1 /*
2 * Copyright 2016 The Android Open Source Project
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
8 #include "SkAnalyticEdge.h"
9 #include "SkAntiRun.h"
10 #include "SkAutoMalloc.h"
11 #include "SkBlitter.h"
12 #include "SkCoverageDelta.h"
13 #include "SkEdge.h"
14 #include "SkEdgeBuilder.h"
15 #include "SkGeometry.h"
16 #include "SkMask.h"
17 #include "SkPath.h"
18 #include "SkQuadClipper.h"
19 #include "SkRasterClip.h"
20 #include "SkRegion.h"
21 #include "SkScan.h"
22 #include "SkScanPriv.h"
23 #include "SkTSort.h"
24 #include "SkTemplates.h"
25 #include "SkUTF.h"
26
27 #if defined(SK_DISABLE_DAA)
DAAFillPath(const SkPath & path,SkBlitter * blitter,const SkIRect & ir,const SkIRect & clipBounds,bool forceRLE,SkDAARecord * record)28 void SkScan::DAAFillPath(const SkPath& path, SkBlitter* blitter, const SkIRect& ir,
29 const SkIRect& clipBounds, bool forceRLE, SkDAARecord* record) {
30 SkDEBUGFAIL("DAA Disabled");
31 return;
32 }
33 #else
34 ///////////////////////////////////////////////////////////////////////////////
35
36 /*
37
38 DAA stands for Delta-based Anti-Aliasing.
39
40 This is an improved version of our analytic AA algorithm (in SkScan_AAAPath.cpp)
41 which first scan convert paths into coverage deltas (this step can happen out of order,
42 and we don't seem to be needed to worry about the intersection, clamping, etc.),
43 and then use a single blitter run to convert all those deltas into the final alphas.
44
45 Before we go to the final blitter run, we'll use SkFixed for all delta values so we
46 don't ever have to worry about under/overflow.
47
48 */
49
50 ///////////////////////////////////////////////////////////////////////////////
51
52 // The following helper functions are the same as those from SkScan_AAAPath.cpp
53 // except that we use SkFixed everywhere instead of SkAlpha.
54
SkFixedMul_lowprec(SkFixed a,SkFixed b)55 static inline SkFixed SkFixedMul_lowprec(SkFixed a, SkFixed b) {
56 return (a >> 8) * (b >> 8);
57 }
58
59 // Return the alpha of a trapezoid whose height is 1
trapezoidToAlpha(SkFixed l1,SkFixed l2)60 static inline SkFixed trapezoidToAlpha(SkFixed l1, SkFixed l2) {
61 SkASSERT(l1 >= 0 && l2 >= 0);
62 return (l1 + l2) >> 1;
63 }
64
65 // The alpha of right-triangle (a, a*b)
partialTriangleToAlpha(SkFixed a,SkFixed b)66 static inline SkFixed partialTriangleToAlpha(SkFixed a, SkFixed b) {
67 SkASSERT(a <= SK_Fixed1);
68 // SkFixedMul(SkFixedMul(a, a), b) >> 1
69 // return ((((a >> 8) * (a >> 8)) >> 8) * (b >> 8)) >> 1;
70 return (a >> 11) * (a >> 11) * (b >> 11);
71 }
72
getPartialAlpha(SkFixed alpha,SkFixed partialMultiplier)73 static inline SkFixed getPartialAlpha(SkFixed alpha, SkFixed partialMultiplier) {
74 // DAA should not be so sensitive to the precision (compared to AAA).
75 return SkFixedMul_lowprec(alpha, partialMultiplier);
76 }
77
78 ///////////////////////////////////////////////////////////////////////////////
79
80 template<bool isPartial, class Deltas>
add_coverage_delta_segment(int y,SkFixed rowHeight,const SkAnalyticEdge * edge,SkFixed nextX,Deltas * deltas)81 static inline void add_coverage_delta_segment(int y, SkFixed rowHeight, const SkAnalyticEdge* edge,
82 SkFixed nextX, Deltas* deltas) { // rowHeight=fullAlpha
83 SkASSERT(rowHeight <= SK_Fixed1 && rowHeight >= 0);
84
85 // Let's see if multiplying sign is faster than multiplying edge->fWinding.
