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 44struct 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 */ 63static 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 */ 79static 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 */ 109static 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 132static void 133setup_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 */ 166static 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 236static 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 */ 245static void 246print_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 265static inline boolean sign(float x){ 266 return x >= 0; 267} 268 269 270/* Used on positive floats only: 271 */ 272static inline float fracf(float f) 273{ 274 return f - floorf(f); 275} 276 277 278 279static boolean 280try_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 const struct u_rect *scissor; 292 struct u_rect bbox, bboxpos; 293 boolean s_planes[4]; 294 unsigned tri_bytes; 295 int x[4]; 296 int y[4]; 297 int i; 298 int nr_planes = 4; 299 unsigned viewport_index = 0; 300 unsigned layer = 0; 301 302 /* linewidth should be interpreted as integer */ 303 int fixed_width = util_iround(width) * FIXED_ONE; 304 305 float x_offset=0; 306 float y_offset=0; 307 float x_offset_end=0; 308 float y_offset_end=0; 309 310 float x1diff; 311 float y1diff; 312 float x2diff; 313 float y2diff; 314 float dx, dy; 315 float area; 316 const float (*pv)[4]; 317 318 boolean draw_start; 319 boolean draw_end; 320 boolean will_draw_start; 321 boolean will_draw_end; 322 323 if (0) 324 print_line(setup, v1, v2); 325 326 if (setup->flatshade_first) { 327 pv = v1; 328 } 329 else { 330 pv = v2; 331 } 332 if (setup->viewport_index_slot > 0) { 333 unsigned *udata = (unsigned*)pv[setup->viewport_index_slot]; 334 viewport_index = lp_clamp_viewport_idx(*udata); 335 } 336 if (setup->layer_slot > 0) { 337 layer = *(unsigned*)pv[setup->layer_slot]; 338 layer = MIN2(layer, scene->fb_max_layer); 339 } 340 341 dx = v1[0][0] - v2[0][0]; 342 dy = v1[0][1] - v2[0][1]; 343 area = (dx * dx + dy * dy); 344 if (area == 0) { 345 LP_COUNT(nr_culled_tris); 346 return TRUE; 347 } 348 349 info.oneoverarea = 1.0f / area; 350 info.dx = dx; 351 info.dy = dy; 352 info.v1 = v1; 353 info.v2 = v2; 354 355 356 /* X-MAJOR LINE */ 357 if (fabsf(dx) >= fabsf(dy)) { 358 float dydx = dy / dx; 359 360 x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5; 361 y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5; 362 x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5; 363 y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5; 364 365 if (y2diff==-0.5 && dy<0){ 366 y2diff = 0.5; 367 } 368 369 /* 370 * Diamond exit rule test for starting point 371 */ 372 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) { 373 draw_start = TRUE; 374 } 375 else if (sign(x1diff) == sign(-dx)) { 376 draw_start = FALSE; 377 } 378 else if (sign(-y1diff) != sign(dy)) { 379 draw_start = TRUE; 380 } 381 else { 382 /* do intersection test */ 383 float yintersect = fracf(v1[0][1]) + x1diff * dydx; 384 draw_start = (yintersect < 1.0 && yintersect > 0.0); 385 } 386 387 388 /* 389 * Diamond exit rule test for ending point 390 */ 391 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) { 392 draw_end = FALSE; 393 } 394 else if (sign(x2diff) != sign(-dx)) { 395 draw_end = FALSE; 396 } 397 else if (sign(-y2diff) == sign(dy)) { 398 draw_end = TRUE; 399 } 400 else { 401 /* do intersection test */ 402 float yintersect = fracf(v2[0][1]) + x2diff * dydx; 403 draw_end = (yintersect < 1.0 && yintersect > 0.0); 404 } 405 406 /* Are we already drawing start/end? 407 */ 408 will_draw_start = sign(-x1diff) != sign(dx); 409 will_draw_end = (sign(x2diff) == sign(-dx)) || x2diff==0; 