radeonsi: use shader_info::cs::local_size_variable to clean up some code
[mesa.git] / src / gallium / drivers / freedreno / freedreno_query_acc.c
1 /*
2 * Copyright (C) 2017 Rob Clark <robclark@freedesktop.org>
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Rob Clark <robclark@freedesktop.org>
25 */
26
27 #include "util/u_memory.h"
28 #include "util/u_inlines.h"
29
30 #include "freedreno_query_acc.h"
31 #include "freedreno_context.h"
32 #include "freedreno_resource.h"
33 #include "freedreno_util.h"
34
35 static void
36 fd_acc_destroy_query(struct fd_context *ctx, struct fd_query *q)
37 {
38 struct fd_acc_query *aq = fd_acc_query(q);
39
40 DBG("%p", q);
41
42 pipe_resource_reference(&aq->prsc, NULL);
43 list_del(&aq->node);
44
45 free(aq->query_data);
46 free(aq);
47 }
48
49 static void
50 realloc_query_bo(struct fd_context *ctx, struct fd_acc_query *aq)
51 {
52 struct fd_resource *rsc;
53 void *map;
54
55 pipe_resource_reference(&aq->prsc, NULL);
56
57 aq->prsc = pipe_buffer_create(&ctx->screen->base,
58 PIPE_BIND_QUERY_BUFFER, 0, 0x1000);
59
60 /* don't assume the buffer is zero-initialized: */
61 rsc = fd_resource(aq->prsc);
62
63 fd_bo_cpu_prep(rsc->bo, ctx->pipe, DRM_FREEDRENO_PREP_WRITE);
64
65 map = fd_bo_map(rsc->bo);
66 memset(map, 0, aq->size);
67 fd_bo_cpu_fini(rsc->bo);
68 }
69
70 static void
71 fd_acc_query_pause(struct fd_acc_query *aq)
72 {
73 const struct fd_acc_sample_provider *p = aq->provider;
74
75 if (!aq->batch)
76 return;
77
78 p->pause(aq, aq->batch);
79 aq->batch = NULL;
80 }
81
82 static void
83 fd_acc_query_resume(struct fd_acc_query *aq, struct fd_batch *batch)
84 {
85 const struct fd_acc_sample_provider *p = aq->provider;
86
87 aq->batch = batch;
88 p->resume(aq, aq->batch);
89
90 fd_screen_lock(batch->ctx->screen);
91 fd_batch_resource_write(batch, fd_resource(aq->prsc));
92 fd_screen_unlock(batch->ctx->screen);
93 }
94
95 static void
96 fd_acc_begin_query(struct fd_context *ctx, struct fd_query *q)
97 {
98 struct fd_acc_query *aq = fd_acc_query(q);
99
100 DBG("%p", q);
101
102 /* ->begin_query() discards previous results, so realloc bo: */
103 realloc_query_bo(ctx, aq);
104
105 /* Signal that we need to update the active queries on the next draw */
106 ctx->update_active_queries = true;
107
108 /* add to active list: */
109 assert(list_is_empty(&aq->node));
110 list_addtail(&aq->node, &ctx->acc_active_queries);
111
112 /* TIMESTAMP/GPU_FINISHED and don't do normal bracketing at draw time, we
113 * need to just emit the capture at this moment.
