src/gallium/drivers/svga/svga_resource_texture.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "svga3d_reg.h"
  27 #include "svga3d_surfacedefs.h"
  28
  29 #include "pipe/p_state.h"
  30 #include "pipe/p_defines.h"
  31 #include "os/os_thread.h"
  32 #include "os/os_time.h"
  33 #include "util/u_format.h"
  34 #include "util/u_inlines.h"
  35 #include "util/u_math.h"
  36 #include "util/u_memory.h"
  37 #include "util/u_resource.h"
  38
  39 #include "svga_cmd.h"
  40 #include "svga_format.h"
  41 #include "svga_screen.h"
  42 #include "svga_context.h"
  43 #include "svga_resource_texture.h"
  44 #include "svga_resource_buffer.h"
  45 #include "svga_sampler_view.h"
  46 #include "svga_winsys.h"
  47 #include "svga_debug.h"
  48
  49
  50 static void
  51 svga_transfer_dma_band(struct svga_context *svga,
  52                        struct svga_transfer *st,
  53                        SVGA3dTransferType transfer,
  54                        unsigned x, unsigned y, unsigned z,
  55                        unsigned w, unsigned h, unsigned d,
  56                        unsigned srcx, unsigned srcy, unsigned srcz,
  57                        SVGA3dSurfaceDMAFlags flags)
  58 {
  59    struct svga_texture *texture = svga_texture(st->base.resource);
  60    SVGA3dCopyBox box;
  61    enum pipe_error ret;
  62
  63    assert(!st->use_direct_map);
  64
  65    box.x = x;
  66    box.y = y;
  67    box.z = z;
  68    box.w = w;
  69    box.h = h;
  70    box.d = d;
  71    box.srcx = srcx;
  72    box.srcy = srcy;
  73    box.srcz = srcz;
  74
  75    SVGA_DBG(DEBUG_DMA, "dma %s sid %p, face %u, (%u, %u, %u) - "
  76             "(%u, %u, %u), %ubpp\n",
  77             transfer == SVGA3D_WRITE_HOST_VRAM ? "to" : "from",
  78             texture->handle,
  79             st->slice,
  80             x,
  81             y,
  82             z,
  83             x + w,
  84             y + h,
  85             z + 1,
  86             util_format_get_blocksize(texture->b.b.format) * 8 /
  87             (util_format_get_blockwidth(texture->b.b.format)
  88              * util_format_get_blockheight(texture->b.b.format)));
  89
  90    ret = SVGA3D_SurfaceDMA(svga->swc, st, transfer, &box, 1, flags);
  91    if (ret != PIPE_OK) {
  92       svga_context_flush(svga, NULL);
  93       ret = SVGA3D_SurfaceDMA(svga->swc, st, transfer, &box, 1, flags);
  94       assert(ret == PIPE_OK);
  95    }
  96 }
  97
  98
  99 static void
 100 svga_transfer_dma(struct svga_context *svga,
 101                   struct svga_transfer *st,
 102                   SVGA3dTransferType transfer,
 103                   SVGA3dSurfaceDMAFlags flags)
 104 {
 105    struct svga_texture *texture = svga_texture(st->base.resource);
 106    struct svga_screen *screen = svga_screen(texture->b.b.screen);
 107    struct svga_winsys_screen *sws = screen->sws;
 108    struct pipe_fence_handle *fence = NULL;
 109
 110    assert(!st->use_direct_map);
 111
 112    if (transfer == SVGA3D_READ_HOST_VRAM) {
 113       SVGA_DBG(DEBUG_PERF, "%s: readback transfer\n", __FUNCTION__);
 114    }
 115
 116    /* Ensure any pending operations on host surfaces are queued on the command
 117     * buffer first.
 118     */
 119    svga_surfaces_flush( svga );
 120
 121    if (!st->swbuf) {
 122       /* Do the DMA transfer in a single go */
 123       svga_transfer_dma_band(svga, st, transfer,
 124                              st->base.box.x, st->base.box.y, st->base.box.z,
 125                              st->base.box.width, st->base.box.height, st->base.box.depth,
 126                              0, 0, 0,
 127                              flags);
 128
 129       if (transfer == SVGA3D_READ_HOST_VRAM) {
 130          svga_context_flush(svga, &fence);
 131          sws->fence_finish(sws, fence, 0);
 132          sws->fence_reference(sws, &fence, NULL);
 133       }
 134    }
 135    else {
 136       int y, h, srcy;
 137       unsigned blockheight =
 138          util_format_get_blockheight(st->base.resource->format);
 139
 140       h = st->hw_nblocksy * blockheight;
 141       srcy = 0;
 142
 143       for (y = 0; y < st->base.box.height; y += h) {
 144          unsigned offset, length;
 145          void *hw, *sw;
 146
 147          if (y + h > st->base.box.height)
 148             h = st->base.box.height - y;
 149
 150          /* Transfer band must be aligned to pixel block boundaries */
 151          assert(y % blockheight == 0);
 152          assert(h % blockheight == 0);
 153
 154          offset = y * st->base.stride / blockheight;
 155          length = h * st->base.stride / blockheight;
 156
 157          sw = (uint8_t *) st->swbuf + offset;
 158
 159          if (transfer == SVGA3D_WRITE_HOST_VRAM) {
 160             unsigned usage = PIPE_TRANSFER_WRITE;
 161
 162             /* Wait for the previous DMAs to complete */
 163             /* TODO: keep one DMA (at half the size) in the background */
 164             if (y) {
 165                svga_context_flush(svga, NULL);
 166                usage |= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE;
 167             }
 168
 169             hw = sws->buffer_map(sws, st->hwbuf, usage);
 170             assert(hw);
 171             if (hw) {
 172                memcpy(hw, sw, length);
 173                sws->buffer_unmap(sws, st->hwbuf);
 174             }
 175          }
 176
 177          svga_transfer_dma_band(svga, st, transfer,
 178                                 st->base.box.x, y, st->base.box.z,
 179                                 st->base.box.width, h, st->base.box.depth,
 180                                 0, srcy, 0, flags);
 181
 182          /*
 183           * Prevent the texture contents to be discarded on the next band
 184           * upload.
