1 /*
2 * Copyright © 2015 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 DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "nir.h"
25
26 /**
27 * \file nir_sweep.c
28 *
29 * The nir_sweep() pass performs a mark and sweep pass over a nir_shader's associated
30 * memory - anything still connected to the program will be kept, and any dead memory
31 * we dropped on the floor will be freed.
32 *
33 * The expectation is that drivers should call this when finished compiling the shader
34 * (after any optimization, lowering, and so on). However, it's also fine to call it
35 * earlier, and even many times, trading CPU cycles for memory savings.
36 */
37
38 #define steal_list(mem_ctx, type, list) \
39 foreach_list_typed(type, obj, node, list) { ralloc_steal(mem_ctx, obj); }
40
41 static void sweep_cf_node(nir_shader *nir, nir_cf_node *cf_node);
42
43 static bool
sweep_src_indirect(nir_src * src,void * nir)44 sweep_src_indirect(nir_src *src, void *nir)
45 {
46 if (!src->is_ssa && src->reg.indirect)
47 ralloc_steal(nir, src->reg.indirect);
48
49 return true;
50 }
51
52 static bool
sweep_dest_indirect(nir_dest * dest,void * nir)53 sweep_dest_indirect(nir_dest *dest, void *nir)
54 {
55 if (!dest->is_ssa && dest->reg.indirect)
56 ralloc_steal(nir, dest->reg.indirect);
57
58 return true;
59 }
60
61 static void
sweep_block(nir_shader * nir,nir_block * block)62 sweep_block(nir_shader *nir, nir_block *block)
63 {
64 ralloc_steal(nir, block);
65
66 /* sweep_impl will mark all metadata invalid. We can safely release all of
67 * this here.
68 */
69 ralloc_free(block->live_in);
70 block->live_in = NULL;
71
72 ralloc_free(block->live_out);
73 block->live_out = NULL;
74
75 nir_foreach_instr(instr, block) {
76 ralloc_steal(nir, instr);
77
78 nir_foreach_src(instr, sweep_src_indirect, nir);
79 nir_foreach_dest(instr, sweep_dest_indirect, nir);
80 }
81 }
82
83 static void
sweep_if(nir_shader * nir,nir_if * iff)84 sweep_if(nir_shader *nir, nir_if *iff)
85 {
86 ralloc_steal(nir, iff);
87
88 foreach_list_typed(nir_cf_node, cf_node, node, &iff->then_list) {
89 sweep_cf_node(nir, cf_node);
90 }
91
92 foreach_list_typed(nir_cf_node, cf_node, node, &iff->else_list) {
93 sweep_cf_node(nir, cf_node);
94 }
95 }
96
97 static void
sweep_loop(nir_shader * nir,nir_loop * loop)98 sweep_loop(nir_shader *nir, nir_loop *loop)
99 {
100 ralloc_steal(nir, loop);
101
102 foreach_list_typed(nir_cf_node, cf_node, node, &loop->body) {
103 sweep_cf_node(nir, cf_node);
104 }
105 }
106
107 static void
sweep_cf_node(nir_shader * nir,nir_cf_node * cf_node)108 sweep_cf_node(nir_shader *nir, nir_cf_node *cf_node)
109 {
110 switch (cf_node->type) {
111 case nir_cf_node_block:
112 sweep_block(nir, nir_cf_node_as_block(cf_node));
113 break;
114 case nir_cf_node_if:
115 sweep_if(nir, nir_cf_node_as_if(cf_node));
116 break;
117 case nir_cf_node_loop:
118 sweep_loop(nir, nir_cf_node_as_loop(cf_node));
119 break;
120 default:
121 unreachable("Invalid CF node type");
122 }
123 }
124
125 static void
sweep_impl(nir_shader * nir,nir_function_impl * impl)126 sweep_impl(nir_shader *nir, nir_function_impl *impl)
127 {
128 ralloc_steal(nir, impl);
129
130 steal_list(nir, nir_variable, &impl->locals);
131 steal_list(nir, nir_register, &impl->registers);
132
133 foreach_list_typed(nir_cf_node, cf_node, node, &impl->body) {
134 sweep_cf_node(nir, cf_node);
135 }
136
137 sweep_block(nir, impl->end_block);
138
139 /* Wipe out all the metadata, if any. */
140 nir_metadata_preserve(impl, nir_metadata_none);
141 }
142
143 static void
sweep_function(nir_shader * nir,nir_function * f)144 sweep_function(nir_shader *nir, nir_function *f)
145 {
146 ralloc_steal(nir, f);
147 ralloc_steal(nir, f->params);
148
149 if (f->impl)
150 sweep_impl(nir, f->impl);
151 }
152
153 void
nir_sweep(nir_shader * nir)154 nir_sweep(nir_shader *nir)
155 {
156 void *rubbish = ralloc_context(NULL);
157
158 /* First, move ownership of all the memory to a temporary context; assume dead. */
159 ralloc_adopt(rubbish, nir);
160
161 ralloc_steal(nir, (char *)nir->info.name);
162 if (nir->info.label)
163 ralloc_steal(nir, (char *)nir->info.label);
164
165 /* Variables and registers are not dead. Steal them back. */
166 steal_list(nir, nir_variable, &nir->variables);
167
168 /* Recurse into functions, stealing their contents back. */
169 foreach_list_typed(nir_function, func, node, &nir->functions) {
170 sweep_function(nir, func);
171 }
172
173 ralloc_steal(nir, nir->constant_data);
174
175 /* Free everything we didn't steal back. */
176 ralloc_free(rubbish);
177 }
178