src/gallium/drivers/freedreno/a4xx/fd4_format.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #include "pipe/p_defines.h"
  30 #include "util/u_format.h"
  31
  32 #include "fd4_format.h"
  33
  34
  35 /* Specifies the table of all the formats and their features. Also supplies
  36  * the helpers that look up various data in those tables.
  37  */
  38
  39 struct fd4_format {
  40         enum a4xx_vtx_fmt vtx;
  41         enum a4xx_tex_fmt tex;
  42         enum a4xx_color_fmt rb;
  43         enum a3xx_color_swap swap;
  44         boolean present;
  45 };
  46
  47 #define RB4_NONE ~0
  48
  49 /* vertex + texture */
  50 #define VT(pipe, fmt, rbfmt, swapfmt) \
  51         [PIPE_FORMAT_ ## pipe] = { \
  52                 .present = 1, \
  53                 .vtx = VFMT4_ ## fmt, \
  54                 .tex = TFMT4_ ## fmt, \
  55                 .rb = RB4_ ## rbfmt, \
  56                 .swap = swapfmt \
  57         }
  58
  59 /* texture-only */
  60 #define _T(pipe, fmt, rbfmt, swapfmt) \
  61         [PIPE_FORMAT_ ## pipe] = { \
  62                 .present = 1, \
  63                 .vtx = ~0, \
  64                 .tex = TFMT4_ ## fmt, \
  65                 .rb = RB4_ ## rbfmt, \
  66                 .swap = swapfmt \
  67         }
  68
  69 /* vertex-only */
  70 #define V_(pipe, fmt, rbfmt, swapfmt) \
  71         [PIPE_FORMAT_ ## pipe] = { \
  72                 .present = 1, \
  73                 .vtx = VFMT4_ ## fmt, \
  74                 .tex = ~0, \
  75                 .rb = RB4_ ## rbfmt, \
  76                 .swap = swapfmt \
  77         }
  78
  79 static struct fd4_format formats[PIPE_FORMAT_COUNT] = {
  80         /* 8-bit */
  81         VT(R8_UNORM,   8_UNORM, R8_UNORM, WZYX),
  82         V_(R8_SNORM,   8_SNORM, NONE,     WZYX),
  83         V_(R8_UINT,    8_UINT,  NONE,     WZYX),
  84         V_(R8_SINT,    8_SINT,  NONE,     WZYX),
  85         V_(R8_USCALED, 8_UINT,  NONE,     WZYX),
  86         V_(R8_SSCALED, 8_UINT,  NONE,     WZYX),
  87
  88         _T(A8_UNORM,   8_UNORM, A8_UNORM, WZYX),
  89         _T(L8_UNORM,   8_UNORM, R8_UNORM, WZYX),
  90         _T(I8_UNORM,   8_UNORM, NONE,     WZYX),
  91
  92         /* NOTE: should be TFMT_8_UINT (which then gets remapped to
  93          * TFMT_8_UNORM for mem2gmem in _gmem_restore_format()), but
  94          * we don't know TFMT_8_UINT yet.. so just use TFMT_8_UNORM
  95          * for now.. sampling from stencil as a texture might not
  96          * work right, but at least should be fine for zsbuf..
  97          */
  98         _T(S8_UINT,    8_UNORM,  R8_UNORM, WZYX),
  99
 100         /* 16-bit */
 101         V_(R16_UNORM,   16_UNORM, NONE,     WZYX),
 102         V_(R16_SNORM,   16_SNORM, NONE,     WZYX),
 103         VT(R16_UINT,    16_UINT,  R16_UINT, WZYX),
 104         VT(R16_SINT,    16_SINT,  R16_SINT, WZYX),
 105         V_(R16_USCALED, 16_UINT,  NONE,     WZYX),
 106         V_(R16_SSCALED, 16_UINT,  NONE,     WZYX),
 107         VT(R16_FLOAT,   16_FLOAT, NONE,     WZYX),
 108
 109         _T(A16_UINT,    16_UINT,  NONE,     WZYX),
 110         _T(A16_SINT,    16_SINT,  NONE,     WZYX),
 111         _T(L16_UINT,    16_UINT,  NONE,     WZYX),
 112         _T(L16_SINT,    16_SINT,  NONE,     WZYX),
 113         _T(I16_UINT,    16_UINT,  NONE,     WZYX),
 114         _T(I16_SINT,    16_SINT,  NONE,     WZYX),
 115
 116         VT(R8G8_UNORM,   8_8_UNORM, R8G8_UNORM, WZYX),
 117         VT(R8G8_SNORM,   8_8_SNORM, R8G8_SNORM, WZYX),
 118         VT(R8G8_UINT,    8_8_UINT,  NONE,       WZYX),
 119         VT(R8G8_SINT,    8_8_SINT,  NONE,       WZYX),
 120         V_(R8G8_USCALED, 8_8_UINT,  NONE,       WZYX),
 121         V_(R8G8_SSCALED, 8_8_SINT,  NONE,       WZYX),
 122
 123         _T(L8A8_UINT,    8_8_UINT,  NONE,       WZYX),
 124         _T(L8A8_SINT,    8_8_SINT,  NONE,       WZYX),
 125
 126         _T(B5G5R5A1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
 127         _T(B5G5R5X1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
