src/gallium/drivers/swr/rasterizer/memory/LoadTile.cpp

   1 /****************************************************************************
   2 * Copyright (C) 2014-2015 Intel Corporation.   All Rights Reserved.
   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 * @file LoadTile.cpp
  24 *
  25 * @brief Functionality for Load
  26 *
  27 ******************************************************************************/
  28 #include "common/os.h"
  29 #include "common/formats.h"
  30 #include "core/context.h"
  31 #include "core/rdtsc_core.h"
  32 #include "memory/TilingFunctions.h"
  33 #include "memory/tilingtraits.h"
  34 #include "memory/Convert.h"
  35
  36 typedef void(*PFN_LOAD_TILES)(SWR_SURFACE_STATE*, uint8_t*, uint32_t, uint32_t, uint32_t);
  37
  38 //////////////////////////////////////////////////////////////////////////
  39 /// Load Raster Tile Function Tables.
  40 //////////////////////////////////////////////////////////////////////////
  41 static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_NONE[NUM_SWR_FORMATS];
  42 static PFN_LOAD_TILES sLoadTilesDepthTable_SWR_TILE_NONE[NUM_SWR_FORMATS];
  43
  44 static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_MODE_YMAJOR[NUM_SWR_FORMATS];
  45 static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_MODE_XMAJOR[NUM_SWR_FORMATS];
  46
  47 static PFN_LOAD_TILES sLoadTilesDepthTable_SWR_TILE_MODE_YMAJOR[NUM_SWR_FORMATS];
  48
  49 //////////////////////////////////////////////////////////////////////////
  50 /// LoadRasterTile
  51 //////////////////////////////////////////////////////////////////////////
  52 template<typename TTraits, SWR_FORMAT SrcFormat, SWR_FORMAT DstFormat>
  53 struct LoadRasterTile
  54 {
  55     //////////////////////////////////////////////////////////////////////////
  56     /// @brief Retrieve color from hot tile source which is always float.
  57     /// @param pSrc - Pointer to raster tile.
  58     /// @param x, y - Coordinates to raster tile.
  59     /// @param output - output color
  60     INLINE static void SetSwizzledDstColor(
  61         const float srcColor[4],
  62         uint32_t x, uint32_t y,
  63         uint8_t* pDst)
  64     {
  65         typedef SimdTile<DstFormat, SrcFormat> SimdT;
  66
  67         SimdT* pDstSimdTiles = (SimdT*)pDst;
  68
  69         // Compute which simd tile we're accessing within 8x8 tile.
  70         //   i.e. Compute linear simd tile coordinate given (x, y) in pixel coordinates.
  71         uint32_t simdIndex = (y / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM) + (x / SIMD_TILE_X_DIM);
  72
  73         SimdT* pSimdTile = &pDstSimdTiles[simdIndex];
  74
  75         uint32_t simdOffset = (y % SIMD_TILE_Y_DIM) * SIMD_TILE_X_DIM + (x % SIMD_TILE_X_DIM);
  76
  77         pSimdTile->SetSwizzledColor(simdOffset, srcColor);
  78     }
  79
  80     //////////////////////////////////////////////////////////////////////////
  81     /// @brief Loads an 8x8 raster tile from the src surface.
  82     /// @param pSrcSurface - Src surface state
  83     /// @param pDst - Destination hot tile pointer
  84     /// @param x, y - Coordinates to raster tile.
  85     INLINE static void Load(
  86         SWR_SURFACE_STATE* pSrcSurface,
  87         uint8_t* pDst,
  88         uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) // (x, y) pixel coordinate to start of raster tile.