86 // (Compiler should be able to optimize multiplication with 1/-1?)
87 int sign = edge->fWinding == 1 ? 1 : -1;
88
89 SkFixed l = SkTMin(edge->fX, nextX);
90 SkFixed r = edge->fX + nextX - l;
91 int L = SkFixedFloorToInt(l);
92 int R = SkFixedCeilToInt(r);
93 int len = R - L;
94
95 switch (len) {
96 case 0: {
97 deltas->addDelta(L, y, rowHeight * sign);
98 return;
99 }
100 case 1: {
101 SkFixed fixedR = SkIntToFixed(R);
102 SkFixed alpha = trapezoidToAlpha(fixedR - l, fixedR - r);
103 if (isPartial) {
104 alpha = getPartialAlpha(alpha, rowHeight);
105 }
106 deltas->addDelta(L, y, alpha * sign);
107 deltas->addDelta(L + 1, y, (rowHeight - alpha) * sign);
108 return;
109 }
110 case 2: {
111 SkFixed middle = SkIntToFixed(L + 1);
112 SkFixed x1 = middle - l;
113 SkFixed x2 = r - middle;
114 SkFixed alpha1 = partialTriangleToAlpha(x1, edge->fDY);
115 SkFixed alpha2 = rowHeight - partialTriangleToAlpha(x2, edge->fDY);
116 deltas->addDelta(L, y, alpha1 * sign);
117 deltas->addDelta(L + 1, y, (alpha2 - alpha1) * sign);
118 deltas->addDelta(L + 2, y, (rowHeight - alpha2) * sign);
119 return;
120 }
121 }
122
123 // When len > 2, computations are similar to computeAlphaAboveLine in SkScan_AAAPath.cpp
124 SkFixed dY = edge->fDY;
125 SkFixed fixedL = SkIntToFixed(L);
126 SkFixed fixedR = SkIntToFixed(R);
127 SkFixed first = SK_Fixed1 + fixedL - l; // horizontal edge of the left-most triangle
128 SkFixed last = r - (fixedR - SK_Fixed1); // horizontal edge of the right-most triangle
129 SkFixed firstH = SkFixedMul_lowprec(first, dY); // vertical edge of the left-most triangle
130
131 SkFixed alpha0 = SkFixedMul_lowprec(first, firstH) >> 1; // triangle alpha
132 SkFixed alpha1 = firstH + (dY >> 1); // rectangle plus triangle
133 deltas->addDelta(L, y, alpha0 * sign);
134 deltas->addDelta(L + 1, y, (alpha1 - alpha0) * sign);
135 for(int i = 2; i < len - 1; ++i) {
136 deltas->addDelta(L + i, y, dY * sign); // the increment is always a rect of height = dY
137 }
138
139 SkFixed alphaR2 = alpha1 + dY * (len - 3); // the alpha at R - 2
140 SkFixed lastAlpha = rowHeight - partialTriangleToAlpha(last, dY); // the alpha at R - 1
141 deltas->addDelta(R - 1, y, (lastAlpha - alphaR2) * sign);
142 deltas->addDelta(R, y, (rowHeight - lastAlpha) * sign);
143 }
144
145 class XLessThan {
146 public:
operator ()(const SkBezier * a,const SkBezier * b)147 bool operator()(const SkBezier* a, const SkBezier* b) {
148 return a->fP0.fX + a->fP1.fX < b->fP0.fX + b->fP1.fX;
149 }
150 };
151
152 class YLessThan {
153 public:
operator ()(const SkBezier * a,const SkBezier * b)154 bool operator()(const SkBezier* a, const SkBezier* b) {
155 return a->fP0.fY + a->fP1.fY < b->fP0.fY + b->fP1.fY;
156 }
157 };
158
159 template<class Deltas> static SK_ALWAYS_INLINE
gen_alpha_deltas(const SkPath & path,const SkIRect & clippedIR,const SkIRect & clipBounds,Deltas & result,SkBlitter * blitter,bool skipRect,bool pathContainedInClip)160 void gen_alpha_deltas(const SkPath& path, const SkIRect& clippedIR, const SkIRect& clipBounds,
161 Deltas& result, SkBlitter* blitter, bool skipRect, bool pathContainedInClip) {
162 // 1. Build edges
163 SkBezierEdgeBuilder builder;
164 // We have to use clipBounds instead of clippedIR to build edges because of "canCullToTheRight":
165 // if the builder finds a right edge past the right clip, it won't build that right edge.
