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 "util/u_math.h"
  33 #include "util/u_memory.h"
  34 #include "lp_setup_context.h"
  35 #include "lp_perf.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 "tgsi/tgsi_scan.h"
  41 #include "draw/draw_context.h"
  42
  43 #define NUM_CHANNELS 4
  44
  45 struct point_info {
  46    /* x,y deltas */
  47    int dy01, dy12;
  48    int dx01, dx12;
  49
  50    const float (*v0)[4];
  51
  52    float (*a0)[4];
  53    float (*dadx)[4];
  54    float (*dady)[4];
  55
  56    boolean frontfacing;
  57 };
  58
  59
  60 /**
  61  * Compute a0 for a constant-valued coefficient (GL_FLAT shading).
  62  */
  63 static void
  64 constant_coef(struct lp_setup_context *setup,
  65               struct point_info *info,
  66               unsigned slot,
  67               const float value,
  68               unsigned i)
  69 {
  70    info->a0[slot][i] = value;
  71    info->dadx[slot][i] = 0.0f;
  72    info->dady[slot][i] = 0.0f;
  73 }
  74
  75
  76 static void
  77 point_persp_coeff(struct lp_setup_context *setup,
  78                   const struct point_info *info,
  79                   unsigned slot,
  80                   unsigned i)
  81 {
  82    /*
  83     * Fragment shader expects pre-multiplied w for LP_INTERP_PERSPECTIVE. A
  84     * better stratergy would be to take the primitive in consideration when
  85     * generating the fragment shader key, and therefore avoid the per-fragment
  86     * perspective divide.
  87     */
  88
  89    float w0 = info->v0[0][3];
  90
  91    assert(i < 4);
  92
  93    info->a0[slot][i] = info->v0[slot][i]*w0;
  94    info->dadx[slot][i] = 0.0f;
  95    info->dady[slot][i] = 0.0f;
  96 }
  97
  98
  99 /**
 100  * Setup automatic texcoord coefficients (for sprite rendering).
 101  * \param slot  the vertex attribute slot to setup
 102  * \param i  the attribute channel in [0,3]
 103  * \param sprite_coord_origin  one of PIPE_SPRITE_COORD_x
 104  * \param perspective  does the shader expects pre-multiplied w, i.e.,
 105  *    LP_INTERP_PERSPECTIVE is specified in the shader key
 106  */
 107 static void
 108 texcoord_coef(struct lp_setup_context *setup,
 109               const struct point_info *info,
 110               unsigned slot,
 111               unsigned i,
 112               unsigned sprite_coord_origin,
 113               boolean perspective)
 114 {
 115    float w0 = info->v0[0][3];
 116
 117    assert(i < 4);
 118
 119    if (i == 0) {
 120       float dadx = FIXED_ONE / (float)info->dx12;
 121       float dady =  0.0f;
 122       float x0 = info->v0[0][0] - setup->pixel_offset;
 123       float y0 = info->v0[0][1] - setup->pixel_offset;
 124
 125       info->dadx[slot][0] = dadx;
 126       info->dady[slot][0] = dady;
 127       info->a0[slot][0] = 0.5 - (dadx * x0 + dady * y0);
 128
 129       if (perspective) {
 130          info->dadx[slot][0] *= w0;
 131          info->dady[slot][0] *= w0;
 132          info->a0[slot][0] *= w0;
 133       }
 134    }
 135    else if (i == 1) {
 136       float dadx = 0.0f;
 137       float dady = FIXED_ONE / (float)info->dx12;
 138       float x0 = info->v0[0][0] - setup->pixel_offset;
 139       float y0 = info->v0[0][1] - setup->pixel_offset;
 140
 141       if (sprite_coord_origin == PIPE_SPRITE_COORD_LOWER_LEFT) {
 142          dady = -dady;
 143       }
 144
 145       info->dadx[slot][1] = dadx;
 146       info->dady[slot][1] = dady;
 147       info->a0[slot][1] = 0.5 - (dadx * x0 + dady * y0);
 148
 149       if (perspective) {
 150          info->dadx[slot][1] *= w0;
 151          info->dady[slot][1] *= w0;
 152          info->a0[slot][1] *= w0;
 153       }
 154    }
 155    else if (i == 2) {
 156       info->a0[slot][2] = 0.0f;
 157       info->dadx[slot][2] = 0.0f;
 158       info->dady[slot][2] = 0.0f;
 159    }
 160    else {
 161       info->a0[slot][3] = perspective ? w0 : 1.0f;
 162       info->dadx[slot][3] = 0.0f;
 163       info->dady[slot][3] = 0.0f;
 164    }
 165 }
 166
 167
 168 /**
 169  * Special coefficient setup for gl_FragCoord.
