src/gallium/drivers/llvmpipe/lp_rast_tri_tmp.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007-2010 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
  33
  34 /**
  35  * Prototype for a 8 plane rasterizer function.  Will codegenerate
  36  * several of these.
  37  *
  38  * XXX: Varients for more/fewer planes.
  39  * XXX: Need ways of dropping planes as we descend.
  40  * XXX: SIMD
  41  */
  42 static void
  43 TAG(do_block_4)(struct lp_rasterizer_task *task,
  44                 const struct lp_rast_triangle *tri,
  45                 const struct lp_rast_plane *plane,
  46                 int x, int y,
  47                 const int *c)
  48 {
  49    unsigned mask = 0xffff;
  50    int j;
  51
  52    for (j = 0; j < NR_PLANES; j++) {
  53       mask &= ~build_mask_linear(c[j] - 1,
  54                                  -plane[j].dcdx,
  55                                  plane[j].dcdy);
  56    }
  57
  58    /* Now pass to the shader:
  59     */
  60    if (mask)
  61       lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
  62 }
  63
  64 /**
  65  * Evaluate a 16x16 block of pixels to determine which 4x4 subblocks are in/out
  66  * of the triangle's bounds.
  67  */
  68 static void
  69 TAG(do_block_16)(struct lp_rasterizer_task *task,
  70                  const struct lp_rast_triangle *tri,
  71                  const struct lp_rast_plane *plane,
  72                  int x, int y,
  73                  const int *c)
  74 {
  75    unsigned outmask, inmask, partmask, partial_mask;
  76    unsigned j;
  77
  78    outmask = 0;                 /* outside one or more trivial reject planes */
  79    partmask = 0;                /* outside one or more trivial accept planes */
  80
  81    for (j = 0; j < NR_PLANES; j++) {
  82       const int dcdx = -plane[j].dcdx * 4;
  83       const int dcdy = plane[j].dcdy * 4;
  84       const int cox = plane[j].eo * 4;
  85       const int cio = plane[j].ei * 4 - 1;
  86
  87       build_masks(c[j] + cox,
  88                   cio - cox,
  89                   dcdx, dcdy,
  90                   &outmask,   /* sign bits from c[i][0..15] + cox */
  91                   &partmask); /* sign bits from c[i][0..15] + cio */
  92    }
  93
  94    if (outmask == 0xffff)
  95       return;
  96
  97    /* Mask of sub-blocks which are inside all trivial accept planes:
  98     */
  99    inmask = ~partmask & 0xffff;
 100
 101    /* Mask of sub-blocks which are inside all trivial reject planes,
 102     * but outside at least one trivial accept plane:
 103     */
 104    partial_mask = partmask & ~outmask;
 105
 106    assert((partial_mask & inmask) == 0);
 107
 108    LP_COUNT_ADD(nr_empty_4, util_bitcount(0xffff & ~(partial_mask | inmask)));
 109
 110    /* Iterate over partials:
 111     */
 112    while (partial_mask) {
 113       int i = ffs(partial_mask) - 1;
 114       int ix = (i & 3) * 4;
 115       int iy = (i >> 2) * 4;
 116       int px = x + ix;
 117       int py = y + iy;
 118       int cx[NR_PLANES];
 119
 120       partial_mask &= ~(1 << i);
 121
 122       LP_COUNT(nr_partially_covered_4);
 123
 124       for (j = 0; j < NR_PLANES; j++)
 125          cx[j] = (c[j]
 126                   - plane[j].dcdx * ix
 127                   + plane[j].dcdy * iy);
 128
 129       TAG(do_block_4)(task, tri, plane, px, py, cx);
 130    }
 131
 132    /* Iterate over fulls:
 133     */
 134    while (inmask) {
 135       int i = ffs(inmask) - 1;
 136       int ix = (i & 3) * 4;
 137       int iy = (i >> 2) * 4;
 138       int px = x + ix;
 139       int py = y + iy;
 140
 141       inmask &= ~(1 << i);
 142
 143       LP_COUNT(nr_fully_covered_4);
 144       block_full_4(task, tri, px, py);
 145    }
 146 }
 147
 148
 149 /**
 150  * Scan the tile in chunks and figure out which pixels to rasterize
 151  * for this triangle.
