src/gallium/drivers/r300/r300_texture_desc.c

   1 /*
   2  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  * Copyright 2010 Marek Olšák <maraeo@gmail.com>
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  23
  24 #include "r300_texture_desc.h"
  25
  26 #include "r300_context.h"
  27 #include "r300_winsys.h"
  28
  29 #include "util/u_format.h"
  30
  31 /* Returns the number of pixels that the texture should be aligned to
  32  * in the given dimension. */
  33 unsigned r300_get_pixel_alignment(enum pipe_format format,
  34                                   unsigned num_samples,
  35                                   enum r300_buffer_tiling microtile,
  36                                   enum r300_buffer_tiling macrotile,
  37                                   enum r300_dim dim)
  38 {
  39     static const unsigned table[2][5][3][2] =
  40     {
  41         {
  42     /* Macro: linear    linear    linear
  43        Micro: linear    tiled  square-tiled */
  44             {{ 32, 1}, { 8,  4}, { 0,  0}}, /*   8 bits per pixel */
  45             {{ 16, 1}, { 8,  2}, { 4,  4}}, /*  16 bits per pixel */
  46             {{  8, 1}, { 4,  2}, { 0,  0}}, /*  32 bits per pixel */
  47             {{  4, 1}, { 2,  2}, { 0,  0}}, /*  64 bits per pixel */
  48             {{  2, 1}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
  49         },
  50         {
  51     /* Macro: tiled     tiled     tiled
  52        Micro: linear    tiled  square-tiled */
  53             {{256, 8}, {64, 32}, { 0,  0}}, /*   8 bits per pixel */
  54             {{128, 8}, {64, 16}, {32, 32}}, /*  16 bits per pixel */
  55             {{ 64, 8}, {32, 16}, { 0,  0}}, /*  32 bits per pixel */
  56             {{ 32, 8}, {16, 16}, { 0,  0}}, /*  64 bits per pixel */
  57             {{ 16, 8}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
  58         }
  59     };
  60     static const unsigned aa_block[2] = {4, 8};
  61     unsigned tile = 0;
  62     unsigned pixsize = util_format_get_blocksize(format);
  63
  64     assert(macrotile <= R300_BUFFER_TILED);
  65     assert(microtile <= R300_BUFFER_SQUARETILED);
  66     assert(pixsize <= 16);
  67     assert(dim <= DIM_HEIGHT);
  68
  69     if (num_samples > 1) {
  70         /* Multisampled textures have their own alignment scheme. */
  71         if (pixsize == 4)
  72             tile = aa_block[dim];
  73         /* XXX FP16 AA. */
  74     } else {
  75         /* Standard alignment. */
  76         tile = table[macrotile][util_logbase2(pixsize)][microtile][dim];
  77     }
  78
  79     assert(tile);
  80     return tile;
  81 }
  82
  83 /* Return true if macrotiling should be enabled on the miplevel. */
  84 static boolean r300_texture_macro_switch(struct r300_texture_desc *desc,
  85                                          unsigned level,
  86                                          boolean rv350_mode,
  87                                          enum r300_dim dim)
  88 {
  89     unsigned tile, texdim;
  90
  91     tile = r300_get_pixel_alignment(desc->b.b.format, desc->b.b.nr_samples,
  92                                     desc->microtile, R300_BUFFER_TILED, dim);
  93     if (dim == DIM_WIDTH) {
  94         texdim = u_minify(desc->width0, level);
  95     } else {
  96         texdim = u_minify(desc->height0, level);
  97     }
  98
  99     /* See TX_FILTER1_n.MACRO_SWITCH. */
 100     if (rv350_mode) {
 101         return texdim >= tile;
 102     } else {
 103         return texdim > tile;
 104     }
 105 }
 106
 107 /**
 108  * Return the stride, in bytes, of the texture image of the given texture
 109  * at the given level.
