Merge remote-tracking branch 'mesa-public/master' into vulkan
[mesa.git] / src / glsl / link_atomics.cpp
1 /*
2 * Copyright © 2013 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
21 * DEALINGS IN THE SOFTWARE.
22 */
23
24 #include "glsl_parser_extras.h"
25 #include "ir.h"
26 #include "ir_uniform.h"
27 #include "linker.h"
28 #include "program/hash_table.h"
29 #include "main/macros.h"
30
31 namespace {
32 /*
33 * Atomic counter as seen by the program.
34 */
35 struct active_atomic_counter {
36 unsigned id;
37 ir_variable *var;
38 };
39
40 /*
41 * Atomic counter buffer referenced by the program. There is a one
42 * to one correspondence between these and the objects that can be
43 * queried using glGetActiveAtomicCounterBufferiv().
44 */
45 struct active_atomic_buffer {
46 active_atomic_buffer()
47 : counters(0), num_counters(0), stage_references(), size(0)
48 {}
49
50 ~active_atomic_buffer()
51 {
52 free(counters);
53 }
54
55 void push_back(unsigned id, ir_variable *var)
56 {
57 active_atomic_counter *new_counters;
58
59 new_counters = (active_atomic_counter *)
60 realloc(counters, sizeof(active_atomic_counter) *
61 (num_counters + 1));
62
63 if (new_counters == NULL) {
64 _mesa_error_no_memory(__func__);
65 return;
66 }
67
68 counters = new_counters;
69 counters[num_counters].id = id;
70 counters[num_counters].var = var;
71 num_counters++;
72 }
73
74 active_atomic_counter *counters;
75 unsigned num_counters;
76 unsigned stage_references[MESA_SHADER_STAGES];
77 unsigned size;
78 };
79
80 int
81 cmp_actives(const void *a, const void *b)
82 {
83 const active_atomic_counter *const first = (active_atomic_counter *) a;
84 const active_atomic_counter *const second = (active_atomic_counter *) b;
85
86 return int(first->var->data.atomic.offset) - int(second->var->data.atomic.offset);
87 }
88
89 bool
90 check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
91 {
92 return ((x->data.atomic.offset >= y->data.atomic.offset &&
93 x->data.atomic.offset < y->data.atomic.offset + y->type->atomic_size()) ||
94 (y->data.atomic.offset >= x->data.atomic.offset &&
95 y->data.atomic.offset < x->data.atomic.offset + x->type->atomic_size()));
96 }
97
98 active_atomic_buffer *
99 find_active_atomic_counters(struct gl_context *ctx,
100 struct gl_shader_program *prog,
101 unsigned *num_buffers)
102 {
103 active_atomic_buffer *const buffers =
104 new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
105
106 *num_buffers = 0;
107
108 for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
109 struct gl_shader *sh = prog->_LinkedShaders[i];
110 if (sh == NULL)
111 continue;
112
113 foreach_in_list(ir_instruction, node, sh->ir) {
114 ir_variable *var = node->as_variable();
115
116 if (var && var->type->contains_atomic()) {
117 unsigned id = 0;
118 bool found = prog->UniformHash->get(id, var->name);
119 assert(found);
120 (void) found;
121 active_atomic_buffer *buf = &buffers[var->data.binding];
122
123 /* If this is the first time the buffer is used, increment
124 * the counter of buffers used.
125 */
126 if (buf->size == 0)
127 (*num_buffers)++;
128
129 buf->push_back(id, var);
130
131 buf->stage_references[i]++;
132 buf->size = MAX2(buf->size, var->data.atomic.offset +
133 var->type->atomic_size());
134 }
135 }
136 }
137
138 for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
139 if (buffers[i].size == 0)
140 continue;
141
142 qsort(buffers[i].counters, buffers[i].num_counters,
143 sizeof(active_atomic_counter),
144 cmp_actives);
145
146 for (unsigned j = 1; j < buffers[i].num_counters; j++) {
147 /* If an overlapping counter found, it must be a reference to the
148 * same counter from a different shader stage.