 128         _T(B4G4R4A4_UNORM, 4_4_4_4_UNORM, R4G4B4A4_UNORM, WXYZ),
 129
 130         /* 24-bit */
 131         V_(R8G8B8_UNORM,   8_8_8_UNORM, NONE, WZYX),
 132         V_(R8G8B8_SNORM,   8_8_8_SNORM, NONE, WZYX),
 133         V_(R8G8B8_UINT,    8_8_8_UINT,  NONE, WZYX),
 134         V_(R8G8B8_SINT,    8_8_8_SINT,  NONE, WZYX),
 135         V_(R8G8B8_USCALED, 8_8_8_UINT,  NONE, WZYX),
 136         V_(R8G8B8_SSCALED, 8_8_8_SINT,  NONE, WZYX),
 137
 138         /* 32-bit */
 139         VT(R32_UINT,    32_UINT,  R32_UINT, WZYX),
 140         VT(R32_SINT,    32_SINT,  R32_SINT, WZYX),
 141         V_(R32_USCALED, 32_UINT,  NONE,     WZYX),
 142         V_(R32_SSCALED, 32_UINT,  NONE,     WZYX),
 143         VT(R32_FLOAT,   32_FLOAT, NONE,     WZYX),
 144         V_(R32_FIXED,   32_FIXED, NONE,     WZYX),
 145
 146         _T(A32_UINT,    32_UINT,  NONE,     WZYX),
 147         _T(A32_SINT,    32_SINT,  NONE,     WZYX),
 148         _T(L32_UINT,    32_UINT,  NONE,     WZYX),
 149         _T(L32_SINT,    32_SINT,  NONE,     WZYX),
 150         _T(I32_UINT,    32_UINT,  NONE,     WZYX),
 151         _T(I32_SINT,    32_SINT,  NONE,     WZYX),
 152
 153         V_(R16G16_UNORM,   16_16_UNORM, NONE,        WZYX),
 154         V_(R16G16_SNORM,   16_16_SNORM, NONE,        WZYX),
 155         VT(R16G16_UINT,    16_16_UINT,  R16G16_UINT, WZYX),
 156         VT(R16G16_SINT,    16_16_SINT,  R16G16_SINT, WZYX),
 157         V_(R16G16_USCALED, 16_16_UINT,  NONE,        WZYX),
 158         V_(R16G16_SSCALED, 16_16_SINT,  NONE,        WZYX),
 159         VT(R16G16_FLOAT,   16_16_FLOAT, NONE,        WZYX),
 160
 161         _T(L16A16_UINT,    16_16_UINT,  NONE,        WZYX),
 162         _T(L16A16_SINT,    16_16_SINT,  NONE,        WZYX),
 163
 164         VT(R8G8B8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 165         _T(R8G8B8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 166         _T(R8G8B8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 167         _T(R8G8B8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 168         VT(R8G8B8A8_SNORM,   8_8_8_8_SNORM, R8G8B8A8_SNORM, WZYX),
 169         VT(R8G8B8A8_UINT,    8_8_8_8_UINT,  R8G8B8A8_UINT,  WZYX),
 170         VT(R8G8B8A8_SINT,    8_8_8_8_SINT,  R8G8B8A8_SINT,  WZYX),
 171         V_(R8G8B8A8_USCALED, 8_8_8_8_UINT,  NONE,           WZYX),
 172         V_(R8G8B8A8_SSCALED, 8_8_8_8_SINT,  NONE,           WZYX),
 173
 174         VT(B8G8R8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 175         _T(B8G8R8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 176         VT(B8G8R8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 177         _T(B8G8R8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 178
 179         VT(A8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 180         _T(X8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 181         _T(A8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 182         _T(X8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 183
 184         VT(A8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 185         _T(X8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 186         _T(A8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 187         _T(X8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 188
 189         VT(R10G10B10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WZYX),
 190         VT(B10G10R10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
 191         _T(B10G10R10X2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
 192         V_(R10G10B10A2_SNORM,   10_10_10_2_SNORM, NONE,              WZYX),
 193         V_(R10G10B10A2_UINT,    10_10_10_2_UINT,  NONE,              WZYX),
 194         V_(R10G10B10A2_USCALED, 10_10_10_2_UINT,  NONE,              WZYX),
 195         V_(R10G10B10A2_SSCALED, 10_10_10_2_SINT,  NONE,              WZYX),
 196
 197         _T(R11G11B10_FLOAT, 11_11_10_FLOAT, R11G11B10_FLOAT, WZYX),
 198         _T(R9G9B9E5_FLOAT,  9_9_9_E5_FLOAT, NONE,            WZYX),
 199
