src/gallium/drivers/nouveau/nv50/nv50_miptree.c

   1 /*
   2  * Copyright 2008 Ben Skeggs
   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
  23 #include "pipe/p_state.h"
  24 #include "pipe/p_defines.h"
  25 #include "util/u_inlines.h"
  26 #include "util/format/u_format.h"
  27
  28 #include "nv50/nv50_context.h"
  29 #include "nv50/nv50_resource.h"
  30
  31 uint32_t
  32 nv50_tex_choose_tile_dims_helper(unsigned nx, unsigned ny, unsigned nz,
  33                                  bool is_3d)
  34 {
  35    uint32_t tile_mode = 0x000;
  36
  37    if (ny > 64) tile_mode = 0x040; /* height 128 tiles */
  38    else
  39    if (ny > 32) tile_mode = 0x030; /* height 64 tiles */
  40    else
  41    if (ny > 16) tile_mode = 0x020; /* height 32 tiles */
  42    else
  43    if (ny >  8) tile_mode = 0x010; /* height 16 tiles */
  44
  45    if (!is_3d)
  46       return tile_mode;
  47    else
  48       if (tile_mode > 0x020)
  49          tile_mode = 0x020;
  50
  51    if (nz > 16 && tile_mode < 0x020)
  52       return tile_mode | 0x500; /* depth 32 tiles */
  53    if (nz > 8) return tile_mode | 0x400; /* depth 16 tiles */
  54    if (nz > 4) return tile_mode | 0x300; /* depth 8 tiles */
  55    if (nz > 2) return tile_mode | 0x200; /* depth 4 tiles */
  56    if (nz > 1) return tile_mode | 0x100; /* depth 2 tiles */
  57
  58    return tile_mode;
  59 }
  60
  61 static uint32_t
  62 nv50_tex_choose_tile_dims(unsigned nx, unsigned ny, unsigned nz, bool is_3d)
  63 {
  64    return nv50_tex_choose_tile_dims_helper(nx, ny * 2, nz, is_3d);
  65 }
  66
  67 static uint32_t
  68 nv50_mt_choose_storage_type(struct nv50_miptree *mt, bool compressed)
  69 {
  70    const unsigned ms = util_logbase2(mt->base.base.nr_samples);
  71    uint32_t tile_flags;
  72
  73    if (unlikely(mt->base.base.flags & NOUVEAU_RESOURCE_FLAG_LINEAR))
  74       return 0;
  75    if (unlikely(mt->base.base.bind & PIPE_BIND_CURSOR))
  76       return 0;
  77
  78    switch (mt->base.base.format) {
  79    case PIPE_FORMAT_Z16_UNORM:
  80       tile_flags = 0x6c + ms;
  81       break;
  82    case PIPE_FORMAT_X8Z24_UNORM:
  83    case PIPE_FORMAT_S8X24_UINT:
  84    case PIPE_FORMAT_S8_UINT_Z24_UNORM:
  85       tile_flags = 0x18 + ms;
  86       break;
  87    case PIPE_FORMAT_X24S8_UINT:
  88    case PIPE_FORMAT_Z24X8_UNORM:
  89    case PIPE_FORMAT_Z24_UNORM_S8_UINT:
  90       tile_flags = 0x128 + ms;
  91       break;
  92    case PIPE_FORMAT_Z32_FLOAT:
  93       tile_flags = 0x40 + ms;
  94       break;
  95    case PIPE_FORMAT_X32_S8X24_UINT:
  96    case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
  97       tile_flags = 0x60 + ms;
  98       break;
  99    default:
 100       /* Most color formats don't work with compression. */
 101       compressed = false;
 102       /* fallthrough */
 103    case PIPE_FORMAT_R8G8B8A8_UNORM:
 104    case PIPE_FORMAT_R8G8B8A8_SRGB:
 105    case PIPE_FORMAT_R8G8B8X8_UNORM:
 106    case PIPE_FORMAT_R8G8B8X8_SRGB:
 107    case PIPE_FORMAT_B8G8R8A8_UNORM:
 108    case PIPE_FORMAT_B8G8R8A8_SRGB:
 109    case PIPE_FORMAT_B8G8R8X8_UNORM:
