1 /*
2 * Copyright © 2019 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
27 static void
build_write_masked_store(nir_builder * b,nir_deref_instr * vec_deref,nir_ssa_def * value,unsigned component)28 build_write_masked_store(nir_builder *b, nir_deref_instr *vec_deref,
29 nir_ssa_def *value, unsigned component)
30 {
31 assert(value->num_components == 1);
32 unsigned num_components = glsl_get_components(vec_deref->type);
33 assert(num_components > 1 && num_components <= NIR_MAX_VEC_COMPONENTS);
34
35 nir_ssa_def *u = nir_ssa_undef(b, 1, value->bit_size);
36 nir_ssa_def *comps[NIR_MAX_VEC_COMPONENTS];
37 for (unsigned i = 0; i < num_components; i++)
38 comps[i] = (i == component) ? value : u;
39
40 nir_ssa_def *vec = nir_vec(b, comps, num_components);
41 nir_store_deref(b, vec_deref, vec, (1u << component));
42 }
43
44 static void
build_write_masked_stores(nir_builder * b,nir_deref_instr * vec_deref,nir_ssa_def * value,nir_ssa_def * index,unsigned start,unsigned end)45 build_write_masked_stores(nir_builder *b, nir_deref_instr *vec_deref,
46 nir_ssa_def *value, nir_ssa_def *index,
47 unsigned start, unsigned end)
48 {
49 if (start == end - 1) {
50 build_write_masked_store(b, vec_deref, value, start);
51 } else {
52 unsigned mid = start + (end - start) / 2;
53 nir_push_if(b, nir_ilt(b, index, nir_imm_int(b, mid)));
54 build_write_masked_stores(b, vec_deref, value, index, start, mid);
55 nir_push_else(b, NULL);
56 build_write_masked_stores(b, vec_deref, value, index, mid, end);
57 nir_pop_if(b, NULL);
58 }
59 }
60
61 static bool
nir_lower_array_deref_of_vec_impl(nir_function_impl * impl,nir_variable_mode modes,nir_lower_array_deref_of_vec_options options)62 nir_lower_array_deref_of_vec_impl(nir_function_impl *impl,
63 nir_variable_mode modes,
64 nir_lower_array_deref_of_vec_options options)
65 {
66 bool progress = false;
67
68 nir_builder b;
69 nir_builder_init(&b, impl);
70
71 nir_foreach_block(block, impl) {
72 nir_foreach_instr_safe(instr, block) {
73 if (instr->type != nir_instr_type_intrinsic)
74 continue;
75
76 nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
77 assert(intrin->intrinsic != nir_intrinsic_copy_deref);
78
79 if (intrin->intrinsic != nir_intrinsic_load_deref &&
80 intrin->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
81 intrin->intrinsic != nir_intrinsic_interp_deref_at_sample &&
82 intrin->intrinsic != nir_intrinsic_interp_deref_at_offset &&
83 intrin->intrinsic != nir_intrinsic_interp_deref_at_vertex &&
84 intrin->intrinsic != nir_intrinsic_store_deref)
85 continue;
86
87 nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
88
89 /* We choose to be conservative here. If the deref contains any
90 * modes which weren't specified, we bail and don't bother lowering.
91 */
92 if (!nir_deref_mode_must_be(deref, modes))
93 continue;
94
95 /* We only care about array derefs that act on vectors */
96 if (deref->deref_type != nir_deref_type_array)
97 continue;
98
99 nir_deref_instr *vec_deref = nir_deref_instr_parent(deref);
100 if (!glsl_type_is_vector(vec_deref->type))
101 continue;
102
103 assert(intrin->num_components == 1);
104 unsigned num_components = glsl_get_components(vec_deref->type);
105 assert(num_components > 1 && num_components <= NIR_MAX_VEC_COMPONENTS);
106
107 b.cursor = nir_after_instr(&intrin->instr);
108
109 if (intrin->intrinsic == nir_intrinsic_store_deref) {
110 assert(intrin->src[1].is_ssa);
111 nir_ssa_def *value = intrin->src[1].ssa;
112
113 if (nir_src_is_const(deref->arr.index)) {
114 if (!(options & nir_lower_direct_array_deref_of_vec_store))
115 continue;
116
117 unsigned index = nir_src_as_uint(deref->arr.index);
118 /* If index is OOB, we throw the old store away and don't
119 * replace it with anything.
120 */
121 if (index < num_components)
122 build_write_masked_store(&b, vec_deref, value, index);
123 } else {
124 if (!(options & nir_lower_indirect_array_deref_of_vec_store))
125 continue;
126
127 nir_ssa_def *index = nir_ssa_for_src(&b, deref->arr.index, 1);
128 build_write_masked_stores(&b, vec_deref, value, index,
129 0, num_components);
130 }
131 nir_instr_remove(&intrin->instr);
132
133 progress = true;
134 } else {
135 if (nir_src_is_const(deref->arr.index)) {
136 if (!(options & nir_lower_direct_array_deref_of_vec_load))
137 continue;
138 } else {
139 if (!(options & nir_lower_indirect_array_deref_of_vec_load))
140 continue;
141 }
142
143 /* Turn the load into a vector load */
144 nir_instr_rewrite_src(&intrin->instr, &intrin->src[0],
145 nir_src_for_ssa(&vec_deref->dest.ssa));
146 intrin->dest.ssa.num_components = num_components;
147 intrin->num_components = num_components;
148
149 nir_ssa_def *index = nir_ssa_for_src(&b, deref->arr.index, 1);
150 nir_ssa_def *scalar =
151 nir_vector_extract(&b, &intrin->dest.ssa, index);
152 if (scalar->parent_instr->type == nir_instr_type_ssa_undef) {
153 nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
154 scalar);
155 nir_instr_remove(&intrin->instr);
156 } else {
157 nir_ssa_def_rewrite_uses_after(&intrin->dest.ssa,
158 scalar,
159 scalar->parent_instr);
160 }
161 progress = true;
162 }
163 }
164 }
165
166 if (progress) {
167 nir_metadata_preserve(impl, nir_metadata_block_index |
168 nir_metadata_dominance);
169 } else {
170 nir_metadata_preserve(impl, nir_metadata_all);
171 }
172
173 return progress;
174 }
175
176 /* Lowers away array dereferences on vectors
177 *
178 * These are allowed on certain variable types such as SSBOs and TCS outputs.
179 * However, not everyone can actually handle them everywhere. There are also
180 * cases where we want to lower them for performance reasons.
181 *
182 * This patch assumes that copy_deref instructions have already been lowered.
183 */
184 bool
nir_lower_array_deref_of_vec(nir_shader * shader,nir_variable_mode modes,nir_lower_array_deref_of_vec_options options)185 nir_lower_array_deref_of_vec(nir_shader *shader, nir_variable_mode modes,
186 nir_lower_array_deref_of_vec_options options)
187 {
188 bool progress = false;
189
190 nir_foreach_function(function, shader) {
191 if (function->impl &&
192 nir_lower_array_deref_of_vec_impl(function->impl, modes, options))
193 progress = true;
194 }
195
196 return progress;
197 }
198