 152  */
 153 void
 154 TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
 155                       const union lp_rast_cmd_arg arg)
 156 {
 157    const struct lp_rast_triangle *tri = arg.triangle.tri;
 158    unsigned plane_mask = arg.triangle.plane_mask;
 159    const struct lp_rast_plane *tri_plane = GET_PLANES(tri);
 160    const int x = task->x, y = task->y;
 161    struct lp_rast_plane plane[NR_PLANES];
 162    int c[NR_PLANES];
 163    unsigned outmask, inmask, partmask, partial_mask;
 164    unsigned j = 0;
 165
 166    if (tri->inputs.disable) {
 167       /* This triangle was partially binned and has been disabled */
 168       return;
 169    }
 170
 171    outmask = 0;                 /* outside one or more trivial reject planes */
 172    partmask = 0;                /* outside one or more trivial accept planes */
 173
 174    while (plane_mask) {
 175       int i = ffs(plane_mask) - 1;
 176       plane[j] = tri_plane[i];
 177       plane_mask &= ~(1 << i);
 178       c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
 179
 180       {
 181          const int dcdx = -plane[j].dcdx * 16;
 182          const int dcdy = plane[j].dcdy * 16;
 183          const int cox = plane[j].eo * 16;
 184          const int cio = plane[j].ei * 16 - 1;
 185
 186          build_masks(c[j] + cox,
 187                      cio - cox,
 188                      dcdx, dcdy,
 189                      &outmask,   /* sign bits from c[i][0..15] + cox */
 190                      &partmask); /* sign bits from c[i][0..15] + cio */
 191       }
 192
 193       j++;
 194    }
 195
 196    if (outmask == 0xffff)
 197       return;
 198
 199    /* Mask of sub-blocks which are inside all trivial accept planes:
 200     */
 201    inmask = ~partmask & 0xffff;
 202
 203    /* Mask of sub-blocks which are inside all trivial reject planes,
 204     * but outside at least one trivial accept plane:
 205     */
 206    partial_mask = partmask & ~outmask;
 207
 208    assert((partial_mask & inmask) == 0);
 209
 210    LP_COUNT_ADD(nr_empty_16, util_bitcount(0xffff & ~(partial_mask | inmask)));
 211
 212    /* Iterate over partials:
 213     */
 214    while (partial_mask) {
 215       int i = ffs(partial_mask) - 1;
 216       int ix = (i & 3) * 16;
 217       int iy = (i >> 2) * 16;
 218       int px = x + ix;
 219       int py = y + iy;
 220       int cx[NR_PLANES];
 221
 222       for (j = 0; j < NR_PLANES; j++)
 223          cx[j] = (c[j]
 224                   - plane[j].dcdx * ix
 225                   + plane[j].dcdy * iy);
 226
 227       partial_mask &= ~(1 << i);
 228
 229       LP_COUNT(nr_partially_covered_16);
 230       TAG(do_block_16)(task, tri, plane, px, py, cx);
 231    }
 232
 233    /* Iterate over fulls:
 234     */
 235    while (inmask) {
 236       int i = ffs(inmask) - 1;
 237       int ix = (i & 3) * 16;
 238       int iy = (i >> 2) * 16;
 239       int px = x + ix;
 240       int py = y + iy;
 241
 242       inmask &= ~(1 << i);
 243
 244       LP_COUNT(nr_fully_covered_16);
 245       block_full_16(task, tri, px, py);
 246    }
 247 }
 248
 249 #if defined(PIPE_ARCH_SSE) && defined(TRI_16)
 250 /* XXX: special case this when intersection is not required.
 251  *      - tile completely within bbox,
 252  *      - bbox completely within tile.