 110  */
 111 static unsigned r300_texture_get_stride(struct r300_screen *screen,
 112                                         struct r300_texture_desc *desc,
 113                                         unsigned level)
 114 {
 115     unsigned tile_width, width, stride;
 116
 117     if (desc->stride_in_bytes_override)
 118         return desc->stride_in_bytes_override;
 119
 120     /* Check the level. */
 121     if (level > desc->b.b.last_level) {
 122         SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n",
 123                    __FUNCTION__, level, desc->b.b.last_level);
 124         return 0;
 125     }
 126
 127     width = u_minify(desc->width0, level);
 128
 129     if (util_format_is_plain(desc->b.b.format)) {
 130         tile_width = r300_get_pixel_alignment(desc->b.b.format,
 131                                               desc->b.b.nr_samples,
 132                                               desc->microtile,
 133                                               desc->macrotile[level],
 134                                               DIM_WIDTH);
 135         width = align(width, tile_width);
 136
 137         stride = util_format_get_stride(desc->b.b.format, width);
 138
 139         /* Some IGPs need a minimum stride of 64 bytes, hmm... */
 140         if (!desc->macrotile[level] &&
 141             (screen->caps.family == CHIP_FAMILY_RS600 ||
 142              screen->caps.family == CHIP_FAMILY_RS690 ||
 143              screen->caps.family == CHIP_FAMILY_RS740)) {
 144             unsigned min_stride;
 145
 146             if (desc->microtile) {
 147                 unsigned tile_height =
 148                         r300_get_pixel_alignment(desc->b.b.format,
 149                                                  desc->b.b.nr_samples,
 150                                                  desc->microtile,
 151                                                  desc->macrotile[level],
 152                                                  DIM_HEIGHT);
 153
 154                 min_stride = 64 / tile_height;
 155             } else {
 156                 min_stride = 64;
 157             }
 158
 159             return stride < min_stride ? min_stride : stride;
 160         }
 161
 162         /* The alignment to 32 bytes is sort of implied by the layout... */
 163         return stride;
 164     } else {
 165         return align(util_format_get_stride(desc->b.b.format, width), 32);
 166     }
 167 }
 168
 169 static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc,
 170                                           unsigned level,
 171                                           boolean *out_aligned_for_cbzb)
 172 {
 173     unsigned height, tile_height;
 174
 175     height = u_minify(desc->height0, level);
 176
 177     if (util_format_is_plain(desc->b.b.format)) {
 178         tile_height = r300_get_pixel_alignment(desc->b.b.format,
 179                                                desc->b.b.nr_samples,
 180                                                desc->microtile,
 181                                                desc->macrotile[level],
 182                                                DIM_HEIGHT);
 183         height = align(height, tile_height);
 184
 185         /* This is needed for the kernel checker, unfortunately. */
 186         if ((desc->b.b.target != PIPE_TEXTURE_1D &&
 187              desc->b.b.target != PIPE_TEXTURE_2D &&
 188              desc->b.b.target != PIPE_TEXTURE_RECT) ||
 189             desc->b.b.last_level != 0) {
 190             height = util_next_power_of_two(height);
 191         }
 192
 193         /* See if the CBZB clear can be used on the buffer,
 194          * taking the texture size into account. */
 195         if (out_aligned_for_cbzb) {
 196             if (desc->macrotile[level]) {
 197                 /* When clearing, the layer (width*height) is horizontally split
 198                  * into two, and the upper and lower halves are cleared by the CB
 199                  * and ZB units, respectively. Therefore, the number of macrotiles
 200                  * in the Y direction must be even. */
 201
 202                 /* Align the height so that there is an even number of macrotiles.
