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
  72 static INLINE unsigned
  73 build_mask_linear(int c, int dcdx, int dcdy)
  74 {
  75    int mask = 0;
  76
  77    int c0 = c;
  78    int c1 = c0 + dcdy;
  79    int c2 = c1 + dcdy;
  80    int c3 = c2 + dcdy;
  81
  82    mask |= ((c0 + 0 * dcdx) >> 31) & (1 << 0);
  83    mask |= ((c0 + 1 * dcdx) >> 31) & (1 << 1);
  84    mask |= ((c0 + 2 * dcdx) >> 31) & (1 << 2);
  85    mask |= ((c0 + 3 * dcdx) >> 31) & (1 << 3);
  86    mask |= ((c1 + 0 * dcdx) >> 31) & (1 << 4);
  87    mask |= ((c1 + 1 * dcdx) >> 31) & (1 << 5);
  88    mask |= ((c1 + 2 * dcdx) >> 31) & (1 << 6);
  89    mask |= ((c1 + 3 * dcdx) >> 31) & (1 << 7);
  90    mask |= ((c2 + 0 * dcdx) >> 31) & (1 << 8);
  91    mask |= ((c2 + 1 * dcdx) >> 31) & (1 << 9);
  92    mask |= ((c2 + 2 * dcdx) >> 31) & (1 << 10);
  93    mask |= ((c2 + 3 * dcdx) >> 31) & (1 << 11);
  94    mask |= ((c3 + 0 * dcdx) >> 31) & (1 << 12);
  95    mask |= ((c3 + 1 * dcdx) >> 31) & (1 << 13);
  96    mask |= ((c3 + 2 * dcdx) >> 31) & (1 << 14);
  97    mask |= ((c3 + 3 * dcdx) >> 31) & (1 << 15);
  98
  99    return mask;
 100 }
 101
 102
 103 static INLINE void
 104 build_masks(int c,
 105             int cdiff,
 106             int dcdx,
 107             int dcdy,
 108             unsigned *outmask,
 109             unsigned *partmask)
 110 {
 111    *outmask |= build_mask_linear(c, dcdx, dcdy);
 112    *partmask |= build_mask_linear(c + cdiff, dcdx, dcdy);
 113 }
 114
 115 void
 116 lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 117                       const union lp_rast_cmd_arg arg)
 118 {
 119    union lp_rast_cmd_arg arg2;
 120    arg2.triangle.tri = arg.triangle.tri;
 121    arg2.triangle.plane_mask = (1<<3)-1;
 122    lp_rast_triangle_3(task, arg2);
 123 }
 124
 125 void
 126 lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 127                       const union lp_rast_cmd_arg arg)
 128 {
 129    lp_rast_triangle_3_16(task, arg);
 130 }
 131
 132 #else
 133 #include <emmintrin.h>
 134 #include "util/u_sse.h"
 135
 136
 137 static INLINE void
 138 build_masks(int c,
 139             int cdiff,
 140             int dcdx,
 141             int dcdy,
 142             unsigned *outmask,
 143             unsigned *partmask)
 144 {
 145    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
 146    __m128i xdcdy = _mm_set1_epi32(dcdy);
 147
 148    /* Get values across the quad
 149     */
 150    __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 151    __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 152    __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 153
 154    {
 155       __m128i cstep01, cstep23, result;
 156
 157       cstep01 = _mm_packs_epi32(cstep0, cstep1);
 158       cstep23 = _mm_packs_epi32(cstep2, cstep3);
 159       result = _mm_packs_epi16(cstep01, cstep23);
 160
 161       *outmask |= _mm_movemask_epi8(result);
 162    }
 163
 164
 165    {
 166       __m128i cio4 = _mm_set1_epi32(cdiff);
 167       __m128i cstep01, cstep23, result;
 168
 169       cstep0 = _mm_add_epi32(cstep0, cio4);
 170       cstep1 = _mm_add_epi32(cstep1, cio4);
 171       cstep2 = _mm_add_epi32(cstep2, cio4);
 172       cstep3 = _mm_add_epi32(cstep3, cio4);
 173
 174       cstep01 = _mm_packs_epi32(cstep0, cstep1);
 175       cstep23 = _mm_packs_epi32(cstep2, cstep3);
 176       result = _mm_packs_epi16(cstep01, cstep23);
 177
 178       *partmask |= _mm_movemask_epi8(result);
 179    }
 180 }
 181
 182
 183 static INLINE unsigned
 184 build_mask_linear(int c, int dcdx, int dcdy)
 185 {
 186    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
 187    __m128i xdcdy = _mm_set1_epi32(dcdy);
 188
 189    /* Get values across the quad
 190     */
 191    __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 192    __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 193    __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 194
 195    /* pack pairs of results into epi16
 196     */
 197    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 198    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 199
 200    /* pack into epi8, preserving sign bits
 201     */
 202    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 203
 204    /* extract sign bits to create mask
 205     */
 206    return _mm_movemask_epi8(result);
 207 }
 208
 209 static INLINE unsigned
 210 sign_bits4(const __m128i *cstep, int cdiff)
 211 {
 212
 213    /* Adjust the step values
 214     */
 215    __m128i cio4 = _mm_set1_epi32(cdiff);
 216    __m128i cstep0 = _mm_add_epi32(cstep[0], cio4);
 217    __m128i cstep1 = _mm_add_epi32(cstep[1], cio4);
 218    __m128i cstep2 = _mm_add_epi32(cstep[2], cio4);
 219    __m128i cstep3 = _mm_add_epi32(cstep[3], cio4);
 220
 221    /* Pack down to epi8
 222     */
 223    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 224    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 225    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 226
 227    /* Extract the sign bits
 228     */
 229    return _mm_movemask_epi8(result);
 230 }
 231
 232
 233 /* Special case for 3 plane triangle which is contained entirely
 234  * within a 16x16 block.
