• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
3  Intel funded Tungsten Graphics to
4  develop this 3D driver.
5 
6  Permission is hereby granted, free of charge, to any person obtaining
7  a 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, sublicense, 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
16  portions of the Software.
17 
18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 
26  **********************************************************************/
27  /*
28   * Authors:
29   *   Keith Whitwell <keithw@vmware.com>
30   */
31 
32 #include "main/macros.h"
33 #include "main/enums.h"
34 #include "program/program.h"
35 
36 #include "intel_batchbuffer.h"
37 
38 #include "brw_defines.h"
39 #include "brw_context.h"
40 #include "brw_eu.h"
41 #include "brw_clip.h"
42 
43 
44 
brw_clip_line_alloc_regs(struct brw_clip_compile * c)45 static void brw_clip_line_alloc_regs( struct brw_clip_compile *c )
46 {
47    const struct gen_device_info *devinfo = c->func.devinfo;
48    GLuint i = 0,j;
49 
50    /* Register usage is static, precompute here:
51     */
52    c->reg.R0 = retype(brw_vec8_grf(i, 0), BRW_REGISTER_TYPE_UD); i++;
53 
54    if (c->key.nr_userclip) {
55       c->reg.fixed_planes = brw_vec4_grf(i, 0);
56       i += (6 + c->key.nr_userclip + 1) / 2;
57 
58       c->prog_data.curb_read_length = (6 + c->key.nr_userclip + 1) / 2;
59    }
60    else
61       c->prog_data.curb_read_length = 0;
62 
63 
64    /* Payload vertices plus space for more generated vertices:
65     */
66    for (j = 0; j < 4; j++) {
67       c->reg.vertex[j] = brw_vec4_grf(i, 0);
68       i += c->nr_regs;
69    }
70 
71    c->reg.t           = brw_vec1_grf(i, 0);
72    c->reg.t0          = brw_vec1_grf(i, 1);
73    c->reg.t1          = brw_vec1_grf(i, 2);
74    c->reg.planemask   = retype(brw_vec1_grf(i, 3), BRW_REGISTER_TYPE_UD);
75    c->reg.plane_equation = brw_vec4_grf(i, 4);
76    i++;
77 
78    c->reg.dp0         = brw_vec1_grf(i, 0); /* fixme - dp4 will clobber r.1,2,3 */
79    c->reg.dp1         = brw_vec1_grf(i, 4);
80    i++;
81 
82    if (!c->key.nr_userclip) {
83       c->reg.fixed_planes = brw_vec8_grf(i, 0);
84       i++;
85    }
86 
87    c->reg.vertex_src_mask = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD);
88    c->reg.clipdistance_offset = retype(brw_vec1_grf(i, 1), BRW_REGISTER_TYPE_W);
89    i++;
90 
91    if (devinfo->gen == 5) {
92       c->reg.ff_sync = retype(brw_vec1_grf(i, 0), BRW_REGISTER_TYPE_UD);
93       i++;
94    }
95 
96    c->first_tmp = i;
97    c->last_tmp = i;
98 
99    c->prog_data.urb_read_length = c->nr_regs; /* ? */
100    c->prog_data.total_grf = i;
101 }
102 
103 
104 /* Line clipping, more or less following the following algorithm:
105  *
106  *  for (p=0;p<MAX_PLANES;p++) {
107  *     if (clipmask & (1 << p)) {
108  *        GLfloat dp0 = DOTPROD( vtx0, plane[p] );
109  *        GLfloat dp1 = DOTPROD( vtx1, plane[p] );
110  *
111  *        if (dp1 < 0.0f) {
112  *           GLfloat t = dp1 / (dp1 - dp0);
113  *           if (t > t1) t1 = t;
114  *        } else {
115  *           GLfloat t = dp0 / (dp0 - dp1);
116  *           if (t > t0) t0 = t;
117  *        }
118  *
119  *        if (t0 + t1 >= 1.0)
120  *           return;
121  *     }
122  *  }
123  *
124  *  interp( ctx, newvtx0, vtx0, vtx1, t0 );
125  *  interp( ctx, newvtx1, vtx1, vtx0, t1 );
126  *
127  */
clip_and_emit_line(struct brw_clip_compile * c)128 static void clip_and_emit_line( struct brw_clip_compile *c )
129 {
130    struct brw_codegen *p = &c->func;
131    struct brw_indirect vtx0     = brw_indirect(0, 0);
132    struct brw_indirect vtx1      = brw_indirect(1, 0);
133    struct brw_indirect newvtx0   = brw_indirect(2, 0);