166 int count = builder.buildEdges(path, pathContainedInClip ? nullptr : &clipBounds);
167
168 if (count == 0) {
169 return;
170 }
171 SkBezier** list = builder.bezierList();
172
173 // 2. Try to find the rect part because blitAntiRect is so much faster than blitCoverageDeltas
174 int rectTop = clippedIR.fBottom; // the rect is initialized to be empty as top = bot
175 int rectBot = clippedIR.fBottom;
176 if (skipRect) { // only find that rect is skipRect == true
177 YLessThan lessThan; // sort edges in YX order
178 SkTQSort(list, list + count - 1, lessThan);
179 for(int i = 0; i < count - 1; ++i) {
180 SkBezier* lb = list[i];
181 SkBezier* rb = list[i + 1];
182
183 // fCount == 2 ensures that lb and rb are lines instead of quads or cubics.
184 bool lDX0 = lb->fP0.fX == lb->fP1.fX && lb->fCount == 2;
185 bool rDX0 = rb->fP0.fX == rb->fP1.fX && rb->fCount == 2;
186 if (!lDX0 || !rDX0) { // make sure that the edges are vertical
187 continue;
188 }
189
190 SkAnalyticEdge l, r;
191 if (!l.setLine(lb->fP0, lb->fP1) || !r.setLine(rb->fP0, rb->fP1)) {
192 continue;
193 }
194
195 SkFixed xorUpperY = l.fUpperY ^ r.fUpperY;
196 SkFixed xorLowerY = l.fLowerY ^ r.fLowerY;
197 if ((xorUpperY | xorLowerY) == 0) { // equal upperY and lowerY
198 rectTop = SkFixedCeilToInt(l.fUpperY);
199 rectBot = SkFixedFloorToInt(l.fLowerY);
200 if (rectBot > rectTop) { // if bot == top, the rect is too short for blitAntiRect
201 int L = SkFixedCeilToInt(l.fUpperX);
202 int R = SkFixedFloorToInt(r.fUpperX);
203 if (L <= R) {
204 SkAlpha la = (SkIntToFixed(L) - l.fUpperX) >> 8;
205 SkAlpha ra = (r.fUpperX - SkIntToFixed(R)) >> 8;
206 result.setAntiRect(L - 1, rectTop, R - L, rectBot - rectTop, la, ra);
207 } else { // too thin to use blitAntiRect; reset the rect region to be emtpy
208 rectTop = rectBot = clippedIR.fBottom;
209 }
210 }
211 break;
212 }
213
214 }
215 }
216
217 // 3. Sort edges in x so we may need less sorting for delta based on x. This only helps
218 // SkCoverageDeltaList. And we don't want to sort more than SORT_THRESHOLD edges where
219 // the log(count) factor of the quick sort may become a bottleneck; when there are so
220 // many edges, we're unlikely to make deltas sorted anyway.
221 constexpr int SORT_THRESHOLD = 256;
222 if (std::is_same<Deltas, SkCoverageDeltaList>::value && count < SORT_THRESHOLD) {
223 XLessThan lessThan;
224 SkTQSort(list, list + count - 1, lessThan);
225 }
226
227 // Future todo: parallize and SIMD the following code.
