Reland https://skia-review.googlesource.com/c/6091/
authorYuqian Li <liyuqian@google.com>
Fri, 13 Jan 2017 15:13:13 +0000 (10:13 -0500)
committerSkia Commit-Bot <skia-commit-bot@chromium.org>
Fri, 13 Jan 2017 16:07:48 +0000 (16:07 +0000)
The only difference is that we now also put the guard flag
SK_SUPPORT_LEGACY_AAA in SkUserConfig.h. Previously, SkAnalyticEdge.cpp doesn't
get that flag from SkScan.h and that caused many problems.

BUG=skia:
TBR=reed@google.com,caryclark@google.com

Change-Id: I134bb76cebd6fffa712f438076668765321bba3b
Reviewed-on: https://skia-review.googlesource.com/6992
Reviewed-by: Yuqian Li <liyuqian@google.com>
Commit-Queue: Yuqian Li <liyuqian@google.com>

12 files changed:
bench/nanobench.cpp
dm/DM.cpp
samplecode/SampleApp.cpp
src/core/SkAnalyticEdge.cpp
src/core/SkAnalyticEdge.h
src/core/SkScan.cpp
src/core/SkScan.h
src/core/SkScanPriv.h
src/core/SkScan_AAAPath.cpp
src/core/SkScan_AntiPath.cpp
tools/flags/SkCommonFlags.cpp
tools/flags/SkCommonFlags.h

index 34aaee8..7e05319 100644 (file)
@@ -1198,6 +1198,10 @@ int nanobench_main() {
 
     gSkUseAnalyticAA = FLAGS_analyticAA;
 
+    if (FLAGS_forceAnalyticAA) {
+        gSkForceAnalyticAA = true;
+    }
+
     int runs = 0;
     BenchmarkStream benchStream;
     while (Benchmark* b = benchStream.next()) {
index 7774d5e..26883b1 100644 (file)
--- a/dm/DM.cpp
+++ b/dm/DM.cpp
@@ -1289,6 +1289,10 @@ int dm_main() {
 
     gSkUseAnalyticAA = FLAGS_analyticAA;
 
+    if (FLAGS_forceAnalyticAA) {
+        gSkForceAnalyticAA = true;
+    }
+
     if (FLAGS_verbose) {
         gVLog = stderr;
     } else if (!FLAGS_writePath.isEmpty()) {
index 8797c4d..a1dabc2 100644 (file)
@@ -1822,7 +1822,12 @@ bool SampleWindow::onHandleChar(SkUnichar uni) {
             }
             break;
         case 'A':
-            gSkUseAnalyticAA = !gSkUseAnalyticAA.load();
+            if (gSkUseAnalyticAA.load() && !gSkForceAnalyticAA.load()) {
+                gSkForceAnalyticAA = true;
+            } else {
+                gSkUseAnalyticAA = !gSkUseAnalyticAA.load();
+                gSkForceAnalyticAA = false;
+            }
             this->inval(nullptr);
             this->updateTitle();
             break;
@@ -2176,7 +2181,11 @@ void SampleWindow::updateTitle() {
     title.prepend(gDeviceTypePrefix[fDeviceType]);
 
     if (gSkUseAnalyticAA) {
-        title.prepend("<AAA> ");
+        if (gSkForceAnalyticAA) {
+            title.prepend("<FAAA> ");
+        } else {
+            title.prepend("<AAA> ");
+        }
     }
     if (fTilingMode != kNo_Tiling) {
         title.prependf("<T: %s> ", gTilingInfo[fTilingMode].label);
index 057bc21..199fc4e 100644 (file)
@@ -48,34 +48,22 @@ bool SkAnalyticEdge::updateLine(SkFixed x0, SkFixed y0, SkFixed x1, SkFixed y1,
     return true;
 }
 
-void SkAnalyticEdge::chopLineWithClip(const SkIRect& clip) {
-    int top = SkFixedFloorToInt(fUpperY);
-
-    SkASSERT(top < clip.fBottom);
-
-    // clip the line to the clip top
-    if (top < clip.fTop) {
-        SkASSERT(SkFixedCeilToInt(fLowerY) > clip.fTop);
-        SkFixed newY = SkIntToFixed(clip.fTop);
-        this->goY(newY);
-        fUpperY = newY;
-    }
-}
-
 bool SkAnalyticQuadraticEdge::setQuadratic(const SkPoint pts[3]) {
-    if (!fQEdge.setQuadraticWithoutUpdate(pts, 2)) {
+    fRiteE = nullptr;
+
+    if (!fQEdge.setQuadraticWithoutUpdate(pts, kDefaultAccuracy)) {
         return false;
     }
-    fQEdge.fQx >>= 2;
-    fQEdge.fQy >>= 2;
-    fQEdge.fQDx >>= 2;
-    fQEdge.fQDy >>= 2;
-    fQEdge.fQDDx >>= 2;
-    fQEdge.fQDDy >>= 2;
-    fQEdge.fQLastX >>= 2;
-    fQEdge.fQLastY >>= 2;
-    fQEdge.fQy = snapY(fQEdge.fQy);
-    fQEdge.fQLastY = snapY(fQEdge.fQLastY);
+    fQEdge.fQx >>= kDefaultAccuracy;
+    fQEdge.fQy >>= kDefaultAccuracy;
+    fQEdge.fQDx >>= kDefaultAccuracy;
+    fQEdge.fQDy >>= kDefaultAccuracy;
+    fQEdge.fQDDx >>= kDefaultAccuracy;
+    fQEdge.fQDDy >>= kDefaultAccuracy;
+    fQEdge.fQLastX >>= kDefaultAccuracy;
+    fQEdge.fQLastY >>= kDefaultAccuracy;
+    fQEdge.fQy = SnapY(fQEdge.fQy);
+    fQEdge.fQLastY = SnapY(fQEdge.fQLastY);
 
     fWinding = fQEdge.fWinding;
     fCurveCount = fQEdge.fCurveCount;
@@ -105,15 +93,28 @@ bool SkAnalyticQuadraticEdge::updateQuadratic() {
         {
             newx    = oldx + (dx >> shift);
             newy    = oldy + (dy >> shift);
-            SkFDot6 diffY = (newy - fSnappedY) >> 10;
-            slope = diffY ? QuickSkFDot6Div((newx - fSnappedX) >> 10, diffY) : SK_MaxS32;
             if (SkAbs32(dy >> shift) >= SK_Fixed1 * 2) { // only snap when dy is large enough
+                SkFDot6 diffY = SkFixedToFDot6(newy - fSnappedY);
+                slope = diffY ? QuickSkFDot6Div(SkFixedToFDot6(newx - fSnappedX), diffY)
+                              : SK_MaxS32;
                 newSnappedY = SkTMin<SkFixed>(fQEdge.fQLastY, SkFixedRoundToFixed(newy));
                 newSnappedX = newx - SkFixedMul(slope, newy - newSnappedY);
             } else {
-                newSnappedY = SkTMin(fQEdge.fQLastY, snapY(newy));
+                newSnappedY = SkTMin(fQEdge.fQLastY, SnapY(newy));
                 newSnappedX = newx;
+#ifdef SK_SUPPORT_LEGACY_AAA
+                SkFDot6 diffY = SkFixedToFDot6(newy - fSnappedY);
+#else
+                SkFDot6 diffY = SkFixedToFDot6(newSnappedY - fSnappedY);
+#endif
+                slope = diffY ? QuickSkFDot6Div(SkFixedToFDot6(newx - fSnappedX), diffY)
+                              : SK_MaxS32;
             }
+#ifndef SK_SUPPORT_LEGACY_AAA
+            SkASSERT(slope == SK_MaxS32 ||
+                    SkAbs32(fSnappedX + SkFixedMul(slope, newSnappedY - fSnappedY) - newSnappedX)
+                    < SK_FixedHalf);
+#endif
             dx += fQEdge.fQDDx;
             dy += fQEdge.fQDDy;
         }
@@ -146,28 +147,32 @@ bool SkAnalyticQuadraticEdge::updateQuadratic() {
 }
 
 bool SkAnalyticCubicEdge::setCubic(const SkPoint pts[4]) {
-    if (!fCEdge.setCubicWithoutUpdate(pts, 2)) {
+    fRiteE = nullptr;
+
+    if (!fCEdge.setCubicWithoutUpdate(pts, kDefaultAccuracy)) {
         return false;
     }
 