 170  * X and Y are trivial
 171  * Z and W are copied from position_coef which should have already been computed.
 172  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
 173  */
 174 static void
 175 setup_point_fragcoord_coef(struct lp_setup_context *setup,
 176                            struct point_info *info,
 177                            unsigned slot,
 178                            unsigned usage_mask)
 179 {
 180    /*X*/
 181    if (usage_mask & TGSI_WRITEMASK_X) {
 182       info->a0[slot][0] = 0.0;
 183       info->dadx[slot][0] = 1.0;
 184       info->dady[slot][0] = 0.0;
 185    }
 186
 187    /*Y*/
 188    if (usage_mask & TGSI_WRITEMASK_Y) {
 189       info->a0[slot][1] = 0.0;
 190       info->dadx[slot][1] = 0.0;
 191       info->dady[slot][1] = 1.0;
 192    }
 193
 194    /*Z*/
 195    if (usage_mask & TGSI_WRITEMASK_Z) {
 196       constant_coef(setup, info, slot, info->v0[0][2], 2);
 197    }
 198
 199    /*W*/
 200    if (usage_mask & TGSI_WRITEMASK_W) {
 201       constant_coef(setup, info, slot, info->v0[0][3], 3);
 202    }
 203 }
 204
 205
 206 /**
 207  * Compute the point->coef[] array dadx, dady, a0 values.
 208  */
 209 static void
 210 setup_point_coefficients( struct lp_setup_context *setup,
 211                           struct point_info *info)
 212 {
 213    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
 214    const struct lp_fragment_shader *shader = setup->fs.current.variant->shader;
 215    unsigned fragcoord_usage_mask = TGSI_WRITEMASK_XYZ;
 216    unsigned slot;
 217
 218    /* setup interpolation for all the remaining attributes:
 219     */
 220    for (slot = 0; slot < key->num_inputs; slot++) {
 221       unsigned vert_attr = key->inputs[slot].src_index;
 222       unsigned usage_mask = key->inputs[slot].usage_mask;
 223       enum lp_interp interp = key->inputs[slot].interp;
 224       boolean perspective = !!(interp == LP_INTERP_PERSPECTIVE);
 225       unsigned i;
 226
 227       if (perspective & usage_mask) {
 228          fragcoord_usage_mask |= TGSI_WRITEMASK_W;
 229       }
 230
 231       switch (interp) {
 232       case LP_INTERP_POSITION:
 233          /*
 234           * The generated pixel interpolators will pick up the coeffs from
 235           * slot 0, so all need to ensure that the usage mask is covers all
 236           * usages.
 237           */
 238          fragcoord_usage_mask |= usage_mask;
 239          break;
 240
 241       case LP_INTERP_LINEAR:
 242          /* Sprite tex coords may use linear interpolation someday */
 243          /* fall-through */
 244       case LP_INTERP_PERSPECTIVE:
 245          /* check if the sprite coord flag is set for this attribute.
 246           * If so, set it up so it up so x and y vary from 0 to 1.
 247           */
 248          if (shader->info.base.input_semantic_name[slot] == TGSI_SEMANTIC_GENERIC) {
 249             unsigned semantic_index = shader->info.base.input_semantic_index[slot];
 250             /* Note that sprite_coord enable is a bitfield of
 251              * PIPE_MAX_SHADER_OUTPUTS bits.
 252              */
 253             if (semantic_index < PIPE_MAX_SHADER_OUTPUTS &&
 254                 (setup->sprite_coord_enable & (1 << semantic_index))) {
 255                for (i = 0; i < NUM_CHANNELS; i++) {
 256                   if (usage_mask & (1 << i)) {
 257                      texcoord_coef(setup, info, slot + 1, i,
 258                                    setup->sprite_coord_origin,
 259                                    perspective);
 260                   }
 261                }
 262                break;
 263             }
 264          }
 265          /* fall-through */
 266       case LP_INTERP_CONSTANT:
 267          for (i = 0; i < NUM_CHANNELS; i++) {
 268             if (usage_mask & (1 << i)) {
 269                if (perspective) {
 270                   point_persp_coeff(setup, info, slot+1, i);
 271                }
 272                else {
 273                   constant_coef(setup, info, slot+1, info->v0[vert_attr][i], i);
 274                }
 275             }
 276          }
 277          break;
 278
 279       case LP_INTERP_FACING:
 280          for (i = 0; i < NUM_CHANNELS; i++)
 281             if (usage_mask & (1 << i))
 282                constant_coef(setup, info, slot+1,
 283                              info->frontfacing ? 1.0f : -1.0f, i);
 284          break;
 285
 286       default:
 287          assert(0);
 288          break;
 289       }
 290    }
 291
 292    /* The internal position input is in slot zero:
 293     */
 294    setup_point_fragcoord_coef(setup, info, 0,
 295                               fragcoord_usage_mask);
 296 }
 297
 298
 299 static inline int
 300 subpixel_snap(float a)
 301 {
 302    return util_iround(FIXED_ONE * a);
 303 }
 304
 305 /**
 306  * Print point vertex attribs (for debug).