 185           */
 186          flags.discard = FALSE;
 187
 188          if (transfer == SVGA3D_READ_HOST_VRAM) {
 189             svga_context_flush(svga, &fence);
 190             sws->fence_finish(sws, fence, 0);
 191
 192             hw = sws->buffer_map(sws, st->hwbuf, PIPE_TRANSFER_READ);
 193             assert(hw);
 194             if (hw) {
 195                memcpy(sw, hw, length);
 196                sws->buffer_unmap(sws, st->hwbuf);
 197             }
 198          }
 199       }
 200    }
 201 }
 202
 203
 204 static boolean
 205 svga_texture_get_handle(struct pipe_screen *screen,
 206                         struct pipe_resource *texture,
 207                         struct winsys_handle *whandle)
 208 {
 209    struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
 210    unsigned stride;
 211
 212    assert(svga_texture(texture)->key.cachable == 0);
 213    svga_texture(texture)->key.cachable = 0;
 214
 215    stride = util_format_get_nblocksx(texture->format, texture->width0) *
 216             util_format_get_blocksize(texture->format);
 217
 218    return sws->surface_get_handle(sws, svga_texture(texture)->handle,
 219                                   stride, whandle);
 220 }
 221
 222
 223 static void
 224 svga_texture_destroy(struct pipe_screen *screen,
 225                      struct pipe_resource *pt)
 226 {
 227    struct svga_screen *ss = svga_screen(screen);
 228    struct svga_texture *tex = svga_texture(pt);
 229
 230    ss->texture_timestamp++;
 231
 232    svga_sampler_view_reference(&tex->cached_view, NULL);
 233
 234    /*
 235      DBG("%s deleting %p\n", __FUNCTION__, (void *) tex);
 236    */
 237    SVGA_DBG(DEBUG_DMA, "unref sid %p (texture)\n", tex->handle);
 238    svga_screen_surface_destroy(ss, &tex->key, &tex->handle);
 239
 240    ss->hud.total_resource_bytes -= tex->size;
 241
 242    FREE(tex->defined);
 243    FREE(tex->rendered_to);
 244    FREE(tex->dirty);
 245    FREE(tex);
 246
 247    assert(ss->hud.num_resources > 0);
 248    if (ss->hud.num_resources > 0)
 249       ss->hud.num_resources--;
 250 }
 251
 252
 253 /**
 254  * Determine if we need to read back a texture image before mapping it.
 255  */
 256 static boolean
 257 need_tex_readback(struct pipe_transfer *transfer)
 258 {
 259    struct svga_texture *t = svga_texture(transfer->resource);
 260
 261    if (transfer->usage & PIPE_TRANSFER_READ)
 262       return TRUE;
 263
 264    if ((transfer->usage & PIPE_TRANSFER_WRITE) &&
 265        ((transfer->usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) == 0)) {
 266       unsigned face;
 267
 268       if (transfer->resource->target == PIPE_TEXTURE_CUBE) {
 269          assert(transfer->box.depth == 1);
 270          face = transfer->box.z;
 271       }
 272       else {
 273          face = 0;
 274       }
 275       if (svga_was_texture_rendered_to(t, face, transfer->level)) {
 276          return TRUE;
 277       }
 278    }
 279
 280    return FALSE;
 281 }
 282
 283
 284 static enum pipe_error
 285 readback_image_vgpu9(struct svga_context *svga,
 286                    struct svga_winsys_surface *surf,
 287                    unsigned slice,
 288                    unsigned level)
 289 {
 290    enum pipe_error ret;
 291
 292    ret = SVGA3D_ReadbackGBImage(svga->swc, surf, slice, level);
 293    if (ret != PIPE_OK) {
 294       svga_context_flush(svga, NULL);
 295       ret = SVGA3D_ReadbackGBImage(svga->swc, surf, slice, level);
 296    }
 297    return ret;
 298 }
 299
 300
 301 static enum pipe_error
 302 readback_image_vgpu10(struct svga_context *svga,
 303                     struct svga_winsys_surface *surf,
 304                     unsigned slice,
 305                     unsigned level,
 306                     unsigned numMipLevels)
 307 {
 308    enum pipe_error ret;
 309    unsigned subResource;
 310
 311    subResource = slice * numMipLevels + level;
 312    ret = SVGA3D_vgpu10_ReadbackSubResource(svga->swc, surf, subResource);
 313    if (ret != PIPE_OK) {
 314       svga_context_flush(svga, NULL);
 315       ret = SVGA3D_vgpu10_ReadbackSubResource(svga->swc, surf, subResource);
 316    }
 317    return ret;
 318 }
 319
 320
 321 static void *
 322 svga_texture_transfer_map(struct pipe_context *pipe,
 323                           struct pipe_resource *texture,
 324                           unsigned level,
 325                           unsigned usage,
 326                           const struct pipe_box *box,
 327                           struct pipe_transfer **ptransfer)
 328 {
 329    struct svga_context *svga = svga_context(pipe);
 330    struct svga_screen *ss = svga_screen(pipe->screen);
