src/gallium/drivers/llvmpipe/lp_setup_point.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2010, 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 points
  30  */
  31
  32 #include "lp_setup_context.h"
  33 #include "util/u_math.h"
  34 #include "util/u_memory.h"
  35 #include "lp_perf.h"
  36 #include "lp_setup_context.h"
  37 #include "lp_rast.h"
  38 #include "lp_state_fs.h"
  39 #include "tgsi/tgsi_scan.h"
  40
  41 #define NUM_CHANNELS 4
  42
  43 struct point_info {
  44    /* x,y deltas */
  45    int dy01, dy12;
  46    int dx01, dx12;
  47
  48    const float (*v0)[4];
  49 };
  50
  51
  52 /**
  53  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  54  */
  55 static void
  56 constant_coef(struct lp_setup_context *setup,
  57               struct lp_rast_triangle *point,
  58               unsigned slot,
  59               const float value,
  60               unsigned i)
  61 {
  62    point->inputs.a0[slot][i] = value;
  63    point->inputs.dadx[slot][i] = 0.0f;
  64    point->inputs.dady[slot][i] = 0.0f;
  65 }
  66
  67
  68 static void
  69 perspective_coef(struct lp_setup_context *setup,
  70                  struct lp_rast_triangle *point,
  71                  const struct point_info *info,
  72                  unsigned slot,
  73                  unsigned vert_attr,
  74                  unsigned i)
  75 {
  76    if (i == 0) {
  77       float dadx = FIXED_ONE / (float)info->dx12;
  78       float dady =  0.0f;
  79       point->inputs.dadx[slot][i] = dadx;
  80       point->inputs.dady[slot][i] = dady;
  81       point->inputs.a0[slot][i] = (0.5 -
  82                                   (dadx * ((float)info->v0[0][0] - setup->pixel_offset) +
  83                                    dady * ((float)info->v0[0][1] - setup->pixel_offset)));
  84    }
  85    else if (i == 1) {
  86       float dadx =  0.0f;
  87       float dady =  FIXED_ONE / (float)info->dx12;
  88
  89       point->inputs.dadx[slot][i] = dadx;
  90       point->inputs.dady[slot][i] = dady;
  91       point->inputs.a0[slot][i] = (0.5 -
  92                                   (dadx * ((float)info->v0[0][0] - setup->pixel_offset) +
  93                                    dady * ((float)info->v0[0][1] - setup->pixel_offset)));
  94    }
  95    else if (i == 2) {
  96       point->inputs.a0[slot][i] = 0.0f;
  97       point->inputs.dadx[slot][i] = 0.0f;
  98       point->inputs.dady[slot][i] = 0.0f;
  99    }
 100    else if (i == 3) {
 101       point->inputs.a0[slot][i] = 1.0f;
 102       point->inputs.dadx[slot][i] = 0.0f;
 103       point->inputs.dady[slot][i] = 0.0f;
 104    }
 105 }
 106
 107
 108 /**
 109  * Special coefficient setup for gl_FragCoord.
 110  * X and Y are trivial
 111  * Z and W are copied from position_coef which should have already been computed.
 112  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
 113  */
 114 static void
 115 setup_point_fragcoord_coef(struct lp_setup_context *setup,
 116                            struct lp_rast_triangle *point,
 117                            const struct point_info *info,
 118                            unsigned slot,
 119                            unsigned usage_mask)
 120 {
 121    /*X*/
 122    if (usage_mask & TGSI_WRITEMASK_X) {
 123       point->inputs.a0[slot][0] = 0.0;
 124       point->inputs.dadx[slot][0] = 1.0;
 125       point->inputs.dady[slot][0] = 0.0;
 126    }
 127
 128    /*Y*/
 129    if (usage_mask & TGSI_WRITEMASK_Y) {
 130       point->inputs.a0[slot][1] = 0.0;
 131       point->inputs.dadx[slot][1] = 0.0;
 132       point->inputs.dady[slot][1] = 1.0;
 133    }
 134
 135    /*Z*/
 136    if (usage_mask & TGSI_WRITEMASK_Z) {
 137       constant_coef(setup, point, slot, info->v0[0][2], 2);
 138    }
 139
 140    /*W*/
 141    if (usage_mask & TGSI_WRITEMASK_W) {
 142       constant_coef(setup, point, slot, info->v0[0][3], 3);
 143    }
 144 }
 145
 146
 147 /**
 148  * Compute the point->coef[] array dadx, dady, a0 values.
