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         VT(R8_SNORM,   8_SNORM, NONE,     WZYX),
  83         VT(R8_UINT,    8_UINT,  NONE,     WZYX),
  84         VT(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         _T(A8_UINT,    8_UINT,  NONE,     WZYX),
  93         _T(A8_SINT,    8_SINT,  NONE,     WZYX),
  94         _T(L8_UINT,    8_UINT,  NONE,     WZYX),
  95         _T(L8_SINT,    8_SINT,  NONE,     WZYX),
  96         _T(I8_UINT,    8_UINT,  NONE,     WZYX),
  97         _T(I8_SINT,    8_SINT,  NONE,     WZYX),
  98
  99         _T(S8_UINT,    8_UINT,  R8_UNORM, WZYX),
 100
 101         /* 16-bit */
 102         V_(R16_UNORM,   16_UNORM, NONE,     WZYX),
 103         V_(R16_SNORM,   16_SNORM, NONE,     WZYX),
 104         VT(R16_UINT,    16_UINT,  R16_UINT, WZYX),
 105         VT(R16_SINT,    16_SINT,  R16_SINT, WZYX),
 106         V_(R16_USCALED, 16_UINT,  NONE,     WZYX),
 107         V_(R16_SSCALED, 16_UINT,  NONE,     WZYX),
 108         VT(R16_FLOAT,   16_FLOAT, R16_FLOAT,WZYX),
 109
 110         _T(A16_UINT,    16_UINT,  NONE,     WZYX),
 111         _T(A16_SINT,    16_SINT,  NONE,     WZYX),
 112         _T(L16_UINT,    16_UINT,  NONE,     WZYX),
 113         _T(L16_SINT,    16_SINT,  NONE,     WZYX),
 114         _T(I16_UINT,    16_UINT,  NONE,     WZYX),
 115         _T(I16_SINT,    16_SINT,  NONE,     WZYX),
 116
 117         VT(R8G8_UNORM,   8_8_UNORM, R8G8_UNORM, WZYX),
 118         VT(R8G8_SNORM,   8_8_SNORM, R8G8_SNORM, WZYX),
 119         VT(R8G8_UINT,    8_8_UINT,  R8G8_UINT,  WZYX),
 120         VT(R8G8_SINT,    8_8_SINT,  R8G8_SINT,  WZYX),
 121         V_(R8G8_USCALED, 8_8_UINT,  NONE,       WZYX),
 122         V_(R8G8_SSCALED, 8_8_SINT,  NONE,       WZYX),
 123
 124         _T(L8A8_UINT,    8_8_UINT,  NONE,       WZYX),
 125         _T(L8A8_SINT,    8_8_SINT,  NONE,       WZYX),
 126
 127         _T(B5G5R5A1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
 128         _T(B5G5R5X1_UNORM, 5_5_5_1_UNORM, R5G5B5A1_UNORM, WXYZ),
 129         _T(B4G4R4A4_UNORM, 4_4_4_4_UNORM, R4G4B4A4_UNORM, WXYZ),
 130
 131         /* 24-bit */
 132         V_(R8G8B8_UNORM,   8_8_8_UNORM, NONE, WZYX),
 133         V_(R8G8B8_SNORM,   8_8_8_SNORM, NONE, WZYX),
 134         V_(R8G8B8_UINT,    8_8_8_UINT,  NONE, WZYX),
 135         V_(R8G8B8_SINT,    8_8_8_SINT,  NONE, WZYX),
 136         V_(R8G8B8_USCALED, 8_8_8_UINT,  NONE, WZYX),
 137         V_(R8G8B8_SSCALED, 8_8_8_SINT,  NONE, WZYX),
 138
 139         /* 32-bit */
 140         VT(R32_UINT,    32_UINT,  R32_UINT, WZYX),
 141         VT(R32_SINT,    32_SINT,  R32_SINT, WZYX),
 142         V_(R32_USCALED, 32_UINT,  NONE,     WZYX),
 143         V_(R32_SSCALED, 32_UINT,  NONE,     WZYX),
 144         VT(R32_FLOAT,   32_FLOAT, R32_FLOAT,WZYX),
 145         V_(R32_FIXED,   32_FIXED, NONE,     WZYX),
 146
 147         _T(A32_UINT,    32_UINT,  NONE,     WZYX),
 148         _T(A32_SINT,    32_SINT,  NONE,     WZYX),
 149         _T(L32_UINT,    32_UINT,  NONE,     WZYX),
 150         _T(L32_SINT,    32_SINT,  NONE,     WZYX),
 151         _T(I32_UINT,    32_UINT,  NONE,     WZYX),
 152         _T(I32_SINT,    32_SINT,  NONE,     WZYX),
 153
 154         V_(R16G16_UNORM,   16_16_UNORM, NONE,        WZYX),
 155         V_(R16G16_SNORM,   16_16_SNORM, NONE,        WZYX),
 156         VT(R16G16_UINT,    16_16_UINT,  R16G16_UINT, WZYX),
 157         VT(R16G16_SINT,    16_16_SINT,  R16G16_SINT, WZYX),
 158         V_(R16G16_USCALED, 16_16_UINT,  NONE,        WZYX),
 159         V_(R16G16_SSCALED, 16_16_SINT,  NONE,        WZYX),
 160         VT(R16G16_FLOAT,   16_16_FLOAT, R16G16_FLOAT,WZYX),
 161
 162         _T(L16A16_UINT,    16_16_UINT,  NONE,        WZYX),
 163         _T(L16A16_SINT,    16_16_SINT,  NONE,        WZYX),
 164
