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_rast_priv.h"
  36 #include "lp_tile_soa.h"
  37
  38
  39 /**
  40  * Map an index in [0,15] to an x,y position, multiplied by 4.
  41  * This is used to get the position of each subtile in a 4x4
  42  * grid of edge step values.
  43  * Note: we can use some bit twiddling to compute these values instead
  44  * of using a look-up table, but there's no measurable performance
  45  * difference.
  46  */
  47 static const int pos_table4[16][2] = {
  48    { 0, 0 },
  49    { 4, 0 },
  50    { 0, 4 },
  51    { 4, 4 },
  52    { 8, 0 },
  53    { 12, 0 },
  54    { 8, 4 },
  55    { 12, 4 },
  56    { 0, 8 },
  57    { 4, 8 },
  58    { 0, 12 },
  59    { 4, 12 },
  60    { 8, 8 },
  61    { 12, 8 },
  62    { 8, 12 },
  63    { 12, 12 }
  64 };
  65
  66
  67 static const int pos_table16[16][2] = {
  68    { 0, 0 },
  69    { 16, 0 },
  70    { 0, 16 },
  71    { 16, 16 },
  72    { 32, 0 },
  73    { 48, 0 },
  74    { 32, 16 },
  75    { 48, 16 },
  76    { 0, 32 },
  77    { 16, 32 },
  78    { 0, 48 },
  79    { 16, 48 },
  80    { 32, 32 },
  81    { 48, 32 },
  82    { 32, 48 },
  83    { 48, 48 }
  84 };
  85
  86
  87 /**
  88  * Shade all pixels in a 4x4 block.
  89  */
  90 static void
  91 block_full_4( struct lp_rasterizer_task *rast_task,
  92               const struct lp_rast_triangle *tri,
  93               int x, int y )
  94 {
  95    lp_rast_shade_quads_all(rast_task->rast,
  96                            rast_task->thread_index,
  97                            &tri->inputs,
  98                            x, y);
  99 }
 100
 101
 102 /**
 103  * Shade all pixels in a 16x16 block.
 104  */
 105 static void
 106 block_full_16( struct lp_rasterizer_task *rast_task,
 107                const struct lp_rast_triangle *tri,
 108                int x, int y )
 109 {
 110    unsigned ix, iy;
 111    assert(x % 16 == 0);
 112    assert(y % 16 == 0);
 113    for (iy = 0; iy < 16; iy += 4)
 114       for (ix = 0; ix < 16; ix += 4)
 115          block_full_4(rast_task, tri, x + ix, y + iy);
 116 }
 117
 118
 119 /**
 120  * Pass the 4x4 pixel block to the shader function.
 121  * Determination of which of the 16 pixels lies inside the triangle
 122  * will be done as part of the fragment shader.
 123  */
 124 static void
 125 do_block_4( struct lp_rasterizer_task *rast_task,
 126             const struct lp_rast_triangle *tri,
 127             int x, int y,
 128             int c1,
 129             int c2,
 130             int c3 )
 131 {
 132    lp_rast_shade_quads(rast_task->rast,
 133                        rast_task->thread_index,
 134                        &tri->inputs,
 135                        x, y,
 136                        -c1, -c2, -c3);
 137 }
 138
 139
 140 /**
 141  * Evaluate a 16x16 block of pixels to determine which 4x4 subblocks are in/out
 142  * of the triangle's bounds.
 143  */
 144 static void
 145 do_block_16( struct lp_rasterizer_task *rast_task,
 146              const struct lp_rast_triangle *tri,
 147              int x, int y,
 148              int c1,
 149              int c2,
 150              int c3 )
 151 {
 152    const int eo1 = tri->eo1 * 4;
 153    const int eo2 = tri->eo2 * 4;
 154    const int eo3 = tri->eo3 * 4;
 155
 156    int i;
 157
 158    assert(x % 16 == 0);
 159    assert(y % 16 == 0);
 160
 161    for (i = 0; i < 16; i++) {
 162       int cx1 = c1 + (tri->inputs.step[0][i] * 4);
 163       int cx2 = c2 + (tri->inputs.step[1][i] * 4);
 164       int cx3 = c3 + (tri->inputs.step[2][i] * 4);
 165
 166       if (cx1 + eo1 < 0 ||
 167           cx2 + eo2 < 0 ||
 168           cx3 + eo3 < 0) {
 169          /* the block is completely outside the triangle - nop */
 170       }
 171       else {
 172          int px = x + pos_table4[i][0];
 173          int py = y + pos_table4[i][1];
 174          /* Don't bother testing if the 4x4 block is entirely in/out of
 175           * the triangle.  It's a little faster to do it in the jit code.
 176           */
 177          do_block_4(rast_task, tri, px, py, cx1, cx2, cx3);
 178       }
 179    }
 180 }
 181
 182
 183 /**
 184  * Scan the tile in chunks and figure out which pixels to rasterize
 185  * for this triangle.
 186  */
 187 void
 188 lp_rast_triangle( struct lp_rasterizer *rast,
 189                   unsigned thread_index,
 190                   const union lp_rast_cmd_arg arg )
 191 {
 192    struct lp_rasterizer_task *rast_task = &rast->tasks[thread_index];
 193    const struct lp_rast_triangle *tri = arg.triangle;
 194
 195    int x = rast_task->x;
 196    int y = rast_task->y;
 197    unsigned i;
 198
 199    int c1 = tri->c1 + tri->dx12 * y - tri->dy12 * x;
 200    int c2 = tri->c2 + tri->dx23 * y - tri->dy23 * x;
 201    int c3 = tri->c3 + tri->dx31 * y - tri->dy31 * x;
 202
 203    int ei1 = tri->ei1 * 16;
 204    int ei2 = tri->ei2 * 16;
 205    int ei3 = tri->ei3 * 16;
 206
 207    int eo1 = tri->eo1 * 16;
 208    int eo2 = tri->eo2 * 16;
 209    int eo3 = tri->eo3 * 16;
 210
 211    LP_DBG(DEBUG_RAST, "lp_rast_triangle\n");
 212
 213    /* Walk over the tile to build a list of 4x4 pixel blocks which will
 214     * be filled/shaded.  We do this at two granularities: 16x16 blocks
 215     * and then 4x4 blocks.
 216     */
 217    for (i = 0; i < 16; i++) {
 218       int cx1 = c1 + (tri->inputs.step[0][i] * 16);
 219       int cx2 = c2 + (tri->inputs.step[1][i] * 16);
 220       int cx3 = c3 + (tri->inputs.step[2][i] * 16);
 221
 222       if (cx1 + eo1 < 0 ||
 223           cx2 + eo2 < 0 ||
 224           cx3 + eo3 < 0) {
 225          /* the block is completely outside the triangle - nop */
 226       }
 227       else {
 228          int px = x + pos_table16[i][0];
 229          int py = y + pos_table16[i][1];
 230
 231          if (cx1 + ei1 > 0 &&
 232              cx2 + ei2 > 0 &&
 233              cx3 + ei3 > 0) {
 234             /* the block is completely inside the triangle */
 235             block_full_16(rast_task, tri, px, py);
 236          }
 237          else {
 238             /* the block is partially in/out of the triangle */
 239             do_block_16(rast_task, tri, px, py, cx1, cx2, cx3);
 240          }
 241       }
 242    }
 243 }