 203                  * Do so for 3 or more macrotiles in the Y direction. */
 204                 if (level == 0 && desc->b.b.last_level == 0 &&
 205                     (desc->b.b.target == PIPE_TEXTURE_1D ||
 206                      desc->b.b.target == PIPE_TEXTURE_2D ||
 207                      desc->b.b.target == PIPE_TEXTURE_RECT) &&
 208                     height >= tile_height * 3) {
 209                     height = align(height, tile_height * 2);
 210                 }
 211
 212                 *out_aligned_for_cbzb = height % (tile_height * 2) == 0;
 213             } else {
 214                 *out_aligned_for_cbzb = FALSE;
 215             }
 216         }
 217     }
 218
 219     return util_format_get_nblocksy(desc->b.b.format, height);
 220 }
 221
 222 static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
 223                                            struct r300_texture_desc *desc)
 224 {
 225     /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures
 226      * incorrectly. This is a workaround to prevent CS from being rejected. */
 227
 228     unsigned i, size;
 229
 230     if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) &&
 231         desc->b.b.target == PIPE_TEXTURE_3D &&
 232         desc->b.b.last_level > 0) {
 233         size = 0;
 234
 235         for (i = 0; i <= desc->b.b.last_level; i++) {
 236             size += desc->stride_in_bytes[i] *
 237                     r300_texture_get_nblocksy(desc, i, FALSE);
 238         }
 239
 240         size *= desc->depth0;
 241         desc->size_in_bytes = size;
 242     }
 243 }
 244
 245 /* Get a width in pixels from a stride in bytes. */
 246 static unsigned stride_to_width(enum pipe_format format,
 247                                 unsigned stride_in_bytes)
 248 {
 249     return (stride_in_bytes / util_format_get_blocksize(format)) *
 250             util_format_get_blockwidth(format);
 251 }
 252
 253 static void r300_setup_miptree(struct r300_screen *screen,
 254                                struct r300_texture_desc *desc,
 255                                boolean align_for_cbzb)
 256 {
 257     struct pipe_resource *base = &desc->b.b;
 258     unsigned stride, size, layer_size, nblocksy, i;
 259     boolean rv350_mode = screen->caps.family >= CHIP_FAMILY_R350;
 260     boolean aligned_for_cbzb;
 261
 262     desc->size_in_bytes = 0;
 263
 264     SCREEN_DBG(screen, DBG_TEXALLOC,
 265         "r300: Making miptree for texture, format %s\n",
 266         util_format_short_name(base->format));
 267
 268     for (i = 0; i <= base->last_level; i++) {
 269         /* Let's see if this miplevel can be macrotiled. */
 270         desc->macrotile[i] =
 271             (desc->macrotile[0] == R300_BUFFER_TILED &&
 272              r300_texture_macro_switch(desc, i, rv350_mode, DIM_WIDTH) &&
 273              r300_texture_macro_switch(desc, i, rv350_mode, DIM_HEIGHT)) ?
 274              R300_BUFFER_TILED : R300_BUFFER_LINEAR;
 275
 276         stride = r300_texture_get_stride(screen, desc, i);
 277
 278         /* Compute the number of blocks in Y, see if the CBZB clear can be
 279          * used on the texture. */
 280         aligned_for_cbzb = FALSE;
 281         if (align_for_cbzb && desc->cbzb_allowed[i])
 282             nblocksy = r300_texture_get_nblocksy(desc, i, &aligned_for_cbzb);
 283         else
 284             nblocksy = r300_texture_get_nblocksy(desc, i, NULL);
 285
 286         layer_size = stride * nblocksy;
 287
 288         if (base->nr_samples) {
 289             layer_size *= base->nr_samples;
 290         }
 291
 292         if (base->target == PIPE_TEXTURE_CUBE)
 293             size = layer_size * 6;
 294         else
 295             size = layer_size * u_minify(desc->depth0, i);
 296
 297         desc->offset_in_bytes[i] = desc->size_in_bytes;
 298         desc->size_in_bytes = desc->offset_in_bytes[i] + size;
