src/gallium/drivers/nouveau/nv50/nv50_query.c

   1 /*
   2  * Copyright 2011 Nouveau Project
   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 shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * Authors: Christoph Bumiller
  23  */
  24
  25 #define NV50_PUSH_EXPLICIT_SPACE_CHECKING
  26
  27 #include "nv50/nv50_context.h"
  28 #include "nv_object.xml.h"
  29
  30 /* XXX: Nested queries, and simultaneous queries on multiple gallium contexts
  31  * (since we use only a single GPU channel per screen) will not work properly.
  32  *
  33  * The first is not that big of an issue because OpenGL does not allow nested
  34  * queries anyway.
  35  */
  36
  37 struct nv50_query {
  38    uint32_t *data;
  39    uint16_t type;
  40    uint16_t index;
  41    uint32_t sequence;
  42    struct nouveau_bo *bo;
  43    uint32_t base;
  44    uint32_t offset; /* base + i * 16 */
  45    boolean ready;
  46    boolean flushed;
  47    boolean is64bit;
  48    struct nouveau_mm_allocation *mm;
  49 };
  50
  51 #define NV50_QUERY_ALLOC_SPACE 256
  52
  53 static INLINE struct nv50_query *
  54 nv50_query(struct pipe_query *pipe)
  55 {
  56    return (struct nv50_query *)pipe;
  57 }
  58
  59 static boolean
  60 nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
  61 {
  62    struct nv50_screen *screen = nv50->screen;
  63    int ret;
  64
  65    if (q->bo) {
  66       nouveau_bo_ref(NULL, &q->bo);
  67       if (q->mm) {
  68          if (q->ready)
  69             nouveau_mm_free(q->mm);
  70          else
  71             nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
  72                                q->mm);
  73       }
  74    }
  75    if (size) {
  76       q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
  77       if (!q->bo)
  78          return FALSE;
  79       q->offset = q->base;
  80
  81       ret = nouveau_bo_map(q->bo, 0, screen->base.client);
  82       if (ret) {
  83          nv50_query_allocate(nv50, q, 0);
  84          return FALSE;
  85       }
  86       q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
  87    }
  88    return TRUE;
  89 }
  90
  91 static void
  92 nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
  93 {
  94    nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
  95    FREE(nv50_query(pq));
  96 }
  97
  98 static struct pipe_query *
  99 nv50_query_create(struct pipe_context *pipe, unsigned type, unsigned index)
 100 {
 101    struct nv50_context *nv50 = nv50_context(pipe);
 102    struct nv50_query *q;
 103
 104    q = CALLOC_STRUCT(nv50_query);
 105    if (!q)
 106       return NULL;
 107
 108    if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
 109       FREE(q);
 110       return NULL;
 111    }
 112
 113    q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
 114                  type == PIPE_QUERY_PRIMITIVES_EMITTED ||
 115                  type == PIPE_QUERY_SO_STATISTICS);
 116    q->type = type;
 117
 118    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 119       q->offset -= 16;
 120       q->data -= 16 / sizeof(*q->data); /* we advance before query_begin ! */
 121    }
 122
 123    return (struct pipe_query *)q;
 124 }
 125
 126 static void
 127 nv50_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
 128                unsigned offset, uint32_t get)
 129 {
 130    offset += q->offset;
 131
 132    PUSH_SPACE(push, 5);
 133    PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 134    BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
 135    PUSH_DATAh(push, q->bo->offset + offset);
 136    PUSH_DATA (push, q->bo->offset + offset);
 137    PUSH_DATA (push, q->sequence);
 138    PUSH_DATA (push, get);
 139 }
 140
 141 static void
 142 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 143 {
 144    struct nv50_context *nv50 = nv50_context(pipe);
 145    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 146    struct nv50_query *q = nv50_query(pq);
 147
 148    /* For occlusion queries we have to change the storage, because a previous
 149     * query might set the initial render conition to FALSE even *after* we re-
 150     * initialized it to TRUE.
