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 *rast_task,
  93               const struct lp_rast_triangle *tri,
  94               int x, int y )
  95 {
  96    lp_rast_shade_quads_all(rast_task->rast,
  97                            rast_task->thread_index,
  98                            &tri->inputs,
  99                            x, y);
 100 }
 101
 102
 103 /**
 104  * Shade all pixels in a 16x16 block.
 105  */
 106 static void
 107 block_full_16( struct lp_rasterizer_task *rast_task,
 108                const struct lp_rast_triangle *tri,
 109                int x, int y )
 110 {
 111    unsigned ix, iy;
 112    assert(x % 16 == 0);
 113    assert(y % 16 == 0);
 114    for (iy = 0; iy < 16; iy += 4)
 115       for (ix = 0; ix < 16; ix += 4)
 116          block_full_4(rast_task, tri, x + ix, y + iy);
 117 }
 118
 119
 120 /**
 121  * Pass the 4x4 pixel block to the shader function.
 122  * Determination of which of the 16 pixels lies inside the triangle
 123  * will be done as part of the fragment shader.
 124  */
 125 static void
 126 do_block_4( struct lp_rasterizer_task *rast_task,
 127             const struct lp_rast_triangle *tri,
 128             int x, int y,
 129             int c1,
 130             int c2,
 131             int c3 )
 132 {
 133    lp_rast_shade_quads(rast_task->rast,
 134                        rast_task->thread_index,
 135                        &tri->inputs,
 136                        x, y,
 137                        -c1, -c2, -c3);
 138 }
 139
 140
 141 /**
 142  * Evaluate a 16x16 block of pixels to determine which 4x4 subblocks are in/out
 143  * of the triangle's bounds.
 144  */
 145 static void
 146 do_block_16( struct lp_rasterizer_task *rast_task,
 147              const struct lp_rast_triangle *tri,
 148              int x, int y,
 149              int c0,
 150              int c1,
 151              int c2 )
 152 {
 153    unsigned mask = 0;
 154    int eo[3];
 155    int c[3];
 156    int i, j;
 157
 158    assert(x % 16 == 0);
 159    assert(y % 16 == 0);
 160
 161    eo[0] = tri->eo1 * 4;
 162    eo[1] = tri->eo2 * 4;
 163    eo[2] = tri->eo3 * 4;
 164
 165    c[0] = c0;
 166    c[1] = c1;
 167    c[2] = c2;
 168
 169    for (j = 0; j < 3; j++) {
 170       const int *step = tri->inputs.step[j];
 171       const int cx = c[j] + eo[j];
 172
 173       /* Mask has bits set whenever we are outside any of the edges.
 174        */
 175       for (i = 0; i < 16; i++) {
 176          int out = cx + step[i] * 4;
 177          mask |= (out >> 31) & (1 << i);
 178       }
 179    }
 180
 181    mask = ~mask & 0xffff;
 182    while (mask) {
 183       int i = ffs(mask) - 1;
 184       int px = x + pos_table4[i][0];
 185       int py = y + pos_table4[i][1];
 186       int cx1 = c0 + tri->inputs.step[0][i] * 4;
 187       int cx2 = c1 + tri->inputs.step[1][i] * 4;
 188       int cx3 = c2 + tri->inputs.step[2][i] * 4;
 189
 190       mask &= ~(1 << i);
 191
 192       /* Don't bother testing if the 4x4 block is entirely in/out of
 193        * the triangle.  It's a little faster to do it in the jit code.
 194        */
 195       LP_COUNT(nr_non_empty_4);
 196       do_block_4(rast_task, tri, px, py, cx1, cx2, cx3);
 197    }
 198 }
 199
 200
 201 /**
 202  * Scan the tile in chunks and figure out which pixels to rasterize
 203  * for this triangle.
 204  */
 205 void
 206 lp_rast_triangle( struct lp_rasterizer *rast,
 207                   unsigned thread_index,
 208                   const union lp_rast_cmd_arg arg )
 209 {
 210    struct lp_rasterizer_task *rast_task = &rast->tasks[thread_index];
 211    const struct lp_rast_triangle *tri = arg.triangle;
 212
 213    int x = rast_task->x;
 214    int y = rast_task->y;
 215    int ei[3], eo[3], c[3];
 216    unsigned outmask, inmask, partial_mask;
 217    unsigned i, j;
 218
 219    c[0] = tri->c1 + tri->dx12 * y - tri->dy12 * x;
 220    c[1] = tri->c2 + tri->dx23 * y - tri->dy23 * x;
 221    c[2] = tri->c3 + tri->dx31 * y - tri->dy31 * x;
 222
 223    eo[0] = tri->eo1 * 16;
 224    eo[1] = tri->eo2 * 16;
 225    eo[2] = tri->eo3 * 16;
 226
 227    ei[0] = tri->ei1 * 16;
 228    ei[1] = tri->ei2 * 16;
 229    ei[2] = tri->ei3 * 16;
 230
 231    outmask = 0;
 232    inmask = 0xffff;
 233
 234    for (j = 0; j < 3; j++) {
 235       const int *step = tri->inputs.step[j];
 236       const int cox = c[j] + eo[j];
 237       const int cio = ei[j]- eo[j];
 238
 239       /* Outmask has bits set whenever we are outside any of the
 240        * edges.
 241        */
 242       /* Inmask has bits set whenever we are inside all of the edges.
 243        */
 244       for (i = 0; i < 16; i++) {
 245          int out = cox + step[i] * 16;
 246          int in = out + cio;
 247          outmask |= (out >> 31) & (1 << i);
 248          inmask &= ~((in >> 31) & (1 << i));
 249       }
 250    }
 251
 252    assert((outmask & inmask) == 0);
 253
 254    if (outmask == 0xffff)
 255       return;
 256
 257    /* Invert mask, so that bits are set whenever we are at least
 258     * partially inside all of the edges:
 259     */
 260    partial_mask = ~inmask & ~outmask & 0xffff;
 261
 262    /* Iterate over partials:
 263     */
 264    while (partial_mask) {
 265       int i = ffs(partial_mask) - 1;
 266       int px = x + pos_table16[i][0];
 267       int py = y + pos_table16[i][1];
 268       int cx1 = c[0] + tri->inputs.step[0][i] * 16;
 269       int cx2 = c[1] + tri->inputs.step[1][i] * 16;
 270       int cx3 = c[2] + tri->inputs.step[2][i] * 16;
 271
 272       partial_mask &= ~(1 << i);
 273
 274       LP_COUNT(nr_partially_covered_16);
 275       do_block_16(rast_task, tri, px, py, cx1, cx2, cx3);
 276    }
 277
 278    /* Iterate over fulls:
 279     */
 280    while (inmask) {
 281       int i = ffs(inmask) - 1;
 282       int px = x + pos_table16[i][0];
 283       int py = y + pos_table16[i][1];
 284
 285       inmask &= ~(1 << i);
 286
 287       LP_COUNT(nr_fully_covered_16);
 288       block_full_16(rast_task, tri, px, py);
 289    }
 290 }