 299         desc->layer_size_in_bytes[i] = layer_size;
 300         desc->stride_in_bytes[i] = stride;
 301         desc->stride_in_pixels[i] = stride_to_width(desc->b.b.format, stride);
 302         desc->cbzb_allowed[i] = desc->cbzb_allowed[i] && aligned_for_cbzb;
 303
 304         SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d "
 305                 "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n",
 306                 i, u_minify(desc->width0, i), u_minify(desc->height0, i),
 307                 u_minify(desc->depth0, i), stride, desc->size_in_bytes,
 308                 desc->macrotile[i] ? "TRUE" : "FALSE");
 309     }
 310 }
 311
 312 static void r300_setup_flags(struct r300_texture_desc *desc)
 313 {
 314     desc->uses_stride_addressing =
 315         !util_is_power_of_two(desc->b.b.width0) ||
 316         (desc->stride_in_bytes_override &&
 317          stride_to_width(desc->b.b.format,
 318                          desc->stride_in_bytes_override) != desc->b.b.width0);
 319
 320     desc->is_npot =
 321         desc->uses_stride_addressing ||
 322         !util_is_power_of_two(desc->b.b.height0) ||
 323         !util_is_power_of_two(desc->b.b.depth0);
 324 }
 325
 326 static void r300_setup_cbzb_flags(struct r300_screen *rscreen,
 327                                   struct r300_texture_desc *desc)
 328 {
 329     unsigned i, bpp;
 330     boolean first_level_valid;
 331
 332     bpp = util_format_get_blocksizebits(desc->b.b.format);
 333
 334     /* 1) The texture must be point-sampled,
 335      * 2) The depth must be 16 or 32 bits.
 336      * 3) If the midpoint ZB offset is not aligned to 2048, it returns garbage
 337      *    with certain texture sizes. Macrotiling ensures the alignment. */
 338     first_level_valid = desc->b.b.nr_samples <= 1 &&
 339                        (bpp == 16 || bpp == 32) &&
 340                        desc->macrotile[0];
 341
 342     if (SCREEN_DBG_ON(rscreen, DBG_NO_CBZB))
 343         first_level_valid = FALSE;
 344
 345     for (i = 0; i <= desc->b.b.last_level; i++)
 346         desc->cbzb_allowed[i] = first_level_valid && desc->macrotile[i];
 347 }
 348
 349 static void r300_setup_tiling(struct r300_screen *screen,
 350                               struct r300_texture_desc *desc)
 351 {
 352     struct r300_winsys_screen *rws = screen->rws;
 353     enum pipe_format format = desc->b.b.format;
 354     boolean rv350_mode = screen->caps.family >= CHIP_FAMILY_R350;
 355     boolean is_zb = util_format_is_depth_or_stencil(format);
 356     boolean dbg_no_tiling = SCREEN_DBG_ON(screen, DBG_NO_TILING);
 357
 358     if (!util_format_is_plain(format)) {
 359         return;
 360     }
 361
 362     /* If height == 1, disable microtiling except for zbuffer. */
 363     if (!is_zb && (desc->b.b.height0 == 1 || dbg_no_tiling)) {
 364         return;
 365     }
 366
 367     /* Set microtiling. */
 368     switch (util_format_get_blocksize(format)) {
 369         case 1:
 370         case 4:
 371         case 8:
 372             desc->microtile = R300_BUFFER_TILED;
 373             break;
 374
 375         case 2:
 376             if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
 377                 desc->microtile = R300_BUFFER_SQUARETILED;
 378             }
 379             break;
 380     }
 381
 382     if (dbg_no_tiling) {
 383         return;
 384     }
 385
 386     /* Set macrotiling. */
 387     if (r300_texture_macro_switch(desc, 0, rv350_mode, DIM_WIDTH) &&
 388         r300_texture_macro_switch(desc, 0, rv350_mode, DIM_HEIGHT)) {
 389         desc->macrotile[0] = R300_BUFFER_TILED;
 390     }
 391 }
 392
 393 static void r300_tex_print_info(struct r300_screen *rscreen,
 394                                 struct r300_texture_desc *desc,
 395                                 const char *func)
 396 {
 397     fprintf(stderr,
 398             "r300: %s: Macro: %s, Micro: %s, Pitch: %i, Dim: %ix%ix%i, "
