src/mesa/drivers/dri/i965/brw_queryobj.c

   1 /*
   2  * Copyright © 2008 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  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 /** @file support for ARB_query_object
  29  *
  30  * ARB_query_object is implemented by using the PIPE_CONTROL command to stall
  31  * execution on the completion of previous depth tests, and write the
  32  * current PS_DEPTH_COUNT to a buffer object.
  33  *
  34  * We use before and after counts when drawing during a query so that
  35  * we don't pick up other clients' query data in ours.  To reduce overhead,
  36  * a single BO is used to record the query data for all active queries at
  37  * once.  This also gives us a simple bound on how much batchbuffer space is
  38  * required for handling queries, so that we can be sure that we won't
  39  * have to emit a batchbuffer without getting the ending PS_DEPTH_COUNT.
  40  */
  41 #include "main/simple_list.h"
  42 #include "main/imports.h"
  43
  44 #include "brw_context.h"
  45 #include "brw_state.h"
  46 #include "intel_batchbuffer.h"
  47 #include "intel_reg.h"
  48
  49 /** Waits on the query object's BO and totals the results for this query */
  50 static void
  51 brw_queryobj_get_results(struct brw_query_object *query)
  52 {
  53    int i;
  54    uint64_t *results;
  55
  56    if (query->bo == NULL)
  57       return;
  58
  59    /* Map and count the pixels from the current query BO */
  60    dri_bo_map(query->bo, GL_FALSE);
  61    results = query->bo->virtual;
  62    for (i = query->first_index; i <= query->last_index; i++) {
  63       query->Base.Result += results[i * 2 + 1] - results[i * 2];
  64    }
  65    dri_bo_unmap(query->bo);
  66
  67    dri_bo_unreference(query->bo);
  68    query->bo = NULL;
  69 }
  70
  71 static struct gl_query_object *
  72 brw_new_query_object(GLcontext *ctx, GLuint id)
  73 {
  74    struct brw_query_object *query;
  75
  76    query = _mesa_calloc(sizeof(struct brw_query_object));
  77
  78    query->Base.Id = id;
  79    query->Base.Result = 0;
  80    query->Base.Active = GL_FALSE;
  81    query->Base.Ready = GL_TRUE;
  82
  83    return &query->Base;
  84 }
  85
  86 static void
  87 brw_delete_query(GLcontext *ctx, struct gl_query_object *q)
  88 {
  89    struct brw_query_object *query = (struct brw_query_object *)q;
  90
  91    dri_bo_unreference(query->bo);
  92    _mesa_free(query);
  93 }
  94
  95 static void
  96 brw_begin_query(GLcontext *ctx, struct gl_query_object *q)
  97 {
  98    struct brw_context *brw = brw_context(ctx);
  99    struct intel_context *intel = intel_context(ctx);
 100    struct brw_query_object *query = (struct brw_query_object *)q;
 101
 102    /* Reset our driver's tracking of query state. */
 103    dri_bo_unreference(query->bo);
 104    query->bo = NULL;
 105    query->first_index = -1;
 106    query->last_index = -1;
 107
 108    insert_at_head(&brw->query.active_head, query);
 109    intel->stats_wm++;
 110 }
 111
 112 /**
 113  * Begin the ARB_occlusion_query query on a query object.
 114  */
 115 static void
 116 brw_end_query(GLcontext *ctx, struct gl_query_object *q)
 117 {
 118    struct brw_context *brw = brw_context(ctx);
 119    struct intel_context *intel = intel_context(ctx);
 120    struct brw_query_object *query = (struct brw_query_object *)q;
 121
 122    /* Flush the batchbuffer in case it has writes to our query BO.
 123     * Have later queries write to a new query BO so that further rendering
 124     * doesn't delay the collection of our results.