 110    case PIPE_FORMAT_B8G8R8X8_SRGB:
 111    case PIPE_FORMAT_R10G10B10A2_UNORM:
 112    case PIPE_FORMAT_B10G10R10A2_UNORM:
 113    case PIPE_FORMAT_R16G16B16A16_FLOAT:
 114    case PIPE_FORMAT_R16G16B16X16_FLOAT:
 115    case PIPE_FORMAT_R11G11B10_FLOAT:
 116       switch (util_format_get_blocksizebits(mt->base.base.format)) {
 117       case 128:
 118          assert(ms < 3);
 119          tile_flags = 0x74;
 120          break;
 121       case 64:
 122          switch (ms) {
 123          case 2: tile_flags = 0xfc; break;
 124          case 3: tile_flags = 0xfd; break;
 125          default:
 126             tile_flags = 0x70;
 127             break;
 128          }
 129          break;
 130       case 32:
 131          if (mt->base.base.bind & PIPE_BIND_SCANOUT) {
 132             assert(ms == 0);
 133             tile_flags = 0x7a;
 134          } else {
 135             switch (ms) {
 136             case 2: tile_flags = 0xf8; break;
 137             case 3: tile_flags = 0xf9; break;
 138             default:
 139                tile_flags = 0x70;
 140                break;
 141             }
 142          }
 143          break;
 144       case 16:
 145       case 8:
 146          tile_flags = 0x70;
 147          break;
 148       default:
 149          return 0;
 150       }
 151       if (mt->base.base.bind & PIPE_BIND_CURSOR)
 152          tile_flags = 0;
 153    }
 154
 155    if (!compressed)
 156       tile_flags &= ~0x180;
 157
 158    return tile_flags;
 159 }
 160
 161 void
 162 nv50_miptree_destroy(struct pipe_screen *pscreen, struct pipe_resource *pt)
 163 {
 164    struct nv50_miptree *mt = nv50_miptree(pt);
 165
 166    if (mt->base.fence && mt->base.fence->state < NOUVEAU_FENCE_STATE_FLUSHED)
 167       nouveau_fence_work(mt->base.fence, nouveau_fence_unref_bo, mt->base.bo);
 168    else
 169       nouveau_bo_ref(NULL, &mt->base.bo);
 170
 171    nouveau_fence_ref(NULL, &mt->base.fence);
 172    nouveau_fence_ref(NULL, &mt->base.fence_wr);
 173
 174    NOUVEAU_DRV_STAT(nouveau_screen(pscreen), tex_obj_current_count, -1);
 175    NOUVEAU_DRV_STAT(nouveau_screen(pscreen), tex_obj_current_bytes,
 176                     -(uint64_t)mt->total_size);
 177
 178    FREE(mt);
 179 }
 180
 181 bool
 182 nv50_miptree_get_handle(struct pipe_screen *pscreen,
 183                         struct pipe_resource *pt,
 184                         struct winsys_handle *whandle)
 185 {
 186    struct nv50_miptree *mt = nv50_miptree(pt);
 187    unsigned stride;
 188
 189    if (!mt || !mt->base.bo)
 190       return false;
 191
 192    stride = mt->level[0].pitch;
 193
 194    return nouveau_screen_bo_get_handle(pscreen,
 195                                        mt->base.bo,
 196                                        stride,
 197                                        whandle);
 198 }
 199
 200 const struct u_resource_vtbl nv50_miptree_vtbl =
 201 {
 202    nv50_miptree_get_handle,         /* get_handle */
 203    nv50_miptree_destroy,            /* resource_destroy */
 204    nv50_miptree_transfer_map,       /* transfer_map */