  89     {
  90         uint32_t lodWidth = (pSrcSurface->width == 1) ? 1 : pSrcSurface->width >> pSrcSurface->lod;
  91         uint32_t lodHeight = (pSrcSurface->height == 1) ? 1 : pSrcSurface->height >> pSrcSurface->lod;
  92
  93         // For each raster tile pixel (rx, ry)
  94         for (uint32_t ry = 0; ry < KNOB_TILE_Y_DIM; ++ry)
  95         {
  96             for (uint32_t rx = 0; rx < KNOB_TILE_X_DIM; ++rx)
  97             {
  98                 if (((x + rx) < lodWidth) &&
  99                     ((y + ry) < lodHeight))
 100                 {
 101                     uint8_t* pSrc = (uint8_t*)ComputeSurfaceAddress<false>(x + rx, y + ry, pSrcSurface->arrayIndex + renderTargetArrayIndex,
 102                                                                            pSrcSurface->arrayIndex + renderTargetArrayIndex, sampleNum,
 103                                                                            pSrcSurface->lod, pSrcSurface);
 104
 105                     float srcColor[4];
 106                     ConvertPixelToFloat<SrcFormat>(srcColor, pSrc);
 107
 108                     // store pixel to hottile
 109                     SetSwizzledDstColor(srcColor, rx, ry, pDst);
 110                 }
 111             }
 112         }
 113     }
 114 };
 115
 116 //////////////////////////////////////////////////////////////////////////
 117 /// LoadMacroTile - Loads a macro tile which consists of raster tiles.
 118 //////////////////////////////////////////////////////////////////////////
 119 template<typename TTraits, SWR_FORMAT SrcFormat, SWR_FORMAT DstFormat>
 120 struct LoadMacroTile
 121 {
 122     //////////////////////////////////////////////////////////////////////////
 123     /// @brief Load a macrotile to the destination surface.
 124     /// @param pSrc - Pointer to macro tile.
 125     /// @param pDstSurface - Destination surface state
 126     /// @param x, y - Coordinates to macro tile
 127     static void Load(
 128         SWR_SURFACE_STATE* pSrcSurface,
 129         uint8_t *pDstHotTile,
 130         uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex)
 131     {
 132         // Load each raster tile from the hot tile to the destination surface.
 133         for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM)
 134         {
 135             for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM)
 136             {
 137                 for (uint32_t sampleNum = 0; sampleNum < pSrcSurface->numSamples; sampleNum++)
 138                 {
 139                     LoadRasterTile<TTraits, SrcFormat, DstFormat>::Load(pSrcSurface, pDstHotTile,
 140                         (x + col), (y + row), sampleNum, renderTargetArrayIndex);
 141                     pDstHotTile += KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits<DstFormat>::bpp / 8);
 142                 }
 143             }
 144         }
 145     }
 146 };
 147
 148 static void BUCKETS_START(UINT id)
 149 {
 150 #ifdef KNOB_ENABLE_RDTSC
 151     gBucketMgr.StartBucket(id);
 152 #endif
 153 }
 154
 155 static void BUCKETS_STOP(UINT id)
 156 {
 157 #ifdef KNOB_ENABLE_RDTSC
 158     gBucketMgr.StopBucket(id);
 159 #endif
 160 }
 161
 162 // on demand buckets for load tiles
 163 static std::vector<int> sBuckets(NUM_SWR_FORMATS, -1);
 164 static std::mutex sBucketMutex;
 165
 166 //////////////////////////////////////////////////////////////////////////
 167 /// @brief Loads a full hottile from a render surface
 168 /// @param hPrivateContext - Handle to private DC
 169 /// @param dstFormat - Format for hot tile.
 170 /// @param renderTargetIndex - Index to src render target
 171 /// @param x, y - Coordinates to raster tile.