 165         VT(R8G8B8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 166         _T(R8G8B8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 167         _T(R8G8B8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 168         _T(R8G8B8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WZYX),
 169         VT(R8G8B8A8_SNORM,   8_8_8_8_SNORM, R8G8B8A8_SNORM, WZYX),
 170         VT(R8G8B8A8_UINT,    8_8_8_8_UINT,  R8G8B8A8_UINT,  WZYX),
 171         VT(R8G8B8A8_SINT,    8_8_8_8_SINT,  R8G8B8A8_SINT,  WZYX),
 172         V_(R8G8B8A8_USCALED, 8_8_8_8_UINT,  NONE,           WZYX),
 173         V_(R8G8B8A8_SSCALED, 8_8_8_8_SINT,  NONE,           WZYX),
 174
 175         VT(B8G8R8A8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 176         _T(B8G8R8X8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 177         VT(B8G8R8A8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 178         _T(B8G8R8X8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, WXYZ),
 179
 180         VT(A8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 181         _T(X8B8G8R8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 182         _T(A8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 183         _T(X8B8G8R8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, XYZW),
 184
 185         VT(A8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 186         _T(X8R8G8B8_UNORM,   8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 187         _T(A8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 188         _T(X8R8G8B8_SRGB,    8_8_8_8_UNORM, R8G8B8A8_UNORM, ZYXW),
 189
 190         VT(R10G10B10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WZYX),
 191         VT(B10G10R10A2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
 192         _T(B10G10R10X2_UNORM,   10_10_10_2_UNORM, R10G10B10A2_UNORM, WXYZ),
 193         V_(R10G10B10A2_SNORM,   10_10_10_2_SNORM, NONE,              WZYX),
 194         V_(R10G10B10A2_UINT,    10_10_10_2_UINT,  NONE,              WZYX),
 195         V_(R10G10B10A2_USCALED, 10_10_10_2_UINT,  NONE,              WZYX),
 196         V_(R10G10B10A2_SSCALED, 10_10_10_2_SINT,  NONE,              WZYX),
 197
 198         _T(R11G11B10_FLOAT, 11_11_10_FLOAT, R11G11B10_FLOAT, WZYX),
 199         _T(R9G9B9E5_FLOAT,  9_9_9_E5_FLOAT, NONE,            WZYX),
 200
 201         _T(Z24X8_UNORM,       X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 202         _T(Z24_UNORM_S8_UINT, X8Z24_UNORM, R8G8B8A8_UNORM, WZYX),
 203         _T(Z32_FLOAT,         32_FLOAT,   R8G8B8A8_UNORM, WZYX),
 204         _T(Z32_FLOAT_S8X24_UINT, 32_FLOAT,R8G8B8A8_UNORM, WZYX),
 205
 206         /* 48-bit */
 207         V_(R16G16B16_UNORM,   16_16_16_UNORM, NONE, WZYX),
 208         V_(R16G16B16_SNORM,   16_16_16_SNORM, NONE, WZYX),
 209         V_(R16G16B16_UINT,    16_16_16_UINT,  NONE, WZYX),
 210         V_(R16G16B16_SINT,    16_16_16_SINT,  NONE, WZYX),
 211         V_(R16G16B16_USCALED, 16_16_16_UINT,  NONE, WZYX),
 212         V_(R16G16B16_SSCALED, 16_16_16_SINT,  NONE, WZYX),
 213         V_(R16G16B16_FLOAT,   16_16_16_FLOAT, NONE, WZYX),
 214
 215         /* 64-bit */
 216         V_(R16G16B16A16_UNORM,   16_16_16_16_UNORM, NONE,               WZYX),
 217         V_(R16G16B16A16_SNORM,   16_16_16_16_SNORM, NONE,               WZYX),
 218         VT(R16G16B16A16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
 219         _T(R16G16B16X16_UINT,    16_16_16_16_UINT,  R16G16B16A16_UINT,  WZYX),
 220         VT(R16G16B16A16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
 221         _T(R16G16B16X16_SINT,    16_16_16_16_SINT,  R16G16B16A16_SINT,  WZYX),
 222         V_(R16G16B16A16_USCALED, 16_16_16_16_UINT,  NONE,               WZYX),
 223         V_(R16G16B16A16_SSCALED, 16_16_16_16_SINT,  NONE,               WZYX),
 224         VT(R16G16B16A16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
 225         _T(R16G16B16X16_FLOAT,   16_16_16_16_FLOAT, R16G16B16A16_FLOAT, WZYX),
 226
