1 /* 2 * Copyright (c) 2007, 2016, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. Oracle designates this 8 * particular file as subject to the "Classpath" exception as provided 9 * by Oracle in the LICENSE file that accompanied this code. 10 * 11 * This code is distributed in the hope that it will be useful, but WITHOUT 12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 14 * version 2 for more details (a copy is included in the LICENSE file that 15 * accompanied this code). 16 * 17 * You should have received a copy of the GNU General Public License version 18 * 2 along with this work; if not, write to the Free Software Foundation, 19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 20 * 21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 22 * or visit www.oracle.com if you need additional information or have any 23 * questions. 24 */ 25 26 package sun.java2d.marlin; 27 28 import java.util.Arrays; 29 import sun.awt.geom.PathConsumer2D; 30 import static sun.java2d.marlin.OffHeapArray.SIZE_INT; 31 import jdk.internal.misc.Unsafe; 32 33 final class Renderer implements PathConsumer2D, MarlinConst { 34 35 static final boolean DISABLE_RENDER = false; 36 37 static final boolean ENABLE_BLOCK_FLAGS = MarlinProperties.isUseTileFlags(); 38 static final boolean ENABLE_BLOCK_FLAGS_HEURISTICS = MarlinProperties.isUseTileFlagsWithHeuristics(); 39 40 private static final int ALL_BUT_LSB = 0xfffffffe; 41 private static final int ERR_STEP_MAX = 0x7fffffff; // = 2^31 - 1 42 43 private static final double POWER_2_TO_32 = 0x1.0p32; 44 45 // use float to make tosubpix methods faster (no int to float conversion) 46 public static final float F_SUBPIXEL_POSITIONS_X 47 = (float) SUBPIXEL_POSITIONS_X; 48 public static final float F_SUBPIXEL_POSITIONS_Y 49 = (float) SUBPIXEL_POSITIONS_Y; 50 public static final int SUBPIXEL_MASK_X = SUBPIXEL_POSITIONS_X - 1; 51 public static final int SUBPIXEL_MASK_Y = SUBPIXEL_POSITIONS_Y - 1; 52 53 // number of subpixels corresponding to a tile line 54 private static final int SUBPIXEL_TILE 55 = TILE_SIZE << SUBPIXEL_LG_POSITIONS_Y; 56 57 // 2048 (pixelSize) pixels (height) x 8 subpixels = 64K 58 static final int INITIAL_BUCKET_ARRAY 59 = INITIAL_PIXEL_DIM * SUBPIXEL_POSITIONS_Y; 60 61 // crossing capacity = edges count / 8 ~ 512 62 static final int INITIAL_CROSSING_COUNT = INITIAL_EDGES_COUNT >> 3; 63 64 public static final int WIND_EVEN_ODD = 0; 65 public static final int WIND_NON_ZERO = 1; 66 67 // common to all types of input path segments. 68 // OFFSET as bytes 69 // only integer values: 70 public static final long OFF_CURX_OR = 0; 71 public static final long OFF_ERROR = OFF_CURX_OR + SIZE_INT; 72 public static final long OFF_BUMP_X = OFF_ERROR + SIZE_INT; 73 public static final long OFF_BUMP_ERR = OFF_BUMP_X + SIZE_INT; 74 public static final long OFF_NEXT = OFF_BUMP_ERR + SIZE_INT; 75 public static final long OFF_YMAX = OFF_NEXT + SIZE_INT; 76 77 // size of one edge in bytes 78 public static final int SIZEOF_EDGE_BYTES = (int)(OFF_YMAX + SIZE_INT); 79 80 // curve break into lines 81 // cubic error in subpixels to decrement step 82 private static final float CUB_DEC_ERR_SUBPIX 83 = 2.5f * (NORM_SUBPIXELS / 8f); // 2.5 subpixel for typical 8x8 subpixels 84 // cubic error in subpixels to increment step 85 private static final float CUB_INC_ERR_SUBPIX 86 = 1f * (NORM_SUBPIXELS / 8f); // 1 subpixel for typical 8x8 subpixels 87 88 // cubic bind length to decrement step = 8 * error in subpixels 89 // pisces: 20 / 8 90 // openjfx pisces: 8 / 3.2 91 // multiply by 8 = error scale factor: 92 public static final float CUB_DEC_BND 93 = 8f * CUB_DEC_ERR_SUBPIX; // 20f means 2.5 subpixel error 94 // cubic bind length to increment step = 8 * error in subpixels 95 public static final float CUB_INC_BND 96 = 8f * CUB_INC_ERR_SUBPIX; // 8f means 1 subpixel error 97 98 // cubic countlg 99 public static final int CUB_COUNT_LG = 2; 100 // cubic count = 2^countlg 101 private static final int CUB_COUNT = 1 << CUB_COUNT_LG; 102 // cubic count^2 = 4^countlg 103 private static final int CUB_COUNT_2 = 1 << (2 * CUB_COUNT_LG); 104 // cubic count^3 = 8^countlg 105 private static final int CUB_COUNT_3 = 1 << (3 * CUB_COUNT_LG); 106 // cubic dt = 1 / count 107 private static final float CUB_INV_COUNT = 1f / CUB_COUNT; 108 // cubic dt^2 = 1 / count^2 = 1 / 4^countlg 109 private static final float CUB_INV_COUNT_2 = 1f / CUB_COUNT_2; 110 // cubic dt^3 = 1 / count^3 = 1 / 8^countlg 111 private static final float CUB_INV_COUNT_3 = 1f / CUB_COUNT_3; 112 113 // quad break into lines 114 // quadratic error in subpixels 115 private static final float QUAD_DEC_ERR_SUBPIX 116 = 1f * (NORM_SUBPIXELS / 8f); // 1 subpixel for typical 8x8 subpixels 117 118 // quadratic bind length to decrement step = 8 * error in subpixels 119 // pisces and openjfx pisces: 32 120 public static final float QUAD_DEC_BND 121 = 8f * QUAD_DEC_ERR_SUBPIX; // 8f means 1 subpixel error 122 123 ////////////////////////////////////////////////////////////////////////////// 124 // SCAN LINE 125 ////////////////////////////////////////////////////////////////////////////// 126 // crossings ie subpixel edge x coordinates 127 private int[] crossings; 128 // auxiliary storage for crossings (merge sort) 129 private int[] aux_crossings; 130 131 // indices into the segment pointer lists. They indicate the "active" 132 // sublist in the segment lists (the portion of the list that contains 133 // all the segments that cross the next scan line). 134 private int edgeCount; 135 private int[] edgePtrs; 136 // auxiliary storage for edge pointers (merge sort) 137 private int[] aux_edgePtrs; 138 139 // max used for both edgePtrs and crossings (stats only) 140 private int activeEdgeMaxUsed; 141 142 // crossings ref (dirty) 143 private final IntArrayCache.Reference crossings_ref; 144 // edgePtrs ref (dirty) 145 private final IntArrayCache.Reference edgePtrs_ref; 146 // merge sort initial arrays (large enough to satisfy most usages) (1024) 147 // aux_crossings ref (dirty) 148 private final IntArrayCache.Reference aux_crossings_ref; 149 // aux_edgePtrs ref (dirty) 150 private final IntArrayCache.Reference aux_edgePtrs_ref; 151 152 ////////////////////////////////////////////////////////////////////////////// 153 // EDGE LIST 154 ////////////////////////////////////////////////////////////////////////////// 155 private int edgeMinY = Integer.MAX_VALUE; 156 private int edgeMaxY = Integer.MIN_VALUE; 157 private float edgeMinX = Float.POSITIVE_INFINITY; 158 private float edgeMaxX = Float.NEGATIVE_INFINITY; 159 160 // edges [floats|ints] stored in off-heap memory 161 private final OffHeapArray edges; 162 163 private int[] edgeBuckets; 164 private int[] edgeBucketCounts; // 2*newedges + (1 if pruning needed) 165 // used range for edgeBuckets / edgeBucketCounts 166 private int buckets_minY; 167 private int buckets_maxY; 168 // sum of each edge delta Y (subpixels) 169 private int edgeSumDeltaY; 170 171 // edgeBuckets ref (clean) 172 private final IntArrayCache.Reference edgeBuckets_ref; 173 // edgeBucketCounts ref (clean) 174 private final IntArrayCache.Reference edgeBucketCounts_ref; 175 176 // Flattens using adaptive forward differencing. This only carries out 177 // one iteration of the AFD loop. All it does is update AFD variables (i.e. 178 // X0, Y0, D*[X|Y], COUNT; not variables used for computing scanline crossings). 179 private void quadBreakIntoLinesAndAdd(float x0, float y0, 180 final Curve c, 181 final float x2, final float y2) 182 { 183 int count = 1; // dt = 1 / count 184 185 // maximum(ddX|Y) = norm(dbx, dby) * dt^2 (= 1) 186 float maxDD = FloatMath.max(Math.abs(c.dbx), Math.abs(c.dby)); 187 188 final float _DEC_BND = QUAD_DEC_BND; 189 190 while (maxDD >= _DEC_BND) { 191 // divide step by half: 192 maxDD /= 4f; // error divided by 2^2 = 4 193 194 count <<= 1; 195 if (DO_STATS) { 196 rdrCtx.stats.stat_rdr_quadBreak_dec.add(count); 197 } 198 } 199 200 int nL = 0; // line count 201 if (count > 1) { 202 final float icount = 1f / count; // dt 203 final float icount2 = icount * icount; // dt^2 204 205 final float ddx = c.dbx * icount2; 206 final float ddy = c.dby * icount2; 207 float dx = c.bx * icount2 + c.cx * icount; 208 float dy = c.by * icount2 + c.cy * icount; 209 210 float x1, y1; 211 212 while (--count > 0) { 213 x1 = x0 + dx; 214 dx += ddx; 215 y1 = y0 + dy; 216 dy += ddy; 217 218 addLine(x0, y0, x1, y1); 219 220 if (DO_STATS) { nL++; } 221 x0 = x1; 222 y0 = y1; 229 } 230 } 231 232 // x0, y0 and x3,y3 are the endpoints of the curve. We could compute these 233 // using c.xat(0),c.yat(0) and c.xat(1),c.yat(1), but this might introduce 234 // numerical errors, and our callers already have the exact values. 