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 "main/macros.h"
29
30 namespace {
31 /*
32 * Atomic counter uniform as seen by the program.
33 */
34 struct active_atomic_counter_uniform {
35 unsigned uniform_loc;
36 ir_variable *var;
37 };
38
39 /*
40 * Atomic counter buffer referenced by the program. There is a one
41 * to one correspondence between these and the objects that can be
42 * queried using glGetActiveAtomicCounterBufferiv().
43 */
44 struct active_atomic_buffer {
active_atomic_buffer__anon54a60dec0111::active_atomic_buffer45 active_atomic_buffer()
46 : uniforms(0), num_uniforms(0), stage_counter_references(), size(0)
47 {}
48
~active_atomic_buffer__anon54a60dec0111::active_atomic_buffer49 ~active_atomic_buffer()
50 {
51 free(uniforms);
52 }
53
push_back__anon54a60dec0111::active_atomic_buffer54 void push_back(unsigned uniform_loc, ir_variable *var)
55 {
56 active_atomic_counter_uniform *new_uniforms;
57
58 new_uniforms = (active_atomic_counter_uniform *)
59 realloc(uniforms, sizeof(active_atomic_counter_uniform) *
60 (num_uniforms + 1));
61
62 if (new_uniforms == NULL) {
63 _mesa_error_no_memory(__func__);
64 return;
65 }
66
67 uniforms = new_uniforms;
68 uniforms[num_uniforms].uniform_loc = uniform_loc;
69 uniforms[num_uniforms].var = var;
70 num_uniforms++;
71 }
72
73 active_atomic_counter_uniform *uniforms;
74 unsigned num_uniforms;
75 unsigned stage_counter_references[MESA_SHADER_STAGES];
76 unsigned size;
77 };
78
79 int
cmp_actives(const void * a,const void * b)80 cmp_actives(const void *a, const void *b)
81 {
82 const active_atomic_counter_uniform *const first = (active_atomic_counter_uniform *) a;
83 const active_atomic_counter_uniform *const second = (active_atomic_counter_uniform *) b;
84
85 return int(first->var->data.offset) - int(second->var->data.offset);
86 }
87
88 bool
check_atomic_counters_overlap(const ir_variable * x,const ir_variable * y)89 check_atomic_counters_overlap(const ir_variable *x, const ir_variable *y)
90 {
91 return ((x->data.offset >= y->data.offset &&
92 x->data.offset < y->data.offset + y->type->atomic_size()) ||
93 (y->data.offset >= x->data.offset &&
94 y->data.offset < x->data.offset + x->type->atomic_size()));
95 }
96
97 void
process_atomic_variable(const glsl_type * t,struct gl_shader_program * prog,unsigned * uniform_loc,ir_variable * var,active_atomic_buffer * const buffers,unsigned * num_buffers,int * offset,const unsigned shader_stage)98 process_atomic_variable(const glsl_type *t, struct gl_shader_program *prog,
99 unsigned *uniform_loc, ir_variable *var,
100 active_atomic_buffer *const buffers,
101 unsigned *num_buffers, int *offset,
102 const unsigned shader_stage)
103 {
104 /* FIXME: Arrays of arrays get counted separately. For example:
105 * x1[3][3][2] = 9 uniforms, 18 atomic counters
106 * x2[3][2] = 3 uniforms, 6 atomic counters
107 * x3[2] = 1 uniform, 2 atomic counters
108 *
109 * However this code marks all the counters as active even when they
110 * might not be used.
111 */
112 if (t->is_array() && t->fields.array->is_array()) {
113 for (unsigned i = 0; i < t->length; i++) {
114 process_atomic_variable(t->fields.array, prog, uniform_loc,
115 var, buffers, num_buffers, offset,
116 shader_stage);
117 }
118 } else {
119 active_atomic_buffer *buf = &buffers[var->data.binding];
120 gl_uniform_storage *const storage =
121 &prog->data->UniformStorage[*uniform_loc];
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(*uniform_loc, var);
130
131 /* When checking for atomic counters we should count every member in
132 * an array as an atomic counter reference.
