Home | History | Annotate | Download | only in llvmpipe
      1 /**************************************************************************
      2  *
      3  * Copyright 2007 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  * 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 #include "lp_state_setup.h"
     39 #include "lp_context.h"
     40 #include "draw/draw_context.h"
     41 
     42 #define NUM_CHANNELS 4
     43 
     44 struct lp_line_info {
     45 
     46    float dx;
     47    float dy;
     48    float oneoverarea;
     49    boolean frontfacing;
     50 
     51    const float (*v1)[4];
     52    const float (*v2)[4];
     53 
     54    float (*a0)[4];
     55    float (*dadx)[4];
     56    float (*dady)[4];
     57 };
     58 
     59 
     60 /**
     61  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
     62  */
     63 static void constant_coef( struct lp_setup_context *setup,
     64                            struct lp_line_info *info,
     65                            unsigned slot,
     66                            const float value,
     67                            unsigned i )
     68 {
     69    info->a0[slot][i] = value;
     70    info->dadx[slot][i] = 0.0f;
     71    info->dady[slot][i] = 0.0f;
     72 }
     73 
     74 
     75 /**
     76  * Compute a0, dadx and dady for a linearly interpolated coefficient,
     77  * for a triangle.
     78  */
     79 static void linear_coef( struct lp_setup_context *setup,
     80                          struct lp_line_info *info,
     81                          unsigned slot,
     82                          unsigned vert_attr,
     83                          unsigned i)
     84 {
     85    float a1 = info->v1[vert_attr][i];
     86    float a2 = info->v2[vert_attr][i];
     87 
     88    float da21 = a1 - a2;
     89    float dadx = da21 * info->dx * info->oneoverarea;
     90    float dady = da21 * info->dy * info->oneoverarea;
     91 
     92    info->dadx[slot][i] = dadx;
     93    info->dady[slot][i] = dady;
     94 
     95    info->a0[slot][i] = (a1 -
     96                               (dadx * (info->v1[0][0] - setup->pixel_offset) +
     97                                dady * (info->v1[0][1] - setup->pixel_offset)));
     98 }
     99 
    100 
    101 /**
    102  * Compute a0, dadx and dady for a perspective-corrected interpolant,
    103  * for a triangle.
    104  * We basically multiply the vertex value by 1/w before computing
    105  * the plane coefficients (a0, dadx, dady).
    106  * Later, when we compute the value at a particular fragment position we'll
    107  * divide the interpolated value by the interpolated W at that fragment.
    108  */
    109 static void perspective_coef( struct lp_setup_context *setup,
    110                               struct lp_line_info *info,
    111                               unsigned slot,
    112                               unsigned vert_attr,
    113                               unsigned i)
    114 {
    115    /* premultiply by 1/w  (v[0][3] is always 1/w):
    116     */
    117    float a1 = info->v1[vert_attr][i] * info->v1[0][3];
    118    float a2 = info->v2[vert_attr][i] * info->v2[0][3];
    119 
    120    float da21 = a1 - a2;
    121    float dadx = da21 * info->dx * info->oneoverarea;
    122    float dady = da21 * info->dy * info->oneoverarea;
    123 
    124    info->dadx[slot][i] = dadx;
    125    info->dady[slot][i] = dady;
    126 
    127    info->a0[slot][i] = (a1 -
    128                         (dadx * (info->v1[0][0] - setup->pixel_offset) +
    129                          dady * (info->v1[0][1] - setup->pixel_offset)));
    130 }
    131 
    132 static void
    133 setup_fragcoord_coef( struct lp_setup_context *setup,
    134                       struct lp_line_info *info,
    135                       unsigned slot,
    136                       unsigned usage_mask)
    137 {
    138    /*X*/
    139    if (usage_mask & TGSI_WRITEMASK_X) {
    140       info->a0[slot][0] = 0.0;
    141       info->dadx[slot][0] = 1.0;
    142       info->dady[slot][0] = 0.0;
    143    }
    144 
    145    /*Y*/
    146    if (usage_mask & TGSI_WRITEMASK_Y) {
    147       info->a0[slot][1] = 0.0;
    148       info->dadx[slot][1] = 0.0;
    149       info->dady[slot][1] = 1.0;
    150    }
    151 
    152    /*Z*/
    153    if (usage_mask & TGSI_WRITEMASK_Z) {
    154       linear_coef(setup, info, slot, 0, 2);
    155    }
    156 
    157    /*W*/
    158    if (usage_mask & TGSI_WRITEMASK_W) {
    159       linear_coef(setup, info, slot, 0, 3);
    160    }
    161 }
    162 
    163 /**
    164  * Compute the tri->coef[] array dadx, dady, a0 values.
