src/gallium/drivers/llvmpipe/lp_setup_line.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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  * Binning code for lines
  30  */
  31
  32 #include "util/u_math.h"
  33 #include "util/u_memory.h"
  34 #include "lp_perf.h"
  35 #include "lp_setup_context.h"
  36 #include "lp_rast.h"
  37 #include "lp_state_fs.h"
  38
  39 #define NUM_CHANNELS 4
  40
  41 struct lp_line_info {
  42
  43    float dx;
  44    float dy;
  45    float oneoverarea;
  46
  47    const float (*v1)[4];
  48    const float (*v2)[4];
  49 };
  50
  51
  52 /**
  53  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  54  */
  55 static void constant_coef( struct lp_setup_context *setup,
  56                            struct lp_rast_triangle *tri,
  57                            unsigned slot,
  58                            const float value,
  59                            unsigned i )
  60 {
  61    tri->inputs.a0[slot][i] = value;
  62    tri->inputs.dadx[slot][i] = 0.0f;
  63    tri->inputs.dady[slot][i] = 0.0f;
  64 }
  65
  66
  67 /**
  68  * Compute a0, dadx and dady for a linearly interpolated coefficient,
  69  * for a triangle.
  70  */
  71 static void linear_coef( struct lp_setup_context *setup,
  72                          struct lp_rast_triangle *tri,
  73                          struct lp_line_info *info,
  74                          unsigned slot,
  75                          unsigned vert_attr,
  76                          unsigned i)
  77 {
  78    float a1 = info->v1[vert_attr][i];
  79    float a2 = info->v2[vert_attr][i];
  80
  81    float da21 = a1 - a2;
  82    float dadx = da21 * info->dx * info->oneoverarea;
  83    float dady = da21 * info->dy * info->oneoverarea;
  84
  85    tri->inputs.dadx[slot][i] = dadx;
  86    tri->inputs.dady[slot][i] = dady;
  87
  88    tri->inputs.a0[slot][i] = (a1 -
  89                               (dadx * (info->v1[0][0] - setup->pixel_offset) +
  90                                dady * (info->v1[0][1] - setup->pixel_offset)));
  91 }
  92
  93
  94 /**
  95  * Compute a0, dadx and dady for a perspective-corrected interpolant,
  96  * for a triangle.
  97  * We basically multiply the vertex value by 1/w before computing
  98  * the plane coefficients (a0, dadx, dady).
  99  * Later, when we compute the value at a particular fragment position we'll
 100  * divide the interpolated value by the interpolated W at that fragment.
 101  */
 102 static void perspective_coef( struct lp_setup_context *setup,
 103                               struct lp_rast_triangle *tri,
 104                               struct lp_line_info *info,
 105                               unsigned slot,
 106                               unsigned vert_attr,
 107                               unsigned i)
 108 {
 109    /* premultiply by 1/w  (v[0][3] is always 1/w):
 110     */
 111    float a1 = info->v1[vert_attr][i] * info->v1[0][3];
 112    float a2 = info->v2[vert_attr][i] * info->v2[0][3];
 113
 114    float da21 = a1 - a2;
 115    float dadx = da21 * info->dx * info->oneoverarea;
 116    float dady = da21 * info->dy * info->oneoverarea;
 117
 118    tri->inputs.dadx[slot][i] = dadx;
 119    tri->inputs.dady[slot][i] = dady;
 120
 121    tri->inputs.a0[slot][i] = (a1 -
 122                               (dadx * (info->v1[0][0] - setup->pixel_offset) +
 123                                dady * (info->v1[0][1] - setup->pixel_offset)));
 124 }
 125
 126 static void
 127 setup_fragcoord_coef( struct lp_setup_context *setup,
 128                       struct lp_rast_triangle *tri,
 129                       struct lp_line_info *info,
 130                       unsigned slot,
 131                       unsigned usage_mask)
 132 {
 133    /*X*/
 134    if (usage_mask & TGSI_WRITEMASK_X) {
 135       tri->inputs.a0[slot][0] = 0.0;
 136       tri->inputs.dadx[slot][0] = 1.0;
 137       tri->inputs.dady[slot][0] = 0.0;
 138    }
 139
 140    /*Y*/
 141    if (usage_mask & TGSI_WRITEMASK_Y) {
 142       tri->inputs.a0[slot][1] = 0.0;
 143       tri->inputs.dadx[slot][1] = 0.0;
 144       tri->inputs.dady[slot][1] = 1.0;
 145    }
 146
 147    /*Z*/
 148    if (usage_mask & TGSI_WRITEMASK_Z) {
 149       linear_coef(setup, tri, info, slot, 0, 2);
 150    }
 151
 152    /*W*/
 153    if (usage_mask & TGSI_WRITEMASK_W) {
 154       linear_coef(setup, tri, info, slot, 0, 3);
 155    }
 156 }
 157
 158 /**
 159  * Compute the tri->coef[] array dadx, dady, a0 values.
