src/gallium/drivers/freedreno/freedreno_query_hw.h

   1 /*
   2  * Copyright (C) 2014 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 #ifndef FREEDRENO_QUERY_HW_H_
  28 #define FREEDRENO_QUERY_HW_H_
  29
  30 #include "util/list.h"
  31
  32 #include "freedreno_query.h"
  33 #include "freedreno_context.h"
  34
  35
  36 /*
  37  * HW Queries:
  38  *
  39  * See: https://github.com/freedreno/freedreno/wiki/Queries#hardware-queries
  40  *
  41  * Hardware queries will be specific to gpu generation, but they need
  42  * some common infrastructure for triggering start/stop samples at
  43  * various points (for example, to exclude mem2gmem/gmem2mem or clear)
  44  * as well as per tile tracking.
  45  *
  46  * NOTE: in at least some cases hw writes sample values to memory addr
  47  * specified in some register.  So we don't really have the option to
  48  * just sample the same counter multiple times for multiple different
  49  * queries with the same query_type.  So we cache per sample provider
  50  * the most recent sample since the last draw.  This way multiple
  51  * sample periods for multiple queries can reference the same sample.
  52  *
  53  * fd_hw_sample_provider:
  54  *   - one per query type, registered/implemented by gpu generation
  55  *     specific code
  56  *   - can construct fd_hw_samples on demand
  57  *   - most recent sample (since last draw) cached so multiple
  58  *     different queries can ref the same sample
  59  *
  60  * fd_hw_sample:
  61  *   - abstracts one snapshot of counter value(s) across N tiles
  62  *   - backing object not allocated until submit time when number
  63  *     of samples and number of tiles is known
  64  *
  65  * fd_hw_sample_period:
  66  *   - consists of start and stop sample
  67  *   - a query accumulates a list of sample periods
  68  *   - the query result is the sum of the sample periods
  69  */
  70
  71 struct fd_hw_sample_provider {
  72         unsigned query_type;
  73
  74         /* stages applicable to the query type: */
  75         enum fd_render_stage active;
  76
  77         /* Optional hook for enabling a counter.  Guaranteed to happen
  78          * at least once before the first ->get_sample() in a batch.
  79          */
  80         void (*enable)(struct fd_context *ctx, struct fd_ringbuffer *ring);
  81
  82         /* when a new sample is required, emit appropriate cmdstream
  83          * and return a sample object:
  84          */
  85         struct fd_hw_sample *(*get_sample)(struct fd_batch *batch,
  86                         struct fd_ringbuffer *ring);
  87
  88         /* accumulate the results from specified sample period: */
  89         void (*accumulate_result)(struct fd_context *ctx,
  90                         const void *start, const void *end,
  91                         union pipe_query_result *result);
  92 };
  93
  94 struct fd_hw_sample {
  95         struct pipe_reference reference;  /* keep this first */
  96
  97         /* offset and size of the sample are know at the time the
  98          * sample is constructed.
  99          */
 100         uint32_t size;
 101         uint32_t offset;
 102
 103         /* backing object, offset/stride/etc are determined not when
 104          * the sample is constructed, but when the batch is submitted.
 105          * This way we can defer allocation until total # of requested
 106          * samples, and total # of tiles, is known.
 107          */
 108         struct pipe_resource *prsc;
 109         uint32_t num_tiles;
 110         uint32_t tile_stride;
 111 };
 112
 113 struct fd_hw_sample_period;
 114
 115 struct fd_hw_query {
 116         struct fd_query base;
 117
 118         const struct fd_hw_sample_provider *provider;
 119
 120         /* list of fd_hw_sample_periods: */
 121         struct list_head periods;
 122
 123         /* if active and not paused, the current sample period (not
 124          * yet added to current_periods):
 125          */
 126         struct fd_hw_sample_period *period;
 127
 128         struct list_head list;   /* list-node in batch->active_queries */
 129
 130         int no_wait_cnt;         /* see fd_hw_get_query_result */
 131 };
 132
 133 static inline struct fd_hw_query *
 134 fd_hw_query(struct fd_query *q)
 135 {
 136         return (struct fd_hw_query *)q;
 137 }
 138
 139 struct fd_query * fd_hw_create_query(struct fd_context *ctx, unsigned query_type, unsigned index);
 140 /* helper for sample providers: */
 141 struct fd_hw_sample * fd_hw_sample_init(struct fd_batch *batch, uint32_t size);
 142 /* don't call directly, use fd_hw_sample_reference() */
 143 void __fd_hw_sample_destroy(struct fd_context *ctx, struct fd_hw_sample *samp);
 144 void fd_hw_query_prepare(struct fd_batch *batch, uint32_t num_tiles);
 145 void fd_hw_query_prepare_tile(struct fd_batch *batch, uint32_t n,
 146                 struct fd_ringbuffer *ring);
 147 void fd_hw_query_set_stage(struct fd_batch *batch, enum fd_render_stage stage);
 148 void fd_hw_query_enable(struct fd_batch *batch, struct fd_ringbuffer *ring);
 149 void fd_hw_query_register_provider(struct pipe_context *pctx,
 150                 const struct fd_hw_sample_provider *provider);
 151 void fd_hw_query_init(struct pipe_context *pctx);
 152 void fd_hw_query_fini(struct pipe_context *pctx);
 153
 154 static inline void
 155 fd_hw_sample_reference(struct fd_context *ctx,
 156                 struct fd_hw_sample **ptr, struct fd_hw_sample *samp)
 157 {
 158         struct fd_hw_sample *old_samp = *ptr;
 159
 160         if (pipe_reference(&(*ptr)->reference, &samp->reference))
 161                 __fd_hw_sample_destroy(ctx, old_samp);
 162         *ptr = samp;
 163 }
 164
 165 #endif /* FREEDRENO_QUERY_HW_H_ */