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 "compiler/glsl/ir_uniform.h"
29 #include "nir.h"
30 #include "main/config.h"
31 #include <assert.h>
32
33 /*
34 * replace atomic counter intrinsics that use a variable with intrinsics
35 * that directly store the buffer index and byte offset
36 */
37
38 static bool
lower_instr(nir_intrinsic_instr * instr,const struct gl_shader_program * shader_program,nir_shader * shader)39 lower_instr(nir_intrinsic_instr *instr,
40 const struct gl_shader_program *shader_program,
41 nir_shader *shader)
42 {
43 nir_intrinsic_op op;
44 switch (instr->intrinsic) {
45 case nir_intrinsic_atomic_counter_read_var:
46 op = nir_intrinsic_atomic_counter_read;
47 break;
48
49 case nir_intrinsic_atomic_counter_inc_var:
50 op = nir_intrinsic_atomic_counter_inc;
51 break;
52
53 case nir_intrinsic_atomic_counter_dec_var:
54 op = nir_intrinsic_atomic_counter_dec;
55 break;
56
57 case nir_intrinsic_atomic_counter_add_var:
58 op = nir_intrinsic_atomic_counter_add;
59 break;
60
61 case nir_intrinsic_atomic_counter_min_var:
62 op = nir_intrinsic_atomic_counter_min;
63 break;
64
65 case nir_intrinsic_atomic_counter_max_var:
66 op = nir_intrinsic_atomic_counter_max;
67 break;
68
69 case nir_intrinsic_atomic_counter_and_var:
70 op = nir_intrinsic_atomic_counter_and;
71 break;
72
73 case nir_intrinsic_atomic_counter_or_var:
74 op = nir_intrinsic_atomic_counter_or;
75 break;
76
77 case nir_intrinsic_atomic_counter_xor_var:
78 op = nir_intrinsic_atomic_counter_xor;
79 break;
80
81 case nir_intrinsic_atomic_counter_exchange_var:
82 op = nir_intrinsic_atomic_counter_exchange;
83 break;
84
85 case nir_intrinsic_atomic_counter_comp_swap_var:
86 op = nir_intrinsic_atomic_counter_comp_swap;
87 break;
88
89 default:
90 return false;
91 }
92
93 if (instr->variables[0]->var->data.mode != nir_var_uniform &&
94 instr->variables[0]->var->data.mode != nir_var_shader_storage &&
95 instr->variables[0]->var->data.mode != nir_var_shared)
96 return false; /* atomics passed as function arguments can't be lowered */
97
98 void *mem_ctx = ralloc_parent(instr);
99 unsigned uniform_loc = instr->variables[0]->var->data.location;
100
101 nir_intrinsic_instr *new_instr = nir_intrinsic_instr_create(mem_ctx, op);
102 nir_intrinsic_set_base(new_instr,
103 shader_program->data->UniformStorage[uniform_loc].opaque[shader->info.stage].index);
104
105 nir_load_const_instr *offset_const =
106 nir_load_const_instr_create(mem_ctx, 1, 32);
107 offset_const->value.u32[0] = instr->variables[0]->var->data.offset;
108
109 nir_instr_insert_before(&instr->instr, &offset_const->instr);
110
111 nir_ssa_def *offset_def = &offset_const->def;
112
113 nir_deref *tail = &instr->variables[0]->deref;
114 while (tail->child != NULL) {
115 nir_deref_array *deref_array = nir_deref_as_array(tail->child);
116 tail = tail->child;
117
118 unsigned child_array_elements = tail->child != NULL ?
119 glsl_get_aoa_size(tail->type) : 1;
120
121 offset_const->value.u32[0] += deref_array->base_offset *
122 child_array_elements * ATOMIC_COUNTER_SIZE;
123
124 if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
125 nir_load_const_instr *atomic_counter_size =
126 nir_load_const_instr_create(mem_ctx, 1, 32);
127 atomic_counter_size->value.u32[0] = child_array_elements * ATOMIC_COUNTER_SIZE;
128 nir_instr_insert_before(&instr->instr, &atomic_counter_size->instr);
129
130 nir_alu_instr *mul = nir_alu_instr_create(mem_ctx, nir_op_imul);
131 nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, 32, NULL);
132 mul->dest.write_mask = 0x1;
133 nir_src_copy(&mul->src[0].src, &deref_array->indirect, mul);
134 mul->src[1].src.is_ssa = true;
135 mul->src[1].src.ssa = &atomic_counter_size->def;
136 nir_instr_insert_before(&instr->instr, &mul->instr);
137
138 nir_alu_instr *add = nir_alu_instr_create(mem_ctx, nir_op_iadd);
139 nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, 32, NULL);
140 add->dest.write_mask = 0x1;
141 add->src[0].src.is_ssa = true;
142 add->src[0].src.ssa = &mul->dest.dest.ssa;
143 add->src[1].src.is_ssa = true;
144 add->src[1].src.ssa = offset_def;
145 nir_instr_insert_before(&instr->instr, &add->instr);
146
147 offset_def = &add->dest.dest.ssa;
148 }
149 }
150
151 new_instr->src[0].is_ssa = true;
152 new_instr->src[0].ssa = offset_def;
153
154 /* Copy the other sources, if any, from the original instruction to the new
155 * instruction.
156 */
157 for (unsigned i = 0; i < nir_intrinsic_infos[instr->intrinsic].num_srcs; i++)
158 nir_src_copy(&new_instr->src[i + 1], &instr->src[i], new_instr);
159
160 if (instr->dest.is_ssa) {
161 nir_ssa_dest_init(&new_instr->instr, &new_instr->dest,
162 instr->dest.ssa.num_components, 32, NULL);
163 nir_ssa_def_rewrite_uses(&instr->dest.ssa,
164 nir_src_for_ssa(&new_instr->dest.ssa));
165 } else {
166 nir_dest_copy(&new_instr->dest, &instr->dest, mem_ctx);
167 }
168
169 nir_instr_insert_before(&instr->instr, &new_instr->instr);
170 nir_instr_remove(&instr->instr);
171
172 return true;
173 }
174
175 bool
nir_lower_atomics(nir_shader * shader,const struct gl_shader_program * shader_program)176 nir_lower_atomics(nir_shader *shader,
177 const struct gl_shader_program *shader_program)
178 {
179 bool progress = false;
180
181 nir_foreach_function(function, shader) {
182 if (function->impl) {
183 nir_foreach_block(block, function->impl) {
184 nir_foreach_instr_safe(instr, block) {
185 if (instr->type == nir_instr_type_intrinsic)
186 progress |= lower_instr(nir_instr_as_intrinsic(instr),
187 shader_program, shader);
188 }
189 }
190
191 nir_metadata_preserve(function->impl, nir_metadata_block_index |
192 nir_metadata_dominance);
193 }
194 }
195
196 return progress;
197 }
198