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 * \brief Primitive rasterization/rendering (points, lines, triangles) 30 * 31 * \author Keith Whitwell <keithw (at) vmware.com> 32 * \author Brian Paul 33 */ 34 35 #include "sp_context.h" 36 #include "sp_quad.h" 37 #include "sp_quad_pipe.h" 38 #include "sp_setup.h" 39 #include "sp_state.h" 40 #include "draw/draw_context.h" 41 #include "pipe/p_shader_tokens.h" 42 #include "util/u_math.h" 43 #include "util/u_memory.h" 44 45 46 #define DEBUG_VERTS 0 47 #define DEBUG_FRAGS 0 48 49 50 /** 51 * Triangle edge info 52 */ 53 struct edge { 54 float dx; /**< X(v1) - X(v0), used only during setup */ 55 float dy; /**< Y(v1) - Y(v0), used only during setup */ 56 float dxdy; /**< dx/dy */ 57 float sx, sy; /**< first sample point coord */ 58 int lines; /**< number of lines on this edge */ 59 }; 60 61 62 /** 63 * Max number of quads (2x2 pixel blocks) to process per batch. 64 * This can't be arbitrarily increased since we depend on some 32-bit 65 * bitmasks (two bits per quad). 66 */ 67 #define MAX_QUADS 16 68 69 70 /** 71 * Triangle setup info. 72 * Also used for line drawing (taking some liberties). 73 */ 74 struct setup_context { 75 struct softpipe_context *softpipe; 76 77 /* Vertices are just an array of floats making up each attribute in 78 * turn. Currently fixed at 4 floats, but should change in time. 79 * Codegen will help cope with this. 80 */ 81 const float (*vmax)[4]; 82 const float (*vmid)[4]; 83 const float (*vmin)[4]; 84 const float (*vprovoke)[4]; 85 86 struct edge ebot; 87 struct edge etop; 88 struct edge emaj; 89 90 float oneoverarea; 91 int facing; 92 93 float pixel_offset; 94 unsigned max_layer; 95 96 struct quad_header quad[MAX_QUADS]; 97 struct quad_header *quad_ptrs[MAX_QUADS]; 98 unsigned count; 99 100 struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; 101 struct tgsi_interp_coef posCoef; /* For Z, W */ 102 103 struct { 104 int left[2]; /**< [0] = row0, [1] = row1 */ 105 int right[2]; 106 int y; 107 } span; 108 109 #if DEBUG_FRAGS 110 uint numFragsEmitted; /**< per primitive */ 111 uint numFragsWritten; /**< per primitive */ 112 #endif 113 114 unsigned cull_face; /* which faces cull */ 115 unsigned nr_vertex_attrs; 116 }; 117 118 119 120 121 122 123 124 /** 125 * Clip setup->quad against the scissor/surface bounds. 126 */ 127 static inline void 128 quad_clip(struct setup_context *setup, struct quad_header *quad) 129 { 130 unsigned viewport_index = quad[0].input.viewport_index; 131 const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect[viewport_index]; 132 const int minx = (int) cliprect->minx; 133 const int maxx = (int) cliprect->maxx; 134 const int miny = (int) cliprect->miny; 135 const int maxy = (int) cliprect->maxy; 136 137 if (quad->input.x0 >= maxx || 138 quad->input.y0 >= maxy || 139 quad->input.x0 + 1 < minx || 140 quad->input.y0 + 1 < miny) { 141 /* totally clipped */ 142 quad->inout.mask = 0x0; 143 return; 144 } 145 if (quad->input.x0 < minx) 146 quad->inout.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); 147 if (quad->input.y0 < miny) 148 quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); 149 if (quad->input.x0 == maxx - 1) 150 quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); 151 if (quad->input.y0 == maxy - 1) 152 quad->inout.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); 153 } 154 155 156 /** 157 * Emit a quad (pass to next stage) with clipping. 158 */ 159 static inline void 160 clip_emit_quad(struct setup_context *setup, struct quad_header *quad) 161 { 162 quad_clip(setup, quad); 163 164 if (quad->inout.mask) { 165 struct softpipe_context *sp = setup->softpipe; 166 167 #if DEBUG_FRAGS 168 setup->numFragsEmitted += util_bitcount(quad->inout.mask); 169 #endif 170 171 sp->quad.first->run( sp->quad.first, &quad, 1 ); 172 } 173 } 174 175 176 177 /** 178 * Given an X or Y coordinate, return the block/quad coordinate that it 179 * belongs to. 180 */ 181 static inline int 182 block(int x) 183 { 184 return x & ~(2-1); 185 } 186 187 188 static inline int 189 block_x(int x) 190 { 191 return x & ~(16-1); 192 } 193 194 195 /** 196 * Render a horizontal span of quads 197 */ 198 static void 199 flush_spans(struct setup_context *setup) 200 { 201 const int step = MAX_QUADS; 202 const int xleft0 = setup->span.left[0]; 203 const int xleft1 = setup->span.left[1]; 204 const int xright0 = setup->span.right[0]; 205 const int xright1 = setup->span.right[1]; 206 struct quad_stage *pipe = setup->softpipe->quad.first; 207 208 const int minleft = block_x(MIN2(xleft0, xleft1)); 209 const int maxright = MAX2(xright0, xright1); 210 int x; 211 212 /* process quads in horizontal chunks of 16 */ 213 for (x = minleft; x < maxright; x += step) { 214 unsigned skip_left0 = CLAMP(xleft0 - x, 0, step); 215 unsigned skip_left1 = CLAMP(xleft1 - x, 0, step); 216 unsigned skip_right0 = CLAMP(x + step - xright0, 0, step); 217 unsigned skip_right1 = CLAMP(x + step - xright1, 0, step); 218 unsigned lx = x; 219 unsigned q = 0; 220 221 unsigned skipmask_left0 = (1U << skip_left0) - 1U; 222 unsigned skipmask_left1 = (1U << skip_left1) - 1U; 223 224 /* These calculations fail when step == 32 and skip_right == 0. 