 227         VT(R32G32_UINT,    32_32_UINT,  R32G32_UINT, WZYX),
 228         VT(R32G32_SINT,    32_32_SINT,  R32G32_SINT, WZYX),
 229         V_(R32G32_USCALED, 32_32_UINT,  NONE,        WZYX),
 230         V_(R32G32_SSCALED, 32_32_SINT,  NONE,        WZYX),
 231         VT(R32G32_FLOAT,   32_32_FLOAT, R32G32_FLOAT,WZYX),
 232         V_(R32G32_FIXED,   32_32_FIXED, NONE,        WZYX),
 233
 234         _T(L32A32_UINT,    32_32_UINT,  NONE,        WZYX),
 235         _T(L32A32_SINT,    32_32_SINT,  NONE,        WZYX),
 236
 237         /* 96-bit */
 238         V_(R32G32B32_UINT,    32_32_32_UINT,  NONE, WZYX),
 239         V_(R32G32B32_SINT,    32_32_32_SINT,  NONE, WZYX),
 240         V_(R32G32B32_USCALED, 32_32_32_UINT,  NONE, WZYX),
 241         V_(R32G32B32_SSCALED, 32_32_32_SINT,  NONE, WZYX),
 242         V_(R32G32B32_FLOAT,   32_32_32_FLOAT, NONE, WZYX),
 243         V_(R32G32B32_FIXED,   32_32_32_FIXED, NONE, WZYX),
 244
 245         /* 128-bit */
 246         VT(R32G32B32A32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
 247         _T(R32G32B32X32_UINT,    32_32_32_32_UINT,  R32G32B32A32_UINT,  WZYX),
 248         VT(R32G32B32A32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
 249         _T(R32G32B32X32_SINT,    32_32_32_32_SINT,  R32G32B32A32_SINT,  WZYX),
 250         V_(R32G32B32A32_USCALED, 32_32_32_32_UINT,  NONE,               WZYX),
 251         V_(R32G32B32A32_SSCALED, 32_32_32_32_SINT,  NONE,               WZYX),
 252         VT(R32G32B32A32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
 253         _T(R32G32B32X32_FLOAT,   32_32_32_32_FLOAT, R32G32B32A32_FLOAT, WZYX),
 254         V_(R32G32B32A32_FIXED,   32_32_32_32_FIXED, NONE,               WZYX),
 255 };
 256
 257 /* convert pipe format to vertex buffer format: */
 258 enum a4xx_vtx_fmt
 259 fd4_pipe2vtx(enum pipe_format format)
 260 {
 261         if (!formats[format].present)
 262                 return ~0;
 263         return formats[format].vtx;
 264 }
 265
 266 /* convert pipe format to texture sampler format: */
 267 enum a4xx_tex_fmt
 268 fd4_pipe2tex(enum pipe_format format)
 269 {
 270         if (!formats[format].present)
 271                 return ~0;
 272         return formats[format].tex;
 273 }
 274
 275 /* convert pipe format to MRT / copydest format used for render-target: */
 276 enum a4xx_color_fmt
 277 fd4_pipe2color(enum pipe_format format)
 278 {
 279         if (!formats[format].present)
 280                 return ~0;
 281         return formats[format].rb;
 282 }
 283
 284 enum a3xx_color_swap
 285 fd4_pipe2swap(enum pipe_format format)
 286 {
 287         if (!formats[format].present)
 288                 return WZYX;
 289         return formats[format].swap;
 290 }
 291
 292 enum a4xx_tex_fetchsize
 293 fd4_pipe2fetchsize(enum pipe_format format)
 294 {
 295         if (format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
 296                 format = PIPE_FORMAT_Z32_FLOAT;
 297
 298         switch (util_format_get_blocksizebits(format)) {
 299         case 8:   return TFETCH4_1_BYTE;
 300         case 16:  return TFETCH4_2_BYTE;
 301         case 32:  return TFETCH4_4_BYTE;
 302         case 64:  return TFETCH4_8_BYTE;
 303         case 128: return TFETCH4_16_BYTE;
 304         default:
 305                 debug_printf("Unknown block size for format %s: %d\n",
 306                                 util_format_name(format),
 307                                 util_format_get_blocksizebits(format));
 308                 return TFETCH4_1_BYTE;
 309         }
 310 }
 311
 312 /* we need to special case a bit the depth/stencil restore, because we are
 313  * using the texture sampler to blit into the depth/stencil buffer, *not*
 314  * into a color buffer.  Otherwise fd4_tex_swiz() will do the wrong thing,
 315  * as it is assuming that you are sampling into normal render target..
