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 */
42static void
43TAG(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 */
68static void
69TAG(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 ei = plane[j].dcdy - plane[j].dcdx - plane[j].eo;
86      const int cio = ei * 4 - 1;
87
88      build_masks(c[j] + cox,
89		  cio - cox,
90		  dcdx, dcdy,
91		  &outmask,   /* sign bits from c[i][0..15] + cox */
92		  &partmask); /* sign bits from c[i][0..15] + cio */
93   }
94
95   if (outmask == 0xffff)
96      return;
97
98   /* Mask of sub-blocks which are inside all trivial accept planes:
99    */
100   inmask = ~partmask & 0xffff;
101
102   /* Mask of sub-blocks which are inside all trivial reject planes,
103    * but outside at least one trivial accept plane:
104    */
105   partial_mask = partmask & ~outmask;
106
107   assert((partial_mask & inmask) == 0);
108
109   LP_COUNT_ADD(nr_empty_4, util_bitcount(0xffff & ~(partial_mask | inmask)));
110
111   /* Iterate over partials:
112    */
113   while (partial_mask) {
114      int i = ffs(partial_mask) - 1;
115      int ix = (i & 3) * 4;
116      int iy = (i >> 2) * 4;
117      int px = x + ix;
118      int py = y + iy;
119      int cx[NR_PLANES];
120
121      partial_mask &= ~(1 << i);
122
123      LP_COUNT(nr_partially_covered_4);
124
125      for (j = 0; j < NR_PLANES; j++)
126         cx[j] = (c[j]
127		  - plane[j].dcdx * ix
128		  + plane[j].dcdy * iy);
129
130      TAG(do_block_4)(task, tri, plane, px, py, cx);
131   }
132
133   /* Iterate over fulls:
134    */
135   while (inmask) {
136      int i = ffs(inmask) - 1;
137      int ix = (i & 3) * 4;
138      int iy = (i >> 2) * 4;
139      int px = x + ix;
140      int py = y + iy;
141
142      inmask &= ~(1 << i);
143
144      LP_COUNT(nr_fully_covered_4);
145      block_full_4(task, tri, px, py);
146   }
147}
148
149
150/**
151 * Scan the tile in chunks and figure out which pixels to rasterize
152 * for this triangle.
153 */
154void
155TAG(lp_rast_triangle)(struct lp_rasterizer_task *task,
156                      const union lp_rast_cmd_arg arg)
157{
158   const struct lp_rast_triangle *tri = arg.triangle.tri;
159   unsigned plane_mask = arg.triangle.plane_mask;
160   const struct lp_rast_plane *tri_plane = GET_PLANES(tri);
161   const int x = task->x, y = task->y;
162   struct lp_rast_plane plane[NR_PLANES];
163   int c[NR_PLANES];
164   unsigned outmask, inmask, partmask, partial_mask;
165   unsigned j = 0;
166
167   if (tri->inputs.disable) {
168      /* This triangle was partially binned and has been disabled */
169      return;
170   }
171
172   outmask = 0;                 /* outside one or more trivial reject planes */
173   partmask = 0;                /* outside one or more trivial accept planes */
174
175   while (plane_mask) {
176      int i = ffs(plane_mask) - 1;
177      plane[j] = tri_plane[i];
178      plane_mask &= ~(1 << i);
179      c[j] = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
180
181      {
182	 const int dcdx = -plane[j].dcdx * 16;
183	 const int dcdy = plane[j].dcdy * 16;
184	 const int cox = plane[j].eo * 16;
185         const int ei = plane[j].dcdy - plane[j].dcdx - plane[j].eo;
186         const int cio = ei * 16 - 1;
187
188	 build_masks(c[j] + cox,
189		     cio - cox,
190		     dcdx, dcdy,
191		     &outmask,   /* sign bits from c[i][0..15] + cox */
192		     &partmask); /* sign bits from c[i][0..15] + cio */
193      }
194
195      j++;
196   }
197
198   if (outmask == 0xffff)
199      return;
200
201   /* Mask of sub-blocks which are inside all trivial accept planes:
202    */
203   inmask = ~partmask & 0xffff;
204
205   /* Mask of sub-blocks which are inside all trivial reject planes,
206    * but outside at least one trivial accept plane:
207    */
208   partial_mask = partmask & ~outmask;
209
210   assert((partial_mask & inmask) == 0);
211
212   LP_COUNT_ADD(nr_empty_16, util_bitcount(0xffff & ~(partial_mask | inmask)));
213
214   /* Iterate over partials:
215    */
216   while (partial_mask) {
217      int i = ffs(partial_mask) - 1;
218      int ix = (i & 3) * 16;
219      int iy = (i >> 2) * 16;
220      int px = x + ix;
221      int py = y + iy;
222      int cx[NR_PLANES];
223
224      for (j = 0; j < NR_PLANES; j++)
225         cx[j] = (c[j]
226		  - plane[j].dcdx * ix
227		  + plane[j].dcdy * iy);
228
229      partial_mask &= ~(1 << i);
230
231      LP_COUNT(nr_partially_covered_16);
232      TAG(do_block_16)(task, tri, plane, px, py, cx);
233   }
234
235   /* Iterate over fulls:
236    */
237   while (inmask) {
238      int i = ffs(inmask) - 1;
239      int ix = (i & 3) * 16;
240      int iy = (i >> 2) * 16;
241      int px = x + ix;
242      int py = y + iy;
243
244      inmask &= ~(1 << i);
245
246      LP_COUNT(nr_fully_covered_16);
247      block_full_16(task, tri, px, py);
248   }
249}
250
251#if defined(PIPE_ARCH_SSE) && defined(TRI_16)
252/* XXX: special case this when intersection is not required.