225 */ 226 unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0); 227 unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1); 228 229 unsigned mask0 = ~skipmask_left0 & ~skipmask_right0; 230 unsigned mask1 = ~skipmask_left1 & ~skipmask_right1; 231 232 if (mask0 | mask1) { 233 do { 234 unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2); 235 if (quadmask) { 236 setup->quad[q].input.x0 = lx; 237 setup->quad[q].input.y0 = setup->span.y; 238 setup->quad[q].input.facing = setup->facing; 239 setup->quad[q].inout.mask = quadmask; 240 setup->quad_ptrs[q] = &setup->quad[q]; 241 q++; 242 #if DEBUG_FRAGS 243 setup->numFragsEmitted += util_bitcount(quadmask); 244 #endif 245 } 246 mask0 >>= 2; 247 mask1 >>= 2; 248 lx += 2; 249 } while (mask0 | mask1); 250 251 pipe->run( pipe, setup->quad_ptrs, q ); 252 } 253 } 254 255 256 setup->span.y = 0; 257 setup->span.right[0] = 0; 258 setup->span.right[1] = 0; 259 setup->span.left[0] = 1000000; /* greater than right[0] */ 260 setup->span.left[1] = 1000000; /* greater than right[1] */ 261 } 262 263 264 #if DEBUG_VERTS 265 static void 266 print_vertex(const struct setup_context *setup, 267 const float (*v)[4]) 268 { 269 int i; 270 debug_printf(" Vertex: (%p)\n", (void *) v); 271 for (i = 0; i < setup->nr_vertex_attrs; i++) { 272 debug_printf(" %d: %f %f %f %f\n", i, 273 v[i][0], v[i][1], v[i][2], v[i][3]); 274 if (util_is_inf_or_nan(v[i][0])) { 275 debug_printf(" NaN!\n"); 276 } 277 } 278 } 279 #endif 280 281 282 /** 283 * Sort the vertices from top to bottom order, setting up the triangle 284 * edge fields (ebot, emaj, etop). 285 * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise 286 */ 287 static boolean 288 setup_sort_vertices(struct setup_context *setup, 289 float det, 290 const float (*v0)[4], 291 const float (*v1)[4], 292 const float (*v2)[4]) 293 { 294 if (setup->softpipe->rasterizer->flatshade_first) 295 setup->vprovoke = v0; 296 else 297 setup->vprovoke = v2; 298 299 /* determine bottom to top order of vertices */ 300 { 301 float y0 = v0[0][1]; 302 float y1 = v1[0][1]; 303 float y2 = v2[0][1]; 304 if (y0 <= y1) { 305 if (y1 <= y2) { 306 /* y0<=y1<=y2 */ 307 setup->vmin = v0; 308 setup->vmid = v1; 309 setup->vmax = v2; 310 } 311 else if (y2 <= y0) { 312 /* y2<=y0<=y1 */ 313 setup->vmin = v2; 314 setup->vmid = v0; 315 setup->vmax = v1; 316 } 317 else { 318 /* y0<=y2<=y1 */ 319 setup->vmin = v0; 320 setup->vmid = v2; 321 setup->vmax = v1; 322 } 323 } 324 else { 325 if (y0 <= y2) { 326 /* y1<=y0<=y2 */ 327 setup->vmin = v1; 328 setup->vmid = v0; 329 setup->vmax = v2; 330 } 331 else if (y2 <= y1) { 332 /* y2<=y1<=y0 */ 333 setup->vmin = v2; 334 setup->vmid = v1; 335 setup->vmax = v0; 336 } 337 else { 338 /* y1<=y2<=y0 */ 339 setup->vmin = v1; 340 setup->vmid = v2; 341 setup->vmax = v0; 342 } 343 } 344 } 345 346 setup->ebot.dx = setup->vmid[0][0] - setup->vmin[0][0]; 347 setup->ebot.dy = setup->vmid[0][1] - setup->vmin[0][1]; 348 setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0]; 349 setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1]; 350 setup->etop.dx = setup->vmax[0][0] - setup->vmid[0][0]; 351 setup->etop.dy = setup->vmax[0][1] - setup->vmid[0][1]; 352 353 /* 354 * Compute triangle's area. Use 1/area to compute partial 355 * derivatives of attributes later. 356 * 357 * The area will be the same as prim->det, but the sign may be 358 * different depending on how the vertices get sorted above. 359 * 360 * To determine whether the primitive is front or back facing we 361 * use the prim->det value because its sign is correct. 362 */ 363 { 364 const float area = (setup->emaj.dx * setup->ebot.dy - 365 setup->ebot.dx * setup->emaj.dy); 366 367 setup->oneoverarea = 1.0f / area; 368 369 /* 370 debug_printf("%s one-over-area %f area %f det %f\n", 371 __FUNCTION__, setup->oneoverarea, area, det ); 372 */ 373 if (util_is_inf_or_nan(setup->oneoverarea)) 374 return FALSE; 375 } 376 377 /* We need to know if this is a front or back-facing triangle for: 378 * - the GLSL gl_FrontFacing fragment attribute (bool) 379 * - two-sided stencil test 380 * 0 = front-facing, 1 = back-facing 381 */ 382 setup->facing = 383 ((det < 0.0) ^ 384 (setup->softpipe->rasterizer->front_ccw)); 385 386 { 387 unsigned face = setup->facing == 0 ? PIPE_FACE_FRONT : PIPE_FACE_BACK; 388 389 if (face & setup->cull_face) 390 return FALSE; 391 } 392 393 394 /* Prepare pixel offset for rasterisation: 395 * - pixel center (0.5, 0.5) for GL, or 396 * - assume (0.0, 0.0) for other APIs. 397 */ 398 if (setup->softpipe->rasterizer->half_pixel_center) { 399 setup->pixel_offset = 0.5f; 400 } else { 401 setup->pixel_offset = 0.0f; 402 } 403 404 return TRUE; 405 } 406 407 408 /* Apply cylindrical wrapping to v0, v1, v2 coordinates, if enabled. 409 * Input coordinates must be in [0, 1] range, otherwise results are undefined. 410 * Some combinations of coordinates produce invalid results, 411 * but this behaviour is acceptable. 412 */ 413 static void 414 tri_apply_cylindrical_wrap(float v0, 415 float v1, 416 float v2, 417 uint cylindrical_wrap, 418 float output[3]) 419 { 420 if (cylindrical_wrap) { 421 float delta; 422 423 delta = v1 - v0; 424 if (delta > 0.5f) { 425 v0 += 1.0f; 426 } 427 else if (delta < -0.5f) { 428 v1 += 1.0f; 429 } 430 431 delta = v2 - v1; 432 if (delta > 0.5f) { 433 v1 += 1.0f; 434 } 435 else if (delta < -0.5f) { 436 v2 += 1.0f; 437 } 438 439 delta = v0 - v2; 440 if (delta > 0.5f) { 441 v2 += 1.0f; 442 } 443 else if (delta < -0.5f) { 444 v0 += 1.0f; 445 } 446 } 447 448 output[0] = v0; 449 output[1] = v1; 450 output[2] = v2; 451 } 452 453 454 /** 455 * Compute a0 for a constant-valued coefficient (GL_FLAT shading). 