src/mesa/drivers/dri/i965/brw_performance_query.h

   1 /*
   2  * Copyright © 2015 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 DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #ifndef BRW_PERFORMANCE_QUERY_H
  25 #define BRW_PERFORMANCE_QUERY_H
  26
  27 #include <stdint.h>
  28
  29 #include "brw_context.h"
  30
  31 struct gen_perf_query_info;
  32
  33 /*
  34  * When currently allocate only one page for pipeline statistics queries. Here
  35  * we derived the maximum number of counters for that amount.
  36  */
  37 #define STATS_BO_SIZE               4096
  38 #define STATS_BO_END_OFFSET_BYTES   (STATS_BO_SIZE / 2)
  39 #define MAX_STAT_COUNTERS           (STATS_BO_END_OFFSET_BYTES / 8)
  40
  41 /*
  42  * The largest OA formats we can use include:
  43  * For Haswell:
  44  *   1 timestamp, 45 A counters, 8 B counters and 8 C counters.
  45  * For Gen8+
  46  *   1 timestamp, 1 clock, 36 A counters, 8 B counters and 8 C counters
  47  */
  48 #define MAX_OA_REPORT_COUNTERS 62
  49
  50 /**
  51  * i965 representation of a performance query object.
  52  *
  53  * NB: We want to keep this structure relatively lean considering that
  54  * applications may expect to allocate enough objects to be able to
  55  * query around all draw calls in a frame.
  56  */
  57 struct brw_perf_query_object
  58 {
  59    struct gl_perf_query_object base;
  60
  61    const struct gen_perf_query_info *query;
  62
  63    /* See query->kind to know which state below is in use... */
  64    union {
  65       struct {
  66
  67          /**
  68           * BO containing OA counter snapshots at query Begin/End time.
  69           */
  70          struct brw_bo *bo;
  71
  72          /**
  73           * Address of mapped of @bo
  74           */
  75          void *map;
  76
  77          /**
  78           * The MI_REPORT_PERF_COUNT command lets us specify a unique
  79           * ID that will be reflected in the resulting OA report
  80           * that's written by the GPU. This is the ID we're expecting
  81           * in the begin report and the the end report should be
  82           * @begin_report_id + 1.
  83           */
  84          int begin_report_id;
  85
  86          /**
  87           * Reference the head of the brw->perfquery.sample_buffers
  88           * list at the time that the query started (so we only need
  89           * to look at nodes after this point when looking for samples
  90           * related to this query)
  91           *
  92           * (See struct brw_oa_sample_buf description for more details)
  93           */
  94          struct exec_node *samples_head;
  95
  96          /**
  97           * Storage for the final accumulated OA counters.
  98           */
  99          uint64_t accumulator[MAX_OA_REPORT_COUNTERS];
 100
 101          /**
 102           * Hw ID used by the context on which the query was running.
 103           */
 104          uint32_t hw_id;
 105
 106          /**
 107           * false while in the unaccumulated_elements list, and set to
 108           * true when the final, end MI_RPC snapshot has been
 109           * accumulated.
 110           */
 111          bool results_accumulated;
 112
 113          /**
 114           * Number of reports accumulated to produce the results.
 115           */
 116          uint32_t reports_accumulated;
 117
 118          /**
 119           * Frequency of the GT at begin and end of the query.
 120           */
 121          uint64_t gt_frequency[2];
 122
 123          /**
 124           * Frequency in the slices of the GT at the begin and end of the
 125           * query.
 126           */
 127          uint64_t slice_frequency[2];
 128
 129          /**
 130           * Frequency in the unslice of the GT at the begin and end of the
 131           * query.
 132           */
 133          uint64_t unslice_frequency[2];
 134       } oa;
 135
 136       struct {
 137          /**
 138           * BO containing starting and ending snapshots for the
 139           * statistics counters.
 140           */
 141          struct brw_bo *bo;
 142       } pipeline_stats;
 143    };
 144 };
 145
 146 int brw_perf_query_get_mdapi_oa_data(struct brw_context *brw,
 147                                      struct brw_perf_query_object *obj,
 148                                      size_t data_size,
 149                                      uint8_t *data);
 150 void brw_perf_query_register_mdapi_oa_query(struct brw_context *brw);
 151 void brw_perf_query_register_mdapi_statistic_query(struct brw_context *brw);
 152
 153 #endif /* BRW_PERFORMANCE_QUERY_H */