134    struct brw_indirect newvtx1   = brw_indirect(3, 0);
135    struct brw_indirect plane_ptr = brw_indirect(4, 0);
136    struct brw_reg v1_null_ud = retype(vec1(brw_null_reg()), BRW_REGISTER_TYPE_UD);
137    GLuint hpos_offset = brw_varying_to_offset(&c->vue_map, VARYING_SLOT_POS);
138    GLint clipdist0_offset = c->key.nr_userclip
139       ? brw_varying_to_offset(&c->vue_map, VARYING_SLOT_CLIP_DIST0)
140       : 0;
141 
142    brw_MOV(p, get_addr_reg(vtx0),      brw_address(c->reg.vertex[0]));
143    brw_MOV(p, get_addr_reg(vtx1),      brw_address(c->reg.vertex[1]));
144    brw_MOV(p, get_addr_reg(newvtx0),   brw_address(c->reg.vertex[2]));
145    brw_MOV(p, get_addr_reg(newvtx1),   brw_address(c->reg.vertex[3]));
146    brw_MOV(p, get_addr_reg(plane_ptr), brw_clip_plane0_address(c));
147 
148    /* Note: init t0, t1 together:
149     */
150    brw_MOV(p, vec2(c->reg.t0), brw_imm_f(0));
151 
152    brw_clip_init_planes(c);
153    brw_clip_init_clipmask(c);
154 
155    /* -ve rhw workaround */
156    if (p->devinfo->has_negative_rhw_bug) {
157       brw_AND(p, brw_null_reg(), get_element_ud(c->reg.R0, 2),
158               brw_imm_ud(1<<20));
159       brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
160       brw_OR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(0x3f));
161       brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
162    }
163 
164    /* Set the initial vertex source mask: The first 6 planes are the bounds
165     * of the view volume; the next 8 planes are the user clipping planes.
166     */
167    brw_MOV(p, c->reg.vertex_src_mask, brw_imm_ud(0x3fc0));
168 
169    /* Set the initial clipdistance offset to be 6 floats before gl_ClipDistance[0].
170     * We'll increment 6 times before we start hitting actual user clipping. */
171    brw_MOV(p, c->reg.clipdistance_offset, brw_imm_d(clipdist0_offset - 6*sizeof(float)));
172 
173    brw_DO(p, BRW_EXECUTE_1);
174    {
175       /* if (planemask & 1)
176        */
177       brw_AND(p, v1_null_ud, c->reg.planemask, brw_imm_ud(1));
178       brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
179 
180       brw_IF(p, BRW_EXECUTE_1);
181       {
182          brw_AND(p, v1_null_ud, c->reg.vertex_src_mask, brw_imm_ud(1));
183          brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
184          brw_IF(p, BRW_EXECUTE_1);
185          {
186             /* user clip distance: just fetch the correct float from each vertex */
187             struct brw_indirect temp_ptr = brw_indirect(7, 0);
188             brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx0), c->reg.clipdistance_offset);
189             brw_MOV(p, c->reg.dp0, deref_1f(temp_ptr, 0));
190             brw_ADD(p, get_addr_reg(temp_ptr), get_addr_reg(vtx1), c->reg.clipdistance_offset);
191             brw_MOV(p, c->reg.dp1, deref_1f(temp_ptr, 0));
192          }
193          brw_ELSE(p);
194          {
195             /* fixed plane: fetch the hpos, dp4 against the plane. */
196             if (c->key.nr_userclip)
197                brw_MOV(p, c->reg.plane_equation, deref_4f(plane_ptr, 0));
198             else
199                brw_MOV(p, c->reg.plane_equation, deref_4b(plane_ptr, 0));
200 
201             brw_DP4(p, vec4(c->reg.dp0), deref_4f(vtx0, hpos_offset), c->reg.plane_equation);
202             brw_DP4(p, vec4(c->reg.dp1), deref_4f(vtx1, hpos_offset), c->reg.plane_equation);
203          }
204          brw_ENDIF(p);
205 
206          brw_CMP(p, brw_null_reg(), BRW_CONDITIONAL_L, vec1(c->reg.dp1), brw_imm_f(0.0f));
207 
208          brw_IF(p, BRW_EXECUTE_1);
209          {
210              /*
211               * Both can be negative on GM965/G965 due to RHW workaround
212               * if so, this object should be rejected.