-    fCEdge.fCx >>= 2;
-    fCEdge.fCy >>= 2;
-    fCEdge.fCDx >>= 2;
-    fCEdge.fCDy >>= 2;
-    fCEdge.fCDDx >>= 2;
-    fCEdge.fCDDy >>= 2;
-    fCEdge.fCDDDx >>= 2;
-    fCEdge.fCDDDy >>= 2;
-    fCEdge.fCLastX >>= 2;
-    fCEdge.fCLastY >>= 2;
-    fCEdge.fCy = snapY(fCEdge.fCy);
-    fCEdge.fCLastY = snapY(fCEdge.fCLastY);
+    fCEdge.fCx >>= kDefaultAccuracy;
+    fCEdge.fCy >>= kDefaultAccuracy;
+    fCEdge.fCDx >>= kDefaultAccuracy;
+    fCEdge.fCDy >>= kDefaultAccuracy;
+    fCEdge.fCDDx >>= kDefaultAccuracy;
+    fCEdge.fCDDy >>= kDefaultAccuracy;
+    fCEdge.fCDDDx >>= kDefaultAccuracy;
+    fCEdge.fCDDDy >>= kDefaultAccuracy;
+    fCEdge.fCLastX >>= kDefaultAccuracy;
+    fCEdge.fCLastY >>= kDefaultAccuracy;
+    fCEdge.fCy = SnapY(fCEdge.fCy);
+    fCEdge.fCLastY = SnapY(fCEdge.fCLastY);
 
     fWinding = fCEdge.fWinding;
     fCurveCount = fCEdge.fCurveCount;
     fCurveShift = fCEdge.fCurveShift;
     fCubicDShift = fCEdge.fCubicDShift;
 
+    fSnappedY = fCEdge.fCy;
+
     return this->updateCubic();
 }
 
@@ -203,17 +208,24 @@ bool SkAnalyticCubicEdge::updateCubic() {
             newy = oldy;
         }
 
-        SkFixed snappedOldY = SkAnalyticEdge::snapY(oldy);
-        SkFixed snappedNewY = SkAnalyticEdge::snapY(newy);
-        SkFixed slope = SkFixedToFDot6(snappedNewY - snappedOldY) == 0
+        SkFixed newSnappedY = SnapY(newy);
+        // we want to SkASSERT(snappedNewY <= fCEdge.fCLastY), but our finite fixedpoint
+        // doesn't always achieve that, so we have to explicitly pin it here.
+        if (fCEdge.fCLastY < newSnappedY) {
+            newSnappedY = fCEdge.fCLastY;
+            count = 0;
+        }
+
+        SkFixed slope = SkFixedToFDot6(newSnappedY - fSnappedY) == 0
                         ? SK_MaxS32
                         : SkFDot6Div(SkFixedToFDot6(newx - oldx),
-                                     SkFixedToFDot6(snappedNewY - snappedOldY));
+                                     SkFixedToFDot6(newSnappedY - fSnappedY));
 
-        success = this->updateLine(oldx, snappedOldY, newx, snappedNewY, slope);
+        success = this->updateLine(oldx, fSnappedY, newx, newSnappedY, slope);
 
         oldx = newx;
         oldy = newy;
+        fSnappedY = newSnappedY;
     } while (count < 0 && !success);
 
     fCEdge.fCx  = newx;
index 7f5e62c..1ed25f3 100644 (file)
 
 #include "SkEdge.h"
 
+// Use this to check that we successfully guard the change against Chromium layout tests
+#ifndef  SK_SUPPORT_LEGACY_AAA
+# define SK_SUPPORT_LEGACY_AAA
+#endif
+
 struct SkAnalyticEdge {
     // Similar to SkEdge, the conic edges will be converted to quadratic edges
     enum Type {
@@ -21,6 +26,10 @@ struct SkAnalyticEdge {
     SkAnalyticEdge* fNext;
     SkAnalyticEdge* fPrev;
 
+    // During aaa_walk_edges, if this edge is a left edge,
+    // then fRiteE is its corresponding right edge. Otherwise it's nullptr.
+    SkAnalyticEdge* fRiteE;
+
     SkFixed fX;
     SkFixed fDX;
     SkFixed fUpperX;        // The x value when y = fUpperY
@@ -30,6 +39,10 @@ struct SkAnalyticEdge {
     SkFixed fDY;            // abs(1/fDX); may be SK_MaxS32 when fDX is close to 0.
                             // fDY is only used for blitting trapezoids.
 
+    SkFixed fSavedX;        // For deferred blitting
+    SkFixed fSavedY;        // For deferred blitting
+    SkFixed fSavedDY;       // For deferred blitting
+
     int8_t  fCurveCount;    // only used by kQuad(+) and kCubic(-)
     uint8_t fCurveShift;    // appled to all Dx/DDx/DDDx except for fCubicDShift exception
     uint8_t fCubicDShift;   // applied to fCDx and fCDy only in cubic
@@ -37,7 +50,8 @@ struct SkAnalyticEdge {
 
     static const int kDefaultAccuracy = 2; // default accuracy for snapping
 
-    static inline SkFixed snapY(SkFixed y, int accuracy = kDefaultAccuracy) {
+    static inline SkFixed SnapY(SkFixed y) {
+        const int accuracy = kDefaultAccuracy;
         // This approach is safer than left shift, round, then right shift
         return ((unsigned)y + (SK_Fixed1 >> (accuracy + 1))) >> (16 - accuracy) << (16 - accuracy);
     }
@@ -55,15 +69,22 @@ struct SkAnalyticEdge {
         }
     }
 
-    inline bool setLine(const SkPoint& p0, const SkPoint& p1, const SkIRect* clip = nullptr);
-    inline bool updateLine(SkFixed ax, SkFixed ay, SkFixed bx, SkFixed by, SkFixed slope);
-    void chopLineWithClip(const SkIRect& clip);
+    inline void goY(SkFixed y, int yShift) {
+        SkASSERT(yShift >= 0 && yShift <= kDefaultAccuracy);
+        SkASSERT(fDX == 0 || y - fY == SK_Fixed1 >> yShift);
+        fY = y;
+        fX += fDX >> yShift;
+    }
 
-    inline bool intersectsClip(const SkIRect& clip) const {
-        SkASSERT(SkFixedFloorToInt(fUpperY) < clip.fBottom);
-        return SkFixedCeilToInt(fLowerY) > clip.fTop;
+    inline void saveXY(SkFixed x, SkFixed y, SkFixed dY) {
+        fSavedX = x;
+        fSavedY = y;
+        fSavedDY = dY;
     }
 
+    inline bool setLine(const SkPoint& p0, const SkPoint& p1);
+    inline bool updateLine(SkFixed ax, SkFixed ay, SkFixed bx, SkFixed by, SkFixed slope);
+
 #ifdef SK_DEBUG
     void dump() const {
         SkDebugf("edge: upperY:%d lowerY:%d y:%g x:%g dx:%g w:%d\n",
@@ -90,36 +111,47 @@ struct SkAnalyticQuadraticEdge : public SkAnalyticEdge {
 
     bool setQuadratic(const SkPoint pts[3]);
     bool updateQuadratic();
+    inline void keepContinuous() {
+        // We use fX as the starting x to ensure the continuouty.
+        // Without it, we may break the sorted edge list.
+        SkASSERT(SkAbs32(fX - SkFixedMul(fY - fSnappedY, fDX) - fSnappedX) < SK_Fixed1);
+        SkASSERT(SkAbs32(fY - fSnappedY) < SK_Fixed1); // This may differ due to smooth jump
+        fSnappedX = fX;
+        fSnappedY = fY;
+    }
 };
 
 struct SkAnalyticCubicEdge : public SkAnalyticEdge {
     SkCubicEdge fCEdge;
 
+    SkFixed fSnappedY; // to make sure that y is increasing with smooth jump and snapping
+
     bool setCubic(const SkPoint pts[4]);
     bool updateCubic();
+    inline void keepContinuous() {
+        SkASSERT(SkAbs32(fX - SkFixedMul(fDX, fY - SnapY(fCEdge.fCy)) - fCEdge.fCx) < SK_Fixed1);
+        fCEdge.fCx = fX;
+        fSnappedY = fY;
+    }
 };
 
-bool SkAnalyticEdge::setLine(const SkPoint& p0, const SkPoint& p1, const SkIRect* clip) {
-    // We must set X/Y using the same way (times 4, to FDot6, then to Fixed) as Quads/Cubics.
+bool SkAnalyticEdge::setLine(const SkPoint& p0, const SkPoint& p1) {
+    fRiteE = nullptr;
+
+    // We must set X/Y using the same way (e.g., times 4, to FDot6, then to Fixed) as Quads/Cubics.
     // Otherwise the order of the edge might be wrong due to precision limit.
-    SkFixed x0 = SkFDot6ToFixed(SkScalarToFDot6(p0.fX * 4)) >> 2;
-    SkFixed y0 = snapY(SkFDot6ToFixed(SkScalarToFDot6(p0.fY * 4)) >> 2);
-    SkFixed x1 = SkFDot6ToFixed(SkScalarToFDot6(p1.fX * 4)) >> 2;
-    SkFixed y1 = snapY(SkFDot6ToFixed(SkScalarToFDot6(p1.fY * 4)) >> 2);
+    const int accuracy = kDefaultAccuracy;
+    const int multiplier = (1 << kDefaultAccuracy);
+    SkFixed x0 = SkFDot6ToFixed(SkScalarToFDot6(p0.fX * multiplier)) >> accuracy;
+    SkFixed y0 = SnapY(SkFDot6ToFixed(SkScalarToFDot6(p0.fY * multiplier)) >> accuracy);
+    SkFixed x1 = SkFDot6ToFixed(SkScalarToFDot6(p1.fX * multiplier)) >> accuracy;
+    SkFixed y1 = SnapY(SkFDot6ToFixed(SkScalarToFDot6(p1.fY * multiplier)) >> accuracy);
 
