src/gallium/drivers/freedreno/freedreno_query_hw.h

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #ifndef FREEDRENO_QUERY_HW_H_
  30 #define FREEDRENO_QUERY_HW_H_
  31
  32 #include "util/list.h"
  33
  34 #include "freedreno_query.h"
  35 #include "freedreno_context.h"
  36
  37
  38 /*
  39  * HW Queries:
  40  *
  41  * See: https://github.com/freedreno/freedreno/wiki/Queries#hardware-queries
  42  *
  43  * Hardware queries will be specific to gpu generation, but they need
  44  * some common infrastructure for triggering start/stop samples at
  45  * various points (for example, to exclude mem2gmem/gmem2mem or clear)
  46  * as well as per tile tracking.
  47  *
  48  * NOTE: in at least some cases hw writes sample values to memory addr
  49  * specified in some register.  So we don't really have the option to
  50  * just sample the same counter multiple times for multiple different
  51  * queries with the same query_type.  So we cache per sample provider
  52  * the most recent sample since the last draw.  This way multiple
  53  * sample periods for multiple queries can reference the same sample.
  54  *
  55  * fd_hw_sample_provider:
  56  *   - one per query type, registered/implemented by gpu generation
  57  *     specific code
  58  *   - can construct fd_hw_samples on demand
  59  *   - most recent sample (since last draw) cached so multiple
  60  *     different queries can ref the same sample
  61  *
  62  * fd_hw_sample:
  63  *   - abstracts one snapshot of counter value(s) across N tiles
  64  *   - backing object not allocated until submit time when number
  65  *     of samples and number of tiles is known
  66  *
  67  * fd_hw_sample_period:
  68  *   - consists of start and stop sample
  69  *   - a query accumulates a list of sample periods
  70  *   - the query result is the sum of the sample periods
  71  */
  72
  73 struct fd_hw_sample_provider {
  74         unsigned query_type;
  75
  76         /* stages applicable to the query type: */
  77         enum fd_render_stage active;
  78
  79         /* Optional hook for enabling a counter.  Guaranteed to happen
  80          * at least once before the first ->get_sample() in a batch.
  81          */
  82         void (*enable)(struct fd_context *ctx, struct fd_ringbuffer *ring);
  83
  84         /* when a new sample is required, emit appropriate cmdstream
  85          * and return a sample object:
  86          */
  87         struct fd_hw_sample *(*get_sample)(struct fd_context *ctx,
  88                         struct fd_ringbuffer *ring);
  89
  90         /* accumulate the results from specified sample period: */
  91         void (*accumulate_result)(struct fd_context *ctx,
  92                         const void *start, const void *end,
  93                         union pipe_query_result *result);
  94 };
  95
  96 struct fd_hw_sample {
  97         struct pipe_reference reference;  /* keep this first */
  98
  99         /* offset and size of the sample are know at the time the
 100          * sample is constructed.
 101          */
 102         uint32_t size;
 103         uint32_t offset;
 104
 105         /* backing object, offset/stride/etc are determined not when
 106          * the sample is constructed, but when the batch is submitted.
 107          * This way we can defer allocation until total # of requested
 108          * samples, and total # of tiles, is known.
 109          */
 110         struct fd_bo *bo;
 111         uint32_t num_tiles;
 112         uint32_t tile_stride;
 113 };
 114
 115 struct fd_hw_sample_period;
 116
 117 struct fd_hw_query {
 118         struct fd_query base;
 119
 120         const struct fd_hw_sample_provider *provider;
 121
 122         /* list of fd_hw_sample_period in previous submits: */
 123         struct list_head periods;
 124
 125         /* list of fd_hw_sample_period's in current submit: */
 126         struct list_head current_periods;
 127
 128         /* if active and not paused, the current sample period (not
 129          * yet added to current_periods):
 130          */
 131         struct fd_hw_sample_period *period;
 132
 133         struct list_head list;  /* list-node in ctx->active_queries */
 134 };
 135
 136 static inline struct fd_hw_query *
 137 fd_hw_query(struct fd_query *q)
 138 {
 139         return (struct fd_hw_query *)q;
 140 }
 141
 142 struct fd_query * fd_hw_create_query(struct fd_context *ctx, unsigned query_type);
 143 /* helper for sample providers: */
 144 struct fd_hw_sample * fd_hw_sample_init(struct fd_context *ctx, uint32_t size);
 145 /* don't call directly, use fd_hw_sample_reference() */
 146 void __fd_hw_sample_destroy(struct fd_context *ctx, struct fd_hw_sample *samp);
 147 void fd_hw_query_prepare(struct fd_context *ctx, uint32_t num_tiles);
 148 void fd_hw_query_prepare_tile(struct fd_context *ctx, uint32_t n,
 149                 struct fd_ringbuffer *ring);
 150 void fd_hw_query_set_stage(struct fd_context *ctx,
 151                 struct fd_ringbuffer *ring, enum fd_render_stage stage);
 152 void fd_hw_query_enable(struct fd_context *ctx, struct fd_ringbuffer *ring);
 153 void fd_hw_query_register_provider(struct pipe_context *pctx,
 154                 const struct fd_hw_sample_provider *provider);
 155 void fd_hw_query_init(struct pipe_context *pctx);
 156 void fd_hw_query_fini(struct pipe_context *pctx);
 157
 158 static inline void
 159 fd_hw_sample_reference(struct fd_context *ctx,
 160                 struct fd_hw_sample **ptr, struct fd_hw_sample *samp)
 161 {
 162         struct fd_hw_sample *old_samp = *ptr;
 163
 164         if (pipe_reference(&(*ptr)->reference, &samp->reference))
 165                 __fd_hw_sample_destroy(ctx, old_samp);
 166         if (ptr)
 167                 *ptr = samp;
 168 }
 169
 170 #endif /* FREEDRENO_QUERY_HW_H_ */