 331    struct svga_winsys_screen *sws = ss->sws;
 332    struct svga_texture *tex = svga_texture(texture);
 333    struct svga_transfer *st;
 334    unsigned nblocksx, nblocksy;
 335    boolean use_direct_map = svga_have_gb_objects(svga) &&
 336       !svga_have_gb_dma(svga);
 337    unsigned d;
 338    void *returnVal;
 339    int64_t begin = os_time_get();
 340
 341    /* We can't map texture storage directly unless we have GB objects */
 342    if (usage & PIPE_TRANSFER_MAP_DIRECTLY) {
 343       if (svga_have_gb_objects(svga))
 344          use_direct_map = TRUE;
 345       else
 346          return NULL;
 347    }
 348
 349    st = CALLOC_STRUCT(svga_transfer);
 350    if (!st)
 351       return NULL;
 352
 353    st->base.level = level;
 354    st->base.usage = usage;
 355    st->base.box = *box;
 356
 357    switch (tex->b.b.target) {
 358    case PIPE_TEXTURE_CUBE:
 359       st->slice = st->base.box.z;
 360       st->base.box.z = 0;   /* so we don't apply double offsets below */
 361       break;
 362    case PIPE_TEXTURE_2D_ARRAY:
 363    case PIPE_TEXTURE_1D_ARRAY:
 364       st->slice = st->base.box.z;
 365       st->base.box.z = 0;   /* so we don't apply double offsets below */
 366
 367       /* Force direct map for transfering multiple slices */
 368       if (st->base.box.depth > 1)
 369          use_direct_map = svga_have_gb_objects(svga);
 370
 371       break;
 372    default:
 373       st->slice = 0;
 374       break;
 375    }
 376
 377    {
 378       unsigned w, h;
 379       if (use_direct_map) {
 380          /* we'll directly access the guest-backed surface */
 381          w = u_minify(texture->width0, level);
 382          h = u_minify(texture->height0, level);
 383          d = u_minify(texture->depth0, level);
 384       }
 385       else {
 386          /* we'll put the data into a tightly packed buffer */
 387          w = box->width;
 388          h = box->height;
 389          d = box->depth;
 390       }
 391       nblocksx = util_format_get_nblocksx(texture->format, w);
 392       nblocksy = util_format_get_nblocksy(texture->format, h);
 393    }
 394
 395    pipe_resource_reference(&st->base.resource, texture);
 396
 397    st->base.stride = nblocksx*util_format_get_blocksize(texture->format);
 398    st->base.layer_stride = st->base.stride * nblocksy;
 399
 400    if (usage & PIPE_TRANSFER_WRITE) {
 401       /* record texture upload for HUD */
 402       svga->hud.num_bytes_uploaded +=
 403          nblocksx * nblocksy * d * util_format_get_blocksize(texture->format);
 404    }
 405
 406    if (!use_direct_map) {
 407       /* Use a DMA buffer */
 408       st->hw_nblocksy = nblocksy;
 409
 410       st->hwbuf = svga_winsys_buffer_create(svga, 1, 0,
 411                                    st->hw_nblocksy * st->base.stride * d);
 412       while(!st->hwbuf && (st->hw_nblocksy /= 2)) {
 413          st->hwbuf = svga_winsys_buffer_create(svga, 1, 0,
 414                                    st->hw_nblocksy * st->base.stride * d);
 415       }
 416
 417       if (!st->hwbuf) {
 418          FREE(st);
 419          return NULL;
 420       }
 421
 422       if (st->hw_nblocksy < nblocksy) {
 423          /* We couldn't allocate a hardware buffer big enough for the transfer,
 424           * so allocate regular malloc memory instead */
 425          if (0) {
 426             debug_printf("%s: failed to allocate %u KB of DMA, "
 427                          "splitting into %u x %u KB DMA transfers\n",
 428                          __FUNCTION__,
 429                          (nblocksy*st->base.stride + 1023)/1024,
 430                          (nblocksy + st->hw_nblocksy - 1)/st->hw_nblocksy,
 431                          (st->hw_nblocksy*st->base.stride + 1023)/1024);
 432          }
 433
 434          st->swbuf = MALLOC(nblocksy * st->base.stride * d);
 435          if (!st->swbuf) {
 436             sws->buffer_destroy(sws, st->hwbuf);
 437             FREE(st);
 438             return NULL;
 439          }
 440       }
 441
 442       if (usage & PIPE_TRANSFER_READ) {
 443          SVGA3dSurfaceDMAFlags flags;
 444          memset(&flags, 0, sizeof flags);
 445          svga_transfer_dma(svga, st, SVGA3D_READ_HOST_VRAM, flags);
 446       }
 447    } else {
 448       struct pipe_transfer *transfer = &st->base;
 449       struct svga_winsys_surface *surf = tex->handle;
 450
 451       if (!surf) {
 452          FREE(st);
 453          return NULL;
 454       }
 455
 456       /* If this is the first time mapping to the surface in this
 457        * command buffer, clear the dirty masks of this surface.