 172 /// @param pDstHotTile - Pointer to Hot Tile
 173 void LoadHotTile(
 174     SWR_SURFACE_STATE *pSrcSurface,
 175     SWR_FORMAT dstFormat,
 176     SWR_RENDERTARGET_ATTACHMENT renderTargetIndex,
 177     uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex,
 178     uint8_t *pDstHotTile)
 179 {
 180     PFN_LOAD_TILES pfnLoadTiles = NULL;
 181
 182     // don't need to load null surfaces
 183     if (pSrcSurface->type == SURFACE_NULL)
 184     {
 185         return;
 186     }
 187
 188     // force 0 if requested renderTargetArrayIndex is OOB
 189     if (renderTargetArrayIndex >= pSrcSurface->depth)
 190     {
 191         renderTargetArrayIndex = 0;
 192     }
 193
 194     if (renderTargetIndex < SWR_ATTACHMENT_DEPTH)
 195     {
 196         switch (pSrcSurface->tileMode)
 197         {
 198         case SWR_TILE_NONE:
 199             pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_NONE[pSrcSurface->format];
 200             break;
 201         case SWR_TILE_MODE_YMAJOR:
 202             pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_MODE_YMAJOR[pSrcSurface->format];
 203             break;
 204         case SWR_TILE_MODE_XMAJOR:
 205             pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_MODE_XMAJOR[pSrcSurface->format];
 206             break;
 207         case SWR_TILE_MODE_WMAJOR:
 208             SWR_ASSERT(pSrcSurface->format == R8_UINT);
 209             pfnLoadTiles = LoadMacroTile<TilingTraits<SWR_TILE_MODE_WMAJOR, 8>, R8_UINT, R8_UINT>::Load;
 210             break;
 211         default:
 212             SWR_ASSERT(0, "Unsupported tiling mode");
 213             break;
 214         }
 215     }
 216     else if (renderTargetIndex == SWR_ATTACHMENT_DEPTH)
 217     {
 218         // Currently depth can map to linear and tile-y.
 219         switch (pSrcSurface->tileMode)
 220         {
 221         case SWR_TILE_NONE:
 222             pfnLoadTiles = sLoadTilesDepthTable_SWR_TILE_NONE[pSrcSurface->format];
 223             break;
 224         case SWR_TILE_MODE_YMAJOR:
 225             pfnLoadTiles = sLoadTilesDepthTable_SWR_TILE_MODE_YMAJOR[pSrcSurface->format];
 226             break;
 227         default:
 228             SWR_ASSERT(0, "Unsupported tiling mode");
 229             break;
 230         }
 231     }
 232     else
 233     {
 234         SWR_ASSERT(renderTargetIndex == SWR_ATTACHMENT_STENCIL);
 235         SWR_ASSERT(pSrcSurface->format == R8_UINT);
 236         switch (pSrcSurface->tileMode)
 237         {
 238         case SWR_TILE_NONE:
 239             pfnLoadTiles = LoadMacroTile<TilingTraits<SWR_TILE_NONE, 8>, R8_UINT, R8_UINT>::Load;
 240             break;
 241         case SWR_TILE_MODE_WMAJOR:
 242             pfnLoadTiles = LoadMacroTile<TilingTraits<SWR_TILE_MODE_WMAJOR, 8>, R8_UINT, R8_UINT>::Load;
 243             break;
 244         default:
 245             SWR_ASSERT(0, "Unsupported tiling mode");
 246             break;
 247         }
 248     }
 249
 250     if (pfnLoadTiles == nullptr)
 251     {
 252         SWR_ASSERT(false, "Unsupported format for load tile");
 253         return;
 254     }
 255
 256     // Load a macro tile.
 257 #ifdef KNOB_ENABLE_RDTSC
 258     if (sBuckets[pSrcSurface->format] == -1)
 259     {
 260         // guard sBuckets update since storetiles is called by multiple threads
 261         sBucketMutex.lock();
 262         if (sBuckets[pSrcSurface->format] == -1)
 263         {
 264             const SWR_FORMAT_INFO& info = GetFormatInfo(pSrcSurface->format);
 265             BUCKET_DESC desc{ info.name, "", false, 0xffffffff };
 266             sBuckets[pSrcSurface->format] = gBucketMgr.RegisterBucket(desc);
 267         }
 268         sBucketMutex.unlock();
 269     }
 270 #endif
 271
 272     BUCKETS_START(sBuckets[pSrcSurface->format]);
 273     pfnLoadTiles(pSrcSurface, pDstHotTile, x, y, renderTargetArrayIndex);
 274     BUCKETS_STOP(sBuckets[pSrcSurface->format]);
 275 }
 276
 277 //////////////////////////////////////////////////////////////////////////
 278 /// INIT_LOAD_TILES_TABLE - Helper macro for setting up the tables.