 160  */
 161 static void setup_line_coefficients( struct lp_setup_context *setup,
 162                                      struct lp_rast_triangle *tri,
 163                                      struct lp_line_info *info)
 164 {
 165    unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
 166    unsigned slot;
 167
 168    /* setup interpolation for all the remaining attributes:
 169     */
 170    for (slot = 0; slot < setup->fs.nr_inputs; slot++) {
 171       unsigned vert_attr = setup->fs.input[slot].src_index;
 172       unsigned usage_mask = setup->fs.input[slot].usage_mask;
 173       unsigned i;
 174
 175       switch (setup->fs.input[slot].interp) {
 176       case LP_INTERP_CONSTANT:
 177          if (setup->flatshade_first) {
 178             for (i = 0; i < NUM_CHANNELS; i++)
 179                if (usage_mask & (1 << i))
 180                   constant_coef(setup, tri, slot+1, info->v1[vert_attr][i], i);
 181          }
 182          else {
 183             for (i = 0; i < NUM_CHANNELS; i++)
 184                if (usage_mask & (1 << i))
 185                   constant_coef(setup, tri, slot+1, info->v2[vert_attr][i], i);
 186          }
 187          break;
 188
 189       case LP_INTERP_LINEAR:
 190          for (i = 0; i < NUM_CHANNELS; i++)
 191             if (usage_mask & (1 << i))
 192                linear_coef(setup, tri, info, slot+1, vert_attr, i);
 193          break;
 194
 195       case LP_INTERP_PERSPECTIVE:
 196          for (i = 0; i < NUM_CHANNELS; i++)
 197             if (usage_mask & (1 << i))
 198                perspective_coef(setup, tri, info, slot+1, vert_attr, i);
 199          fragcoord_usage_mask |= TGSI_WRITEMASK_W;
 200          break;
 201
 202       case LP_INTERP_POSITION:
 203          /*
 204           * The generated pixel interpolators will pick up the coeffs from
 205           * slot 0, so all need to ensure that the usage mask is covers all
 206           * usages.
 207           */
 208          fragcoord_usage_mask |= usage_mask;
 209          break;
 210
 211       case LP_INTERP_FACING:
 212          for (i = 0; i < NUM_CHANNELS; i++)
 213             if (usage_mask & (1 << i))
 214                constant_coef(setup, tri, slot+1, 1.0, i);
 215          break;
 216
 217       default:
 218          assert(0);
 219       }
 220    }
 221
 222    /* The internal position input is in slot zero:
 223     */
 224    setup_fragcoord_coef(setup, tri, info, 0,
 225                         fragcoord_usage_mask);
 226 }
 227
 228
 229
 230 static INLINE int subpixel_snap( float a )
 231 {
 232    return util_iround(FIXED_ONE * a);
 233 }
 234
 235
 236 /**
 237  * Print line vertex attribs (for debug).