     // are we a zero-height line?
     if (y0 == y1) {
         return false;
     }
 
-    int top = SkFixedFloorToInt(y0);
-    int bot = SkFixedCeilToInt(y1);
-
-    // are we completely above or below the clip?
-    if (clip && (top >= clip->fBottom || bot <= clip->fTop)) {
-        return false;
-    }
-
     int winding = 1;
 
     if (y0 > y1) {
@@ -128,7 +160,15 @@ bool SkAnalyticEdge::setLine(const SkPoint& p0, const SkPoint& p1, const SkIRect
         winding = -1;
     }
 
+#ifdef SK_SUPPORT_LEGACY_AAA
     SkFixed slope = SkFixedDiv(x1 - x0, y1 - y0);
+#else
+    SkFDot6 dy = SkFixedToFDot6(y1 - y0);
+    SkFDot6 dx = SkFixedToFDot6(x1 - x0);
+    SkFixed slope = dy ? QuickSkFDot6Div(dx, dy) : SK_MaxS32;
+    SkASSERT(dx == 0 || slope != 0);
+    SkFixed absSlope = SkAbs32(slope);
+#endif
 
     fX          = x0;
     fDX         = slope;
@@ -136,14 +176,17 @@ bool SkAnalyticEdge::setLine(const SkPoint& p0, const SkPoint& p1, const SkIRect
     fY          = y0;
     fUpperY     = y0;
     fLowerY     = y1;
+#ifdef SK_SUPPORT_LEGACY_AAA
     fDY         = x1 != x0 ? SkAbs32(SkFixedDiv(y1 - y0, x1 - x0)) : SK_MaxS32;
+#else
+    fDY         = dx == 0 ? SK_MaxS32 : absSlope < kInverseTableSize
+                                                 ? QuickFDot6Inverse::Lookup(absSlope)
+                                                 : SkAbs32(QuickSkFDot6Div(dy, dx));
+#endif
     fCurveCount = 0;
     fWinding    = SkToS8(winding);
     fCurveShift = 0;
 
-    if (clip) {
-        this->chopLineWithClip(*clip);
-    }
     return true;
 }
 
index 5a02c61..c2f00f3 100644 (file)
@@ -16,6 +16,8 @@
     std::atomic<bool> gSkUseAnalyticAA{true};
 #endif
 
+std::atomic<bool> gSkForceAnalyticAA{false};
+
 static inline void blitrect(SkBlitter* blitter, const SkIRect& r) {
     blitter->blitRect(r.fLeft, r.fTop, r.width(), r.height());
 }
index 24e0088..d0c10fa 100644 (file)
@@ -23,7 +23,13 @@ class SkPath;
 */
 typedef SkIRect SkXRect;
 
+// Use this to check that we successfully guard the change against Chromium layout tests
+#ifndef  SK_SUPPORT_LEGACY_AAA
+# define SK_SUPPORT_LEGACY_AAA
+#endif
+
 extern std::atomic<bool> gSkUseAnalyticAA;
+extern std::atomic<bool> gSkForceAnalyticAA;
 
 class AdditiveBlitter;
 
index 4c237e0..606929d 100644 (file)
@@ -34,10 +34,6 @@ void sk_fill_path(const SkPath& path, const SkIRect& clipRect,
                   SkBlitter* blitter, int start_y, int stop_y, int shiftEdgesUp,
                   bool pathContainedInClip);
 
-void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter*,
-                   int start_y, int stop_y, bool pathContainedInClip, bool isUsingMask,
-                   bool forceRLE);
-
 // blit the rects above and below avoid, clipped to clip
 void sk_blit_above(SkBlitter*, const SkIRect& avoid, const SkRegion& clip);
 void sk_blit_below(SkBlitter*, const SkIRect& avoid, const SkRegion& clip);
index 8c0a748..1263228 100644 (file)
@@ -84,9 +84,13 @@ number of scan lines in our algorithm is only about 3 + H while the
 
 ///////////////////////////////////////////////////////////////////////////////
 
-static inline void addAlpha(SkAlpha& alpha, SkAlpha delta) {
-    SkASSERT(alpha + (int)delta <= 256);
-    alpha = SkAlphaRuns::CatchOverflow(alpha + (int)delta);
+static inline void addAlpha(SkAlpha* alpha, SkAlpha delta) {
+    SkASSERT(*alpha + (int)delta <= 256);
+    *alpha = SkAlphaRuns::CatchOverflow(*alpha + (int)delta);
+}
+
+static inline void safelyAddAlpha(SkAlpha* alpha, SkAlpha delta) {
+    *alpha = SkTMin(0xFF, *alpha + (int)delta);
 }
 
 class AdditiveBlitter : public SkBlitter {
@@ -147,7 +151,7 @@ public:
     void blitAntiH(int x, int y, const SkAlpha antialias[], int len) override;
 
     // Allowing following methods are used to blit rectangles during aaa_walk_convex_edges
-    // Since there aren't many rectangles, we can still break the slow speed of virtual functions.
+    // Since there aren't many rectangles, we can still bear the slow speed of virtual functions.
     void blitAntiH(int x, int y, const SkAlpha alpha) override;
     void blitAntiH(int x, int y, int width, const SkAlpha alpha) override;
     void blitV(int x, int y, int height, SkAlpha alpha) override;
@@ -227,14 +231,14 @@ void MaskAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha antialias[], int
 
 void MaskAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha alpha) {
     SkASSERT(x >= fMask.fBounds.fLeft -1);
-    addAlpha(this->getRow(y)[x], alpha);
+    addAlpha(&this->getRow(y)[x], alpha);
 }
 
 void MaskAdditiveBlitter::blitAntiH(int x, int y, int width, const SkAlpha alpha) {
     SkASSERT(x >= fMask.fBounds.fLeft -1);
     uint8_t* row = this->getRow(y);
-    for (int i=0; i<width; i++) {
-        addAlpha(row[x + i], alpha);
+    for (int i = 0; i < width; ++i) {
+        addAlpha(&row[x + i], alpha);
     }
 }
 
@@ -246,7 +250,7 @@ void MaskAdditiveBlitter::blitV(int x, int y, int height, SkAlpha alpha) {
     // This must be called as if this is a real blitter.
     // So we directly set alpha rather than adding it.
     uint8_t* row = this->getRow(y);
-    for (int i=0; i<height; i++) {
+    for (int i = 0; i < height; ++i) {
         row[x] = alpha;
         row += fMask.fRowBytes;
     }
@@ -257,7 +261,7 @@ void MaskAdditiveBlitter::blitRect(int x, int y, int width, int height) {
     // This must be called as if this is a real blitter.
     // So we directly set alpha rather than adding it.
     uint8_t* row = this->getRow(y);
-    for (int i=0; i<height; i++) {
+    for (int i = 0; i < height; ++i) {
         memset(row + x, 0xFF, width);
         row += fMask.fRowBytes;
     }
@@ -290,7 +294,7 @@ public:
         }
     }
 
-private:
+protected:
     SkBlitter* fRealBlitter;
 
     /// Current y coordinate
@@ -313,7 +317,7 @@ private:
 
     int         fOffsetX;
 
-    inline bool check(int x, int width) {
+    inline bool check(int x, int width) const {
         #ifdef SK_DEBUG
         if (x < 0 || x + width > fWidth) {
             // SkDebugf("Ignore x = %d, width = %d\n", x, width);
@@ -431,8 +435,8 @@ void RunBasedAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha antialias[],
         }
         fRuns.fRuns[x + i] = 1;
     }
-    for (int i=0; i<len; i++) {
-        addAlpha(fRuns.fAlpha[x + i], antialias[i]);
+    for (int i = 0; i < len; ++i) {
+        addAlpha(&fRuns.fAlpha[x + i], antialias[i]);
     }
 }
 void RunBasedAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha alpha) {
@@ -463,12 +467,85 @@ void RunBasedAdditiveBlitter::blitAntiH(int x, int y, int width, const SkAlpha a
 
