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
  42 /**
  43  * Shade all pixels in a 4x4 block.
  44  */
  45 static void
  46 block_full_4(struct lp_rasterizer_task *task,
  47              const struct lp_rast_triangle *tri,
  48              int x, int y)
  49 {
  50    lp_rast_shade_quads_all(task, &tri->inputs, x, y);
  51 }
  52
  53
  54 /**
  55  * Shade all pixels in a 16x16 block.
  56  */
  57 static void
  58 block_full_16(struct lp_rasterizer_task *task,
  59               const struct lp_rast_triangle *tri,
  60               int x, int y)
  61 {
  62    unsigned ix, iy;
  63    assert(x % 16 == 0);
  64    assert(y % 16 == 0);
  65    for (iy = 0; iy < 16; iy += 4)
  66       for (ix = 0; ix < 16; ix += 4)
  67          block_full_4(task, tri, x + ix, y + iy);
  68 }
  69
  70 #if !defined(PIPE_ARCH_SSE)
  71 static INLINE unsigned
  72 build_mask(int c, int dcdx, int dcdy)
  73 {
  74    int mask = 0;
  75
  76    int c0 = c;
  77    int c1 = c0 + dcdx;
  78    int c2 = c1 + dcdx;
  79    int c3 = c2 + dcdx;
  80
  81    mask |= ((c0 + 0 * dcdy) >> 31) & (1 << 0);
  82    mask |= ((c0 + 1 * dcdy) >> 31) & (1 << 2);
  83    mask |= ((c0 + 2 * dcdy) >> 31) & (1 << 8);
  84    mask |= ((c0 + 3 * dcdy) >> 31) & (1 << 10);
  85    mask |= ((c1 + 0 * dcdy) >> 31) & (1 << 1);
  86    mask |= ((c1 + 1 * dcdy) >> 31) & (1 << 3);
  87    mask |= ((c1 + 2 * dcdy) >> 31) & (1 << 9);
  88    mask |= ((c1 + 3 * dcdy) >> 31) & (1 << 11);
  89    mask |= ((c2 + 0 * dcdy) >> 31) & (1 << 4);
  90    mask |= ((c2 + 1 * dcdy) >> 31) & (1 << 6);
  91    mask |= ((c2 + 2 * dcdy) >> 31) & (1 << 12);
  92    mask |= ((c2 + 3 * dcdy) >> 31) & (1 << 14);
  93    mask |= ((c3 + 0 * dcdy) >> 31) & (1 << 5);
  94    mask |= ((c3 + 1 * dcdy) >> 31) & (1 << 7);
  95    mask |= ((c3 + 2 * dcdy) >> 31) & (1 << 13);
  96    mask |= ((c3 + 3 * dcdy) >> 31) & (1 << 15);
  97
  98    return mask;
  99 }
 100
 101
 102 static INLINE unsigned
 103 build_mask_linear(int c, int dcdx, int dcdy)
 104 {
 105    int mask = 0;
 106
 107    int c0 = c;
 108    int c1 = c0 + dcdy;
 109    int c2 = c1 + dcdy;
 110    int c3 = c2 + dcdy;
 111
 112    mask |= ((c0 + 0 * dcdx) >> 31) & (1 << 0);
 113    mask |= ((c0 + 1 * dcdx) >> 31) & (1 << 1);
 114    mask |= ((c0 + 2 * dcdx) >> 31) & (1 << 2);
 115    mask |= ((c0 + 3 * dcdx) >> 31) & (1 << 3);
 116    mask |= ((c1 + 0 * dcdx) >> 31) & (1 << 4);
 117    mask |= ((c1 + 1 * dcdx) >> 31) & (1 << 5);
 118    mask |= ((c1 + 2 * dcdx) >> 31) & (1 << 6);
 119    mask |= ((c1 + 3 * dcdx) >> 31) & (1 << 7);
 120    mask |= ((c2 + 0 * dcdx) >> 31) & (1 << 8);
 121    mask |= ((c2 + 1 * dcdx) >> 31) & (1 << 9);
 122    mask |= ((c2 + 2 * dcdx) >> 31) & (1 << 10);
 123    mask |= ((c2 + 3 * dcdx) >> 31) & (1 << 11);
 124    mask |= ((c3 + 0 * dcdx) >> 31) & (1 << 12);
 125    mask |= ((c3 + 1 * dcdx) >> 31) & (1 << 13);
 126    mask |= ((c3 + 2 * dcdx) >> 31) & (1 << 14);
 127    mask |= ((c3 + 3 * dcdx) >> 31) & (1 << 15);
 128
 129    return mask;
 130 }
 131
 132