235 // Another alternative would be to pass all the control points, and call 236 // c.set here, but then too many numbers are passed around. 237 private void curveBreakIntoLinesAndAdd(float x0, float y0, 238 final Curve c, 239 final float x3, final float y3) 240 { 241 int count = CUB_COUNT; 242 final float icount = CUB_INV_COUNT; // dt 243 final float icount2 = CUB_INV_COUNT_2; // dt^2 244 final float icount3 = CUB_INV_COUNT_3; // dt^3 245 246 // the dx and dy refer to forward differencing variables, not the last 247 // coefficients of the "points" polynomial 248 float dddx, dddy, ddx, ddy, dx, dy; 249 dddx = 2f * c.dax * icount3; 250 dddy = 2f * c.day * icount3; 251 ddx = dddx + c.dbx * icount2; 252 ddy = dddy + c.dby * icount2; 253 dx = c.ax * icount3 + c.bx * icount2 + c.cx * icount; 254 dy = c.ay * icount3 + c.by * icount2 + c.cy * icount; 255 256 // we use x0, y0 to walk the line 257 float x1 = x0, y1 = y0; 258 int nL = 0; // line count 259 260 final float _DEC_BND = CUB_DEC_BND; 261 final float _INC_BND = CUB_INC_BND; 262 263 while (count > 0) { 264 // divide step by half: 265 while (Math.abs(ddx) >= _DEC_BND || Math.abs(ddy) >= _DEC_BND) { 266 dddx /= 8f; 267 dddy /= 8f; 268 ddx = ddx/4f - dddx; 269 ddy = ddy/4f - dddy; 270 dx = (dx - ddx) / 2f; 271 dy = (dy - ddy) / 2f; 272 273 count <<= 1; 274 if (DO_STATS) { 275 rdrCtx.stats.stat_rdr_curveBreak_dec.add(count); 276 } 277 } 278 279 // double step: 280 // TODO: why use first derivative dX|Y instead of second ddX|Y ? 281 // both scale changes should use speed or acceleration to have the same metric. 282 283 // can only do this on even "count" values, because we must divide count by 2 284 while (count % 2 == 0 285 && Math.abs(dx) <= _INC_BND && Math.abs(dy) <= _INC_BND) 286 { 287 dx = 2f * dx + ddx; 288 dy = 2f * dy + ddy; 289 ddx = 4f * (ddx + dddx); 290 ddy = 4f * (ddy + dddy); 291 dddx *= 8f; 292 dddy *= 8f; 293 294 count >>= 1; 295 if (DO_STATS) { 296 rdrCtx.stats.stat_rdr_curveBreak_inc.add(count); 297 } 298 } 299 if (--count > 0) { 300 x1 += dx; 301 dx += ddx; 302 ddx += dddx; 303 y1 += dy; 304 dy += ddy; 305 ddy += dddy; 306 } else { 307 x1 = x3; 308 y1 = y3; 309 } 310 311 addLine(x0, y0, x1, y1); 312 320 } 321 322 private void addLine(float x1, float y1, float x2, float y2) { 323 if (DO_MONITORS) { 324 rdrCtx.stats.mon_rdr_addLine.start(); 325 } 326 if (DO_STATS) { 327 rdrCtx.stats.stat_rdr_addLine.add(1); 328 } 329 int or = 1; // orientation of the line. 1 if y increases, 0 otherwise. 330 if (y2 < y1) { 331 or = 0; 332 float tmp = y2; 333 y2 = y1; 334 y1 = tmp; 335 tmp = x2; 336 x2 = x1; 337 x1 = tmp; 338 } 339 340 // convert subpixel coordinates (float) into pixel positions (int) 341 342 // The index of the pixel that holds the next HPC is at ceil(trueY - 0.5) 343 // Since y1 and y2 are biased by -0.5 in tosubpixy(), this is simply 344 // ceil(y1) or ceil(y2) 345 // upper integer (inclusive) 346 final int firstCrossing = FloatMath.max(FloatMath.ceil_int(y1), boundsMinY); 347 348 // note: use boundsMaxY (last Y exclusive) to compute correct coverage 349 // upper integer (exclusive) 350 final int lastCrossing = FloatMath.min(FloatMath.ceil_int(y2), boundsMaxY); 351 352 /* skip horizontal lines in pixel space and clip edges 353 out of y range [boundsMinY; boundsMaxY] */ 354 if (firstCrossing >= lastCrossing) { 355 if (DO_MONITORS) { 356 rdrCtx.stats.mon_rdr_addLine.stop(); 357 } 358 if (DO_STATS) { 359 rdrCtx.stats.stat_rdr_addLine_skip.add(1); 360 } 361 return; 362 } 363 364 // edge min/max X/Y are in subpixel space (inclusive) within bounds: 365 // note: Use integer crossings to ensure consistent range within 366 // edgeBuckets / edgeBucketCounts arrays in case of NaN values (int = 0) 367 if (firstCrossing < edgeMinY) { 368 edgeMinY = firstCrossing; 369 } 370 if (lastCrossing > edgeMaxY) { 371 edgeMaxY = lastCrossing; 372 } 373 374 // Use double-precision for improved accuracy: 375 final double x1d = x1; 376 final double y1d = y1; 377 final double slope = (x1d - x2) / (y1d - y2); 378 379 if (slope >= 0.0) { // <==> x1 < x2 380 if (x1 < edgeMinX) { 381 edgeMinX = x1; 382 } 383 if (x2 > edgeMaxX) { 384 edgeMaxX = x2; 385 } 386 } else { 387 if (x2 < edgeMinX) { 388 edgeMinX = x2; 389 } 390 if (x1 > edgeMaxX) { 391 edgeMaxX = x1; 392 } 393 } 394 395 // local variables for performance: 396 final int _SIZEOF_EDGE_BYTES = SIZEOF_EDGE_BYTES; 397 398 final OffHeapArray _edges = edges; 399 422 423 // The x value must be bumped up to its position at the next HPC we will evaluate. 424 // "firstcrossing" is the (sub)pixel number where the next crossing occurs 425 // thus, the actual coordinate of the next HPC is "firstcrossing + 0.5" 426 // so the Y distance we cover is "firstcrossing + 0.5 - trueY". 427 // Note that since y1 (and y2) are already biased by -0.5 in tosubpixy(), we have 428 // y1 = trueY - 0.5 429 // trueY = y1 + 0.5 430 // firstcrossing + 0.5 - trueY = firstcrossing + 0.5 - (y1 + 0.5) 431 // = firstcrossing - y1 432 // The x coordinate at that HPC is then: 433 // x1_intercept = x1 + (firstcrossing - y1) * slope 434 // The next VPC is then given by: 435 // VPC index = ceil(x1_intercept - 0.5), or alternately 436 // VPC index = floor(x1_intercept - 0.5 + 1 - epsilon) 437 // epsilon is hard to pin down in floating point, but easy in fixed point, so if 438 // we convert to fixed point then these operations get easier: 439 // long x1_fixed = x1_intercept * 2^32; (fixed point 32.32 format) 440 // curx = next VPC = fixed_floor(x1_fixed - 2^31 + 2^32 - 1) 441 // = fixed_floor(x1_fixed + 2^31 - 1) 442 // = fixed_floor(x1_fixed + 0x7fffffff) 443 // and error = fixed_fract(x1_fixed + 0x7fffffff) 444 final double x1_intercept = x1d + (firstCrossing - y1d) * slope; 445 446 // inlined scalb(x1_intercept, 32): 447 final long x1_fixed_biased = ((long) (POWER_2_TO_32 * x1_intercept)) 448 + 0x7fffffffL; 449 // curx: 450 // last bit corresponds to the orientation 451 _unsafe.putInt(addr, (((int) (x1_fixed_biased >> 31L)) & ALL_BUT_LSB) | or); 452 addr += SIZE_INT; 453 _unsafe.putInt(addr, ((int) x1_fixed_biased) >>> 1); 454 addr += SIZE_INT; 455 456 // inlined scalb(slope, 32): 457 final long slope_fixed = (long) (POWER_2_TO_32 * slope); 458 459 // last bit set to 0 to keep orientation: 460 _unsafe.putInt(addr, (((int) (slope_fixed >> 31L)) & ALL_BUT_LSB)); 461 addr += SIZE_INT; 462 _unsafe.putInt(addr, ((int) slope_fixed) >>> 1); 463 addr += SIZE_INT; 464 465 final int[] _edgeBuckets = edgeBuckets; 466 final int[] _edgeBucketCounts = edgeBucketCounts; 467 468 final int _boundsMinY = boundsMinY; 469 470 // each bucket is a linked list. this method adds ptr to the 471 // start of the "bucket"th linked list. 472 final int bucketIdx = firstCrossing - _boundsMinY; 473 474 // pointer from bucket 475 _unsafe.putInt(addr, _edgeBuckets[bucketIdx]); 476 addr += SIZE_INT; 477 // y max (inclusive) 478 _unsafe.putInt(addr, lastCrossing); 479 480 // Update buckets: 481 // directly the edge struct "pointer" 482 _edgeBuckets[bucketIdx] = edgePtr; 483 _edgeBucketCounts[bucketIdx] += 2; // 1 << 1 484 // last bit means edge end 485 _edgeBucketCounts[lastCrossing - _boundsMinY] |= 0x1; 486 487 // update sum of delta Y (subpixels): 488 edgeSumDeltaY += (lastCrossing - firstCrossing); 489 490 // update free pointer (ie length in bytes) 491 _edges.used += _SIZEOF_EDGE_BYTES; 492 493 if (DO_MONITORS) { 494 rdrCtx.stats.mon_rdr_addLine.stop(); 495 } 496 } 497 498 // END EDGE LIST 499 ////////////////////////////////////////////////////////////////////////////// 500 501 // Cache to store RLE-encoded coverage mask of the current primitive 502 final MarlinCache cache; 503 504 // Bounds of the drawing region, at subpixel precision. 