src/gallium/drivers/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 BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  18  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
  19  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  20  * SOFTWARE.
  21  *
  22  * Authors: Christoph Bumiller
  23  */
  24
  25 #define NV50_PUSH_EXPLICIT_SPACE_CHECKING
  26
  27 #include "nv50_context.h"
  28 #include "nouveau/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    uint32_t type;
  40    uint32_t sequence;
  41    struct nouveau_bo *bo;
  42    uint32_t base;
  43    uint32_t offset; /* base + i * 16 */
  44    boolean ready;
  45    boolean flushed;
  46    boolean is64bit;
  47    struct nouveau_mm_allocation *mm;
  48 };
  49
  50 #define NV50_QUERY_ALLOC_SPACE 128
  51
  52 static INLINE struct nv50_query *
  53 nv50_query(struct pipe_query *pipe)
  54 {
  55    return (struct nv50_query *)pipe;
  56 }
  57
  58 static boolean
  59 nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
  60 {
  61    struct nv50_screen *screen = nv50->screen;
  62    int ret;
  63
  64    if (q->bo) {
  65       nouveau_bo_ref(NULL, &q->bo);
  66       if (q->mm) {
  67          if (q->ready)
  68             nouveau_mm_free(q->mm);
  69          else
  70             nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
  71                                q->mm);
  72       }
  73    }
  74    if (size) {
  75       q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
  76       if (!q->bo)
  77          return FALSE;
  78       q->offset = q->base;
  79
  80       ret = nouveau_bo_map(q->bo, 0, screen->base.client);
  81       if (ret) {
  82          nv50_query_allocate(nv50, q, 0);
  83          return FALSE;
  84       }
  85       q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
  86    }
  87    return TRUE;
  88 }
  89
  90 static void
  91 nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
  92 {
  93    nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
  94    FREE(nv50_query(pq));
  95 }
  96
  97 static struct pipe_query *
  98 nv50_query_create(struct pipe_context *pipe, unsigned type)
  99 {
 100    struct nv50_context *nv50 = nv50_context(pipe);
 101    struct nv50_query *q;
 102
 103    q = CALLOC_STRUCT(nv50_query);
 104    if (!q)
 105       return NULL;
 106
 107    if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
 108       FREE(q);
 109       return NULL;
 110    }
 111
 112    q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
 113                  type == PIPE_QUERY_PRIMITIVES_EMITTED ||
 114                  type == PIPE_QUERY_SO_STATISTICS);
 115    q->type = type;
 116
 117    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 118       q->offset -= 16;
 119       q->data -= 16 / sizeof(*q->data); /* we advance before query_begin ! */
 120    }
 121
 122    return (struct pipe_query *)q;
 123 }
 124
 125 static void
 126 nv50_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
 127                unsigned offset, uint32_t get)
 128 {
 129    offset += q->offset;
 130
 131    PUSH_SPACE(push, 5);
 132    PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 133    BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
 134    PUSH_DATAh(push, q->bo->offset + offset);
 135    PUSH_DATA (push, q->bo->offset + offset);
 136    PUSH_DATA (push, q->sequence);
 137    PUSH_DATA (push, get);
 138 }
 139
 140 static void
 141 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 142 {
 143    struct nv50_context *nv50 = nv50_context(pipe);
 144    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 145    struct nv50_query *q = nv50_query(pq);
 146
 147    /* For occlusion queries we have to change the storage, because a previous
 148     * query might set the initial render conition to FALSE even *after* we re-
 149     * initialized it to TRUE.
 150     */
 151    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 152       q->offset += 16;
 153       q->data += 16 / sizeof(*q->data);
 154       if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
 155          nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
 156
 157       /* XXX: can we do this with the GPU, and sync with respect to a previous
 158        *  query ?
