src/mesa/drivers/dri/i965/brw_clip_line.c

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