228 // 4. iterate through edges and generate deltas
229 for(int index = 0; index < count; ++index) {
230 SkAnalyticCubicEdge storage;
231 SkASSERT(sizeof(SkAnalyticQuadraticEdge) >= sizeof(SkAnalyticEdge));
232 SkASSERT(sizeof(SkAnalyticCubicEdge) >= sizeof(SkAnalyticQuadraticEdge));
233
234 SkBezier* bezier = list[index];
235 SkAnalyticEdge* currE = &storage;
236 bool edgeSet = false;
237
238 int originalWinding = 1;
239 bool sortY = true;
240 switch (bezier->fCount) {
241 case 2: {
242 edgeSet = currE->setLine(bezier->fP0, bezier->fP1);
243 originalWinding = currE->fWinding;
244 break;
245 }
246 case 3: {
247 SkQuad* quad = static_cast<SkQuad*>(bezier);
248 SkPoint pts[3] = {quad->fP0, quad->fP1, quad->fP2};
249 edgeSet = static_cast<SkAnalyticQuadraticEdge*>(currE)->setQuadratic(pts);
250 originalWinding = static_cast<SkAnalyticQuadraticEdge*>(currE)->fQEdge.fWinding;
251 break;
252 }
253 case 4: {
254 sortY = false;
255 SkCubic* cubic = static_cast<SkCubic*>(bezier);
256 SkPoint pts[4] = {cubic->fP0, cubic->fP1, cubic->fP2, cubic->fP3};
257 edgeSet = static_cast<SkAnalyticCubicEdge*>(currE)->setCubic(pts, sortY);
258 originalWinding = static_cast<SkAnalyticCubicEdge*>(currE)->fCEdge.fWinding;
259 break;
260 }
261 }
262
263 if (!edgeSet) {
264 continue;
265 }
266
267 do {
268 currE->fX = currE->fUpperX;
269
270 SkFixed upperFloor = SkFixedFloorToFixed(currE->fUpperY);
271 SkFixed lowerCeil = SkFixedCeilToFixed(currE->fLowerY);
272 int iy = SkFixedFloorToInt(upperFloor);
273
274 if (lowerCeil <= upperFloor + SK_Fixed1) { // only one row is affected by the currE
275 SkFixed rowHeight = currE->fLowerY - currE->fUpperY;
276 SkFixed nextX = currE->fX + SkFixedMul(currE->fDX, rowHeight);
277 if (iy >= clippedIR.fTop && iy < clippedIR.fBottom) {
278 add_coverage_delta_segment<true>(iy, rowHeight, currE, nextX, &result);
279 }
280 continue;
281 }
282
283 // check first row
284 SkFixed rowHeight = upperFloor + SK_Fixed1 - currE->fUpperY;
285 SkFixed nextX;
286 if (rowHeight != SK_Fixed1) { // it's a partial row
287 nextX = currE->fX + SkFixedMul(currE->fDX, rowHeight);
288 add_coverage_delta_segment<true>(iy, rowHeight, currE, nextX, &result);
289 } else { // it's a full row so we can leave it to the while loop
290 iy--; // compensate the iy++ in the while loop
291 nextX = currE->fX;
292 }
293
294 while (true) { // process the full rows in the middle
295 iy++;
296 SkFixed y = SkIntToFixed(iy);
297 currE->fX = nextX;
298 nextX += currE->fDX;
299
300 if (y + SK_Fixed1 > currE->fLowerY) {
301 break; // no full rows left, break
302 }
303
304 // Check whether we're in the rect part that will be covered by blitAntiRect
305 if (iy >= rectTop && iy < rectBot) {
306 SkASSERT(currE->fDX == 0); // If yes, we must be on an edge with fDX = 0.
307 iy = rectBot - 1; // Skip the rect part by advancing iy to the bottom.
