1 /*
2 * Copyright (C) 2018-2020 Collabora, Ltd.
3 * Copyright (C) 2019-2020 Icecream95
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24
25 #include "pan_ir.h"
26 #include "compiler/nir/nir_builder.h"
27
28 /* Midgard can write all of color, depth and stencil in a single writeout
29 * operation, so we merge depth/stencil stores with color stores.
30 * If there are no color stores, we add a write to the "depth RT".
31 *
32 * For Bifrost, we want these combined so we can properly order
33 * +ZS_EMIT with respect to +ATEST and +BLEND, as well as combining
34 * depth/stencil stores into a single +ZS_EMIT op.
35 */
36 bool
pan_nir_lower_zs_store(nir_shader * nir)37 pan_nir_lower_zs_store(nir_shader *nir)
38 {
39 if (nir->info.stage != MESA_SHADER_FRAGMENT)
40 return false;
41
42 nir_variable *z_var = NULL, *s_var = NULL;
43
44 nir_foreach_shader_out_variable(var, nir) {
45 if (var->data.location == FRAG_RESULT_DEPTH)
46 z_var = var;
47 else if (var->data.location == FRAG_RESULT_STENCIL)
48 s_var = var;
49 }
50
51 if (!z_var && !s_var)
52 return false;
53
54 bool progress = false;
55
56 nir_foreach_function(function, nir) {
57 if (!function->impl) continue;
58
59 nir_intrinsic_instr *z_store = NULL, *s_store = NULL;
60
61 nir_foreach_block(block, function->impl) {
62 nir_foreach_instr_safe(instr, block) {
63 if (instr->type != nir_instr_type_intrinsic)
64 continue;
65
66 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
67 if (intr->intrinsic != nir_intrinsic_store_output)
68 continue;
69
70 if (z_var && nir_intrinsic_base(intr) == z_var->data.driver_location) {
71 assert(!z_store);
72 z_store = intr;
73 }
74
75 if (s_var && nir_intrinsic_base(intr) == s_var->data.driver_location) {
76 assert(!s_store);
77 s_store = intr;
78 }
79 }
80 }
81
82 if (!z_store && !s_store) continue;
83
84 bool replaced = false;
85
86 nir_foreach_block(block, function->impl) {
87 nir_foreach_instr_safe(instr, block) {
88 if (instr->type != nir_instr_type_intrinsic)
89 continue;
90
91 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
92 if (intr->intrinsic != nir_intrinsic_store_output)
93 continue;
94
95 const nir_variable *var = nir_find_variable_with_driver_location(nir, nir_var_shader_out, nir_intrinsic_base(intr));
96 assert(var);
97
98 if (var->data.location < FRAG_RESULT_DATA0)
99 continue;
100
101 if (var->data.index)
102 continue;
103
104 assert(nir_src_is_const(intr->src[1]) && "no indirect outputs");
105
106 nir_builder b;
107 nir_builder_init(&b, function->impl);
108
109 assert(!z_store || z_store->instr.block == instr->block);
110 assert(!s_store || s_store->instr.block == instr->block);
111 b.cursor = nir_after_block_before_jump(instr->block);
112
113 nir_intrinsic_instr *combined_store;
114 combined_store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_combined_output_pan);
115
116 combined_store->num_components = intr->src[0].ssa->num_components;
117
118 nir_intrinsic_set_base(combined_store, nir_intrinsic_base(intr));
119 nir_intrinsic_set_src_type(combined_store, nir_intrinsic_src_type(intr));
120
121 unsigned writeout = PAN_WRITEOUT_C;
122 if (z_store)
123 writeout |= PAN_WRITEOUT_Z;
124 if (s_store)
125 writeout |= PAN_WRITEOUT_S;
126
127 nir_intrinsic_set_component(combined_store, writeout);
128
129 struct nir_ssa_def *zero = nir_imm_int(&b, 0);