    165  */
    166 static void setup_line_coefficients( struct lp_setup_context *setup,
    167                                      struct lp_line_info *info)
    168 {
    169    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
    170    unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
    171    unsigned slot;
    172 
    173    /* setup interpolation for all the remaining attributes:
    174     */
    175    for (slot = 0; slot < key->num_inputs; slot++) {
    176       unsigned vert_attr = key->inputs[slot].src_index;
    177       unsigned usage_mask = key->inputs[slot].usage_mask;
    178       unsigned i;
    179 
    180       switch (key->inputs[slot].interp) {
    181       case LP_INTERP_CONSTANT:
    182          if (key->flatshade_first) {
    183             for (i = 0; i < NUM_CHANNELS; i++)
    184                if (usage_mask & (1 << i))
    185                   constant_coef(setup, info, slot+1, info->v1[vert_attr][i], i);
    186          }
    187          else {
    188             for (i = 0; i < NUM_CHANNELS; i++)
    189                if (usage_mask & (1 << i))
    190                   constant_coef(setup, info, slot+1, info->v2[vert_attr][i], i);
    191          }
    192          break;
    193 
    194       case LP_INTERP_LINEAR:
    195          for (i = 0; i < NUM_CHANNELS; i++)
    196             if (usage_mask & (1 << i))
    197                linear_coef(setup, info, slot+1, vert_attr, i);
    198          break;
    199 
    200       case LP_INTERP_PERSPECTIVE:
    201          for (i = 0; i < NUM_CHANNELS; i++)
    202             if (usage_mask & (1 << i))
    203                perspective_coef(setup, info, slot+1, vert_attr, i);
    204          fragcoord_usage_mask |= TGSI_WRITEMASK_W;
    205          break;
    206 
    207       case LP_INTERP_POSITION:
    208          /*
    209           * The generated pixel interpolators will pick up the coeffs from
    210           * slot 0, so all need to ensure that the usage mask is covers all
    211           * usages.
    212           */
    213          fragcoord_usage_mask |= usage_mask;
    214          break;
    215 
    216       case LP_INTERP_FACING:
    217          for (i = 0; i < NUM_CHANNELS; i++)
    218             if (usage_mask & (1 << i))
    219                constant_coef(setup, info, slot+1,
    220                              info->frontfacing ? 1.0f : -1.0f, i);
    221          break;
    222 
    223       default:
    224          assert(0);
    225       }
    226    }
    227 
    228    /* The internal position input is in slot zero:
    229     */
    230    setup_fragcoord_coef(setup, info, 0,
    231                         fragcoord_usage_mask);
    232 }
    233 
    234 
    235 
    236 static inline int subpixel_snap( float a )
    237 {
    238    return util_iround(FIXED_ONE * a);
    239 }
    240 
    241 
    242 /**
    243  * Print line vertex attribs (for debug).
    244  */
    245 static void
    246 print_line(struct lp_setup_context *setup,
    247            const float (*v1)[4],
    248            const float (*v2)[4])
    249 {
    250    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
    251    uint i;
    252 
    253    debug_printf("llvmpipe line\n");
    254    for (i = 0; i < 1 + key->num_inputs; i++) {
    255       debug_printf("  v1[%d]:  %f %f %f %f\n", i,
    256                    v1[i][0], v1[i][1], v1[i][2], v1[i][3]);
    257    }
    258    for (i = 0; i < 1 + key->num_inputs; i++) {
    259       debug_printf("  v2[%d]:  %f %f %f %f\n", i,
    260                    v2[i][0], v2[i][1], v2[i][2], v2[i][3]);
    261    }
    262 }
    263 
    264 
    265 static inline boolean sign(float x){
    266    return x >= 0;
    267 }
    268 
    269 
    270 /* Used on positive floats only:
    271  */
    272 static inline float fracf(float f)
    273 {
    274    return f - floorf(f);
    275 }
    276 
    277 
    278 
    279 static boolean
    280 try_setup_line( struct lp_setup_context *setup,
    281                const float (*v1)[4],