 458        */
 459       if (sws->surface_is_flushed(sws, surf)) {
 460          svga_clear_texture_dirty(tex);
 461       }
 462
 463       if (need_tex_readback(transfer)) {
 464          enum pipe_error ret;
 465
 466          svga_surfaces_flush(svga);
 467
 468          if (svga_have_vgpu10(svga)) {
 469             ret = readback_image_vgpu10(svga, surf, st->slice, transfer->level,
 470                                         tex->b.b.last_level + 1);
 471          } else {
 472             ret = readback_image_vgpu9(svga, surf, st->slice, transfer->level);
 473          }
 474
 475          svga->hud.num_readbacks++;
 476
 477          assert(ret == PIPE_OK);
 478          (void) ret;
 479
 480          svga_context_flush(svga, NULL);
 481
 482          /*
 483           * Note: if PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE were specified
 484           * we could potentially clear the flag for all faces/layers/mips.
 485           */
 486          svga_clear_texture_rendered_to(tex, st->slice, transfer->level);
 487       }
 488       else {
 489          assert(transfer->usage & PIPE_TRANSFER_WRITE);
 490          if ((transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) == 0) {
 491             if (svga_is_texture_dirty(tex, st->slice, transfer->level)) {
 492                /*
 493                 * do a surface flush if the subresource has been modified
 494                 * in this command buffer.
 495                 */
 496                svga_surfaces_flush(svga);
 497                if (!sws->surface_is_flushed(sws, surf)) {
 498                   svga->hud.surface_write_flushes++;
 499                   svga_context_flush(svga, NULL);
 500                }
 501             }
 502          }
 503       }
 504       if (transfer->usage & PIPE_TRANSFER_WRITE) {
 505          /* mark this texture level as dirty */
 506          svga_set_texture_dirty(tex, st->slice, transfer->level);
 507       }
 508    }
 509
 510    st->use_direct_map = use_direct_map;
 511
 512    *ptransfer = &st->base;
 513
 514    /*
 515     * Begin mapping code
 516     */
 517    if (st->swbuf) {
 518       returnVal = st->swbuf;
 519    }
 520    else if (!st->use_direct_map) {
 521       returnVal = sws->buffer_map(sws, st->hwbuf, usage);
 522    }
 523    else {
 524       SVGA3dSize baseLevelSize;
 525       struct svga_texture *tex = svga_texture(texture);
 526       struct svga_winsys_surface *surf = tex->handle;
 527       uint8_t *map;
 528       boolean retry;
 529       unsigned offset, mip_width, mip_height;
 530       unsigned xoffset = st->base.box.x;
 531       unsigned yoffset = st->base.box.y;
 532       unsigned zoffset = st->base.box.z;
 533
 534       map = svga->swc->surface_map(svga->swc, surf, usage, &retry);
 535       if (map == NULL && retry) {
 536          /*
 537           * At this point, the svga_surfaces_flush() should already have
 538           * called in svga_texture_get_transfer().
 539           */
 540          svga_context_flush(svga, NULL);
 541          map = svga->swc->surface_map(svga->swc, surf, usage, &retry);
 542       }
 543
 544       /*
 545        * Make sure we return NULL if the map fails
 546        */
 547       if (!map) {
 548          FREE(st);
 549          return map;
 550       }
 551
 552       /**
 553        * Compute the offset to the specific texture slice in the buffer.
 554        */
 555       baseLevelSize.width = tex->b.b.width0;
 556       baseLevelSize.height = tex->b.b.height0;
 557       baseLevelSize.depth = tex->b.b.depth0;
 558
 559       if ((tex->b.b.target == PIPE_TEXTURE_1D_ARRAY) ||
 560           (tex->b.b.target == PIPE_TEXTURE_2D_ARRAY)) {
 561          st->base.layer_stride =
 562             svga3dsurface_get_image_offset(tex->key.format, baseLevelSize,
 563                                            tex->b.b.last_level + 1, 1, 0);
 564       }
 565
 566       offset = svga3dsurface_get_image_offset(tex->key.format, baseLevelSize,
 567                                               tex->b.b.last_level + 1, /* numMips */
 568                                               st->slice, level);
 569       if (level > 0) {
 570          assert(offset > 0);
 571       }
 572
 573       mip_width = u_minify(tex->b.b.width0, level);
 574       mip_height = u_minify(tex->b.b.height0, level);
 575
 576       offset += svga3dsurface_get_pixel_offset(tex->key.format,
 577                                                mip_width, mip_height,
 578                                                xoffset, yoffset, zoffset);
 579       returnVal = (void *) (map + offset);
 580    }
 581
 582    svga->hud.map_buffer_time += (os_time_get() - begin);
 583    svga->hud.num_resources_mapped++;
 584
 585    return returnVal;
 586 }
 587
 588
 589 /**
 590  * Unmap a GB texture surface.