 279 #define INIT_LOAD_TILES_COLOR_TABLE(tilemode) \
 280     memset(sLoadTilesColorTable_##tilemode, 0, sizeof(sLoadTilesColorTable_##tilemode)); \
 281     \
 282     sLoadTilesColorTable_##tilemode[R32G32B32A32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 128>, R32G32B32A32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 283     sLoadTilesColorTable_##tilemode[R32G32B32A32_SINT]      = LoadMacroTile<TilingTraits<tilemode, 128>, R32G32B32A32_SINT, R32G32B32A32_FLOAT>::Load; \
 284     sLoadTilesColorTable_##tilemode[R32G32B32A32_UINT]      = LoadMacroTile<TilingTraits<tilemode, 128>, R32G32B32A32_UINT, R32G32B32A32_FLOAT>::Load; \
 285     sLoadTilesColorTable_##tilemode[R32G32B32X32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 128>, R32G32B32X32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 286     sLoadTilesColorTable_##tilemode[R32G32B32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 96>, R32G32B32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 287     sLoadTilesColorTable_##tilemode[R32G32B32_SINT]      = LoadMacroTile<TilingTraits<tilemode, 96>, R32G32B32_SINT, R32G32B32A32_FLOAT>::Load; \
 288     sLoadTilesColorTable_##tilemode[R32G32B32_UINT]      = LoadMacroTile<TilingTraits<tilemode, 96>, R32G32B32_UINT, R32G32B32A32_FLOAT>::Load; \
 289     sLoadTilesColorTable_##tilemode[R16G16B16A16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16A16_UNORM, R32G32B32A32_FLOAT>::Load; \
 290     sLoadTilesColorTable_##tilemode[R16G16B16A16_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16A16_SNORM, R32G32B32A32_FLOAT>::Load; \
 291     sLoadTilesColorTable_##tilemode[R16G16B16A16_SINT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16A16_SINT, R32G32B32A32_FLOAT>::Load; \
 292     sLoadTilesColorTable_##tilemode[R16G16B16A16_UINT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16A16_UINT, R32G32B32A32_FLOAT>::Load; \
 293     sLoadTilesColorTable_##tilemode[R16G16B16A16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16A16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 294     sLoadTilesColorTable_##tilemode[R32G32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R32G32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 295     sLoadTilesColorTable_##tilemode[R32G32_SINT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R32G32_SINT, R32G32B32A32_FLOAT>::Load; \
 296     sLoadTilesColorTable_##tilemode[R32G32_UINT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R32G32_UINT, R32G32B32A32_FLOAT>::Load; \
 297     sLoadTilesColorTable_##tilemode[R16G16B16X16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16X16_UNORM, R32G32B32A32_FLOAT>::Load; \
 298     sLoadTilesColorTable_##tilemode[R16G16B16X16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 64>, R16G16B16X16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 299     sLoadTilesColorTable_##tilemode[B8G8R8A8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, B8G8R8A8_UNORM, R32G32B32A32_FLOAT>::Load; \
 300     sLoadTilesColorTable_##tilemode[B8G8R8A8_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, B8G8R8A8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 301     sLoadTilesColorTable_##tilemode[R10G10B10A2_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R10G10B10A2_UNORM, R32G32B32A32_FLOAT>::Load; \
 302     sLoadTilesColorTable_##tilemode[R10G10B10A2_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, R10G10B10A2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 303     sLoadTilesColorTable_##tilemode[R10G10B10A2_UINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R10G10B10A2_UINT, R32G32B32A32_FLOAT>::Load; \
 304     sLoadTilesColorTable_##tilemode[R8G8B8A8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8A8_UNORM, R32G32B32A32_FLOAT>::Load; \
