1 /*
2 * Copyright © 2014 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 * Authors:
24 * Connor Abbott (cwabbott0@gmail.com)
25 *
26 */
27
28 #include "nir.h"
29 #include "nir_builder.h"
30
31 /**
32 * SSA-based copy propagation
33 */
34
35 static bool
is_swizzleless_move(nir_alu_instr * instr)36 is_swizzleless_move(nir_alu_instr *instr)
37 {
38 unsigned num_comp = instr->dest.dest.ssa.num_components;
39
40 if (instr->src[0].src.ssa->num_components != num_comp)
41 return false;
42
43 if (instr->op == nir_op_mov) {
44 for (unsigned i = 0; i < num_comp; i++) {
45 if (instr->src[0].swizzle[i] != i)
46 return false;
47 }
48 } else {
49 for (unsigned i = 0; i < num_comp; i++) {
50 if (instr->src[i].swizzle[0] != i ||
51 instr->src[i].src.ssa != instr->src[0].src.ssa)
52 return false;
53 }
54 }
55
56 return true;
57 }
58
59 static bool
rewrite_to_vec(nir_function_impl * impl,nir_alu_instr * mov,nir_alu_instr * vec)60 rewrite_to_vec(nir_function_impl *impl, nir_alu_instr *mov, nir_alu_instr *vec)
61 {
62 if (mov->op != nir_op_mov)
63 return false;
64
65 nir_builder b;
66 nir_builder_init(&b, impl);
67 b.cursor = nir_after_instr(&mov->instr);
68
69 unsigned num_comp = mov->dest.dest.ssa.num_components;
70 nir_alu_instr *new_vec = nir_alu_instr_create(b.shader, nir_op_vec(num_comp));
71 for (unsigned i = 0; i < num_comp; i++)
72 new_vec->src[i] = vec->src[mov->src[0].swizzle[i]];
73
74 nir_ssa_def *new = nir_builder_alu_instr_finish_and_insert(&b, new_vec);
75 nir_ssa_def_rewrite_uses(&mov->dest.dest.ssa, new);
76
77 /* If we remove "mov" and it's the next instruction in the
78 * nir_foreach_instr_safe() loop, then we would end copy-propagation early. */
79
80 return true;
81 }
82
83 static bool
copy_propagate_alu(nir_function_impl * impl,nir_alu_src * src,nir_alu_instr * copy)84 copy_propagate_alu(nir_function_impl *impl, nir_alu_src *src, nir_alu_instr *copy)
85 {
86 nir_ssa_def *def = NULL;
87 nir_alu_instr *user = nir_instr_as_alu(src->src.parent_instr);
88 unsigned src_idx = src - user->src;
89 assert(src_idx < nir_op_infos[user->op].num_inputs);
90 unsigned num_comp = nir_ssa_alu_instr_src_components(user, src_idx);
91
92 if (copy->op == nir_op_mov) {
93 def = copy->src[0].src.ssa;
94
95 for (unsigned i = 0; i < num_comp; i++)
96 src->swizzle[i] = copy->src[0].swizzle[src->swizzle[i]];
97 } else {
98 def = copy->src[src->swizzle[0]].src.ssa;
99
100 for (unsigned i = 1; i < num_comp; i++) {
101 if (copy->src[src->swizzle[i]].src.ssa != def)
102 return rewrite_to_vec(impl, user, copy);
103 }
104
105 for (unsigned i = 0; i < num_comp; i++)
106 src->swizzle[i] = copy->src[src->swizzle[i]].swizzle[0];
107 }
108
109 nir_instr_rewrite_src_ssa(src->src.parent_instr, &src->src, def);
110
111 return true;
112 }
113
114 static bool
copy_propagate(nir_src * src,nir_alu_instr * copy)115 copy_propagate(nir_src *src, nir_alu_instr *copy)
116 {
117 if (!is_swizzleless_move(copy))
118 return false;
119
120 nir_instr_rewrite_src_ssa(src->parent_instr, src, copy->src[0].src.ssa);
121
122 return true;
123 }
124
125 static bool
copy_propagate_if(nir_src * src,nir_alu_instr * copy)126 copy_propagate_if(nir_src *src, nir_alu_instr *copy)
127 {
128 if (!is_swizzleless_move(copy))
129 return false;
130
131 nir_if_rewrite_condition_ssa(src->parent_if, src, copy->src[0].src.ssa);
132
133 return true;
134 }
135
136 static bool
copy_prop_instr(nir_function_impl * impl,nir_instr * instr)137 copy_prop_instr(nir_function_impl *impl, nir_instr *instr)
138 {
139 if (instr->type != nir_instr_type_alu)
140 return false;
141
142 nir_alu_instr *mov = nir_instr_as_alu(instr);
143
144 if (!nir_alu_instr_is_copy(mov))
145 return false;
146
147 bool progress = false;
148
149 nir_foreach_use_safe(src, &mov->dest.dest.ssa) {
150 if (src->parent_instr->type == nir_instr_type_alu)
151 progress |= copy_propagate_alu(impl, container_of(src, nir_alu_src, src), mov);
152 else
153 progress |= copy_propagate(src, mov);
154 }
155
156 nir_foreach_if_use_safe(src, &mov->dest.dest.ssa)
157 progress |= copy_propagate_if(src, mov);
158
159 if (progress && nir_ssa_def_is_unused(&mov->dest.dest.ssa))
160 nir_instr_remove(&mov->instr);
161
162 return progress;
163 }
164
165 bool
nir_copy_prop_impl(nir_function_impl * impl)166 nir_copy_prop_impl(nir_function_impl *impl)
167 {
168 bool progress = false;
169
170 nir_foreach_block(block, impl) {
171 nir_foreach_instr_safe(instr, block) {
172 progress |= copy_prop_instr(impl, instr);
173 }
174 }
175
176 if (progress) {
177 nir_metadata_preserve(impl, nir_metadata_block_index |
178 nir_metadata_dominance);
179 } else {
180 nir_metadata_preserve(impl, nir_metadata_all);
181 }
182
183 return progress;
184 }
185
186 bool
nir_copy_prop(nir_shader * shader)187 nir_copy_prop(nir_shader *shader)
188 {
189 bool progress = false;
190
191 nir_foreach_function(function, shader) {
192 if (function->impl && nir_copy_prop_impl(function->impl))
193 progress = true;
194 }
195
196 return progress;
197 }
198