vc4_tiling.c

   1 /*
   2  * Copyright © 2014 Broadcom
   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 (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 /** @file vc4_tiling.c
  25  *
  26  * Handles information about the VC4 tiling formats, and loading and storing
  27  * from them.
  28  *
  29  * Texture mipmap levels on VC4 are (with the exception of 32-bit RGBA raster
  30  * textures for scanout) stored as groups of microtiles.  If the texture is at
  31  * least 4x4 microtiles (utiles), then those microtiles are arranged in a sort
  32  * of Hilbert-fractal-ish layout (T), otherwise the microtiles are in raster
  33  * order (LT).
  34  *
  35  * Specifically, the T format has:
  36  *
  37  * - 64b utiles of pixels in a raster-order grid according to cpp.  It's 4x4
  38  *   pixels at 32 bit depth.
  39  *
  40  * - 1k subtiles made of a 4x4 raster-order grid of 64b utiles (so usually
  41  *   16x16 pixels).
  42  *
  43  * - 4k tiles made of a 2x2 grid of 1k subtiles (so usually 32x32 pixels).  On
  44  *   even 4k tile rows, they're arranged as (BL, TL, TR, BR), and on odd rows
  45  *   they're (TR, BR, BL, TL), where bottom left is start of memory.
  46  *
  47  * - an image made of 4k tiles in rows either left-to-right (even rows of 4k
  48  *   tiles) or right-to-left (odd rows of 4k tiles).
  49  */
  50
  51 #include "vc4_screen.h"
  52 #include "vc4_context.h"
  53 #include "vc4_tiling.h"
  54
  55 /** Return the width in pixels of a 64-byte microtile. */
  56 uint32_t
  57 vc4_utile_width(int cpp)
  58 {
  59         switch (cpp) {
  60         case 1:
  61         case 2:
  62                 return 8;
  63         case 4:
  64                 return 4;
  65         case 8:
  66                 return 2;
  67         default:
  68                 fprintf(stderr, "unknown cpp: %d\n", cpp);
  69                 abort();
  70         }
  71 }
  72
  73 /** Return the height in pixels of a 64-byte microtile. */
  74 uint32_t
  75 vc4_utile_height(int cpp)
  76 {
  77         switch (cpp) {
  78         case 1:
  79                 return 8;
  80         case 2:
  81         case 4:
  82         case 8:
  83                 return 4;
  84         default:
  85                 fprintf(stderr, "unknown cpp: %d\n", cpp);
  86                 abort();
  87         }
  88 }
  89
  90 /**
  91  * The texture unit decides what tiling format a particular miplevel is using
  92  * this function, so we lay out our miptrees accordingly.
  93  */
  94 bool
  95 vc4_size_is_lt(uint32_t width, uint32_t height, int cpp)
  96 {
  97         return (width <= 4 * vc4_utile_width(cpp) ||
  98                 height <= 4 * vc4_utile_height(cpp));
  99 }
 100
 101 void
 102 vc4_load_utile(void *dst, void *src, uint32_t dst_stride, uint32_t cpp)
 103 {
 104         uint32_t utile_h = vc4_utile_height(cpp);
 105         uint32_t row_size = 64 / utile_h;
 106
 107         for (int y = 0; y < utile_h; y++) {
 108                 memcpy(dst, src, row_size);
 109                 dst += dst_stride;
 110                 src += row_size;
 111         }
 112 }
 113
 114 void
 115 vc4_store_utile(void *dst, void *src, uint32_t src_stride, uint32_t cpp)
 116 {
 117         uint32_t utile_h = vc4_utile_height(cpp);
 118         uint32_t row_size = 64 / utile_h;
 119
 120         for (int y = 0; y < utile_h; y++) {
 121                 memcpy(dst, src, row_size);
 122                 dst += row_size;
 123                 src += src_stride;
 124         }
 125 }
 126
 127 static void