 305     sLoadTilesColorTable_##tilemode[R8G8B8A8_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8A8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 306     sLoadTilesColorTable_##tilemode[R8G8B8A8_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8A8_SNORM, R32G32B32A32_FLOAT>::Load; \
 307     sLoadTilesColorTable_##tilemode[R8G8B8A8_SINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8A8_SINT, R32G32B32A32_FLOAT>::Load; \
 308     sLoadTilesColorTable_##tilemode[R8G8B8A8_UINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8A8_UINT, R32G32B32A32_FLOAT>::Load; \
 309     sLoadTilesColorTable_##tilemode[R16G16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R16G16_UNORM, R32G32B32A32_FLOAT>::Load; \
 310     sLoadTilesColorTable_##tilemode[R16G16_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R16G16_SNORM, R32G32B32A32_FLOAT>::Load; \
 311     sLoadTilesColorTable_##tilemode[R16G16_SINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R16G16_SINT, R32G32B32A32_FLOAT>::Load; \
 312     sLoadTilesColorTable_##tilemode[R16G16_UINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R16G16_UINT, R32G32B32A32_FLOAT>::Load; \
 313     sLoadTilesColorTable_##tilemode[R16G16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R16G16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 314     sLoadTilesColorTable_##tilemode[B10G10R10A2_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10A2_UNORM, R32G32B32A32_FLOAT>::Load; \
 315     sLoadTilesColorTable_##tilemode[B10G10R10A2_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10A2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 316     sLoadTilesColorTable_##tilemode[R11G11B10_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R11G11B10_FLOAT, R32G32B32A32_FLOAT>::Load; \
 317     sLoadTilesColorTable_##tilemode[R32_SINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R32_SINT, R32G32B32A32_FLOAT>::Load; \
 318     sLoadTilesColorTable_##tilemode[R32_UINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R32_UINT, R32G32B32A32_FLOAT>::Load; \
 319     sLoadTilesColorTable_##tilemode[R32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 320     sLoadTilesColorTable_##tilemode[A32_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 32>, A32_FLOAT, R32G32B32A32_FLOAT>::Load; \
 321     sLoadTilesColorTable_##tilemode[B8G8R8X8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, B8G8R8X8_UNORM, R32G32B32A32_FLOAT>::Load; \
 322     sLoadTilesColorTable_##tilemode[B8G8R8X8_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, B8G8R8X8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 323     sLoadTilesColorTable_##tilemode[R8G8B8X8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8X8_UNORM, R32G32B32A32_FLOAT>::Load; \
 324     sLoadTilesColorTable_##tilemode[R8G8B8X8_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 32>, R8G8B8X8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 325     sLoadTilesColorTable_##tilemode[B10G10R10X2_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10X2_UNORM, R32G32B32A32_FLOAT>::Load; \
 326     sLoadTilesColorTable_##tilemode[B5G6R5_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G6R5_UNORM, R32G32B32A32_FLOAT>::Load; \
 327     sLoadTilesColorTable_##tilemode[B5G6R5_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G6R5_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 328     sLoadTilesColorTable_##tilemode[B5G5R5A1_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G5R5A1_UNORM, R32G32B32A32_FLOAT>::Load; \
 329     sLoadTilesColorTable_##tilemode[B5G5R5A1_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G5R5A1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 330     sLoadTilesColorTable_##tilemode[B4G4R4A4_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, B4G4R4A4_UNORM, R32G32B32A32_FLOAT>::Load; \