    282                const float (*v2)[4])
    283 {
    284    struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
    285    struct lp_scene *scene = setup->scene;
    286    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
    287    struct lp_rast_triangle *line;
    288    struct lp_rast_plane *plane;
    289    struct lp_line_info info;
    290    float width = MAX2(1.0, setup->line_width);
    291    struct u_rect bbox;
    292    unsigned tri_bytes;
    293    int x[4];
    294    int y[4];
    295    int i;
    296    int nr_planes = 4;
    297    unsigned viewport_index = 0;
    298    unsigned layer = 0;
    299 
    300    /* linewidth should be interpreted as integer */
    301    int fixed_width = util_iround(width) * FIXED_ONE;
    302 
    303    float x_offset=0;
    304    float y_offset=0;
    305    float x_offset_end=0;
    306    float y_offset_end=0;
    307 
    308    float x1diff;
    309    float y1diff;
    310    float x2diff;
    311    float y2diff;
    312    float dx, dy;
    313    float area;
    314    const float (*pv)[4];
    315 
    316    boolean draw_start;
    317    boolean draw_end;
    318    boolean will_draw_start;
    319    boolean will_draw_end;
    320 
    321    if (0)
    322       print_line(setup, v1, v2);
    323 
    324    if (setup->flatshade_first) {
    325       pv = v1;
    326    }
    327    else {
    328       pv = v2;
    329    }
    330    if (setup->viewport_index_slot > 0) {
    331       unsigned *udata = (unsigned*)pv[setup->viewport_index_slot];
    332       viewport_index = lp_clamp_viewport_idx(*udata);
    333    }
    334    if (setup->layer_slot > 0) {
    335       layer = *(unsigned*)pv[setup->layer_slot];
    336       layer = MIN2(layer, scene->fb_max_layer);
    337    }
    338 
    339    dx = v1[0][0] - v2[0][0];
    340    dy = v1[0][1] - v2[0][1];
    341    area = (dx * dx  + dy * dy);
    342    if (area == 0) {
    343       LP_COUNT(nr_culled_tris);
    344       return TRUE;
    345    }
    346 
    347    info.oneoverarea = 1.0f / area;
    348    info.dx = dx;
    349    info.dy = dy;
    350    info.v1 = v1;
    351    info.v2 = v2;
    352 
    353 
    354    /* X-MAJOR LINE */
    355    if (fabsf(dx) >= fabsf(dy)) {
    356       float dydx = dy / dx;
    357 
    358       x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
    359       y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
    360       x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
    361       y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
    362 
    363       if (y2diff==-0.5 && dy<0){
    364          y2diff = 0.5;
    365       }
    366 
    367       /*
    368        * Diamond exit rule test for starting point
    369        */
    370       if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
    371          draw_start = TRUE;
    372       }
    373       else if (sign(x1diff) == sign(-dx)) {
    374          draw_start = FALSE;
    375       }
    376       else if (sign(-y1diff) != sign(dy)) {
    377          draw_start = TRUE;
    378       }
    379       else {
    380          /* do intersection test */
    381          float yintersect = fracf(v1[0][1]) + x1diff * dydx;
    382          draw_start = (yintersect < 1.0 && yintersect > 0.0);
    383       }
    384 
    385 
    386       /*
    387        * Diamond exit rule test for ending point
    388        */
    389       if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
    390          draw_end = FALSE;
    391       }
    392       else if (sign(x2diff) != sign(-dx)) {
    393          draw_end = FALSE;
    394       }
    395       else if (sign(-y2diff) == sign(dy)) {
    396          draw_end = TRUE;
    397       }
    398       else {
    399          /* do intersection test */
    400          float yintersect = fracf(v2[0][1]) + x2diff * dydx;
    401          draw_end = (yintersect < 1.0 && yintersect > 0.0);
    402       }
    403 
    404       /* Are we already drawing start/end?