 591  */
 592 static void
 593 svga_texture_surface_unmap(struct svga_context *svga,
 594                            struct pipe_transfer *transfer)
 595 {
 596    struct svga_winsys_surface *surf = svga_texture(transfer->resource)->handle;
 597    struct svga_winsys_context *swc = svga->swc;
 598    boolean rebind;
 599
 600    assert(surf);
 601
 602    swc->surface_unmap(swc, surf, &rebind);
 603    if (rebind) {
 604       enum pipe_error ret;
 605       ret = SVGA3D_BindGBSurface(swc, surf);
 606       if (ret != PIPE_OK) {
 607          /* flush and retry */
 608          svga_context_flush(svga, NULL);
 609          ret = SVGA3D_BindGBSurface(swc, surf);
 610          assert(ret == PIPE_OK);
 611       }
 612    }
 613 }
 614
 615
 616 static enum pipe_error
 617 update_image_vgpu9(struct svga_context *svga,
 618                    struct svga_winsys_surface *surf,
 619                    const SVGA3dBox *box,
 620                    unsigned slice,
 621                    unsigned level)
 622 {
 623    enum pipe_error ret;
 624
 625    ret = SVGA3D_UpdateGBImage(svga->swc, surf, box, slice, level);
 626    if (ret != PIPE_OK) {
 627       svga_context_flush(svga, NULL);
 628       ret = SVGA3D_UpdateGBImage(svga->swc, surf, box, slice, level);
 629    }
 630    return ret;
 631 }
 632
 633
 634 static enum pipe_error
 635 update_image_vgpu10(struct svga_context *svga,
 636                     struct svga_winsys_surface *surf,
 637                     const SVGA3dBox *box,
 638                     unsigned slice,
 639                     unsigned level,
 640                     unsigned numMipLevels)
 641 {
 642    enum pipe_error ret;
 643    unsigned subResource;
 644
 645    subResource = slice * numMipLevels + level;
 646    ret = SVGA3D_vgpu10_UpdateSubResource(svga->swc, surf, box, subResource);
 647    if (ret != PIPE_OK) {
 648       svga_context_flush(svga, NULL);
 649       ret = SVGA3D_vgpu10_UpdateSubResource(svga->swc, surf, box, subResource);
 650    }
 651    return ret;
 652 }
 653
 654
 655 static void
 656 svga_texture_transfer_unmap(struct pipe_context *pipe,
 657                             struct pipe_transfer *transfer)
 658 {
 659    struct svga_context *svga = svga_context(pipe);
 660    struct svga_screen *ss = svga_screen(pipe->screen);
 661    struct svga_winsys_screen *sws = ss->sws;
 662    struct svga_transfer *st = svga_transfer(transfer);
 663    struct svga_texture *tex = svga_texture(transfer->resource);
 664
 665    if (!st->swbuf) {
 666       if (st->use_direct_map) {
 667          svga_texture_surface_unmap(svga, transfer);
 668       }
 669       else {
 670          sws->buffer_unmap(sws, st->hwbuf);
 671       }
 672    }
 673
 674    if (!st->use_direct_map && (st->base.usage & PIPE_TRANSFER_WRITE)) {
 675       /* Use DMA to transfer texture data */
 676       SVGA3dSurfaceDMAFlags flags;
 677
 678       memset(&flags, 0, sizeof flags);
 679       if (transfer->usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
 680          flags.discard = TRUE;
 681       }
 682       if (transfer->usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
 683          flags.unsynchronized = TRUE;
 684       }
 685
 686       svga_transfer_dma(svga, st, SVGA3D_WRITE_HOST_VRAM, flags);
 687    } else if (transfer->usage & PIPE_TRANSFER_WRITE) {
 688       struct svga_winsys_surface *surf =
 689          svga_texture(transfer->resource)->handle;
 690       SVGA3dBox box;
 691       enum pipe_error ret;
 692       unsigned nlayers = 1;
 693
 694       assert(svga_have_gb_objects(svga));
 695
 696       /* update the effected region */
 697       box.x = transfer->box.x;
 698       box.y = transfer->box.y;
 699       box.w = transfer->box.width;
 700       box.h = transfer->box.height;
 701       box.d = transfer->box.depth;
 702
 703       switch (tex->b.b.target) {
 704       case PIPE_TEXTURE_CUBE:
 705          box.z = 0;
 706          break;
 707       case PIPE_TEXTURE_2D_ARRAY:
 708          nlayers = box.d;
 709          box.z = 0;
 710          box.d = 1;
 711          break;
 712       case PIPE_TEXTURE_1D_ARRAY:
 713          nlayers = box.d;
 714          box.y = box.z = 0;
 715          box.d = 1;
 716          break;
 717       default:
 718          box.z = transfer->box.z;
 719          break;
 720       }
 721
 722       if (0)
 723          debug_printf("%s %d, %d, %d  %d x %d x %d\n",
 724                       __FUNCTION__,
 725                       box.x, box.y, box.z,
 726                       box.w, box.h, box.d);
 727
 728       if (svga_have_vgpu10(svga)) {
 729          unsigned i;
 730          for (i = 0; i < nlayers; i++) {
 731             ret = update_image_vgpu10(svga, surf, &box,
 732                                       st->slice + i, transfer->level,
 733                                       tex->b.b.last_level + 1);
 734             assert(ret == PIPE_OK);
 735          }
 736       } else {
 737          assert(nlayers == 1);
 738          ret = update_image_vgpu9(svga, surf, &box, st->slice, transfer->level);
 739          assert(ret == PIPE_OK);
 740       }
 741
 742       svga->hud.num_resource_updates++;
 743
 744       (void) ret;
 745    }
 746
 747    ss->texture_timestamp++;
 748    svga_age_texture_view(tex, transfer->level);
 749    if (transfer->resource->target == PIPE_TEXTURE_CUBE)
 750       svga_define_texture_level(tex, st->slice, transfer->level);
 751    else
 752       svga_define_texture_level(tex, 0, transfer->level);
 753
 754    pipe_resource_reference(&st->base.resource, NULL);
 755
 756    FREE(st->swbuf);
 757    if (!st->use_direct_map) {
 758       sws->buffer_destroy(sws, st->hwbuf);
 759    }
 760    FREE(st);
 761 }
 762
 763
 764 /**
 765  * Does format store depth values?