133 */
134 if (t->is_array())
135 buf->stage_counter_references[shader_stage] += t->length;
136 else
137 buf->stage_counter_references[shader_stage]++;
138 buf->size = MAX2(buf->size, *offset + t->atomic_size());
139
140 storage->offset = *offset;
141 *offset += t->atomic_size();
142
143 (*uniform_loc)++;
144 }
145 }
146
147 active_atomic_buffer *
find_active_atomic_counters(struct gl_context * ctx,struct gl_shader_program * prog,unsigned * num_buffers)148 find_active_atomic_counters(struct gl_context *ctx,
149 struct gl_shader_program *prog,
150 unsigned *num_buffers)
151 {
152 active_atomic_buffer *const buffers =
153 new active_atomic_buffer[ctx->Const.MaxAtomicBufferBindings];
154
155 *num_buffers = 0;
156
157 for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
158 struct gl_linked_shader *sh = prog->_LinkedShaders[i];
159 if (sh == NULL)
160 continue;
161
162 foreach_in_list(ir_instruction, node, sh->ir) {
163 ir_variable *var = node->as_variable();
164
165 if (var && var->type->contains_atomic()) {
166 int offset = var->data.offset;
167 unsigned uniform_loc = var->data.location;
168 process_atomic_variable(var->type, prog, &uniform_loc,
169 var, buffers, num_buffers, &offset, i);
170 }
171 }
172 }
173
174 for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
175 if (buffers[i].size == 0)
176 continue;
177
178 qsort(buffers[i].uniforms, buffers[i].num_uniforms,
179 sizeof(active_atomic_counter_uniform),
180 cmp_actives);
181
182 for (unsigned j = 1; j < buffers[i].num_uniforms; j++) {
183 /* If an overlapping counter found, it must be a reference to the
184 * same counter from a different shader stage.
185 */
186 if (check_atomic_counters_overlap(buffers[i].uniforms[j-1].var,
187 buffers[i].uniforms[j].var)
188 && strcmp(buffers[i].uniforms[j-1].var->name,
189 buffers[i].uniforms[j].var->name) != 0) {
190 linker_error(prog, "Atomic counter %s declared at offset %d "
191 "which is already in use.",
192 buffers[i].uniforms[j].var->name,
193 buffers[i].uniforms[j].var->data.offset);
194 }
195 }
196 }
197 return buffers;
198 }
199 }
200
201 void
link_assign_atomic_counter_resources(struct gl_context * ctx,struct gl_shader_program * prog)202 link_assign_atomic_counter_resources(struct gl_context *ctx,
203 struct gl_shader_program *prog)
204 {
205 unsigned num_buffers;
206 unsigned num_atomic_buffers[MESA_SHADER_STAGES] = {};
207 active_atomic_buffer *abs =
208 find_active_atomic_counters(ctx, prog, &num_buffers);
209
210 prog->data->AtomicBuffers = rzalloc_array(prog, gl_active_atomic_buffer,
211 num_buffers);
212 prog->data->NumAtomicBuffers = num_buffers;
213
214 unsigned i = 0;
215 for (unsigned binding = 0;
216 binding < ctx->Const.MaxAtomicBufferBindings;
217 binding++) {
218
219 /* If the binding was not used, skip.
220 */
221 if (abs[binding].size == 0)
222 continue;
223
224 active_atomic_buffer &ab = abs[binding];
225 gl_active_atomic_buffer &mab = prog->data->AtomicBuffers[i];
226
227 /* Assign buffer-specific fields. */
228 mab.Binding = binding;
229 mab.MinimumSize = ab.size;
230 mab.Uniforms = rzalloc_array(prog->data->AtomicBuffers, GLuint,
231 ab.num_uniforms);
232 mab.NumUniforms = ab.num_uniforms;
233
234 /* Assign counter-specific fields. */
235 for (unsigned j = 0; j < ab.num_uniforms; j++) {
236 ir_variable *const var = ab.uniforms[j].var;
237 gl_uniform_storage *const storage =
238 &prog->data->UniformStorage[ab.uniforms[j].uniform_loc];
239
240 mab.Uniforms[j] = ab.uniforms[j].uniform_loc;
241 if (!var->data.explicit_binding)
242 var->data.binding = i;
243
244 storage->atomic_buffer_index = i;
245 storage->offset = var->data.offset;
246 storage->array_stride = (var->type->is_array() ?