 238  */
 239 static void
 240 print_line(struct lp_setup_context *setup,
 241            const float (*v1)[4],
 242            const float (*v2)[4])
 243 {
 244    uint i;
 245
 246    debug_printf("llvmpipe line\n");
 247    for (i = 0; i < 1 + setup->fs.nr_inputs; i++) {
 248       debug_printf("  v1[%d]:  %f %f %f %f\n", i,
 249                    v1[i][0], v1[i][1], v1[i][2], v1[i][3]);
 250    }
 251    for (i = 0; i < 1 + setup->fs.nr_inputs; i++) {
 252       debug_printf("  v2[%d]:  %f %f %f %f\n", i,
 253                    v2[i][0], v2[i][1], v2[i][2], v2[i][3]);
 254    }
 255 }
 256
 257
 258 static INLINE boolean sign(float x){
 259    return x >= 0;
 260 }
 261
 262
 263 /* Used on positive floats only:
 264  */
 265 static INLINE float fracf(float f)
 266 {
 267    return f - floorf(f);
 268 }
 269
 270
 271
 272 static boolean
 273 try_setup_line( struct lp_setup_context *setup,
 274                const float (*v1)[4],
 275                const float (*v2)[4])
 276 {
 277    struct lp_scene *scene = setup->scene;
 278    struct lp_rast_triangle *line;
 279    struct lp_line_info info;
 280    float width = MAX2(1.0, setup->line_width);
 281    struct u_rect bbox;
 282    unsigned tri_bytes;
 283    int x[4];
 284    int y[4];
 285    int i;
 286    int nr_planes = 4;
 287
 288    /* linewidth should be interpreted as integer */
 289    int fixed_width = util_iround(width) * FIXED_ONE;
 290
 291    float x_offset=0;
 292    float y_offset=0;
 293    float x_offset_end=0;
 294    float y_offset_end=0;
 295
 296    float x1diff;
 297    float y1diff;
 298    float x2diff;
 299    float y2diff;
 300    float dx, dy;
 301    float area;
 302
 303    boolean draw_start;
 304    boolean draw_end;
 305    boolean will_draw_start;
 306    boolean will_draw_end;
 307
 308    if (0)
 309       print_line(setup, v1, v2);
 310
 311    if (setup->scissor_test) {
 312       nr_planes = 8;
 313    }
 314    else {
 315       nr_planes = 4;
 316    }
 317
 318
 319    dx = v1[0][0] - v2[0][0];
 320    dy = v1[0][1] - v2[0][1];
 321    area = (dx * dx  + dy * dy);
 322    if (area == 0) {
 323       LP_COUNT(nr_culled_tris);
 324       return TRUE;
 325    }
 326
 327    info.oneoverarea = 1.0f / area;
 328    info.dx = dx;
 329    info.dy = dy;
 330    info.v1 = v1;
 331    info.v2 = v2;
 332
 333
 334    /* X-MAJOR LINE */
 335    if (fabsf(dx) >= fabsf(dy)) {
 336       float dydx = dy / dx;
 337
 338       x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
 339       y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
 340       x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
 341       y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
 342
 343       if (y2diff==-0.5 && dy<0){
 344          y2diff = 0.5;
 345       }
 346
 347       /*
 348        * Diamond exit rule test for starting point
 349        */
 350       if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
 351          draw_start = TRUE;
 352       }
 353       else if (sign(x1diff) == sign(-dx)) {
 354          draw_start = FALSE;
 355       }
 356       else if (sign(-y1diff) != sign(dy)) {
 357          draw_start = TRUE;
 358       }
 359       else {
 360          /* do intersection test */
 361          float yintersect = fracf(v1[0][1]) + x1diff * dydx;
 362          draw_start = (yintersect < 1.0 && yintersect > 0.0);
 363       }
 364
 365
 366       /*
 367        * Diamond exit rule test for ending point
 368        */
 369       if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
 370          draw_end = FALSE;
 371       }
 372       else if (sign(x2diff) != sign(-dx)) {
 373          draw_end = FALSE;
 374       }
 375       else if (sign(-y2diff) == sign(dy)) {
 376          draw_end = TRUE;
 377       }
 378       else {
 379          /* do intersection test */
 380          float yintersect = fracf(v2[0][1]) + x2diff * dydx;
 381          draw_end = (yintersect < 1.0 && yintersect > 0.0);
 382       }
 383
 384       /* Are we already drawing start/end?
