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 gl_context *ctx,
  51                          struct brw_query_object *query)
  52 {
  53    struct intel_context *intel = intel_context(ctx);
  54
  55    int i;
  56    uint64_t *results;
  57
  58    if (query->bo == NULL)
  59       return;
  60
  61    drm_intel_bo_map(query->bo, false);
  62    results = query->bo->virtual;
  63    switch (query->Base.Target) {
  64    case GL_TIME_ELAPSED_EXT:
  65       if (intel->gen >= 6)
  66          query->Base.Result += 80 * (results[1] - results[0]);
  67       else
  68          query->Base.Result += 1000 * ((results[1] >> 32) - (results[0] >> 32));
  69       break;
  70
  71    case GL_SAMPLES_PASSED_ARB:
  72       /* Map and count the pixels from the current query BO */
  73       for (i = query->first_index; i <= query->last_index; i++) {
  74          query->Base.Result += results[i * 2 + 1] - results[i * 2];
  75       }
  76       break;
  77
  78    default:
  79       assert(!"Unrecognized query target in brw_queryobj_get_results()");
  80       break;
  81    }
  82    drm_intel_bo_unmap(query->bo);
  83
  84    drm_intel_bo_unreference(query->bo);
  85    query->bo = NULL;
  86 }
  87
  88 static struct gl_query_object *
  89 brw_new_query_object(struct gl_context *ctx, GLuint id)
  90 {
  91    struct brw_query_object *query;
  92
  93    query = calloc(1, sizeof(struct brw_query_object));
  94
  95    query->Base.Id = id;
  96    query->Base.Result = 0;
  97    query->Base.Active = false;
  98    query->Base.Ready = true;
  99
 100    return &query->Base;
 101 }
 102
 103 static void
 104 brw_delete_query(struct gl_context *ctx, struct gl_query_object *q)
 105 {
 106    struct brw_query_object *query = (struct brw_query_object *)q;
 107
 108    drm_intel_bo_unreference(query->bo);
 109    free(query);
 110 }
 111
 112 static void
 113 brw_begin_query(struct gl_context *ctx, struct gl_query_object *q)
 114 {
 115    struct brw_context *brw = brw_context(ctx);
 116    struct intel_context *intel = intel_context(ctx);
 117    struct brw_query_object *query = (struct brw_query_object *)q;
 118
 119    switch (query->Base.Target) {
 120    case GL_TIME_ELAPSED_EXT:
 121       drm_intel_bo_unreference(query->bo);
 122       query->bo = drm_intel_bo_alloc(intel->bufmgr, "timer query",
 123                                      4096, 4096);
 124
 125       if (intel->gen >= 6) {
 126           BEGIN_BATCH(4);
 127           OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 128           OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
 129           OUT_RELOC(query->bo,
 130                   I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 131                   PIPE_CONTROL_GLOBAL_GTT_WRITE |
 132                   0);
 133           OUT_BATCH(0);
 134           ADVANCE_BATCH();
 135
 136       } else {
 137           BEGIN_BATCH(4);
 138           OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 139                   PIPE_CONTROL_WRITE_TIMESTAMP);
 140           OUT_RELOC(query->bo,
 141                   I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 142                   PIPE_CONTROL_GLOBAL_GTT_WRITE |
 143                   0);
 144           OUT_BATCH(0);
 145           OUT_BATCH(0);
 146           ADVANCE_BATCH();
 147       }
 148       break;
 149
 150    case GL_SAMPLES_PASSED_ARB:
 151       /* Reset our driver's tracking of query state. */
 152       drm_intel_bo_unreference(query->bo);
 153       query->bo = NULL;
 154       query->first_index = -1;
 155       query->last_index = -1;
 156
 157       brw->query.obj = query;
 158       intel->stats_wm++;
 159       break;
 160
 161    default:
 162       assert(!"Unrecognized query target in brw_begin_query()");
 163       break;
 164    }
 165 }
 166
 167 /**
 168  * Begin the ARB_occlusion_query query on a query object.
