2 * Copyright © 2010 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.
24 #include "glsl_types.h"
25 #include "loop_analysis.h"
26 #include "ir_hierarchical_visitor.h"
28 class loop_unroll_visitor
: public ir_hierarchical_visitor
{
30 loop_unroll_visitor(loop_state
*state
, unsigned max_iterations
)
33 this->progress
= false;
34 this->max_iterations
= max_iterations
;
37 virtual ir_visitor_status
visit_leave(ir_loop
*ir
);
42 unsigned max_iterations
;
47 loop_unroll_visitor::visit_leave(ir_loop
*ir
)
49 loop_variable_state
*const ls
= this->state
->get(ir
);
52 /* If we've entered a loop that hasn't been analyzed, something really,
53 * really bad has happened.
57 return visit_continue
;
60 iterations
= ls
->max_iterations
;
62 /* Don't try to unroll loops where the number of iterations is not known
66 return visit_continue
;
68 /* Don't try to unroll loops that have zillions of iterations either.
70 if (iterations
> (int) max_iterations
)
71 return visit_continue
;
73 if (ls
->num_loop_jumps
> 1)
74 return visit_continue
;
75 else if (ls
->num_loop_jumps
) {
76 /* recognize loops in the form produced by ir_lower_jumps */
77 ir_instruction
*last_ir
=
78 ((ir_instruction
*)ir
->body_instructions
.get_tail());
80 assert(last_ir
!= NULL
);
82 ir_if
*last_if
= last_ir
->as_if();
84 bool continue_from_then_branch
;
86 /* Determine which if-statement branch, if any, ends with a break.
87 * The branch that did *not* have the break will get a temporary
88 * continue inserted in each iteration of the loop unroll.
90 * Note that since ls->num_loop_jumps is <= 1, it is impossible for
91 * both branches to end with a break.
93 ir_instruction
*last
=
94 (ir_instruction
*) last_if
->then_instructions
.get_tail();
96 if (last
&& last
->ir_type
== ir_type_loop_jump
97 && ((ir_loop_jump
*) last
)->is_break()) {
98 continue_from_then_branch
= false;
100 last
= (ir_instruction
*) last_if
->then_instructions
.get_tail();
102 if (last
&& last
->ir_type
== ir_type_loop_jump
103 && ((ir_loop_jump
*) last
)->is_break())
104 continue_from_then_branch
= true;
106 /* Bail out if neither if-statement branch ends with a break.
108 return visit_continue
;
111 /* Remove the break from the if-statement.
115 void *const mem_ctx
= talloc_parent(ir
);
116 ir_instruction
*ir_to_replace
= ir
;
118 for (int i
= 0; i
< iterations
; i
++) {
121 copy_list
.make_empty();
122 clone_ir_list(mem_ctx
, ©_list
, &ir
->body_instructions
);
124 last_if
= ((ir_instruction
*)copy_list
.get_tail())->as_if();
127 ir_to_replace
->insert_before(©_list
);
128 ir_to_replace
->remove();
130 /* placeholder that will be removed in the next iteration */
132 new(mem_ctx
) ir_loop_jump(ir_loop_jump::jump_continue
);
134 exec_list
*const list
= (continue_from_then_branch
)
135 ? &last_if
->then_instructions
: &last_if
->else_instructions
;
137 list
->push_tail(ir_to_replace
);
140 ir_to_replace
->remove();
142 this->progress
= true;
143 return visit_continue
;
144 } else if (last_ir
->ir_type
== ir_type_loop_jump
145 && ((ir_loop_jump
*)last_ir
)->is_break()) {
146 /* If the only loop-jump is a break at the end of the loop, the loop
147 * will execute exactly once. Remove the break, set the iteration
148 * count, and fall through to the normal unroller.
153 this->progress
= true;
155 return visit_continue
;
158 void *const mem_ctx
= talloc_parent(ir
);
160 for (int i
= 0; i
< iterations
; i
++) {
163 copy_list
.make_empty();
164 clone_ir_list(mem_ctx
, ©_list
, &ir
->body_instructions
);
166 ir
->insert_before(©_list
);
169 /* The loop has been replaced by the unrolled copies. Remove the original
170 * loop from the IR sequence.
174 this->progress
= true;
175 return visit_continue
;
180 unroll_loops(exec_list
*instructions
, loop_state
*ls
, unsigned max_iterations
)
182 loop_unroll_visitor
v(ls
, max_iterations
);