 200         _T(Z24X8_UNORM,       X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 201         _T(Z24_UNORM_S8_UINT, X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 202         _T(Z32_FLOAT,         32_FLOAT,   R8G8B8A8_UNORM, WZYX),
 203         _T(Z32_FLOAT_S8X24_UINT, 32_FLOAT,R8G8B8A8_UNORM, WZYX),
 204
 205         /* 48-bit */
 206         V_(R16G16B16_UNORM,   16_16_16_UNORM, NONE, WZYX),
 207         V_(R16G16B16_SNORM,   16_16_16_SNORM, NONE, WZYX),
 208         V_(R16G16B16_UINT,    16_16_16_UINT,  NONE, WZYX),
 209         V_(R16G16B16_SINT,    16_16_16_SINT,  NONE, WZYX),
 210         V_(R16G16B16_USCALED, 16_16_16_UINT,  NONE, WZYX),
 211         V_(R16G16B16_SSCALED, 16_16_16_SINT,  NONE, WZYX),
 212         V_(R16G16B16_FLOAT,   16_16_16_FLOAT, NONE, WZYX),
 213
 214         /* 64-bit */
 215         V_(R16G16B16A16_UNORM,   16_16_16_16_UNORM, NONE,               WZYX),
 216         V_(R16G16B16A16_SNORM,   16_16_16_16_SNORM, NONE,               WZYX),
 217         VT(R16G16B16A16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
 218         _T(R16G16B16X16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
 219         VT(R16G16B16A16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
 220         _T(R16G16B16X16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
 221         V_(R16G16B16A16_USCALED, 16_16_16_16_UINT,  NONE,               WZYX),
 222         V_(R16G16B16A16_SSCALED, 16_16_16_16_SINT,  NONE,               WZYX),
 223         VT(R16G16B16A16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
 224         _T(R16G16B16X16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
 225
 226         VT(R32G32_UINT,    32_32_UINT,  R32G32_UINT, WZYX),
 227         VT(R32G32_SINT,    32_32_SINT,  R32G32_SINT, WZYX),
 228         V_(R32G32_USCALED, 32_32_UINT,  NONE,        WZYX),
 229         V_(R32G32_SSCALED, 32_32_SINT,  NONE,        WZYX),
 230         VT(R32G32_FLOAT,   32_32_FLOAT, NONE,        WZYX),
 231         V_(R32G32_FIXED,   32_32_FIXED, NONE,        WZYX),
 232
 233         _T(L32A32_UINT,    32_32_UINT,  NONE,        WZYX),
 234         _T(L32A32_SINT,    32_32_SINT,  NONE,        WZYX),
 235
 236         /* 96-bit */
 237         V_(R32G32B32_UINT,    32_32_32_UINT,  NONE, WZYX),
 238         V_(R32G32B32_SINT,    32_32_32_SINT,  NONE, WZYX),
 239         V_(R32G32B32_USCALED, 32_32_32_UINT,  NONE, WZYX),
 240         V_(R32G32B32_SSCALED, 32_32_32_SINT,  NONE, WZYX),
 241         V_(R32G32B32_FLOAT,   32_32_32_FLOAT, NONE, WZYX),
 242         V_(R32G32B32_FIXED,   32_32_32_FIXED, NONE, WZYX),
 243
 244         /* 128-bit */
 245         VT(R32G32B32A32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
 246         _T(R32G32B32X32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
 247         VT(R32G32B32A32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
 248         _T(R32G32B32X32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
 249         V_(R32G32B32A32_USCALED, 32_32_32_32_UINT,  NONE,               WZYX),
 250         V_(R32G32B32A32_SSCALED, 32_32_32_32_SINT,  NONE,               WZYX),
 251         VT(R32G32B32A32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
 252         _T(R32G32B32X32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
 253         V_(R32G32B32A32_FIXED,   32_32_32_32_FIXED, NONE,               WZYX),
 254 };
 255
 256 /* convert pipe format to vertex buffer format: */
 257 enum a4xx_vtx_fmt
 258 fd4_pipe2vtx(enum pipe_format format)
 259 {
 260         if (!formats[format].present)
 261                 return ~0;
 262         return formats[format].vtx;
 263 }
 264
 265 /* convert pipe format to texture sampler format: */
 266 enum a4xx_tex_fmt
 267 fd4_pipe2tex(enum pipe_format format)
 268 {
 269         if (!formats[format].present)
 270                 return ~0;
 271         return formats[format].tex;
 272 }
 273
 274 /* convert pipe format to MRT / copydest format used for render-target: */
 275 enum a4xx_color_fmt
 276 fd4_pipe2color(enum pipe_format format)