 149  */
 150 static void
 151 setup_point_coefficients( struct lp_setup_context *setup,
 152                           struct lp_rast_triangle *point,
 153                           const struct point_info *info)
 154 {
 155    const struct lp_fragment_shader *shader = setup->fs.current.variant->shader;
 156    unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
 157    unsigned slot;
 158
 159    /* setup interpolation for all the remaining attributes:
 160     */
 161    for (slot = 0; slot < setup->fs.nr_inputs; slot++) {
 162       unsigned vert_attr = setup->fs.input[slot].src_index;
 163       unsigned usage_mask = setup->fs.input[slot].usage_mask;
 164       unsigned i;
 165
 166       switch (setup->fs.input[slot].interp) {
 167       case LP_INTERP_POSITION:
 168          /*
 169           * The generated pixel interpolators will pick up the coeffs from
 170           * slot 0, so all need to ensure that the usage mask is covers all
 171           * usages.
 172           */
 173          fragcoord_usage_mask |= usage_mask;
 174          break;
 175
 176       case LP_INTERP_LINEAR:
 177          /* Sprite tex coords may use linear interpolation someday */
 178          /* fall-through */
 179
 180       case LP_INTERP_PERSPECTIVE:
 181          /* check if the sprite coord flag is set for this attribute.
 182           * If so, set it up so it up so x any y vary from 0 to 1.
 183           */
 184          if (shader->info.input_semantic_name[slot] == TGSI_SEMANTIC_GENERIC) {
 185             const int index = shader->info.input_semantic_index[slot];
 186
 187             if (setup->sprite & (1 << index)) {
 188                for (i = 0; i < NUM_CHANNELS; i++)
 189                   if (usage_mask & (1 << i))
 190                      perspective_coef(setup, point, info, slot+1, vert_attr, i);
 191                fragcoord_usage_mask |= TGSI_WRITEMASK_W;
 192                break;
 193             }
 194          }
 195
 196          /* Otherwise fallthrough */
 197       default:
 198          for (i = 0; i < NUM_CHANNELS; i++) {
 199             if (usage_mask & (1 << i))
 200                constant_coef(setup, point, slot+1, info->v0[vert_attr][i], i);
 201          }
 202       }
 203    }
 204
 205    /* The internal position input is in slot zero:
 206     */
 207    setup_point_fragcoord_coef(setup, point, info, 0,
 208                               fragcoord_usage_mask);
 209 }
 210
 211
 212 static INLINE int
 213 subpixel_snap(float a)
 214 {
 215    return util_iround(FIXED_ONE * a);
 216 }
 217
 218
 219 static boolean
 220 try_setup_point( struct lp_setup_context *setup,
 221                  const float (*v0)[4] )
 222 {
 223    /* x/y positions in fixed point */
 224    const int sizeAttr = setup->psize;
 225    const float size
 226       = (setup->point_size_per_vertex && sizeAttr > 0) ? v0[sizeAttr][0]
 227       : setup->point_size;
 228
 229    /* Point size as fixed point integer, remove rounding errors
 230     * and gives minimum width for very small points
 231     */
 232    int fixed_width = MAX2(FIXED_ONE,
 233                           (subpixel_snap(size) + FIXED_ONE/2 - 1) & ~(FIXED_ONE-1));
 234
 235    const int x0 = subpixel_snap(v0[0][0] - setup->pixel_offset) - fixed_width/2;
 236    const int y0 = subpixel_snap(v0[0][1] - setup->pixel_offset) - fixed_width/2;
 237
 238    struct lp_scene *scene = setup->scene;
 239    struct lp_rast_triangle *point;
 240    unsigned bytes;
 241    struct u_rect bbox;
 242    unsigned nr_planes = 4;
 243    struct point_info info;
 244
 245
 246    /* Bounding rectangle (in pixels) */
 247    {
 248       /* Yes this is necessary to accurately calculate bounding boxes
 249        * with the two fill-conventions we support.  GL (normally) ends
 250        * up needing a bottom-left fill convention, which requires
 251        * slightly different rounding.