 133 static INLINE void
 134 build_masks(int c,
 135             int cdiff,
 136             int dcdx,
 137             int dcdy,
 138             unsigned *outmask,
 139             unsigned *partmask)
 140 {
 141    *outmask |= build_mask_linear(c, dcdx, dcdy);
 142    *partmask |= build_mask_linear(c + cdiff, dcdx, dcdy);
 143 }
 144
 145 #else
 146 #include <emmintrin.h>
 147 #include "util/u_sse.h"
 148
 149
 150 static INLINE void
 151 build_masks(int c,
 152             int cdiff,
 153             int dcdx,
 154             int dcdy,
 155             unsigned *outmask,
 156             unsigned *partmask)
 157 {
 158    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
 159    __m128i xdcdy = _mm_set1_epi32(dcdy);
 160
 161    /* Get values across the quad
 162     */
 163    __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 164    __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 165    __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 166
 167    {
 168       __m128i cstep01, cstep23, result;
 169
 170       cstep01 = _mm_packs_epi32(cstep0, cstep1);
 171       cstep23 = _mm_packs_epi32(cstep2, cstep3);
 172       result = _mm_packs_epi16(cstep01, cstep23);
 173
 174       *outmask |= _mm_movemask_epi8(result);
 175    }
 176
 177
 178    {
 179       __m128i cio4 = _mm_set1_epi32(cdiff);
 180       __m128i cstep01, cstep23, result;
 181
 182       cstep0 = _mm_add_epi32(cstep0, cio4);
 183       cstep1 = _mm_add_epi32(cstep1, cio4);
 184       cstep2 = _mm_add_epi32(cstep2, cio4);
 185       cstep3 = _mm_add_epi32(cstep3, cio4);
 186
 187       cstep01 = _mm_packs_epi32(cstep0, cstep1);
 188       cstep23 = _mm_packs_epi32(cstep2, cstep3);
 189       result = _mm_packs_epi16(cstep01, cstep23);
 190
 191       *partmask |= _mm_movemask_epi8(result);
 192    }
 193 }
 194
 195
 196 static INLINE unsigned
 197 build_mask_linear(int c, int dcdx, int dcdy)
 198 {
 199    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
 200    __m128i xdcdy = _mm_set1_epi32(dcdy);
 201
 202    /* Get values across the quad
 203     */
 204    __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 205    __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 206    __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 207
 208    /* pack pairs of results into epi16
 209     */
 210    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 211    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 212
 213    /* pack into epi8, preserving sign bits
 214     */
 215    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 216
 217    /* extract sign bits to create mask
 218     */
 219    return _mm_movemask_epi8(result);
 220 }
 221
 222 static INLINE unsigned
 223 build_mask(int c, int dcdx, int dcdy)
 224 {
 225    __m128i step = _mm_setr_epi32(0, dcdx, dcdy, dcdx + dcdy);
 226    __m128i c0 = _mm_set1_epi32(c);
 227
 228    /* Get values across the quad
 229     */
 230    __m128i cstep0 = _mm_add_epi32(c0, step);
 231
 232    /* Scale up step for moving between quads.
