index Selects one of the vec4's in gl_ClipDistanceMESA
126 * \param swizzle_index Selects a component within the vec4 selected by
130 lower_clip_distance_visitor::create_indices(ir_rvalue
*old_index
,
131 ir_rvalue
*&array_index
,
132 ir_rvalue
*&swizzle_index
)
134 void *ctx
= ralloc_parent(old_index
);
136 /* Make sure old_index is a signed int so that the bitwise "shift" and
137 * "and" operations below type check properly.
139 if (old_index
->type
!= glsl_type::int_type
) {
140 assert (old_index
->type
== glsl_type::uint_type
);
141 old_index
= new(ctx
) ir_expression(ir_unop_u2i
, old_index
);
144 ir_constant
*old_index_constant
= old_index
->constant_expression_value();
145 if (old_index_constant
) {
146 /* gl_ClipDistance is being accessed via a constant index. Don't bother
147 * creating expressions to calculate the lowered indices. Just create
150 int const_val
= old_index_constant
->get_int_component(0);
151 array_index
= new(ctx
) ir_constant(const_val
/ 4);
152 swizzle_index
= new(ctx
) ir_constant(const_val
% 4);
154 /* Create a variable to hold the value of old_index (so that we
155 * don't compute it twice).
157 ir_variable
*old_index_var
= new(ctx
) ir_variable(
158 glsl_type::int_type
, "clip_distance_index", ir_var_temporary
);
159 this->base_ir
->insert_before(old_index_var
);
160 this->base_ir
->insert_before(new(ctx
) ir_assignment(
161 new(ctx
) ir_dereference_variable(old_index_var
), old_index
));
163 /* Create the expression clip_distance_index / 4. Do this as a bit
164 * shift because that's likely to be more efficient.
166 array_index
= new(ctx
) ir_expression(
167 ir_binop_rshift
, new(ctx
) ir_dereference_variable(old_index_var
),
168 new(ctx
) ir_constant(2));
170 /* Create the expression clip_distance_index % 4. Do this as a bitwise
171 * AND because that's likely to be more efficient.
173 swizzle_index
= new(ctx
) ir_expression(
174 ir_binop_bit_and
, new(ctx
) ir_dereference_variable(old_index_var
),
175 new(ctx
) ir_constant(3));
181 lower_clip_distance_visitor::handle_rvalue(ir_rvalue
**rv
)
183 /* If the gl_ClipDistance var hasn't been declared yet, then
184 * there's no way this deref can refer to it.
186 if (!this->old_clip_distance_var
|| *rv
== NULL
)
189 ir_dereference_array
*const array_deref
= (*rv
)->as_dereference_array();
190 if (array_deref
== NULL
)
193 /* Replace any expression that indexes into the gl_ClipDistance array
194 * with an expression that indexes into one of the vec4's in
195 * gl_ClipDistanceMESA and accesses the appropriate component.
197 ir_dereference_variable
*old_var_ref
=
198 array_deref
->array
->as_dereference_variable();
199 if (old_var_ref
&& old_var_ref
->var
== this->old_clip_distance_var
) {
200 this->progress
= true;
201 ir_rvalue
*array_index
;
202 ir_rvalue
*swizzle_index
;
203 this->create_indices(array_deref
->array_index
, array_index
, swizzle_index
);
204 void *mem_ctx
= ralloc_parent(array_deref
);
206 ir_dereference_array
*const ClipDistanceMESA_deref
=
207 new(mem_ctx
) ir_dereference_array(this->new_clip_distance_var
,
210 ir_expression
*const expr
=
211 new(mem_ctx
) ir_expression(ir_binop_vector_extract
,
212 ClipDistanceMESA_deref
,
220 lower_clip_distance_visitor::fix_lhs(ir_assignment
*ir
)
222 if (ir
->lhs
->ir_type
== ir_type_expression
) {
223 void *mem_ctx
= ralloc_parent(ir
);
224 ir_expression
*const expr
= (ir_expression
*) ir
->lhs
;
226 /* The expression must be of the form:
228 * (vector_extract gl_ClipDistanceMESA[i], j).
