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/nir/nir.h"
29#include "compiler/nir/nir_builder.h"
30#include "gl_nir.h"
31#include "ir_uniform.h"
32#include "main/config.h"
33#include "main/shader_types.h"
34#include <assert.h>
35
36/*
37 * replace atomic counter intrinsics that use a variable with intrinsics
38 * that directly store the buffer index and byte offset
39 */
40
41static bool
42lower_deref_instr(nir_builder *b, nir_intrinsic_instr *instr,
43                  const struct gl_shader_program *shader_program,
44                  nir_shader *shader, bool use_binding_as_idx)
45{
46   nir_intrinsic_op op;
47   switch (instr->intrinsic) {
48   case nir_intrinsic_atomic_counter_read_deref:
49      op = nir_intrinsic_atomic_counter_read;
50      break;
51
52   case nir_intrinsic_atomic_counter_inc_deref:
53      op = nir_intrinsic_atomic_counter_inc;
54      break;
55
56   case nir_intrinsic_atomic_counter_pre_dec_deref:
57      op = nir_intrinsic_atomic_counter_pre_dec;
58      break;
59
60   case nir_intrinsic_atomic_counter_post_dec_deref:
61      op = nir_intrinsic_atomic_counter_post_dec;
62      break;
63
64   case nir_intrinsic_atomic_counter_add_deref:
65      op = nir_intrinsic_atomic_counter_add;
66      break;
67
68   case nir_intrinsic_atomic_counter_min_deref:
69      op = nir_intrinsic_atomic_counter_min;
70      break;
71
72   case nir_intrinsic_atomic_counter_max_deref:
73      op = nir_intrinsic_atomic_counter_max;
74      break;
75
76   case nir_intrinsic_atomic_counter_and_deref:
77      op = nir_intrinsic_atomic_counter_and;
78      break;
79
80   case nir_intrinsic_atomic_counter_or_deref:
81      op = nir_intrinsic_atomic_counter_or;
82      break;
83
84   case nir_intrinsic_atomic_counter_xor_deref:
85      op = nir_intrinsic_atomic_counter_xor;
86      break;
87
88   case nir_intrinsic_atomic_counter_exchange_deref:
89      op = nir_intrinsic_atomic_counter_exchange;
90      break;
91
92   case nir_intrinsic_atomic_counter_comp_swap_deref:
93      op = nir_intrinsic_atomic_counter_comp_swap;
94      break;
95
96   default:
97      return false;
98   }
99
100   nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
101   nir_variable *var = nir_deref_instr_get_variable(deref);
102
103   if (var->data.mode != nir_var_uniform &&
104       var->data.mode != nir_var_mem_ssbo &&
105       var->data.mode != nir_var_mem_shared)
106      return false; /* atomics passed as function arguments can't be lowered */
107
108   const unsigned uniform_loc = var->data.location;
109   const unsigned idx = use_binding_as_idx ? var->data.binding :
110      shader_program->data->UniformStorage[uniform_loc].opaque[shader->info.stage].index;
111
112   b->cursor = nir_before_instr(&instr->instr);
113
114   nir_ssa_def *offset = nir_imm_int(b, var->data.offset);
115   for (nir_deref_instr *d = deref; d->deref_type != nir_deref_type_var;
116        d = nir_deref_instr_parent(d)) {
117      assert(d->deref_type == nir_deref_type_array);
118      assert(d->arr.index.is_ssa);
119
120      unsigned array_stride = ATOMIC_COUNTER_SIZE;
121      if (glsl_type_is_array(d->type))
122         array_stride *= glsl_get_aoa_size(d->type);
123
124      offset = nir_iadd(b, offset, nir_imul(b, d->arr.index.ssa,
125                                            nir_imm_int(b, array_stride)));
126   }
127
128   /* Since the first source is a deref and the first source in the lowered
129    * instruction is the offset, we can just swap it out and change the
130    * opcode.
131    */
132   instr->intrinsic = op;
133   nir_instr_rewrite_src(&instr->instr, &instr->src[0],
134                         nir_src_for_ssa(offset));
135   nir_intrinsic_set_base(instr, idx);
136
137   nir_deref_instr_remove_if_unused(deref);
138
139   return true;
140}
141
142bool
143gl_nir_lower_atomics(nir_shader *shader,
144                     const struct gl_shader_program *shader_program,
145                     bool use_binding_as_idx)
146{
147   bool progress = false;
148
149   nir_foreach_function(function, shader) {
150      if (!function->impl)
151         continue;
152
153      bool impl_progress = false;
154
155      nir_builder build;
156      nir_builder_init(&build, function->impl);
157
158      nir_foreach_block(block, function->impl) {
159         nir_foreach_instr_safe(instr, block) {
160            if (instr->type != nir_instr_type_intrinsic)
161               continue;
162
163            impl_progress |= lower_deref_instr(&build,
164                                               nir_instr_as_intrinsic(instr),
165                                               shader_program, shader,
166                                               use_binding_as_idx);
167         }
168      }
169
170      if (impl_progress) {
171         nir_metadata_preserve(function->impl, nir_metadata_block_index |
172                                               nir_metadata_dominance);
173         progress = true;
174      }
175   }
176
177   return progress;
178}
179