 385        */
 386       will_draw_start = sign(-x1diff) != sign(dx);
 387       will_draw_end = (sign(x2diff) == sign(-dx)) || x2diff==0;
 388
 389       if (dx < 0) {
 390          /* if v2 is to the right of v1, swap pointers */
 391          const float (*temp)[4] = v1;
 392          v1 = v2;
 393          v2 = temp;
 394          dx = -dx;
 395          dy = -dy;
 396          /* Otherwise shift planes appropriately */
 397          if (will_draw_start != draw_start) {
 398             x_offset_end = - x1diff - 0.5;
 399             y_offset_end = x_offset_end * dydx;
 400
 401          }
 402          if (will_draw_end != draw_end) {
 403             x_offset = - x2diff - 0.5;
 404             y_offset = x_offset * dydx;
 405          }
 406
 407       }
 408       else{
 409          /* Otherwise shift planes appropriately */
 410          if (will_draw_start != draw_start) {
 411             x_offset = - x1diff + 0.5;
 412             y_offset = x_offset * dydx;
 413          }
 414          if (will_draw_end != draw_end) {
 415             x_offset_end = - x2diff + 0.5;
 416             y_offset_end = x_offset_end * dydx;
 417          }
 418       }
 419
 420       /* x/y positions in fixed point */
 421       x[0] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset);
 422       x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
 423       x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
 424       x[3] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset);
 425
 426       y[0] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset) - fixed_width/2;
 427       y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) - fixed_width/2;
 428       y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) + fixed_width/2;
 429       y[3] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset) + fixed_width/2;
 430
 431    }
 432    else {
 433       const float dxdy = dx / dy;
 434
 435       /* Y-MAJOR LINE */
 436       x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
 437       y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
 438       x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
 439       y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
 440
 441       if (x2diff==-0.5 && dx<0) {
 442          x2diff = 0.5;
 443       }
 444
 445       /*
 446        * Diamond exit rule test for starting point
 447        */
 448       if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
 449          draw_start = TRUE;
 450       }
 451       else if (sign(-y1diff) == sign(dy)) {
 452          draw_start = FALSE;
 453       }
 454       else if (sign(x1diff) != sign(-dx)) {
 455          draw_start = TRUE;
 456       }
 457       else {
 458          /* do intersection test */
 459          float xintersect = fracf(v1[0][0]) + y1diff * dxdy;
 460          draw_start = (xintersect < 1.0 && xintersect > 0.0);
 461       }
 462
 463       /*
 464        * Diamond exit rule test for ending point
 465        */
 466       if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
 467          draw_end = FALSE;
 468       }
 469       else if (sign(-y2diff) != sign(dy) ) {
 470          draw_end = FALSE;
 471       }
 472       else if (sign(x2diff) == sign(-dx) ) {
 473          draw_end = TRUE;
 474       }
 475       else {
 476          /* do intersection test */
 477          float xintersect = fracf(v2[0][0]) + y2diff * dxdy;
 478          draw_end = (xintersect < 1.0 && xintersect > 0.0);
 479       }
 480
 481       /* Are we already drawing start/end?