505 private int boundsMinX, boundsMinY, boundsMaxX, boundsMaxY; 506 507 // Current winding rule 508 private int windingRule; 509 551 alphaLine = alphaLine_ref.initial; 552 553 this.cache = rdrCtx.cache; 554 555 crossings_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 556 aux_crossings_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 557 edgePtrs_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 558 aux_edgePtrs_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 559 560 crossings = crossings_ref.initial; 561 aux_crossings = aux_crossings_ref.initial; 562 edgePtrs = edgePtrs_ref.initial; 563 aux_edgePtrs = aux_edgePtrs_ref.initial; 564 565 blkFlags_ref = rdrCtx.newCleanIntArrayRef(INITIAL_ARRAY); // 1K = 1 tile line 566 blkFlags = blkFlags_ref.initial; 567 } 568 569 Renderer init(final int pix_boundsX, final int pix_boundsY, 570 final int pix_boundsWidth, final int pix_boundsHeight, 571 final int windingRule) { 572 573 this.windingRule = windingRule; 574 575 // bounds as half-open intervals: minX <= x < maxX and minY <= y < maxY 576 this.boundsMinX = pix_boundsX << SUBPIXEL_LG_POSITIONS_X; 577 this.boundsMaxX = 578 (pix_boundsX + pix_boundsWidth) << SUBPIXEL_LG_POSITIONS_X; 579 this.boundsMinY = pix_boundsY << SUBPIXEL_LG_POSITIONS_Y; 580 this.boundsMaxY = 581 (pix_boundsY + pix_boundsHeight) << SUBPIXEL_LG_POSITIONS_Y; 582 583 if (DO_LOG_BOUNDS) { 584 MarlinUtils.logInfo("boundsXY = [" + boundsMinX + " ... " 585 + boundsMaxX + "[ [" + boundsMinY + " ... " 586 + boundsMaxY + "["); 587 } 588 589 // see addLine: ceil(boundsMaxY) => boundsMaxY + 1 590 // +1 for edgeBucketCounts 591 final int edgeBucketsLength = (boundsMaxY - boundsMinY) + 1; 592 594 if (DO_STATS) { 595 rdrCtx.stats.stat_array_renderer_edgeBuckets 596 .add(edgeBucketsLength); 597 rdrCtx.stats.stat_array_renderer_edgeBucketCounts 598 .add(edgeBucketsLength); 599 } 600 edgeBuckets = edgeBuckets_ref.getArray(edgeBucketsLength); 601 edgeBucketCounts = edgeBucketCounts_ref.getArray(edgeBucketsLength); 602 } 603 604 edgeMinY = Integer.MAX_VALUE; 605 edgeMaxY = Integer.MIN_VALUE; 606 edgeMinX = Float.POSITIVE_INFINITY; 607 edgeMaxX = Float.NEGATIVE_INFINITY; 608 609 // reset used mark: 610 edgeCount = 0; 611 activeEdgeMaxUsed = 0; 612 edges.used = 0; 613 614 edgeSumDeltaY = 0; 615 616 return this; // fluent API 617 } 618 619 /** 620 * Disposes this renderer and recycle it clean up before reusing this instance 621 */ 622 void dispose() { 623 if (DO_STATS) { 624 rdrCtx.stats.stat_rdr_activeEdges.add(activeEdgeMaxUsed); 625 rdrCtx.stats.stat_rdr_edges.add(edges.used); 626 rdrCtx.stats.stat_rdr_edges_count.add(edges.used / SIZEOF_EDGE_BYTES); 627 rdrCtx.stats.hist_rdr_edges_count.add(edges.used / SIZEOF_EDGE_BYTES); 628 rdrCtx.stats.totalOffHeap += edges.length; 629 } 630 // Return arrays: 631 crossings = crossings_ref.putArray(crossings); 632 aux_crossings = aux_crossings_ref.putArray(aux_crossings); 633 634 edgePtrs = edgePtrs_ref.putArray(edgePtrs); 635 aux_edgePtrs = aux_edgePtrs_ref.putArray(aux_edgePtrs); 652 buckets_minY, 653 buckets_maxY + 1); 654 } else { 655 // unused arrays 656 edgeBuckets = edgeBuckets_ref.putArray(edgeBuckets, 0, 0); 657 edgeBucketCounts = edgeBucketCounts_ref.putArray(edgeBucketCounts, 0, 0); 658 } 659 660 // At last: resize back off-heap edges to initial size 661 if (edges.length != INITIAL_EDGES_CAPACITY) { 662 // note: may throw OOME: 663 edges.resize(INITIAL_EDGES_CAPACITY); 664 } 665 if (DO_CLEAN_DIRTY) { 666 // Force zero-fill dirty arrays: 667 edges.fill(BYTE_0); 668 } 669 if (DO_MONITORS) { 670 rdrCtx.stats.mon_rdr_endRendering.stop(); 671 } 672 } 673 674 private static float tosubpixx(final float pix_x) { 675 return F_SUBPIXEL_POSITIONS_X * pix_x; 676 } 677 678 private static float tosubpixy(final float pix_y) { 679 // shift y by -0.5 for fast ceil(y - 0.5): 680 return F_SUBPIXEL_POSITIONS_Y * pix_y - 0.5f; 681 } 682 683 @Override 684 public void moveTo(float pix_x0, float pix_y0) { 685 closePath(); 686 final float sx = tosubpixx(pix_x0); 687 final float sy = tosubpixy(pix_y0); 688 this.sx0 = sx; 689 this.sy0 = sy; 690 this.x0 = sx; 691 this.y0 = sy; 692 } 693 694 @Override 695 public void lineTo(float pix_x1, float pix_y1) { 696 final float x1 = tosubpixx(pix_x1); 697 final float y1 = tosubpixy(pix_y1); 698 addLine(x0, y0, x1, y1); 699 x0 = x1; 700 y0 = y1; 701 } 702 703 @Override 704 public void curveTo(float x1, float y1, 705 float x2, float y2, 706 float x3, float y3) 707 { 708 final float xe = tosubpixx(x3); 709 final float ye = tosubpixy(y3); 710 curve.set(x0, y0, tosubpixx(x1), tosubpixy(y1), 711 tosubpixx(x2), tosubpixy(y2), xe, ye); 712 curveBreakIntoLinesAndAdd(x0, y0, curve, xe, ye); 713 x0 = xe; 714 y0 = ye; 715 } 716 717 @Override 718 public void quadTo(float x1, float y1, float x2, float y2) { 719 final float xe = tosubpixx(x2); 720 final float ye = tosubpixy(y2); 721 curve.set(x0, y0, tosubpixx(x1), tosubpixy(y1), xe, ye); 722 quadBreakIntoLinesAndAdd(x0, y0, curve, xe, ye); 723 x0 = xe; 724 y0 = ye; 725 } 726 952 */ 953 if ((ptrLen < 10) || (numCrossings < 40)) { 954 if (DO_STATS) { 955 rdrCtx.stats.hist_rdr_crossings.add(numCrossings); 956 rdrCtx.stats.hist_rdr_crossings_adds.add(ptrLen); 957 } 958 959 /* 960 * threshold to use binary insertion sort instead of 961 * straight insertion sort (to reduce minimize comparisons). 962 */ 963 useBinarySearch = (numCrossings >= 20); 964 965 // if small enough: 966 lastCross = _MIN_VALUE; 967 968 for (i = 0; i < numCrossings; i++) { 969 // get the pointer to the edge 970 ecur = _edgePtrs[i]; 971 972 /* convert subpixel coordinates (float) into pixel 973 positions (int) for coming scanline */ 974 /* note: it is faster to always update edges even 975 if it is removed from AEL for coming or last scanline */ 976 977 // random access so use unsafe: 978 addr = addr0 + ecur; // ecur + OFF_F_CURX 979 980 // get current crossing: 981 curx = _unsafe.getInt(addr); 982 983 // update crossing with orientation at last bit: 984 cross = curx; 985 986 // Increment x using DDA (fixed point): 987 curx += _unsafe.getInt(addr + _OFF_BUMP_X); 988 989 // Increment error: 990 err = _unsafe.getInt(addr + _OFF_ERROR) 991 + _unsafe.getInt(addr + _OFF_BUMP_ERR); 992 993 // Manual carry handling: 1052 } 1053 } else { 1054 if (DO_STATS) { 1055 rdrCtx.stats.stat_rdr_crossings_msorts.add(numCrossings); 1056 rdrCtx.stats.hist_rdr_crossings_ratio 1057 .add((1000 * ptrLen) / numCrossings); 1058 rdrCtx.stats.hist_rdr_crossings_msorts.add(numCrossings); 1059 rdrCtx.stats.hist_rdr_crossings_msorts_adds.add(ptrLen); 1060 } 1061 1062 // Copy sorted data in auxiliary arrays 1063 // and perform insertion sort on almost sorted data 1064 // (ie i < prevNumCrossings): 1065 1066 lastCross = _MIN_VALUE; 1067 1068 for (i = 0; i < numCrossings; i++) { 1069 // get the pointer to the edge 1070 ecur = _edgePtrs[i]; 1071 1072 /* convert subpixel coordinates (float) into pixel 1073 positions (int) for coming scanline */ 1074 /* note: it is faster to always update edges even 1075 if it is removed from AEL for coming or last scanline */ 1076 1077 // random access so use unsafe: 1078 addr = addr0 + ecur; // ecur + OFF_F_CURX 1079 1080 // get current crossing: 1081 curx = _unsafe.getInt(addr); 1082 1083 // update crossing with orientation at last bit: 1084 cross = curx; 1085 1086 // Increment x using DDA (fixed point): 1087 curx += _unsafe.getInt(addr + _OFF_BUMP_X); 1088 1089 // Increment error: 1090 err = _unsafe.getInt(addr + _OFF_ERROR) 1091 + _unsafe.getInt(addr + _OFF_BUMP_ERR); 1092 1093 // Manual carry handling: 1159 prev = curx = x0; 1160 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1161 // last bit contains orientation (0 or 1) 1162 crorientation = ((curxo & 0x1) << 1) - 1; 1163 1164 if (windingRuleEvenOdd) { 1165 sum = crorientation; 1166 1167 // Even Odd winding rule: take care of mask ie sum(orientations) 1168 for (i = 1; i < numCrossings; i++) { 1169 curxo = _crossings[i]; 1170 curx = curxo >> 1; 1171 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1172 // last bit contains orientation (0 or 1) 1173 crorientation = ((curxo & 0x1) << 1) - 1; 1174 1175 if ((sum & 0x1) != 0) { 1176 // TODO: perform line clipping on left-right sides 1177 // to avoid such bound checks: 1178 x0 = (prev > bboxx0) ? prev : bboxx0; 1179 x1 = (curx < bboxx1) ? curx : bboxx1; 1180 1181 if (x0 < x1) { 1182 x0 -= bboxx0; // turn x0, x1 from coords to indices 1183 x1 -= bboxx0; // in the alpha array. 1184 1185 pix_x = x0 >> _SUBPIXEL_LG_POSITIONS_X; 1186 pix_xmaxm1 = (x1 - 1) >> _SUBPIXEL_LG_POSITIONS_X; 1187 1188 if (pix_x == pix_xmaxm1) { 1189 // Start and end in same pixel 1190 tmp = (x1 - x0); // number of subpixels 1191 _alpha[pix_x ] += tmp; 1192 _alpha[pix_x + 1] -= tmp; 1193 1194 if (useBlkFlags) { 1195 // flag used blocks: 1196 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1197 } 1198 } else { 1199 tmp = (x0 & _SUBPIXEL_MASK_X); 1200 _alpha[pix_x ] 1201 += (_SUBPIXEL_POSITIONS_X - tmp); 1202 _alpha[pix_x + 1] 1203 += tmp; 1204 1205 pix_xmax = x1 >> _SUBPIXEL_LG_POSITIONS_X; 1206 1207 tmp = (x1 & _SUBPIXEL_MASK_X); 1208 _alpha[pix_xmax ] 1209 -= (_SUBPIXEL_POSITIONS_X - tmp); 1210 _alpha[pix_xmax + 1] 1211 -= tmp; 1212 1213 if (useBlkFlags) { 1214 // flag used blocks: 1215 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1216 _blkFlags[pix_xmax >> _BLK_SIZE_LG] = 1; 1217 } 1218 } 1219 } 1220 } 1221 1222 sum += crorientation; 1223 prev = curx; 1224 } 1225 } else { 1226 // Non-zero winding rule: optimize that case (default) 1227 // and avoid processing intermediate crossings 1228 for (i = 1, sum = 0;; i++) { 1229 sum += crorientation; 1230 1231 if (sum != 0) { 1232 // prev = min(curx) 1233 if (prev > curx) { 1234 prev = curx; 1235 } 1236 } else { 1237 // TODO: perform line clipping on left-right sides 1238 // to avoid such bound checks: 1239 x0 = (prev > bboxx0) ? prev : bboxx0; 1240 x1 = (curx < bboxx1) ? curx : bboxx1; 1241 1242 if (x0 < x1) { 1243 x0 -= bboxx0; // turn x0, x1 from coords to indices 1244 x1 -= bboxx0; // in the alpha array. 