 205    u_default_transfer_flush_region, /* transfer_flush_region */
 206    nv50_miptree_transfer_unmap,     /* transfer_unmap */
 207 };
 208
 209 static inline bool
 210 nv50_miptree_init_ms_mode(struct nv50_miptree *mt)
 211 {
 212    switch (mt->base.base.nr_samples) {
 213    case 8:
 214       mt->ms_mode = NV50_3D_MULTISAMPLE_MODE_MS8;
 215       mt->ms_x = 2;
 216       mt->ms_y = 1;
 217       break;
 218    case 4:
 219       mt->ms_mode = NV50_3D_MULTISAMPLE_MODE_MS4;
 220       mt->ms_x = 1;
 221       mt->ms_y = 1;
 222       break;
 223    case 2:
 224       mt->ms_mode = NV50_3D_MULTISAMPLE_MODE_MS2;
 225       mt->ms_x = 1;
 226       break;
 227    case 1:
 228    case 0:
 229       mt->ms_mode = NV50_3D_MULTISAMPLE_MODE_MS1;
 230       break;
 231    default:
 232       NOUVEAU_ERR("invalid nr_samples: %u\n", mt->base.base.nr_samples);
 233       return false;
 234    }
 235    return true;
 236 }
 237
 238 bool
 239 nv50_miptree_init_layout_linear(struct nv50_miptree *mt, unsigned pitch_align)
 240 {
 241    struct pipe_resource *pt = &mt->base.base;
 242    const unsigned blocksize = util_format_get_blocksize(pt->format);
 243    unsigned h = pt->height0;
 244
 245    if (util_format_is_depth_or_stencil(pt->format))
 246       return false;
 247
 248    if ((pt->last_level > 0) || (pt->depth0 > 1) || (pt->array_size > 1))
 249       return false;
 250    if (mt->ms_x | mt->ms_y)
 251       return false;
 252
 253    mt->level[0].pitch = align(pt->width0 * blocksize, pitch_align);
 254
 255    /* Account for very generous prefetch (allocate size as if tiled). */
 256    h = MAX2(h, 8);
 257    h = util_next_power_of_two(h);
 258
 259    mt->total_size = mt->level[0].pitch * h;
 260
 261    return true;
 262 }
 263
 264 static void
 265 nv50_miptree_init_layout_video(struct nv50_miptree *mt)
 266 {
 267    const struct pipe_resource *pt = &mt->base.base;
 268    const unsigned blocksize = util_format_get_blocksize(pt->format);
 269
 270    assert(pt->last_level == 0);
 271    assert(mt->ms_x == 0 && mt->ms_y == 0);
 272    assert(!util_format_is_compressed(pt->format));
 273
 274    mt->layout_3d = pt->target == PIPE_TEXTURE_3D;
 275
 276    mt->level[0].tile_mode = 0x20;
 277    mt->level[0].pitch = align(pt->width0 * blocksize, 64);
 278    mt->total_size = align(pt->height0, 16) * mt->level[0].pitch * (mt->layout_3d ? pt->depth0 : 1);
 279
 280    if (pt->array_size > 1) {
 281       mt->layer_stride = align(mt->total_size, NV50_TILE_SIZE(0x20));
 282       mt->total_size = mt->layer_stride * pt->array_size;
 283    }
 284 }
 285
 286 static void
 287 nv50_miptree_init_layout_tiled(struct nv50_miptree *mt)
 288 {
 289    struct pipe_resource *pt = &mt->base.base;
 290    unsigned w, h, d, l;
 291    const unsigned blocksize = util_format_get_blocksize(pt->format);
 292
 293    mt->layout_3d = pt->target == PIPE_TEXTURE_3D;
 294
 295    w = pt->width0 << mt->ms_x;
 296    h = pt->height0 << mt->ms_y;
 297
 298    /* For 3D textures, a mipmap is spanned by all the layers, for array
 299     * textures and cube maps, each layer contains its own mipmaps.