 151     */
 152    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 153       q->offset += 16;
 154       q->data += 16 / sizeof(*q->data);
 155       if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
 156          nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
 157
 158       /* XXX: can we do this with the GPU, and sync with respect to a previous
 159        *  query ?
 160        */
 161       q->data[0] = q->sequence; /* initialize sequence */
 162       q->data[1] = 1; /* initial render condition = TRUE */
 163       q->data[4] = q->sequence + 1; /* for comparison COND_MODE */
 164       q->data[5] = 0;
 165    }
 166    if (!q->is64bit)
 167       q->data[0] = q->sequence++; /* the previously used one */
 168
 169    switch (q->type) {
 170    case PIPE_QUERY_OCCLUSION_COUNTER:
 171       PUSH_SPACE(push, 4);
 172       BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
 173       PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
 174       BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
 175       PUSH_DATA (push, 1);
 176       break;
 177    case PIPE_QUERY_PRIMITIVES_GENERATED:
 178       nv50_query_get(push, q, 0x10, 0x06805002);
 179       break;
 180    case PIPE_QUERY_PRIMITIVES_EMITTED:
 181       nv50_query_get(push, q, 0x10, 0x05805002);
 182       break;
 183    case PIPE_QUERY_SO_STATISTICS:
 184       nv50_query_get(push, q, 0x20, 0x05805002);
 185       nv50_query_get(push, q, 0x30, 0x06805002);
 186       break;
 187    case PIPE_QUERY_PIPELINE_STATISTICS:
 188       nv50_query_get(push, q, 0x80, 0x00801002); /* VFETCH, VERTICES */
 189       nv50_query_get(push, q, 0x90, 0x01801002); /* VFETCH, PRIMS */
 190       nv50_query_get(push, q, 0xa0, 0x02802002); /* VP, LAUNCHES */
 191       nv50_query_get(push, q, 0xb0, 0x03806002); /* GP, LAUNCHES */
 192       nv50_query_get(push, q, 0xc0, 0x04806002); /* GP, PRIMS_OUT */
 193       nv50_query_get(push, q, 0xd0, 0x07804002); /* RAST, PRIMS_IN */
 194       nv50_query_get(push, q, 0xe0, 0x08804002); /* RAST, PRIMS_OUT */
 195       nv50_query_get(push, q, 0xf0, 0x0980a002); /* ROP, PIXELS */
 196       break;
 197    case PIPE_QUERY_TIME_ELAPSED:
 198       nv50_query_get(push, q, 0x10, 0x00005002);
 199       break;
 200    default:
 201       break;
 202    }
 203    q->ready = FALSE;
 204 }
 205
 206 static void
 207 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 208 {
 209    struct nv50_context *nv50 = nv50_context(pipe);
 210    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 211    struct nv50_query *q = nv50_query(pq);
 212
 213    switch (q->type) {
 214    case PIPE_QUERY_OCCLUSION_COUNTER:
 215       nv50_query_get(push, q, 0, 0x0100f002);
 216       PUSH_SPACE(push, 2);
 217       BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
 218       PUSH_DATA (push, 0);
 219       break;
 220    case PIPE_QUERY_PRIMITIVES_GENERATED:
 221       nv50_query_get(push, q, 0, 0x06805002);
 222       break;
 223    case PIPE_QUERY_PRIMITIVES_EMITTED:
 224       nv50_query_get(push, q, 0, 0x05805002);
 225       break;
 226    case PIPE_QUERY_SO_STATISTICS:
 227       nv50_query_get(push, q, 0x00, 0x05805002);
 228       nv50_query_get(push, q, 0x10, 0x06805002);
 229       break;
 230    case PIPE_QUERY_PIPELINE_STATISTICS:
 231       nv50_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
 232       nv50_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
 233       nv50_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
 234       nv50_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
 235       nv50_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
 236       nv50_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
 237       nv50_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
 238       nv50_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
 239       break;
 240    case PIPE_QUERY_TIMESTAMP:
 241       q->sequence++;