    405        */
    406       will_draw_start = sign(-x1diff) != sign(dx);
    407       will_draw_end = (sign(x2diff) == sign(-dx)) || x2diff==0;
    408 
    409       if (dx < 0) {
    410          /* if v2 is to the right of v1, swap pointers */
    411          const float (*temp)[4] = v1;
    412          v1 = v2;
    413          v2 = temp;
    414          dx = -dx;
    415          dy = -dy;
    416          /* Otherwise shift planes appropriately */
    417          if (will_draw_start != draw_start) {
    418             x_offset_end = - x1diff - 0.5;
    419             y_offset_end = x_offset_end * dydx;
    420 
    421          }
    422          if (will_draw_end != draw_end) {
    423             x_offset = - x2diff - 0.5;
    424             y_offset = x_offset * dydx;
    425          }
    426 
    427       }
    428       else{
    429          /* Otherwise shift planes appropriately */
    430          if (will_draw_start != draw_start) {
    431             x_offset = - x1diff + 0.5;
    432             y_offset = x_offset * dydx;
    433          }
    434          if (will_draw_end != draw_end) {
    435             x_offset_end = - x2diff + 0.5;
    436             y_offset_end = x_offset_end * dydx;
    437          }
    438       }
    439 
    440       /* x/y positions in fixed point */
    441       x[0] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset);
    442       x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
    443       x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset);
    444       x[3] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset);
    445 
    446       y[0] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset) - fixed_width/2;
    447       y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) - fixed_width/2;
    448       y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) + fixed_width/2;
    449       y[3] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset) + fixed_width/2;
    450 
    451    }
    452    else {
    453       const float dxdy = dx / dy;
    454 
    455       /* Y-MAJOR LINE */
    456       x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5;
    457       y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5;
    458       x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5;
    459       y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5;
    460 
    461       if (x2diff==-0.5 && dx<0) {
    462          x2diff = 0.5;
    463       }
    464 
    465       /*
    466        * Diamond exit rule test for starting point
    467        */
    468       if (fabsf(x1diff) + fabsf(y1diff) < 0.5) {
    469          draw_start = TRUE;
    470       }
    471       else if (sign(-y1diff) == sign(dy)) {
    472          draw_start = FALSE;
    473       }
    474       else if (sign(x1diff) != sign(-dx)) {
    475          draw_start = TRUE;
    476       }
    477       else {
    478          /* do intersection test */
    479          float xintersect = fracf(v1[0][0]) + y1diff * dxdy;
    480          draw_start = (xintersect < 1.0 && xintersect > 0.0);
    481       }
    482 
    483       /*
    484        * Diamond exit rule test for ending point
    485        */
    486       if (fabsf(x2diff) + fabsf(y2diff) < 0.5) {
    487          draw_end = FALSE;
    488       }
    489       else if (sign(-y2diff) != sign(dy) ) {
    490          draw_end = FALSE;
    491       }
    492       else if (sign(x2diff) == sign(-dx) ) {
    493          draw_end = TRUE;
    494       }
    495       else {
    496          /* do intersection test */
    497          float xintersect = fracf(v2[0][0]) + y2diff * dxdy;
    498          draw_end = (xintersect < 1.0 && xintersect >= 0.0);
    499       }
    500 
    501       /* Are we already drawing start/end?
    502        */
    503       will_draw_start = sign(y1diff) == sign(dy);
    504       will_draw_end = (sign(-y2diff) == sign(dy)) || y2diff==0;
    505 
    506       if (dy > 0) {
    507          /* if v2 is on top of v1, swap pointers */
    508          const float (*temp)[4] = v1;
    509          v1 = v2;
    510          v2 = temp;
    511          dx = -dx;
    512          dy = -dy;
    513 
    514          /* Otherwise shift planes appropriately */
    515          if (will_draw_start != draw_start) {
    516             y_offset_end = - y1diff + 0.5;
    517             x_offset_end = y_offset_end * dxdy;
    518          }
    519          if (will_draw_end != draw_end) {
    520             y_offset = - y2diff + 0.5;
    521             x_offset = y_offset * dxdy;
    522          }
    523       }
    524       else {
    525          /* Otherwise shift planes appropriately */
    526          if (will_draw_start != draw_start) {
    527             y_offset = - y1diff - 0.5;
    528             x_offset = y_offset * dxdy;
    529 
    530          }
    531          if (will_draw_end != draw_end) {
    532             y_offset_end = - y2diff - 0.5;
    533             x_offset_end = y_offset_end * dxdy;
    534          }
    535       }
    536 
    537       /* x/y positions in fixed point */
    538       x[0] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset) - fixed_width/2;
    539       x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) - fixed_width/2;
    540       x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) + fixed_width/2;
    541       x[3] = subpixel_snap(v1[0][0] + x_offset     - setup->pixel_offset) + fixed_width/2;
    542 
    543       y[0] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset);
    544       y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
    545       y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset);
    546       y[3] = subpixel_snap(v1[0][1] + y_offset     - setup->pixel_offset);
    547    }
    548 
    549    /* Bounding rectangle (in pixels) */
    550    {
    551       /* Yes this is necessary to accurately calculate bounding boxes
    552        * with the two fill-conventions we support.  GL (normally) ends
    553        * up needing a bottom-left fill convention, which requires
    554        * slightly different rounding.