 233     */
 234    __m128i step4 = _mm_add_epi32(step, step);
 235
 236    /* Get values for the remaining quads:
 237     */
 238    __m128i cstep1 = _mm_add_epi32(cstep0,
 239                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1)));
 240    __m128i cstep2 = _mm_add_epi32(cstep0,
 241                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(2,2,2,2)));
 242    __m128i cstep3 = _mm_add_epi32(cstep2,
 243                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1)));
 244
 245    /* pack pairs of results into epi16
 246     */
 247    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 248    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 249
 250    /* pack into epi8, preserving sign bits
 251     */
 252    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 253
 254    /* extract sign bits to create mask
 255     */
 256    return _mm_movemask_epi8(result);
 257 }
 258
 259 #endif
 260
 261
 262
 263
 264 #define TAG(x) x##_1
 265 #define NR_PLANES 1
 266 #include "lp_rast_tri_tmp.h"
 267
 268 #define TAG(x) x##_2
 269 #define NR_PLANES 2
 270 #include "lp_rast_tri_tmp.h"
 271
 272 #define TAG(x) x##_3
 273 #define NR_PLANES 3
 274 #include "lp_rast_tri_tmp.h"
 275
 276 #define TAG(x) x##_4
 277 #define NR_PLANES 4
 278 #include "lp_rast_tri_tmp.h"
 279
 280 #define TAG(x) x##_5
 281 #define NR_PLANES 5
 282 #include "lp_rast_tri_tmp.h"
 283
 284 #define TAG(x) x##_6
 285 #define NR_PLANES 6
 286 #include "lp_rast_tri_tmp.h"
 287
 288 #define TAG(x) x##_7
 289 #define NR_PLANES 7
 290 #include "lp_rast_tri_tmp.h"
 291
 292 #define TAG(x) x##_8
 293 #define NR_PLANES 8
 294 #include "lp_rast_tri_tmp.h"
 295
 296
 297 /* Special case for 3 plane triangle which is contained entirely
 298  * within a 16x16 block.
 299  */
 300 void
 301 lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 302                       const union lp_rast_cmd_arg arg)
 303 {
 304    const struct lp_rast_triangle *tri = arg.triangle.tri;
 305    const struct lp_rast_plane *plane = tri->plane;
 306    unsigned mask = arg.triangle.plane_mask;
 307    const int x = task->x + (mask & 0xf) * 16;
 308    const int y = task->y + (mask >> 4) * 16;
 309    unsigned outmask, inmask, partmask, partial_mask;
 310    unsigned j;
 311    int c[3];
 312
 313    outmask = 0;                 /* outside one or more trivial reject planes */
 314    partmask = 0;                /* outside one or more trivial accept planes */
 315
 316    for (j = 0; j < 3; j++) {
 317       c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
 318
 319       {
 320          const int dcdx = -plane[j].dcdx * 4;
 321          const int dcdy = plane[j].dcdy * 4;
 322          const int cox = plane[j].eo * 4;
 323          const int cio = plane[j].ei * 4 - 1;
 324
 325          build_masks(c[j] + cox,
 326                      cio - cox,
 327                      dcdx, dcdy,
 328                      &outmask,   /* sign bits from c[i][0..15] + cox */
 329                      &partmask); /* sign bits from c[i][0..15] + cio */
 330       }
 331    }
 332
 333    if (outmask == 0xffff)
 334       return;
 335
 336    /* Mask of sub-blocks which are inside all trivial accept planes:
 337     */
 338    inmask = ~partmask & 0xffff;
 339
 340    /* Mask of sub-blocks which are inside all trivial reject planes,
 341     * but outside at least one trivial accept plane:
 342     */
 343    partial_mask = partmask & ~outmask;
 344
 345    assert((partial_mask & inmask) == 0);
 346
 347    /* Iterate over partials:
 348     */
 349    while (partial_mask) {
 350       int i = ffs(partial_mask) - 1;
 351       int ix = (i & 3) * 4;
 352       int iy = (i >> 2) * 4;
 353       int px = x + ix;
 354       int py = y + iy;
 355       int cx[3];
 356
 357       partial_mask &= ~(1 << i);
 358
 359       for (j = 0; j < 3; j++)
 360          cx[j] = (c[j]
 361                   - plane[j].dcdx * ix
 362                   + plane[j].dcdy * iy);
 363
 364       do_block_4_3(task, tri, plane, px, py, cx);
 365    }
 366
 367    /* Iterate over fulls:
 368     */
 369    while (inmask) {
 370       int i = ffs(inmask) - 1;
 371       int ix = (i & 3) * 4;
 372       int iy = (i >> 2) * 4;
 373       int px = x + ix;
 374       int py = y + iy;
 375
 376       inmask &= ~(1 << i);
 377
 378       block_full_4(task, tri, px, py);
 379    }
 380 }