1245 1246 pix_x = x0 >> _SUBPIXEL_LG_POSITIONS_X; 1247 pix_xmaxm1 = (x1 - 1) >> _SUBPIXEL_LG_POSITIONS_X; 1248 1249 if (pix_x == pix_xmaxm1) { 1250 // Start and end in same pixel 1251 tmp = (x1 - x0); // number of subpixels 1252 _alpha[pix_x ] += tmp; 1253 _alpha[pix_x + 1] -= tmp; 1254 1255 if (useBlkFlags) { 1256 // flag used blocks: 1257 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1258 } 1259 } else { 1260 tmp = (x0 & _SUBPIXEL_MASK_X); 1261 _alpha[pix_x ] 1262 += (_SUBPIXEL_POSITIONS_X - tmp); 1263 _alpha[pix_x + 1] 1264 += tmp; 1265 1266 pix_xmax = x1 >> _SUBPIXEL_LG_POSITIONS_X; 1267 1268 tmp = (x1 & _SUBPIXEL_MASK_X); 1269 _alpha[pix_xmax ] 1270 -= (_SUBPIXEL_POSITIONS_X - tmp); 1271 _alpha[pix_xmax + 1] 1272 -= tmp; 1273 1274 if (useBlkFlags) { 1275 // flag used blocks: 1276 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1277 _blkFlags[pix_xmax >> _BLK_SIZE_LG] = 1; 1278 } 1279 } 1280 } 1281 prev = _MAX_VALUE; 1282 } 1283 1284 if (i == numCrossings) { 1285 break; 1286 } 1287 1288 curxo = _crossings[i]; 1289 curx = curxo >> 1; 1290 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1291 // last bit contains orientation (0 or 1) 1292 crorientation = ((curxo & 0x1) << 1) - 1; 1293 } 1294 } 1295 } // numCrossings > 0 1296 1297 // even if this last row had no crossings, alpha will be zeroed 1298 // from the last emitRow call. But this doesn't matter because 1299 // maxX < minX, so no row will be emitted to the MarlinCache. 1300 if ((y & _SUBPIXEL_MASK_Y) == _SUBPIXEL_MASK_Y) { 1301 lastY = y >> _SUBPIXEL_LG_POSITIONS_Y; 1302 1303 // convert subpixel to pixel coordinate within boundaries: 1304 minX = FloatMath.max(minX, bboxx0) >> _SUBPIXEL_LG_POSITIONS_X; 1305 maxX = FloatMath.min(maxX, bboxx1) >> _SUBPIXEL_LG_POSITIONS_X; 1306 1307 if (maxX >= minX) { 1308 // note: alpha array will be zeroed by copyAARow() 1309 // +2 because alpha [pix_minX; pix_maxX+1] 1310 // fix range [x0; x1[ 1311 copyAARow(_alpha, lastY, minX, maxX + 2, useBlkFlags); 1312 1313 // speculative for next pixel row (scanline coherence): 1314 if (_enableBlkFlagsHeuristics) { 1315 // Use block flags if large pixel span and few crossings: 1316 // ie mean(distance between crossings) is larger than 1317 // 1 block size; 1318 1319 // fast check width: 1320 maxX -= minX; 1321 1322 // if stroking: numCrossings /= 2 1323 // => shift numCrossings by 1 1324 // condition = (width / (numCrossings - 1)) > blockSize 1325 useBlkFlags = (maxX > _BLK_SIZE) && (maxX > 1326 (((numCrossings >> stroking) - 1) << _BLK_SIZE_LG)); 1327 1328 if (DO_STATS) { 1329 tmp = FloatMath.max(1, 1330 ((numCrossings >> stroking) - 1)); 1331 rdrCtx.stats.hist_tile_generator_encoding_dist 1333 } 1334 } 1335 } else { 1336 _cache.clearAARow(lastY); 1337 } 1338 minX = _MAX_VALUE; 1339 maxX = _MIN_VALUE; 1340 } 1341 } // scan line iterator 1342 1343 // Emit final row 1344 y--; 1345 y >>= _SUBPIXEL_LG_POSITIONS_Y; 1346 1347 // convert subpixel to pixel coordinate within boundaries: 1348 minX = FloatMath.max(minX, bboxx0) >> _SUBPIXEL_LG_POSITIONS_X; 1349 maxX = FloatMath.min(maxX, bboxx1) >> _SUBPIXEL_LG_POSITIONS_X; 1350 1351 if (maxX >= minX) { 1352 // note: alpha array will be zeroed by copyAARow() 1353 // +2 because alpha [pix_minX; pix_maxX+1] 1354 // fix range [x0; x1[ 1355 copyAARow(_alpha, y, minX, maxX + 2, useBlkFlags); 1356 } else if (y != lastY) { 1357 _cache.clearAARow(y); 1358 } 1359 1360 // update member: 1361 edgeCount = numCrossings; 1362 prevUseBlkFlags = useBlkFlags; 1363 1364 if (DO_STATS) { 1365 // update max used mark 1366 activeEdgeMaxUsed = _arrayMaxUsed; 1367 } 1368 } 1369 1370 boolean endRendering() { 1371 if (DO_MONITORS) { 1372 rdrCtx.stats.mon_rdr_endRendering.start(); 1373 } 1374 if (edgeMinY == Integer.MAX_VALUE) { 1375 return false; // undefined edges bounds 1376 } 1377 1378 final int _boundsMinY = boundsMinY; 1379 final int _boundsMaxY = boundsMaxY; 1380 1381 // bounds as inclusive intervals 1382 final int spminX = FloatMath.max(FloatMath.ceil_int(edgeMinX - 0.5f), boundsMinX); 1383 final int spmaxX = FloatMath.min(FloatMath.ceil_int(edgeMaxX - 0.5f), boundsMaxX - 1); 1384 1385 // edge Min/Max Y are already rounded to subpixels within bounds: 1386 final int spminY = edgeMinY; 1387 final int spmaxY; 1388 int maxY = edgeMaxY; 1389 1390 if (maxY <= _boundsMaxY - 1) { 1391 spmaxY = maxY; 1392 } else { 1393 spmaxY = _boundsMaxY - 1; 1394 maxY = _boundsMaxY; 1395 } 1396 buckets_minY = spminY - _boundsMinY; 1397 buckets_maxY = maxY - _boundsMinY; 1398 1399 if (DO_LOG_BOUNDS) { 1400 MarlinUtils.logInfo("edgesXY = [" + edgeMinX + " ... " + edgeMaxX 1401 + "][" + edgeMinY + " ... " + edgeMaxY + "]"); 1402 MarlinUtils.logInfo("spXY = [" + spminX + " ... " + spmaxX 1403 + "][" + spminY + " ... " + spmaxY + "]"); 1404 } 1405 1406 // test clipping for shapes out of bounds 1407 if ((spminX > spmaxX) || (spminY > spmaxY)) { 1408 return false; 1409 } 1410 1411 // half open intervals 1412 // inclusive: 1413 final int pminX = spminX >> SUBPIXEL_LG_POSITIONS_X; 1414 // exclusive: 1415 final int pmaxX = (spmaxX + SUBPIXEL_MASK_X) >> SUBPIXEL_LG_POSITIONS_X; 1416 // inclusive: 1417 final int pminY = spminY >> SUBPIXEL_LG_POSITIONS_Y; 1418 // exclusive: 1419 final int pmaxY = (spmaxY + SUBPIXEL_MASK_Y) >> SUBPIXEL_LG_POSITIONS_Y; 1420 1421 // store BBox to answer ptg.getBBox(): 1422 this.cache.init(pminX, pminY, pmaxX, pmaxY, edgeSumDeltaY); 1423 1424 // Heuristics for using block flags: 1425 if (ENABLE_BLOCK_FLAGS) { 1426 enableBlkFlags = this.cache.useRLE; 1427 prevUseBlkFlags = enableBlkFlags && !ENABLE_BLOCK_FLAGS_HEURISTICS; 1428 1429 if (enableBlkFlags) { 1430 // ensure blockFlags array is large enough: 1431 // note: +2 to ensure enough space left at end 1432 final int nxTiles = ((pmaxX - pminX) >> TILE_SIZE_LG) + 2; 1433 if (nxTiles > INITIAL_ARRAY) { 1434 blkFlags = blkFlags_ref.getArray(nxTiles); 1435 } 1436 } 1437 } 1438 1439 // memorize the rendering bounding box: 1440 /* note: bbox_spminX and bbox_spmaxX must be pixel boundaries 1441 to have correct coverage computation */ 1442 // inclusive: 1443 bbox_spminX = pminX << SUBPIXEL_LG_POSITIONS_X; 1444 // exclusive: 1445 bbox_spmaxX = pmaxX << SUBPIXEL_LG_POSITIONS_X; 1446 // inclusive: 1447 bbox_spminY = spminY; 1448 // exclusive: 1449 bbox_spmaxY = FloatMath.min(spmaxY + 1, pmaxY << SUBPIXEL_LG_POSITIONS_Y); 1450 1451 if (DO_LOG_BOUNDS) { 1452 MarlinUtils.logInfo("pXY = [" + pminX + " ... " + pmaxX 1453 + "[ [" + pminY + " ... " + pmaxY + "["); 1454 MarlinUtils.logInfo("bbox_spXY = [" + bbox_spminX + " ... " 1455 + bbox_spmaxX + "[ [" + bbox_spminY + " ... " 1456 + bbox_spmaxY + "["); 1457 } 1458 1459 // Prepare alpha line: 1460 // add 2 to better deal with the last pixel in a pixel row. 1461 final int width = (pmaxX - pminX) + 2; 1462 1463 // Useful when processing tile line by tile line 1464 if (width > INITIAL_AA_ARRAY) { 1465 if (DO_STATS) { 1466 rdrCtx.stats.stat_array_renderer_alphaline.add(width); 1467 } 1468 alphaLine = alphaLine_ref.getArray(width); 1469 } 1487 // avoid rendering for last call to nextTile() 1488 if (fixed_spminY < bbox_spmaxY) { 1489 // process a complete tile line ie scanlines for 32 rows 1490 final int spmaxY = FloatMath.min(bbox_spmaxY, spminY + SUBPIXEL_TILE); 1491 1492 // process tile line [0 - 32] 1493 cache.resetTileLine(pminY); 1494 1495 // Process only one tile line: 1496 _endRendering(fixed_spminY, spmaxY); 1497 } 1498 if (DO_MONITORS) { 1499 rdrCtx.stats.mon_rdr_endRendering_Y.stop(); 1500 } 1501 } 1502 1503 void copyAARow(final int[] alphaRow, 1504 final int pix_y, final int pix_from, final int pix_to, 1505 final boolean useBlockFlags) 1506 { 1507 if (useBlockFlags) { 1508 if (DO_STATS) { 1509 rdrCtx.stats.hist_tile_generator_encoding.add(1); 1510 } 1511 cache.copyAARowRLE_WithBlockFlags(blkFlags, alphaRow, pix_y, pix_from, pix_to); 1512 } else { 1513 if (DO_STATS) { 1514 rdrCtx.stats.hist_tile_generator_encoding.add(0); 1515 } 1516 cache.copyAARowNoRLE(alphaRow, pix_y, pix_from, pix_to); 1517 } 1518 } 1519 } | 1 /* 2 * Copyright (c) 2007, 2017, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. Oracle designates this 8 * particular file as subject to the "Classpath" exception as provided 9 * by Oracle in the LICENSE file that accompanied this code. 10 * 11 * This code is distributed in the hope that it will be useful, but WITHOUT 12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 14 * version 2 for more details (a copy is included in the LICENSE file that 15 * accompanied this code). 