    555        */
    556       int adj = (setup->bottom_edge_rule != 0) ? 1 : 0;
    557 
    558       bbox.x0 = (MIN4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
    559       bbox.x1 = (MAX4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER;
    560       bbox.y0 = (MIN4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
    561       bbox.y1 = (MAX4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
    562 
    563       /* Inclusive coordinates:
    564        */
    565       bbox.x1--;
    566       bbox.y1--;
    567    }
    568 
    569    if (bbox.x1 < bbox.x0 ||
    570        bbox.y1 < bbox.y0) {
    571       if (0) debug_printf("empty bounding box\n");
    572       LP_COUNT(nr_culled_tris);
    573       return TRUE;
    574    }
    575 
    576    if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) {
    577       if (0) debug_printf("offscreen\n");
    578       LP_COUNT(nr_culled_tris);
    579       return TRUE;
    580    }
    581 
    582    /* Can safely discard negative regions:
    583     */
    584    bbox.x0 = MAX2(bbox.x0, 0);
    585    bbox.y0 = MAX2(bbox.y0, 0);
    586 
    587    nr_planes = 4;
    588    /*
    589     * Determine how many scissor planes we need, that is drop scissor
    590     * edges if the bounding box of the tri is fully inside that edge.
    591     */
    592    if (setup->scissor_test) {
    593       /* why not just use draw_regions */
    594       boolean s_planes[4];
    595       scissor_planes_needed(s_planes, &bbox, &setup->scissors[viewport_index]);
    596       nr_planes += s_planes[0] + s_planes[1] + s_planes[2] + s_planes[3];
    597    }
    598 
    599    line = lp_setup_alloc_triangle(scene,
    600                                   key->num_inputs,
    601                                   nr_planes,
    602                                   &tri_bytes);
    603    if (!line)
    604       return FALSE;
    605 
    606 #ifdef DEBUG
    607    line->v[0][0] = v1[0][0];
    608    line->v[1][0] = v2[0][0];
    609    line->v[0][1] = v1[0][1];
    610    line->v[1][1] = v2[0][1];
    611 #endif
    612 
    613    LP_COUNT(nr_tris);
    614 
    615    if (lp_context->active_statistics_queries &&
    616        !llvmpipe_rasterization_disabled(lp_context)) {
    617       lp_context->pipeline_statistics.c_primitives++;
    618    }
    619 
    620    /* calculate the deltas */
    621    plane = GET_PLANES(line);
    622    plane[0].dcdy = x[0] - x[1];
    623    plane[1].dcdy = x[1] - x[2];
    624    plane[2].dcdy = x[2] - x[3];
    625    plane[3].dcdy = x[3] - x[0];
    626 
    627    plane[0].dcdx = y[0] - y[1];
    628    plane[1].dcdx = y[1] - y[2];
    629    plane[2].dcdx = y[2] - y[3];
    630    plane[3].dcdx = y[3] - y[0];
    631 
    632    if (draw_will_inject_frontface(lp_context->draw) &&
    633        setup->face_slot > 0) {
    634       line->inputs.frontfacing = v1[setup->face_slot][0];
    635    } else {
    636       line->inputs.frontfacing = TRUE;
    637    }
    638 
    639    /* Setup parameter interpolants:
    640     */
    641    info.a0 = GET_A0(&line->inputs);
    642    info.dadx = GET_DADX(&line->inputs);
    643    info.dady = GET_DADY(&line->inputs);
    644    info.frontfacing = line->inputs.frontfacing;
    645    setup_line_coefficients(setup, &info);
    646 
    647    line->inputs.disable = FALSE;
    648    line->inputs.opaque = FALSE;
    649    line->inputs.layer = layer;
    650    line->inputs.viewport_index = viewport_index;
    651 
    652    /*
    653     * XXX: this code is mostly identical to the one in lp_setup_tri, except it
    654     * uses 4 planes instead of 3. Could share the code (including the sse
    655     * assembly, in fact we'd get the 4th plane for free).
    656     * The only difference apart from storing the 4th plane would be some
    657     * different shuffle for calculating dcdx/dcdy.
    658     */
    659    for (i = 0; i < 4; i++) {
    660 
    661       /* half-edge constants, will be iterated over the whole render
    662        * target.