410 411 if (dx < 0) { 412 /* if v2 is to the right of v1, swap pointers */ 413 const float (*temp)[4] = v1; 414 v1 = v2; 415 v2 = temp; 416 dx = -dx; 417 dy = -dy; 418 /* Otherwise shift planes appropriately */ 419 if (will_draw_start != draw_start) { 420 x_offset_end = - x1diff - 0.5; 421 y_offset_end = x_offset_end * dydx; 422 423 } 424 if (will_draw_end != draw_end) { 425 x_offset = - x2diff - 0.5; 426 y_offset = x_offset * dydx; 427 } 428 429 } 430 else{ 431 /* Otherwise shift planes appropriately */ 432 if (will_draw_start != draw_start) { 433 x_offset = - x1diff + 0.5; 434 y_offset = x_offset * dydx; 435 } 436 if (will_draw_end != draw_end) { 437 x_offset_end = - x2diff + 0.5; 438 y_offset_end = x_offset_end * dydx; 439 } 440 } 441 442 /* x/y positions in fixed point */ 443 x[0] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset); 444 x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset); 445 x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset); 446 x[3] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset); 447 448 y[0] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset) - fixed_width/2; 449 y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) - fixed_width/2; 450 y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset) + fixed_width/2; 451 y[3] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset) + fixed_width/2; 452 453 } 454 else { 455 const float dxdy = dx / dy; 456 457 /* Y-MAJOR LINE */ 458 x1diff = v1[0][0] - (float) floor(v1[0][0]) - 0.5; 459 y1diff = v1[0][1] - (float) floor(v1[0][1]) - 0.5; 460 x2diff = v2[0][0] - (float) floor(v2[0][0]) - 0.5; 461 y2diff = v2[0][1] - (float) floor(v2[0][1]) - 0.5; 462 463 if (x2diff==-0.5 && dx<0) { 464 x2diff = 0.5; 465 } 466 467 /* 468 * Diamond exit rule test for starting point 469 */ 470 if (fabsf(x1diff) + fabsf(y1diff) < 0.5) { 471 draw_start = TRUE; 472 } 473 else if (sign(-y1diff) == sign(dy)) { 474 draw_start = FALSE; 475 } 476 else if (sign(x1diff) != sign(-dx)) { 477 draw_start = TRUE; 478 } 479 else { 480 /* do intersection test */ 481 float xintersect = fracf(v1[0][0]) + y1diff * dxdy; 482 draw_start = (xintersect < 1.0 && xintersect > 0.0); 483 } 484 485 /* 486 * Diamond exit rule test for ending point 487 */ 488 if (fabsf(x2diff) + fabsf(y2diff) < 0.5) { 489 draw_end = FALSE; 490 } 491 else if (sign(-y2diff) != sign(dy) ) { 492 draw_end = FALSE; 493 } 494 else if (sign(x2diff) == sign(-dx) ) { 495 draw_end = TRUE; 496 } 497 else { 498 /* do intersection test */ 499 float xintersect = fracf(v2[0][0]) + y2diff * dxdy; 500 draw_end = (xintersect < 1.0 && xintersect >= 0.0); 501 } 502 503 /* Are we already drawing start/end? 504 */ 505 will_draw_start = sign(y1diff) == sign(dy); 506 will_draw_end = (sign(-y2diff) == sign(dy)) || y2diff==0; 507 508 if (dy > 0) { 509 /* if v2 is on top of v1, swap pointers */ 510 const float (*temp)[4] = v1; 511 v1 = v2; 512 v2 = temp; 513 dx = -dx; 514 dy = -dy; 515 516 /* Otherwise shift planes appropriately */ 517 if (will_draw_start != draw_start) { 518 y_offset_end = - y1diff + 0.5; 519 x_offset_end = y_offset_end * dxdy; 520 } 521 if (will_draw_end != draw_end) { 522 y_offset = - y2diff + 0.5; 523 x_offset = y_offset * dxdy; 524 } 525 } 526 else { 527 /* Otherwise shift planes appropriately */ 528 if (will_draw_start != draw_start) { 529 y_offset = - y1diff - 0.5; 530 x_offset = y_offset * dxdy; 531 532 } 533 if (will_draw_end != draw_end) { 534 y_offset_end = - y2diff - 0.5; 535 x_offset_end = y_offset_end * dxdy; 536 } 537 } 538 539 /* x/y positions in fixed point */ 540 x[0] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset) - fixed_width/2; 541 x[1] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) - fixed_width/2; 