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 /**
 118  * Pass the 4x4 pixel block to the shader function.
 119  * Determination of which of the 16 pixels lies inside the triangle
 120  * will be done as part of the fragment shader.
 121  */
 122 static void
 123 do_block_4(struct lp_rasterizer_task *task,
 124            const struct lp_rast_triangle *tri,
 125            int x, int y,
 126            int c1, int c2, int c3)
 127 {
 128    assert(x >= 0);
 129    assert(y >= 0);
 130
 131    lp_rast_shade_quads(task, &tri->inputs, x, y, -c1, -c2, -c3);
 132 }
 133
 134
 135 /**
 136  * Evaluate a 16x16 block of pixels to determine which 4x4 subblocks are in/out
 137  * of the triangle's bounds.
 138  */
 139 static void
 140 do_block_16(struct lp_rasterizer_task *task,
 141             const struct lp_rast_triangle *tri,
 142             int x, int y,
 143             int c0, int c1, int c2)
 144 {
 145    unsigned mask = 0;
 146    int eo[3];
 147    int c[3];
 148    int i, j;
 149
 150    assert(x >= 0);
 151    assert(y >= 0);
 152    assert(x % 16 == 0);
 153    assert(y % 16 == 0);
 154
 155    eo[0] = tri->eo1 * 4;
 156    eo[1] = tri->eo2 * 4;
 157    eo[2] = tri->eo3 * 4;
 158
 159    c[0] = c0;
 160    c[1] = c1;
 161    c[2] = c2;
 162
 163    for (j = 0; j < 3; j++) {
 164       const int *step = tri->inputs.step[j];
 165       const int cx = c[j] + eo[j];
 166
 167       /* Mask has bits set whenever we are outside any of the edges.
 168        */
 169       for (i = 0; i < 16; i++) {
 170          int out = cx + step[i] * 4;
 171          mask |= (out >> 31) & (1 << i);
 172       }
 173    }
 174
 175    mask = ~mask & 0xffff;
 176    while (mask) {
 177       int i = ffs(mask) - 1;
 178       int px = x + pos_table4[i][0];
 179       int py = y + pos_table4[i][1];
 180       int cx1 = c0 + tri->inputs.step[0][i] * 4;
 181       int cx2 = c1 + tri->inputs.step[1][i] * 4;
 182       int cx3 = c2 + tri->inputs.step[2][i] * 4;
 183
 184       mask &= ~(1 << i);
 185
 186       /* Don't bother testing if the 4x4 block is entirely in/out of
 187        * the triangle.  It's a little faster to do it in the jit code.
 188        */
 189       LP_COUNT(nr_non_empty_4);
 190       do_block_4(task, tri, px, py, cx1, cx2, cx3);
 191    }
 192 }
 193
 194
 195 /**
 196  * Scan the tile in chunks and figure out which pixels to rasterize
 197  * for this triangle.
 198  */
 199 void
 200 lp_rast_triangle(struct lp_rasterizer_task *task,
 201                  const union lp_rast_cmd_arg arg)
 202 {
 203    const struct lp_rast_triangle *tri = arg.triangle;
 204    const int x = task->x, y = task->y;
 205    int ei[3], eo[3], c[3];
 206    unsigned outmask, inmask, partial_mask;
 207    unsigned i, j;
 208
 209    c[0] = tri->c1 + tri->dx12 * y - tri->dy12 * x;
 210    c[1] = tri->c2 + tri->dx23 * y - tri->dy23 * x;
 211    c[2] = tri->c3 + tri->dx31 * y - tri->dy31 * x;
 212
 213    eo[0] = tri->eo1 * 16;
 214    eo[1] = tri->eo2 * 16;
 215    eo[2] = tri->eo3 * 16;
 216
 217    ei[0] = tri->ei1 * 16;
 218    ei[1] = tri->ei2 * 16;
 219    ei[2] = tri->ei3 * 16;
 220
 221    outmask = 0;
 222    inmask = 0xffff;
 223
 224    for (j = 0; j < 3; j++) {
 225       const int *step = tri->inputs.step[j];
 226       const int cox = c[j] + eo[j];
 227       const int cio = ei[j]- eo[j];
 228
 229       /* Outmask has bits set whenever we are outside any of the
 230        * edges.
 231        */
 232       /* Inmask has bits set whenever we are inside all of the edges.
 233        */
 234       for (i = 0; i < 16; i++) {
 235          int out = cox + step[i] * 16;
 236          int in = out + cio;
 237          outmask |= (out >> 31) & (1 << i);
 238          inmask &= ~((in >> 31) & (1 << i));
 239       }
 240    }
 241
 242    assert((outmask & inmask) == 0);
 243
 244    if (outmask == 0xffff)
 245       return;
 246
 247    /* Invert mask, so that bits are set whenever we are at least
 248     * partially inside all of the edges:
 249     */
 250    partial_mask = ~inmask & ~outmask & 0xffff;
 251
 252    /* Iterate over partials:
 253     */
 254    while (partial_mask) {
 255       int i = ffs(partial_mask) - 1;
 256       int px = x + pos_table16[i][0];
 257       int py = y + pos_table16[i][1];
 258       int cx1 = c[0] + tri->inputs.step[0][i] * 16;
 259       int cx2 = c[1] + tri->inputs.step[1][i] * 16;
 260       int cx3 = c[2] + tri->inputs.step[2][i] * 16;
 261
 262       partial_mask &= ~(1 << i);
 263
 264       LP_COUNT(nr_partially_covered_16);
 265       do_block_16(task, tri, px, py, cx1, cx2, cx3);
 266    }
 267
 268    /* Iterate over fulls:
 269     */
 270    while (inmask) {
 271       int i = ffs(inmask) - 1;
 272       int px = x + pos_table16[i][0];
 273       int py = y + pos_table16[i][1];
 274
 275       inmask &= ~(1 << i);
 276
 277       LP_COUNT(nr_fully_covered_16);
 278       block_full_16(task, tri, px, py);
 279    }
 280 }