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
41static void sweep_cf_node(nir_shader *nir, nir_cf_node *cf_node);
42
43static void
44sweep_block(nir_shader *nir, nir_block *block)
45{
46   ralloc_steal(nir, block);
47
48   /* sweep_impl will mark all metadata invalid.  We can safely release all of
49    * this here.
50    */
51   ralloc_free(block->live_in);
52   block->live_in = NULL;
53
54   ralloc_free(block->live_out);
55   block->live_out = NULL;
56
57   nir_foreach_instr(instr, block) {
58      list_del(&instr->gc_node);
59      list_add(&instr->gc_node, &nir->gc_list);
60   }
61}
62
63static void
64sweep_if(nir_shader *nir, nir_if *iff)
65{
66   ralloc_steal(nir, iff);
67
68   foreach_list_typed(nir_cf_node, cf_node, node, &iff->then_list) {
69      sweep_cf_node(nir, cf_node);
70   }
71
72   foreach_list_typed(nir_cf_node, cf_node, node, &iff->else_list) {
73      sweep_cf_node(nir, cf_node);
74   }
75}
76
77static void
78sweep_loop(nir_shader *nir, nir_loop *loop)
79{
80   ralloc_steal(nir, loop);
81
82   foreach_list_typed(nir_cf_node, cf_node, node, &loop->body) {
83      sweep_cf_node(nir, cf_node);
84   }
85}
86
87static void
88sweep_cf_node(nir_shader *nir, nir_cf_node *cf_node)
89{
90   switch (cf_node->type) {
91   case nir_cf_node_block:
92      sweep_block(nir, nir_cf_node_as_block(cf_node));
93      break;
94   case nir_cf_node_if:
95      sweep_if(nir, nir_cf_node_as_if(cf_node));
96      break;
97   case nir_cf_node_loop:
98      sweep_loop(nir, nir_cf_node_as_loop(cf_node));
99      break;
100   default:
101      unreachable("Invalid CF node type");
102   }
103}
104
105static void
106sweep_impl(nir_shader *nir, nir_function_impl *impl)
107{
108   ralloc_steal(nir, impl);
109
110   steal_list(nir, nir_variable, &impl->locals);
111   steal_list(nir, nir_register, &impl->registers);
112
113   foreach_list_typed(nir_cf_node, cf_node, node, &impl->body) {
114      sweep_cf_node(nir, cf_node);
115   }
116
117   sweep_block(nir, impl->end_block);
118
119   /* Wipe out all the metadata, if any. */
120   nir_metadata_preserve(impl, nir_metadata_none);
121}
122
123static void
124sweep_function(nir_shader *nir, nir_function *f)
125{
126   ralloc_steal(nir, f);
127   ralloc_steal(nir, f->params);
128
129   if (f->impl)
130      sweep_impl(nir, f->impl);
131}
132
133void
134nir_sweep(nir_shader *nir)
135{
136   void *rubbish = ralloc_context(NULL);
137
138   struct list_head instr_gc_list;
139   list_inithead(&instr_gc_list);
140
141   list_replace(&nir->gc_list, &instr_gc_list);
142   list_inithead(&nir->gc_list);
143
144   /* First, move ownership of all the memory to a temporary context; assume dead. */
145   ralloc_adopt(rubbish, nir);
146
147   ralloc_steal(nir, (char *)nir->info.name);
148   if (nir->info.label)
149      ralloc_steal(nir, (char *)nir->info.label);
150
151   /* Variables and registers are not dead.  Steal them back. */
152   steal_list(nir, nir_variable, &nir->variables);
153
154   /* Recurse into functions, stealing their contents back. */
155   foreach_list_typed(nir_function, func, node, &nir->functions) {
156      sweep_function(nir, func);
157   }
158
159   /* Sweep instrs not found while walking the shader. */
160   list_for_each_entry_safe(nir_instr, instr, &instr_gc_list, gc_node) {
161      nir_instr_free(instr);
162   }
163   assert(list_is_empty(&instr_gc_list));
164
165   ralloc_steal(nir, nir->constant_data);
166   ralloc_steal(nir, nir->xfb_info);
167   ralloc_steal(nir, nir->printf_info);
168   for (int i = 0; i < nir->printf_info_count; i++) {
169      ralloc_steal(nir, nir->printf_info[i].arg_sizes);
170      ralloc_steal(nir, nir->printf_info[i].strings);
171   }
172
173   /* Free everything we didn't steal back. */
174   ralloc_free(rubbish);
175}
176