1 /*
2 * Copyright (C) 2018 Jonathan Marek <jonathan@marek.ca>
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 * Jonathan Marek <jonathan@marek.ca>
25 * Rob Clark <robclark@freedesktop.org>
26 */
27
28 /* NOTE: perfcntrs are 48-bits but we only have 32-bit accumulate (?)
29 * so we work with 32-bits only. we accumulate start/stop separately,
30 * which differs from a5xx but works with only accumulate (no add/neg)
31 */
32
33 #include "freedreno_query_acc.h"
34 #include "freedreno_resource.h"
35
36 #include "fd2_context.h"
37 #include "fd2_query.h"
38
39 struct PACKED fd2_query_sample {
40 uint32_t start;
41 uint32_t stop;
42 };
43
44 /* offset of a single field of an array of fd2_query_sample: */
45 #define query_sample_idx(aq, idx, field) \
46 fd_resource((aq)->prsc)->bo, \
47 (idx * sizeof(struct fd2_query_sample)) + \
48 offsetof(struct fd2_query_sample, field), \
49 0, 0
50
51 /* offset of a single field of fd2_query_sample: */
52 #define query_sample(aq, field) \
53 query_sample_idx(aq, 0, field)
54
55 /*
56 * Performance Counter (batch) queries:
57 *
58 * Only one of these is active at a time, per design of the gallium
59 * batch_query API design. On perfcntr query tracks N query_types,
60 * each of which has a 'fd_batch_query_entry' that maps it back to
61 * the associated group and counter.
62 */
63
64 struct fd_batch_query_entry {
65 uint8_t gid; /* group-id */
66 uint8_t cid; /* countable-id within the group */
67 };
68
69 struct fd_batch_query_data {
70 struct fd_screen *screen;
71 unsigned num_query_entries;
72 struct fd_batch_query_entry query_entries[];
73 };
74
75 static void
perfcntr_resume(struct fd_acc_query * aq,struct fd_batch * batch)76 perfcntr_resume(struct fd_acc_query *aq, struct fd_batch *batch)
77 {
78 struct fd_batch_query_data *data = aq->query_data;
79 struct fd_screen *screen = data->screen;
80 struct fd_ringbuffer *ring = batch->draw;
81
82 unsigned counters_per_group[screen->num_perfcntr_groups];
83 memset(counters_per_group, 0, sizeof(counters_per_group));
84
85 fd_wfi(batch, ring);
86
87 /* configure performance counters for the requested queries: */
88 for (unsigned i = 0; i < data->num_query_entries; i++) {
89 struct fd_batch_query_entry *entry = &data->query_entries[i];
90 const struct fd_perfcntr_group *g = &screen->perfcntr_groups[entry->gid];
91 unsigned counter_idx = counters_per_group[entry->gid]++;
92
93 debug_assert(counter_idx < g->num_counters);
94
95 OUT_PKT0(ring, g->counters[counter_idx].select_reg, 1);
96 OUT_RING(ring, g->countables[entry->cid].selector);
97 }
98
99 memset(counters_per_group, 0, sizeof(counters_per_group));
100
101 /* and snapshot the start values */
102 for (unsigned i = 0; i < data->num_query_entries; i++) {
103 struct fd_batch_query_entry *entry = &data->query_entries[i];
104 const struct fd_perfcntr_group *g = &screen->perfcntr_groups[entry->gid];
105 unsigned counter_idx = counters_per_group[entry->gid]++;
106 const struct fd_perfcntr_counter *counter = &g->counters[counter_idx];
107
108 OUT_PKT3(ring, CP_REG_TO_MEM, 2);
109 OUT_RING(ring, counter->counter_reg_lo | CP_REG_TO_MEM_0_ACCUMULATE);
110 OUT_RELOC(ring, query_sample_idx(aq, i, start));
111 }
112 }
113
114 static void
perfcntr_pause(struct fd_acc_query * aq,struct fd_batch * batch)115 perfcntr_pause(struct fd_acc_query *aq, struct fd_batch *batch)
116 {
117 struct fd_batch_query_data *data = aq->query_data;
118 struct fd_screen *screen = data->screen;
119 struct fd_ringbuffer *ring = batch->draw;
120
121 unsigned counters_per_group[screen->num_perfcntr_groups];
122 memset(counters_per_group, 0, sizeof(counters_per_group));
123
124 fd_wfi(batch, ring);
125
126 /* TODO do we need to bother to turn anything off? */
127
128 /* snapshot the end values: */
129 for (unsigned i = 0; i < data->num_query_entries; i++) {
130 struct fd_batch_query_entry *entry = &data->query_entries[i];