213               */
214              if (p->devinfo->has_negative_rhw_bug) {
215                  brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_LE, c->reg.dp0, brw_imm_f(0.0));
216                  brw_IF(p, BRW_EXECUTE_1);
217                  {
218                      brw_clip_kill_thread(c);
219                  }
220                  brw_ENDIF(p);
221              }
222 
223              brw_ADD(p, c->reg.t, c->reg.dp1, negate(c->reg.dp0));
224              brw_math_invert(p, c->reg.t, c->reg.t);
225              brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp1);
226 
227              brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t1 );
228              brw_MOV(p, c->reg.t1, c->reg.t);
229              brw_inst_set_pred_control(p->devinfo, brw_last_inst,
230                                        BRW_PREDICATE_NORMAL);
231 	 }
232 	 brw_ELSE(p);
233 	 {
234              /* Coming back in.  We know that both cannot be negative
235               * because the line would have been culled in that case.
236               */
237 
238              /* If both are positive, do nothing */
239              /* Only on GM965/G965 */
240              if (p->devinfo->has_negative_rhw_bug) {
241                  brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.dp0, brw_imm_f(0.0));
242                  brw_IF(p, BRW_EXECUTE_1);
243              }
244 
245              {
246                  brw_ADD(p, c->reg.t, c->reg.dp0, negate(c->reg.dp1));
247                  brw_math_invert(p, c->reg.t, c->reg.t);
248                  brw_MUL(p, c->reg.t, c->reg.t, c->reg.dp0);
249 
250                  brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_G, c->reg.t, c->reg.t0 );
251                  brw_MOV(p, c->reg.t0, c->reg.t);
252                  brw_inst_set_pred_control(p->devinfo, brw_last_inst,
253                                            BRW_PREDICATE_NORMAL);
254              }
255 
256              if (p->devinfo->has_negative_rhw_bug) {
257                  brw_ENDIF(p);
258              }
259          }
260 	 brw_ENDIF(p);
261       }
262       brw_ENDIF(p);
263 
264       /* plane_ptr++;
265        */
266       brw_ADD(p, get_addr_reg(plane_ptr), get_addr_reg(plane_ptr), brw_clip_plane_stride(c));
267 
268       /* while (planemask>>=1) != 0
269        */
270       brw_SHR(p, c->reg.planemask, c->reg.planemask, brw_imm_ud(1));
271       brw_inst_set_cond_modifier(p->devinfo, brw_last_inst, BRW_CONDITIONAL_NZ);
272       brw_SHR(p, c->reg.vertex_src_mask, c->reg.vertex_src_mask, brw_imm_ud(1));
273       brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
274       brw_ADD(p, c->reg.clipdistance_offset, c->reg.clipdistance_offset, brw_imm_w(sizeof(float)));
275       brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
276    }
277    brw_WHILE(p);
278    brw_inst_set_pred_control(p->devinfo, brw_last_inst, BRW_PREDICATE_NORMAL);
279 
280    brw_ADD(p, c->reg.t, c->reg.t0, c->reg.t1);
281    brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.t, brw_imm_f(1.0));
282    brw_IF(p, BRW_EXECUTE_1);
283    {
284       brw_clip_interp_vertex(c, newvtx0, vtx0, vtx1, c->reg.t0, false);
285       brw_clip_interp_vertex(c, newvtx1, vtx1, vtx0, c->reg.t1, false);
286 
287       brw_clip_emit_vue(c, newvtx0, BRW_URB_WRITE_ALLOCATE_COMPLETE,
288                         (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
289                         | URB_WRITE_PRIM_START);
290       brw_clip_emit_vue(c, newvtx1, BRW_URB_WRITE_EOT_COMPLETE,
291                         (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
292                         | URB_WRITE_PRIM_END);
293    }
294    brw_ENDIF(p);
295    brw_clip_kill_thread(c);
296 }
297 
298 
299 
brw_emit_line_clip(struct brw_clip_compile * c)300 void brw_emit_line_clip( struct brw_clip_compile *c )
301 {
302    brw_clip_line_alloc_regs(c);
303    brw_clip_init_ff_sync(c);
304 
305    if (c->key.contains_flat_varying) {
306       if (c->key.pv_first)
307          brw_clip_copy_flatshaded_attributes(c, 1, 0);
308       else
309          brw_clip_copy_flatshaded_attributes(c, 0, 1);
310    }
311 
312    clip_and_emit_line(c);
313 }
314