 169  */
 170 static void
 171 brw_end_query(struct gl_context *ctx, struct gl_query_object *q)
 172 {
 173    struct brw_context *brw = brw_context(ctx);
 174    struct intel_context *intel = intel_context(ctx);
 175    struct brw_query_object *query = (struct brw_query_object *)q;
 176
 177    switch (query->Base.Target) {
 178    case GL_TIME_ELAPSED_EXT:
 179       if (intel->gen >= 6) {
 180           BEGIN_BATCH(4);
 181           OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 182           OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
 183           OUT_RELOC(query->bo,
 184                   I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 185                   PIPE_CONTROL_GLOBAL_GTT_WRITE |
 186                   8);
 187           OUT_BATCH(0);
 188           ADVANCE_BATCH();
 189
 190       } else {
 191           BEGIN_BATCH(4);
 192           OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 193                   PIPE_CONTROL_WRITE_TIMESTAMP);
 194           OUT_RELOC(query->bo,
 195                   I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 196                   PIPE_CONTROL_GLOBAL_GTT_WRITE |
 197                   8);
 198           OUT_BATCH(0);
 199           OUT_BATCH(0);
 200           ADVANCE_BATCH();
 201       }
 202
 203       intel_batchbuffer_flush(intel);
 204       break;
 205
 206    case GL_SAMPLES_PASSED_ARB:
 207       /* Flush the batchbuffer in case it has writes to our query BO.
 208        * Have later queries write to a new query BO so that further rendering
 209        * doesn't delay the collection of our results.
 210        */
 211       if (query->bo) {
 212          brw_emit_query_end(brw);
 213          intel_batchbuffer_flush(intel);
 214
 215          drm_intel_bo_unreference(brw->query.bo);
 216          brw->query.bo = NULL;
 217       }
 218
 219       brw->query.obj = NULL;
 220
 221       intel->stats_wm--;
 222       break;
 223
 224    default:
 225       assert(!"Unrecognized query target in brw_end_query()");
 226       break;
 227    }
 228 }
 229
 230 static void brw_wait_query(struct gl_context *ctx, struct gl_query_object *q)
 231 {
 232    struct brw_query_object *query = (struct brw_query_object *)q;
 233
 234    brw_queryobj_get_results(ctx, query);
 235    query->Base.Ready = true;
 236 }
 237
 238 static void brw_check_query(struct gl_context *ctx, struct gl_query_object *q)
 239 {
 240    struct brw_query_object *query = (struct brw_query_object *)q;
 241
 242    if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
 243       brw_queryobj_get_results(ctx, query);
 244       query->Base.Ready = true;
 245    }
 246 }
 247
 248 /** Called to set up the query BO and account for its aperture space */
 249 void
 250 brw_prepare_query_begin(struct brw_context *brw)
 251 {
 252    struct intel_context *intel = &brw->intel;
 253
 254    /* Skip if we're not doing any queries. */
 255    if (!brw->query.obj)
 256       return;
 257
 258    /* Get a new query BO if we're going to need it. */
 259    if (brw->query.bo == NULL ||
 260        brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) {
 261       drm_intel_bo_unreference(brw->query.bo);
 262       brw->query.bo = NULL;
 263
 264       brw->query.bo = drm_intel_bo_alloc(intel->bufmgr, "query", 4096, 1);
 265
 266       /* clear target buffer */
 267       drm_intel_bo_map(brw->query.bo, true);
 268       memset((char *)brw->query.bo->virtual, 0, 4096);
 269       drm_intel_bo_unmap(brw->query.bo);
 270
 271       brw->query.index = 0;
 272    }
 273 }
 274
 275 /** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */
 276 void
 277 brw_emit_query_begin(struct brw_context *brw)
 278 {
 279    struct intel_context *intel = &brw->intel;
 280    struct gl_context *ctx = &intel->ctx;
 281    struct brw_query_object *query = brw->query.obj;
 282
 283    /* Skip if we're not doing any queries, or we've emitted the start. */
 284    if (!query || brw->query.active)
 285       return;
 286
 287    if (intel->gen >= 6) {
 288        BEGIN_BATCH(8);
 289
 290        /* workaround: CS stall required before depth stall. */
 291        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 292        OUT_BATCH(PIPE_CONTROL_CS_STALL);
 293        OUT_BATCH(0); /* write address */
 294        OUT_BATCH(0); /* write data */
 295
 296        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 297        OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
 298                  PIPE_CONTROL_WRITE_DEPTH_COUNT);
 299        OUT_RELOC(brw->query.bo,
 300                  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 301                  PIPE_CONTROL_GLOBAL_GTT_WRITE |
 302                  ((brw->query.index * 2) * sizeof(uint64_t)));
 303        OUT_BATCH(0);
 304        ADVANCE_BATCH();
 305
 306    } else {
 307        BEGIN_BATCH(4);
 308        OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 309                PIPE_CONTROL_DEPTH_STALL |
 310                PIPE_CONTROL_WRITE_DEPTH_COUNT);
 311        /* This object could be mapped cacheable, but we don't have an exposed
 312         * mechanism to support that.  Since it's going uncached, tell GEM that
 313         * we're writing to it.  The usual clflush should be all that's required
 314         * to pick up the results.
