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}, { 0,  0}, { 2,  2}}, /*  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}, { 0,  0}, {16, 16}}, /*  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->b.b.width0, level);
  95     } else {
  96         texdim = u_minify(desc->b.b.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->b.b.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->b.b.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.last_level != 0) {
 189             height = util_next_power_of_two(height);
 190         }
 191
 192         /* See if the CBZB clear can be used on the buffer,
 193          * taking the texture size into account. */
 194         if (out_aligned_for_cbzb) {
 195             if (desc->macrotile[level]) {
 196                 /* When clearing, the layer (width*height) is horizontally split
 197                  * into two, and the upper and lower halves are cleared by the CB
 198                  * and ZB units, respectively. Therefore, the number of macrotiles
 199                  * in the Y direction must be even. */
 200
 201                 /* Align the height so that there is an even number of macrotiles.
 202                  * Do so for 3 or more macrotiles in the Y direction. */
 203                 if (level == 0 && desc->b.b.last_level == 0 &&
 204                     (desc->b.b.target == PIPE_TEXTURE_1D ||
 205                      desc->b.b.target == PIPE_TEXTURE_2D) &&
 206                     height >= tile_height * 3) {
 207                     height = align(height, tile_height * 2);
 208                 }
 209
 210                 *out_aligned_for_cbzb = height % (tile_height * 2) == 0;
 211             } else {
 212                 *out_aligned_for_cbzb = FALSE;
 213             }
 214         }
 215     }
 216
 217     return util_format_get_nblocksy(desc->b.b.format, height);
 218 }
 219
 220 static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
 221                                            struct r300_texture_desc *desc)
 222 {
 223     /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures
 224      * incorrectly. This is a workaround to prevent CS from being rejected. */
 225
 226     unsigned i, size;
 227
 228     if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) &&
 229         desc->b.b.target == PIPE_TEXTURE_3D &&
 230         desc->b.b.last_level > 0) {
 231         size = 0;
 232
 233         for (i = 0; i <= desc->b.b.last_level; i++) {
 234             size += desc->stride_in_bytes[i] *
 235                     r300_texture_get_nblocksy(desc, i, FALSE);
 236         }
 237
 238         size *= desc->b.b.depth0;
 239         desc->size_in_bytes = size;
 240     }
 241 }
 242
 243 /* Get a width in pixels from a stride in bytes. */
 244 static unsigned stride_to_width(enum pipe_format format,
 245                                 unsigned stride_in_bytes)
 246 {
 247     return (stride_in_bytes / util_format_get_blocksize(format)) *
 248             util_format_get_blockwidth(format);
 249 }
 250
 251 static void r300_setup_miptree(struct r300_screen *screen,
 252                                struct r300_texture_desc *desc,
 253                                boolean align_for_cbzb)
 254 {
 255     struct pipe_resource *base = &desc->b.b;
 256     unsigned stride, size, layer_size, nblocksy, i;
 257     boolean rv350_mode = screen->caps.is_rv350;
 258     boolean aligned_for_cbzb;
 259
 260     desc->size_in_bytes = 0;
 261
 262     SCREEN_DBG(screen, DBG_TEXALLOC,
 263         "r300: Making miptree for texture, format %s\n",
 264         util_format_short_name(base->format));
 265
 266     for (i = 0; i <= base->last_level; i++) {
 267         /* Let's see if this miplevel can be macrotiled. */
 268         desc->macrotile[i] =
 269             (desc->macrotile[0] == R300_BUFFER_TILED &&
 270              r300_texture_macro_switch(desc, i, rv350_mode, DIM_WIDTH) &&
 271              r300_texture_macro_switch(desc, i, rv350_mode, DIM_HEIGHT)) ?
