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 /**
  69  * Setup automatic texcoord coefficients (for sprite rendering).
  70  * \param slot  the vertex attribute slot to setup
  71  * \param i  the attribute channel in [0,3]
  72  * \param sprite_coord_origin  one of PIPE_SPRITE_COORD_x
  73  * \param perspective_proj  will the TEX instruction do a divide by Q?
  74  */
  75 static void
  76 texcoord_coef(struct lp_setup_context *setup,
  77               struct lp_rast_triangle *point,
  78               const struct point_info *info,
  79               unsigned slot,
  80               unsigned i,
  81               unsigned sprite_coord_origin,
  82               boolean perspective_proj)
  83 {
  84    assert(i < 4);
  85
  86    if (i == 0) {
  87       float dadx = FIXED_ONE / (float)info->dx12;
  88       float dady =  0.0f;
  89       float x0 = info->v0[0][0] - setup->pixel_offset;
  90       float y0 = info->v0[0][1] - setup->pixel_offset;
  91
  92       point->inputs.dadx[slot][0] = dadx;
  93       point->inputs.dady[slot][0] = dady;
  94       point->inputs.a0[slot][0] = 0.5 - (dadx * x0 + dady * y0);
  95
  96       if (!perspective_proj) {
  97          /* Divide coefficients by vertex.w here.
  98           *
  99           * It would be clearer to always multiply by w0 above and
 100           * then divide it out for perspective projection here, but
 101           * doing it this way involves less algebra.
 102           */
 103          float w0 = info->v0[0][3];
 104          point->inputs.dadx[slot][0] *= w0;
 105          point->inputs.dady[slot][0] *= w0;
 106          point->inputs.a0[slot][0] *= w0;
 107       }
 108    }
 109    else if (i == 1) {
 110       float dadx = 0.0f;
 111       float dady = FIXED_ONE / (float)info->dx12;
 112       float x0 = info->v0[0][0] - setup->pixel_offset;
 113       float y0 = info->v0[0][1] - setup->pixel_offset;
 114
 115       if (sprite_coord_origin == PIPE_SPRITE_COORD_LOWER_LEFT) {
 116          dady = -dady;
 117       }
 118
 119       point->inputs.dadx[slot][1] = dadx;
 120       point->inputs.dady[slot][1] = dady;
 121       point->inputs.a0[slot][1] = 0.5 - (dadx * x0 + dady * y0);
 122
 123       if (!perspective_proj) {
 124          float w0 = info->v0[0][3];
 125          point->inputs.dadx[slot][1] *= w0;
 126          point->inputs.dady[slot][1] *= w0;
 127          point->inputs.a0[slot][1] *= w0;
 128       }
 129    }
 130    else if (i == 2) {
 131       point->inputs.a0[slot][2] = 0.0f;
 132       point->inputs.dadx[slot][2] = 0.0f;
 133       point->inputs.dady[slot][2] = 0.0f;
 134    }
 135    else {
 136       point->inputs.a0[slot][3] = 1.0f;
 137       point->inputs.dadx[slot][3] = 0.0f;
 138       point->inputs.dady[slot][3] = 0.0f;
 139    }
 140 }
 141
 142
 143 /**
 144  * Special coefficient setup for gl_FragCoord.
 145  * X and Y are trivial
 146  * Z and W are copied from position_coef which should have already been computed.
 147  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
 148  */
 149 static void
 150 setup_point_fragcoord_coef(struct lp_setup_context *setup,
 151                            struct lp_rast_triangle *point,
 152                            const struct point_info *info,
 153                            unsigned slot,
 154                            unsigned usage_mask)
 155 {
 156    /*X*/
 157    if (usage_mask & TGSI_WRITEMASK_X) {
 158       point->inputs.a0[slot][0] = 0.0;
 159       point->inputs.dadx[slot][0] = 1.0;
 160       point->inputs.dady[slot][0] = 0.0;
 161    }
 162
 163    /*Y*/
 164    if (usage_mask & TGSI_WRITEMASK_Y) {
 165       point->inputs.a0[slot][1] = 0.0;
 166       point->inputs.dadx[slot][1] = 0.0;
 167       point->inputs.dady[slot][1] = 1.0;
 168    }
 169
 170    /*Z*/
 171    if (usage_mask & TGSI_WRITEMASK_Z) {
 172       constant_coef(setup, point, slot, info->v0[0][2], 2);
 173    }
 174
 175    /*W*/
 176    if (usage_mask & TGSI_WRITEMASK_W) {
 177       constant_coef(setup, point, slot, info->v0[0][3], 3);
 178    }
 179 }
 180
 181
 182 /**
 183  * Compute the point->coef[] array dadx, dady, a0 values.
