nir/lower_indirect_derefs: Add a threshold
[mesa.git] / src / compiler / nir / nir_lower_indirect_derefs.c
1 /*
2 * Copyright © 2016 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 #include "nir_builder.h"
26 #include "nir_deref.h"
27
28 static void
29 emit_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
30 nir_deref_instr *parent,
31 nir_deref_instr **deref_arr,
32 nir_ssa_def **dest, nir_ssa_def *src);
33
34 static void
35 emit_indirect_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
36 nir_deref_instr *parent,
37 nir_deref_instr **deref_arr,
38 int start, int end,
39 nir_ssa_def **dest, nir_ssa_def *src)
40 {
41 assert(start < end);
42 if (start == end - 1) {
43 emit_load_store_deref(b, orig_instr,
44 nir_build_deref_array_imm(b, parent, start),
45 deref_arr + 1, dest, src);
46 } else {
47 int mid = start + (end - start) / 2;
48
49 nir_ssa_def *then_dest, *else_dest;
50
51 nir_deref_instr *deref = *deref_arr;
52 assert(deref->deref_type == nir_deref_type_array);
53
54 nir_push_if(b, nir_ilt(b, deref->arr.index.ssa, nir_imm_intN_t(b, mid, parent->dest.ssa.bit_size)));
55 emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
56 start, mid, &then_dest, src);
57 nir_push_else(b, NULL);
58 emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
59 mid, end, &else_dest, src);
60 nir_pop_if(b, NULL);
61
62 if (src == NULL)
63 *dest = nir_if_phi(b, then_dest, else_dest);
64 }
65 }
66
67 static void
68 emit_load_store_deref(nir_builder *b, nir_intrinsic_instr *orig_instr,
69 nir_deref_instr *parent,
70 nir_deref_instr **deref_arr,
71 nir_ssa_def **dest, nir_ssa_def *src)
72 {
73 for (; *deref_arr; deref_arr++) {
74 nir_deref_instr *deref = *deref_arr;
75 if (deref->deref_type == nir_deref_type_array &&
76 !nir_src_is_const(deref->arr.index)) {
77 int length = glsl_get_length(parent->type);
78
79 emit_indirect_load_store_deref(b, orig_instr, parent, deref_arr,
80 0, length, dest, src);
81 return;
82 }
83
84 parent = nir_build_deref_follower(b, parent, deref);
85 }
86
87 /* We reached the end of the deref chain. Emit the instruction */
88 assert(*deref_arr == NULL);
89
90 if (src == NULL) {
91 /* This is a load instruction */
92 nir_intrinsic_instr *load =
93 nir_intrinsic_instr_create(b->shader, orig_instr->intrinsic);
94 load->num_components = orig_instr->num_components;
95
96 load->src[0] = nir_src_for_ssa(&parent->dest.ssa);
97
98 /* Copy over any other sources. This is needed for interp_deref_at */
99 for (unsigned i = 1;
100 i < nir_intrinsic_infos[orig_instr->intrinsic].num_srcs; i++)
101 nir_src_copy(&load->src[i], &orig_instr->src[i], load);
102
103 nir_ssa_dest_init(&load->instr, &load->dest,
104 orig_instr->dest.ssa.num_components,
105 orig_instr->dest.ssa.bit_size, NULL);
106 nir_builder_instr_insert(b, &load->instr);
107 *dest = &load->dest.ssa;
108 } else {
109 assert(orig_instr->intrinsic == nir_intrinsic_store_deref);
110 nir_store_deref(b, parent, src, nir_intrinsic_write_mask(orig_instr));
111 }
112 }
113
114 static bool
115 lower_indirect_derefs_block(nir_block *block, nir_builder *b,
116 nir_variable_mode modes,
117 uint32_t max_lower_array_len)
118 {
119 bool progress = false;
120
121 nir_foreach_instr_safe(instr, block) {
122 if (instr->type != nir_instr_type_intrinsic)
123 continue;
124
125 nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
126 if (intrin->intrinsic != nir_intrinsic_load_deref &&
127 intrin->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
128 intrin->intrinsic != nir_intrinsic_interp_deref_at_sample &&
129 intrin->intrinsic != nir_intrinsic_interp_deref_at_offset &&
130 intrin->intrinsic != nir_intrinsic_interp_deref_at_vertex &&
131 intrin->intrinsic != nir_intrinsic_store_deref)
132 continue;
133
134 nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
135
136 /* Walk the deref chain back to the base and look for indirects */
137 uint32_t indirect_array_len = 1;
138 bool has_indirect = false;
139 nir_deref_instr *base = deref;
140 while (base && base->deref_type != nir_deref_type_var) {
141 nir_deref_instr *parent = nir_deref_instr_parent(base);
142 if (base->deref_type == nir_deref_type_array &&
143 !nir_src_is_const(base->arr.index)) {
144 indirect_array_len *= glsl_get_length(parent->type);
145 has_indirect = true;
146 }
147
148 base = parent;
149 }
150
151 if (!has_indirect || !base || indirect_array_len > max_lower_array_len)
152 continue;
153
154 /* Only lower variables whose mode is in the mask, or compact
155 * array variables. (We can't handle indirects on tightly packed
156 * scalar arrays, so we need to lower them regardless.)
157 */
158 if (!(modes & base->var->data.mode) && !base->var->data.compact)
159 continue;
160
161 b->cursor = nir_instr_remove(&intrin->instr);
162
163 nir_deref_path path;
164 nir_deref_path_init(&path, deref, NULL);
165 assert(path.path[0] == base);
166
167 if (intrin->intrinsic == nir_intrinsic_store_deref) {
168 assert(intrin->src[1].is_ssa);
169 emit_load_store_deref(b, intrin, base, &path.path[1],
170 NULL, intrin->src[1].ssa);
171 } else {
172 nir_ssa_def *result;
173 emit_load_store_deref(b, intrin, base, &path.path[1],
174 &result, NULL);
175 nir_ssa_def_rewrite_uses(&intrin->dest.ssa, nir_src_for_ssa(result));
176 }
177
178 nir_deref_path_finish(&path);
179
180 progress = true;
181 }
182
183 return progress;
184 }
185
186 static bool
187 lower_indirects_impl(nir_function_impl *impl, nir_variable_mode modes,
188 uint32_t max_lower_array_len)
189 {
190 nir_builder builder;
191 nir_builder_init(&builder, impl);
192 bool progress = false;
193
194 nir_foreach_block_safe(block, impl) {
195 progress |= lower_indirect_derefs_block(block, &builder, modes,
196 max_lower_array_len);
197 }
198
199 if (progress)
200 nir_metadata_preserve(impl, nir_metadata_none);
201 else
202 nir_metadata_preserve(impl, nir_metadata_all);
203
204 return progress;
205 }
206
207 /** Lowers indirect variable loads/stores to direct loads/stores.
208 *
209 * The pass works by replacing any indirect load or store with an if-ladder
210 * that does a binary search on the array index.
211 */
212 bool
213 nir_lower_indirect_derefs(nir_shader *shader, nir_variable_mode modes,
214 uint32_t max_lower_array_len)
215 {
216 bool progress = false;
217
218 nir_foreach_function(function, shader) {
219 if (function->impl) {
220 progress = lower_indirects_impl(function->impl, modes,
221 max_lower_array_len) || progress;
222 }
223 }
224
225 return progress;
226 }