16 * 17 * You should have received a copy of the GNU General Public License version 18 * 2 along with this work; if not, write to the Free Software Foundation, 19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 20 * 21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 22 * or visit www.oracle.com if you need additional information or have any 23 * questions. 24 */ 25 26 package sun.java2d.marlin; 27 28 import sun.awt.geom.PathConsumer2D; 29 import static sun.java2d.marlin.OffHeapArray.SIZE_INT; 30 import jdk.internal.misc.Unsafe; 31 32 final class Renderer implements PathConsumer2D, MarlinRenderer { 33 34 static final boolean DISABLE_RENDER = false; 35 36 static final boolean ENABLE_BLOCK_FLAGS = MarlinProperties.isUseTileFlags(); 37 static final boolean ENABLE_BLOCK_FLAGS_HEURISTICS = MarlinProperties.isUseTileFlagsWithHeuristics(); 38 39 private static final int ALL_BUT_LSB = 0xFFFFFFFE; 40 private static final int ERR_STEP_MAX = 0x7FFFFFFF; // = 2^31 - 1 41 42 private static final double POWER_2_TO_32 = 0x1.0p32d; 43 44 // use float to make tosubpix methods faster (no int to float conversion) 45 static final float SUBPIXEL_SCALE_X = (float) SUBPIXEL_POSITIONS_X; 46 static final float SUBPIXEL_SCALE_Y = (float) SUBPIXEL_POSITIONS_Y; 47 static final int SUBPIXEL_MASK_X = SUBPIXEL_POSITIONS_X - 1; 48 static final int SUBPIXEL_MASK_Y = SUBPIXEL_POSITIONS_Y - 1; 49 50 // number of subpixels corresponding to a tile line 51 private static final int SUBPIXEL_TILE 52 = TILE_H << SUBPIXEL_LG_POSITIONS_Y; 53 54 // 2048 (pixelSize) pixels (height) x 8 subpixels = 64K 55 static final int INITIAL_BUCKET_ARRAY 56 = INITIAL_PIXEL_DIM * SUBPIXEL_POSITIONS_Y; 57 58 // crossing capacity = edges count / 4 ~ 1024 59 static final int INITIAL_CROSSING_COUNT = INITIAL_EDGES_COUNT >> 2; 60 61 public static final int WIND_EVEN_ODD = 0; 62 public static final int WIND_NON_ZERO = 1; 63 64 // common to all types of input path segments. 65 // OFFSET as bytes 66 // only integer values: 67 public static final long OFF_CURX_OR = 0; 68 public static final long OFF_ERROR = OFF_CURX_OR + SIZE_INT; 69 public static final long OFF_BUMP_X = OFF_ERROR + SIZE_INT; 70 public static final long OFF_BUMP_ERR = OFF_BUMP_X + SIZE_INT; 71 public static final long OFF_NEXT = OFF_BUMP_ERR + SIZE_INT; 72 public static final long OFF_YMAX = OFF_NEXT + SIZE_INT; 73 74 // size of one edge in bytes 75 public static final int SIZEOF_EDGE_BYTES = (int)(OFF_YMAX + SIZE_INT); 76 77 // curve break into lines 78 // cubic error in subpixels to decrement step 79 private static final float CUB_DEC_ERR_SUBPIX 80 = MarlinProperties.getCubicDecD2() * (NORM_SUBPIXELS / 8.0f); // 1 pixel 81 // cubic error in subpixels to increment step 82 private static final float CUB_INC_ERR_SUBPIX 83 = MarlinProperties.getCubicIncD1() * (NORM_SUBPIXELS / 8.0f); // 0.4 pixel 84 85 // TestNonAARasterization (JDK-8170879): cubics 86 // bad paths (59294/100000 == 59,29%, 94335 bad pixels (avg = 1,59), 3966 warnings (avg = 0,07) 87 88 // cubic bind length to decrement step 89 public static final float CUB_DEC_BND 90 = 8.0f * CUB_DEC_ERR_SUBPIX; 91 // cubic bind length to increment step 92 public static final float CUB_INC_BND 93 = 8.0f * CUB_INC_ERR_SUBPIX; 94 95 // cubic countlg 96 public static final int CUB_COUNT_LG = 2; 97 // cubic count = 2^countlg 98 private static final int CUB_COUNT = 1 << CUB_COUNT_LG; 99 // cubic count^2 = 4^countlg 100 private static final int CUB_COUNT_2 = 1 << (2 * CUB_COUNT_LG); 101 // cubic count^3 = 8^countlg 102 private static final int CUB_COUNT_3 = 1 << (3 * CUB_COUNT_LG); 103 // cubic dt = 1 / count 104 private static final float CUB_INV_COUNT = 1.0f / CUB_COUNT; 105 // cubic dt^2 = 1 / count^2 = 1 / 4^countlg 106 private static final float CUB_INV_COUNT_2 = 1.0f / CUB_COUNT_2; 107 // cubic dt^3 = 1 / count^3 = 1 / 8^countlg 108 private static final float CUB_INV_COUNT_3 = 1.0f / CUB_COUNT_3; 109 110 // quad break into lines 111 // quadratic error in subpixels 112 private static final float QUAD_DEC_ERR_SUBPIX 113 = MarlinProperties.getQuadDecD2() * (NORM_SUBPIXELS / 8.0f); // 0.5 pixel 114 115 // TestNonAARasterization (JDK-8170879): quads 116 // bad paths (62916/100000 == 62,92%, 103818 bad pixels (avg = 1,65), 6514 warnings (avg = 0,10) 117 118 // quadratic bind length to decrement step 119 public static final float QUAD_DEC_BND 120 = 8.0f * QUAD_DEC_ERR_SUBPIX; 121 122 ////////////////////////////////////////////////////////////////////////////// 123 // SCAN LINE 124 ////////////////////////////////////////////////////////////////////////////// 125 // crossings ie subpixel edge x coordinates 126 private int[] crossings; 127 // auxiliary storage for crossings (merge sort) 128 private int[] aux_crossings; 129 130 // indices into the segment pointer lists. They indicate the "active" 131 // sublist in the segment lists (the portion of the list that contains 132 // all the segments that cross the next scan line). 133 private int edgeCount; 134 private int[] edgePtrs; 135 // auxiliary storage for edge pointers (merge sort) 136 private int[] aux_edgePtrs; 137 138 // max used for both edgePtrs and crossings (stats only) 139 private int activeEdgeMaxUsed; 140 141 // crossings ref (dirty) 142 private final IntArrayCache.Reference crossings_ref; 143 // edgePtrs ref (dirty) 144 private final IntArrayCache.Reference edgePtrs_ref; 145 // merge sort initial arrays (large enough to satisfy most usages) (1024) 146 // aux_crossings ref (dirty) 147 private final IntArrayCache.Reference aux_crossings_ref; 148 // aux_edgePtrs ref (dirty) 149 private final IntArrayCache.Reference aux_edgePtrs_ref; 150 151 ////////////////////////////////////////////////////////////////////////////// 152 // EDGE LIST 153 ////////////////////////////////////////////////////////////////////////////// 154 private int edgeMinY = Integer.MAX_VALUE; 155 private int edgeMaxY = Integer.MIN_VALUE; 156 private float edgeMinX = Float.POSITIVE_INFINITY; 157 private float edgeMaxX = Float.NEGATIVE_INFINITY; 158 159 // edges [ints] stored in off-heap memory 160 private final OffHeapArray edges; 161 162 private int[] edgeBuckets; 163 private int[] edgeBucketCounts; // 2*newedges + (1 if pruning needed) 164 // used range for edgeBuckets / edgeBucketCounts 165 private int buckets_minY; 166 private int buckets_maxY; 167 168 // edgeBuckets ref (clean) 169 private final IntArrayCache.Reference edgeBuckets_ref; 170 // edgeBucketCounts ref (clean) 171 private final IntArrayCache.Reference edgeBucketCounts_ref; 172 173 // Flattens using adaptive forward differencing. This only carries out 174 // one iteration of the AFD loop. All it does is update AFD variables (i.e. 175 // X0, Y0, D*[X|Y], COUNT; not variables used for computing scanline crossings). 176 private void quadBreakIntoLinesAndAdd(float x0, float y0, 177 final Curve c, 178 final float x2, final float y2) 179 { 180 int count = 1; // dt = 1 / count 181 182 // maximum(ddX|Y) = norm(dbx, dby) * dt^2 (= 1) 183 float maxDD = Math.abs(c.dbx) + Math.abs(c.dby); 184 185 final float _DEC_BND = QUAD_DEC_BND; 186 187 while (maxDD >= _DEC_BND) { 188 // divide step by half: 189 maxDD /= 4.0f; // error divided by 2^2 = 4 190 191 count <<= 1; 192 if (DO_STATS) { 193 rdrCtx.stats.stat_rdr_quadBreak_dec.add(count); 194 } 195 } 196 197 int nL = 0; // line count 198 if (count > 1) { 199 final float icount = 1.0f / count; // dt 200 final float icount2 = icount * icount; // dt^2 201 202 final float ddx = c.dbx * icount2; 203 final float ddy = c.dby * icount2; 204 float dx = c.bx * icount2 + c.cx * icount; 205 float dy = c.by * icount2 + c.cy * icount; 206 207 float x1, y1; 208 209 while (--count > 0) { 210 x1 = x0 + dx; 211 dx += ddx; 212 y1 = y0 + dy; 213 dy += ddy; 214 215 addLine(x0, y0, x1, y1); 216 217 if (DO_STATS) { nL++; } 218 x0 = x1; 219 y0 = y1; 226 } 227 } 228 229 // x0, y0 and x3,y3 are the endpoints of the curve. We could compute these 230 // using c.xat(0),c.yat(0) and c.xat(1),c.yat(1), but this might introduce 231 // numerical errors, and our callers already have the exact values. 232 // Another alternative would be to pass all the control points, and call 233 // c.set here, but then too many numbers are passed around. 