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 * Jason Ekstrand (jason@jlekstrand.net)
25 *
26 */
27
28 #include "nir.h"
29
30 /*
31 * This pass lowers the neg, abs, and sat operations to source modifiers on
32 * ALU operations to make things nicer for the backend. It's just much
33 * easier to not have them when we're doing optimizations.
34 */
35
36 static void
alu_src_consume_abs(nir_alu_src * src)37 alu_src_consume_abs(nir_alu_src *src)
38 {
39 src->abs = true;
40 }
41
42 static void
alu_src_consume_negate(nir_alu_src * src)43 alu_src_consume_negate(nir_alu_src *src)
44 {
45 /* If abs is set on the source, the negate goes away */
46 if (!src->abs)
47 src->negate = !src->negate;
48 }
49
50 static bool
nir_lower_to_source_mods_block(nir_block * block,nir_lower_to_source_mods_flags options)51 nir_lower_to_source_mods_block(nir_block *block,
52 nir_lower_to_source_mods_flags options)
53 {
54 bool progress = false;
55
56 nir_foreach_instr(instr, block) {
57 if (instr->type != nir_instr_type_alu)
58 continue;
59
60 nir_alu_instr *alu = nir_instr_as_alu(instr);
61
62 bool lower_abs = (nir_op_infos[alu->op].num_inputs < 3) ||
63 (options & nir_lower_triop_abs);
64
65 for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) {
66 if (!alu->src[i].src.is_ssa)
67 continue;
68
69 if (alu->src[i].src.ssa->parent_instr->type != nir_instr_type_alu)
70 continue;
71
72 nir_alu_instr *parent = nir_instr_as_alu(alu->src[i].src.ssa->parent_instr);
73
74 if (parent->dest.saturate)
75 continue;
76
77 switch (nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[i])) {
78 case nir_type_float:
79 if (!(options & nir_lower_float_source_mods))
80 continue;
81 if (parent->op != nir_op_fabs && parent->op != nir_op_fneg)
82 continue;
83 break;
84 case nir_type_int:
85 if (!(options & nir_lower_int_source_mods))
86 continue;
87 if (parent->op != nir_op_iabs && parent->op != nir_op_ineg)
88 continue;
89 break;
90 default:
91 continue;
92 }
93
94 if (nir_src_bit_size(alu->src[i].src) == 64 &&
95 !(options & nir_lower_64bit_source_mods)) {
96 continue;
97 }
98
99 /* We can only do a rewrite if the source we are copying is SSA.
100 * Otherwise, moving the read might invalidly reorder reads/writes
101 * on a register.
102 */
103 if (!parent->src[0].src.is_ssa)
104 continue;
105
106 if (!lower_abs && (parent->op == nir_op_fabs ||
107 parent->op == nir_op_iabs))
108 continue;
109
110 nir_instr_rewrite_src(instr, &alu->src[i].src, parent->src[0].src);
111
112 /* Apply any modifiers that come from the parent opcode */
113 if (parent->op == nir_op_fneg || parent->op == nir_op_ineg)
114 alu_src_consume_negate(&alu->src[i]);
115 if (parent->op == nir_op_fabs || parent->op == nir_op_iabs)
116 alu_src_consume_abs(&alu->src[i]);
117
118 /* Apply modifiers from the parent source */
119 if (parent->src[0].negate)
120 alu_src_consume_negate(&alu->src[i]);
121 if (parent->src[0].abs)
122 alu_src_consume_abs(&alu->src[i]);
123
124 for (int j = 0; j < 4; ++j) {
125 if (!nir_alu_instr_channel_used(alu, i, j))
126 continue;
127 alu->src[i].swizzle[j] = parent->src[0].swizzle[alu->src[i].swizzle[j]];
128 }
129
130 if (list_is_empty(&parent->dest.dest.ssa.uses) &&
131 list_is_empty(&parent->dest.dest.ssa.if_uses))
132 nir_instr_remove(&parent->instr);
133
134 progress = true;
135 }
136
137 /* We've covered sources. Now we're going to try and saturate the
138 * destination if we can.
139 */
140
141 if (!alu->dest.dest.is_ssa)
142 continue;
143
144 if (nir_dest_bit_size(alu->dest.dest) == 64 &&
145 !(options & nir_lower_64bit_source_mods)) {
146 continue;
147 }
148
149 /* We can only saturate float destinations */
150 if (nir_alu_type_get_base_type(nir_op_infos[alu->op].output_type) !=
151 nir_type_float)
152 continue;
153
154 if (!(options & nir_lower_float_source_mods))
155 continue;
156
157 if (!list_is_empty(&alu->dest.dest.ssa.if_uses))
158 continue;
159
160 bool all_children_are_sat = true;
161 nir_foreach_use(child_src, &alu->dest.dest.ssa) {
162 assert(child_src->is_ssa);
163 nir_instr *child = child_src->parent_instr;
164 if (child->type != nir_instr_type_alu) {
165 all_children_are_sat = false;
166 continue;
167 }
168
169 nir_alu_instr *child_alu = nir_instr_as_alu(child);
170 if (child_alu->src[0].negate || child_alu->src[0].abs) {
171 all_children_are_sat = false;
172 continue;
173 }
174
175 if (child_alu->op != nir_op_fsat) {
176 all_children_are_sat = false;
177 continue;
178 }
179 }
180
181 if (!all_children_are_sat)
182 continue;
183
184 alu->dest.saturate = true;
185 progress = true;
186
187 nir_foreach_use(child_src, &alu->dest.dest.ssa) {
188 assert(child_src->is_ssa);
189 nir_alu_instr *child_alu = nir_instr_as_alu(child_src->parent_instr);
190
191 child_alu->op = nir_op_mov;
192 child_alu->dest.saturate = false;
193 /* We could propagate the dest of our instruction to the
194 * destinations of the uses here. However, one quick round of
195 * copy propagation will clean that all up and then we don't have
196 * the complexity.
197 */
198 }
199 }
200
201 return progress;
202 }
203
204 static bool
nir_lower_to_source_mods_impl(nir_function_impl * impl,nir_lower_to_source_mods_flags options)205 nir_lower_to_source_mods_impl(nir_function_impl *impl,
206 nir_lower_to_source_mods_flags options)
207 {
208 bool progress = false;
209
210 nir_foreach_block(block, impl) {
211 progress |= nir_lower_to_source_mods_block(block, options);
212 }
213
214 if (progress)
215 nir_metadata_preserve(impl, nir_metadata_block_index |
216 nir_metadata_dominance);
217
218 return progress;
219 }
220
221 bool
nir_lower_to_source_mods(nir_shader * shader,nir_lower_to_source_mods_flags options)222 nir_lower_to_source_mods(nir_shader *shader,
223 nir_lower_to_source_mods_flags options)
224 {
225 bool progress = false;
226
227 nir_foreach_function(function, shader) {
228 if (function->impl) {
229 progress |= nir_lower_to_source_mods_impl(function->impl, options);
230 }
231 }
232
233 return progress;
234 }
235