 277 {
 278         if (!formats[format].present)
 279                 return ~0;
 280         return formats[format].rb;
 281 }
 282
 283 enum a3xx_color_swap
 284 fd4_pipe2swap(enum pipe_format format)
 285 {
 286         if (!formats[format].present)
 287                 return WZYX;
 288         return formats[format].swap;
 289 }
 290
 291 enum a4xx_tex_fetchsize
 292 fd4_pipe2fetchsize(enum pipe_format format)
 293 {
 294         if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
 295                 format = PIPE_FORMAT_Z32_FLOAT;
 296
 297         switch (util_format_get_blocksizebits(format)) {
 298         case 8:   return TFETCH4_1_BYTE;
 299         case 16:  return TFETCH4_2_BYTE;
 300         case 32:  return TFETCH4_4_BYTE;
 301         case 64:  return TFETCH4_8_BYTE;
 302         case 128: return TFETCH4_16_BYTE;
 303         default:
 304                 debug_printf("Unknown block size for format %s: %d\n",
 305                                 util_format_name(format),
 306                                 util_format_get_blocksizebits(format));
 307                 return TFETCH4_1_BYTE;
 308         }
 309 }
 310
 311 /* we need to special case a bit the depth/stencil restore, because we are
 312  * using the texture sampler to blit into the depth/stencil buffer, *not*
 313  * into a color buffer.  Otherwise fd4_tex_swiz() will do the wrong thing,
 314  * as it is assuming that you are sampling into normal render target..
 315  *
 316  * TODO looks like we can probably share w/ a3xx..
 317  */
 318 enum pipe_format
 319 fd4_gmem_restore_format(enum pipe_format format)
 320 {
 321         switch (format) {
 322         case PIPE_FORMAT_Z24X8_UNORM:
 323         case PIPE_FORMAT_Z24_UNORM_S8_UINT:
 324                 return PIPE_FORMAT_R8G8B8A8_UNORM;
 325         case PIPE_FORMAT_Z16_UNORM:
 326                 return PIPE_FORMAT_R8G8_UNORM;
 327         case PIPE_FORMAT_S8_UINT:
 328                 return PIPE_FORMAT_R8_UNORM;
 329         default:
 330                 return format;
 331         }
 332 }
 333
 334 enum a4xx_depth_format
 335 fd4_pipe2depth(enum pipe_format format)
 336 {
 337         switch (format) {
 338         case PIPE_FORMAT_Z16_UNORM:
 339                 return DEPTH4_16;
 340         case PIPE_FORMAT_Z24X8_UNORM:
 341         case PIPE_FORMAT_Z24_UNORM_S8_UINT:
 342         case PIPE_FORMAT_X8Z24_UNORM:
 343         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
 344                 return DEPTH4_24_8;
 345         case PIPE_FORMAT_Z32_FLOAT:
 346         case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
 347                 return DEPTH4_32;
 348         default:
 349                 return ~0;
 350         }
 351 }
 352
 353 static inline enum a4xx_tex_swiz
 354 tex_swiz(unsigned swiz)
 355 {
 356         switch (swiz) {
 357         default:
 358         case PIPE_SWIZZLE_RED:   return A4XX_TEX_X;
 359         case PIPE_SWIZZLE_GREEN: return A4XX_TEX_Y;
 360         case PIPE_SWIZZLE_BLUE:  return A4XX_TEX_Z;
 361         case PIPE_SWIZZLE_ALPHA: return A4XX_TEX_W;
 362         case PIPE_SWIZZLE_ZERO:  return A4XX_TEX_ZERO;
 363         case PIPE_SWIZZLE_ONE:   return A4XX_TEX_ONE;
 364         }
 365 }
 366
 367 uint32_t
 368 fd4_tex_swiz(enum pipe_format format, unsigned swizzle_r, unsigned swizzle_g,
 369                 unsigned swizzle_b, unsigned swizzle_a)
 370 {
 371         const struct util_format_description *desc =
 372                         util_format_description(format);
 373         unsigned char swiz[4] = {
 374                         swizzle_r, swizzle_g, swizzle_b, swizzle_a,
 375         }, rswiz[4];
 376
 377         util_format_compose_swizzles(desc->swizzle, swiz, rswiz);
 378
 379         return A4XX_TEX_CONST_0_SWIZ_X(tex_swiz(rswiz[0])) |
 380                         A4XX_TEX_CONST_0_SWIZ_Y(tex_swiz(rswiz[1])) |
 381                         A4XX_TEX_CONST_0_SWIZ_Z(tex_swiz(rswiz[2])) |
 382                         A4XX_TEX_CONST_0_SWIZ_W(tex_swiz(rswiz[3]));
 383 }