230 assert(expr
->operation
== ir_binop_vector_extract
);
231 assert(expr
->operands
[0]->ir_type
== ir_type_dereference_array
);
232 assert(expr
->operands
[0]->type
== glsl_type::vec4_type
);
234 ir_dereference
*const new_lhs
= (ir_dereference
*) expr
->operands
[0];
235 ir
->rhs
= new(mem_ctx
) ir_expression(ir_triop_vector_insert
,
236 glsl_type::vec4_type
,
237 new_lhs
->clone(mem_ctx
, NULL
),
240 ir
->set_lhs(new_lhs
);
241 ir
->write_mask
= WRITEMASK_XYZW
;
246 * Replace any assignment having gl_ClipDistance (undereferenced) as its LHS
247 * or RHS with a sequence of assignments, one for each component of the array.
248 * Each of these assignments is lowered to refer to gl_ClipDistanceMESA as
252 lower_clip_distance_visitor::visit_leave(ir_assignment
*ir
)
254 ir_dereference_variable
*lhs_var
= ir
->lhs
->as_dereference_variable();
255 ir_dereference_variable
*rhs_var
= ir
->rhs
->as_dereference_variable();
256 if ((lhs_var
&& lhs_var
->var
== this->old_clip_distance_var
)
257 || (rhs_var
&& rhs_var
->var
== this->old_clip_distance_var
)) {
258 /* LHS or RHS of the assignment is the entire gl_ClipDistance array.
259 * Since we are reshaping gl_ClipDistance from an array of floats to an
260 * array of vec4's, this isn't going to work as a bulk assignment
261 * anymore, so unroll it to element-by-element assignments and lower
264 * Note: to unroll into element-by-element assignments, we need to make
265 * clones of the LHS and RHS. This is safe because expressions and
266 * l-values are side-effect free.
268 void *ctx
= ralloc_parent(ir
);
269 int array_size
= this->old_clip_distance_var
->type
->array_size();
270 for (int i
= 0; i
< array_size
; ++i
) {
271 ir_dereference_array
*new_lhs
= new(ctx
) ir_dereference_array(
272 ir
->lhs
->clone(ctx
, NULL
), new(ctx
) ir_constant(i
));
273 ir_dereference_array
*new_rhs
= new(ctx
) ir_dereference_array(
274 ir
->rhs
->clone(ctx
, NULL
), new(ctx
) ir_constant(i
));
275 this->handle_rvalue((ir_rvalue
**) &new_rhs
);
277 /* Handle the LHS after creating the new assignment. This must
278 * happen in this order because handle_rvalue may replace the old LHS
279 * with an ir_expression of ir_binop_vector_extract. Since this is
280 * not a valide l-value, this will cause an assertion in the
281 * ir_assignment constructor to fail.
283 * If this occurs, replace the mangled LHS with a dereference of the
284 * vector, and replace the RHS with an ir_triop_vector_insert.
286 ir_assignment
*const assign
= new(ctx
) ir_assignment(new_lhs
, new_rhs
);
287 this->handle_rvalue((ir_rvalue
**) &assign
->lhs
);
288 this->fix_lhs(assign
);
290 this->base_ir
->insert_before(assign
);
294 return visit_continue
;
297 /* Handle the LHS as if it were an r-value. Normally
298 * rvalue_visit(ir_assignment *) only visits the RHS, but we need to lower
299 * expressions in the LHS as well.
301 * This may cause the LHS to get replaced with an ir_expression of
302 * ir_binop_vector_extract. If this occurs, replace it with a dereference
303 * of the vector, and replace the RHS with an ir_triop_vector_insert.