 int RunBasedAdditiveBlitter::getWidth() { return fWidth; }
 
+// This exists specifically for concave path filling.
+// In those cases, we can easily accumulate alpha greater than 0xFF.
+class SafeRLEAdditiveBlitter : public RunBasedAdditiveBlitter {
+public:
+    SafeRLEAdditiveBlitter(SkBlitter* realBlitter, const SkIRect& ir, const SkRegion& clip,
+            bool isInverse) : RunBasedAdditiveBlitter(realBlitter, ir, clip, isInverse) {}
+
+    void blitAntiH(int x, int y, const SkAlpha antialias[], int len) override;
+    void blitAntiH(int x, int y, const SkAlpha alpha) override;
+    void blitAntiH(int x, int y, int width, const SkAlpha alpha) override;
+};
+
+void SafeRLEAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha antialias[], int len) {
+    checkY(y);
+    x -= fLeft;
+
+    if (x < 0) {
+        len += x;
+        antialias -= x;
+        x = 0;
+    }
+    len = SkTMin(len, fWidth - x);
+    SkASSERT(check(x, len));
+
+    if (x < fOffsetX) {
+        fOffsetX = 0;
+    }
+
+    fOffsetX = fRuns.add(x, 0, len, 0, 0, fOffsetX); // Break the run
+    for (int i = 0; i < len; i += fRuns.fRuns[x + i]) {
+        for (int j = 1; j < fRuns.fRuns[x + i]; j++) {
+            fRuns.fRuns[x + i + j] = 1;
+            fRuns.fAlpha[x + i + j] = fRuns.fAlpha[x + i];
+        }
+        fRuns.fRuns[x + i] = 1;
+    }
+    for (int i = 0; i < len; ++i) {
+        safelyAddAlpha(&fRuns.fAlpha[x + i], antialias[i]);
+    }
+}
+
+void SafeRLEAdditiveBlitter::blitAntiH(int x, int y, const SkAlpha alpha) {
+    checkY(y);
+    x -= fLeft;
+
+    if (x < fOffsetX) {
+        fOffsetX = 0;
+    }
+
+    if (check(x, 1)) {
+        // Break the run
+        fOffsetX = fRuns.add(x, 0, 1, 0, 0, fOffsetX);
+        safelyAddAlpha(&fRuns.fAlpha[x], alpha);
+    }
+}
+
+void SafeRLEAdditiveBlitter::blitAntiH(int x, int y, int width, const SkAlpha alpha) {
+    checkY(y);
+    x -= fLeft;
+
+    if (x < fOffsetX) {
+        fOffsetX = 0;
+    }
+
+    if (check(x, width)) {
+        // Break the run
+        fOffsetX = fRuns.add(x, 0, width, 0, 0, fOffsetX);
+        for(int i = x; i < x + width; i += fRuns.fRuns[i]) {
+            safelyAddAlpha(&fRuns.fAlpha[i], alpha);
+        }
+    }
+}
+
 ///////////////////////////////////////////////////////////////////////////////
 
 // Return the alpha of a trapezoid whose height is 1
 static inline SkAlpha trapezoidToAlpha(SkFixed l1, SkFixed l2) {
     SkASSERT(l1 >= 0 && l2 >= 0);
-    return ((l1 + l2) >> 9);
+    return (l1 + l2) >> 9;
 }
 
 // The alpha of right-triangle (a, a*b), in 16 bits
@@ -524,7 +601,7 @@ static inline void computeAlphaAboveLine(SkAlpha* alphas, SkFixed l, SkFixed r,
         SkFixed firstH = SkFixedMul(first, dY); // vertical edge of the left-most triangle
         alphas[0] = SkFixedMul(first, firstH) >> 9; // triangle alpha
         SkFixed alpha16 = firstH + (dY >> 1); // rectangle plus triangle
-        for (int i = 1; i < R - 1; i++) {
+        for (int i = 1; i < R - 1; ++i) {
             alphas[i] = alpha16 >> 8;
             alpha16 += dY;
         }
@@ -557,17 +634,19 @@ static inline void computeAlphaBelowLine(
 }
 
 // Note that if fullAlpha != 0xFF, we'll multiply alpha by fullAlpha
-static inline void blit_single_alpha(AdditiveBlitter* blitter, int y, int x,
+static SK_ALWAYS_INLINE void blit_single_alpha(AdditiveBlitter* blitter, int y, int x,
                               SkAlpha alpha, SkAlpha fullAlpha, SkAlpha* maskRow,
-                              bool isUsingMask) {
+                              bool isUsingMask, bool noRealBlitter, bool needSafeCheck) {
     if (isUsingMask) {
-        if (fullAlpha == 0xFF) {
+        if (fullAlpha == 0xFF && !noRealBlitter) { // noRealBlitter is needed for concave paths
             maskRow[x] = alpha;
+        } else if (needSafeCheck) {
+            safelyAddAlpha(&maskRow[x], getPartialAlpha(alpha, fullAlpha));
         } else {
-            addAlpha(maskRow[x], getPartialAlpha(alpha, fullAlpha));
+            addAlpha(&maskRow[x], getPartialAlpha(alpha, fullAlpha));
         }
     } else {
-        if (fullAlpha == 0xFF) {
+        if (fullAlpha == 0xFF && !noRealBlitter) {
             blitter->getRealBlitter()->blitV(x, y, 1, alpha);
         } else {
             blitter->blitAntiH(x, y, getPartialAlpha(alpha, fullAlpha));
@@ -575,14 +654,19 @@ static inline void blit_single_alpha(AdditiveBlitter* blitter, int y, int x,
     }
 }
 
-static inline void blit_two_alphas(AdditiveBlitter* blitter, int y, int x,
+static SK_ALWAYS_INLINE void blit_two_alphas(AdditiveBlitter* blitter, int y, int x,
                             SkAlpha a1, SkAlpha a2, SkAlpha fullAlpha, SkAlpha* maskRow,
-                            bool isUsingMask) {
+                            bool isUsingMask, bool noRealBlitter, bool needSafeCheck) {
     if (isUsingMask) {
-        addAlpha(maskRow[x], a1);
-        addAlpha(maskRow[x + 1], a2);
+        if (needSafeCheck) {
+            safelyAddAlpha(&maskRow[x], a1);
+            safelyAddAlpha(&maskRow[x + 1], a2);
+        } else {
+            addAlpha(&maskRow[x], a1);
+            addAlpha(&maskRow[x + 1], a2);
+        }
     } else {
-        if (fullAlpha == 0xFF) {
+        if (fullAlpha == 0xFF && !noRealBlitter) {
             blitter->getRealBlitter()->blitAntiH2(x, y, a1, a2);
         } else {
             blitter->blitAntiH(x, y, a1);
@@ -593,13 +677,18 @@ static inline void blit_two_alphas(AdditiveBlitter* blitter, int y, int x,
 
 // It's important that this is inline. Otherwise it'll be much slower.
 static SK_ALWAYS_INLINE void blit_full_alpha(AdditiveBlitter* blitter, int y, int x, int len,
-                            SkAlpha fullAlpha, SkAlpha* maskRow, bool isUsingMask) {
+                            SkAlpha fullAlpha, SkAlpha* maskRow, bool isUsingMask,
+                            bool noRealBlitter, bool needSafeCheck) {
     if (isUsingMask) {
-        for (int i=0; i<len; i++) {
-            addAlpha(maskRow[x + i], fullAlpha);
+        for (int i = 0; i < len; ++i) {
+            if (needSafeCheck) {
+                safelyAddAlpha(&maskRow[x + i], fullAlpha);
+            } else {
+                addAlpha(&maskRow[x + i], fullAlpha);
+            }
         }
     } else {
-        if (fullAlpha == 0xFF) {
+        if (fullAlpha == 0xFF && !noRealBlitter) {
             blitter->getRealBlitter()->blitH(x, y, len);
         } else {
             blitter->blitAntiH(x, y, len, fullAlpha);
@@ -610,13 +699,14 @@ static SK_ALWAYS_INLINE void blit_full_alpha(AdditiveBlitter* blitter, int y, in
 static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
                                    SkFixed ul, SkFixed ur, SkFixed ll, SkFixed lr,
                                    SkFixed lDY, SkFixed rDY, SkAlpha fullAlpha, SkAlpha* maskRow,
-                                   bool isUsingMask) {
+                                   bool isUsingMask, bool noRealBlitter, bool needSafeCheck) {
     int L = SkFixedFloorToInt(ul), R = SkFixedCeilToInt(lr);
     int len = R - L;
 
     if (len == 1) {
         SkAlpha alpha = trapezoidToAlpha(ur - ul, lr - ll);
-        blit_single_alpha(blitter, y, L, alpha, fullAlpha, maskRow, isUsingMask);
+        blit_single_alpha(blitter, y, L, alpha, fullAlpha, maskRow, isUsingMask, noRealBlitter,
+                needSafeCheck);
         return;
     }
 
@@ -637,7 +727,7 @@ static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
     SkAlpha* tempAlphas = alphas + len + 1;
     int16_t* runs = (int16_t*)(alphas + (len + 1) * 2);
 