 766  */
 767 static inline boolean
 768 format_has_depth(enum pipe_format format)
 769 {
 770    const struct util_format_description *desc = util_format_description(format);
 771    return util_format_has_depth(desc);
 772 }
 773
 774
 775 struct u_resource_vtbl svga_texture_vtbl =
 776 {
 777    svga_texture_get_handle,           /* get_handle */
 778    svga_texture_destroy,              /* resource_destroy */
 779    svga_texture_transfer_map,         /* transfer_map */
 780    u_default_transfer_flush_region,   /* transfer_flush_region */
 781    svga_texture_transfer_unmap,       /* transfer_unmap */
 782 };
 783
 784
 785 struct pipe_resource *
 786 svga_texture_create(struct pipe_screen *screen,
 787                     const struct pipe_resource *template)
 788 {
 789    struct svga_screen *svgascreen = svga_screen(screen);
 790    struct svga_texture *tex;
 791    unsigned bindings = template->bind;
 792
 793    assert(template->last_level < SVGA_MAX_TEXTURE_LEVELS);
 794    if (template->last_level >= SVGA_MAX_TEXTURE_LEVELS) {
 795       return NULL;
 796    }
 797
 798    tex = CALLOC_STRUCT(svga_texture);
 799    if (!tex) {
 800       return NULL;
 801    }
 802
 803    tex->defined = CALLOC(template->depth0 * template->array_size,
 804                          sizeof(tex->defined[0]));
 805    if (!tex->defined) {
 806       FREE(tex);
 807       return NULL;
 808    }
 809
 810    tex->rendered_to = CALLOC(template->depth0 * template->array_size,
 811                              sizeof(tex->rendered_to[0]));
 812    if (!tex->rendered_to) {
 813       goto fail;
 814    }
 815
 816    tex->dirty = CALLOC(template->depth0 * template->array_size,
 817                              sizeof(tex->dirty[0]));
 818    if (!tex->dirty) {
 819       goto fail;
 820    }
 821
 822    tex->b.b = *template;
 823    tex->b.vtbl = &svga_texture_vtbl;
 824    pipe_reference_init(&tex->b.b.reference, 1);
 825    tex->b.b.screen = screen;
 826
 827    tex->key.flags = 0;
 828    tex->key.size.width = template->width0;
 829    tex->key.size.height = template->height0;
 830    tex->key.size.depth = template->depth0;
 831    tex->key.arraySize = 1;
 832    tex->key.numFaces = 1;
 833    tex->key.sampleCount = template->nr_samples;
 834
 835    if (template->nr_samples > 1) {
 836       tex->key.flags |= SVGA3D_SURFACE_MASKABLE_ANTIALIAS;
 837    }
 838
 839    if (svgascreen->sws->have_vgpu10) {
 840       switch (template->target) {
 841       case PIPE_TEXTURE_1D:
 842          tex->key.flags |= SVGA3D_SURFACE_1D;
 843          break;
 844       case PIPE_TEXTURE_1D_ARRAY:
 845          tex->key.flags |= SVGA3D_SURFACE_1D;
 846          /* fall-through */
 847       case PIPE_TEXTURE_2D_ARRAY:
 848          tex->key.flags |= SVGA3D_SURFACE_ARRAY;
 849          tex->key.arraySize = template->array_size;
 850          break;
 851       case PIPE_TEXTURE_3D:
 852          tex->key.flags |= SVGA3D_SURFACE_VOLUME;
 853          break;
 854       case PIPE_TEXTURE_CUBE:
 855          tex->key.flags |= (SVGA3D_SURFACE_CUBEMAP | SVGA3D_SURFACE_ARRAY);
 856          tex->key.numFaces = 6;
 857          break;
 858       default:
 859          break;
 860       }
 861    }
 862    else {
 863       switch (template->target) {
 864       case PIPE_TEXTURE_3D:
 865          tex->key.flags |= SVGA3D_SURFACE_VOLUME;
 866          break;
 867       case PIPE_TEXTURE_CUBE:
 868          tex->key.flags |= SVGA3D_SURFACE_CUBEMAP;
 869          tex->key.numFaces = 6;
 870          break;
 871       default:
 872          break;
 873       }
 874    }
 875
 876    tex->key.cachable = 1;
 877
 878    if ((bindings & (PIPE_BIND_RENDER_TARGET | PIPE_BIND_DEPTH_STENCIL)) &&
 879        !(bindings & PIPE_BIND_SAMPLER_VIEW)) {
 880       /* Also check if the format can be sampled from */
 881       if (screen->is_format_supported(screen, template->format,
 882                                       template->target,
 883                                       template->nr_samples,
 884                                       PIPE_BIND_SAMPLER_VIEW)) {
 885          bindings |= PIPE_BIND_SAMPLER_VIEW;
 886       }
 887    }
 888
 889    if (bindings & PIPE_BIND_SAMPLER_VIEW) {
 890       tex->key.flags |= SVGA3D_SURFACE_HINT_TEXTURE;
 891       tex->key.flags |= SVGA3D_SURFACE_BIND_SHADER_RESOURCE;
 892
 893       if (!(bindings & PIPE_BIND_RENDER_TARGET)) {
 894          /* Also check if the format is renderable */
 895          if (screen->is_format_supported(screen, template->format,
 896                                          template->target,
 897                                          template->nr_samples,
 898                                          PIPE_BIND_RENDER_TARGET)) {
 899             bindings |= PIPE_BIND_RENDER_TARGET;
 900          }
 901       }
 902    }
 903
 904    if (bindings & PIPE_BIND_DISPLAY_TARGET) {
 905       tex->key.cachable = 0;
 906    }
 907
 908    if (bindings & PIPE_BIND_SHARED) {
 909       tex->key.cachable = 0;
 910    }
 911
 912    if (bindings & (PIPE_BIND_SCANOUT | PIPE_BIND_CURSOR)) {
 913       tex->key.scanout = 1;
 914       tex->key.cachable = 0;
 915    }
 916
 917    /*
 918     * Note: Previously we never passed the
 919     * SVGA3D_SURFACE_HINT_RENDERTARGET hint. Mesa cannot
 920     * know beforehand whether a texture will be used as a rendertarget or not
 921     * and it always requests PIPE_BIND_RENDER_TARGET, therefore
 922     * passing the SVGA3D_SURFACE_HINT_RENDERTARGET here defeats its purpose.