 331     sLoadTilesColorTable_##tilemode[B4G4R4A4_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 16>, B4G4R4A4_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 332     sLoadTilesColorTable_##tilemode[R8G8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, R8G8_UNORM, R32G32B32A32_FLOAT>::Load; \
 333     sLoadTilesColorTable_##tilemode[R8G8_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, R8G8_SNORM, R32G32B32A32_FLOAT>::Load; \
 334     sLoadTilesColorTable_##tilemode[R8G8_SINT]      = LoadMacroTile<TilingTraits<tilemode, 16>, R8G8_SINT, R32G32B32A32_FLOAT>::Load; \
 335     sLoadTilesColorTable_##tilemode[R8G8_UINT]      = LoadMacroTile<TilingTraits<tilemode, 16>, R8G8_UINT, R32G32B32A32_FLOAT>::Load; \
 336     sLoadTilesColorTable_##tilemode[R16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, R16_UNORM, R32G32B32A32_FLOAT>::Load; \
 337     sLoadTilesColorTable_##tilemode[R16_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, R16_SNORM, R32G32B32A32_FLOAT>::Load; \
 338     sLoadTilesColorTable_##tilemode[R16_SINT]      = LoadMacroTile<TilingTraits<tilemode, 16>, R16_SINT, R32G32B32A32_FLOAT>::Load; \
 339     sLoadTilesColorTable_##tilemode[R16_UINT]      = LoadMacroTile<TilingTraits<tilemode, 16>, R16_UINT, R32G32B32A32_FLOAT>::Load; \
 340     sLoadTilesColorTable_##tilemode[R16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 16>, R16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 341     sLoadTilesColorTable_##tilemode[A16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, A16_UNORM, R32G32B32A32_FLOAT>::Load; \
 342     sLoadTilesColorTable_##tilemode[A16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 16>, A16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 343     sLoadTilesColorTable_##tilemode[B5G5R5X1_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G5R5X1_UNORM, R32G32B32A32_FLOAT>::Load; \
 344     sLoadTilesColorTable_##tilemode[B5G5R5X1_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 16>, B5G5R5X1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 345     sLoadTilesColorTable_##tilemode[R8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 8>, R8_UNORM, R32G32B32A32_FLOAT>::Load; \
 346     sLoadTilesColorTable_##tilemode[R8_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 8>, R8_SNORM, R32G32B32A32_FLOAT>::Load; \
 347     sLoadTilesColorTable_##tilemode[R8_SINT]      = LoadMacroTile<TilingTraits<tilemode, 8>, R8_SINT, R32G32B32A32_FLOAT>::Load; \
 348     sLoadTilesColorTable_##tilemode[R8_UINT]      = LoadMacroTile<TilingTraits<tilemode, 8>, R8_UINT, R32G32B32A32_FLOAT>::Load; \
 349     sLoadTilesColorTable_##tilemode[A8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 8>, A8_UNORM, R32G32B32A32_FLOAT>::Load; \
 350     sLoadTilesColorTable_##tilemode[BC1_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, BC1_UNORM, R32G32B32A32_FLOAT>::Load; \
 351     sLoadTilesColorTable_##tilemode[BC2_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC2_UNORM, R32G32B32A32_FLOAT>::Load; \
 352     sLoadTilesColorTable_##tilemode[BC3_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC3_UNORM, R32G32B32A32_FLOAT>::Load; \
 353     sLoadTilesColorTable_##tilemode[BC4_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, BC4_UNORM, R32G32B32A32_FLOAT>::Load; \
 354     sLoadTilesColorTable_##tilemode[BC5_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC5_UNORM, R32G32B32A32_FLOAT>::Load; \
 355     sLoadTilesColorTable_##tilemode[BC1_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 64>, BC1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 356     sLoadTilesColorTable_##tilemode[BC2_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 357     sLoadTilesColorTable_##tilemode[BC3_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC3_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 358     sLoadTilesColorTable_##tilemode[R8G8B8_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 24>, R8G8B8_UNORM, R32G32B32A32_FLOAT>::Load; \