308 continue;
309 }
310
311 // Add current edge's coverage deltas on this full row
312 add_coverage_delta_segment<false>(iy, SK_Fixed1, currE, nextX, &result);
313 }
314
315 // last partial row
316 if (SkIntToFixed(iy) < currE->fLowerY &&
317 iy >= clippedIR.fTop && iy < clippedIR.fBottom) {
318 rowHeight = currE->fLowerY - SkIntToFixed(iy);
319 nextX = currE->fX + SkFixedMul(currE->fDX, rowHeight);
320 add_coverage_delta_segment<true>(iy, rowHeight, currE, nextX, &result);
321 }
322 // Intended assignment to fWinding to restore the maybe-negated winding (during updateLine)
323 } while ((currE->fWinding = originalWinding) && currE->update(currE->fLowerY, sortY));
324 }
325 }
326
DAAFillPath(const SkPath & path,SkBlitter * blitter,const SkIRect & ir,const SkIRect & clipBounds,bool forceRLE,SkDAARecord * record)327 void SkScan::DAAFillPath(const SkPath& path, SkBlitter* blitter, const SkIRect& ir,
328 const SkIRect& clipBounds, bool forceRLE, SkDAARecord* record) {
329 bool containedInClip = clipBounds.contains(ir);
330 bool isEvenOdd = path.getFillType() & 1;
331 bool isConvex = path.isConvex();
332 bool isInverse = path.isInverseFillType();
333 bool skipRect = isConvex && !isInverse;
334 bool isInitOnce = record && record->fType == SkDAARecord::Type::kToBeComputed;
335
336 SkIRect clippedIR = ir;
337 clippedIR.intersect(clipBounds);
338
339 // The overhead of even constructing SkCoverageDeltaList/Mask is too big.
340 // So TryBlitFatAntiRect and return if it's successful.
341 if (!isInverse && TryBlitFatAntiRect(blitter, path, clipBounds)) {
342 SkDAARecord::SetEmpty(record);
343 return;
344 }
345
346 #ifdef SK_BUILD_FOR_GOOGLE3
347 constexpr int STACK_SIZE = 12 << 10; // 12K stack size alloc; Google3 has 16K limit.
348 #else
349 constexpr int STACK_SIZE = 64 << 10; // 64k stack size to avoid heap allocation
350 #endif
351 SkSTArenaAlloc<STACK_SIZE> stackAlloc; // avoid heap allocation with SkSTArenaAlloc
352
353 // Set alloc to record's alloc if and only if we're in the init-once phase. We have to do that
354 // during init phase because the mask or list needs to live longer. We can't do that during blit
355 // phase because the same record could be accessed by multiple threads simultaneously.
356 SkArenaAlloc* alloc = isInitOnce ? record->fAlloc : &stackAlloc;
357
358 if (record == nullptr) {
359 record = alloc->make<SkDAARecord>(alloc);
360 }
361
362 // Only blitter->blitXXX needs to be done in order in the threaded backend. Everything else can
363 // be done out of order in the init-once phase. We do that by calling DAAFillPath twice: first
364 // with a null blitter, and then second with the real blitter and the SkMask/SkCoverageDeltaList
365 // generated in the first step.
366 if (record->fType == SkDAARecord::Type::kToBeComputed) {
367 if (!forceRLE && !isInverse && SkCoverageDeltaMask::Suitable(clippedIR)) {
368 record->fType = SkDAARecord::Type::kMask;
369 SkCoverageDeltaMask deltaMask(alloc, clippedIR);
370 gen_alpha_deltas(path, clippedIR, clipBounds, deltaMask, blitter, skipRect,
371 containedInClip);
372 deltaMask.convertCoverageToAlpha(isEvenOdd, isInverse, isConvex);
373 record->fMask = deltaMask.prepareSkMask();
374 } else {
375 record->fType = SkDAARecord::Type::kList;
376 SkCoverageDeltaList* deltaList = alloc->make<SkCoverageDeltaList>(
377 alloc, clippedIR, forceRLE);
378 gen_alpha_deltas(path, clippedIR, clipBounds, *deltaList, blitter, skipRect,
379 containedInClip);
380 record->fList = deltaList;
381 }
382 }
383
384 if (!isInitOnce) {
385 SkASSERT(record->fType != SkDAARecord::Type::kToBeComputed);
386 if (record->fType == SkDAARecord::Type::kMask) {
387 blitter->blitMask(record->fMask, clippedIR);
388 } else {
389 blitter->blitCoverageDeltas(record->fList, clipBounds, isEvenOdd, isInverse, isConvex);
390 }
391 }
392 }
393 #endif //defined(SK_DISABLE_DAA)
394