 315         */
 316        OUT_RELOC(brw->query.bo,
 317                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 318                PIPE_CONTROL_GLOBAL_GTT_WRITE |
 319                ((brw->query.index * 2) * sizeof(uint64_t)));
 320        OUT_BATCH(0);
 321        OUT_BATCH(0);
 322        ADVANCE_BATCH();
 323    }
 324
 325    if (query->bo != brw->query.bo) {
 326       if (query->bo != NULL)
 327          brw_queryobj_get_results(ctx, query);
 328       drm_intel_bo_reference(brw->query.bo);
 329       query->bo = brw->query.bo;
 330       query->first_index = brw->query.index;
 331    }
 332    query->last_index = brw->query.index;
 333    brw->query.active = true;
 334 }
 335
 336 /** Called at batchbuffer flush to get an ending PS_DEPTH_COUNT */
 337 void
 338 brw_emit_query_end(struct brw_context *brw)
 339 {
 340    struct intel_context *intel = &brw->intel;
 341
 342    if (!brw->query.active)
 343       return;
 344
 345    if (intel->gen >= 6) {
 346        BEGIN_BATCH(8);
 347        /* workaround: CS stall required before depth stall. */
 348        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 349        OUT_BATCH(PIPE_CONTROL_CS_STALL);
 350        OUT_BATCH(0); /* write address */
 351        OUT_BATCH(0); /* write data */
 352
 353        OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 354        OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
 355                  PIPE_CONTROL_WRITE_DEPTH_COUNT);
 356        OUT_RELOC(brw->query.bo,
 357                  I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 358                  PIPE_CONTROL_GLOBAL_GTT_WRITE |
 359                  ((brw->query.index * 2 + 1) * sizeof(uint64_t)));
 360        OUT_BATCH(0);
 361        ADVANCE_BATCH();
 362
 363    } else {
 364        BEGIN_BATCH(4);
 365        OUT_BATCH(_3DSTATE_PIPE_CONTROL |
 366                PIPE_CONTROL_DEPTH_STALL |
 367                PIPE_CONTROL_WRITE_DEPTH_COUNT);
 368        OUT_RELOC(brw->query.bo,
 369                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
 370                PIPE_CONTROL_GLOBAL_GTT_WRITE |
 371                ((brw->query.index * 2 + 1) * sizeof(uint64_t)));
 372        OUT_BATCH(0);
 373        OUT_BATCH(0);
 374        ADVANCE_BATCH();
 375    }
 376
 377    brw->query.active = false;
 378    brw->query.index++;
 379 }
 380
 381 void brw_init_queryobj_functions(struct dd_function_table *functions)
 382 {
 383    functions->NewQueryObject = brw_new_query_object;
 384    functions->DeleteQuery = brw_delete_query;
 385    functions->BeginQuery = brw_begin_query;
 386    functions->EndQuery = brw_end_query;
 387    functions->CheckQuery = brw_check_query;
 388    functions->WaitQuery = brw_wait_query;
 389 }