src/gallium/drivers/nvc0/nvc0_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 #include "nvc0_context.h"
  26 #include "nouveau/nv_object.xml.h"
  27
  28 /* XXX: Nested queries, and simultaneous queries on multiple gallium contexts
  29  * (since we use only a single GPU channel per screen) will not work properly.
  30  *
  31  * The first is not that big of an issue because OpenGL does not allow nested
  32  * queries anyway.
  33  */
  34
  35 struct nvc0_query {
  36    uint32_t *data;
  37    uint32_t type;
  38    uint32_t sequence;
  39    struct nouveau_bo *bo;
  40    uint32_t base;
  41    uint32_t offset; /* base + i * 16 */
  42    boolean ready;
  43    boolean is64bit;
  44    struct nvc0_mm_allocation *mm;
  45 };
  46
  47 #define NVC0_QUERY_ALLOC_SPACE 128
  48
  49 static INLINE struct nvc0_query *
  50 nvc0_query(struct pipe_query *pipe)
  51 {
  52    return (struct nvc0_query *)pipe;
  53 }
  54
  55 static boolean
  56 nvc0_query_allocate(struct nvc0_context *nvc0, struct nvc0_query *q, int size)
  57 {
  58    struct nvc0_screen *screen = nvc0->screen;
  59    int ret;
  60
  61    if (q->bo) {
  62       nouveau_bo_ref(NULL, &q->bo);
  63       if (q->mm) {
  64          if (q->ready)
  65             nvc0_mm_free(q->mm);
  66          else
  67             nvc0_fence_sched_release(screen->fence.current, q->mm);
  68       }
  69    }
  70    if (size) {
  71       q->mm = nvc0_mm_allocate(screen->mm_GART, size, &q->bo, &q->base);
  72       if (!q->bo)
  73          return FALSE;
  74       q->offset = q->base;
  75
  76       ret = nouveau_bo_map_range(q->bo, q->base, size, NOUVEAU_BO_RD |
  77                                  NOUVEAU_BO_NOSYNC);
  78       if (ret) {
  79          nvc0_query_allocate(nvc0, q, 0);
  80          return FALSE;
  81       }
  82       q->data = q->bo->map;
  83       nouveau_bo_unmap(q->bo);
  84    }
  85    return TRUE;
  86 }
  87
  88 static void
  89 nvc0_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
  90 {
  91    nvc0_query_allocate(nvc0_context(pipe), nvc0_query(pq), 0);
  92    FREE(nvc0_query(pq));
  93 }
  94
  95 static struct pipe_query *
  96 nvc0_query_create(struct pipe_context *pipe, unsigned type)
  97 {
  98    struct nvc0_context *nvc0 = nvc0_context(pipe);
  99    struct nvc0_query *q;
 100
 101    q = CALLOC_STRUCT(nvc0_query);
 102    if (!q)
 103       return NULL;
 104
 105    if (!nvc0_query_allocate(nvc0, q, NVC0_QUERY_ALLOC_SPACE)) {
 106       FREE(q);
 107       return NULL;
 108    }
 109
 110    q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
 111                  type == PIPE_QUERY_PRIMITIVES_EMITTED ||
 112                  type == PIPE_QUERY_SO_STATISTICS);
 113    q->type = type;
 114
 115    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 116       q->offset -= 16;
 117       q->data -= 16 / sizeof(*q->data); /* we advance before query_begin ! */
 118    }
 119
 120    return (struct pipe_query *)q;
 121 }
 122
 123 static void
 124 nvc0_query_get(struct nouveau_channel *chan, struct nvc0_query *q,
 125                unsigned offset, uint32_t get)
 126 {
 127    offset += q->offset;
 128
 129    MARK_RING (chan, 5, 2);
 130    BEGIN_RING(chan, RING_3D(QUERY_ADDRESS_HIGH), 4);
 131    OUT_RELOCh(chan, q->bo, offset, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 132    OUT_RELOCl(chan, q->bo, offset, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 133    OUT_RING  (chan, q->sequence);
 134    OUT_RING  (chan, get);
 135 }
 136
 137 static void
 138 nvc0_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 139 {
 140    struct nvc0_context *nvc0 = nvc0_context(pipe);
 141    struct nouveau_channel *chan = nvc0->screen->base.channel;
 142    struct nvc0_query *q = nvc0_query(pq);
 143
 144    /* For occlusion queries we have to change the storage, because a previous
 145     * query might set the initial render conition to FALSE even *after* we re-
 146     * initialized it to TRUE.