114 */
115 if (skip_begin_query(q->type))
116 fd_acc_query_resume(aq, fd_context_batch(ctx));
117 }
118
119 static void
120 fd_acc_end_query(struct fd_context *ctx, struct fd_query *q)
121 {
122 struct fd_acc_query *aq = fd_acc_query(q);
123
124 DBG("%p", q);
125
126 fd_acc_query_pause(aq);
127
128 /* remove from active list: */
129 list_delinit(&aq->node);
130 }
131
132 static bool
133 fd_acc_get_query_result(struct fd_context *ctx, struct fd_query *q,
134 bool wait, union pipe_query_result *result)
135 {
136 struct fd_acc_query *aq = fd_acc_query(q);
137 const struct fd_acc_sample_provider *p = aq->provider;
138 struct fd_resource *rsc = fd_resource(aq->prsc);
139
140 DBG("%p: wait=%d", q, wait);
141
142 assert(list_is_empty(&aq->node));
143
144 /* if !wait, then check the last sample (the one most likely to
145 * not be ready yet) and bail if it is not ready:
146 */
147 if (!wait) {
148 int ret;
149
150 if (pending(rsc, false)) {
151 /* piglit spec@arb_occlusion_query@occlusion_query_conform
152 * test, and silly apps perhaps, get stuck in a loop trying
153 * to get query result forever with wait==false.. we don't
154 * wait to flush unnecessarily but we also don't want to
155 * spin forever:
156 */
157 if (aq->no_wait_cnt++ > 5)
158 fd_batch_flush(rsc->write_batch);
159 return false;
160 }
161
162 ret = fd_bo_cpu_prep(rsc->bo, ctx->pipe,
163 DRM_FREEDRENO_PREP_READ | DRM_FREEDRENO_PREP_NOSYNC);
164 if (ret)
165 return false;
166
167 fd_bo_cpu_fini(rsc->bo);
168 }
169
170 if (rsc->write_batch)
171 fd_batch_flush(rsc->write_batch);
172
173 /* get the result: */
174 fd_bo_cpu_prep(rsc->bo, ctx->pipe, DRM_FREEDRENO_PREP_READ);
175
176 void *ptr = fd_bo_map(rsc->bo);
177 p->result(aq, ptr, result);
178 fd_bo_cpu_fini(rsc->bo);
179
180 return true;
181 }
182
183 static const struct fd_query_funcs acc_query_funcs = {
184 .destroy_query = fd_acc_destroy_query,
185 .begin_query = fd_acc_begin_query,
186 .end_query = fd_acc_end_query,
187 .get_query_result = fd_acc_get_query_result,
188 };
189
190 struct fd_query *
191 fd_acc_create_query2(struct fd_context *ctx, unsigned query_type,
192 unsigned index, const struct fd_acc_sample_provider *provider)
193 {
194 struct fd_acc_query *aq;
195 struct fd_query *q;
196
197 aq = CALLOC_STRUCT(fd_acc_query);
198 if (!aq)
199 return NULL;
200
201 DBG("%p: query_type=%u", aq, query_type);
202
203 aq->provider = provider;
204 aq->size = provider->size;
205
206 list_inithead(&aq->node);
207
208 q = &aq->base;
209 q->funcs = &acc_query_funcs;
210 q->type = query_type;
211 q->index = index;
212
213 return q;
214 }
215
216 struct fd_query *
217 fd_acc_create_query(struct fd_context *ctx, unsigned query_type,
218 unsigned index)
219 {
220 int idx = pidx(query_type);
221
222 if ((idx < 0) || !ctx->acc_sample_providers[idx])
223 return NULL;
224
225 return fd_acc_create_query2(ctx, query_type, index,
226 ctx->acc_sample_providers[idx]);
227 }
228
229 /* Called at clear/draw/blit time to enable/disable the appropriate queries in
230 * the batch (and transfer active querying between batches in the case of
231 * batch reordering).
232 */
233 void
234 fd_acc_query_set_stage(struct fd_batch *batch, enum fd_render_stage stage)
235 {
236 struct fd_context *ctx = batch->ctx;
237
238 if (stage != batch->stage || ctx->update_active_queries) {
239 struct fd_acc_query *aq;
240 LIST_FOR_EACH_ENTRY(aq, &ctx->acc_active_queries, node) {
241 bool batch_change = aq->batch != batch;
242 bool was_active = aq->batch != NULL;
243 bool now_active = stage != FD_STAGE_NULL &&
244 (ctx->active_queries || aq->provider->always);
245
246 if (was_active && (!now_active || batch_change))
247 fd_acc_query_pause(aq);
248 if (now_active && (!was_active || batch_change))
249 fd_acc_query_resume(aq, batch);
250 }
251 }
252
253 ctx->update_active_queries = false;
254 }
255
256 void
257 fd_acc_query_register_provider(struct pipe_context *pctx,
258 const struct fd_acc_sample_provider *provider)
259 {
260 struct fd_context *ctx = fd_context(pctx);
261 int idx = pidx(provider->query_type);
262
263 assert((0 <= idx) && (idx < MAX_HW_SAMPLE_PROVIDERS));
264 assert(!ctx->acc_sample_providers[idx]);
265
266 ctx->acc_sample_providers[idx] = provider;
267 }