234 private void curveBreakIntoLinesAndAdd(float x0, float y0, 235 final Curve c, 236 final float x3, final float y3) 237 { 238 int count = CUB_COUNT; 239 final float icount = CUB_INV_COUNT; // dt 240 final float icount2 = CUB_INV_COUNT_2; // dt^2 241 final float icount3 = CUB_INV_COUNT_3; // dt^3 242 243 // the dx and dy refer to forward differencing variables, not the last 244 // coefficients of the "points" polynomial 245 float dddx, dddy, ddx, ddy, dx, dy; 246 dddx = 2.0f * c.dax * icount3; 247 dddy = 2.0f * c.day * icount3; 248 ddx = dddx + c.dbx * icount2; 249 ddy = dddy + c.dby * icount2; 250 dx = c.ax * icount3 + c.bx * icount2 + c.cx * icount; 251 dy = c.ay * icount3 + c.by * icount2 + c.cy * icount; 252 253 // we use x0, y0 to walk the line 254 float x1 = x0, y1 = y0; 255 int nL = 0; // line count 256 257 final float _DEC_BND = CUB_DEC_BND; 258 final float _INC_BND = CUB_INC_BND; 259 260 while (count > 0) { 261 // divide step by half: 262 while (Math.abs(ddx) + Math.abs(ddy) >= _DEC_BND) { 263 dddx /= 8.0f; 264 dddy /= 8.0f; 265 ddx = ddx / 4.0f - dddx; 266 ddy = ddy / 4.0f - dddy; 267 dx = (dx - ddx) / 2.0f; 268 dy = (dy - ddy) / 2.0f; 269 270 count <<= 1; 271 if (DO_STATS) { 272 rdrCtx.stats.stat_rdr_curveBreak_dec.add(count); 273 } 274 } 275 276 // double step: 277 // can only do this on even "count" values, because we must divide count by 2 278 while (count % 2 == 0 279 && Math.abs(dx) + Math.abs(dy) <= _INC_BND) 280 { 281 dx = 2.0f * dx + ddx; 282 dy = 2.0f * dy + ddy; 283 ddx = 4.0f * (ddx + dddx); 284 ddy = 4.0f * (ddy + dddy); 285 dddx *= 8.0f; 286 dddy *= 8.0f; 287 288 count >>= 1; 289 if (DO_STATS) { 290 rdrCtx.stats.stat_rdr_curveBreak_inc.add(count); 291 } 292 } 293 if (--count > 0) { 294 x1 += dx; 295 dx += ddx; 296 ddx += dddx; 297 y1 += dy; 298 dy += ddy; 299 ddy += dddy; 300 } else { 301 x1 = x3; 302 y1 = y3; 303 } 304 305 addLine(x0, y0, x1, y1); 306 314 } 315 316 private void addLine(float x1, float y1, float x2, float y2) { 317 if (DO_MONITORS) { 318 rdrCtx.stats.mon_rdr_addLine.start(); 319 } 320 if (DO_STATS) { 321 rdrCtx.stats.stat_rdr_addLine.add(1); 322 } 323 int or = 1; // orientation of the line. 1 if y increases, 0 otherwise. 324 if (y2 < y1) { 325 or = 0; 326 float tmp = y2; 327 y2 = y1; 328 y1 = tmp; 329 tmp = x2; 330 x2 = x1; 331 x1 = tmp; 332 } 333 334 // convert subpixel coordinates [float] into pixel positions [int] 335 336 // The index of the pixel that holds the next HPC is at ceil(trueY - 0.5) 337 // Since y1 and y2 are biased by -0.5 in tosubpixy(), this is simply 338 // ceil(y1) or ceil(y2) 339 // upper integer (inclusive) 340 final int firstCrossing = FloatMath.max(FloatMath.ceil_int(y1), boundsMinY); 341 342 // note: use boundsMaxY (last Y exclusive) to compute correct coverage 343 // upper integer (exclusive) 344 final int lastCrossing = FloatMath.min(FloatMath.ceil_int(y2), boundsMaxY); 345 346 /* skip horizontal lines in pixel space and clip edges 347 out of y range [boundsMinY; boundsMaxY] */ 348 if (firstCrossing >= lastCrossing) { 349 if (DO_MONITORS) { 350 rdrCtx.stats.mon_rdr_addLine.stop(); 351 } 352 if (DO_STATS) { 353 rdrCtx.stats.stat_rdr_addLine_skip.add(1); 354 } 355 return; 356 } 357 358 // edge min/max X/Y are in subpixel space (half-open interval): 359 // note: Use integer crossings to ensure consistent range within 360 // edgeBuckets / edgeBucketCounts arrays in case of NaN values (int = 0) 361 if (firstCrossing < edgeMinY) { 362 edgeMinY = firstCrossing; 363 } 364 if (lastCrossing > edgeMaxY) { 365 edgeMaxY = lastCrossing; 366 } 367 368 // Use double-precision for improved accuracy: 369 final double x1d = x1; 370 final double y1d = y1; 371 final double slope = (x1d - x2) / (y1d - y2); 372 373 if (slope >= 0.0d) { // <==> x1 < x2 374 if (x1 < edgeMinX) { 375 edgeMinX = x1; 376 } 377 if (x2 > edgeMaxX) { 378 edgeMaxX = x2; 379 } 380 } else { 381 if (x2 < edgeMinX) { 382 edgeMinX = x2; 383 } 384 if (x1 > edgeMaxX) { 385 edgeMaxX = x1; 386 } 387 } 388 389 // local variables for performance: 390 final int _SIZEOF_EDGE_BYTES = SIZEOF_EDGE_BYTES; 391 392 final OffHeapArray _edges = edges; 393 416 417 // The x value must be bumped up to its position at the next HPC we will evaluate. 418 // "firstcrossing" is the (sub)pixel number where the next crossing occurs 419 // thus, the actual coordinate of the next HPC is "firstcrossing + 0.5" 420 // so the Y distance we cover is "firstcrossing + 0.5 - trueY". 421 // Note that since y1 (and y2) are already biased by -0.5 in tosubpixy(), we have 422 // y1 = trueY - 0.5 423 // trueY = y1 + 0.5 424 // firstcrossing + 0.5 - trueY = firstcrossing + 0.5 - (y1 + 0.5) 425 // = firstcrossing - y1 426 // The x coordinate at that HPC is then: 427 // x1_intercept = x1 + (firstcrossing - y1) * slope 428 // The next VPC is then given by: 429 // VPC index = ceil(x1_intercept - 0.5), or alternately 430 // VPC index = floor(x1_intercept - 0.5 + 1 - epsilon) 431 // epsilon is hard to pin down in floating point, but easy in fixed point, so if 432 // we convert to fixed point then these operations get easier: 433 // long x1_fixed = x1_intercept * 2^32; (fixed point 32.32 format) 434 // curx = next VPC = fixed_floor(x1_fixed - 2^31 + 2^32 - 1) 435 // = fixed_floor(x1_fixed + 2^31 - 1) 436 // = fixed_floor(x1_fixed + 0x7FFFFFFF) 437 // and error = fixed_fract(x1_fixed + 0x7FFFFFFF) 438 final double x1_intercept = x1d + (firstCrossing - y1d) * slope; 439 440 // inlined scalb(x1_intercept, 32): 441 final long x1_fixed_biased = ((long) (POWER_2_TO_32 * x1_intercept)) 442 + 0x7FFFFFFFL; 443 // curx: 444 // last bit corresponds to the orientation 445 _unsafe.putInt(addr, (((int) (x1_fixed_biased >> 31L)) & ALL_BUT_LSB) | or); 446 addr += SIZE_INT; 447 _unsafe.putInt(addr, ((int) x1_fixed_biased) >>> 1); 448 addr += SIZE_INT; 449 450 // inlined scalb(slope, 32): 451 final long slope_fixed = (long) (POWER_2_TO_32 * slope); 452 453 // last bit set to 0 to keep orientation: 454 _unsafe.putInt(addr, (((int) (slope_fixed >> 31L)) & ALL_BUT_LSB)); 455 addr += SIZE_INT; 456 _unsafe.putInt(addr, ((int) slope_fixed) >>> 1); 457 addr += SIZE_INT; 458 459 final int[] _edgeBuckets = edgeBuckets; 460 final int[] _edgeBucketCounts = edgeBucketCounts; 461 462 final int _boundsMinY = boundsMinY; 463 464 // each bucket is a linked list. this method adds ptr to the 465 // start of the "bucket"th linked list. 466 final int bucketIdx = firstCrossing - _boundsMinY; 467 468 // pointer from bucket 469 _unsafe.putInt(addr, _edgeBuckets[bucketIdx]); 470 addr += SIZE_INT; 471 // y max (exclusive) 472 _unsafe.putInt(addr, lastCrossing); 473 474 // Update buckets: 475 // directly the edge struct "pointer" 476 _edgeBuckets[bucketIdx] = edgePtr; 477 _edgeBucketCounts[bucketIdx] += 2; // 1 << 1 478 // last bit means edge end 479 _edgeBucketCounts[lastCrossing - _boundsMinY] |= 0x1; 480 481 // update free pointer (ie length in bytes) 482 _edges.used += _SIZEOF_EDGE_BYTES; 483 484 if (DO_MONITORS) { 485 rdrCtx.stats.mon_rdr_addLine.stop(); 486 } 487 } 488 489 // END EDGE LIST 490 ////////////////////////////////////////////////////////////////////////////// 491 492 // Cache to store RLE-encoded coverage mask of the current primitive 493 final MarlinCache cache; 494 495 // Bounds of the drawing region, at subpixel precision. 496 private int boundsMinX, boundsMinY, boundsMaxX, boundsMaxY; 497 498 // Current winding rule 499 private int windingRule; 500 542 alphaLine = alphaLine_ref.initial; 543 544 this.cache = rdrCtx.cache; 545 546 crossings_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 547 aux_crossings_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 548 edgePtrs_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 549 aux_edgePtrs_ref = rdrCtx.newDirtyIntArrayRef(INITIAL_CROSSING_COUNT); // 2K 550 551 crossings = crossings_ref.initial; 552 aux_crossings = aux_crossings_ref.initial; 553 edgePtrs = edgePtrs_ref.initial; 554 aux_edgePtrs = aux_edgePtrs_ref.initial; 555 556 blkFlags_ref = rdrCtx.newCleanIntArrayRef(INITIAL_ARRAY); // 1K = 1 tile line 557 blkFlags = blkFlags_ref.initial; 558 } 559 560 Renderer init(final int pix_boundsX, final int pix_boundsY, 561 final int pix_boundsWidth, final int pix_boundsHeight, 562 final int windingRule) 563 { 564 this.windingRule = windingRule; 565 566 // bounds as half-open intervals: minX <= x < maxX and minY <= y < maxY 567 this.boundsMinX = pix_boundsX << SUBPIXEL_LG_POSITIONS_X; 568 this.boundsMaxX = 569 (pix_boundsX + pix_boundsWidth) << SUBPIXEL_LG_POSITIONS_X; 570 this.boundsMinY = pix_boundsY << SUBPIXEL_LG_POSITIONS_Y; 571 this.boundsMaxY = 572 (pix_boundsY + pix_boundsHeight) << SUBPIXEL_LG_POSITIONS_Y; 573 574 if (DO_LOG_BOUNDS) { 575 