 252        */
 253       int adj = (setup->pixel_offset != 0) ? 1 : 0;
 254
 255       bbox.x0 = (x0 + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 256       bbox.x1 = (x0 + fixed_width + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 257       bbox.y0 = (y0 + (FIXED_ONE-1)) >> FIXED_ORDER;
 258       bbox.y1 = (y0 + fixed_width + (FIXED_ONE-1)) >> FIXED_ORDER;
 259
 260       /* Inclusive coordinates:
 261        */
 262       bbox.x1--;
 263       bbox.y1--;
 264    }
 265
 266    if (!u_rect_test_intersection(&setup->draw_region, &bbox)) {
 267       if (0) debug_printf("offscreen\n");
 268       LP_COUNT(nr_culled_tris);
 269       return TRUE;
 270    }
 271
 272    u_rect_find_intersection(&setup->draw_region, &bbox);
 273
 274    point = lp_setup_alloc_triangle(scene,
 275                                    setup->fs.nr_inputs,
 276                                    nr_planes,
 277                                    &bytes);
 278    if (!point)
 279       return FALSE;
 280
 281 #ifdef DEBUG
 282    point->v[0][0] = v0[0][0];
 283    point->v[0][1] = v0[0][1];
 284 #endif
 285
 286    info.v0 = v0;
 287    info.dx01 = 0;
 288    info.dx12 = fixed_width;
 289    info.dy01 = fixed_width;
 290    info.dy12 = 0;
 291
 292    /* Setup parameter interpolants:
 293     */
 294    setup_point_coefficients(setup, point, &info);
 295
 296    point->inputs.facing = 1.0F;
 297    point->inputs.state = setup->fs.stored;
 298    point->inputs.disable = FALSE;
 299    point->inputs.opaque = FALSE;
 300
 301    {
 302       point->plane[0].dcdx = -1;
 303       point->plane[0].dcdy = 0;
 304       point->plane[0].c = 1-bbox.x0;
 305       point->plane[0].ei = 0;
 306       point->plane[0].eo = 1;
 307
 308       point->plane[1].dcdx = 1;
 309       point->plane[1].dcdy = 0;
 310       point->plane[1].c = bbox.x1+1;
 311       point->plane[1].ei = -1;
 312       point->plane[1].eo = 0;
 313
 314       point->plane[2].dcdx = 0;
 315       point->plane[2].dcdy = 1;
 316       point->plane[2].c = 1-bbox.y0;
 317       point->plane[2].ei = 0;
 318       point->plane[2].eo = 1;
 319
 320       point->plane[3].dcdx = 0;
 321       point->plane[3].dcdy = -1;
 322       point->plane[3].c = bbox.y1+1;
 323       point->plane[3].ei = -1;
 324       point->plane[3].eo = 0;
 325    }
 326
 327    return lp_setup_bin_triangle(setup, point, &bbox, nr_planes);
 328 }
 329
 330
 331 static void
 332 lp_setup_point(struct lp_setup_context *setup,
 333                const float (*v0)[4])
 334 {
 335    if (!try_setup_point( setup, v0 ))
 336    {
 337       lp_setup_flush_and_restart(setup);
 338
 339       if (!try_setup_point( setup, v0 ))
 340          assert(0);
 341    }
 342 }
 343
 344
 345 void
 346 lp_setup_choose_point( struct lp_setup_context *setup )
 347 {
 348    setup->point = lp_setup_point;
 349 }
 350
 351