 184  */
 185 static void
 186 setup_point_coefficients( struct lp_setup_context *setup,
 187                           struct lp_rast_triangle *point,
 188                           const struct point_info *info)
 189 {
 190    const struct lp_fragment_shader *shader = setup->fs.current.variant->shader;
 191    unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
 192    unsigned slot;
 193
 194    /* setup interpolation for all the remaining attributes:
 195     */
 196    for (slot = 0; slot < setup->fs.nr_inputs; slot++) {
 197       unsigned vert_attr = setup->fs.input[slot].src_index;
 198       unsigned usage_mask = setup->fs.input[slot].usage_mask;
 199       unsigned i;
 200
 201       switch (setup->fs.input[slot].interp) {
 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_LINEAR:
 212          /* Sprite tex coords may use linear interpolation someday */
 213          /* fall-through */
 214
 215       case LP_INTERP_PERSPECTIVE:
 216          /* check if the sprite coord flag is set for this attribute.
 217           * If so, set it up so it up so x and y vary from 0 to 1.
 218           */
 219          if (shader->info.input_semantic_name[slot] == TGSI_SEMANTIC_GENERIC) {
 220             const int index = shader->info.input_semantic_index[slot];
 221             /* Note that sprite_coord enable is a bitfield of
 222              * PIPE_MAX_SHADER_OUTPUTS bits.
 223              */
 224             if (index < PIPE_MAX_SHADER_OUTPUTS &&
 225                 (setup->sprite_coord_enable & (1 << index))) {
 226                for (i = 0; i < NUM_CHANNELS; i++)
 227                   if (usage_mask & (1 << i))
 228                      texcoord_coef(setup, point, info, slot + 1, i,
 229                                    setup->sprite_coord_origin,
 230                                    (usage_mask & TGSI_WRITEMASK_W));
 231                fragcoord_usage_mask |= TGSI_WRITEMASK_W;
 232                break;
 233             }
 234          }
 235
 236          /* Otherwise fallthrough */
 237       default:
 238          for (i = 0; i < NUM_CHANNELS; i++) {
 239             if (usage_mask & (1 << i))
 240                constant_coef(setup, point, slot+1, info->v0[vert_attr][i], i);
 241          }
 242       }
 243    }
 244
 245    /* The internal position input is in slot zero:
 246     */
 247    setup_point_fragcoord_coef(setup, point, info, 0,
 248                               fragcoord_usage_mask);
 249 }
 250
 251
 252 static INLINE int
 253 subpixel_snap(float a)
 254 {
 255    return util_iround(FIXED_ONE * a);
 256 }
 257
 258
 259 static boolean
 260 try_setup_point( struct lp_setup_context *setup,
 261                  const float (*v0)[4] )
 262 {
 263    /* x/y positions in fixed point */
 264    const int sizeAttr = setup->psize;
 265    const float size
 266       = (setup->point_size_per_vertex && sizeAttr > 0) ? v0[sizeAttr][0]
 267       : setup->point_size;
 268
 269    /* Point size as fixed point integer, remove rounding errors
 270     * and gives minimum width for very small points
 271     */
 272    int fixed_width = MAX2(FIXED_ONE,
 273                           (subpixel_snap(size) + FIXED_ONE/2 - 1) & ~(FIXED_ONE-1));
 274
 275    const int x0 = subpixel_snap(v0[0][0] - setup->pixel_offset) - fixed_width/2;
 276    const int y0 = subpixel_snap(v0[0][1] - setup->pixel_offset) - fixed_width/2;
 277
 278    struct lp_scene *scene = setup->scene;
 279    struct lp_rast_triangle *point;
 280    unsigned bytes;
 281    struct u_rect bbox;
 282    unsigned nr_planes = 4;
 283    struct point_info info;
 284
 285
 286    /* Bounding rectangle (in pixels) */
 287    {
 288       /* Yes this is necessary to accurately calculate bounding boxes
 289        * with the two fill-conventions we support.  GL (normally) ends
 290        * up needing a bottom-left fill convention, which requires
 291        * slightly different rounding.