    663        */
    664       plane[i].c = IMUL64(plane[i].dcdx, x[i]) - IMUL64(plane[i].dcdy, y[i]);
    665 
    666       /* correct for top-left vs. bottom-left fill convention.
    667        */
    668       if (plane[i].dcdx < 0) {
    669          /* both fill conventions want this - adjust for left edges */
    670          plane[i].c++;
    671       }
    672       else if (plane[i].dcdx == 0) {
    673          if (setup->pixel_offset == 0) {
    674             /* correct for top-left fill convention:
    675              */
    676             if (plane[i].dcdy > 0) plane[i].c++;
    677          }
    678          else {
    679             /* correct for bottom-left fill convention:
    680              */
    681             if (plane[i].dcdy < 0) plane[i].c++;
    682          }
    683       }
    684 
    685       plane[i].dcdx *= FIXED_ONE;
    686       plane[i].dcdy *= FIXED_ONE;
    687 
    688       /* find trivial reject offsets for each edge for a single-pixel
    689        * sized block.  These will be scaled up at each recursive level to
    690        * match the active blocksize.  Scaling in this way works best if
    691        * the blocks are square.
    692        */
    693       plane[i].eo = 0;
    694       if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx;
    695       if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy;
    696    }
    697 
    698 
    699    /*
    700     * When rasterizing scissored tris, use the intersection of the
    701     * triangle bounding box and the scissor rect to generate the
    702     * scissor planes.
    703     *
    704     * This permits us to cut off the triangle "tails" that are present
    705     * in the intermediate recursive levels caused when two of the
    706     * triangles edges don't diverge quickly enough to trivially reject
    707     * exterior blocks from the triangle.
    708     *
    709     * It's not really clear if it's worth worrying about these tails,
    710     * but since we generate the planes for each scissored tri, it's
    711     * free to trim them in this case.
    712     *
    713     * Note that otherwise, the scissor planes only vary in 'C' value,
    714     * and even then only on state-changes.  Could alternatively store
    715     * these planes elsewhere.
    716     * (Or only store the c value together with a bit indicating which
    717     * scissor edge this is, so rasterization would treat them differently
    718     * (easier to evaluate) to ordinary planes.)
    719     */
    720    if (nr_planes > 4) {
    721       /* why not just use draw_regions */
    722       const struct u_rect *scissor = &setup->scissors[viewport_index];
    723       struct lp_rast_plane *plane_s = &plane[4];
    724       boolean s_planes[4];
    725       scissor_planes_needed(s_planes, &bbox, scissor);
    726 
    727       if (s_planes[0]) {
    728          plane_s->dcdx = -1 << 8;
    729          plane_s->dcdy = 0;
    730          plane_s->c = (1-scissor->x0) << 8;
    731          plane_s->eo = 1 << 8;
    732          plane_s++;
    733       }
    734       if (s_planes[1]) {
    735          plane_s->dcdx = 1 << 8;
    736          plane_s->dcdy = 0;
    737          plane_s->c = (scissor->x1+1) << 8;
    738          plane_s->eo = 0 << 8;
    739          plane_s++;
    740       }
    741       if (s_planes[2]) {
    742          plane_s->dcdx = 0;
    743          plane_s->dcdy = 1 << 8;
    744          plane_s->c = (1-scissor->y0) << 8;
    745          plane_s->eo = 1 << 8;
    746          plane_s++;
    747       }
    748       if (s_planes[3]) {
    749          plane_s->dcdx = 0;
    750          plane_s->dcdy = -1 << 8;
    751          plane_s->c = (scissor->y1+1) << 8;
    752          plane_s->eo = 0;
    753          plane_s++;
    754       }
    755       assert(plane_s == &plane[nr_planes]);
    756    }
    757 
    758    return lp_setup_bin_triangle(setup, line, &bbox, nr_planes, viewport_index);
    759 }
    760 
    761 
    762 static void lp_setup_line( struct lp_setup_context *setup,
    763                            const float (*v0)[4],
    764                            const float (*v1)[4] )
    765 {
    766    if (!try_setup_line( setup, v0, v1 ))
    767    {
    768       if (!lp_setup_flush_and_restart(setup))
    769          return;
    770 
    771       if (!try_setup_line( setup, v0, v1 ))
    772          return;
    773    }
    774 }
    775 
    776 
    777 void lp_setup_choose_line( struct lp_setup_context *setup )
    778 {
    779    setup->line = lp_setup_line;
    780 }
    781 
    782 
    783