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 #include "nv50_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 nv50_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 nouveau_mm_allocation *mm;
  45 };
  46
  47 #define NV50_QUERY_ALLOC_SPACE 128
  48
  49 static INLINE struct nv50_query *
  50 nv50_query(struct pipe_query *pipe)
  51 {
  52    return (struct nv50_query *)pipe;
  53 }
  54
  55 static boolean
  56 nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
  57 {
  58    struct nv50_screen *screen = nv50->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             nouveau_mm_free(q->mm);
  66          else
  67             nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work, q->mm);
  68       }
  69    }
  70    if (size) {
  71       q->mm = nouveau_mm_allocate(screen->base.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          nv50_query_allocate(nv50, 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 nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
  90 {
  91    nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
  92    FREE(nv50_query(pq));
  93 }
  94
  95 static struct pipe_query *
  96 nv50_query_create(struct pipe_context *pipe, unsigned type)
  97 {
  98    struct nv50_context *nv50 = nv50_context(pipe);
  99    struct nv50_query *q;
 100
 101    q = CALLOC_STRUCT(nv50_query);
 102    if (!q)
 103       return NULL;
 104
 105    if (!nv50_query_allocate(nv50, q, NV50_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 nv50_query_get(struct nouveau_channel *chan, struct nv50_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 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
 139 {
 140    struct nv50_context *nv50 = nv50_context(pipe);
 141    struct nouveau_channel *chan = nv50->screen->base.channel;
 142    struct nv50_query *q = nv50_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 == NV50_QUERY_ALLOC_SPACE)
 152          nv50_query_allocate(nv50, q, NV50_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       BEGIN_RING(chan, RING_3D(COUNTER_RESET), 1);
 165       OUT_RING  (chan, NV50_3D_COUNTER_RESET_SAMPLECNT);
 166       BEGIN_RING(chan, RING_3D(SAMPLECNT_ENABLE), 1);
 167       OUT_RING  (chan, 1);
 168       break;
 169    case PIPE_QUERY_PRIMITIVES_GENERATED: /* store before & after instead ? */
 170       BEGIN_RING(chan, RING_3D(COUNTER_RESET), 1);
 171       OUT_RING  (chan, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 172       break;
 173    case PIPE_QUERY_PRIMITIVES_EMITTED:
 174       BEGIN_RING(chan, RING_3D(COUNTER_RESET), 1);
 175       OUT_RING  (chan, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
 176       break;
 177    case PIPE_QUERY_SO_STATISTICS:
 178       BEGIN_RING_NI(chan, RING_3D(COUNTER_RESET), 2);
 179       OUT_RING  (chan, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
 180       OUT_RING  (chan, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
 181       break;
 182    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 183    case PIPE_QUERY_TIME_ELAPSED:
 184       nv50_query_get(chan, q, 0x10, 0x00005002);
 185       break;
 186    default:
 187       break;
 188    }
 189    q->ready = FALSE;
 190 }
 191
 192 static void
 193 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
 194 {
 195    struct nv50_context *nv50 = nv50_context(pipe);
 196    struct nouveau_channel *chan = nv50->screen->base.channel;
 197    struct nv50_query *q = nv50_query(pq);
 198
 199    switch (q->type) {
 200    case PIPE_QUERY_OCCLUSION_COUNTER:
 201       nv50_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       nv50_query_get(chan, q, 0, 0x06805002);
 207       break;
 208    case PIPE_QUERY_PRIMITIVES_EMITTED:
 209       nv50_query_get(chan, q, 0, 0x05805002);
 210       break;
 211    case PIPE_QUERY_SO_STATISTICS:
 212       nv50_query_get(chan, q, 0x00, 0x05805002);
 213       nv50_query_get(chan, q, 0x10, 0x06805002);
 214       break;
 215    case PIPE_QUERY_TIMESTAMP_DISJOINT:
 216    case PIPE_QUERY_TIME_ELAPSED:
 217       nv50_query_get(chan, q, 0, 0x00005002);
 218       break;
 219    case PIPE_QUERY_GPU_FINISHED:
 220       nv50_query_get(chan, q, 0, 0x1000f010);
 221       break;
 222    default:
 223       assert(0);
 224       break;
 225    }
 226 }
 227
 228 static INLINE boolean
 229 nv50_query_ready(struct nv50_query *q)
 230 {
 231    return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
 232 }
 233
 234 static INLINE boolean
 235 nv50_query_wait(struct nv50_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 nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
 246                   boolean wait, void *result)
 247 {
 248    struct nv50_query *q = nv50_query(pq);
 249    uint64_t *res64 = result;
 250    boolean *res8 = result;
 251    uint64_t *data64 = (uint64_t *)q->data;
 252
 253    if (q->type == PIPE_QUERY_GPU_FINISHED) {
 254       res8[0] = nv50_query_ready(q);
 255       return TRUE;
 256    }
 257
 258    if (!q->ready) /* update ? */
 259       q->ready = nv50_query_ready(q);
 260    if (!q->ready) {
 261       struct nouveau_channel *chan = nv50_context(pipe)->screen->base.channel;
 262       if (!wait) {
 263          if (nouveau_bo_pending(q->bo) & NOUVEAU_BO_WR) /* for daft apps */
 264             FIRE_RING(chan);
 265          return FALSE;
 266       }
 267       if (!nv50_query_wait(q))
 268          return FALSE;
 269    }
 270    q->ready = TRUE;
 271
 272    switch (q->type) {
 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_channel *chan = nv50->screen->base.channel;
 304    struct nv50_query *q;
 305
 306    if (!pq) {
 307       BEGIN_RING(chan, RING_3D(COND_MODE), 1);
 308       OUT_RING  (chan, NV50_3D_COND_MODE_ALWAYS);
 309       return;
 310    }
 311    q = nv50_query(pq);
 312
 313    if (mode == PIPE_RENDER_COND_WAIT ||
 314        mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
 315       BEGIN_RING(chan, RING_3D_(NV50_GRAPH_WAIT_FOR_IDLE), 1);
 316       OUT_RING  (chan, 0);
 317    }
 318
 319    MARK_RING (chan, 4, 2);
 320    BEGIN_RING(chan, RING_3D(COND_ADDRESS_HIGH), 3);
 321    OUT_RELOCh(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 322    OUT_RELOCl(chan, q->bo, q->offset, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
 323    OUT_RING  (chan, NV50_3D_COND_MODE_RES_NON_ZERO);
 324 }
 325
 326 void
 327 nv50_init_query_functions(struct nv50_context *nv50)
 328 {
 329    struct pipe_context *pipe = &nv50->base.pipe;
 330
 331    pipe->create_query = nv50_query_create;
 332    pipe->destroy_query = nv50_query_destroy;
 333    pipe->begin_query = nv50_query_begin;
 334    pipe->end_query = nv50_query_end;
 335    pipe->get_query_result = nv50_query_result;
 336    pipe->render_condition = nv50_render_condition;
 337 }