MarlinUtils.logInfo("boundsXY = [" + boundsMinX + " ... " 576 + boundsMaxX + "[ [" + boundsMinY + " ... " 577 + boundsMaxY + "["); 578 } 579 580 // see addLine: ceil(boundsMaxY) => boundsMaxY + 1 581 // +1 for edgeBucketCounts 582 final int edgeBucketsLength = (boundsMaxY - boundsMinY) + 1; 583 585 if (DO_STATS) { 586 rdrCtx.stats.stat_array_renderer_edgeBuckets 587 .add(edgeBucketsLength); 588 rdrCtx.stats.stat_array_renderer_edgeBucketCounts 589 .add(edgeBucketsLength); 590 } 591 edgeBuckets = edgeBuckets_ref.getArray(edgeBucketsLength); 592 edgeBucketCounts = edgeBucketCounts_ref.getArray(edgeBucketsLength); 593 } 594 595 edgeMinY = Integer.MAX_VALUE; 596 edgeMaxY = Integer.MIN_VALUE; 597 edgeMinX = Float.POSITIVE_INFINITY; 598 edgeMaxX = Float.NEGATIVE_INFINITY; 599 600 // reset used mark: 601 edgeCount = 0; 602 activeEdgeMaxUsed = 0; 603 edges.used = 0; 604 605 return this; // fluent API 606 } 607 608 /** 609 * Disposes this renderer and recycle it clean up before reusing this instance 610 */ 611 void dispose() { 612 if (DO_STATS) { 613 rdrCtx.stats.stat_rdr_activeEdges.add(activeEdgeMaxUsed); 614 rdrCtx.stats.stat_rdr_edges.add(edges.used); 615 rdrCtx.stats.stat_rdr_edges_count.add(edges.used / SIZEOF_EDGE_BYTES); 616 rdrCtx.stats.hist_rdr_edges_count.add(edges.used / SIZEOF_EDGE_BYTES); 617 rdrCtx.stats.totalOffHeap += edges.length; 618 } 619 // Return arrays: 620 crossings = crossings_ref.putArray(crossings); 621 aux_crossings = aux_crossings_ref.putArray(aux_crossings); 622 623 edgePtrs = edgePtrs_ref.putArray(edgePtrs); 624 aux_edgePtrs = aux_edgePtrs_ref.putArray(aux_edgePtrs); 641 buckets_minY, 642 buckets_maxY + 1); 643 } else { 644 // unused arrays 645 edgeBuckets = edgeBuckets_ref.putArray(edgeBuckets, 0, 0); 646 edgeBucketCounts = edgeBucketCounts_ref.putArray(edgeBucketCounts, 0, 0); 647 } 648 649 // At last: resize back off-heap edges to initial size 650 if (edges.length != INITIAL_EDGES_CAPACITY) { 651 // note: may throw OOME: 652 edges.resize(INITIAL_EDGES_CAPACITY); 653 } 654 if (DO_CLEAN_DIRTY) { 655 // Force zero-fill dirty arrays: 656 edges.fill(BYTE_0); 657 } 658 if (DO_MONITORS) { 659 rdrCtx.stats.mon_rdr_endRendering.stop(); 660 } 661 // recycle the RendererContext instance 662 MarlinRenderingEngine.returnRendererContext(rdrCtx); 663 } 664 665 private static float tosubpixx(final float pix_x) { 666 return SUBPIXEL_SCALE_X * pix_x; 667 } 668 669 private static float tosubpixy(final float pix_y) { 670 // shift y by -0.5 for fast ceil(y - 0.5): 671 return SUBPIXEL_SCALE_Y * pix_y - 0.5f; 672 } 673 674 @Override 675 public void moveTo(float pix_x0, float pix_y0) { 676 closePath(); 677 final float sx = tosubpixx(pix_x0); 678 final float sy = tosubpixy(pix_y0); 679 this.sx0 = sx; 680 this.sy0 = sy; 681 this.x0 = sx; 682 this.y0 = sy; 683 } 684 685 @Override 686 public void lineTo(float pix_x1, float pix_y1) { 687 final float x1 = tosubpixx(pix_x1); 688 final float y1 = tosubpixy(pix_y1); 689 addLine(x0, y0, x1, y1); 690 x0 = x1; 691 y0 = y1; 692 } 693 694 @Override 695 public void curveTo(float x1, float y1, 696 float x2, float y2, 697 float x3, float y3) 698 { 699 final float xe = tosubpixx(x3); 700 final float ye = tosubpixy(y3); 701 curve.set(x0, y0, tosubpixx(x1), tosubpixy(y1), 702 tosubpixx(x2), tosubpixy(y2), xe, ye); 703 curveBreakIntoLinesAndAdd(x0, y0, curve, xe, ye); 704 x0 = xe; 705 y0 = ye; 706 } 707 708 @Override 709 public void quadTo(float x1, float y1, float x2, float y2) { 710 final float xe = tosubpixx(x2); 711 final float ye = tosubpixy(y2); 712 curve.set(x0, y0, tosubpixx(x1), tosubpixy(y1), xe, ye); 713 quadBreakIntoLinesAndAdd(x0, y0, curve, xe, ye); 714 x0 = xe; 715 y0 = ye; 716 } 717 943 */ 944 if ((ptrLen < 10) || (numCrossings < 40)) { 945 if (DO_STATS) { 946 rdrCtx.stats.hist_rdr_crossings.add(numCrossings); 947 rdrCtx.stats.hist_rdr_crossings_adds.add(ptrLen); 948 } 949 950 /* 951 * threshold to use binary insertion sort instead of 952 * straight insertion sort (to reduce minimize comparisons). 953 */ 954 useBinarySearch = (numCrossings >= 20); 955 956 // if small enough: 957 lastCross = _MIN_VALUE; 958 959 for (i = 0; i < numCrossings; i++) { 960 // get the pointer to the edge 961 ecur = _edgePtrs[i]; 962 963 /* convert subpixel coordinates into pixel 964 positions for coming scanline */ 965 /* note: it is faster to always update edges even 966 if it is removed from AEL for coming or last scanline */ 967 968 // random access so use unsafe: 969 addr = addr0 + ecur; // ecur + OFF_F_CURX 970 971 // get current crossing: 972 curx = _unsafe.getInt(addr); 973 974 // update crossing with orientation at last bit: 975 cross = curx; 976 977 // Increment x using DDA (fixed point): 978 curx += _unsafe.getInt(addr + _OFF_BUMP_X); 979 980 // Increment error: 981 err = _unsafe.getInt(addr + _OFF_ERROR) 982 + _unsafe.getInt(addr + _OFF_BUMP_ERR); 983 984 // Manual carry handling: 1043 } 1044 } else { 1045 if (DO_STATS) { 1046 rdrCtx.stats.stat_rdr_crossings_msorts.add(numCrossings); 1047 rdrCtx.stats.hist_rdr_crossings_ratio 1048 .add((1000 * ptrLen) / numCrossings); 1049 rdrCtx.stats.hist_rdr_crossings_msorts.add(numCrossings); 1050 rdrCtx.stats.hist_rdr_crossings_msorts_adds.add(ptrLen); 1051 } 1052 1053 // Copy sorted data in auxiliary arrays 1054 // and perform insertion sort on almost sorted data 1055 // (ie i < prevNumCrossings): 1056 1057 lastCross = _MIN_VALUE; 1058 1059 for (i = 0; i < numCrossings; i++) { 1060 // get the pointer to the edge 1061 ecur = _edgePtrs[i]; 1062 1063 /* convert subpixel coordinates into pixel 1064 positions for coming scanline */ 1065 /* note: it is faster to always update edges even 1066 if it is removed from AEL for coming or last scanline */ 1067 1068 // random access so use unsafe: 1069 addr = addr0 + ecur; // ecur + OFF_F_CURX 1070 1071 // get current crossing: 1072 curx = _unsafe.getInt(addr); 1073 1074 // update crossing with orientation at last bit: 1075 cross = curx; 1076 1077 // Increment x using DDA (fixed point): 1078 curx += _unsafe.getInt(addr + _OFF_BUMP_X); 1079 1080 // Increment error: 1081 err = _unsafe.getInt(addr + _OFF_ERROR) 1082 + _unsafe.getInt(addr + _OFF_BUMP_ERR); 1083 1084 // Manual carry handling: 1150 prev = curx = x0; 1151 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1152 // last bit contains orientation (0 or 1) 1153 crorientation = ((curxo & 0x1) << 1) - 1; 1154 1155 if (windingRuleEvenOdd) { 1156 sum = crorientation; 1157 1158 // Even Odd winding rule: take care of mask ie sum(orientations) 1159 for (i = 1; i < numCrossings; i++) { 1160 curxo = _crossings[i]; 1161 curx = curxo >> 1; 1162 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1163 // last bit contains orientation (0 or 1) 1164 crorientation = ((curxo & 0x1) << 1) - 1; 1165 1166 if ((sum & 0x1) != 0) { 1167 // TODO: perform line clipping on left-right sides 1168 // to avoid such bound checks: 1169 x0 = (prev > bboxx0) ? prev : bboxx0; 1170 1171 if (curx < bboxx1) { 1172 x1 = curx; 1173 } else { 1174 x1 = bboxx1; 1175 // skip right side (fast exit loop): 1176 i = numCrossings; 1177 } 1178 1179 if (x0 < x1) { 1180 x0 -= bboxx0; // turn x0, x1 from coords to indices 1181 x1 -= bboxx0; // in the alpha array. 1182 1183 pix_x = x0 >> _SUBPIXEL_LG_POSITIONS_X; 1184 pix_xmaxm1 = (x1 - 1) >> _SUBPIXEL_LG_POSITIONS_X; 1185 1186 if (pix_x == pix_xmaxm1) { 1187 // Start and end in same pixel 1188 tmp = (x1 - x0); // number of subpixels 1189 _alpha[pix_x ] += tmp; 1190 _alpha[pix_x + 1] -= tmp; 1191 1192 if (useBlkFlags) { 1193 // flag used blocks: 1194 // note: block processing handles extra pixel: 1195 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1196 } 1197 } else { 1198 tmp = (x0 & _SUBPIXEL_MASK_X); 1199 _alpha[pix_x ] 1200 += (_SUBPIXEL_POSITIONS_X - tmp); 1201 _alpha[pix_x + 1] 1202 += tmp; 1203 1204 pix_xmax = x1 >> _SUBPIXEL_LG_POSITIONS_X; 1205 1206 tmp = (x1 & _SUBPIXEL_MASK_X); 1207 _alpha[pix_xmax ] 1208 -= (_SUBPIXEL_POSITIONS_X - tmp); 1209 _alpha[pix_xmax + 1] 1210 -= tmp; 1211 1212 if (useBlkFlags) { 1213 // flag used blocks: 1214 // note: block processing handles extra pixel: 1215 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1216 _blkFlags[pix_xmax >> _BLK_SIZE_LG] = 1; 1217 } 1218 } 1219 } 1220 } 1221 1222 sum += crorientation; 1223 prev = curx; 1224 } 1225 } else { 1226 // Non-zero winding rule: optimize that case (default) 1227 // and avoid processing intermediate crossings 1228 for (i = 1, sum = 0;; i++) { 1229 sum += crorientation; 1230 1231 if (sum != 0) { 1232 // prev = min(curx) 1233 if (prev > curx) { 1234 prev = curx; 1235 } 1236 } else { 1237 // TODO: perform line clipping on left-right sides 1238 // to avoid such bound checks: 1239 x0 = (prev > bboxx0) ? prev : bboxx0; 1240 1241 if (curx < bboxx1) { 1242 x1 = curx; 1243 } else { 1244 x1 = bboxx1; 1245 // skip right side (fast exit loop): 1246 i = numCrossings; 1247 } 1248 1249 if (x0 < x1) { 1250 x0 -= bboxx0; // turn x0, x1 from coords to indices 1251 x1 -= bboxx0; // in the alpha array. 