 482        */
 483       will_draw_start = sign(y1diff) == sign(dy);
 484       will_draw_end = (sign(-y2diff) == sign(dy)) || y2diff==0;
 485
 486       if (dy > 0) {
 487          /* if v2 is on top of v1, swap pointers */
 488          const float (*temp)[4] = v1;
 489          v1 = v2;
 490          v2 = temp;
 491          dx = -dx;
 492          dy = -dy;
 493
 494          /* Otherwise shift planes appropriately */
 495          if (will_draw_start != draw_start) {
 496             y_offset_end = - y1diff + 0.5;
 497             x_offset_end = y_offset_end * dxdy;
 498          }
 499          if (will_draw_end != draw_end) {
 500             y_offset = - y2diff + 0.5;
 501             x_offset = y_offset * dxdy;
 502          }
 503       }
 504       else {
 505          /* Otherwise shift planes appropriately */
 506          if (will_draw_start != draw_start) {
 507             y_offset = - y1diff - 0.5;
 508             x_offset = y_offset * dxdy;
 509
 510          }
 511          if (will_draw_end != draw_end) {
 512             y_offset_end = - y2diff - 0.5;
 513             x_offset_end = y_offset_end * dxdy;
 514          }
 515       }
 516
 517       /* x/y positions in fixed point */
 518       x[0] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset) - fixed_width/2;
 519       x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) - fixed_width/2;
 520       x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) + fixed_width/2;
 521       x[3] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset) + fixed_width/2;
 522
 523       y[0] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset);
 524       y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
 525       y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
 526       y[3] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset);
 527    }
 528
 529
 530
 531    LP_COUNT(nr_tris);
 532
 533
 534    /* Bounding rectangle (in pixels) */
 535    {
 536       /* Yes this is necessary to accurately calculate bounding boxes
 537        * with the two fill-conventions we support.  GL (normally) ends
 538        * up needing a bottom-left fill convention, which requires
 539        * slightly different rounding.
 540        */
 541       int adj = (setup->pixel_offset != 0) ? 1 : 0;
 542
 543       bbox.x0 = (MIN4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
 544       bbox.x1 = (MAX4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
 545       bbox.y0 = (MIN4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 546       bbox.y1 = (MAX4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 547
 548       /* Inclusive coordinates:
 549        */
 550       bbox.x1--;
 551       bbox.y1--;
 552    }
 553
 554    if (bbox.x1 < bbox.x0 ||
 555        bbox.y1 < bbox.y0) {
 556       if (0) debug_printf("empty bounding box\n");
 557       LP_COUNT(nr_culled_tris);
 558       return TRUE;
 559    }
 560
 561    if (!u_rect_test_intersection(&setup->draw_region, &bbox)) {
 562       if (0) debug_printf("offscreen\n");
 563       LP_COUNT(nr_culled_tris);
 564       return TRUE;
 565    }
 566
 567    u_rect_find_intersection(&setup->draw_region, &bbox);
 568
 569    line = lp_setup_alloc_triangle(scene,
 570                                   setup->fs.nr_inputs,
 571                                   nr_planes,
 572                                   &tri_bytes);
 573    if (!line)
 574       return FALSE;
 575
 576 #ifdef DEBUG
 577    line->v[0][0] = v1[0][0];
 578    line->v[1][0] = v2[0][0];
 579    line->v[0][1] = v1[0][1];
 580    line->v[1][1] = v2[0][1];
 581 #endif
 582
 583    /* calculate the deltas */
 584    line->plane[0].dcdy = x[0] - x[1];
 585    line->plane[1].dcdy = x[1] - x[2];
 586    line->plane[2].dcdy = x[2] - x[3];
 587    line->plane[3].dcdy = x[3] - x[0];
 588
 589    line->plane[0].dcdx = y[0] - y[1];
 590    line->plane[1].dcdx = y[1] - y[2];
 591    line->plane[2].dcdx = y[2] - y[3];
 592    line->plane[3].dcdx = y[3] - y[0];
 593
 594
 595    /* Setup parameter interpolants:
 596     */
 597    setup_line_coefficients( setup, line, &info);
 598
 599    line->inputs.facing = 1.0F;
 600    line->inputs.state = setup->fs.stored;
 601    line->inputs.disable = FALSE;
 602    line->inputs.opaque = FALSE;
 603
 604    for (i = 0; i < 4; i++) {
 605       struct lp_rast_plane *plane = &line->plane[i];
 606
 607       /* half-edge constants, will be interated over the whole render
 608        * target.
 609        */
 610       plane->c = plane->dcdx * x[i] - plane->dcdy * y[i];
 611
 612
 613       /* correct for top-left vs. bottom-left fill convention.
 614        *
 615        * note that we're overloading gl_rasterization_rules to mean
 616        * both (0.5,0.5) pixel centers *and* bottom-left filling
 617        * convention.