542 x[2] = subpixel_snap(v2[0][0] + x_offset_end - setup->pixel_offset) + fixed_width/2; 543 x[3] = subpixel_snap(v1[0][0] + x_offset - setup->pixel_offset) + fixed_width/2; 544 545 y[0] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset); 546 y[1] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset); 547 y[2] = subpixel_snap(v2[0][1] + y_offset_end - setup->pixel_offset); 548 y[3] = subpixel_snap(v1[0][1] + y_offset - setup->pixel_offset); 549 } 550 551 /* Bounding rectangle (in pixels) */ 552 { 553 /* Yes this is necessary to accurately calculate bounding boxes 554 * with the two fill-conventions we support. GL (normally) ends 555 * up needing a bottom-left fill convention, which requires 556 * slightly different rounding. 557 */ 558 int adj = (setup->bottom_edge_rule != 0) ? 1 : 0; 559 560 bbox.x0 = (MIN4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER; 561 bbox.x1 = (MAX4(x[0], x[1], x[2], x[3]) + (FIXED_ONE-1)) >> FIXED_ORDER; 562 bbox.y0 = (MIN4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER; 563 bbox.y1 = (MAX4(y[0], y[1], y[2], y[3]) + (FIXED_ONE-1) + adj) >> FIXED_ORDER; 564 565 /* Inclusive coordinates: 566 */ 567 bbox.x1--; 568 bbox.y1--; 569 } 570 571 if (bbox.x1 < bbox.x0 || 572 bbox.y1 < bbox.y0) { 573 if (0) debug_printf("empty bounding box\n"); 574 LP_COUNT(nr_culled_tris); 575 return TRUE; 576 } 577 578 if (!u_rect_test_intersection(&setup->draw_regions[viewport_index], &bbox)) { 579 if (0) debug_printf("offscreen\n"); 580 LP_COUNT(nr_culled_tris); 581 return TRUE; 582 } 583 584 bboxpos = bbox; 585 586 /* Can safely discard negative regions: 587 */ 588 bboxpos.x0 = MAX2(bboxpos.x0, 0); 589 bboxpos.y0 = MAX2(bboxpos.y0, 0); 590 591 nr_planes = 4; 592 /* 593 * Determine how many scissor planes we need, that is drop scissor 594 * edges if the bounding box of the tri is fully inside that edge. 595 */ 596 if (setup->scissor_test) { 597 /* why not just use draw_regions */ 598 scissor = &setup->scissors[viewport_index]; 599 scissor_planes_needed(s_planes, &bboxpos, scissor); 600 nr_planes += s_planes[0] + s_planes[1] + s_planes[2] + s_planes[3]; 601 } 602 603 line = lp_setup_alloc_triangle(scene, 604 key->num_inputs, 605 nr_planes, 606 &tri_bytes); 607 if (!line) 608 return FALSE; 609 610#ifdef DEBUG 611 line->v[0][0] = v1[0][0]; 612 line->v[1][0] = v2[0][0]; 613 line->v[0][1] = v1[0][1]; 614 line->v[1][1] = v2[0][1]; 615#endif 616 617 LP_COUNT(nr_tris); 618 619 if (lp_context->active_statistics_queries) { 620 lp_context->pipeline_statistics.c_primitives++; 621 } 622 623 /* calculate the deltas */ 624 plane = GET_PLANES(line); 625 plane[0].dcdy = x[0] - x[1]; 626 plane[1].dcdy = x[1] - x[2]; 627 plane[2].dcdy = x[2] - x[3]; 628 plane[3].dcdy = x[3] - x[0]; 629 630 plane[0].dcdx = y[0] - y[1]; 631 plane[1].dcdx = y[1] - y[2]; 632 plane[2].dcdx = y[2] - y[3]; 633 plane[3].dcdx = y[3] - y[0]; 634 635 if (draw_will_inject_frontface(lp_context->draw) && 636 setup->face_slot > 0) { 637 line->inputs.frontfacing = v1[setup->face_slot][0]; 638 } else { 639 line->inputs.frontfacing = TRUE; 640 } 641 642 /* Setup parameter interpolants: 643 */ 644 info.a0 = GET_A0(&line->inputs); 645 info.dadx = GET_DADX(&line->inputs); 646 info.dady = GET_DADY(&line->inputs); 647 info.frontfacing = line->inputs.frontfacing; 648 setup_line_coefficients(setup, &info); 649 650 line->inputs.disable = FALSE; 651 line->inputs.opaque = FALSE; 652 line->inputs.layer = layer; 653 line->inputs.viewport_index = viewport_index; 654 655 /* 656 * XXX: this code is mostly identical to the one in lp_setup_tri, except it 657 * uses 4 planes instead of 3. Could share the code (including the sse 658 * assembly, in fact we'd get the 4th plane for free). 