1 /*
2 * Copyright © 2020 Mike Blumenkrantz
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 * Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
25 */
26
27 #include "nir.h"
28 #include "nir_builder.h"
29
30 bool nir_lower_dynamic_bo_access(nir_shader *shader);
31 /**
32 * This pass converts dynamic UBO/SSBO block indices to constant indices by generating
33 * conditional chains which reduce to single values.
34 *
35 * This is needed by anything which intends to convert GLSL-like shaders to SPIRV,
36 * as SPIRV requires explicit load points for UBO/SSBO variables and has no instruction for
37 * loading based on an offset in the underlying driver's binding table
38 */
39
40
41 /* generate a single ssa value which conditionally selects the right value that
42 * was previously loaded by the load_ubo conditional chain
43 */
44 static nir_ssa_def *
recursive_generate_bo_ssa_def(nir_builder * b,nir_intrinsic_instr * instr,nir_ssa_def * index,unsigned start,unsigned end)45 recursive_generate_bo_ssa_def(nir_builder *b, nir_intrinsic_instr *instr, nir_ssa_def *index, unsigned start, unsigned end)
46 {
47 if (start == end - 1) {
48 /* block index src is 1 for this op */
49 unsigned block_idx = instr->intrinsic == nir_intrinsic_store_ssbo;
50 nir_intrinsic_instr *new_instr = nir_intrinsic_instr_create(b->shader, instr->intrinsic);
51 new_instr->src[block_idx] = nir_src_for_ssa(nir_imm_int(b, start));
52 for (unsigned i = 0; i < nir_intrinsic_infos[instr->intrinsic].num_srcs; i++) {
53 if (i != block_idx)
54 nir_src_copy(&new_instr->src[i], &instr->src[i], &new_instr->instr);
55 }
56 if (instr->intrinsic != nir_intrinsic_load_ubo_vec4) {
57 nir_intrinsic_set_align(new_instr, nir_intrinsic_align_mul(instr), nir_intrinsic_align_offset(instr));
58 if (instr->intrinsic != nir_intrinsic_load_ssbo)
59 nir_intrinsic_set_range(new_instr, nir_intrinsic_range(instr));
60 }
61 new_instr->num_components = instr->num_components;
62 if (instr->intrinsic != nir_intrinsic_store_ssbo)
63 nir_ssa_dest_init(&new_instr->instr, &new_instr->dest,
64 nir_dest_num_components(instr->dest),
65 nir_dest_bit_size(instr->dest), NULL);
66 nir_builder_instr_insert(b, &new_instr->instr);
67 return &new_instr->dest.ssa;
68 }
69
70 unsigned mid = start + (end - start) / 2;
71 return nir_build_alu(b, nir_op_bcsel, nir_build_alu(b, nir_op_ilt, index, nir_imm_int(b, mid), NULL, NULL),
72 recursive_generate_bo_ssa_def(b, instr, index, start, mid),
73 recursive_generate_bo_ssa_def(b, instr, index, mid, end),
74 NULL
75 );
76 }
77
78 static bool
lower_dynamic_bo_access_instr(nir_intrinsic_instr * instr,nir_builder * b)79 lower_dynamic_bo_access_instr(nir_intrinsic_instr *instr, nir_builder *b)
80 {
81 if (instr->intrinsic != nir_intrinsic_load_ubo &&
82 instr->intrinsic != nir_intrinsic_load_ubo_vec4 &&
83 instr->intrinsic != nir_intrinsic_get_ssbo_size &&
84 instr->intrinsic != nir_intrinsic_load_ssbo &&
85 instr->intrinsic != nir_intrinsic_store_ssbo)
86 return false;
87 /* block index src is 1 for this op */
88 unsigned block_idx = instr->intrinsic == nir_intrinsic_store_ssbo;
89 if (nir_src_is_const(instr->src[block_idx]))
90 return false;
91 b->cursor = nir_after_instr(&instr->instr);
92 bool ssbo_mode = instr->intrinsic != nir_intrinsic_load_ubo && instr->intrinsic != nir_intrinsic_load_ubo_vec4;
93 unsigned first_idx = UINT_MAX, last_idx;
94 if (ssbo_mode) {
95 /* ssbo bindings don't always start at 0 */
96 nir_foreach_variable_with_modes(var, b->shader, nir_var_mem_ssbo) {
97 first_idx = var->data.binding;
98 break;
99 }
100 assert(first_idx != UINT_MAX);
101 last_idx = first_idx + b->shader->info.num_ssbos;
102 } else {
103 /* skip 0 index if uniform_0 is one we created previously */
104 first_idx = !b->shader->info.first_ubo_is_default_ubo;
105 last_idx = first_idx + b->shader->info.num_ubos;
106 }
107
108 /* now create the composite dest with a bcsel chain based on the original value */
109 nir_ssa_def *new_dest = recursive_generate_bo_ssa_def(b, instr,
110 instr->src[block_idx].ssa,
111 first_idx, last_idx);
112
113 if (instr->intrinsic != nir_intrinsic_store_ssbo)
114 /* now use the composite dest in all cases where the original dest (from the dynamic index)
115 * was used and remove the dynamically-indexed load_*bo instruction
116 */
117 nir_ssa_def_rewrite_uses_after(&instr->dest.ssa, nir_src_for_ssa(new_dest), &instr->instr);
118
119 nir_instr_remove(&instr->instr);
120 return true;
121 }
122
123 bool
nir_lower_dynamic_bo_access(nir_shader * shader)124 nir_lower_dynamic_bo_access(nir_shader *shader)
125 {
126 bool progress = false;
127
128 nir_foreach_function(function, shader) {
129 if (function->impl) {
130 nir_builder builder;
131 nir_builder_init(&builder, function->impl);
132 nir_foreach_block(block, function->impl) {
133 nir_foreach_instr_safe(instr, block) {
134 if (instr->type == nir_instr_type_intrinsic)
135 progress |= lower_dynamic_bo_access_instr(
136 nir_instr_as_intrinsic(instr),
137 &builder);
138 }
139 }
140
141 nir_metadata_preserve(function->impl, nir_metadata_dominance);
142 }
143 }
144
145 return progress;
146 }
147