-    for (int i = 0; i < len; i++) {
+    for (int i = 0; i < len; ++i) {
         runs[i] = 1;
         alphas[i] = fullAlpha;
     }
@@ -655,7 +745,7 @@ static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
     } else {
         computeAlphaBelowLine(tempAlphas + uL - L, ul - (uL << 16), ll - (uL << 16),
                 lDY, fullAlpha);
-        for (int i = uL; i < lL; i++) {
+        for (int i = uL; i < lL; ++i) {
             if (alphas[i - L] > tempAlphas[i - L]) {
                 alphas[i - L] -= tempAlphas[i - L];
             } else {
@@ -676,7 +766,7 @@ static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
     } else {
         computeAlphaAboveLine(tempAlphas + uR - L, ur - (uR << 16), lr - (uR << 16),
                 rDY, fullAlpha);
-        for (int i = uR; i < lR; i++) {
+        for (int i = uR; i < lR; ++i) {
             if (alphas[i - L] > tempAlphas[i - L]) {
                 alphas[i - L] -= tempAlphas[i - L];
             } else {
@@ -686,11 +776,16 @@ static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
     }
 
     if (isUsingMask) {
-        for (int i=0; i<len; i++) {
-            addAlpha(maskRow[L + i], alphas[i]);
+        for (int i = 0; i < len; ++i) {
+            if (needSafeCheck) {
+                safelyAddAlpha(&maskRow[L + i], alphas[i]);
+            } else {
+                addAlpha(&maskRow[L + i], alphas[i]);
+            }
         }
     } else {
-        if (fullAlpha == 0xFF) { // Real blitter is faster than RunBasedAdditiveBlitter
+        if (fullAlpha == 0xFF && !noRealBlitter) {
+            // Real blitter is faster than RunBasedAdditiveBlitter
             blitter->getRealBlitter()->blitAntiH(L, y, alphas, runs);
         } else {
             blitter->blitAntiH(L, y, alphas, len);
@@ -702,10 +797,11 @@ static void blit_aaa_trapezoid_row(AdditiveBlitter* blitter, int y,
     }
 }
 
-static inline void blit_trapezoid_row(AdditiveBlitter* blitter, int y,
+static SK_ALWAYS_INLINE void blit_trapezoid_row(AdditiveBlitter* blitter, int y,
                                SkFixed ul, SkFixed ur, SkFixed ll, SkFixed lr,
                                SkFixed lDY, SkFixed rDY, SkAlpha fullAlpha,
-                               SkAlpha* maskRow, bool isUsingMask) {
+                               SkAlpha* maskRow, bool isUsingMask, bool noRealBlitter = false,
+                               bool needSafeCheck = false) {
     SkASSERT(lDY >= 0 && rDY >= 0); // We should only send in the absolte value
 
     if (ul > ur) {
@@ -743,16 +839,19 @@ static inline void blit_trapezoid_row(AdditiveBlitter* blitter, int y,
             int len = SkFixedCeilToInt(joinLeft - ul);
             if (len == 1) {
                 SkAlpha alpha = trapezoidToAlpha(joinLeft - ul, joinLeft - ll);
-                blit_single_alpha(blitter, y, ul >> 16, alpha, fullAlpha, maskRow, isUsingMask);
+                blit_single_alpha(blitter, y, ul >> 16, alpha, fullAlpha, maskRow, isUsingMask,
+                        noRealBlitter, needSafeCheck);
             } else if (len == 2) {
                 SkFixed first = joinLeft - SK_Fixed1 - ul;
                 SkFixed second = ll - ul - first;
                 SkAlpha a1 = partialTriangleToAlpha(first, lDY);
                 SkAlpha a2 = fullAlpha - partialTriangleToAlpha(second, lDY);
-                blit_two_alphas(blitter, y, ul >> 16, a1, a2, fullAlpha, maskRow, isUsingMask);
+                blit_two_alphas(blitter, y, ul >> 16, a1, a2, fullAlpha, maskRow, isUsingMask,
+                        noRealBlitter, needSafeCheck);
             } else {
                 blit_aaa_trapezoid_row(blitter, y, ul, joinLeft, ll, joinLeft, lDY, SK_MaxS32,
-                                       fullAlpha, maskRow, isUsingMask);
+                                       fullAlpha, maskRow, isUsingMask, noRealBlitter,
+                                       needSafeCheck);
             }
         }
         // SkAAClip requires that we blit from left to right.
@@ -760,29 +859,30 @@ static inline void blit_trapezoid_row(AdditiveBlitter* blitter, int y,
         if (joinLeft < joinRite) {
             blit_full_alpha(blitter, y, SkFixedFloorToInt(joinLeft),
                             SkFixedFloorToInt(joinRite - joinLeft),
-                            fullAlpha, maskRow, isUsingMask);
+                            fullAlpha, maskRow, isUsingMask, noRealBlitter, needSafeCheck);
         }
         if (lr > joinRite) {
             int len = SkFixedCeilToInt(lr - joinRite);
             if (len == 1) {
                 SkAlpha alpha = trapezoidToAlpha(ur - joinRite, lr - joinRite);
                 blit_single_alpha(blitter, y, joinRite >> 16, alpha, fullAlpha, maskRow,
-                                  isUsingMask);
+                                  isUsingMask, noRealBlitter, needSafeCheck);
             } else if (len == 2) {
                 SkFixed first = joinRite + SK_Fixed1 - ur;
                 SkFixed second = lr - ur - first;
                 SkAlpha a1 = fullAlpha - partialTriangleToAlpha(first, rDY);
                 SkAlpha a2 = partialTriangleToAlpha(second, rDY);
                 blit_two_alphas(blitter, y, joinRite >> 16, a1, a2, fullAlpha, maskRow,
-                                isUsingMask);
+                                isUsingMask, noRealBlitter, needSafeCheck);
             } else {
                 blit_aaa_trapezoid_row(blitter, y, joinRite, ur, joinRite, lr, SK_MaxS32, rDY,
-                                       fullAlpha, maskRow, isUsingMask);
+                                       fullAlpha, maskRow, isUsingMask, noRealBlitter,
+                                       needSafeCheck);
             }
         }
     } else {
         blit_aaa_trapezoid_row(blitter, y, ul, ur, ll, lr, lDY, rDY, fullAlpha, maskRow,
-                               isUsingMask);
+                               isUsingMask, noRealBlitter, needSafeCheck);
     }
 }
 
@@ -809,7 +909,7 @@ static SkAnalyticEdge* sort_edges(SkAnalyticEdge* list[], int count, SkAnalyticE
     SkTQSort(list, list + count - 1);
 
     // now make the edges linked in sorted order
-    for (int i = 1; i < count; i++) {
+    for (int i = 1; i < count; ++i) {
         list[i - 1]->fNext = list[i];
         list[i]->fPrev = list[i - 1];
     }
@@ -899,9 +999,9 @@ static inline bool isSmoothEnough(SkAnalyticEdge* leftE, SkAnalyticEdge* riteE,
     return isSmoothEnough(leftE, currE, stop_y) && isSmoothEnough(riteE, nextCurrE, stop_y);
 }
 
-static inline void aaa_walk_convex_edges(SkAnalyticEdge* prevHead, AdditiveBlitter* blitter,
-                           int start_y, int stop_y, SkFixed leftBound, SkFixed riteBound,
-                           bool isUsingMask) {
+static inline void aaa_walk_convex_edges(SkAnalyticEdge* prevHead,
+        AdditiveBlitter* blitter, int start_y, int stop_y, SkFixed leftBound, SkFixed riteBound,
+        bool isUsingMask) {
     validate_sort((SkAnalyticEdge*)prevHead->fNext);
 
     SkAnalyticEdge* leftE = (SkAnalyticEdge*) prevHead->fNext;
@@ -1130,13 +1230,407 @@ END_WALK:
     #endif
 }
 