149 */
150 if (check_atomic_counters_overlap(buffers[i].counters[j-1].var,
151 buffers[i].counters[j].var)
152 && strcmp(buffers[i].counters[j-1].var->name,
153 buffers[i].counters[j].var->name) != 0) {
154 linker_error(prog, "Atomic counter %s declared at offset %d "
155 "which is already in use.",
156 buffers[i].counters[j].var->name,
157 buffers[i].counters[j].var->data.atomic.offset);
158 }
159 }
160 }
161 return buffers;
162 }
163 }
164
165 void
166 link_assign_atomic_counter_resources(struct gl_context *ctx,
167 struct gl_shader_program *prog)
168 {
169 unsigned num_buffers;
170 active_atomic_buffer *abs =
171 find_active_atomic_counters(ctx, prog, &num_buffers);
172
173 prog->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
174 num_buffers);
175 prog->NumAtomicBuffers = num_buffers;
176
177 unsigned i = 0;
178 for (unsigned binding = 0;
179 binding < ctx->Const.MaxAtomicBufferBindings;
180 binding++) {
181
182 /* If the binding was not used, skip.
183 */
184 if (abs[binding].size == 0)
185 continue;
186
187 active_atomic_buffer &ab = abs[binding];
188 gl_active_atomic_buffer &mab = prog->AtomicBuffers[i];
189
190 /* Assign buffer-specific fields. */
191 mab.Binding = binding;
192 mab.MinimumSize = ab.size;
193 mab.Uniforms = rzalloc_array(prog->AtomicBuffers, GLuint,
194 ab.num_counters);
195 mab.NumUniforms = ab.num_counters;
196
197 /* Assign counter-specific fields. */
198 for (unsigned j = 0; j < ab.num_counters; j++) {
199 ir_variable *const var = ab.counters[j].var;
200 const unsigned id = ab.counters[j].id;
201 gl_uniform_storage *const storage = &prog->UniformStorage[id];
202
203 mab.Uniforms[j] = id;
204 if (!var->data.explicit_binding)
205 var->data.binding = i;
206
207 storage->atomic_buffer_index = i;
208 storage->offset = var->data.atomic.offset;
209 storage->array_stride = (var->type->is_array() ?
210 var->type->without_array()->atomic_size() : 0);
211 }
212
213 /* Assign stage-specific fields. */
214 for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j)
215 mab.StageReferences[j] =
216 (ab.stage_references[j] ? GL_TRUE : GL_FALSE);
217
218 i++;
219 }
220
221 delete [] abs;
222 assert(i == num_buffers);
223 }
224
225 void
226 link_check_atomic_counter_resources(struct gl_context *ctx,
227 struct gl_shader_program *prog)
228 {
229 unsigned num_buffers;
230 active_atomic_buffer *const abs =
231 find_active_atomic_counters(ctx, prog, &num_buffers);
232 unsigned atomic_counters[MESA_SHADER_STAGES] = {};
233 unsigned atomic_buffers[MESA_SHADER_STAGES] = {};
234 unsigned total_atomic_counters = 0;
235 unsigned total_atomic_buffers = 0;
236
237 /* Sum the required resources. Note that this counts buffers and
238 * counters referenced by several shader stages multiple times
239 * against the combined limit -- That's the behavior the spec
240 * requires.
241 */
242 for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
243 if (abs[i].size == 0)
244 continue;
245
246 for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
247 const unsigned n = abs[i].stage_references[j];
248
249 if (n) {
250 atomic_counters[j] += n;
251 total_atomic_counters += n;
252 atomic_buffers[j]++;
253 total_atomic_buffers++;
254 }
255 }
256 }
257
258 /* Check that they are within the supported limits. */
259 for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
260 if (atomic_counters[i] > ctx->Const.Program[i].MaxAtomicCounters)
261 linker_error(prog, "Too many %s shader atomic counters",
262 _mesa_shader_stage_to_string(i));
263
264 if (atomic_buffers[i] > ctx->Const.Program[i].MaxAtomicBuffers)
265 linker_error(prog, "Too many %s shader atomic counter buffers",
266 _mesa_shader_stage_to_string(i));
267 }
268
269 if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
270 linker_error(prog, "Too many combined atomic counters");
271
272 if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
273 linker_error(prog, "Too many combined atomic buffers");
274
275 delete [] abs;
276 }