 147     */
 148    if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
 149       q->offset += 16;
 150       q->data += 16 / sizeof(*q->data);
 151       if (q->offset - q->base == NVC0_QUERY_ALLOC_SPACE)
 152          nvc0_query_allocate(nvc0, q, NVC0_QUERY_ALLOC_SPACE);
 153
 154       /* XXX: can we do this with the GPU, and sync with respect to a previous
 155        *  query ?
 156        */
 157       q->data[1] = 1; /* initial render condition = TRUE */
 158    }
 159    if (!q->is64bit)
 160       q->data[0] = q->sequence++; /* the previously used one */
 161
 162    switch (q->type) {
 163    case PIPE_QUERY_OCCLUSION_COUNTER:
 164       IMMED_RING(chan, RING_3D(COUNTER_RESET), NVC0_3D_COUNTER_RESET_SAMPLECNT);
 165       IMMED_RING(chan, RING_3D(SAMPLECNT_ENABLE), 1);
 166       break;
 167    case PIPE_QUERY_PRIMITIVES_GENERATED: /* store before & after instead ? */
 168       IMMED_RING(chan, RING_3D(COUNTER_RESET),
 169                  NVC0_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 170       break;
 171    case PIPE_QUERY_PRIMITIVES_EMITTED:
 172       IMMED_RING(chan, RING_3D(COUNTER_RESET),
 173                  NVC0_3D_COUNTER_RESET_EMITTED_PRIMITIVES);
 174       break;
 175    case PIPE_QUERY_SO_STATISTICS:
 176       BEGIN_RING_NI(chan, RING_3D(COUNTER_RESET), 2);
 177       OUT_RING  (chan, NVC0_3D_COUNTER_RESET_EMITTED_PRIMITIVES);
 178       OUT_RING  (chan, NVC0_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 179       break;
 180    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 181    case PIPE_QUERY_TIME_ELAPSED:
 182       nvc0_query_get(chan, q, 0x10, 0x00005002);
 183       break;
 184    default:
 185       break;
 186    }
 187    q->ready = FALSE;
 188 }
 189
 190 static void
 191 nvc0_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 192 {
 193    struct nvc0_context *nvc0 = nvc0_context(pipe);
 194    struct nouveau_channel *chan = nvc0->screen->base.channel;
 195    struct nvc0_query *q = nvc0_query(pq);
 196
 197    const int index = 0; /* for multiple vertex streams */
 198
 199    switch (q->type) {
 200    case PIPE_QUERY_OCCLUSION_COUNTER:
 201       nvc0_query_get(chan, q, 0, 0x0100f002);
 202       BEGIN_RING(chan, RING_3D(SAMPLECNT_ENABLE), 1);
 203       OUT_RING  (chan, 0);
 204       break;
 205    case PIPE_QUERY_PRIMITIVES_GENERATED:
 206       nvc0_query_get(chan, q, 0, 0x09005002 | (index << 5));
 207       break;
 208    case PIPE_QUERY_PRIMITIVES_EMITTED:
 209       nvc0_query_get(chan, q, 0, 0x05805002 | (index << 5));
 210       break;
 211    case PIPE_QUERY_SO_STATISTICS:
 212       nvc0_query_get(chan, q, 0x00, 0x05805002 | (index << 5));
 213       nvc0_query_get(chan, q, 0x10, 0x09005002 | (index << 5));
 214       break;
 215    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 216    case PIPE_QUERY_TIME_ELAPSED:
 217       nvc0_query_get(chan, q, 0, 0x00005002);
 218       break;
 219    case PIPE_QUERY_GPU_FINISHED:
 220       nvc0_query_get(chan, q, 0, 0x1000f010);
 221       break;
 222    default:
 223       assert(0);
 224       break;
 225    }
 226 }
 227
 228 static INLINE boolean
 229 nvc0_query_ready(struct nvc0_query *q)
 230 {
 231    return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
 232 }
 233
 234 static INLINE boolean
 235 nvc0_query_wait(struct nvc0_query *q)
 236 {
 237    int ret = nouveau_bo_map(q->bo, NOUVEAU_BO_RD);
 238    if (ret)
 239       return FALSE;