 307  */
 308 static void
 309 print_point(struct lp_setup_context *setup,
 310             const float (*v0)[4],
 311             const float size)
 312 {
 313    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
 314    uint i;
 315
 316    debug_printf("llvmpipe point, width %f\n", size);
 317    for (i = 0; i < 1 + key->num_inputs; i++) {
 318       debug_printf("  v0[%d]:  %f %f %f %f\n", i,
 319                    v0[i][0], v0[i][1], v0[i][2], v0[i][3]);
 320    }
 321 }
 322
 323
 324 static boolean
 325 try_setup_point( struct lp_setup_context *setup,
 326                  const float (*v0)[4] )
 327 {
 328    struct llvmpipe_context *lp_context = (struct llvmpipe_context *)setup->pipe;
 329    /* x/y positions in fixed point */
 330    const struct lp_setup_variant_key *key = &setup->setup.variant->key;
 331    const int sizeAttr = setup->psize_slot;
 332    const float size
 333       = (setup->point_size_per_vertex && sizeAttr > 0) ? v0[sizeAttr][0]
 334       : setup->point_size;
 335
 336    /* Yes this is necessary to accurately calculate bounding boxes
 337     * with the two fill-conventions we support.  GL (normally) ends
 338     * up needing a bottom-left fill convention, which requires
 339     * slightly different rounding.
 340     */
 341    int adj = (setup->bottom_edge_rule != 0) ? 1 : 0;
 342
 343    struct lp_scene *scene = setup->scene;
 344    struct lp_rast_triangle *point;
 345    unsigned bytes;
 346    struct u_rect bbox;
 347    unsigned nr_planes = 4;
 348    struct point_info info;
 349    unsigned viewport_index = 0;
 350    unsigned layer = 0;
 351    int fixed_width;
 352
 353    if (setup->viewport_index_slot > 0) {
 354       unsigned *udata = (unsigned*)v0[setup->viewport_index_slot];
 355       viewport_index = lp_clamp_viewport_idx(*udata);
 356    }
 357    if (setup->layer_slot > 0) {
 358       layer = *(unsigned*)v0[setup->layer_slot];
 359       layer = MIN2(layer, scene->fb_max_layer);
 360    }
 361
 362    if (0)
 363       print_point(setup, v0, size);
 364
 365    /* Bounding rectangle (in pixels) */
 366    if (!lp_context->rasterizer ||
 367        lp_context->rasterizer->point_quad_rasterization) {
 368       /*
 369        * Rasterize points as quads.
 370        */
 371       int x0, y0;
 372       /* Point size as fixed point integer, remove rounding errors
 373        * and gives minimum width for very small points.
 374        */
 375       fixed_width = MAX2(FIXED_ONE, subpixel_snap(size));
 376
 377       x0 = subpixel_snap(v0[0][0] - setup->pixel_offset) - fixed_width/2;
 378       y0 = subpixel_snap(v0[0][1] - setup->pixel_offset) - fixed_width/2;
 379
 380       bbox.x0 = (x0 + (FIXED_ONE-1)) >> FIXED_ORDER;
 381       bbox.x1 = (x0 + fixed_width + (FIXED_ONE-1)) >> FIXED_ORDER;
 382       bbox.y0 = (y0 + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 383       bbox.y1 = (y0 + fixed_width + (FIXED_ONE-1) + adj) >> FIXED_ORDER;
 384
 385       /* Inclusive coordinates:
 386        */
 387       bbox.x1--;
 388       bbox.y1--;
 389    } else {
 390       /*
 391        * OpenGL legacy rasterization rules for non-sprite points.
 392        *
 393        * Per OpenGL 2.1 spec, section 3.3.1, "Basic Point Rasterization".
 394        *
 395        * This type of point rasterization is only available in pre 3.0 contexts
 396        * (or compatibilility contexts which we don't support) anyway.
 397        */
 398
 399       const int x0 = subpixel_snap(v0[0][0]);
 400       const int y0 = subpixel_snap(v0[0][1]) - adj;
 401
 402       int int_width;
 403       /* Point size as fixed point integer. For GL legacy points
 404        * the point size is always a whole integer.