-void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter* blitter,
-                   int start_y, int stop_y, bool pathContainedInClip, bool isUsingMask,
-                   bool forceRLE) { // forceRLE implies that SkAAClip is calling us
-    SkASSERT(blitter);
+///////////////////////////////////////////////////////////////////////////////
+
+static inline void remove_edge(SkAnalyticEdge* edge) {
+    edge->fPrev->fNext = edge->fNext;
+    edge->fNext->fPrev = edge->fPrev;
+}
+
+static inline void insert_edge_after(SkAnalyticEdge* edge, SkAnalyticEdge* afterMe) {
+    edge->fPrev = afterMe;
+    edge->fNext = afterMe->fNext;
+    afterMe->fNext->fPrev = edge;
+    afterMe->fNext = edge;
+}
+
+static void backward_insert_edge_based_on_x(SkAnalyticEdge* edge) {
+    SkFixed x = edge->fX;
+    SkAnalyticEdge* prev = edge->fPrev;
+    while (prev->fPrev && prev->fX > x) {
+        prev = prev->fPrev;
+    }
+    if (prev->fNext != edge) {
+        remove_edge(edge);
+        insert_edge_after(edge, prev);
+    }
+}
+
+static SkAnalyticEdge* backward_insert_start(SkAnalyticEdge* prev, SkFixed x) {
+    while (prev->fPrev && prev->fX > x) {
+        prev = prev->fPrev;
+    }
+    return prev;
+}
+
+static inline void updateNextNextY(SkFixed y, SkFixed nextY, SkFixed* nextNextY) {
+    *nextNextY = y > nextY && y < *nextNextY ? y : *nextNextY;
+}
+
+static inline void checkIntersection(const SkAnalyticEdge* edge, SkFixed nextY, SkFixed* nextNextY)
+{
+    if (edge->fPrev->fPrev && edge->fPrev->fX + edge->fPrev->fDX > edge->fX + edge->fDX) {
+        *nextNextY = nextY + (SK_Fixed1 >> SkAnalyticEdge::kDefaultAccuracy);
+    }
+}
+
+static void insert_new_edges(SkAnalyticEdge* newEdge, SkFixed y, SkFixed* nextNextY) {
+    if (newEdge->fUpperY > y) {
+        updateNextNextY(newEdge->fUpperY, y, nextNextY);
+        return;
+    }
+    SkAnalyticEdge* prev = newEdge->fPrev;
+    if (prev->fX <= newEdge->fX) {
+        while (newEdge->fUpperY <= y) {
+            checkIntersection(newEdge, y, nextNextY);
+            updateNextNextY(newEdge->fLowerY, y, nextNextY);
+            newEdge = newEdge->fNext;
+        }
+        updateNextNextY(newEdge->fUpperY, y, nextNextY);
+        return;
+    }
+    // find first x pos to insert
+    SkAnalyticEdge* start = backward_insert_start(prev, newEdge->fX);
+    //insert the lot, fixing up the links as we go
+    do {
+        SkAnalyticEdge* next = newEdge->fNext;
+        do {
+            if (start->fNext == newEdge) {
+                goto nextEdge;
+            }
+            SkAnalyticEdge* after = start->fNext;
+            if (after->fX >= newEdge->fX) {
+                break;
+            }
+            SkASSERT(start != after);
+            start = after;
+        } while (true);
+        remove_edge(newEdge);
+        insert_edge_after(newEdge, start);
+nextEdge:
+        checkIntersection(newEdge, y, nextNextY);
+        updateNextNextY(newEdge->fLowerY, y, nextNextY);
+        start = newEdge;
+        newEdge = next;
+    } while (newEdge->fUpperY <= y);
+    updateNextNextY(newEdge->fUpperY, y, nextNextY);
+}
+
+static void validate_edges_for_y(const SkAnalyticEdge* edge, SkFixed y) {
+#ifdef SK_DEBUG
+    while (edge->fUpperY <= y) {
+        SkASSERT(edge->fPrev && edge->fNext);
+        SkASSERT(edge->fPrev->fNext == edge);
+        SkASSERT(edge->fNext->fPrev == edge);
+        SkASSERT(edge->fUpperY <= edge->fLowerY);
+        SkASSERT(edge->fPrev->fPrev == nullptr || edge->fPrev->fX <= edge->fX);
+        edge = edge->fNext;
+    }
+#endif
+}
+
+// Return true if prev->fX, next->fX are too close in the current pixel row.
+static inline bool edges_too_close(SkAnalyticEdge* prev, SkAnalyticEdge* next, SkFixed lowerY) {
+    // Note that even if the following test failed, the edges might still be very close to each
+    // other at some point within the current pixel row because of prev->fDX and next->fDX.
+    // However, to handle that case, we have to sacrafice more performance.
+    // I think the current quality is good enough (mainly by looking at Nebraska-StateSeal.svg)
+    // so I'll ignore fDX for performance tradeoff.
+    return next && prev && next->fUpperY < lowerY && prev->fX >= next->fX - SkAbs32(next->fDX);
+    // The following is more accurate but also slower.
+    // return (prev && prev->fPrev && next && next->fNext != nullptr && next->fUpperY < lowerY &&
+    //     prev->fX + SkAbs32(prev->fDX) >= next->fX - SkAbs32(next->fDX));
+}
+
+// This function exists for the case where the previous rite edge is removed because
+// its fLowerY <= nextY
+static inline bool edges_too_close(int prevRite, SkFixed ul, SkFixed ll) {
+    return prevRite > SkFixedFloorToInt(ul) || prevRite > SkFixedFloorToInt(ll);
+}
+
+static inline void blit_saved_trapezoid(SkAnalyticEdge* leftE, SkFixed lowerY,
+        SkFixed lowerLeft, SkFixed lowerRite,
+        AdditiveBlitter* blitter, SkAlpha* maskRow, bool isUsingMask, bool noRealBlitter,
+        SkFixed leftClip, SkFixed rightClip) {
+    SkAnalyticEdge* riteE = leftE->fRiteE;
+    SkASSERT(riteE);
+    SkASSERT(riteE->fNext == nullptr || leftE->fSavedY == riteE->fSavedY);
+    SkASSERT(SkFixedFloorToInt(lowerY - 1) == SkFixedFloorToInt(leftE->fSavedY));
+    int y = SkFixedFloorToInt(leftE->fSavedY);
+    // Instead of using f2a(lowerY - leftE->fSavedY), we use the following fullAlpha
+    // to elimiate cumulative error: if there are many fractional y scan lines within the
+    // same row, the former may accumulate the rounding error while the later won't.
+    SkAlpha fullAlpha = f2a(lowerY - SkIntToFixed(y)) - f2a(leftE->fSavedY - SkIntToFixed(y));
+    // We need fSavedDY because the (quad or cubic) edge might be updated
+    blit_trapezoid_row(blitter, y,
+            SkTMax(leftE->fSavedX, leftClip), SkTMin(riteE->fSavedX, rightClip),
+            SkTMax(lowerLeft, leftClip), SkTMin(lowerRite, rightClip),
+            leftE->fSavedDY, riteE->fSavedDY, fullAlpha, maskRow, isUsingMask,
+            noRealBlitter ||
+                    (fullAlpha == 0xFF && (edges_too_close(leftE->fPrev, leftE, lowerY)
+                            || edges_too_close(riteE, riteE->fNext, lowerY))),
+            true);
+    leftE->fRiteE = nullptr;
+}
+
+static inline void deferred_blit(SkAnalyticEdge* leftE, SkAnalyticEdge* riteE,
+        SkFixed left, SkFixed leftDY, // don't save leftE->fX/fDY as they may have been updated
+        SkFixed y, SkFixed nextY, bool isIntegralNextY, bool leftEnds, bool riteEnds,
+        AdditiveBlitter* blitter, SkAlpha* maskRow, bool isUsingMask, bool noRealBlitter,
+        SkFixed leftClip, SkFixed rightClip, int yShift) {
+    if (leftE->fRiteE && leftE->fRiteE != riteE) {
+        // leftE's right edge changed. Blit the saved trapezoid.
+        SkASSERT(leftE->fRiteE->fNext == nullptr || leftE->fRiteE->fY == y);
+        blit_saved_trapezoid(leftE, y, left, leftE->fRiteE->fX,
+                blitter, maskRow, isUsingMask, noRealBlitter, leftClip, rightClip);
+    }
+    if (!leftE->fRiteE) {
+        // Save and defer blitting the trapezoid
+        SkASSERT(riteE->fRiteE == nullptr);
+        SkASSERT(leftE->fPrev == nullptr || leftE->fY == nextY);
+        SkASSERT(riteE->fNext == nullptr || riteE->fY == y);
+        leftE->saveXY(left, y, leftDY);
+        riteE->saveXY(riteE->fX, y, riteE->fDY);
+        leftE->fRiteE = riteE;
+    }
+    SkASSERT(leftE->fPrev == nullptr || leftE->fY == nextY);
+    riteE->goY(nextY, yShift);
+    // Always blit when edges end or nextY is integral
+    if (isIntegralNextY || leftEnds || riteEnds) {
+        blit_saved_trapezoid(leftE, nextY, leftE->fX, riteE->fX,
+                blitter, maskRow, isUsingMask, noRealBlitter, leftClip, rightClip);
+    }
+}
+
+static void aaa_walk_edges(SkAnalyticEdge* prevHead, SkAnalyticEdge* nextTail,
+        SkPath::FillType fillType, AdditiveBlitter* blitter, int start_y, int stop_y,
+        SkFixed leftClip, SkFixed rightClip, bool isUsingMask, bool forceRLE, bool useDeferred,
+        bool skipIntersect) {
+    prevHead->fX = prevHead->fUpperX = leftClip;
+    nextTail->fX = nextTail->fUpperX = rightClip;
+    SkFixed y = SkTMax(prevHead->fNext->fUpperY, SkIntToFixed(start_y));
+    SkFixed nextNextY = SK_MaxS32;
+
+    {
+        SkAnalyticEdge* edge;
+        for(edge = prevHead->fNext; edge->fUpperY <= y; edge = edge->fNext) {
+            edge->goY(y);
+            updateNextNextY(edge->fLowerY, y, &nextNextY);
+        }
+        updateNextNextY(edge->fUpperY, y, &nextNextY);
+    }
+
+    // returns 1 for evenodd, -1 for winding, regardless of inverse-ness
+    int windingMask = (fillType & 1) ? 1 : -1;
+
+    bool isInverse = SkPath::IsInverseFillType(fillType);