456 * The value value comes from vertex[slot][i]. 457 * The result will be put into setup->coef[slot].a0[i]. 458 * \param slot which attribute slot 459 * \param i which component of the slot (0..3) 460 */ 461 static void 462 const_coeff(struct setup_context *setup, 463 struct tgsi_interp_coef *coef, 464 uint vertSlot, uint i) 465 { 466 assert(i <= 3); 467 468 coef->dadx[i] = 0; 469 coef->dady[i] = 0; 470 471 /* need provoking vertex info! 472 */ 473 coef->a0[i] = setup->vprovoke[vertSlot][i]; 474 } 475 476 477 /** 478 * Compute a0, dadx and dady for a linearly interpolated coefficient, 479 * for a triangle. 480 * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively. 481 */ 482 static void 483 tri_linear_coeff(struct setup_context *setup, 484 struct tgsi_interp_coef *coef, 485 uint i, 486 const float v[3]) 487 { 488 float botda = v[1] - v[0]; 489 float majda = v[2] - v[0]; 490 float a = setup->ebot.dy * majda - botda * setup->emaj.dy; 491 float b = setup->emaj.dx * botda - majda * setup->ebot.dx; 492 float dadx = a * setup->oneoverarea; 493 float dady = b * setup->oneoverarea; 494 495 assert(i <= 3); 496 497 coef->dadx[i] = dadx; 498 coef->dady[i] = dady; 499 500 /* calculate a0 as the value which would be sampled for the 501 * fragment at (0,0), taking into account that we want to sample at 502 * pixel centers, in other words (pixel_offset, pixel_offset). 503 * 504 * this is neat but unfortunately not a good way to do things for 505 * triangles with very large values of dadx or dady as it will 506 * result in the subtraction and re-addition from a0 of a very 507 * large number, which means we'll end up loosing a lot of the 508 * fractional bits and precision from a0. the way to fix this is 509 * to define a0 as the sample at a pixel center somewhere near vmin 510 * instead - i'll switch to this later. 511 */ 512 coef->a0[i] = (v[0] - 513 (dadx * (setup->vmin[0][0] - setup->pixel_offset) + 514 dady * (setup->vmin[0][1] - setup->pixel_offset))); 515 } 516 517 518 /** 519 * Compute a0, dadx and dady for a perspective-corrected interpolant, 520 * for a triangle. 521 * We basically multiply the vertex value by 1/w before computing 522 * the plane coefficients (a0, dadx, dady). 523 * Later, when we compute the value at a particular fragment position we'll 524 * divide the interpolated value by the interpolated W at that fragment. 525 * v[0], v[1] and v[2] are vmin, vmid and vmax, respectively. 526 */ 527 static void 528 tri_persp_coeff(struct setup_context *setup, 529 struct tgsi_interp_coef *coef, 530 uint i, 531 const float v[3]) 532 { 533 /* premultiply by 1/w (v[0][3] is always W): 534 */ 535 float mina = v[0] * setup->vmin[0][3]; 536 float mida = v[1] * setup->vmid[0][3]; 537 float maxa = v[2] * setup->vmax[0][3]; 538 float botda = mida - mina; 539 float majda = maxa - mina; 540 float a = setup->ebot.dy * majda - botda * setup->emaj.dy; 541 float b = setup->emaj.dx * botda - majda * setup->ebot.dx; 542 float dadx = a * setup->oneoverarea; 543 float dady = b * setup->oneoverarea; 544 545 assert(i <= 3); 546 547 coef->dadx[i] = dadx; 548 coef->dady[i] = dady; 549 coef->a0[i] = (mina - 550 (dadx * (setup->vmin[0][0] - setup->pixel_offset) + 551 dady * (setup->vmin[0][1] - setup->pixel_offset))); 552 } 553 554 555 /** 556 * Special coefficient setup for gl_FragCoord. 557 * X and Y are trivial, though Y may have to be inverted for OpenGL. 558 * Z and W are copied from posCoef which should have already been computed. 559 * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask. 560 */ 561 static void 562 setup_fragcoord_coeff(struct setup_context *setup, uint slot) 563 { 564 const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info; 565 boolean origin_lower_left = 566 fsInfo->properties[TGSI_PROPERTY_FS_COORD_ORIGIN]; 567 boolean pixel_center_integer = 568 fsInfo->properties[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER]; 569 570 /*X*/ 571 setup->coef[slot].a0[0] = pixel_center_integer ? 0.0f : 0.5f; 572 setup->coef[slot].dadx[0] = 1.0f; 573 setup->coef[slot].dady[0] = 0.0f; 574 /*Y*/ 575 setup->coef[slot].a0[1] = 576 (origin_lower_left ? setup->softpipe->framebuffer.height-1 : 0) 577 + (pixel_center_integer ? 0.0f : 0.5f); 578 setup->coef[slot].dadx[1] = 0.0f; 579 setup->coef[slot].dady[1] = origin_lower_left ? -1.0f : 1.0f; 580 /*Z*/ 581 setup->coef[slot].a0[2] = setup->posCoef.a0[2]; 582 setup->coef[slot].dadx[2] = setup->posCoef.dadx[2]; 583 setup->coef[slot].dady[2] = setup->posCoef.dady[2]; 584 /*W*/ 585 setup->coef[slot].a0[3] = setup->posCoef.a0[3]; 586 setup->coef[slot].dadx[3] = setup->posCoef.dadx[3]; 587 setup->coef[slot].dady[3] = setup->posCoef.dady[3]; 588 } 589 590 591 592 /** 593 * Compute the setup->coef[] array dadx, dady, a0 values. 594 * Must be called after setup->vmin,vmid,vmax,vprovoke are initialized. 