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 #else
 132 #include <emmintrin.h>
 133 #include "util/u_sse.h"
 134
 135
 136 static INLINE unsigned
 137 build_mask_linear(int c, int dcdx, int dcdy)
 138 {
 139    __m128i cstep0 = _mm_setr_epi32(c, c+dcdx, c+dcdx*2, c+dcdx*3);
 140    __m128i xdcdy = _mm_set1_epi32(dcdy);
 141
 142    /* Get values across the quad
 143     */
 144    __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
 145    __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
 146    __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
 147
 148    /* pack pairs of results into epi16
 149     */
 150    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 151    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 152
 153    /* pack into epi8, preserving sign bits
 154     */
 155    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 156
 157    /* extract sign bits to create mask
 158     */
 159    return _mm_movemask_epi8(result);
 160 }
 161
 162 static INLINE unsigned
 163 build_mask(int c, int dcdx, int dcdy)
 164 {
 165    __m128i step = _mm_setr_epi32(0, dcdx, dcdy, dcdx + dcdy);
 166    __m128i c0 = _mm_set1_epi32(c);
 167
 168    /* Get values across the quad
 169     */
 170    __m128i cstep0 = _mm_add_epi32(c0, step);
 171
 172    /* Scale up step for moving between quads.  This should probably
 173     * be an arithmetic shift left, but there doesn't seem to be
 174     * such a thing in SSE.  It's unlikely that the step value is
 175     * going to be large enough to overflow across 4 pixels, though
 176     * if it is that big, rendering will be incorrect anyway.
 177     */
 178    __m128i step4 = _mm_slli_epi32(step, 1);
 179
 180    /* Get values for the remaining quads:
 181     */
 182    __m128i cstep1 = _mm_add_epi32(cstep0,
 183                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1)));
 184    __m128i cstep2 = _mm_add_epi32(cstep0,
 185                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(2,2,2,2)));
 186    __m128i cstep3 = _mm_add_epi32(cstep2,
 187                                   _mm_shuffle_epi32(step4, _MM_SHUFFLE(1,1,1,1)));
 188
 189    /* pack pairs of results into epi16
 190     */
 191    __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
 192    __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
 193
 194    /* pack into epi8, preserving sign bits
 195     */
 196    __m128i result = _mm_packs_epi16(cstep01, cstep23);
 197
 198    /* extract sign bits to create mask
 199     */
 200    return _mm_movemask_epi8(result);
 201 }
 202
 203 #endif
 204
 205
 206
 207
 208 #define TAG(x) x##_1
 209 #define NR_PLANES 1
 210 #include "lp_rast_tri_tmp.h"
 211
 212 #define TAG(x) x##_2
 213 #define NR_PLANES 2
 214 #include "lp_rast_tri_tmp.h"
 215
 216 #define TAG(x) x##_3
 217 #define NR_PLANES 3
 218 #include "lp_rast_tri_tmp.h"
 219
 220 #define TAG(x) x##_4
 221 #define NR_PLANES 4
 222 #include "lp_rast_tri_tmp.h"
 223
 224 #define TAG(x) x##_5
 225 #define NR_PLANES 5
 226 #include "lp_rast_tri_tmp.h"
 227
 228 #define TAG(x) x##_6
 229 #define NR_PLANES 6
 230 #include "lp_rast_tri_tmp.h"
 231
 232 #define TAG(x) x##_7
 233 #define NR_PLANES 7
 234 #include "lp_rast_tri_tmp.h"
 235
 236 #define TAG(x) x##_8
 237 #define NR_PLANES 8
 238 #include "lp_rast_tri_tmp.h"
 239
 240
 241 /* Special case for 3 plane triangle which is contained entirely
 242  * within a 16x16 block.
 243  */
 244 void
 245 lp_rast_triangle_3_16(struct lp_rasterizer_task *task,
 246                       const union lp_rast_cmd_arg arg)
 247 {
 248    const struct lp_rast_triangle *tri = arg.triangle.tri;
 249    const struct lp_rast_plane *plane = tri->plane;
 250    unsigned mask = arg.triangle.plane_mask;
 251    const int x = task->x + (mask & 0xf) * 16;
 252    const int y = task->y + (mask >> 4) * 16;
 253    unsigned outmask, inmask, partmask, partial_mask;
 254    unsigned j;
 255    int c[3];
 256
 257    outmask = 0;                 /* outside one or more trivial reject planes */
 258    partmask = 0;                /* outside one or more trivial accept planes */
 259
 260    for (j = 0; j < 3; j++) {
 261       c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
 262
 263       {
 264          const int dcdx = -plane[j].dcdx * 4;
 265          const int dcdy = plane[j].dcdy * 4;
 266          const int cox = c[j] + plane[j].eo * 4;
 267          const int cio = c[j] + plane[j].ei * 4 - 1;
 268
 269          outmask |= build_mask_linear(cox, dcdx, dcdy);
 270          partmask |= build_mask_linear(cio, dcdx, dcdy);
 271       }
 272    }
 273
 274    if (outmask == 0xffff)
 275       return;
 276
 277    /* Mask of sub-blocks which are inside all trivial accept planes:
 278     */
 279    inmask = ~partmask & 0xffff;
 280
 281    /* Mask of sub-blocks which are inside all trivial reject planes,
 282     * but outside at least one trivial accept plane:
 283     */
 284    partial_mask = partmask & ~outmask;
 285
 286    assert((partial_mask & inmask) == 0);
 287
 288    /* Iterate over partials:
 289     */
 290    while (partial_mask) {
 291       int i = ffs(partial_mask) - 1;
 292       int ix = (i & 3) * 4;
 293       int iy = (i >> 2) * 4;
 294       int px = x + ix;
 295       int py = y + iy;
 296       int cx[3];
 297
 298       partial_mask &= ~(1 << i);
 299
 300       for (j = 0; j < 3; j++)
 301          cx[j] = (c[j]
 302                   - plane[j].dcdx * ix
 303                   + plane[j].dcdy * iy);
 304
 305       do_block_4_3(task, tri, plane, px, py, cx);
 306    }
 307
 308    /* Iterate over fulls:
 309     */
 310    while (inmask) {
 311       int i = ffs(inmask) - 1;
 312       int ix = (i & 3) * 4;
 313       int iy = (i >> 2) * 4;
 314       int px = x + ix;
 315       int py = y + iy;
 316
 317       inmask &= ~(1 << i);
 318
 319       block_full_4(task, tri, px, py);
 320    }
 321 }