 316  *
 317  * TODO looks like we can probably share w/ a3xx..
 318  */
 319 enum pipe_format
 320 fd4_gmem_restore_format(enum pipe_format format)
 321 {
 322         switch (format) {
 323         case PIPE_FORMAT_Z24X8_UNORM:
 324         case PIPE_FORMAT_Z24_UNORM_S8_UINT:
 325                 return PIPE_FORMAT_R8G8B8A8_UNORM;
 326         case PIPE_FORMAT_Z16_UNORM:
 327                 return PIPE_FORMAT_R8G8_UNORM;
 328         case PIPE_FORMAT_S8_UINT:
 329                 return PIPE_FORMAT_R8_UNORM;
 330         default:
 331                 return format;
 332         }
 333 }
 334
 335 enum a4xx_depth_format
 336 fd4_pipe2depth(enum pipe_format format)
 337 {
 338         switch (format) {
 339         case PIPE_FORMAT_Z16_UNORM:
 340                 return DEPTH4_16;
 341         case PIPE_FORMAT_Z24X8_UNORM:
 342         case PIPE_FORMAT_Z24_UNORM_S8_UINT:
 343         case PIPE_FORMAT_X8Z24_UNORM:
 344         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
 345                 return DEPTH4_24_8;
 346         case PIPE_FORMAT_Z32_FLOAT:
 347         case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
 348                 return DEPTH4_32;
 349         default:
 350                 return ~0;
 351         }
 352 }
 353
 354 static inline enum a4xx_tex_swiz
 355 tex_swiz(unsigned swiz)
 356 {
 357         switch (swiz) {
 358         default:
 359         case PIPE_SWIZZLE_RED:   return A4XX_TEX_X;
 360         case PIPE_SWIZZLE_GREEN: return A4XX_TEX_Y;
 361         case PIPE_SWIZZLE_BLUE:  return A4XX_TEX_Z;
 362         case PIPE_SWIZZLE_ALPHA: return A4XX_TEX_W;
 363         case PIPE_SWIZZLE_ZERO:  return A4XX_TEX_ZERO;
 364         case PIPE_SWIZZLE_ONE:   return A4XX_TEX_ONE;
 365         }
 366 }
 367
 368 uint32_t
 369 fd4_tex_swiz(enum pipe_format format, unsigned swizzle_r, unsigned swizzle_g,
 370                 unsigned swizzle_b, unsigned swizzle_a)
 371 {
 372         const struct util_format_description *desc =
 373                         util_format_description(format);
 374         unsigned char swiz[4] = {
 375                         swizzle_r, swizzle_g, swizzle_b, swizzle_a,
 376         }, rswiz[4];
 377
 378         util_format_compose_swizzles(desc->swizzle, swiz, rswiz);
 379
 380         return A4XX_TEX_CONST_0_SWIZ_X(tex_swiz(rswiz[0])) |
 381                         A4XX_TEX_CONST_0_SWIZ_Y(tex_swiz(rswiz[1])) |
 382                         A4XX_TEX_CONST_0_SWIZ_Z(tex_swiz(rswiz[2])) |
 383                         A4XX_TEX_CONST_0_SWIZ_W(tex_swiz(rswiz[3]));
 384 }