 128 check_box_utile_alignment(const struct pipe_box *box, int cpp)
 129 {
 130         uint32_t utile_w = vc4_utile_width(cpp);
 131         uint32_t utile_h = vc4_utile_height(cpp);
 132
 133         assert(!(box->x & (utile_w - 1)));
 134         assert(!(box->y & (utile_h - 1)));
 135         assert(!(box->width & (utile_w - 1)));
 136         assert(!(box->height & (utile_h - 1)));
 137 }
 138
 139 static void
 140 vc4_load_lt_image(void *dst, uint32_t dst_stride,
 141                   void *src, uint32_t src_stride,
 142                   int cpp, const struct pipe_box *box)
 143 {
 144         uint32_t utile_w = vc4_utile_width(cpp);
 145         uint32_t utile_h = vc4_utile_height(cpp);
 146         uint32_t xstart = box->x / utile_w;
 147         uint32_t ystart = box->y / utile_h;
 148
 149         for (uint32_t y = 0; y < box->height; y += utile_h) {
 150                 for (int x = 0; x < box->width; x += utile_w) {
 151                         vc4_load_utile(dst + (dst_stride * y +
 152                                               x * cpp),
 153                                        src + ((ystart + y) * src_stride +
 154                                               (xstart + x) * 64 / utile_w),
 155                                        dst_stride, cpp);
 156                 }
 157         }
 158 }
 159
 160 static void
 161 vc4_store_lt_image(void *dst, uint32_t dst_stride,
 162                    void *src, uint32_t src_stride,
 163                    int cpp, const struct pipe_box *box)
 164 {
 165         uint32_t utile_w = vc4_utile_width(cpp);
 166         uint32_t utile_h = vc4_utile_height(cpp);
 167         uint32_t xstart = box->x / utile_w;
 168         uint32_t ystart = box->y / utile_h;
 169
 170         for (uint32_t y = 0; y < box->height; y += utile_h) {
 171                 for (int x = 0; x < box->width; x += utile_w) {
 172                         vc4_store_utile(dst + ((ystart + y) * dst_stride +
 173                                                (xstart + x) * 64 / utile_w),
 174                                         src + (src_stride * y +
 175                                                x * cpp),
 176                                         src_stride, cpp);
 177                 }
 178         }
 179 }
 180
 181 /**
 182  * Takes a utile x and y (and the number of utiles of width of the image) and
 183  * returns the offset to the utile within a VC4_TILING_FORMAT_TF image.
 184  */
 185 static uint32_t
 186 t_utile_address(uint32_t utile_x, uint32_t utile_y,
 187                 uint32_t utile_stride)
 188 {
 189         /* T images have to be aligned to 8 utiles (4x4 subtiles, which are
 190          * 2x2 in a 4k tile).
 191          */
 192         assert(!(utile_stride & 7));
 193         uint32_t tile_stride = utile_stride >> 3;
 194         /* 4k tile offsets. */
 195         uint32_t tile_x = utile_x >> 3;
 196         uint32_t tile_y = utile_y >> 3;
 197         bool odd_tile_y = tile_y & 1;
 198
 199         /* Odd lines of 4k tiles go right-to-left. */
 200         if (odd_tile_y)
 201                 tile_x = tile_stride - tile_x - 1;
 202
 203         uint32_t tile_offset = 4096 * (tile_y * tile_stride + tile_x);
 204
 205         uint32_t stile_x = (utile_x >> 2) & 1;
 206         uint32_t stile_y = (utile_y >> 2) & 1;
 207         uint32_t stile_index = (stile_y << 1) + stile_x;
 208         static const uint32_t odd_stile_map[4] = {2, 1, 3, 0};
 209         static const uint32_t even_stile_map[4] = {0, 3, 1, 2};
 210
 211         uint32_t stile_offset = 1024 * (odd_tile_y ?