595 */ 596 static void 597 setup_tri_coefficients(struct setup_context *setup) 598 { 599 struct softpipe_context *softpipe = setup->softpipe; 600 const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info; 601 const struct sp_setup_info *sinfo = &softpipe->setup_info; 602 uint fragSlot; 603 float v[3]; 604 605 assert(sinfo->valid); 606 607 /* z and w are done by linear interpolation: 608 */ 609 v[0] = setup->vmin[0][2]; 610 v[1] = setup->vmid[0][2]; 611 v[2] = setup->vmax[0][2]; 612 tri_linear_coeff(setup, &setup->posCoef, 2, v); 613 614 v[0] = setup->vmin[0][3]; 615 v[1] = setup->vmid[0][3]; 616 v[2] = setup->vmax[0][3]; 617 tri_linear_coeff(setup, &setup->posCoef, 3, v); 618 619 /* setup interpolation for all the remaining attributes: 620 */ 621 for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) { 622 const uint vertSlot = sinfo->attrib[fragSlot].src_index; 623 uint j; 624 625 switch (sinfo->attrib[fragSlot].interp) { 626 case SP_INTERP_CONSTANT: 627 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 628 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); 629 } 630 break; 631 case SP_INTERP_LINEAR: 632 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 633 tri_apply_cylindrical_wrap(setup->vmin[vertSlot][j], 634 setup->vmid[vertSlot][j], 635 setup->vmax[vertSlot][j], 636 fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j), 637 v); 638 tri_linear_coeff(setup, &setup->coef[fragSlot], j, v); 639 } 640 break; 641 case SP_INTERP_PERSPECTIVE: 642 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 643 tri_apply_cylindrical_wrap(setup->vmin[vertSlot][j], 644 setup->vmid[vertSlot][j], 645 setup->vmax[vertSlot][j], 646 fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j), 647 v); 648 tri_persp_coeff(setup, &setup->coef[fragSlot], j, v); 649 } 650 break; 651 case SP_INTERP_POS: 652 setup_fragcoord_coeff(setup, fragSlot); 653 break; 654 default: 655 assert(0); 656 } 657 658 if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { 659 /* convert 0 to 1.0 and 1 to -1.0 */ 660 setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f; 661 setup->coef[fragSlot].dadx[0] = 0.0; 662 setup->coef[fragSlot].dady[0] = 0.0; 663 } 664 665 if (0) { 666 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 667 debug_printf("attr[%d].%c: a0:%f dx:%f dy:%f\n", 668 fragSlot, "xyzw"[j], 669 setup->coef[fragSlot].a0[j], 670 setup->coef[fragSlot].dadx[j], 671 setup->coef[fragSlot].dady[j]); 672 } 673 } 674 } 675 } 676 677 678 static void 679 setup_tri_edges(struct setup_context *setup) 680 { 681 float vmin_x = setup->vmin[0][0] + setup->pixel_offset; 682 float vmid_x = setup->vmid[0][0] + setup->pixel_offset; 683 684 float vmin_y = setup->vmin[0][1] - setup->pixel_offset; 685 float vmid_y = setup->vmid[0][1] - setup->pixel_offset; 686 float vmax_y = setup->vmax[0][1] - setup->pixel_offset; 687 688 setup->emaj.sy = ceilf(vmin_y); 689 setup->emaj.lines = (int) ceilf(vmax_y - setup->emaj.sy); 690 setup->emaj.dxdy = setup->emaj.dy ? setup->emaj.dx / setup->emaj.dy : .0f; 691 setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy; 692 693 setup->etop.sy = ceilf(vmid_y); 694 setup->etop.lines = (int) ceilf(vmax_y - setup->etop.sy); 695 setup->etop.dxdy = setup->etop.dy ? setup->etop.dx / setup->etop.dy : .0f; 696 setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy; 697 698 setup->ebot.sy = ceilf(vmin_y); 699 setup->ebot.lines = (int) ceilf(vmid_y - setup->ebot.sy); 700 setup->ebot.dxdy = setup->ebot.dy ? setup->ebot.dx / setup->ebot.dy : .0f; 701 setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy; 702 } 703 704 705 /** 706 * Render the upper or lower half of a triangle. 707 * Scissoring/cliprect is applied here too. 708 */ 709 static void 710 subtriangle(struct setup_context *setup, 711 struct edge *eleft, 712 struct edge *eright, 713 int lines, 714 unsigned viewport_index) 715 { 716 const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect[viewport_index]; 717 const int minx = (int) cliprect->minx; 718 const int maxx = (int) cliprect->maxx; 719 const int miny = (int) cliprect->miny; 720 const int maxy = (int) cliprect->maxy; 721 int y, start_y, finish_y; 722 int sy = (int)eleft->sy; 723 724 assert((int)eleft->sy == (int) eright->sy); 725 assert(lines >= 0); 726 727 /* clip top/bottom */ 728 start_y = sy; 729 if (start_y < miny) 730 start_y = miny; 731 732 finish_y = sy + lines; 733 if (finish_y > maxy) 734 finish_y = maxy; 735 736 start_y -= sy; 737 finish_y -= sy; 738 739 /* 740 debug_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y); 741 */ 742 743 for (y = start_y; y < finish_y; y++) { 744 745 /* avoid accumulating adds as floats don't have the precision to 746 * accurately iterate large triangle edges that way. luckily we 747 * can just multiply these days. 748 * 749 * this is all drowned out by the attribute interpolation anyway. 750 */ 751 int left = (int)(eleft->sx + y * eleft->dxdy); 752 int right = (int)(eright->sx + y * eright->dxdy); 753 754 /* clip left/right */ 755 if (left < minx) 756 left = minx; 757 if (right > maxx) 758 right = maxx; 759 760 if (left < right) { 761 int _y = sy + y; 762 if (block(_y) != setup->span.y) { 763 flush_spans(setup); 764 setup->span.y = block(_y); 765 } 766 767 setup->span.left[_y&1] = left; 768 setup->span.right[_y&1] = right; 769 } 770 } 771 772 773 /* save the values so that emaj can be restarted: 774 */ 775 eleft->sx += lines * eleft->dxdy; 776 eright->sx += lines * eright->dxdy; 777 eleft->sy += lines; 778 eright->sy += lines; 779 } 780 781 782 /** 783 * Recalculate prim's determinant. This is needed as we don't have 784 * get this information through the vbuf_render interface & we must 785 * calculate it here. 786 */ 787 static float 788 calc_det(const float (*v0)[4], 789 const float (*v1)[4], 790 const float (*v2)[4]) 791 { 792 /* edge vectors e = v0 - v2, f = v1 - v2 */ 793 const float ex = v0[0][0] - v2[0][0]; 794 const float ey = v0[0][1] - v2[0][1]; 795 const float fx = v1[0][0] - v2[0][0]; 796 const float fy = v1[0][1] - v2[0][1]; 797 798 /* det = cross(e,f).z */ 799 return ex * fy - ey * fx; 800 } 801 802 803 /** 804 * Do setup for triangle rasterization, then render the triangle. 