 159        */
 160       q->data[1] = 1; /* initial render condition = TRUE */
 161    }
 162    if (!q->is64bit)
 163       q->data[0] = q->sequence++; /* the previously used one */
 164
 165    switch (q->type) {
 166    case PIPE_QUERY_OCCLUSION_COUNTER:
 167       PUSH_SPACE(push, 4);
 168       BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
 169       PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
 170       BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
 171       PUSH_DATA (push, 1);
 172       break;
 173    case PIPE_QUERY_PRIMITIVES_GENERATED: /* store before & after instead ? */
 174       PUSH_SPACE(push, 2);
 175       BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
 176       PUSH_DATA (push, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 177       break;
 178    case PIPE_QUERY_PRIMITIVES_EMITTED:
 179       PUSH_SPACE(push, 2);
 180       BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
 181       PUSH_DATA (push, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
 182       break;
 183    case PIPE_QUERY_SO_STATISTICS:
 184       PUSH_SPACE(push, 3);
 185       BEGIN_NI04(push, NV50_3D(COUNTER_RESET), 2);
 186       PUSH_DATA (push, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
 187       PUSH_DATA (push, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 188       break;
 189    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 190    case PIPE_QUERY_TIME_ELAPSED:
 191       nv50_query_get(push, q, 0x10, 0x00005002);
 192       break;
 193    default:
 194       break;
 195    }
 196    q->ready = FALSE;
 197 }
 198
 199 static void
 200 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 201 {
 202    struct nv50_context *nv50 = nv50_context(pipe);
 203    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 204    struct nv50_query *q = nv50_query(pq);
 205
 206    switch (q->type) {
 207    case PIPE_QUERY_OCCLUSION_COUNTER:
 208       nv50_query_get(push, q, 0, 0x0100f002);
 209       PUSH_SPACE(push, 2);
 210       BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
 211       PUSH_DATA (push, 0);
 212       break;
 213    case PIPE_QUERY_PRIMITIVES_GENERATED:
 214       nv50_query_get(push, q, 0, 0x06805002);
 215       break;
 216    case PIPE_QUERY_PRIMITIVES_EMITTED:
 217       nv50_query_get(push, q, 0, 0x05805002);
 218       break;
 219    case PIPE_QUERY_SO_STATISTICS:
 220       nv50_query_get(push, q, 0x00, 0x05805002);
 221       nv50_query_get(push, q, 0x10, 0x06805002);
 222       break;
 223    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 224    case PIPE_QUERY_TIME_ELAPSED:
 225       nv50_query_get(push, q, 0, 0x00005002);
 226       break;
 227    case PIPE_QUERY_GPU_FINISHED:
 228       nv50_query_get(push, q, 0, 0x1000f010);
 229       break;
 230    default:
 231       assert(0);
 232       break;
 233    }
 234    q->flushed = FALSE;
 235 }
 236
 237 static INLINE boolean
 238 nv50_query_ready(struct nv50_query *q)
 239 {
 240    return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
 241 }
 242
 243 static boolean
 244 nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 245                   boolean wait, union pipe_query_result *result)
 246 {
 247    struct nv50_context *nv50 = nv50_context(pipe);
 248    struct nv50_query *q = nv50_query(pq);
 249    uint64_t *res64 = (uint64_t *)result;
 250    boolean *res8 = (boolean *)result;
 251    uint64_t *data64 = (uint64_t *)q->data;
 252
 253    if (!q->ready) /* update ? */
 254       q->ready = nv50_query_ready(q);
 255    if (!q->ready) {
 256       if (!wait) {
 257          /* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
 258          if (!q->flushed) {
 259             q->flushed = TRUE;
 260             PUSH_KICK(nv50->base.pushbuf);
 261          }
 262          return FALSE;
 263       }
 264       if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
 265          return FALSE;
 266    }
 267    q->ready = TRUE;
 268
 269    switch (q->type) {
 270    case PIPE_QUERY_GPU_FINISHED:
 271       res8[0] = TRUE;
 272       break;
 273    case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
 274       res64[0] = q->data[1];
 275       break;
 276    case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
 277    case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
 278       res64[0] = data64[0];
 279       break;
 280    case PIPE_QUERY_SO_STATISTICS:
 281       res64[0] = data64[0];
 282       res64[1] = data64[1];
 283       break;
 284    case PIPE_QUERY_TIMESTAMP_DISJOINT: /* u32 sequence, u32 0, u64 time */
 285       res64[0] = 1000000000;
 286       res8[8] = (data64[0] == data64[2]) ? FALSE : TRUE;
 287       break;
 288    case PIPE_QUERY_TIME_ELAPSED:
 289       res64[0] = data64[1] - data64[3];
 290       break;
 291    default:
 292       return FALSE;
 293    }
 294
 295    return TRUE;
 296 }
 297
 298 static void
 299 nv50_render_condition(struct pipe_context *pipe,
 300                       struct pipe_query *pq, uint mode)
 301 {
 302    struct nv50_context *nv50 = nv50_context(pipe);
 303    struct nouveau_pushbuf *push = nv50->base.pushbuf;
 304    struct nv50_query *q;
 305
 306    PUSH_SPACE(push, 6);
 307
 308    if (!pq) {
 309       BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
 310       PUSH_DATA (push, NV50_3D_COND_MODE_ALWAYS);
 311       return;
 312    }
 313    q = nv50_query(pq);
 314
 315    if (mode == PIPE_RENDER_COND_WAIT ||
 316        mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
 317       BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
 318       PUSH_DATA (push, 0);
 319    }
 320
 321    BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
 322    PUSH_DATAh(push, q->bo->offset + q->offset);
 323    PUSH_DATA (push, q->bo->offset + q->offset);
 324    PUSH_DATA (push, NV50_3D_COND_MODE_RES_NON_ZERO);
 325 }
 326
 327 void
 328 nv50_init_query_functions(struct nv50_context *nv50)
 329 {
 330    struct pipe_context *pipe = &nv50->base.pipe;
 331
 332    pipe->create_query = nv50_query_create;
 333    pipe->destroy_query = nv50_query_destroy;
 334    pipe->begin_query = nv50_query_begin;
 335    pipe->end_query = nv50_query_end;
 336    pipe->get_query_result = nv50_query_result;
 337    pipe->render_condition = nv50_render_condition;
 338 }