305 handle_rvalue((ir_rvalue
**)&ir
->lhs
);
308 return rvalue_visit(ir
);
313 * Set up base_ir properly and call visit_leave() on a newly created
314 * ir_assignment node. This is used in cases where we have to insert an
315 * ir_assignment in a place where we know the hierarchical visitor won't see
319 lower_clip_distance_visitor::visit_new_assignment(ir_assignment
*ir
)
321 ir_instruction
*old_base_ir
= this->base_ir
;
324 this->base_ir
= old_base_ir
;
329 * If gl_ClipDistance appears as an argument in an ir_call expression, replace
330 * it with a temporary variable, and make sure the ir_call is preceded and/or
331 * followed by assignments that copy the contents of the temporary variable to
332 * and/or from gl_ClipDistance. Each of these assignments is then lowered to
333 * refer to gl_ClipDistanceMESA.
336 lower_clip_distance_visitor::visit_leave(ir_call
*ir
)
338 void *ctx
= ralloc_parent(ir
);
340 const exec_node
*formal_param_node
= ir
->callee
->parameters
.head
;
341 const exec_node
*actual_param_node
= ir
->actual_parameters
.head
;
342 while (!actual_param_node
->is_tail_sentinel()) {
343 ir_variable
*formal_param
= (ir_variable
*) formal_param_node
;
344 ir_rvalue
*actual_param
= (ir_rvalue
*) actual_param_node
;
346 /* Advance formal_param_node and actual_param_node now so that we can
347 * safely replace actual_param with another node, if necessary, below.
349 formal_param_node
= formal_param_node
->next
;
350 actual_param_node
= actual_param_node
->next
;
352 ir_dereference_variable
*deref
= actual_param
->as_dereference_variable();
353 if (deref
&& deref
->var
== this->old_clip_distance_var
) {
354 /* User is trying to pass the whole gl_ClipDistance array to a
355 * function call. Since we are reshaping gl_ClipDistance from an
356 * array of floats to an array of vec4's, this isn't going to work
357 * anymore, so use a temporary array instead.
359 ir_variable
*temp_clip_distance
= new(ctx
) ir_variable(
360 actual_param
->type
, "temp_clip_distance", ir_var_temporary
);
361 this->base_ir
->insert_before(temp_clip_distance
);
362 actual_param
->replace_with(
363 new(ctx
) ir_dereference_variable(temp_clip_distance
));
364 if (formal_param
->mode
== ir_var_function_in
365 || formal_param
->mode
== ir_var_function_inout
) {
366 /* Copy from gl_ClipDistance to the temporary before the call.
367 * Since we are going to insert this copy before the current
368 * instruction, we need to visit it afterwards to make sure it
371 ir_assignment
*new_assignment
= new(ctx
) ir_assignment(
372 new(ctx
) ir_dereference_variable(temp_clip_distance
),
373 new(ctx
) ir_dereference_variable(old_clip_distance_var
));
374 this->base_ir
->insert_before(new_assignment
);
375 this->visit_new_assignment(new_assignment
);
377 if (formal_param
->mode
== ir_var_function_out
378 || formal_param
->mode
== ir_var_function_inout
) {
379 /* Copy from the temporary to gl_ClipDistance after the call.
380 * Since visit_list_elements() has already decided which
381 * instruction it's going to visit next, we need to visit
382 * afterwards to make sure it gets lowered.
384 ir_assignment
*new_assignment
= new(ctx
) ir_assignment(
385 new(ctx
) ir_dereference_variable(old_clip_distance_var
),
386 new(ctx
) ir_dereference_variable(temp_clip_distance
));
387 this->base_ir
->insert_after(new_assignment
);
388 this->visit_new_assignment(new_assignment
);
393 return rvalue_visit(ir
);
398 lower_clip_distance(gl_shader
*shader
)
400 lower_clip_distance_visitor v
;
402 visit_list_elements(&v
, shader
->ir
);
404 if (v
.new_clip_distance_var
)
405 shader
->symbols
->add_variable(v
.new_clip_distance_var
);