805 */ 806 void 807 sp_setup_tri(struct setup_context *setup, 808 const float (*v0)[4], 809 const float (*v1)[4], 810 const float (*v2)[4]) 811 { 812 float det; 813 uint layer = 0; 814 unsigned viewport_index = 0; 815 #if DEBUG_VERTS 816 debug_printf("Setup triangle:\n"); 817 print_vertex(setup, v0); 818 print_vertex(setup, v1); 819 print_vertex(setup, v2); 820 #endif 821 822 if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard) 823 return; 824 825 det = calc_det(v0, v1, v2); 826 /* 827 debug_printf("%s\n", __FUNCTION__ ); 828 */ 829 830 #if DEBUG_FRAGS 831 setup->numFragsEmitted = 0; 832 setup->numFragsWritten = 0; 833 #endif 834 835 if (!setup_sort_vertices( setup, det, v0, v1, v2 )) 836 return; 837 838 setup_tri_coefficients( setup ); 839 setup_tri_edges( setup ); 840 841 assert(setup->softpipe->reduced_prim == PIPE_PRIM_TRIANGLES); 842 843 setup->span.y = 0; 844 setup->span.right[0] = 0; 845 setup->span.right[1] = 0; 846 /* setup->span.z_mode = tri_z_mode( setup->ctx ); */ 847 if (setup->softpipe->layer_slot > 0) { 848 layer = *(unsigned *)setup->vprovoke[setup->softpipe->layer_slot]; 849 layer = MIN2(layer, setup->max_layer); 850 } 851 setup->quad[0].input.layer = layer; 852 853 if (setup->softpipe->viewport_index_slot > 0) { 854 unsigned *udata = (unsigned*)v0[setup->softpipe->viewport_index_slot]; 855 viewport_index = sp_clamp_viewport_idx(*udata); 856 } 857 setup->quad[0].input.viewport_index = viewport_index; 858 859 /* init_constant_attribs( setup ); */ 860 861 if (setup->oneoverarea < 0.0) { 862 /* emaj on left: 863 */ 864 subtriangle(setup, &setup->emaj, &setup->ebot, setup->ebot.lines, viewport_index); 865 subtriangle(setup, &setup->emaj, &setup->etop, setup->etop.lines, viewport_index); 866 } 867 else { 868 /* emaj on right: 869 */ 870 subtriangle(setup, &setup->ebot, &setup->emaj, setup->ebot.lines, viewport_index); 871 subtriangle(setup, &setup->etop, &setup->emaj, setup->etop.lines, viewport_index); 872 } 873 874 flush_spans( setup ); 875 876 if (setup->softpipe->active_statistics_queries) { 877 setup->softpipe->pipeline_statistics.c_primitives++; 878 } 879 880 #if DEBUG_FRAGS 881 printf("Tri: %u frags emitted, %u written\n", 882 setup->numFragsEmitted, 883 setup->numFragsWritten); 884 #endif 885 } 886 887 888 /* Apply cylindrical wrapping to v0, v1 coordinates, if enabled. 889 * Input coordinates must be in [0, 1] range, otherwise results are undefined. 890 */ 891 static void 892 line_apply_cylindrical_wrap(float v0, 893 float v1, 894 uint cylindrical_wrap, 895 float output[2]) 896 { 897 if (cylindrical_wrap) { 898 float delta; 899 900 delta = v1 - v0; 901 if (delta > 0.5f) { 902 v0 += 1.0f; 903 } 904 else if (delta < -0.5f) { 905 v1 += 1.0f; 906 } 907 } 908 909 output[0] = v0; 910 output[1] = v1; 911 } 912 913 914 /** 915 * Compute a0, dadx and dady for a linearly interpolated coefficient, 916 * for a line. 917 * v[0] and v[1] are vmin and vmax, respectively. 918 */ 919 static void 920 line_linear_coeff(const struct setup_context *setup, 921 struct tgsi_interp_coef *coef, 922 uint i, 923 const float v[2]) 924 { 925 const float da = v[1] - v[0]; 926 const float dadx = da * setup->emaj.dx * setup->oneoverarea; 927 const float dady = da * setup->emaj.dy * setup->oneoverarea; 928 coef->dadx[i] = dadx; 929 coef->dady[i] = dady; 930 coef->a0[i] = (v[0] - 931 (dadx * (setup->vmin[0][0] - setup->pixel_offset) + 932 dady * (setup->vmin[0][1] - setup->pixel_offset))); 933 } 934 935 936 /** 937 * Compute a0, dadx and dady for a perspective-corrected interpolant, 938 * for a line. 939 * v[0] and v[1] are vmin and vmax, respectively. 940 */ 941 static void 942 line_persp_coeff(const struct setup_context *setup, 943 struct tgsi_interp_coef *coef, 944 uint i, 945 const float v[2]) 946 { 947 const float a0 = v[0] * setup->vmin[0][3]; 948 const float a1 = v[1] * setup->vmax[0][3]; 949 const float da = a1 - a0; 950 const float dadx = da * setup->emaj.dx * setup->oneoverarea; 951 const float dady = da * setup->emaj.dy * setup->oneoverarea; 952 coef->dadx[i] = dadx; 953 coef->dady[i] = dady; 954 coef->a0[i] = (a0 - 955 (dadx * (setup->vmin[0][0] - setup->pixel_offset) + 956 dady * (setup->vmin[0][1] - setup->pixel_offset))); 957 } 958 959 960 /** 961 * Compute the setup->coef[] array dadx, dady, a0 values. 962 * Must be called after setup->vmin,vmax are initialized. 