 272              R300_BUFFER_TILED : R300_BUFFER_LINEAR;
 273
 274         stride = r300_texture_get_stride(screen, desc, i);
 275
 276         /* Compute the number of blocks in Y, see if the CBZB clear can be
 277          * used on the texture. */
 278         aligned_for_cbzb = FALSE;
 279         if (align_for_cbzb && desc->cbzb_allowed[i])
 280             nblocksy = r300_texture_get_nblocksy(desc, i, &aligned_for_cbzb);
 281         else
 282             nblocksy = r300_texture_get_nblocksy(desc, i, NULL);
 283
 284         layer_size = stride * nblocksy;
 285
 286         if (base->nr_samples) {
 287             layer_size *= base->nr_samples;
 288         }
 289
 290         if (base->target == PIPE_TEXTURE_CUBE)
 291             size = layer_size * 6;
 292         else
 293             size = layer_size * u_minify(base->depth0, i);
 294
 295         desc->offset_in_bytes[i] = desc->size_in_bytes;
 296         desc->size_in_bytes = desc->offset_in_bytes[i] + size;
 297         desc->layer_size_in_bytes[i] = layer_size;
 298         desc->stride_in_bytes[i] = stride;
 299         desc->stride_in_pixels[i] = stride_to_width(desc->b.b.format, stride);
 300         desc->cbzb_allowed[i] = desc->cbzb_allowed[i] && aligned_for_cbzb;
 301
 302         SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d "
 303                 "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n",
 304                 i, u_minify(base->width0, i), u_minify(base->height0, i),
 305                 u_minify(base->depth0, i), stride, desc->size_in_bytes,
 306                 desc->macrotile[i] ? "TRUE" : "FALSE");
 307     }
 308 }
 309
 310 static void r300_setup_flags(struct r300_texture_desc *desc)
 311 {
 312     desc->uses_stride_addressing =
 313         !util_is_power_of_two(desc->b.b.width0) ||
 314         !util_is_power_of_two(desc->b.b.height0) ||
 315         (desc->stride_in_bytes_override &&
 316          stride_to_width(desc->b.b.format,
 317                          desc->stride_in_bytes_override) != desc->b.b.width0);
 318
 319     desc->is_npot =
 320         desc->uses_stride_addressing ||
 321         !util_is_power_of_two(desc->b.b.height0);
 322 }
 323
 324 static void r300_setup_cbzb_flags(struct r300_screen *rscreen,
 325                                   struct r300_texture_desc *desc)
 326 {
 327     unsigned i, bpp;
 328     boolean first_level_valid;
 329
 330     bpp = util_format_get_blocksizebits(desc->b.b.format);
 331
 332     /* 1) The texture must be point-sampled,
 333      * 2) The depth must be 16 or 32 bits.
 334      * 3) If the midpoint ZB offset is not aligned to 2048, it returns garbage
 335      *    with certain texture sizes. Macrotiling ensures the alignment. */
 336     first_level_valid = desc->b.b.nr_samples <= 1 &&
 337                        (bpp == 16 || bpp == 32) &&
 338                        desc->macrotile[0];
 339
 340     for (i = 0; i <= desc->b.b.last_level; i++)
 341         desc->cbzb_allowed[i] = first_level_valid && desc->macrotile[i];
 342 }
 343
 344 static void r300_setup_tiling(struct r300_screen *screen,
 345                               struct r300_texture_desc *desc)
 346 {
 347     struct r300_winsys_screen *rws = screen->rws;
 348     enum pipe_format format = desc->b.b.format;
 349     boolean rv350_mode = screen->caps.is_rv350;
 350     boolean is_zb = util_format_is_depth_or_stencil(format);
 351     boolean dbg_no_tiling = SCREEN_DBG_ON(screen, DBG_NO_TILING);
 352
 353     if (!util_format_is_plain(format)) {
 354         return;
 355     }
 356
 357     /* If height == 1, disable microtiling except for zbuffer. */
 358     if (!is_zb && (desc->b.b.height0 == 1 || dbg_no_tiling)) {
 359         return;
 360     }
 361
 362     /* Set microtiling. */
 363     switch (util_format_get_blocksize(format)) {
 364         case 1:
 365         case 4:
 366             desc->microtile = R300_BUFFER_TILED;
 367             break;
 368
 369         case 2:
 370         case 8:
 371             if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
 372                 desc->microtile = R300_BUFFER_SQUARETILED;