 359     sLoadTilesColorTable_##tilemode[R8G8B8_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 24>, R8G8B8_SNORM, R32G32B32A32_FLOAT>::Load; \
 360     sLoadTilesColorTable_##tilemode[BC4_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 64>, BC4_SNORM, R32G32B32A32_FLOAT>::Load; \
 361     sLoadTilesColorTable_##tilemode[BC5_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 128>, BC5_SNORM, R32G32B32A32_FLOAT>::Load; \
 362     sLoadTilesColorTable_##tilemode[R16G16B16_FLOAT]      = LoadMacroTile<TilingTraits<tilemode, 48>, R16G16B16_FLOAT, R32G32B32A32_FLOAT>::Load; \
 363     sLoadTilesColorTable_##tilemode[R16G16B16_UNORM]      = LoadMacroTile<TilingTraits<tilemode, 48>, R16G16B16_UNORM, R32G32B32A32_FLOAT>::Load; \
 364     sLoadTilesColorTable_##tilemode[R16G16B16_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 48>, R16G16B16_SNORM, R32G32B32A32_FLOAT>::Load; \
 365     sLoadTilesColorTable_##tilemode[R8G8B8_UNORM_SRGB]      = LoadMacroTile<TilingTraits<tilemode, 24>, R8G8B8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \
 366     sLoadTilesColorTable_##tilemode[R16G16B16_UINT]      = LoadMacroTile<TilingTraits<tilemode, 48>, R16G16B16_UINT, R32G32B32A32_FLOAT>::Load; \
 367     sLoadTilesColorTable_##tilemode[R16G16B16_SINT]      = LoadMacroTile<TilingTraits<tilemode, 48>, R16G16B16_SINT, R32G32B32A32_FLOAT>::Load; \
 368     sLoadTilesColorTable_##tilemode[R10G10B10A2_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, R10G10B10A2_SNORM, R32G32B32A32_FLOAT>::Load; \
 369     sLoadTilesColorTable_##tilemode[R10G10B10A2_SINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, R10G10B10A2_SINT, R32G32B32A32_FLOAT>::Load; \
 370     sLoadTilesColorTable_##tilemode[B10G10R10A2_SNORM]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10A2_SNORM, R32G32B32A32_FLOAT>::Load; \
 371     sLoadTilesColorTable_##tilemode[B10G10R10A2_UINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10A2_UINT, R32G32B32A32_FLOAT>::Load; \
 372     sLoadTilesColorTable_##tilemode[B10G10R10A2_SINT]      = LoadMacroTile<TilingTraits<tilemode, 32>, B10G10R10A2_SINT, R32G32B32A32_FLOAT>::Load; \
 373     sLoadTilesColorTable_##tilemode[R8G8B8_UINT]      = LoadMacroTile<TilingTraits<tilemode, 24>, R8G8B8_UINT, R32G32B32A32_FLOAT>::Load; \
 374     sLoadTilesColorTable_##tilemode[R8G8B8_SINT]      = LoadMacroTile<TilingTraits<tilemode, 24>, R8G8B8_SINT, R32G32B32A32_FLOAT>::Load; \
 375
 376 //////////////////////////////////////////////////////////////////////////
 377 /// INIT_LOAD_TILES_TABLE - Helper macro for setting up the tables.
 378 #define INIT_LOAD_TILES_DEPTH_TABLE(tilemode) \
 379     memset(sLoadTilesDepthTable_##tilemode, 0, sizeof(sLoadTilesDepthTable_##tilemode)); \
 380     \
 381     sLoadTilesDepthTable_##tilemode[R16_UNORM] = LoadMacroTile<TilingTraits<tilemode, 16>, R16_UNORM, R32_FLOAT>::Load; \
 382     sLoadTilesDepthTable_##tilemode[R32_FLOAT] = LoadMacroTile<TilingTraits<tilemode, 32>, R32_FLOAT, R32_FLOAT>::Load; \
 383     sLoadTilesDepthTable_##tilemode[R24_UNORM_X8_TYPELESS] = LoadMacroTile<TilingTraits<tilemode, 32>, R24_UNORM_X8_TYPELESS, R32_FLOAT>::Load; \
 384
 385 //////////////////////////////////////////////////////////////////////////
 386 /// @brief Sets up tables for LoadTile
 387 void InitSimLoadTilesTable()
 388 {
 389     INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_NONE);
 390     INIT_LOAD_TILES_DEPTH_TABLE(SWR_TILE_NONE);
 391
 392     INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_MODE_YMAJOR);
 393     INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_MODE_XMAJOR);
 394
 395     INIT_LOAD_TILES_DEPTH_TABLE(SWR_TILE_MODE_YMAJOR);
 396 }