 235  */
 236 void
 237 lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 238                       const union lp_rast_cmd_arg arg)
 239 {
 240    const struct lp_rast_triangle *tri = arg.triangle.tri;
 241    const struct lp_rast_plane *plane = tri->plane;
 242    unsigned mask = arg.triangle.plane_mask;
 243    const int x = task->x + (mask & 0xff);
 244    const int y = task->y + (mask >> 8);
 245    unsigned outmask, inmask, partmask, partial_mask;
 246    unsigned j;
 247    __m128i cstep4[3][4];
 248
 249    outmask = 0;                 /* outside one or more trivial reject planes */
 250    partmask = 0;                /* outside one or more trivial accept planes */
 251
 252    for (j = 0; j < 3; j++) {
 253       const int dcdx = -plane[j].dcdx * 4;
 254       const int dcdy = plane[j].dcdy * 4;
 255       __m128i xdcdy = _mm_set1_epi32(dcdy);
 256
 257       cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
 258       cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
 259       cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
 260       cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
 261
 262       {
 263          const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
 264          const int cox = plane[j].eo * 4;
 265          const int cio = plane[j].ei * 4 - 1;
 266
 267          outmask |= sign_bits4(cstep4[j], c + cox);
 268          partmask |= sign_bits4(cstep4[j], c + cio);
 269       }
 270    }
 271
 272    if (outmask == 0xffff)
 273       return;
 274
 275    /* Mask of sub-blocks which are inside all trivial accept planes:
 276     */
 277    inmask = ~partmask & 0xffff;
 278
 279    /* Mask of sub-blocks which are inside all trivial reject planes,
 280     * but outside at least one trivial accept plane:
 281     */
 282    partial_mask = partmask & ~outmask;
 283
 284    assert((partial_mask & inmask) == 0);
 285
 286    /* Iterate over partials:
 287     */
 288    while (partial_mask) {
 289       int i = ffs(partial_mask) - 1;
 290       int ix = (i & 3) * 4;
 291       int iy = (i >> 2) * 4;
 292       int px = x + ix;
 293       int py = y + iy;
 294       unsigned mask = 0xffff;
 295
 296       partial_mask &= ~(1 << i);
 297
 298       for (j = 0; j < 3; j++) {
 299          const int cx = (plane[j].c
 300                          - plane[j].dcdx * px
 301                          + plane[j].dcdy * py) * 4;
 302
 303          mask &= ~sign_bits4(cstep4[j], cx);
 304       }
 305
 306       if (mask)
 307          lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
 308    }
 309
 310    /* Iterate over fulls:
 311     */
 312    while (inmask) {
 313       int i = ffs(inmask) - 1;
 314       int ix = (i & 3) * 4;
 315       int iy = (i >> 2) * 4;
 316       int px = x + ix;
 317       int py = y + iy;
 318
 319       inmask &= ~(1 << i);
 320
 321       block_full_4(task, tri, px, py);
 322    }
 323 }
 324
 325
 326 void
 327 lp_rast_triangle_3_4(struct lp_rasterizer_task *task,
 328                      const union lp_rast_cmd_arg arg)
 329 {
 330    const struct lp_rast_triangle *tri = arg.triangle.tri;
 331    const struct lp_rast_plane *plane = tri->plane;
 332    unsigned mask = arg.triangle.plane_mask;
 333    const int x = task->x + (mask & 0xff);
 334    const int y = task->y + (mask >> 8);
 335    unsigned j;
 336
 337    /* Iterate over partials:
 338     */
 339    {
 340       unsigned mask = 0xffff;
 341
 342       for (j = 0; j < 3; j++) {
 343          const int cx = (plane[j].c
 344                          - plane[j].dcdx * x
 345                          + plane[j].dcdy * y);
 346
 347          const int dcdx = -plane[j].dcdx;
 348          const int dcdy = plane[j].dcdy;
 349          __m128i xdcdy = _mm_set1_epi32(dcdy);
 350
 351          __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
 352          __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 353          __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 354          __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 355
 356          __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 357          __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 358          __m128i result = _mm_packs_epi16(cstep01, cstep23);
 359
 360          /* Extract the sign bits
 361           */
 362          mask &= ~_mm_movemask_epi8(result);
 363       }
 364
 365       if (mask)
 366          lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
 367    }
 368 }
 369
 370
 371 #endif
 372
 373
 374
 375
 376 #define TAG(x) x##_1
 377 #define NR_PLANES 1
 378 #include "lp_rast_tri_tmp.h"
 379
 380 #define TAG(x) x##_2
 381 #define NR_PLANES 2
 382 #include "lp_rast_tri_tmp.h"
 383
 384 #define TAG(x) x##_3
 385 #define NR_PLANES 3
 386 #include "lp_rast_tri_tmp.h"
 387
 388 #define TAG(x) x##_4
 389 #define NR_PLANES 4
 390 #include "lp_rast_tri_tmp.h"
 391
 392 #define TAG(x) x##_5
 393 #define NR_PLANES 5
 394 #include "lp_rast_tri_tmp.h"
 395
 396 #define TAG(x) x##_6
 397 #define NR_PLANES 6
 398 #include "lp_rast_tri_tmp.h"
 399
 400 #define TAG(x) x##_7
 401 #define NR_PLANES 7
 402 #include "lp_rast_tri_tmp.h"
 403
 404 #define TAG(x) x##_8
 405 #define NR_PLANES 8
 406 #include "lp_rast_tri_tmp.h"
 407