 405        */
 406       fixed_width = MAX2(FIXED_ONE,
 407                          (subpixel_snap(size) + FIXED_ONE/2 - 1) & ~(FIXED_ONE-1));
 408       int_width = fixed_width >> FIXED_ORDER;
 409
 410       assert(setup->pixel_offset != 0);
 411
 412       if (int_width == 1) {
 413          bbox.x0 = x0 >> FIXED_ORDER;
 414          bbox.y0 = y0 >> FIXED_ORDER;
 415          bbox.x1 = bbox.x0;
 416          bbox.y1 = bbox.y0;
 417       } else {
 418          if (int_width & 1) {
 419             /* Odd width */
 420             bbox.x0 = (x0 >> FIXED_ORDER) - (int_width - 1)/2;
 421             bbox.y0 = (y0 >> FIXED_ORDER) - (int_width - 1)/2;
 422          } else {
 423             /* Even width */
 424             bbox.x0 = ((x0 + FIXED_ONE/2) >> FIXED_ORDER) - int_width/2;
 425             bbox.y0 = ((y0 + FIXED_ONE/2) >> FIXED_ORDER) - int_width/2;
 426          }
 427
 428          bbox.x1 = bbox.x0 + int_width - 1;
 429          bbox.y1 = bbox.y0 + int_width - 1;
 430       }
 431    }
 432
 433    if (0) {
 434       debug_printf("  bbox: (%i, %i) - (%i, %i)\n",
 435                    bbox.x0, bbox.y0,
 436                    bbox.x1, bbox.y1);
 437    }
 438
 439    if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) {
 440       if (0) debug_printf("offscreen\n");
 441       LP_COUNT(nr_culled_tris);
 442       return TRUE;
 443    }
 444
 445    u_rect_find_intersection(&setup->draw_regions[viewport_index], &bbox);
 446
 447    point = lp_setup_alloc_triangle(scene,
 448                                    key->num_inputs,
 449                                    nr_planes,
 450                                    &bytes);
 451    if (!point)
 452       return FALSE;
 453
 454 #ifdef DEBUG
 455    point->v[0][0] = v0[0][0];
 456    point->v[0][1] = v0[0][1];
 457 #endif
 458
 459    LP_COUNT(nr_tris);
 460
 461    if (lp_context->active_statistics_queries &&
 462        !llvmpipe_rasterization_disabled(lp_context)) {
 463       lp_context->pipeline_statistics.c_primitives++;
 464    }
 465
 466    if (draw_will_inject_frontface(lp_context->draw) &&
 467        setup->face_slot > 0) {
 468       point->inputs.frontfacing = v0[setup->face_slot][0];
 469    } else {
 470       point->inputs.frontfacing = TRUE;
 471    }
 472
 473    info.v0 = v0;
 474    info.dx01 = 0;
 475    info.dx12 = fixed_width;
 476    info.dy01 = fixed_width;
 477    info.dy12 = 0;
 478    info.a0 = GET_A0(&point->inputs);
 479    info.dadx = GET_DADX(&point->inputs);
 480    info.dady = GET_DADY(&point->inputs);
 481    info.frontfacing = point->inputs.frontfacing;
 482
 483    /* Setup parameter interpolants:
 484     */
 485    setup_point_coefficients(setup, &info);
 486
 487    point->inputs.disable = FALSE;
 488    point->inputs.opaque = FALSE;
 489    point->inputs.layer = layer;
 490    point->inputs.viewport_index = viewport_index;
 491
 492    {
 493       struct lp_rast_plane *plane = GET_PLANES(point);
 494
 495       plane[0].dcdx = -1 << 8;
 496       plane[0].dcdy = 0;
 497       plane[0].c = (1-bbox.x0) << 8;
 498       plane[0].eo = 1 << 8;
 499
 500       plane[1].dcdx = 1 << 8;
 501       plane[1].dcdy = 0;
 502       plane[1].c = (bbox.x1+1) << 8;
 503       plane[1].eo = 0;
 504
 505       plane[2].dcdx = 0;
 506       plane[2].dcdy = 1 << 8;
 507       plane[2].c = (1-bbox.y0) << 8;
 508       plane[2].eo = 1 << 8;
 509
 510       plane[3].dcdx = 0;
 511       plane[3].dcdy = -1 << 8;
 512       plane[3].c = (bbox.y1+1) << 8;
 513       plane[3].eo = 0;
 514    }
 515
 516    return lp_setup_bin_triangle(setup, point, &bbox, nr_planes, viewport_index);
 517 }
 518
 519
 520 static void
 521 lp_setup_point(struct lp_setup_context *setup,
 522                const float (*v0)[4])
 523 {
 524    if (!try_setup_point( setup, v0 ))
 525    {
 526       if (!lp_setup_flush_and_restart(setup))
 527          return;
 528
 529       if (!try_setup_point( setup, v0 ))
 530          return;
 531    }
 532 }
 533
 534
 535 void
 536 lp_setup_choose_point( struct lp_setup_context *setup )
 537 {
 538    setup->point = lp_setup_point;
 539 }
 540
 541