 923     *
 924     * However, this was changed since other state trackers
 925     * (XA for example) uses it accurately and certain device versions
 926     * relies on it in certain situations to render correctly.
 927     */
 928    if ((bindings & PIPE_BIND_RENDER_TARGET) &&
 929        !util_format_is_s3tc(template->format)) {
 930       tex->key.flags |= SVGA3D_SURFACE_HINT_RENDERTARGET;
 931       tex->key.flags |= SVGA3D_SURFACE_BIND_RENDER_TARGET;
 932    }
 933
 934    if (bindings & PIPE_BIND_DEPTH_STENCIL) {
 935       tex->key.flags |= SVGA3D_SURFACE_HINT_DEPTHSTENCIL;
 936       tex->key.flags |= SVGA3D_SURFACE_BIND_DEPTH_STENCIL;
 937    }
 938
 939    tex->key.numMipLevels = template->last_level + 1;
 940
 941    tex->key.format = svga_translate_format(svgascreen, template->format,
 942                                            bindings);
 943    if (tex->key.format == SVGA3D_FORMAT_INVALID) {
 944       goto fail;
 945    }
 946
 947    /* The actual allocation is done with a typeless format.  Typeless
 948     * formats can be reinterpreted as other formats.  For example,
 949     * SVGA3D_R8G8B8A8_UNORM_TYPELESS can be interpreted as
 950     * SVGA3D_R8G8B8A8_UNORM_SRGB or SVGA3D_R8G8B8A8_UNORM.
 951     * Do not use typeless formats for SHARED, DISPLAY_TARGET or SCANOUT
 952     * buffers.
 953     */
 954    if (svgascreen->sws->have_vgpu10
 955        && ((bindings & (PIPE_BIND_SHARED |
 956                         PIPE_BIND_DISPLAY_TARGET |
 957                         PIPE_BIND_SCANOUT)) == 0)) {
 958       SVGA3dSurfaceFormat typeless = svga_typeless_format(tex->key.format);
 959       if (0) {
 960          debug_printf("Convert resource type %s -> %s (bind 0x%x)\n",
 961                       svga_format_name(tex->key.format),
 962                       svga_format_name(typeless),
 963                       bindings);
 964       }
 965
 966       if (svga_format_is_uncompressed_snorm(tex->key.format)) {
 967          /* We can't normally render to snorm surfaces, but once we
 968           * substitute a typeless format, we can if the rendertarget view
 969           * is unorm.  This can happen with GL_ARB_copy_image.
 970           */
 971          tex->key.flags |= SVGA3D_SURFACE_HINT_RENDERTARGET;
 972          tex->key.flags |= SVGA3D_SURFACE_BIND_RENDER_TARGET;
 973       }
 974
 975       tex->key.format = typeless;
 976    }
 977
 978    SVGA_DBG(DEBUG_DMA, "surface_create for texture\n", tex->handle);
 979    tex->handle = svga_screen_surface_create(svgascreen, bindings,
 980                                             tex->b.b.usage, &tex->key);
 981    if (!tex->handle) {
 982       goto fail;
 983    }
 984
 985    SVGA_DBG(DEBUG_DMA, "  --> got sid %p (texture)\n", tex->handle);
 986
 987    debug_reference(&tex->b.b.reference,
 988                    (debug_reference_descriptor)debug_describe_resource, 0);
 989
 990    tex->size = util_resource_size(template);
 991    svgascreen->hud.total_resource_bytes += tex->size;
 992    svgascreen->hud.num_resources++;
 993
 994    return &tex->b.b;
 995
 996 fail:
 997    if (tex->dirty)
 998       FREE(tex->dirty);
 999    if (tex->rendered_to)
1000       FREE(tex->rendered_to);
1001    if (tex->defined)
1002       FREE(tex->defined);
1003    FREE(tex);
1004    return NULL;
1005 }
1006
1007
1008 struct pipe_resource *
1009 svga_texture_from_handle(struct pipe_screen *screen,
1010                          const struct pipe_resource *template,
1011                          struct winsys_handle *whandle)
1012 {
1013    struct svga_winsys_screen *sws = svga_winsys_screen(screen);
1014    struct svga_screen *ss = svga_screen(screen);
1015    struct svga_winsys_surface *srf;
1016    struct svga_texture *tex;
1017    enum SVGA3dSurfaceFormat format = 0;
1018    assert(screen);
1019
1020    /* Only supports one type */
1021    if ((template->target != PIPE_TEXTURE_2D &&
1022        template->target != PIPE_TEXTURE_RECT) ||
1023        template->last_level != 0 ||
1024        template->depth0 != 1) {
1025       return NULL;
1026    }
1027
1028    srf = sws->surface_from_handle(sws, whandle, &format);
1029
1030    if (!srf)
1031       return NULL;
1032
1033    if (svga_translate_format(svga_screen(screen), template->format,
1034                              template->bind) != format) {
1035       unsigned f1 = svga_translate_format(svga_screen(screen),
1036                                           template->format, template->bind);
1037       unsigned f2 = format;
1038
1039       /* It's okay for XRGB and ARGB or depth with/out stencil to get mixed up.