 292        */
 293       int adj = (setup->pixel_offset != 0) ? 1 : 0;
 294
 295       bbox.x0 = (x0 + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 296       bbox.x1 = (x0 + fixed_width + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 297       bbox.y0 = (y0 + (FIXED_ONE-1)) >> FIXED_ORDER;
 298       bbox.y1 = (y0 + fixed_width + (FIXED_ONE-1)) >> FIXED_ORDER;
 299
 300       /* Inclusive coordinates:
 301        */
 302       bbox.x1--;
 303       bbox.y1--;
 304    }
 305
 306    if (!u_rect_test_intersection(&setup->draw_region, &bbox)) {
 307       if (0) debug_printf("offscreen\n");
 308       LP_COUNT(nr_culled_tris);
 309       return TRUE;
 310    }
 311
 312    u_rect_find_intersection(&setup->draw_region, &bbox);
 313
 314    point = lp_setup_alloc_triangle(scene,
 315                                    setup->fs.nr_inputs,
 316                                    nr_planes,
 317                                    &bytes);
 318    if (!point)
 319       return FALSE;
 320
 321 #ifdef DEBUG
 322    point->v[0][0] = v0[0][0];
 323    point->v[0][1] = v0[0][1];
 324 #endif
 325
 326    info.v0 = v0;
 327    info.dx01 = 0;
 328    info.dx12 = fixed_width;
 329    info.dy01 = fixed_width;
 330    info.dy12 = 0;
 331
 332    /* Setup parameter interpolants:
 333     */
 334    setup_point_coefficients(setup, point, &info);
 335
 336    point->inputs.facing = 1.0F;
 337    point->inputs.state = setup->fs.stored;
 338    point->inputs.disable = FALSE;
 339    point->inputs.opaque = FALSE;
 340
 341    {
 342       point->plane[0].dcdx = -1;
 343       point->plane[0].dcdy = 0;
 344       point->plane[0].c = 1-bbox.x0;
 345       point->plane[0].ei = 0;
 346       point->plane[0].eo = 1;
 347
 348       point->plane[1].dcdx = 1;
 349       point->plane[1].dcdy = 0;
 350       point->plane[1].c = bbox.x1+1;
 351       point->plane[1].ei = -1;
 352       point->plane[1].eo = 0;
 353
 354       point->plane[2].dcdx = 0;
 355       point->plane[2].dcdy = 1;
 356       point->plane[2].c = 1-bbox.y0;
 357       point->plane[2].ei = 0;
 358       point->plane[2].eo = 1;
 359
 360       point->plane[3].dcdx = 0;
 361       point->plane[3].dcdy = -1;
 362       point->plane[3].c = bbox.y1+1;
 363       point->plane[3].ei = -1;
 364       point->plane[3].eo = 0;
 365    }
 366
 367    return lp_setup_bin_triangle(setup, point, &bbox, nr_planes);
 368 }
 369
 370
 371 static void
 372 lp_setup_point(struct lp_setup_context *setup,
 373                const float (*v0)[4])
 374 {
 375    if (!try_setup_point( setup, v0 ))
 376    {
 377       lp_setup_flush_and_restart(setup);
 378
 379       if (!try_setup_point( setup, v0 ))
 380          assert(0);
 381    }
 382 }
 383
 384
 385 void
 386 lp_setup_choose_point( struct lp_setup_context *setup )
 387 {
 388    setup->point = lp_setup_point;
 389 }
 390
 391