+
+    if (isInverse && SkIntToFixed(start_y) != y) {
+        int width = SkFixedFloorToInt(rightClip - leftClip);
+        if (SkFixedFloorToInt(y) != start_y) {
+            blitter->getRealBlitter()->blitRect(SkFixedFloorToInt(leftClip), start_y,
+                    width, SkFixedFloorToInt(y) - start_y);
+            start_y = SkFixedFloorToInt(y);
+        }
+        SkAlpha* maskRow = isUsingMask ? static_cast<MaskAdditiveBlitter*>(blitter)->getRow(start_y)
+                                       : nullptr;
+        blit_full_alpha(blitter, start_y, SkFixedFloorToInt(leftClip), width,
+                f2a(y - SkIntToFixed(start_y)), maskRow, isUsingMask, false, false);
+    }
+
+    while (true) {
+        int     w = 0;
+        bool    in_interval     = isInverse;
+        SkFixed prevX           = prevHead->fX;
+        SkFixed nextY           = SkTMin(nextNextY, SkFixedCeilToFixed(y + 1));
+        bool isIntegralNextY    = (nextY & (SK_Fixed1 - 1)) == 0;
+        SkAnalyticEdge* currE   = prevHead->fNext;
+        SkAnalyticEdge* leftE   = prevHead;
+        SkFixed left            = leftClip;
+        SkFixed leftDY          = 0;
+        bool leftEnds           = false;
+        int prevRite            = SkFixedFloorToInt(leftClip);
+
+        nextNextY               = SK_MaxS32;
+
+        SkASSERT((nextY & ((SK_Fixed1 >> 2) - 1)) == 0);
+        int yShift = 0;
+        if ((nextY - y) & (SK_Fixed1 >> 2)) {
+            yShift = 2;
+            nextY = y + (SK_Fixed1 >> 2);
+        } else if ((nextY - y) & (SK_Fixed1 >> 1)) {
+            yShift = 1;
+            SkASSERT(nextY == y + (SK_Fixed1 >> 1));
+        }
+
+        SkAlpha fullAlpha = f2a(nextY - y);
+
+        // If we're using mask blitter, we advance the mask row in this function
+        // to save some "if" condition checks.
+        SkAlpha* maskRow = nullptr;
+        if (isUsingMask) {
+            maskRow = static_cast<MaskAdditiveBlitter*>(blitter)->getRow(SkFixedFloorToInt(y));
+        }
+
+        SkASSERT(currE->fPrev == prevHead);
+        validate_edges_for_y(currE, y);
+
+        // Even if next - y == SK_Fixed1, we can still break the left-to-right order requirement
+        // of the SKAAClip: |\| (two trapezoids with overlapping middle wedges)
+        bool noRealBlitter = forceRLE; // forceRLE && (nextY - y != SK_Fixed1);
+
+        while (currE->fUpperY <= y) {
+            SkASSERT(currE->fLowerY >= nextY);
+            SkASSERT(currE->fY == y);
+
+            w += currE->fWinding;
+            bool prev_in_interval = in_interval;
+            in_interval = !(w & windingMask) == isInverse;
+
+            bool isLeft = in_interval && !prev_in_interval;
+            bool isRite = !in_interval && prev_in_interval;
+            bool currEnds = currE->fLowerY == nextY;
+
+            if (useDeferred) {
+                if (currE->fRiteE && !isLeft) {
+                    // currE is a left edge previously, but now it's not.
+                    // Blit the trapezoid between fSavedY and y.
+                    SkASSERT(currE->fRiteE->fY == y);
+                    blit_saved_trapezoid(currE, y, currE->fX, currE->fRiteE->fX,
+                            blitter, maskRow, isUsingMask, noRealBlitter, leftClip, rightClip);
+                }
+                if (leftE->fRiteE == currE && !isRite) {
+                    // currE is a right edge previously, but now it's not.
+                    // Moreover, its corresponding leftE doesn't change (otherwise we'll handle it
+                    // in the previous if clause). Hence we blit the trapezoid.
+                    blit_saved_trapezoid(leftE, y, left, currE->fX,
+                            blitter, maskRow, isUsingMask, noRealBlitter, leftClip, rightClip);
+                }
+            }
+
+            if (isRite) {
+                if (useDeferred) {
+                    deferred_blit(leftE, currE, left, leftDY, y, nextY, isIntegralNextY,
+                            leftEnds, currEnds, blitter, maskRow, isUsingMask, noRealBlitter,
+                            leftClip, rightClip, yShift);
+                } else {
+                    SkFixed rite = currE->fX;
+                    currE->goY(nextY, yShift);
+                    if (leftE->fDX < 0) {
+                        left = SkTMax(leftClip, left);
+                        leftE->fX = SkTMax(leftClip, leftE->fX);
+                    } else {
+                        left = SkTMin(rightClip, left);
+                        leftE->fX = SkTMin(rightClip, leftE->fX);
+                    }
+                    if (currE->fDX < 0) {
+                        rite = SkTMax(leftClip, rite);
+                        currE->fX = SkTMax(leftClip, currE->fX);
+                    } else {
+                        rite = SkTMin(rightClip, rite);
+                        currE->fX = SkTMin(rightClip, currE->fX);
+                    }
+                    blit_trapezoid_row(blitter, y >> 16, left, rite, leftE->fX, currE->fX,
+                            leftDY, currE->fDY, fullAlpha, maskRow, isUsingMask,
+                            noRealBlitter || (fullAlpha == 0xFF && (
+                                    edges_too_close(prevRite, left, leftE->fX) ||
+                                    edges_too_close(currE, currE->fNext, nextY)
+                            )),
+                            true);
+                    prevRite = SkFixedCeilToInt(SkTMax(rite, currE->fX));
+                }
+            } else {
+                if (isLeft) {
+                    left = currE->fX;
+                    leftDY = currE->fDY;
+                    leftE = currE;
+                    leftEnds = leftE->fLowerY == nextY;
+                }
+                currE->goY(nextY, yShift);
+            }
+
+
+            SkAnalyticEdge* next = currE->fNext;
+            SkFixed newX;
+
+            while (currE->fLowerY <= nextY) {
+                if (currE->fCurveCount < 0) {
+                    SkAnalyticCubicEdge* cubicEdge = (SkAnalyticCubicEdge*)currE;
+                    cubicEdge->keepContinuous();
+                    if (!cubicEdge->updateCubic()) {
+                        break;
+                    }
+                } else if (currE->fCurveCount > 0) {
+                    SkAnalyticQuadraticEdge* quadEdge = (SkAnalyticQuadraticEdge*)currE;
+                    quadEdge->keepContinuous();
+                    if (!quadEdge->updateQuadratic()) {
+                        break;
+                    }
+                } else {
+                    break;
+                }
+            }
+            SkASSERT(currE->fY == nextY);
+
+            if (currE->fLowerY <= nextY) {
+                remove_edge(currE);
+            } else {
+                updateNextNextY(currE->fLowerY, nextY, &nextNextY);
+                newX = currE->fX;
+                SkASSERT(currE->fLowerY > nextY);
+                if (newX < prevX) { // ripple currE backwards until it is x-sorted
+                    // If the crossing edge is a right edge, blit the saved trapezoid.
+                    if (leftE->fRiteE == currE && useDeferred) {
+                        SkASSERT(leftE->fY == nextY && currE->fY == nextY);
+                        blit_saved_trapezoid(leftE, nextY, leftE->fX, currE->fX,
+                                blitter, maskRow, isUsingMask, noRealBlitter, leftClip, rightClip);
+                    }
+                    backward_insert_edge_based_on_x(currE);
+                } else {
+                    prevX = newX;
+                }
+                if (!skipIntersect) {
+                   checkIntersection(currE, nextY, &nextNextY);
+                }
+            }
+
+            currE = next;
+            SkASSERT(currE);
+        }
+
+        // was our right-edge culled away?
+        if (in_interval) {
+            if (useDeferred) {
+                deferred_blit(leftE, nextTail, left, leftDY, y, nextY, isIntegralNextY,
+                        leftEnds, false, blitter, maskRow, isUsingMask, noRealBlitter,
+                        leftClip, rightClip, yShift);
+            } else {
+                blit_trapezoid_row(blitter, y >> 16, left, rightClip, leftE->fX, rightClip,
+                        leftDY, 0, fullAlpha, maskRow, isUsingMask,
+                        noRealBlitter ||
+                                (fullAlpha == 0xFF && edges_too_close(leftE->fPrev, leftE, nextY)),
+                        true);
+            }
+        }
+
+        if (forceRLE) {
+            ((RunBasedAdditiveBlitter*)blitter)->flush_if_y_changed(y, nextY);
+        }
+
+        y = nextY;
+        if (y >= SkIntToFixed(stop_y)) {
+            break;
+        }
 
