src/gallium/drivers/llvmpipe/lp_rast_tri.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007-2009 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /*
  29  * Rasterization for binned triangles within a tile
  30  */
  31
  32 #include <limits.h>
  33 #include "util/u_math.h"
  34 #include "lp_debug.h"
  35 #include "lp_perf.h"
  36 #include "lp_rast_priv.h"
  37 #include "lp_tile_soa.h"
  38
  39
  40 /**
  41  * Map an index in [0,15] to an x,y position, multiplied by 4.
  42  * This is used to get the position of each subtile in a 4x4
  43  * grid of edge step values.
  44  * Note: we can use some bit twiddling to compute these values instead
  45  * of using a look-up table, but there's no measurable performance
  46  * difference.
  47  */
  48 static const int pos_table4[16][2] = {
  49    { 0, 0 },
  50    { 4, 0 },
  51    { 0, 4 },
  52    { 4, 4 },
  53    { 8, 0 },
  54    { 12, 0 },
  55    { 8, 4 },
  56    { 12, 4 },
  57    { 0, 8 },
  58    { 4, 8 },
  59    { 0, 12 },
  60    { 4, 12 },
  61    { 8, 8 },
  62    { 12, 8 },
  63    { 8, 12 },
  64    { 12, 12 }
  65 };
  66
  67
  68 static const int pos_table16[16][2] = {
  69    { 0, 0 },
  70    { 16, 0 },
  71    { 0, 16 },
  72    { 16, 16 },
  73    { 32, 0 },
  74    { 48, 0 },
  75    { 32, 16 },
  76    { 48, 16 },
  77    { 0, 32 },
  78    { 16, 32 },
  79    { 0, 48 },
  80    { 16, 48 },
  81    { 32, 32 },
  82    { 48, 32 },
  83    { 32, 48 },
  84    { 48, 48 }
  85 };
  86
  87
  88 /**
  89  * Shade all pixels in a 4x4 block.
  90  */
  91 static void
  92 block_full_4(struct lp_rasterizer_task *task,
  93              const struct lp_rast_triangle *tri,
  94              int x, int y)
  95 {
  96    lp_rast_shade_quads_all(task, &tri->inputs, x, y);
  97 }
  98
  99
 100 /**
 101  * Shade all pixels in a 16x16 block.
 102  */
 103 static void
 104 block_full_16(struct lp_rasterizer_task *task,
 105               const struct lp_rast_triangle *tri,
 106               int x, int y)
 107 {
 108    unsigned ix, iy;
 109    assert(x % 16 == 0);
 110    assert(y % 16 == 0);
 111    for (iy = 0; iy < 16; iy += 4)
 112       for (ix = 0; ix < 16; ix += 4)
 113          block_full_4(task, tri, x + ix, y + iy);
 114 }
 115
 116
 117 static INLINE unsigned
 118 build_mask(int c, int dcdx, int dcdy)
 119 {
 120    int mask = 0;
 121
 122    int c0 = c;
 123    int c1 = c0 + dcdx;
 124    int c2 = c1 + dcdx;
 125    int c3 = c2 + dcdx;
 126
 127    mask |= ((c0 + 0 * dcdy) >> 31) & (1 << 0);
 128    mask |= ((c0 + 1 * dcdy) >> 31) & (1 << 2);
 129    mask |= ((c0 + 2 * dcdy) >> 31) & (1 << 8);
 130    mask |= ((c0 + 3 * dcdy) >> 31) & (1 << 10);
 131    mask |= ((c1 + 0 * dcdy) >> 31) & (1 << 1);
 132    mask |= ((c1 + 1 * dcdy) >> 31) & (1 << 3);
 133    mask |= ((c1 + 2 * dcdy) >> 31) & (1 << 9);
 134    mask |= ((c1 + 3 * dcdy) >> 31) & (1 << 11);
 135    mask |= ((c2 + 0 * dcdy) >> 31) & (1 << 4);
 136    mask |= ((c2 + 1 * dcdy) >> 31) & (1 << 6);
 137    mask |= ((c2 + 2 * dcdy) >> 31) & (1 << 12);
 138    mask |= ((c2 + 3 * dcdy) >> 31) & (1 << 14);
 139    mask |= ((c3 + 0 * dcdy) >> 31) & (1 << 5);
 140    mask |= ((c3 + 1 * dcdy) >> 31) & (1 << 7);
 141    mask |= ((c3 + 2 * dcdy) >> 31) & (1 << 13);
 142    mask |= ((c3 + 3 * dcdy) >> 31) & (1 << 15);
 143
 144    return mask;
 145 }
 146
 147
 148
 149 #define TAG(x) x##_1
 150 #define NR_PLANES 1
 151 #include "lp_rast_tri_tmp.h"
 152
 153 #define TAG(x) x##_2
 154 #define NR_PLANES 2
 155 #include "lp_rast_tri_tmp.h"
 156
 157 #define TAG(x) x##_3
 158 #define NR_PLANES 3
 159 #include "lp_rast_tri_tmp.h"
 160
 161 #define TAG(x) x##_4
 162 #define NR_PLANES 4
 163 #include "lp_rast_tri_tmp.h"
 164
 165 #define TAG(x) x##_5
 166 #define NR_PLANES 5
 167 #include "lp_rast_tri_tmp.h"
 168
 169 #define TAG(x) x##_6
 170 #define NR_PLANES 6
 171 #include "lp_rast_tri_tmp.h"
 172
 173 #define TAG(x) x##_7
 174 #define NR_PLANES 7
 175 #include "lp_rast_tri_tmp.h"
 176