 242       /* fall through */
 243    case PIPE_QUERY_TIME_ELAPSED:
 244       nv50_query_get(push, q, 0, 0x00005002);
 245       break;
 246    case PIPE_QUERY_GPU_FINISHED:
 247       q->sequence++;
 248       nv50_query_get(push, q, 0, 0x1000f010);
 249       break;
 250    case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
 251       nv50_query_get(push, q, 0, 0x0d005002 | (q->index << 5));
 252       break;
 253    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 254       /* This query is not issued on GPU because disjoint is forced to FALSE */
 255       q->ready = TRUE;
 256       break;
 257    default:
 258       assert(0);
 259       break;
 260    }
 261    q->ready = q->flushed = FALSE;
 262 }
 263
 264 static INLINE boolean
 265 nv50_query_ready(struct nv50_query *q)
 266 {
 267    return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
 268 }
 269
 270 static boolean
 271 nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 272                   boolean wait, union pipe_query_result *result)
 273 {
 274    struct nv50_context *nv50 = nv50_context(pipe);
 275    struct nv50_query *q = nv50_query(pq);
 276    uint64_t *res64 = (uint64_t *)result;
 277    uint32_t *res32 = (uint32_t *)result;
 278    boolean *res8 = (boolean *)result;
 279    uint64_t *data64 = (uint64_t *)q->data;
 280    int i;
 281
 282    if (!q->ready) /* update ? */
 283       q->ready = nv50_query_ready(q);
 284    if (!q->ready) {
 285       if (!wait) {
 286          /* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
 287          if (!q->flushed) {
 288             q->flushed = TRUE;
 289             PUSH_KICK(nv50->base.pushbuf);
 290          }
 291          return FALSE;
 292       }
 293       if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
 294          return FALSE;
 295    }
 296    q->ready = TRUE;
 297
 298    switch (q->type) {
 299    case PIPE_QUERY_GPU_FINISHED:
 300       res8[0] = TRUE;
 301       break;
 302    case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
 303       res64[0] = q->data[1];
 304       break;
 305    case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
 306    case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
 307       res64[0] = data64[0] - data64[2];
 308       break;
 309    case PIPE_QUERY_SO_STATISTICS:
 310       res64[0] = data64[0] - data64[4];
 311       res64[1] = data64[2] - data64[6];
 312       break;
 313    case PIPE_QUERY_PIPELINE_STATISTICS:
 314       for (i = 0; i < 8; ++i)
 315          res64[i] = data64[i * 2] - data64[16 + i * 2];
 316       break;
 317    case PIPE_QUERY_TIMESTAMP:
 318       res64[0] = data64[1];
 319       break;
 320    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 321       res64[0] = 1000000000;
 322       res8[8] = FALSE;
 323       break;
 324    case PIPE_QUERY_TIME_ELAPSED:
 325       res64[0] = data64[1] - data64[3];
 326       break;
 327    case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
 328       res32[0] = q->data[1];
 329       break;
 330    default:
 331       return FALSE;
 332    }
 333
 334    return TRUE;
 335 }
 336
 337 void
 338 nv84_query_fifo_wait(struct nouveau_pushbuf *push, struct pipe_query *pq)
 339 {
 340    struct nv50_query *q = nv50_query(pq);
 341    unsigned offset = q->offset;
 342
 343    PUSH_SPACE(push, 5);
 344    PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 345    BEGIN_NV04(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
 346    PUSH_DATAh(push, q->bo->offset + offset);
 347    PUSH_DATA (push, q->bo->offset + offset);
 348    PUSH_DATA (push, q->sequence);
 349    PUSH_DATA (push, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
 350 }
 351
 352 static void
 353 nv50_render_condition(struct pipe_context *pipe,
 354                       struct pipe_query *pq,
 355                       boolean condition, uint mode)
 356 {
 357    struct nv50_context *nv50 = nv50_context(pipe);