1252 1253 pix_x = x0 >> _SUBPIXEL_LG_POSITIONS_X; 1254 pix_xmaxm1 = (x1 - 1) >> _SUBPIXEL_LG_POSITIONS_X; 1255 1256 if (pix_x == pix_xmaxm1) { 1257 // Start and end in same pixel 1258 tmp = (x1 - x0); // number of subpixels 1259 _alpha[pix_x ] += tmp; 1260 _alpha[pix_x + 1] -= tmp; 1261 1262 if (useBlkFlags) { 1263 // flag used blocks: 1264 // note: block processing handles extra pixel: 1265 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1266 } 1267 } else { 1268 tmp = (x0 & _SUBPIXEL_MASK_X); 1269 _alpha[pix_x ] 1270 += (_SUBPIXEL_POSITIONS_X - tmp); 1271 _alpha[pix_x + 1] 1272 += tmp; 1273 1274 pix_xmax = x1 >> _SUBPIXEL_LG_POSITIONS_X; 1275 1276 tmp = (x1 & _SUBPIXEL_MASK_X); 1277 _alpha[pix_xmax ] 1278 -= (_SUBPIXEL_POSITIONS_X - tmp); 1279 _alpha[pix_xmax + 1] 1280 -= tmp; 1281 1282 if (useBlkFlags) { 1283 // flag used blocks: 1284 // note: block processing handles extra pixel: 1285 _blkFlags[pix_x >> _BLK_SIZE_LG] = 1; 1286 _blkFlags[pix_xmax >> _BLK_SIZE_LG] = 1; 1287 } 1288 } 1289 } 1290 prev = _MAX_VALUE; 1291 } 1292 1293 if (i == numCrossings) { 1294 break; 1295 } 1296 1297 curxo = _crossings[i]; 1298 curx = curxo >> 1; 1299 // to turn {0, 1} into {-1, 1}, multiply by 2 and subtract 1. 1300 // last bit contains orientation (0 or 1) 1301 crorientation = ((curxo & 0x1) << 1) - 1; 1302 } 1303 } 1304 } // numCrossings > 0 1305 1306 // even if this last row had no crossings, alpha will be zeroed 1307 // from the last emitRow call. But this doesn't matter because 1308 // maxX < minX, so no row will be emitted to the MarlinCache. 1309 if ((y & _SUBPIXEL_MASK_Y) == _SUBPIXEL_MASK_Y) { 1310 lastY = y >> _SUBPIXEL_LG_POSITIONS_Y; 1311 1312 // convert subpixel to pixel coordinate within boundaries: 1313 minX = FloatMath.max(minX, bboxx0) >> _SUBPIXEL_LG_POSITIONS_X; 1314 maxX = FloatMath.min(maxX, bboxx1) >> _SUBPIXEL_LG_POSITIONS_X; 1315 1316 if (maxX >= minX) { 1317 // note: alpha array will be zeroed by copyAARow() 1318 // +1 because alpha [pix_minX; pix_maxX[ 1319 // fix range [x0; x1[ 1320 // note: if x1=bboxx1, then alpha is written up to bboxx1+1 1321 // inclusive: alpha[bboxx1] ignored, alpha[bboxx1+1] == 0 1322 // (normally so never cleared below) 1323 copyAARow(_alpha, lastY, minX, maxX + 1, useBlkFlags); 1324 1325 // speculative for next pixel row (scanline coherence): 1326 if (_enableBlkFlagsHeuristics) { 1327 // Use block flags if large pixel span and few crossings: 1328 // ie mean(distance between crossings) is larger than 1329 // 1 block size; 1330 1331 // fast check width: 1332 maxX -= minX; 1333 1334 // if stroking: numCrossings /= 2 1335 // => shift numCrossings by 1 1336 // condition = (width / (numCrossings - 1)) > blockSize 1337 useBlkFlags = (maxX > _BLK_SIZE) && (maxX > 1338 (((numCrossings >> stroking) - 1) << _BLK_SIZE_LG)); 1339 1340 if (DO_STATS) { 1341 tmp = FloatMath.max(1, 1342 ((numCrossings >> stroking) - 1)); 1343 rdrCtx.stats.hist_tile_generator_encoding_dist 1345 } 1346 } 1347 } else { 1348 _cache.clearAARow(lastY); 1349 } 1350 minX = _MAX_VALUE; 1351 maxX = _MIN_VALUE; 1352 } 1353 } // scan line iterator 1354 1355 // Emit final row 1356 y--; 1357 y >>= _SUBPIXEL_LG_POSITIONS_Y; 1358 1359 // convert subpixel to pixel coordinate within boundaries: 1360 minX = FloatMath.max(minX, bboxx0) >> _SUBPIXEL_LG_POSITIONS_X; 1361 maxX = FloatMath.min(maxX, bboxx1) >> _SUBPIXEL_LG_POSITIONS_X; 1362 1363 if (maxX >= minX) { 1364 // note: alpha array will be zeroed by copyAARow() 1365 // +1 because alpha [pix_minX; pix_maxX[ 1366 // fix range [x0; x1[ 1367 // note: if x1=bboxx1, then alpha is written up to bboxx1+1 1368 // inclusive: alpha[bboxx1] ignored then cleared and 1369 // alpha[bboxx1+1] == 0 (normally so never cleared after) 1370 copyAARow(_alpha, y, minX, maxX + 1, useBlkFlags); 1371 } else if (y != lastY) { 1372 _cache.clearAARow(y); 1373 } 1374 1375 // update member: 1376 edgeCount = numCrossings; 1377 prevUseBlkFlags = useBlkFlags; 1378 1379 if (DO_STATS) { 1380 // update max used mark 1381 activeEdgeMaxUsed = _arrayMaxUsed; 1382 } 1383 } 1384 1385 boolean endRendering() { 1386 if (DO_MONITORS) { 1387 rdrCtx.stats.mon_rdr_endRendering.start(); 1388 } 1389 if (edgeMinY == Integer.MAX_VALUE) { 1390 return false; // undefined edges bounds 1391 } 1392 1393 // bounds as half-open intervals 1394 final int spminX = FloatMath.max(FloatMath.ceil_int(edgeMinX - 0.5f), boundsMinX); 1395 final int spmaxX = FloatMath.min(FloatMath.ceil_int(edgeMaxX - 0.5f), boundsMaxX); 1396 1397 // edge Min/Max Y are already rounded to subpixels within bounds: 1398 final int spminY = edgeMinY; 1399 final int spmaxY = edgeMaxY; 1400 1401 buckets_minY = spminY - boundsMinY; 1402 buckets_maxY = spmaxY - boundsMinY; 1403 1404 if (DO_LOG_BOUNDS) { 1405 MarlinUtils.logInfo("edgesXY = [" + edgeMinX + " ... " + edgeMaxX 1406 + "[ [" + edgeMinY + " ... " + edgeMaxY + "["); 1407 MarlinUtils.logInfo("spXY = [" + spminX + " ... " + spmaxX 1408 + "[ [" + spminY + " ... " + spmaxY + "["); 1409 } 1410 1411 // test clipping for shapes out of bounds 1412 if ((spminX >= spmaxX) || (spminY >= spmaxY)) { 1413 return false; 1414 } 1415 1416 // half open intervals 1417 // inclusive: 1418 final int pminX = spminX >> SUBPIXEL_LG_POSITIONS_X; 1419 // exclusive: 1420 final int pmaxX = (spmaxX + SUBPIXEL_MASK_X) >> SUBPIXEL_LG_POSITIONS_X; 1421 // inclusive: 1422 final int pminY = spminY >> SUBPIXEL_LG_POSITIONS_Y; 1423 // exclusive: 1424 final int pmaxY = (spmaxY + SUBPIXEL_MASK_Y) >> SUBPIXEL_LG_POSITIONS_Y; 1425 1426 // store BBox to answer ptg.getBBox(): 1427 this.cache.init(pminX, pminY, pmaxX, pmaxY); 1428 1429 // Heuristics for using block flags: 1430 if (ENABLE_BLOCK_FLAGS) { 1431 enableBlkFlags = this.cache.useRLE; 1432 prevUseBlkFlags = enableBlkFlags && !ENABLE_BLOCK_FLAGS_HEURISTICS; 1433 1434 if (enableBlkFlags) { 1435 // ensure blockFlags array is large enough: 1436 // note: +2 to ensure enough space left at end 1437 final int blkLen = ((pmaxX - pminX) >> BLOCK_SIZE_LG) + 2; 1438 if (blkLen > INITIAL_ARRAY) { 1439 blkFlags = blkFlags_ref.getArray(blkLen); 1440 } 1441 } 1442 } 1443 1444 // memorize the rendering bounding box: 1445 /* note: bbox_spminX and bbox_spmaxX must be pixel boundaries 1446 to have correct coverage computation */ 1447 // inclusive: 1448 bbox_spminX = pminX << SUBPIXEL_LG_POSITIONS_X; 1449 // exclusive: 1450 bbox_spmaxX = pmaxX << SUBPIXEL_LG_POSITIONS_X; 1451 // inclusive: 1452 bbox_spminY = spminY; 1453 // exclusive: 1454 bbox_spmaxY = spmaxY; 1455 1456 if (DO_LOG_BOUNDS) { 1457 MarlinUtils.logInfo("pXY = [" + pminX + " ... " + pmaxX 1458 + "[ [" + pminY + " ... " + pmaxY + "["); 1459 MarlinUtils.logInfo("bbox_spXY = [" + bbox_spminX + " ... " 1460 + bbox_spmaxX + "[ [" + bbox_spminY + " ... " 1461 + bbox_spmaxY + "["); 1462 } 1463 1464 // Prepare alpha line: 1465 // add 2 to better deal with the last pixel in a pixel row. 1466 final int width = (pmaxX - pminX) + 2; 1467 1468 // Useful when processing tile line by tile line 1469 if (width > INITIAL_AA_ARRAY) { 1470 if (DO_STATS) { 1471 rdrCtx.stats.stat_array_renderer_alphaline.add(width); 1472 } 1473 alphaLine = alphaLine_ref.getArray(width); 1474 } 1492 // avoid rendering for last call to nextTile() 1493 if (fixed_spminY < bbox_spmaxY) { 1494 // process a complete tile line ie scanlines for 32 rows 1495 final int spmaxY = FloatMath.min(bbox_spmaxY, spminY + SUBPIXEL_TILE); 1496 1497 // process tile line [0 - 32] 1498 cache.resetTileLine(pminY); 1499 1500 // Process only one tile line: 1501 _endRendering(fixed_spminY, spmaxY); 1502 } 1503 if (DO_MONITORS) { 1504 rdrCtx.stats.mon_rdr_endRendering_Y.stop(); 1505 } 1506 } 1507 1508 void copyAARow(final int[] alphaRow, 1509 final int pix_y, final int pix_from, final int pix_to, 1510 final boolean useBlockFlags) 1511 { 1512 if (DO_MONITORS) { 1513 rdrCtx.stats.mon_rdr_copyAARow.start(); 1514 } 1515 if (useBlockFlags) { 1516 if (DO_STATS) { 1517 rdrCtx.stats.hist_tile_generator_encoding.add(1); 1518 } 1519 cache.copyAARowRLE_WithBlockFlags(blkFlags, alphaRow, pix_y, pix_from, pix_to); 1520 } else { 1521 if (DO_STATS) { 1522 rdrCtx.stats.hist_tile_generator_encoding.add(0); 1523 } 1524 cache.copyAARowNoRLE(alphaRow, pix_y, pix_from, pix_to); 1525 } 1526 if (DO_MONITORS) { 1527 rdrCtx.stats.mon_rdr_copyAARow.stop(); 1528 } 1529 } 1530 } |