 618        *
 619        * GL actually has a top-left filling convention, but GL's
 620        * notion of "top" differs from gallium's...
 621        *
 622        * Also, sometimes (in FBO cases) GL will render upside down
 623        * to its usual method, in which case it will probably want
 624        * to use the opposite, top-left convention.
 625        */
 626       if (plane->dcdx < 0) {
 627          /* both fill conventions want this - adjust for left edges */
 628          plane->c++;
 629       }
 630       else if (plane->dcdx == 0) {
 631          if (setup->pixel_offset == 0) {
 632             /* correct for top-left fill convention:
 633              */
 634             if (plane->dcdy > 0) plane->c++;
 635          }
 636          else {
 637             /* correct for bottom-left fill convention:
 638              */
 639             if (plane->dcdy < 0) plane->c++;
 640          }
 641       }
 642
 643       plane->dcdx *= FIXED_ONE;
 644       plane->dcdy *= FIXED_ONE;
 645
 646       /* find trivial reject offsets for each edge for a single-pixel
 647        * sized block.  These will be scaled up at each recursive level to
 648        * match the active blocksize.  Scaling in this way works best if
 649        * the blocks are square.
 650        */
 651       plane->eo = 0;
 652       if (plane->dcdx < 0) plane->eo -= plane->dcdx;
 653       if (plane->dcdy > 0) plane->eo += plane->dcdy;
 654
 655       /* Calculate trivial accept offsets from the above.
 656        */
 657       plane->ei = plane->dcdy - plane->dcdx - plane->eo;
 658    }
 659
 660
 661    /*
 662     * When rasterizing scissored tris, use the intersection of the
 663     * triangle bounding box and the scissor rect to generate the
 664     * scissor planes.
 665     *
 666     * This permits us to cut off the triangle "tails" that are present
 667     * in the intermediate recursive levels caused when two of the
 668     * triangles edges don't diverge quickly enough to trivially reject
 669     * exterior blocks from the triangle.
 670     *
 671     * It's not really clear if it's worth worrying about these tails,
 672     * but since we generate the planes for each scissored tri, it's
 673     * free to trim them in this case.
 674     *
 675     * Note that otherwise, the scissor planes only vary in 'C' value,
 676     * and even then only on state-changes.  Could alternatively store
 677     * these planes elsewhere.
 678     */
 679    if (nr_planes == 8) {
 680       line->plane[4].dcdx = -1;
 681       line->plane[4].dcdy = 0;
 682       line->plane[4].c = 1-bbox.x0;
 683       line->plane[4].ei = 0;
 684       line->plane[4].eo = 1;
 685
 686       line->plane[5].dcdx = 1;
 687       line->plane[5].dcdy = 0;
 688       line->plane[5].c = bbox.x1+1;
 689       line->plane[5].ei = -1;
 690       line->plane[5].eo = 0;
 691
 692       line->plane[6].dcdx = 0;
 693       line->plane[6].dcdy = 1;
 694       line->plane[6].c = 1-bbox.y0;
 695       line->plane[6].ei = 0;
 696       line->plane[6].eo = 1;
 697
 698       line->plane[7].dcdx = 0;
 699       line->plane[7].dcdy = -1;
 700       line->plane[7].c = bbox.y1+1;
 701       line->plane[7].ei = -1;
 702       line->plane[7].eo = 0;
 703    }
 704
 705    return lp_setup_bin_triangle(setup, line, &bbox, nr_planes);
 706 }
 707
 708
 709 static void lp_setup_line( struct lp_setup_context *setup,
 710                            const float (*v0)[4],
 711                            const float (*v1)[4] )
 712 {
 713    if (!try_setup_line( setup, v0, v1 ))
 714    {
 715       lp_setup_flush_and_restart(setup);
 716
 717       if (!try_setup_line( setup, v0, v1 ))
 718          assert(0);
 719    }
 720 }
 721
 722
 723 void lp_setup_choose_line( struct lp_setup_context *setup )
 724 {
 725    setup->line = lp_setup_line;
 726 }
 727
 728