 253  */
 254 void
 255 TRI_16(struct lp_rasterizer_task *task,
 256        const union lp_rast_cmd_arg arg)
 257 {
 258    const struct lp_rast_triangle *tri = arg.triangle.tri;
 259    const struct lp_rast_plane *plane = GET_PLANES(tri);
 260    unsigned mask = arg.triangle.plane_mask;
 261    unsigned outmask, partial_mask;
 262    unsigned j;
 263    __m128i cstep4[NR_PLANES][4];
 264
 265    int x = (mask & 0xff);
 266    int y = (mask >> 8);
 267
 268    outmask = 0;                 /* outside one or more trivial reject planes */
 269
 270    x += task->x;
 271    y += task->y;
 272
 273    for (j = 0; j < NR_PLANES; j++) {
 274       const int dcdx = -plane[j].dcdx * 4;
 275       const int dcdy = plane[j].dcdy * 4;
 276       __m128i xdcdy = _mm_set1_epi32(dcdy);
 277
 278       cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
 279       cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
 280       cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
 281       cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
 282
 283       {
 284          const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
 285          const int cox = plane[j].eo * 4;
 286
 287          outmask |= sign_bits4(cstep4[j], c + cox);
 288       }
 289    }
 290
 291    if (outmask == 0xffff)
 292       return;
 293
 294
 295    /* Mask of sub-blocks which are inside all trivial reject planes,
 296     * but outside at least one trivial accept plane:
 297     */
 298    partial_mask = 0xffff & ~outmask;
 299
 300    /* Iterate over partials:
 301     */
 302    while (partial_mask) {
 303       int i = ffs(partial_mask) - 1;
 304       int ix = (i & 3) * 4;
 305       int iy = (i >> 2) * 4;
 306       int px = x + ix;
 307       int py = y + iy;
 308       unsigned mask = 0xffff;
 309
 310       partial_mask &= ~(1 << i);
 311
 312       for (j = 0; j < NR_PLANES; j++) {
 313          const int cx = (plane[j].c - 1
 314                          - plane[j].dcdx * px
 315                          + plane[j].dcdy * py) * 4;
 316
 317          mask &= ~sign_bits4(cstep4[j], cx);
 318       }
 319
 320       if (mask)
 321          lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
 322    }
 323 }
 324 #endif
 325
 326 #if defined(PIPE_ARCH_SSE) && defined(TRI_4)
 327 void
 328 TRI_4(struct lp_rasterizer_task *task,
 329       const union lp_rast_cmd_arg arg)
 330 {
 331    const struct lp_rast_triangle *tri = arg.triangle.tri;
 332    const struct lp_rast_plane *plane = GET_PLANES(tri);
 333    unsigned mask = arg.triangle.plane_mask;
 334    const int x = task->x + (mask & 0xff);
 335    const int y = task->y + (mask >> 8);
 336    unsigned j;
 337
 338    /* Iterate over partials:
 339     */
 340    {
 341       unsigned mask = 0xffff;
 342
 343       for (j = 0; j < NR_PLANES; j++) {
 344          const int cx = (plane[j].c
 345                          - plane[j].dcdx * x
 346                          + plane[j].dcdy * y);
 347
 348          const int dcdx = -plane[j].dcdx;
 349          const int dcdy = plane[j].dcdy;
 350          __m128i xdcdy = _mm_set1_epi32(dcdy);
 351
 352          __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
 353          __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 354          __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 355          __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 356
 357          __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 358          __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 359          __m128i result = _mm_packs_epi16(cstep01, cstep23);
 360
 361          /* Extract the sign bits
 362           */
 363          mask &= ~_mm_movemask_epi8(result);
 364       }
 365
 366       if (mask)
 367          lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
 368    }
 369 }
 370 #endif
 371
 372
 373
 374 #undef TAG
 375 #undef TRI_4
 376 #undef TRI_16
 377 #undef NR_PLANES
 378