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