659 * The only difference apart from storing the 4th plane would be some 660 * different shuffle for calculating dcdx/dcdy. 661 */ 662 for (i = 0; i < 4; i++) { 663 664 /* half-edge constants, will be iterated over the whole render 665 * target. 666 */ 667 plane[i].c = IMUL64(plane[i].dcdx, x[i]) - IMUL64(plane[i].dcdy, y[i]); 668 669 /* correct for top-left vs. bottom-left fill convention. 670 */ 671 if (plane[i].dcdx < 0) { 672 /* both fill conventions want this - adjust for left edges */ 673 plane[i].c++; 674 } 675 else if (plane[i].dcdx == 0) { 676 if (setup->pixel_offset == 0) { 677 /* correct for top-left fill convention: 678 */ 679 if (plane[i].dcdy > 0) plane[i].c++; 680 } 681 else { 682 /* correct for bottom-left fill convention: 683 */ 684 if (plane[i].dcdy < 0) plane[i].c++; 685 } 686 } 687 688 plane[i].dcdx *= FIXED_ONE; 689 plane[i].dcdy *= FIXED_ONE; 690 691 /* find trivial reject offsets for each edge for a single-pixel 692 * sized block. These will be scaled up at each recursive level to 693 * match the active blocksize. Scaling in this way works best if 694 * the blocks are square. 695 */ 696 plane[i].eo = 0; 697 if (plane[i].dcdx < 0) plane[i].eo -= plane[i].dcdx; 698 if (plane[i].dcdy > 0) plane[i].eo += plane[i].dcdy; 699 } 700 701 702 /* 703 * When rasterizing scissored tris, use the intersection of the 704 * triangle bounding box and the scissor rect to generate the 705 * scissor planes. 706 * 707 * This permits us to cut off the triangle "tails" that are present 708 * in the intermediate recursive levels caused when two of the 709 * triangles edges don't diverge quickly enough to trivially reject 710 * exterior blocks from the triangle. 711 * 712 * It's not really clear if it's worth worrying about these tails, 713 * but since we generate the planes for each scissored tri, it's 714 * free to trim them in this case. 715 * 716 * Note that otherwise, the scissor planes only vary in 'C' value, 717 * and even then only on state-changes. Could alternatively store 718 * these planes elsewhere. 719 * (Or only store the c value together with a bit indicating which 720 * scissor edge this is, so rasterization would treat them differently 721 * (easier to evaluate) to ordinary planes.) 722 */ 723 if (nr_planes > 4) { 724 struct lp_rast_plane *plane_s = &plane[4]; 725 726 if (s_planes[0]) { 727 plane_s->dcdx = ~0U << 8; 728 plane_s->dcdy = 0; 729 plane_s->c = (1-scissor->x0) << 8; 730 plane_s->eo = 1 << 8; 731 plane_s++; 732 } 733 if (s_planes[1]) { 734 plane_s->dcdx = 1 << 8; 735 plane_s->dcdy = 0; 736 plane_s->c = (scissor->x1+1) << 8; 737 plane_s->eo = 0 << 8; 738 plane_s++; 739 } 740 if (s_planes[2]) { 741 plane_s->dcdx = 0; 742 plane_s->dcdy = 1 << 8; 743 plane_s->c = (1-scissor->y0) << 8; 744 plane_s->eo = 1 << 8; 745 plane_s++; 746 } 747 if (s_planes[3]) { 748 plane_s->dcdx = 0; 749 plane_s->dcdy = ~0U << 8; 750 plane_s->c = (scissor->y1+1) << 8; 751 plane_s->eo = 0; 752 plane_s++; 753 } 754 assert(plane_s == &plane[nr_planes]); 755 } 756 757 return lp_setup_bin_triangle(setup, line, &bbox, &bboxpos, nr_planes, viewport_index); 758} 759 760 761static void lp_setup_line_discard(struct lp_setup_context *setup, 762 const float (*v0)[4], 763 const float (*v1)[4]) 764{ 765} 766 767static void lp_setup_line(struct lp_setup_context *setup, 768 const float (*v0)[4], 769 const float (*v1)[4]) 770{ 771 if (!try_setup_line(setup, v0, v1)) { 772 if (!lp_setup_flush_and_restart(setup)) 773 return; 774 775 if (!try_setup_line(setup, v0, v1)) 776 return; 777 } 778} 779 780 781void lp_setup_choose_line(struct lp_setup_context *setup) 782{ 783 if (setup->rasterizer_discard) { 784 setup->line = lp_setup_line_discard; 785 } else { 786 setup->line = lp_setup_line; 787 } 788} 789 790 791