-    // we only implemented the convex shapes yet
-    SkASSERT(!path.isInverseFillType() && path.isConvex());
+        // now currE points to the first edge with a fUpperY larger than the previous y
+        insert_new_edges(currE, y, &nextNextY);
+    }
+}
+
+static void aaa_fill_path(const SkPath& path, const SkIRect& clipRect,
+        AdditiveBlitter* blitter, int start_y, int stop_y, bool pathContainedInClip,
+        bool isUsingMask, bool forceRLE) { // forceRLE implies that SkAAClip is calling us
+    SkASSERT(blitter);
 
     SkEdgeBuilder   builder;
 
@@ -1166,7 +1660,8 @@ void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter*
                 rect.fBottom = stop_y;
             }
             if (!rect.isEmpty()) {
-                blitter->blitRect(rect.fLeft, rect.fTop, rect.width(), rect.height());
+                blitter->getRealBlitter()->blitRect(rect.fLeft, rect.fTop,
+                        rect.width(), rect.height());
             }
         }
         return;
@@ -1176,6 +1671,7 @@ void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter*
     // this returns the first and last edge after they're sorted into a dlink list
     SkAnalyticEdge* edge = sort_edges(list, count, &last);
 
+    headEdge.fRiteE = nullptr;
     headEdge.fPrev = nullptr;
     headEdge.fNext = edge;
     headEdge.fUpperY = headEdge.fLowerY = SK_MinS32;
@@ -1185,13 +1681,14 @@ void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter*
     headEdge.fUpperX = SK_MinS32;
     edge->fPrev = &headEdge;
 
+    tailEdge.fRiteE = nullptr;
     tailEdge.fPrev = last;
     tailEdge.fNext = nullptr;
     tailEdge.fUpperY = tailEdge.fLowerY = SK_MaxS32;
-    headEdge.fX = SK_MaxS32;
-    headEdge.fDX = 0;
-    headEdge.fDY = SK_MaxS32;
-    headEdge.fUpperX = SK_MaxS32;
+    tailEdge.fX = SK_MaxS32;
+    tailEdge.fDX = 0;
+    tailEdge.fDY = SK_MaxS32;
+    tailEdge.fUpperX = SK_MaxS32;
     last->fNext = &tailEdge;
 
     // now edge is the head of the sorted linklist
@@ -1203,22 +1700,32 @@ void aaa_fill_path(const SkPath& path, const SkIRect& clipRect, AdditiveBlitter*
         stop_y = clipRect.fBottom;
     }
 
+    SkFixed leftBound = SkIntToFixed(rect.fLeft);
+    SkFixed rightBound = SkIntToFixed(rect.fRight);
+    if (isUsingMask) {
+        // If we're using mask, then we have to limit the bound within the path bounds.
+        // Otherwise, the edge drift may access an invalid address inside the mask.
+        SkIRect ir;
+        path.getBounds().roundOut(&ir);
+        leftBound = SkTMax(leftBound, SkIntToFixed(ir.fLeft));
+        rightBound = SkTMin(rightBound, SkIntToFixed(ir.fRight));
+    }
+
     if (!path.isInverseFillType() && path.isConvex()) {
         SkASSERT(count >= 2);   // convex walker does not handle missing right edges
-        SkFixed leftBound = SkIntToFixed(rect.fLeft);
-        SkFixed rightBound = SkIntToFixed(rect.fRight);
-        if (isUsingMask) {
-            // If we're using mask, then we have to limit the bound within the path bounds.
-            // Otherwise, the edge drift may access an invalid address inside the mask.
-            SkIRect ir;
-            path.getBounds().roundOut(&ir);
-            leftBound = SkTMax(leftBound, SkIntToFixed(ir.fLeft));
-            rightBound = SkTMin(rightBound, SkIntToFixed(ir.fRight));
-        }
         aaa_walk_convex_edges(&headEdge, blitter, start_y, stop_y,
                               leftBound, rightBound, isUsingMask);
     } else {
-        SkFAIL("Concave AAA is not yet implemented!");
+        // Only use deferred blitting if there are many edges.
+        bool useDeferred = count >
+                (SkFixedFloorToInt(tailEdge.fPrev->fLowerY - headEdge.fNext->fUpperY) + 1) * 4;
+
+        // We skip intersection computation if there are many points which probably already
+        // give us enough fractional scan lines.
+        bool skipIntersect = path.countPoints() > (stop_y - start_y) / 2;
+
+        aaa_walk_edges(&headEdge, &tailEdge, path.getFillType(), blitter, start_y, stop_y,
+               leftBound, rightBound, isUsingMask, forceRLE, useDeferred, skipIntersect);
     }
 }
 
@@ -1268,11 +1775,13 @@ void SkScan::AAAFillPath(const SkPath& path, const SkRegion& origClip, SkBlitter
     if (origClip.isEmpty()) {
         return;
     }
+    #ifdef SK_SUPPORT_LEGACY_AAA
     if (path.isInverseFillType() || !path.isConvex()) {
         // Fall back as we only implemented the algorithm for convex shapes yet.
         SkScan::AntiFillPath(path, origClip, blitter, forceRLE);
         return;
     }
+    #endif
 
     const bool isInverse = path.isInverseFillType();
     SkIRect ir;
@@ -1337,9 +1846,7 @@ void SkScan::AAAFillPath(const SkPath& path, const SkRegion& origClip, SkBlitter
     blitter = clipper.getBlitter();
 
     if (isInverse) {
-        // Currently, we use the old path to render the inverse path,
-        // so we don't need this.
-        // sk_blit_above(blitter, ir, *clipRgn);
+        sk_blit_above(blitter, ir, *clipRgn);
     }
 
     SkASSERT(SkIntToScalar(ir.fTop) <= path.getBounds().fTop);
@@ -1348,16 +1855,18 @@ void SkScan::AAAFillPath(const SkPath& path, const SkRegion& origClip, SkBlitter
         MaskAdditiveBlitter additiveBlitter(blitter, ir, *clipRgn, isInverse);
         aaa_fill_path(path, clipRgn->getBounds(), &additiveBlitter, ir.fTop, ir.fBottom,
                 clipRect == nullptr, true, forceRLE);
-    } else {
+    } else if (!isInverse && path.isConvex()) {
         RunBasedAdditiveBlitter additiveBlitter(blitter, ir, *clipRgn, isInverse);
         aaa_fill_path(path, clipRgn->getBounds(), &additiveBlitter, ir.fTop, ir.fBottom,
                 clipRect == nullptr, false, forceRLE);
+    } else {
+        SafeRLEAdditiveBlitter additiveBlitter(blitter, ir, *clipRgn, isInverse);
+        aaa_fill_path(path, clipRgn->getBounds(), &additiveBlitter, ir.fTop, ir.fBottom,
+                clipRect == nullptr, false, forceRLE);
     }
 
     if (isInverse) {
-        // Currently, we use the old path to render the inverse path,
-        // so we don't need this.
-        // sk_blit_below(blitter, ir, *clipRgn);
+        sk_blit_below(blitter, ir, *clipRgn);
     }
 }
 
index 1d214e3..16c91a9 100644 (file)
@@ -753,9 +753,27 @@ void SkScan::FillPath(const SkPath& path, const SkRasterClip& clip,
     }
 }
 
+static bool suitableForAAA(const SkPath& path) {
+#ifdef SK_SUPPORT_LEGACY_AAA
+    return true;
+#endif
+    if (gSkForceAnalyticAA.load()) {
+        return true;
+    }
+    const SkRect& bounds = path.getBounds();
+    // When the path have so many points compared to the size of its bounds/resolution,
+    // it indicates that the path is not quite smooth in the current resolution:
+    // the expected number of turning points in every pixel row/column is significantly greater than
+    // zero. Hence Aanlytic AA is not likely to produce visible quality improvents, and Analytic AA
+    // might be slower than supersampling.
+    return path.countPoints() < SkTMax(bounds.width(), bounds.height()) / 2 - 10;
+}
+
 void SkScan::AntiFillPath(const SkPath& path, const SkRasterClip& clip,
                           SkBlitter* blitter) {
-    if (gSkUseAnalyticAA.load()) {
+    // Do not use AAA if path is too complicated:
+    // there won't be any speedup or significant visual improvement.
+    if (gSkUseAnalyticAA.load() && suitableForAAA(path)) {
         SkScan::AAAFillPath(path, clip, blitter);
         return;
     }
index 52a5388..0148bcd 100644 (file)
@@ -68,6 +68,10 @@ DEFINE_bool2(pre_log, p, false, "Log before running each test. May be incomprehe
 
 DEFINE_bool(analyticAA, true, "If false, disable analytic anti-aliasing");
 
+DEFINE_bool(forceAnalyticAA, false, "Force analytic anti-aliasing even if the path is complicated: "
+                                    "whether it's concave or convex, we consider a path complicated"
+                                    "if its number of points is comparable to its resolution.");
+
 bool CollectImages(SkCommandLineFlags::StringArray images, SkTArray<SkString>* output) {
     SkASSERT(output);
 
index 1a707c8..92ac141 100644 (file)
@@ -33,6 +33,7 @@ DECLARE_bool(veryVerbose);
 DECLARE_string(writePath);
 DECLARE_bool(pre_log);
 DECLARE_bool(analyticAA);
+DECLARE_bool(forceAnalyticAA);
 
 DECLARE_string(key);
 DECLARE_string(properties);