131 const struct fd_perfcntr_group *g = &screen->perfcntr_groups[entry->gid];
132 unsigned counter_idx = counters_per_group[entry->gid]++;
133 const struct fd_perfcntr_counter *counter = &g->counters[counter_idx];
134
135 OUT_PKT3(ring, CP_REG_TO_MEM, 2);
136 OUT_RING(ring, counter->counter_reg_lo | CP_REG_TO_MEM_0_ACCUMULATE);
137 OUT_RELOC(ring, query_sample_idx(aq, i, stop));
138 }
139 }
140
141 static void
perfcntr_accumulate_result(struct fd_acc_query * aq,void * buf,union pipe_query_result * result)142 perfcntr_accumulate_result(struct fd_acc_query *aq, void *buf,
143 union pipe_query_result *result)
144 {
145 struct fd_batch_query_data *data = aq->query_data;
146 struct fd2_query_sample *sp = buf;
147
148 for (unsigned i = 0; i < data->num_query_entries; i++)
149 result->batch[i].u64 = sp[i].stop - sp[i].start;
150 }
151
152 static const struct fd_acc_sample_provider perfcntr = {
153 .query_type = FD_QUERY_FIRST_PERFCNTR,
154 .always = true,
155 .resume = perfcntr_resume,
156 .pause = perfcntr_pause,
157 .result = perfcntr_accumulate_result,
158 };
159
160 static struct pipe_query *
fd2_create_batch_query(struct pipe_context * pctx,unsigned num_queries,unsigned * query_types)161 fd2_create_batch_query(struct pipe_context *pctx,
162 unsigned num_queries, unsigned *query_types)
163 {
164 struct fd_context *ctx = fd_context(pctx);
165 struct fd_screen *screen = ctx->screen;
166 struct fd_query *q;
167 struct fd_acc_query *aq;
168 struct fd_batch_query_data *data;
169
170 data = CALLOC_VARIANT_LENGTH_STRUCT(fd_batch_query_data,
171 num_queries * sizeof(data->query_entries[0]));
172
173 data->screen = screen;
174 data->num_query_entries = num_queries;
175
176 /* validate the requested query_types and ensure we don't try
177 * to request more query_types of a given group than we have
178 * counters:
179 */
180 unsigned counters_per_group[screen->num_perfcntr_groups];
181 memset(counters_per_group, 0, sizeof(counters_per_group));
182
183 for (unsigned i = 0; i < num_queries; i++) {
184 unsigned idx = query_types[i] - FD_QUERY_FIRST_PERFCNTR;
185
186 /* verify valid query_type, ie. is it actually a perfcntr? */
187 if ((query_types[i] < FD_QUERY_FIRST_PERFCNTR) ||
188 (idx >= screen->num_perfcntr_queries)) {
189 debug_printf("invalid batch query query_type: %u\n", query_types[i]);
190 goto error;
191 }
192
193 struct fd_batch_query_entry *entry = &data->query_entries[i];
194 struct pipe_driver_query_info *pq = &screen->perfcntr_queries[idx];
195
196 entry->gid = pq->group_id;
197
198 /* the perfcntr_queries[] table flattens all the countables
199 * for each group in series, ie:
200 *
201 * (G0,C0), .., (G0,Cn), (G1,C0), .., (G1,Cm), ...
202 *
203 * So to find the countable index just step back through the
204 * table to find the first entry with the same group-id.
205 */
206 while (pq > screen->perfcntr_queries) {
207 pq--;
208 if (pq->group_id == entry->gid)
209 entry->cid++;
210 }
211
212 if (counters_per_group[entry->gid] >=
213 screen->perfcntr_groups[entry->gid].num_counters) {
214 debug_printf("too many counters for group %u\n", entry->gid);
215 goto error;
216 }
217
218 counters_per_group[entry->gid]++;
219 }
220
221 q = fd_acc_create_query2(ctx, 0, 0, &perfcntr);
222 aq = fd_acc_query(q);
223
224 /* sample buffer size is based on # of queries: */
225 aq->size = num_queries * sizeof(struct fd2_query_sample);
226 aq->query_data = data;
227
228 return (struct pipe_query *)q;
229
230 error:
231 free(data);
232 return NULL;
233 }
234
235 void
fd2_query_context_init(struct pipe_context * pctx)236 fd2_query_context_init(struct pipe_context *pctx)
237 {
238 struct fd_context *ctx = fd_context(pctx);
239
240 ctx->create_query = fd_acc_create_query;
241 ctx->query_set_stage = fd_acc_query_set_stage;
242
243 pctx->create_batch_query = fd2_create_batch_query;
244 }
245