 125     */
 126    if (query->bo) {
 127       brw_emit_query_end(brw);
 128       intel_batchbuffer_flush(intel->batch);
 129
 130       dri_bo_unreference(brw->query.bo);
 131       brw->query.bo = NULL;
 132    }
 133
 134    remove_from_list(query);
 135
 136    intel->stats_wm--;
 137 }
 138
 139 static void brw_wait_query(GLcontext *ctx, struct gl_query_object *q)
 140 {
 141    struct brw_query_object *query = (struct brw_query_object *)q;
 142
 143    brw_queryobj_get_results(query);
 144    query->Base.Ready = GL_TRUE;
 145 }
 146
 147 static void brw_check_query(GLcontext *ctx, struct gl_query_object *q)
 148 {
 149    struct brw_query_object *query = (struct brw_query_object *)q;
 150
 151    if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
 152       brw_queryobj_get_results(query);
 153       query->Base.Ready = GL_TRUE;
 154    }
 155 }
 156
 157 /** Called to set up the query BO and account for its aperture space */
 158 void
 159 brw_prepare_query_begin(struct brw_context *brw)
 160 {
 161    struct intel_context *intel = &brw->intel;
 162
 163    /* Skip if we're not doing any queries. */
 164    if (is_empty_list(&brw->query.active_head))
 165       return;
 166
 167    /* Get a new query BO if we're going to need it. */
 168    if (brw->query.bo == NULL ||
 169        brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) {
 170       dri_bo_unreference(brw->query.bo);
 171       brw->query.bo = NULL;
 172
 173       brw->query.bo = dri_bo_alloc(intel->bufmgr, "query", 4096, 1);
 174       brw->query.index = 0;
 175    }
 176
 177    brw_add_validated_bo(brw, brw->query.bo);
 178 }
 179
 180 /** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */
 181 void
 182 brw_emit_query_begin(struct brw_context *brw)
 183 {
 184    struct intel_context *intel = &brw->intel;
 185    struct brw_query_object *query;
 186
 187    /* Skip if we're not doing any queries, or we've emitted the start. */
 188    if (brw->query.active || is_empty_list(&brw->query.active_head))
 189       return;
 190
 191    BEGIN_BATCH(4);
 192    OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 193              PIPE_CONTROL_DEPTH_STALL |
 194              PIPE_CONTROL_WRITE_DEPTH_COUNT);
 195    /* This object could be mapped cacheable, but we don't have an exposed
 196     * mechanism to support that.  Since it's going uncached, tell GEM that
 197     * we're writing to it.  The usual clflush should be all that's required
 198     * to pick up the results.
 199     */
 200    OUT_RELOC(brw->query.bo,
 201              I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 202              PIPE_CONTROL_GLOBAL_GTT_WRITE |
 203              ((brw->query.index * 2) * sizeof(uint64_t)));
 204    OUT_BATCH(0);
 205    OUT_BATCH(0);
 206    ADVANCE_BATCH();
 207
 208    foreach(query, &brw->query.active_head) {
 209       if (query->bo != brw->query.bo) {
 210          if (query->bo != NULL)
 211             brw_queryobj_get_results(query);
 212          dri_bo_reference(brw->query.bo);
 213          query->bo = brw->query.bo;
 214          query->first_index = brw->query.index;
 215       }
 216       query->last_index = brw->query.index;
 217    }
 218    brw->query.active = GL_TRUE;
 219 }
 220
 221 /** Called at batchbuffer flush to get an ending PS_DEPTH_COUNT */
 222 void
 223 brw_emit_query_end(struct brw_context *brw)
 224 {
 225    struct intel_context *intel = &brw->intel;
 226
 227    if (!brw->query.active)
 228       return;
 229
 230    BEGIN_BATCH(4);
 231    OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 232              PIPE_CONTROL_DEPTH_STALL |
 233              PIPE_CONTROL_WRITE_DEPTH_COUNT);
 234    OUT_RELOC(brw->query.bo,
 235              I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 236              PIPE_CONTROL_GLOBAL_GTT_WRITE |
 237              ((brw->query.index * 2 + 1) * sizeof(uint64_t)));
 238    OUT_BATCH(0);
 239    OUT_BATCH(0);
 240    ADVANCE_BATCH();
 241
 242    brw->query.active = GL_FALSE;
 243    brw->query.index++;
 244 }
 245
 246 void brw_init_queryobj_functions(struct dd_function_table *functions)
 247 {
 248    functions->NewQueryObject = brw_new_query_object;
 249    functions->DeleteQuery = brw_delete_query;
 250    functions->BeginQuery = brw_begin_query;
 251    functions->EndQuery = brw_end_query;
 252    functions->CheckQuery = brw_check_query;
 253    functions->WaitQuery = brw_wait_query;
 254 }