 358    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 359    struct nv50_query *q;
 360    uint32_t cond;
 361    boolean wait =
 362       mode != PIPE_RENDER_COND_NO_WAIT &&
 363       mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
 364
 365    if (!pq) {
 366       cond = NV50_3D_COND_MODE_ALWAYS;
 367    }
 368    else {
 369       q = nv50_query(pq);
 370       /* NOTE: comparison of 2 queries only works if both have completed */
 371       switch (q->type) {
 372       case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
 373          cond = condition ? NV50_3D_COND_MODE_EQUAL :
 374                             NV50_3D_COND_MODE_NOT_EQUAL;
 375          wait = TRUE;
 376          break;
 377       case PIPE_QUERY_OCCLUSION_COUNTER:
 378       case PIPE_QUERY_OCCLUSION_PREDICATE:
 379          if (likely(!condition)) {
 380             /* XXX: Placeholder, handle nesting here if available */
 381             if (unlikely(false))
 382                cond = wait ? NV50_3D_COND_MODE_NOT_EQUAL :
 383                              NV50_3D_COND_MODE_ALWAYS;
 384             else
 385                cond = NV50_3D_COND_MODE_RES_NON_ZERO;
 386          } else {
 387             cond = wait ? NV50_3D_COND_MODE_EQUAL : NV50_3D_COND_MODE_ALWAYS;
 388          }
 389          break;
 390       default:
 391          assert(!"render condition query not a predicate");
 392          cond = NV50_3D_COND_MODE_ALWAYS;
 393          break;
 394       }
 395    }
 396
 397    nv50->cond_query = pq;
 398    nv50->cond_cond = condition;
 399    nv50->cond_condmode = cond;
 400    nv50->cond_mode = mode;
 401
 402    if (!pq) {
 403       PUSH_SPACE(push, 2);
 404       BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 405       PUSH_DATA (push, cond);
 406       return;
 407    }
 408
 409    PUSH_SPACE(push, 9);
 410
 411    if (wait) {
 412       BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
 413       PUSH_DATA (push, 0);
 414    }
 415
 416    PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 417    BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
 418    PUSH_DATAh(push, q->bo->offset + q->offset);
 419    PUSH_DATA (push, q->bo->offset + q->offset);
 420    PUSH_DATA (push, cond);
 421
 422    BEGIN_NV04(push, NV50_2D(COND_ADDRESS_HIGH), 2);
 423    PUSH_DATAh(push, q->bo->offset + q->offset);
 424    PUSH_DATA (push, q->bo->offset + q->offset);
 425 }
 426
 427 void
 428 nv50_query_pushbuf_submit(struct nouveau_pushbuf *push,
 429                           struct pipe_query *pq, unsigned result_offset)
 430 {
 431    struct nv50_query *q = nv50_query(pq);
 432
 433    /* XXX: does this exist ? */
 434 #define NV50_IB_ENTRY_1_NO_PREFETCH (0 << (31 - 8))
 435
 436    nouveau_pushbuf_space(push, 0, 0, 1);
 437    nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
 438                         NV50_IB_ENTRY_1_NO_PREFETCH);
 439 }
 440
 441 void
 442 nva0_so_target_save_offset(struct pipe_context *pipe,
 443                            struct pipe_stream_output_target *ptarg,
 444                            unsigned index, boolean serialize)
 445 {
 446    struct nv50_so_target *targ = nv50_so_target(ptarg);
 447
 448    if (serialize) {
 449       struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
 450       PUSH_SPACE(push, 2);
 451       BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
 452       PUSH_DATA (push, 0);
 453    }
 454
 455    nv50_query(targ->pq)->index = index;
 456    nv50_query_end(pipe, targ->pq);
 457 }
 458
 459 void
 460 nv50_init_query_functions(struct nv50_context *nv50)
 461 {
 462    struct pipe_context *pipe = &nv50->base.pipe;
 463
 464    pipe->create_query = nv50_query_create;
 465    pipe->destroy_query = nv50_query_destroy;
 466    pipe->begin_query = nv50_query_begin;
 467    pipe->end_query = nv50_query_end;
 468    pipe->get_query_result = nv50_query_result;
 469    pipe->render_condition = nv50_render_condition;
 470 }