1040        */
1041       if (f1 == SVGA3D_B8G8R8A8_UNORM)
1042          f1 = SVGA3D_A8R8G8B8;
1043       if (f1 == SVGA3D_B8G8R8X8_UNORM)
1044          f1 = SVGA3D_X8R8G8B8;
1045
1046       if ( !( (f1 == f2) ||
1047               (f1 == SVGA3D_X8R8G8B8 && f2 == SVGA3D_A8R8G8B8) ||
1048               (f1 == SVGA3D_X8R8G8B8 && f2 == SVGA3D_B8G8R8X8_UNORM) ||
1049               (f1 == SVGA3D_A8R8G8B8 && f2 == SVGA3D_X8R8G8B8) ||
1050               (f1 == SVGA3D_A8R8G8B8 && f2 == SVGA3D_B8G8R8A8_UNORM) ||
1051               (f1 == SVGA3D_Z_D24X8 && f2 == SVGA3D_Z_D24S8) ||
1052               (f1 == SVGA3D_Z_DF24 && f2 == SVGA3D_Z_D24S8_INT) ) ) {
1053          debug_printf("%s wrong format %s != %s\n", __FUNCTION__,
1054                       svga_format_name(f1), svga_format_name(f2));
1055          return NULL;
1056       }
1057    }
1058
1059    tex = CALLOC_STRUCT(svga_texture);
1060    if (!tex)
1061       return NULL;
1062
1063    tex->defined = CALLOC(template->depth0 * template->array_size,
1064                          sizeof(tex->defined[0]));
1065    if (!tex->defined) {
1066       FREE(tex);
1067       return NULL;
1068    }
1069
1070    tex->b.b = *template;
1071    tex->b.vtbl = &svga_texture_vtbl;
1072    pipe_reference_init(&tex->b.b.reference, 1);
1073    tex->b.b.screen = screen;
1074
1075    SVGA_DBG(DEBUG_DMA, "wrap surface sid %p\n", srf);
1076
1077    tex->key.cachable = 0;
1078    tex->key.format = format;
1079    tex->handle = srf;
1080
1081    tex->rendered_to = CALLOC(1, sizeof(tex->rendered_to[0]));
1082    if (!tex->rendered_to)
1083       goto fail;
1084
1085    tex->dirty = CALLOC(1, sizeof(tex->dirty[0]));
1086    if (!tex->dirty)
1087       goto fail;
1088
1089    tex->imported = TRUE;
1090
1091    ss->hud.num_resources++;
1092
1093    return &tex->b.b;
1094
1095 fail:
1096    if (tex->defined)
1097       FREE(tex->defined);
1098    if (tex->rendered_to)
1099       FREE(tex->rendered_to);
1100    if (tex->dirty)
1101       FREE(tex->dirty);
1102    FREE(tex);
1103    return NULL;
1104 }
1105
1106 boolean
1107 svga_texture_generate_mipmap(struct pipe_context *pipe,
1108                              struct pipe_resource *pt,
1109                              enum pipe_format format,
1110                              unsigned base_level,
1111                              unsigned last_level,
1112                              unsigned first_layer,
1113                              unsigned last_layer)
1114 {
1115    struct pipe_sampler_view templ, *psv;
1116    struct svga_pipe_sampler_view *sv;
1117    struct svga_context *svga = svga_context(pipe);
1118    struct svga_texture *tex = svga_texture(pt);
1119    enum pipe_error ret;
1120
1121    assert(svga_have_vgpu10(svga));
1122
1123    /* Only support 2D texture for now */
1124    if (pt->target != PIPE_TEXTURE_2D)
1125       return FALSE;
1126
1127    /* Fallback to the mipmap generation utility for those formats that
1128     * do not support hw generate mipmap
1129     */
1130    if (!svga_format_support_gen_mips(format))
1131       return FALSE;
1132
1133    /* Make sure the texture surface was created with
1134     * SVGA3D_SURFACE_BIND_RENDER_TARGET
1135     */
1136    if (!tex->handle || !(tex->key.flags & SVGA3D_SURFACE_BIND_RENDER_TARGET))
1137       return FALSE;
1138
1139    templ.format = format;
1140    templ.u.tex.first_layer = first_layer;
1141    templ.u.tex.last_layer = last_layer;
1142    templ.u.tex.first_level = base_level;
1143    templ.u.tex.last_level = last_level;
1144
1145    psv = pipe->create_sampler_view(pipe, pt, &templ);
1146    if (psv == NULL)
1147       return FALSE;
1148
1149    sv = svga_pipe_sampler_view(psv);
1150    ret = svga_validate_pipe_sampler_view(svga, sv);
1151    if (ret != PIPE_OK) {
1152       svga_context_flush(svga, NULL);
1153       ret = svga_validate_pipe_sampler_view(svga, sv);
1154       assert(ret == PIPE_OK);
1155    }
1156
1157    ret = SVGA3D_vgpu10_GenMips(svga->swc, sv->id, tex->handle);
1158    if (ret != PIPE_OK) {
1159       svga_context_flush(svga, NULL);
1160       ret = SVGA3D_vgpu10_GenMips(svga->swc, sv->id, tex->handle);
1161    }
1162    pipe_sampler_view_reference(&psv, NULL);
1163
1164    svga->hud.num_generate_mipmap++;
1165
1166    return TRUE;
1167 }