 399             "LastLevel: %i, Size: %i, Format: %s\n",
 400             func,
 401             desc->macrotile[0] ? "YES" : " NO",
 402             desc->microtile ? "YES" : " NO",
 403             desc->stride_in_pixels[0],
 404             desc->b.b.width0, desc->b.b.height0, desc->b.b.depth0,
 405             desc->b.b.last_level, desc->size_in_bytes,
 406             util_format_short_name(desc->b.b.format));
 407 }
 408
 409 boolean r300_texture_desc_init(struct r300_screen *rscreen,
 410                                struct r300_texture_desc *desc,
 411                                const struct pipe_resource *base,
 412                                enum r300_buffer_tiling microtile,
 413                                enum r300_buffer_tiling macrotile,
 414                                unsigned stride_in_bytes_override,
 415                                unsigned max_buffer_size)
 416 {
 417     desc->b.b = *base;
 418     desc->b.b.screen = &rscreen->screen;
 419     desc->stride_in_bytes_override = stride_in_bytes_override;
 420     desc->width0 = base->width0;
 421     desc->height0 = base->height0;
 422     desc->depth0 = base->depth0;
 423
 424     r300_setup_flags(desc);
 425
 426     /* Align a 3D NPOT texture to POT. */
 427     if (base->target == PIPE_TEXTURE_3D && desc->is_npot) {
 428         desc->width0 = util_next_power_of_two(desc->width0);
 429         desc->height0 = util_next_power_of_two(desc->height0);
 430         desc->depth0 = util_next_power_of_two(desc->depth0);
 431     }
 432
 433     /* Setup tiling. */
 434     if (microtile == R300_BUFFER_SELECT_LAYOUT ||
 435         macrotile == R300_BUFFER_SELECT_LAYOUT) {
 436         r300_setup_tiling(rscreen, desc);
 437     } else {
 438         desc->microtile = microtile;
 439         desc->macrotile[0] = macrotile;
 440         assert(desc->b.b.last_level == 0);
 441     }
 442
 443     r300_setup_cbzb_flags(rscreen, desc);
 444
 445     /* Setup the miptree description. */
 446     r300_setup_miptree(rscreen, desc, TRUE);
 447     /* If the required buffer size is larger the given max size,
 448      * try again without the alignment for the CBZB clear. */
 449     if (max_buffer_size && desc->size_in_bytes > max_buffer_size) {
 450         r300_setup_miptree(rscreen, desc, FALSE);
 451     }
 452
 453     r300_texture_3d_fix_mipmapping(rscreen, desc);
 454
 455     if (max_buffer_size) {
 456         /* Make sure the buffer we got is large enough. */
 457         if (desc->size_in_bytes > max_buffer_size) {
 458             fprintf(stderr, "r300: texture_desc_init: The buffer is not "
 459                             "large enough. Got: %i, Need: %i, Info:\n",
 460                             max_buffer_size, desc->size_in_bytes);
 461             r300_tex_print_info(rscreen, desc, "texture_desc_init");
 462             return FALSE;
 463         }
 464
 465         desc->buffer_size_in_bytes = max_buffer_size;
 466     } else {
 467         desc->buffer_size_in_bytes = desc->size_in_bytes;
 468     }
 469
 470     if (SCREEN_DBG_ON(rscreen, DBG_TEX))
 471         r300_tex_print_info(rscreen, desc, "texture_desc_init");
 472
 473     return TRUE;
 474 }
 475
 476 unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
 477                                  unsigned level, unsigned zslice,
 478                                  unsigned face)
 479 {
 480     unsigned offset = desc->offset_in_bytes[level];
 481
 482     switch (desc->b.b.target) {
 483         case PIPE_TEXTURE_3D:
 484             assert(face == 0);
 485             return offset + zslice * desc->layer_size_in_bytes[level];
 486
 487         case PIPE_TEXTURE_CUBE:
 488             assert(zslice == 0);
 489             return offset + face * desc->layer_size_in_bytes[level];
 490
 491         default:
 492             assert(zslice == 0 && face == 0);
 493             return offset;
 494     }
 495 }