247 var->type->without_array()->atomic_size() : 0);
248 if (!var->type->is_matrix())
249 storage->matrix_stride = 0;
250 }
251
252 /* Assign stage-specific fields. */
253 for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
254 if (ab.stage_counter_references[j]) {
255 mab.StageReferences[j] = GL_TRUE;
256 num_atomic_buffers[j]++;
257 } else {
258 mab.StageReferences[j] = GL_FALSE;
259 }
260 }
261
262 i++;
263 }
264
265 /* Store a list pointers to atomic buffers per stage and store the index
266 * to the intra-stage buffer list in uniform storage.
267 */
268 for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
269 if (prog->_LinkedShaders[j] && num_atomic_buffers[j] > 0) {
270 struct gl_program *gl_prog = prog->_LinkedShaders[j]->Program;
271 gl_prog->info.num_abos = num_atomic_buffers[j];
272 gl_prog->sh.AtomicBuffers =
273 rzalloc_array(prog, gl_active_atomic_buffer *,
274 num_atomic_buffers[j]);
275
276 unsigned intra_stage_idx = 0;
277 for (unsigned i = 0; i < num_buffers; i++) {
278 struct gl_active_atomic_buffer *atomic_buffer =
279 &prog->data->AtomicBuffers[i];
280 if (atomic_buffer->StageReferences[j]) {
281 gl_prog->sh.AtomicBuffers[intra_stage_idx] = atomic_buffer;
282
283 for (unsigned u = 0; u < atomic_buffer->NumUniforms; u++) {
284 prog->data->UniformStorage[atomic_buffer->Uniforms[u]].opaque[j].index =
285 intra_stage_idx;
286 prog->data->UniformStorage[atomic_buffer->Uniforms[u]].opaque[j].active =
287 true;
288 }
289
290 intra_stage_idx++;
291 }
292 }
293 }
294 }
295
296 delete [] abs;
297 assert(i == num_buffers);
298 }
299
300 void
link_check_atomic_counter_resources(struct gl_context * ctx,struct gl_shader_program * prog)301 link_check_atomic_counter_resources(struct gl_context *ctx,
302 struct gl_shader_program *prog)
303 {
304 unsigned num_buffers;
305 active_atomic_buffer *const abs =
306 find_active_atomic_counters(ctx, prog, &num_buffers);
307 unsigned atomic_counters[MESA_SHADER_STAGES] = {};
308 unsigned atomic_buffers[MESA_SHADER_STAGES] = {};
309 unsigned total_atomic_counters = 0;
310 unsigned total_atomic_buffers = 0;
311
312 /* Sum the required resources. Note that this counts buffers and
313 * counters referenced by several shader stages multiple times
314 * against the combined limit -- That's the behavior the spec
315 * requires.
316 */
317 for (unsigned i = 0; i < ctx->Const.MaxAtomicBufferBindings; i++) {
318 if (abs[i].size == 0)
319 continue;
320
321 for (unsigned j = 0; j < MESA_SHADER_STAGES; ++j) {
322 const unsigned n = abs[i].stage_counter_references[j];
323
324 if (n) {
325 atomic_counters[j] += n;
326 total_atomic_counters += n;
327 atomic_buffers[j]++;
328 total_atomic_buffers++;
329 }
330 }
331 }
332
333 /* Check that they are within the supported limits. */
334 for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
335 if (atomic_counters[i] > ctx->Const.Program[i].MaxAtomicCounters)
336 linker_error(prog, "Too many %s shader atomic counters",
337 _mesa_shader_stage_to_string(i));
338
339 if (atomic_buffers[i] > ctx->Const.Program[i].MaxAtomicBuffers)
340 linker_error(prog, "Too many %s shader atomic counter buffers",
341 _mesa_shader_stage_to_string(i));
342 }
343
344 if (total_atomic_counters > ctx->Const.MaxCombinedAtomicCounters)
345 linker_error(prog, "Too many combined atomic counters");
346
347 if (total_atomic_buffers > ctx->Const.MaxCombinedAtomicBuffers)
348 linker_error(prog, "Too many combined atomic buffers");
349
350 delete [] abs;
351 }
352