 300     */
 301    d = mt->layout_3d ? pt->depth0 : 1;
 302
 303    for (l = 0; l <= pt->last_level; ++l) {
 304       struct nv50_miptree_level *lvl = &mt->level[l];
 305       unsigned tsx, tsy, tsz;
 306       unsigned nbx = util_format_get_nblocksx(pt->format, w);
 307       unsigned nby = util_format_get_nblocksy(pt->format, h);
 308
 309       lvl->offset = mt->total_size;
 310
 311       lvl->tile_mode = nv50_tex_choose_tile_dims(nbx, nby, d, mt->layout_3d);
 312
 313       tsx = NV50_TILE_SIZE_X(lvl->tile_mode); /* x is tile row pitch in bytes */
 314       tsy = NV50_TILE_SIZE_Y(lvl->tile_mode);
 315       tsz = NV50_TILE_SIZE_Z(lvl->tile_mode);
 316
 317       lvl->pitch = align(nbx * blocksize, tsx);
 318
 319       mt->total_size += lvl->pitch * align(nby, tsy) * align(d, tsz);
 320
 321       w = u_minify(w, 1);
 322       h = u_minify(h, 1);
 323       d = u_minify(d, 1);
 324    }
 325
 326    if (pt->array_size > 1) {
 327       mt->layer_stride = align(mt->total_size,
 328                                NV50_TILE_SIZE(mt->level[0].tile_mode));
 329       mt->total_size = mt->layer_stride * pt->array_size;
 330    }
 331 }
 332
 333 struct pipe_resource *
 334 nv50_miptree_create(struct pipe_screen *pscreen,
 335                     const struct pipe_resource *templ)
 336 {
 337    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
 338    struct nouveau_drm *drm = nouveau_screen(pscreen)->drm;
 339    struct nv50_miptree *mt = CALLOC_STRUCT(nv50_miptree);
 340    struct pipe_resource *pt = &mt->base.base;
 341    bool compressed = drm->version >= 0x01000101;
 342    int ret;
 343    union nouveau_bo_config bo_config;
 344    uint32_t bo_flags;
 345
 346    if (!mt)
 347       return NULL;
 348
 349    mt->base.vtbl = &nv50_miptree_vtbl;
 350    *pt = *templ;
 351    pipe_reference_init(&pt->reference, 1);
 352    pt->screen = pscreen;
 353
 354    if (pt->bind & PIPE_BIND_LINEAR)
 355       pt->flags |= NOUVEAU_RESOURCE_FLAG_LINEAR;
 356
 357    bo_config.nv50.memtype = nv50_mt_choose_storage_type(mt, compressed);
 358
 359    if (!nv50_miptree_init_ms_mode(mt)) {
 360       FREE(mt);
 361       return NULL;
 362    }
 363
 364    if (unlikely(pt->flags & NV50_RESOURCE_FLAG_VIDEO)) {
 365       nv50_miptree_init_layout_video(mt);
 366       if (pt->flags & NV50_RESOURCE_FLAG_NOALLOC) {
 367          /* BO allocation done by client */
 368          return pt;
 369       }
 370    } else
 371    if (bo_config.nv50.memtype != 0) {
 372       nv50_miptree_init_layout_tiled(mt);
 373    } else
 374    if (!nv50_miptree_init_layout_linear(mt, 64)) {
 375       FREE(mt);
 376       return NULL;
 377    }
 378    bo_config.nv50.tile_mode = mt->level[0].tile_mode;
 379
 380    if (!bo_config.nv50.memtype && (pt->bind & PIPE_BIND_SHARED))
 381       mt->base.domain = NOUVEAU_BO_GART;
 382    else
 383       mt->base.domain = NV_VRAM_DOMAIN(nouveau_screen(pscreen));
 384
 385    bo_flags = mt->base.domain | NOUVEAU_BO_NOSNOOP;
 386    if (mt->base.base.bind & (PIPE_BIND_CURSOR | PIPE_BIND_DISPLAY_TARGET))
 387       bo_flags |= NOUVEAU_BO_CONTIG;
 388
 389    ret = nouveau_bo_new(dev, bo_flags, 4096, mt->total_size, &bo_config,
 390                         &mt->base.bo);
 391    if (ret) {
 392       FREE(mt);
 393       return NULL;
 394    }
 395    mt->base.address = mt->base.bo->offset;
 396
 397    return pt;
 398 }
 399
 400 struct pipe_resource *
 401 nv50_miptree_from_handle(struct pipe_screen *pscreen,
 402                          const struct pipe_resource *templ,
 403                          struct winsys_handle *whandle)
 404 {
 405    struct nv50_miptree *mt;
 406    unsigned stride;
 407
 408    /* only supports 2D, non-mipmapped textures for the moment */
 409    if ((templ->target != PIPE_TEXTURE_2D &&
 410         templ->target != PIPE_TEXTURE_RECT) ||
 411        templ->last_level != 0 ||
 412        templ->depth0 != 1 ||
 413        templ->array_size > 1)
 414       return NULL;
 415
 416    mt = CALLOC_STRUCT(nv50_miptree);
