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/imports.h"
  42
  43 #include "brw_context.h"
  44 #include "brw_state.h"
  45 #include "intel_batchbuffer.h"
  46 #include "intel_reg.h"
  47
  48 /** Waits on the query object's BO and totals the results for this query */
  49 static void
  50 brw_queryobj_get_results(struct brw_query_object *query)
  51 {
  52    int i;
  53    uint64_t *results;
  54
  55    if (query->bo == NULL)
  56       return;
  57
  58    /* Map and count the pixels from the current query BO */
  59    dri_bo_map(query->bo, GL_FALSE);
  60    results = query->bo->virtual;
  61    for (i = query->first_index; i <= query->last_index; i++) {
  62       query->Base.Result += results[i * 2 + 1] - results[i * 2];
  63    }
  64    dri_bo_unmap(query->bo);
  65
  66    dri_bo_unreference(query->bo);
  67    query->bo = NULL;
  68 }
  69
  70 static struct gl_query_object *
  71 brw_new_query_object(GLcontext *ctx, GLuint id)
  72 {
  73    struct brw_query_object *query;
  74
  75    query = calloc(1, sizeof(struct brw_query_object));
  76
  77    query->Base.Id = id;
  78    query->Base.Result = 0;
  79    query->Base.Active = GL_FALSE;
  80    query->Base.Ready = GL_TRUE;
  81
  82    return &query->Base;
  83 }
  84
  85 static void
  86 brw_delete_query(GLcontext *ctx, struct gl_query_object *q)
  87 {
  88    struct brw_query_object *query = (struct brw_query_object *)q;
  89
  90    dri_bo_unreference(query->bo);
  91    free(query);
  92 }
  93
  94 static void
  95 brw_begin_query(GLcontext *ctx, struct gl_query_object *q)
  96 {
  97    struct brw_context *brw = brw_context(ctx);
  98    struct intel_context *intel = intel_context(ctx);
  99    struct brw_query_object *query = (struct brw_query_object *)q;
 100
 101    /* Reset our driver's tracking of query state. */
 102    dri_bo_unreference(query->bo);
 103    query->bo = NULL;
 104    query->first_index = -1;
 105    query->last_index = -1;
 106
 107    brw->query.obj = query;
 108    intel->stats_wm++;
 109 }
 110
 111 /**
 112  * Begin the ARB_occlusion_query query on a query object.
 113  */
 114 static void
 115 brw_end_query(GLcontext *ctx, struct gl_query_object *q)
 116 {
 117    struct brw_context *brw = brw_context(ctx);
 118    struct intel_context *intel = intel_context(ctx);
 119    struct brw_query_object *query = (struct brw_query_object *)q;
 120
 121    /* Flush the batchbuffer in case it has writes to our query BO.
 122     * Have later queries write to a new query BO so that further rendering
 123     * doesn't delay the collection of our results.
 124     */
 125    if (query->bo) {
 126       brw_emit_query_end(brw);
 127       intel_batchbuffer_flush(intel->batch);
 128
 129       dri_bo_unreference(brw->query.bo);
 130       brw->query.bo = NULL;
 131    }
 132
 133    brw->query.obj = NULL;
 134
 135    intel->stats_wm--;
 136 }
 137
 138 static void brw_wait_query(GLcontext *ctx, struct gl_query_object *q)
 139 {
 140    struct brw_query_object *query = (struct brw_query_object *)q;
 141
 142    brw_queryobj_get_results(query);
 143    query->Base.Ready = GL_TRUE;
 144 }
 145
 146 static void brw_check_query(GLcontext *ctx, struct gl_query_object *q)
 147 {
 148    struct brw_query_object *query = (struct brw_query_object *)q;
 149
 150    if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
 151       brw_queryobj_get_results(query);
 152       query->Base.Ready = GL_TRUE;
 153    }
 154 }
 155
 156 /** Called to set up the query BO and account for its aperture space */
 157 void
 158 brw_prepare_query_begin(struct brw_context *brw)
 159 {
 160    struct intel_context *intel = &brw->intel;
 161
 162    /* Skip if we're not doing any queries. */
 163    if (!brw->query.obj)
 164       return;
 165
 166    /* Get a new query BO if we're going to need it. */
 167    if (brw->query.bo == NULL ||
 168        brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) {
 169       dri_bo_unreference(brw->query.bo);
 170       brw->query.bo = NULL;
 171
 172       brw->query.bo = dri_bo_alloc(intel->bufmgr, "query", 4096, 1);
 173       brw->query.index = 0;
 174    }
 175
 176    brw_add_validated_bo(brw, brw->query.bo);
 177 }
 178
 179 /** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */
 180 void
 181 brw_emit_query_begin(struct brw_context *brw)
 182 {
 183    struct intel_context *intel = &brw->intel;
 184    struct brw_query_object *query = brw->query.obj;
 185
 186    /* Skip if we're not doing any queries, or we've emitted the start. */
 187    if (!query || brw->query.active)
 188       return;
 189
 190    BEGIN_BATCH(4);
 191    OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 192              PIPE_CONTROL_DEPTH_STALL |
 193              PIPE_CONTROL_WRITE_DEPTH_COUNT);
 194    /* This object could be mapped cacheable, but we don't have an exposed
 195     * mechanism to support that.  Since it's going uncached, tell GEM that
 196     * we're writing to it.  The usual clflush should be all that's required
 197     * to pick up the results.
 198     */
 199    OUT_RELOC(brw->query.bo,
 200              I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 201              PIPE_CONTROL_GLOBAL_GTT_WRITE |
 202              ((brw->query.index * 2) * sizeof(uint64_t)));
 203    OUT_BATCH(0);
 204    OUT_BATCH(0);
 205    ADVANCE_BATCH();
 206
 207    if (query->bo != brw->query.bo) {
 208       if (query->bo != NULL)
 209          brw_queryobj_get_results(query);
 210       dri_bo_reference(brw->query.bo);
 211       query->bo = brw->query.bo;
 212       query->first_index = brw->query.index;
 213    }
 214    query->last_index = brw->query.index;
 215    brw->query.active = GL_TRUE;
 216 }
 217
 218 /** Called at batchbuffer flush to get an ending PS_DEPTH_COUNT */
 219 void
 220 brw_emit_query_end(struct brw_context *brw)
 221 {
 222    struct intel_context *intel = &brw->intel;
 223
 224    if (!brw->query.active)
 225       return;
 226
 227    BEGIN_BATCH(4);
 228    OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 229              PIPE_CONTROL_DEPTH_STALL |
 230              PIPE_CONTROL_WRITE_DEPTH_COUNT);
 231    OUT_RELOC(brw->query.bo,
 232              I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 233              PIPE_CONTROL_GLOBAL_GTT_WRITE |
 234              ((brw->query.index * 2 + 1) * sizeof(uint64_t)));
 235    OUT_BATCH(0);
 236    OUT_BATCH(0);
 237    ADVANCE_BATCH();
 238
 239    brw->query.active = GL_FALSE;
 240    brw->query.index++;
 241 }
 242
 243 void brw_init_queryobj_functions(struct dd_function_table *functions)
 244 {
 245    functions->NewQueryObject = brw_new_query_object;
 246    functions->DeleteQuery = brw_delete_query;
 247    functions->BeginQuery = brw_begin_query;
 248    functions->EndQuery = brw_end_query;
 249    functions->CheckQuery = brw_check_query;
 250    functions->WaitQuery = brw_wait_query;
 251 }