253 *      - tile completely within bbox,
254 *      - bbox completely within tile.
255 */
256void
257TRI_16(struct lp_rasterizer_task *task,
258       const union lp_rast_cmd_arg arg)
259{
260   const struct lp_rast_triangle *tri = arg.triangle.tri;
261   const struct lp_rast_plane *plane = GET_PLANES(tri);
262   unsigned mask = arg.triangle.plane_mask;
263   unsigned outmask, partial_mask;
264   unsigned j;
265   __m128i cstep4[NR_PLANES][4];
266
267   int x = (mask & 0xff);
268   int y = (mask >> 8);
269
270   outmask = 0;                 /* outside one or more trivial reject planes */
271
272   x += task->x;
273   y += task->y;
274
275   for (j = 0; j < NR_PLANES; j++) {
276      const int dcdx = -plane[j].dcdx * 4;
277      const int dcdy = plane[j].dcdy * 4;
278      __m128i xdcdy = _mm_set1_epi32(dcdy);
279
280      cstep4[j][0] = _mm_setr_epi32(0, dcdx, dcdx*2, dcdx*3);
281      cstep4[j][1] = _mm_add_epi32(cstep4[j][0], xdcdy);
282      cstep4[j][2] = _mm_add_epi32(cstep4[j][1], xdcdy);
283      cstep4[j][3] = _mm_add_epi32(cstep4[j][2], xdcdy);
284
285      {
286	 const int c = plane[j].c + plane[j].dcdy * y - plane[j].dcdx * x;
287	 const int cox = plane[j].eo * 4;
288
289	 outmask |= sign_bits4(cstep4[j], c + cox);
290      }
291   }
292
293   if (outmask == 0xffff)
294      return;
295
296
297   /* Mask of sub-blocks which are inside all trivial reject planes,
298    * but outside at least one trivial accept plane:
299    */
300   partial_mask = 0xffff & ~outmask;
301
302   /* Iterate over partials:
303    */
304   while (partial_mask) {
305      int i = ffs(partial_mask) - 1;
306      int ix = (i & 3) * 4;
307      int iy = (i >> 2) * 4;
308      int px = x + ix;
309      int py = y + iy;
310      unsigned mask = 0xffff;
311
312      partial_mask &= ~(1 << i);
313
314      for (j = 0; j < NR_PLANES; j++) {
315         const int cx = (plane[j].c - 1
316			 - plane[j].dcdx * px
317			 + plane[j].dcdy * py) * 4;
318
319	 mask &= ~sign_bits4(cstep4[j], cx);
320      }
321
322      if (mask)
323	 lp_rast_shade_quads_mask(task, &tri->inputs, px, py, mask);
324   }
325}
326#endif
327
328#if defined(PIPE_ARCH_SSE) && defined(TRI_4)
329void
330TRI_4(struct lp_rasterizer_task *task,
331      const union lp_rast_cmd_arg arg)
332{
333   const struct lp_rast_triangle *tri = arg.triangle.tri;
334   const struct lp_rast_plane *plane = GET_PLANES(tri);
335   unsigned mask = arg.triangle.plane_mask;
336   const int x = task->x + (mask & 0xff);
337   const int y = task->y + (mask >> 8);
338   unsigned j;
339
340   /* Iterate over partials:
341    */
342   {
343      unsigned mask = 0xffff;
344
345      for (j = 0; j < NR_PLANES; j++) {
346	 const int cx = (plane[j].c
347			 - plane[j].dcdx * x
348			 + plane[j].dcdy * y);
349
350	 const int dcdx = -plane[j].dcdx;
351	 const int dcdy = plane[j].dcdy;
352	 __m128i xdcdy = _mm_set1_epi32(dcdy);
353
354	 __m128i cstep0 = _mm_setr_epi32(cx, cx + dcdx, cx + dcdx*2, cx + dcdx*3);
355	 __m128i cstep1 = _mm_add_epi32(cstep0, xdcdy);
356	 __m128i cstep2 = _mm_add_epi32(cstep1, xdcdy);
357	 __m128i cstep3 = _mm_add_epi32(cstep2, xdcdy);
358
359	 __m128i cstep01 = _mm_packs_epi32(cstep0, cstep1);
360	 __m128i cstep23 = _mm_packs_epi32(cstep2, cstep3);
361	 __m128i result = _mm_packs_epi16(cstep01, cstep23);
362
363	 /* Extract the sign bits
364	  */
365	 mask &= ~_mm_movemask_epi8(result);
366      }
367
368      if (mask)
369	 lp_rast_shade_quads_mask(task, &tri->inputs, x, y, mask);
370   }
371}
372#endif
373
374
375
376#undef TAG
377#undef TRI_4
378#undef TRI_16
379#undef NR_PLANES
380
381