963 */ 964 static boolean 965 setup_line_coefficients(struct setup_context *setup, 966 const float (*v0)[4], 967 const float (*v1)[4]) 968 { 969 struct softpipe_context *softpipe = setup->softpipe; 970 const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info; 971 const struct sp_setup_info *sinfo = &softpipe->setup_info; 972 uint fragSlot; 973 float area; 974 float v[2]; 975 976 assert(sinfo->valid); 977 978 /* use setup->vmin, vmax to point to vertices */ 979 if (softpipe->rasterizer->flatshade_first) 980 setup->vprovoke = v0; 981 else 982 setup->vprovoke = v1; 983 setup->vmin = v0; 984 setup->vmax = v1; 985 986 setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0]; 987 setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1]; 988 989 /* NOTE: this is not really area but something proportional to it */ 990 area = setup->emaj.dx * setup->emaj.dx + setup->emaj.dy * setup->emaj.dy; 991 if (area == 0.0f || util_is_inf_or_nan(area)) 992 return FALSE; 993 setup->oneoverarea = 1.0f / area; 994 995 /* z and w are done by linear interpolation: 996 */ 997 v[0] = setup->vmin[0][2]; 998 v[1] = setup->vmax[0][2]; 999 line_linear_coeff(setup, &setup->posCoef, 2, v); 1000 1001 v[0] = setup->vmin[0][3]; 1002 v[1] = setup->vmax[0][3]; 1003 line_linear_coeff(setup, &setup->posCoef, 3, v); 1004 1005 /* setup interpolation for all the remaining attributes: 1006 */ 1007 for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) { 1008 const uint vertSlot = sinfo->attrib[fragSlot].src_index; 1009 uint j; 1010 1011 switch (sinfo->attrib[fragSlot].interp) { 1012 case SP_INTERP_CONSTANT: 1013 for (j = 0; j < TGSI_NUM_CHANNELS; j++) 1014 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); 1015 break; 1016 case SP_INTERP_LINEAR: 1017 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 1018 line_apply_cylindrical_wrap(setup->vmin[vertSlot][j], 1019 setup->vmax[vertSlot][j], 1020 fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j), 1021 v); 1022 line_linear_coeff(setup, &setup->coef[fragSlot], j, v); 1023 } 1024 break; 1025 case SP_INTERP_PERSPECTIVE: 1026 for (j = 0; j < TGSI_NUM_CHANNELS; j++) { 1027 line_apply_cylindrical_wrap(setup->vmin[vertSlot][j], 1028 setup->vmax[vertSlot][j], 1029 fsInfo->input_cylindrical_wrap[fragSlot] & (1 << j), 1030 v); 1031 line_persp_coeff(setup, &setup->coef[fragSlot], j, v); 1032 } 1033 break; 1034 case SP_INTERP_POS: 1035 setup_fragcoord_coeff(setup, fragSlot); 1036 break; 1037 default: 1038 assert(0); 1039 } 1040 1041 if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { 1042 /* convert 0 to 1.0 and 1 to -1.0 */ 1043 setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f; 1044 setup->coef[fragSlot].dadx[0] = 0.0; 1045 setup->coef[fragSlot].dady[0] = 0.0; 1046 } 1047 } 1048 return TRUE; 1049 } 1050 1051 1052 /** 1053 * Plot a pixel in a line segment. 1054 */ 1055 static inline void 1056 plot(struct setup_context *setup, int x, int y) 1057 { 1058 const int iy = y & 1; 1059 const int ix = x & 1; 1060 const int quadX = x - ix; 1061 const int quadY = y - iy; 1062 const int mask = (1 << ix) << (2 * iy); 1063 1064 if (quadX != setup->quad[0].input.x0 || 1065 quadY != setup->quad[0].input.y0) 1066 { 1067 /* flush prev quad, start new quad */ 1068 1069 if (setup->quad[0].input.x0 != -1) 1070 clip_emit_quad(setup, &setup->quad[0]); 1071 1072 setup->quad[0].input.x0 = quadX; 1073 setup->quad[0].input.y0 = quadY; 1074 setup->quad[0].inout.mask = 0x0; 1075 } 1076 1077 setup->quad[0].inout.mask |= mask; 1078 } 1079 1080 1081 /** 1082 * Do setup for line rasterization, then render the line. 1083 * Single-pixel width, no stipple, etc. We rely on the 'draw' module 1084 * to handle stippling and wide lines. 1085 */ 1086 void 1087 sp_setup_line(struct setup_context *setup, 1088 const float (*v0)[4], 1089 const float (*v1)[4]) 1090 { 1091 int x0 = (int) v0[0][0]; 1092 int x1 = (int) v1[0][0]; 1093 int y0 = (int) v0[0][1]; 1094 int y1 = (int) v1[0][1]; 1095 int dx = x1 - x0; 1096 int dy = y1 - y0; 1097 int xstep, ystep; 1098 uint layer = 0; 1099 unsigned viewport_index = 0; 1100 1101 #if DEBUG_VERTS 1102 debug_printf("Setup line:\n"); 1103 print_vertex(setup, v0); 1104 print_vertex(setup, v1); 1105 #endif 1106 1107 if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard) 1108 return; 1109 1110 if (dx == 0 && dy == 0) 1111 return; 1112 1113 if (!setup_line_coefficients(setup, v0, v1)) 1114 return; 1115 1116 assert(v0[0][0] < 1.0e9); 1117 assert(v0[0][1] < 1.0e9); 1118 assert(v1[0][0] < 1.0e9); 1119 assert(v1[0][1] < 1.0e9); 1120 1121 if (dx < 0) { 1122 dx = -dx; /* make positive */ 1123 xstep = -1; 1124 } 1125 else { 1126 xstep = 1; 1127 } 1128 1129 if (dy < 0) { 1130 dy = -dy; /* make positive */ 1131 ystep = -1; 1132 } 1133 else { 1134 ystep = 1; 1135 } 1136 1137 assert(dx >= 0); 1138 assert(dy >= 0); 1139 assert(setup->softpipe->reduced_prim == PIPE_PRIM_LINES); 1140 1141 setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1; 1142 setup->quad[0].inout.mask = 0x0; 1143 if (setup->softpipe->layer_slot > 0) { 1144 layer = *(unsigned *)setup->vprovoke[setup->softpipe->layer_slot]; 1145 layer = MIN2(layer, setup->max_layer); 1146 } 1147 setup->quad[0].input.layer = layer; 1148 1149 if (setup->softpipe->viewport_index_slot > 0) { 1150 unsigned *udata = (unsigned*)setup->vprovoke[setup->softpipe->viewport_index_slot]; 1151 viewport_index = sp_clamp_viewport_idx(*udata); 1152 } 1153 setup->quad[0].input.viewport_index = viewport_index; 1154 1155 /* XXX temporary: set coverage to 1.0 so the line appears 1156 * if AA mode happens to be enabled. 