 240    nouveau_bo_unmap(q->bo);
 241    return TRUE;
 242 }
 243
 244 static boolean
 245 nvc0_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 246                   boolean wait, void *result)
 247 {
 248    struct nvc0_query *q = nvc0_query(pq);
 249    uint64_t *res64 = result;
 250    uint32_t *res32 = result;
 251    boolean *res8 = result;
 252    uint64_t *data64 = (uint64_t *)q->data;
 253
 254    if (q->type == PIPE_QUERY_GPU_FINISHED) {
 255       res8[0] = nvc0_query_ready(q);
 256       return TRUE;
 257    }
 258
 259    if (!q->ready) /* update ? */
 260       q->ready = nvc0_query_ready(q);
 261    if (!q->ready) {
 262       struct nouveau_channel *chan = nvc0_context(pipe)->screen->base.channel;
 263       if (!wait) {
 264          if (nouveau_bo_pending(q->bo) & NOUVEAU_BO_WR) /* for daft apps */
 265             FIRE_RING(chan);
 266          return FALSE;
 267       }
 268       if (!nvc0_query_wait(q))
 269          return FALSE;
 270    }
 271    q->ready = TRUE;
 272
 273    switch (q->type) {
 274    case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
 275       res32[0] = q->data[1];
 276       break;
 277    case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
 278    case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
 279       res64[0] = data64[0];
 280       break;
 281    case PIPE_QUERY_SO_STATISTICS:
 282       res64[0] = data64[0];
 283       res64[1] = data64[1];
 284       break;
 285    case PIPE_QUERY_TIMESTAMP_DISJOINT: /* u32 sequence, u32 0, u64 time */
 286       res64[0] = 1000000000;
 287       res8[8] = (data64[0] == data64[2]) ? FALSE : TRUE;
 288       break;
 289    case PIPE_QUERY_TIME_ELAPSED:
 290       res64[0] = data64[1] - data64[3];
 291       break;
 292    default:
 293       return FALSE;
 294    }
 295
 296    return TRUE;
 297 }
 298
 299 static void
 300 nvc0_render_condition(struct pipe_context *pipe,
 301                       struct pipe_query *pq, uint mode)
 302 {
 303    struct nvc0_context *nvc0 = nvc0_context(pipe);
 304    struct nouveau_channel *chan = nvc0->screen->base.channel;
 305    struct nvc0_query *q;
 306
 307    if (!pq) {
 308       IMMED_RING(chan, RING_3D(COND_MODE), NVC0_3D_COND_MODE_ALWAYS);
 309       return;
 310    }
 311    q = nvc0_query(pq);
 312
 313    if (mode == PIPE_RENDER_COND_WAIT ||
 314        mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
 315       MARK_RING (chan, 5, 2);
 316       BEGIN_RING(chan, RING_3D_(NV84_SUBCHAN_QUERY_ADDRESS_HIGH), 4);
 317       OUT_RELOCh(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 318       OUT_RELOCl(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 319       OUT_RING  (chan, q->sequence);
 320       OUT_RING  (chan, 0x00001001);
 321    }
 322
 323    MARK_RING (chan, 4, 2);
 324    BEGIN_RING(chan, RING_3D(COND_ADDRESS_HIGH), 3);
 325    OUT_RELOCh(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 326    OUT_RELOCl(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 327    OUT_RING  (chan, NVC0_3D_COND_MODE_RES_NON_ZERO);
 328 }
 329
 330 void
 331 nvc0_init_query_functions(struct nvc0_context *nvc0)
 332 {
 333    nvc0->pipe.create_query = nvc0_query_create;
 334    nvc0->pipe.destroy_query = nvc0_query_destroy;
 335    nvc0->pipe.begin_query = nvc0_query_begin;
 336    nvc0->pipe.end_query = nvc0_query_end;
 337    nvc0->pipe.get_query_result = nvc0_query_result;
 338    nvc0->pipe.render_condition = nvc0_render_condition;
 339 }