 373             }
 374             break;
 375     }
 376
 377     if (dbg_no_tiling) {
 378         return;
 379     }
 380
 381     /* Set macrotiling. */
 382     if (r300_texture_macro_switch(desc, 0, rv350_mode, DIM_WIDTH) &&
 383         r300_texture_macro_switch(desc, 0, rv350_mode, DIM_HEIGHT)) {
 384         desc->macrotile[0] = R300_BUFFER_TILED;
 385     }
 386 }
 387
 388 static void r300_tex_print_info(struct r300_screen *rscreen,
 389                                 struct r300_texture_desc *desc,
 390                                 const char *func)
 391 {
 392     fprintf(stderr,
 393             "r300: %s: Macro: %s, Micro: %s, Pitch: %i, Dim: %ix%ix%i, "
 394             "LastLevel: %i, Size: %i, Format: %s\n",
 395             func,
 396             desc->macrotile[0] ? "YES" : " NO",
 397             desc->microtile ? "YES" : " NO",
 398             desc->stride_in_pixels[0],
 399             desc->b.b.width0, desc->b.b.height0, desc->b.b.depth0,
 400             desc->b.b.last_level, desc->size_in_bytes,
 401             util_format_short_name(desc->b.b.format));
 402 }
 403
 404 boolean r300_texture_desc_init(struct r300_screen *rscreen,
 405                                struct r300_texture_desc *desc,
 406                                const struct pipe_resource *base,
 407                                enum r300_buffer_tiling microtile,
 408                                enum r300_buffer_tiling macrotile,
 409                                unsigned stride_in_bytes_override,
 410                                unsigned max_buffer_size)
 411 {
 412     desc->b.b = *base;
 413     desc->b.b.screen = &rscreen->screen;
 414
 415     desc->stride_in_bytes_override = stride_in_bytes_override;
 416
 417     if (microtile == R300_BUFFER_SELECT_LAYOUT ||
 418         macrotile == R300_BUFFER_SELECT_LAYOUT) {
 419         r300_setup_tiling(rscreen, desc);
 420     } else {
 421         desc->microtile = microtile;
 422         desc->macrotile[0] = macrotile;
 423         assert(desc->b.b.last_level == 0);
 424     }
 425
 426     r300_setup_flags(desc);
 427     r300_setup_cbzb_flags(rscreen, desc);
 428
 429     /* Setup the miptree description. */
 430     r300_setup_miptree(rscreen, desc, TRUE);
 431     /* If the required buffer size is larger the given max size,
 432      * try again without the alignment for the CBZB clear. */
 433     if (max_buffer_size && desc->size_in_bytes > max_buffer_size) {
 434         r300_setup_miptree(rscreen, desc, FALSE);
 435     }
 436
 437     r300_texture_3d_fix_mipmapping(rscreen, desc);
 438
 439     if (max_buffer_size) {
 440         /* Make sure the buffer we got is large enough. */
 441         if (desc->size_in_bytes > max_buffer_size) {
 442             fprintf(stderr, "r300: texture_from_handle: The buffer is not "
 443                             "large enough. Got: %i, Need: %i, Info:\n",
 444                             max_buffer_size, desc->size_in_bytes);
 445             r300_tex_print_info(rscreen, desc, "texture_from_handle");
 446             return FALSE;
 447         }
 448
 449         desc->buffer_size_in_bytes = max_buffer_size;
 450     } else {
 451         desc->buffer_size_in_bytes = desc->size_in_bytes;
 452     }
 453
 454     if (SCREEN_DBG_ON(rscreen, DBG_TEX))
 455         r300_tex_print_info(rscreen, desc, "texture_from_handle");
 456
 457     return TRUE;
 458 }
 459
 460 unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
 461                                  unsigned level, unsigned zslice,
 462                                  unsigned face)
 463 {
 464     unsigned offset = desc->offset_in_bytes[level];
 465
 466     switch (desc->b.b.target) {
 467         case PIPE_TEXTURE_3D:
 468             assert(face == 0);
 469             return offset + zslice * desc->layer_size_in_bytes[level];
 470
 471         case PIPE_TEXTURE_CUBE:
 472             assert(zslice == 0);
 473             return offset + face * desc->layer_size_in_bytes[level];
 474
 475         default:
 476             assert(zslice == 0 && face == 0);
 477             return offset;
 478     }
 479 }