1157 */ 1158 setup->quad[0].input.coverage[0] = 1159 setup->quad[0].input.coverage[1] = 1160 setup->quad[0].input.coverage[2] = 1161 setup->quad[0].input.coverage[3] = 1.0; 1162 1163 if (dx > dy) { 1164 /*** X-major line ***/ 1165 int i; 1166 const int errorInc = dy + dy; 1167 int error = errorInc - dx; 1168 const int errorDec = error - dx; 1169 1170 for (i = 0; i < dx; i++) { 1171 plot(setup, x0, y0); 1172 1173 x0 += xstep; 1174 if (error < 0) { 1175 error += errorInc; 1176 } 1177 else { 1178 error += errorDec; 1179 y0 += ystep; 1180 } 1181 } 1182 } 1183 else { 1184 /*** Y-major line ***/ 1185 int i; 1186 const int errorInc = dx + dx; 1187 int error = errorInc - dy; 1188 const int errorDec = error - dy; 1189 1190 for (i = 0; i < dy; i++) { 1191 plot(setup, x0, y0); 1192 1193 y0 += ystep; 1194 if (error < 0) { 1195 error += errorInc; 1196 } 1197 else { 1198 error += errorDec; 1199 x0 += xstep; 1200 } 1201 } 1202 } 1203 1204 /* draw final quad */ 1205 if (setup->quad[0].inout.mask) { 1206 clip_emit_quad(setup, &setup->quad[0]); 1207 } 1208 } 1209 1210 1211 static void 1212 point_persp_coeff(const struct setup_context *setup, 1213 const float (*vert)[4], 1214 struct tgsi_interp_coef *coef, 1215 uint vertSlot, uint i) 1216 { 1217 assert(i <= 3); 1218 coef->dadx[i] = 0.0F; 1219 coef->dady[i] = 0.0F; 1220 coef->a0[i] = vert[vertSlot][i] * vert[0][3]; 1221 } 1222 1223 1224 /** 1225 * Do setup for point rasterization, then render the point. 1226 * Round or square points... 1227 * XXX could optimize a lot for 1-pixel points. 1228 */ 1229 void 1230 sp_setup_point(struct setup_context *setup, 1231 const float (*v0)[4]) 1232 { 1233 struct softpipe_context *softpipe = setup->softpipe; 1234 const struct tgsi_shader_info *fsInfo = &setup->softpipe->fs_variant->info; 1235 const int sizeAttr = setup->softpipe->psize_slot; 1236 const float size 1237 = sizeAttr > 0 ? v0[sizeAttr][0] 1238 : setup->softpipe->rasterizer->point_size; 1239 const float halfSize = 0.5F * size; 1240 const boolean round = (boolean) setup->softpipe->rasterizer->point_smooth; 1241 const float x = v0[0][0]; /* Note: data[0] is always position */ 1242 const float y = v0[0][1]; 1243 const struct sp_setup_info *sinfo = &softpipe->setup_info; 1244 uint fragSlot; 1245 uint layer = 0; 1246 unsigned viewport_index = 0; 1247 #if DEBUG_VERTS 1248 debug_printf("Setup point:\n"); 1249 print_vertex(setup, v0); 1250 #endif 1251 1252 assert(sinfo->valid); 1253 1254 if (setup->softpipe->no_rast || setup->softpipe->rasterizer->rasterizer_discard) 1255 return; 1256 1257 assert(setup->softpipe->reduced_prim == PIPE_PRIM_POINTS); 1258 1259 if (setup->softpipe->layer_slot > 0) { 1260 layer = *(unsigned *)v0[setup->softpipe->layer_slot]; 1261 layer = MIN2(layer, setup->max_layer); 1262 } 1263 setup->quad[0].input.layer = layer; 1264 1265 if (setup->softpipe->viewport_index_slot > 0) { 1266 unsigned *udata = (unsigned*)v0[setup->softpipe->viewport_index_slot]; 1267 viewport_index = sp_clamp_viewport_idx(*udata); 1268 } 1269 setup->quad[0].input.viewport_index = viewport_index; 1270 1271 /* For points, all interpolants are constant-valued. 1272 * However, for point sprites, we'll need to setup texcoords appropriately. 1273 * XXX: which coefficients are the texcoords??? 1274 * We may do point sprites as textured quads... 1275 * 1276 * KW: We don't know which coefficients are texcoords - ultimately 1277 * the choice of what interpolation mode to use for each attribute 1278 * should be determined by the fragment program, using 1279 * per-attribute declaration statements that include interpolation 1280 * mode as a parameter. So either the fragment program will have 1281 * to be adjusted for pointsprite vs normal point behaviour, or 1282 * otherwise a special interpolation mode will have to be defined 1283 * which matches the required behaviour for point sprites. But - 1284 * the latter is not a feature of normal hardware, and as such 1285 * probably should be ruled out on that basis. 1286 */ 1287 setup->vprovoke = v0; 1288 1289 /* setup Z, W */ 1290 const_coeff(setup, &setup->posCoef, 0, 2); 1291 const_coeff(setup, &setup->posCoef, 0, 3); 1292 1293 for (fragSlot = 0; fragSlot < fsInfo->num_inputs; fragSlot++) { 1294 const uint vertSlot = sinfo->attrib[fragSlot].src_index; 1295 uint j; 1296 1297 switch (sinfo->attrib[fragSlot].interp) { 1298 case SP_INTERP_CONSTANT: 1299 /* fall-through */ 1300 case SP_INTERP_LINEAR: 1301 for (j = 0; j < TGSI_NUM_CHANNELS; j++) 1302 const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); 1303 break; 1304 case SP_INTERP_PERSPECTIVE: 1305 for (j = 0; j < TGSI_NUM_CHANNELS; j++) 1306 point_persp_coeff(setup, setup->vprovoke, 1307 &setup->coef[fragSlot], vertSlot, j); 1308 break; 1309 case SP_INTERP_POS: 1310 setup_fragcoord_coeff(setup, fragSlot); 1311 break; 1312 default: 1313 assert(0); 1314 } 1315 1316 if (fsInfo->input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { 1317 /* convert 0 to 1.0 and 1 to -1.0 */ 1318 setup->coef[fragSlot].a0[0] = setup->facing * -2.0f + 1.0f; 1319 setup->coef[fragSlot].dadx[0] = 0.0; 1320 setup->coef[fragSlot].dady[0] = 0.0; 1321 } 1322 } 1323 1324 1325 if (halfSize <= 0.5 && !round) { 1326 /* special case for 1-pixel points */ 1327 const int ix = ((int) x) & 1; 1328 const int iy = ((int) y) & 1; 1329 setup->quad[0].input.x0 = (int) x - ix; 1330 setup->quad[0].input.y0 = (int) y - iy; 1331 setup->quad[0].inout.mask = (1 << ix) << (2 * iy); 1332 clip_emit_quad(setup, &setup->quad[0]); 1333 } 1334 else { 1335 if (round) { 1336 /* rounded points */ 1337 const int ixmin = block((int) (x - halfSize)); 1338 const int ixmax = block((int) (x + halfSize)); 1339 const int iymin = block((int) (y - halfSize)); 1340 const int iymax = block((int) (y + halfSize)); 1341 const float rmin = halfSize - 0.7071F; /* 0.7071 = sqrt(2)/2 */ 1342 const float rmax = halfSize + 0.7071F; 1343 const