 212                                         odd_stile_map[stile_index] :
 213                                         even_stile_map[stile_index]);
 214
 215         uint32_t utile_offset = 64 * ((utile_y & 3) * 4 + (utile_x & 3));
 216
 217 #if 0
 218         fprintf(stderr, "utile %d,%d -> %d + %d + %d (stride %d,%d) = %d\n",
 219                 utile_x, utile_y,
 220                 tile_offset, stile_offset, utile_offset,
 221                 utile_stride, tile_stride,
 222                 tile_offset + stile_offset + utile_offset);
 223 #endif
 224
 225         return tile_offset + stile_offset + utile_offset;
 226 }
 227
 228 static void
 229 vc4_load_t_image(void *dst, uint32_t dst_stride,
 230                  void *src, uint32_t src_stride,
 231                  int cpp, const struct pipe_box *box)
 232 {
 233         uint32_t utile_w = vc4_utile_width(cpp);
 234         uint32_t utile_h = vc4_utile_height(cpp);
 235         uint32_t utile_stride = src_stride / cpp / utile_w;
 236         uint32_t xstart = box->x / utile_w;
 237         uint32_t ystart = box->y / utile_h;
 238
 239         for (uint32_t y = 0; y < box->height / utile_h; y++) {
 240                 for (int x = 0; x < box->width / utile_w; x++) {
 241                         vc4_load_utile(dst + (y * utile_h * dst_stride +
 242                                               x * utile_w * cpp),
 243                                        src + t_utile_address(xstart + x,
 244                                                              ystart + y,
 245                                                              utile_stride),
 246                                        dst_stride, cpp);
 247                 }
 248         }
 249 }
 250
 251 static void
 252 vc4_store_t_image(void *dst, uint32_t dst_stride,
 253                   void *src, uint32_t src_stride,
 254                   int cpp, const struct pipe_box *box)
 255 {
 256         uint32_t utile_w = vc4_utile_width(cpp);
 257         uint32_t utile_h = vc4_utile_height(cpp);
 258         uint32_t utile_stride = dst_stride / cpp / utile_w;
 259         uint32_t xstart = box->x / utile_w;
 260         uint32_t ystart = box->y / utile_h;
 261
 262         for (uint32_t y = 0; y < box->height / utile_h; y++) {
 263                 for (int x = 0; x < box->width / utile_w; x++) {
 264                         vc4_store_utile(dst + t_utile_address(xstart + x,
 265                                                               ystart + y,
 266                                                               utile_stride),
 267                                         src + (y * utile_h * src_stride +
 268                                                x * utile_w * cpp),
 269                                         src_stride, cpp);
 270                 }
 271         }
 272 }
 273
 274 /**
 275  * Loads pixel data from the start (microtile-aligned) box in @src to the
 276  * start of @dst according to the given tiling format.
 277  */
 278 void
 279 vc4_load_tiled_image(void *dst, uint32_t dst_stride,
 280                      void *src, uint32_t src_stride,
 281                      uint8_t tiling_format, int cpp,
 282                      const struct pipe_box *box)
 283 {
 284         check_box_utile_alignment(box, cpp);
 285
 286         if (tiling_format == VC4_TILING_FORMAT_LT) {
 287                 vc4_load_lt_image(dst, dst_stride,
 288                                   src, src_stride,
 289                                   cpp, box);
 290         } else {
 291                 assert(tiling_format == VC4_TILING_FORMAT_T);
 292                 vc4_load_t_image(dst, dst_stride,
 293                                  src, src_stride,
 294                                  cpp, box);
 295         }
 296 }
 297
 298 /**
 299  * Stores pixel data from the start of @src into a (microtile-aligned) box in
 300  * @dst according to the given tiling format.
 301  */
 302 void
 303 vc4_store_tiled_image(void *dst, uint32_t dst_stride,
 304                       void *src, uint32_t src_stride,
 305                       uint8_t tiling_format, int cpp,
 306                       const struct pipe_box *box)
 307 {
 308         check_box_utile_alignment(box, cpp);
 309
 310         if (tiling_format == VC4_TILING_FORMAT_LT) {
 311                 vc4_store_lt_image(dst, dst_stride,
 312                                    src, src_stride,
 313                                    cpp, box);
 314         } else {
 315                 assert(tiling_format == VC4_TILING_FORMAT_T);
 316                 vc4_store_t_image(dst, dst_stride,
 317                                   src, src_stride,
 318                                   cpp, box);
 319         }
 320 }
 321