130
131 struct nir_ssa_def *src[4] = {
132 intr->src[0].ssa,
133 intr->src[1].ssa,
134 z_store ? z_store->src[0].ssa : zero,
135 s_store ? s_store->src[0].ssa : zero,
136 };
137
138 for (int i = 0; i < 4; ++i)
139 combined_store->src[i] = nir_src_for_ssa(src[i]);
140
141 nir_builder_instr_insert(&b, &combined_store->instr);
142
143 nir_instr_remove(instr);
144
145 replaced = true;
146 }
147 }
148
149 /* Insert a store to the depth RT (0xff) if needed */
150 if (!replaced) {
151 nir_builder b;
152 nir_builder_init(&b, function->impl);
153
154 nir_block *block = NULL;
155 if (z_store && s_store)
156 assert(z_store->instr.block == s_store->instr.block);
157
158 if (z_store)
159 block = z_store->instr.block;
160 else
161 block = s_store->instr.block;
162
163 b.cursor = nir_after_block_before_jump(block);
164
165 nir_intrinsic_instr *combined_store;
166 combined_store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_combined_output_pan);
167
168 combined_store->num_components = 4;
169
170 unsigned base;
171 if (z_store)
172 base = nir_intrinsic_base(z_store);
173 else
174 base = nir_intrinsic_base(s_store);
175 nir_intrinsic_set_base(combined_store, base);
176 nir_intrinsic_set_src_type(combined_store, nir_type_float32);
177
178 unsigned writeout = 0;
179 if (z_store)
180 writeout |= PAN_WRITEOUT_Z;
181 if (s_store)
182 writeout |= PAN_WRITEOUT_S;
183
184 nir_intrinsic_set_component(combined_store, writeout);
185
186 struct nir_ssa_def *zero = nir_imm_int(&b, 0);
187
188 struct nir_ssa_def *src[4] = {
189 nir_imm_vec4(&b, 0, 0, 0, 0),
190 zero,
191 z_store ? z_store->src[0].ssa : zero,
192 s_store ? s_store->src[0].ssa : zero,
193 };
194
195 for (int i = 0; i < 4; ++i)
196 combined_store->src[i] = nir_src_for_ssa(src[i]);
197
198 nir_builder_instr_insert(&b, &combined_store->instr);
199 }
200
201 if (z_store)
202 nir_instr_remove(&z_store->instr);
203
204 if (s_store)
205 nir_instr_remove(&s_store->instr);
206
207 nir_metadata_preserve(function->impl, nir_metadata_block_index | nir_metadata_dominance);
208 progress = true;
209 }
210
211 return progress;
212 }
213
214 /* Real writeout stores, which break execution, need to be moved to after
215 * dual-source stores, which are just standard register writes. */
216 bool
pan_nir_reorder_writeout(nir_shader * nir)217 pan_nir_reorder_writeout(nir_shader *nir)
218 {
219 bool progress = false;
220
221 nir_foreach_function(function, nir) {
222 if (!function->impl) continue;
223
224 nir_foreach_block(block, function->impl) {
225 nir_instr *last_writeout = NULL;
226
227 nir_foreach_instr_reverse_safe(instr, block) {
228 if (instr->type != nir_instr_type_intrinsic)
229 continue;
230
231 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
232 if (intr->intrinsic != nir_intrinsic_store_output)
233 continue;
234
235 const nir_variable *var = nir_find_variable_with_driver_location(nir, nir_var_shader_out, nir_intrinsic_base(intr));
236
237 if (var->data.index) {
238 if (!last_writeout)
239 last_writeout = instr;
240 continue;
241 }
242
243 if (!last_writeout)
244 continue;
245
246 /* This is a real store, so move it to after dual-source stores */
247 exec_node_remove(&instr->node);
248 exec_node_insert_after(&last_writeout->node, &instr->node);
249
250 progress = true;
251 }
252 }
253 }
254
255 return progress;
256 }
257