float rmin2 = MAX2(0.0F, rmin * rmin); 1344 const float rmax2 = rmax * rmax; 1345 const float cscale = 1.0F / (rmax2 - rmin2); 1346 int ix, iy; 1347 1348 for (iy = iymin; iy <= iymax; iy += 2) { 1349 for (ix = ixmin; ix <= ixmax; ix += 2) { 1350 float dx, dy, dist2, cover; 1351 1352 setup->quad[0].inout.mask = 0x0; 1353 1354 dx = (ix + 0.5f) - x; 1355 dy = (iy + 0.5f) - y; 1356 dist2 = dx * dx + dy * dy; 1357 if (dist2 <= rmax2) { 1358 cover = 1.0F - (dist2 - rmin2) * cscale; 1359 setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f); 1360 setup->quad[0].inout.mask |= MASK_TOP_LEFT; 1361 } 1362 1363 dx = (ix + 1.5f) - x; 1364 dy = (iy + 0.5f) - y; 1365 dist2 = dx * dx + dy * dy; 1366 if (dist2 <= rmax2) { 1367 cover = 1.0F - (dist2 - rmin2) * cscale; 1368 setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f); 1369 setup->quad[0].inout.mask |= MASK_TOP_RIGHT; 1370 } 1371 1372 dx = (ix + 0.5f) - x; 1373 dy = (iy + 1.5f) - y; 1374 dist2 = dx * dx + dy * dy; 1375 if (dist2 <= rmax2) { 1376 cover = 1.0F - (dist2 - rmin2) * cscale; 1377 setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f); 1378 setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT; 1379 } 1380 1381 dx = (ix + 1.5f) - x; 1382 dy = (iy + 1.5f) - y; 1383 dist2 = dx * dx + dy * dy; 1384 if (dist2 <= rmax2) { 1385 cover = 1.0F - (dist2 - rmin2) * cscale; 1386 setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f); 1387 setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT; 1388 } 1389 1390 if (setup->quad[0].inout.mask) { 1391 setup->quad[0].input.x0 = ix; 1392 setup->quad[0].input.y0 = iy; 1393 clip_emit_quad(setup, &setup->quad[0]); 1394 } 1395 } 1396 } 1397 } 1398 else { 1399 /* square points */ 1400 const int xmin = (int) (x + 0.75 - halfSize); 1401 const int ymin = (int) (y + 0.25 - halfSize); 1402 const int xmax = xmin + (int) size; 1403 const int ymax = ymin + (int) size; 1404 /* XXX could apply scissor to xmin,ymin,xmax,ymax now */ 1405 const int ixmin = block(xmin); 1406 const int ixmax = block(xmax - 1); 1407 const int iymin = block(ymin); 1408 const int iymax = block(ymax - 1); 1409 int ix, iy; 1410 1411 /* 1412 debug_printf("(%f, %f) -> X:%d..%d Y:%d..%d\n", x, y, xmin, xmax,ymin,ymax); 1413 */ 1414 for (iy = iymin; iy <= iymax; iy += 2) { 1415 uint rowMask = 0xf; 1416 if (iy < ymin) { 1417 /* above the top edge */ 1418 rowMask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); 1419 } 1420 if (iy + 1 >= ymax) { 1421 /* below the bottom edge */ 1422 rowMask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); 1423 } 1424 1425 for (ix = ixmin; ix <= ixmax; ix += 2) { 1426 uint mask = rowMask; 1427 1428 if (ix < xmin) { 1429 /* fragment is past left edge of point, turn off left bits */ 1430 mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); 1431 } 1432 if (ix + 1 >= xmax) { 1433 /* past the right edge */ 1434 mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); 1435 } 1436 1437 setup->quad[0].inout.mask = mask; 1438 setup->quad[0].input.x0 = ix; 1439 setup->quad[0].input.y0 = iy; 1440 clip_emit_quad(setup, &setup->quad[0]); 1441 } 1442 } 1443 } 1444 } 1445 } 1446 1447 1448 /** 1449 * Called by vbuf code just before we start buffering primitives. 1450 */ 1451 void 1452 sp_setup_prepare(struct setup_context *setup) 1453 { 1454 struct softpipe_context *sp = setup->softpipe; 1455 int i; 1456 unsigned max_layer = ~0; 1457 if (sp->dirty) { 1458 softpipe_update_derived(sp, sp->reduced_api_prim); 1459 } 1460 1461 /* Note: nr_attrs is only used for debugging (vertex printing) */ 1462 setup->nr_vertex_attrs = draw_num_shader_outputs(sp->draw); 1463 1464 /* 1465 * Determine how many layers the fb has (used for clamping layer value). 1466 * OpenGL (but not d3d10) permits different amount of layers per rt, however 1467 * results are undefined if layer exceeds the amount of layers of ANY 1468 * attachment hence don't need separate per cbuf and zsbuf max. 1469 */ 1470 for (i = 0; i < setup->softpipe->framebuffer.nr_cbufs; i++) { 1471 struct pipe_surface *cbuf = setup->softpipe->framebuffer.cbufs[i]; 1472 if (cbuf) { 1473 max_layer = MIN2(max_layer, 1474 cbuf->u.tex.last_layer - cbuf->u.tex.first_layer); 1475 1476 } 1477 } 1478 1479 setup->max_layer = max_layer; 1480 1481 sp->quad.first->begin( sp->quad.first ); 1482 1483 if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES && 1484 sp->rasterizer->fill_front == PIPE_POLYGON_MODE_FILL && 1485 sp->rasterizer->fill_back == PIPE_POLYGON_MODE_FILL) { 1486 /* we'll do culling */ 1487 setup->cull_face = sp->rasterizer->cull_face; 1488 } 1489 else { 1490 /* 'draw' will do culling */ 1491 setup->cull_face = PIPE_FACE_NONE; 1492 } 1493 } 1494 1495 1496 void 1497 sp_setup_destroy_context(struct setup_context *setup) 1498 { 1499 FREE( setup ); 1500 } 1501 1502 1503 /** 1504 * Create a new primitive setup/render stage. 1505 */ 1506 struct setup_context * 1507 sp_setup_create_context(struct softpipe_context *softpipe) 1508 { 1509 struct setup_context *setup = CALLOC_STRUCT(setup_context); 1510 unsigned i; 1511 1512 setup->softpipe = softpipe; 1513 1514 for (i = 0; i < MAX_QUADS; i++) { 1515 setup->quad[i].coef = setup->coef; 1516 setup->quad[i].posCoef = &setup->posCoef; 1517 } 1518 1519 setup->span.left[0] = 1000000; /* greater than right[0] */ 1520 setup->span.left[1] = 1000000; /* greater than right[1] */ 1521 1522 return setup; 1523 } 1524