 417    if (!mt)
 418       return NULL;
 419
 420    mt->base.bo = nouveau_screen_bo_from_handle(pscreen, whandle, &stride);
 421    if (mt->base.bo == NULL) {
 422       FREE(mt);
 423       return NULL;
 424    }
 425    mt->base.domain = mt->base.bo->flags & NOUVEAU_BO_APER;
 426    mt->base.address = mt->base.bo->offset;
 427
 428    mt->base.base = *templ;
 429    mt->base.vtbl = &nv50_miptree_vtbl;
 430    pipe_reference_init(&mt->base.base.reference, 1);
 431    mt->base.base.screen = pscreen;
 432    mt->level[0].pitch = stride;
 433    mt->level[0].offset = 0;
 434    mt->level[0].tile_mode = mt->base.bo->config.nv50.tile_mode;
 435
 436    NOUVEAU_DRV_STAT(nouveau_screen(pscreen), tex_obj_current_count, 1);
 437
 438    /* no need to adjust bo reference count */
 439    return &mt->base.base;
 440 }
 441
 442
 443 /* Offset of zslice @z from start of level @l. */
 444 inline unsigned
 445 nv50_mt_zslice_offset(const struct nv50_miptree *mt, unsigned l, unsigned z)
 446 {
 447    const struct pipe_resource *pt = &mt->base.base;
 448
 449    unsigned tds = NV50_TILE_SHIFT_Z(mt->level[l].tile_mode);
 450    unsigned ths = NV50_TILE_SHIFT_Y(mt->level[l].tile_mode);
 451
 452    unsigned nby = util_format_get_nblocksy(pt->format,
 453                                            u_minify(pt->height0, l));
 454
 455    /* to next 2D tile slice within a 3D tile */
 456    unsigned stride_2d = NV50_TILE_SIZE_2D(mt->level[l].tile_mode);
 457
 458    /* to slice in the next (in z direction) 3D tile */
 459    unsigned stride_3d = (align(nby, (1 << ths)) * mt->level[l].pitch) << tds;
 460
 461    return (z & ((1 << tds) - 1)) * stride_2d + (z >> tds) * stride_3d;
 462 }
 463
 464 /* Surface functions.
 465  */
 466
 467 struct nv50_surface *
 468 nv50_surface_from_miptree(struct nv50_miptree *mt,
 469                           const struct pipe_surface *templ)
 470 {
 471    struct pipe_surface *ps;
 472    struct nv50_surface *ns = CALLOC_STRUCT(nv50_surface);
 473    if (!ns)
 474       return NULL;
 475    ps = &ns->base;
 476
 477    pipe_reference_init(&ps->reference, 1);
 478    pipe_resource_reference(&ps->texture, &mt->base.base);
 479
 480    ps->format = templ->format;
 481    ps->writable = templ->writable;
 482    ps->u.tex.level = templ->u.tex.level;
 483    ps->u.tex.first_layer = templ->u.tex.first_layer;
 484    ps->u.tex.last_layer = templ->u.tex.last_layer;
 485
 486    ns->width = u_minify(mt->base.base.width0, ps->u.tex.level);
 487    ns->height = u_minify(mt->base.base.height0, ps->u.tex.level);
 488    ns->depth = ps->u.tex.last_layer - ps->u.tex.first_layer + 1;
 489    ns->offset = mt->level[templ->u.tex.level].offset;
 490
 491    /* comment says there are going to be removed, but they're used by the st */
 492    ps->width = ns->width;
 493    ps->height = ns->height;
 494
 495    ns->width <<= mt->ms_x;
 496    ns->height <<= mt->ms_y;
 497
 498    return ns;
 499 }
 500
 501 struct pipe_surface *
 502 nv50_miptree_surface_new(struct pipe_context *pipe,
 503                          struct pipe_resource *pt,
 504                          const struct pipe_surface *templ)
 505 {
 506    struct nv50_miptree *mt = nv50_miptree(pt);
 507    struct nv50_surface *ns = nv50_surface_from_miptree(mt, templ);
 508    if (!ns)
 509       return NULL;
 510    ns->base.context = pipe;
 511
 512    if (ns->base.u.tex.first_layer) {
 513       const unsigned l = ns->base.u.tex.level;
 514       const unsigned z = ns->base.u.tex.first_layer;
 515
 516       if (mt->layout_3d) {
 517          ns->offset += nv50_mt_zslice_offset(mt, l, z);
 518
 519          /* TODO: switch to depth 1 tiles; but actually this shouldn't happen */
 520          if (ns->depth > 1 &&
 521              (z & (NV50_TILE_SIZE_Z(mt->level[l].tile_mode) - 1)))
 522             NOUVEAU_ERR("Creating unsupported 3D surface !\n");
 523       } else {
 524          ns->offset += mt->layer_stride * z;
 525       }
 526    }
 527
 528    return &ns->base;
 529 }