AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
aae533e418
skia2/tests/PathOpsAngleIdeas.cpp

884 lines
35 KiB
C++
Raw Normal View History

Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
/*
* Copyright 2013 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "PathOpsTestCommon.h"
#include "SkIntersections.h"
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
#include "SkOpContour.h"
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
#include "SkOpSegment.h"
#include "SkRandom.h"
#include "SkTArray.h"
#include "SkTSort.h"
#include "Test.h"
static bool gPathOpsAngleIdeasVerbose = false;
static bool gPathOpsAngleIdeasEnableBruteCheck = false;
class PathOpsAngleTester {
public:
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
static int ConvexHullOverlaps(SkOpAngle& lh, SkOpAngle& rh) {
return lh.convexHullOverlaps(&rh);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
}
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
static int EndsIntersect(SkOpAngle& lh, SkOpAngle& rh) {
return lh.endsIntersect(&rh);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
}
};
struct TRange {
double tMin1;
double tMin2;
double t1;
double t2;
double tMin;
double a1;
double a2;
bool ccw;
};
static double testArc(skiatest::Reporter* reporter, const SkDQuad& quad, const SkDQuad& arcRef,
int octant) {
SkDQuad arc = arcRef;
SkDVector offset = {quad[0].fX, quad[0].fY};
arc[0] += offset;
arc[1] += offset;
arc[2] += offset;
SkIntersections i;
i.intersect(arc, quad);
if (i.used() == 0) {
return -1;
}
int smallest = -1;
double t = 2;
for (int idx = 0; idx < i.used(); ++idx) {
if (i[0][idx] > 1 || i[0][idx] < 0) {
i.reset();
i.intersect(arc, quad);
}
if (i[1][idx] > 1 || i[1][idx] < 0) {
i.reset();
i.intersect(arc, quad);
}
if (t > i[1][idx]) {
smallest = idx;
t = i[1][idx];
}
}
REPORTER_ASSERT(reporter, smallest >= 0);
REPORTER_ASSERT(reporter, t >= 0 && t <= 1);
return i[1][smallest];
}
static void orderQuads(skiatest::Reporter* reporter, const SkDQuad& quad, double radius,
SkTArray<double, false>* tArray) {
double r = radius;
double s = r * SK_ScalarTanPIOver8;
double m = r * SK_ScalarRoot2Over2;
// construct circle from quads
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
const QuadPts circle[8] = {{{{ r, 0}, { r, -s}, { m, -m}}},
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{{{ m, -m}, { s, -r}, { 0, -r}}},
{{{ 0, -r}, {-s, -r}, {-m, -m}}},
{{{-m, -m}, {-r, -s}, {-r, 0}}},
{{{-r, 0}, {-r, s}, {-m, m}}},
{{{-m, m}, {-s, r}, { 0, r}}},
{{{ 0, r}, { s, r}, { m, m}}},
{{{ m, m}, { r, s}, { r, 0}}}};
for (int octant = 0; octant < 8; ++octant) {
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
SkDQuad cQuad;
cQuad.debugSet(circle[octant].fPts);
double t = testArc(reporter, quad, cQuad, octant);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
if (t < 0) {
continue;
}
for (int index = 0; index < tArray->count(); ++index) {
double matchT = (*tArray)[index];
if (approximately_equal(t, matchT)) {
goto next;
}
}
tArray->push_back(t);
next: ;
}
}
static double quadAngle(skiatest::Reporter* reporter, const SkDQuad& quad, double t) {
const SkDVector& pt = quad.ptAtT(t) - quad[0];
double angle = (atan2(pt.fY, pt.fX) + SK_ScalarPI) * 8 / (SK_ScalarPI * 2);
REPORTER_ASSERT(reporter, angle >= 0 && angle <= 8);
return angle;
}
static bool angleDirection(double a1, double a2) {
double delta = a1 - a2;
return (delta < 4 && delta > 0) || delta < -4;
}
static void setQuadHullSweep(const SkDQuad& quad, SkDVector sweep[2]) {
sweep[0] = quad[1] - quad[0];
sweep[1] = quad[2] - quad[0];
}
static double distEndRatio(double dist, const SkDQuad& quad) {
SkDVector v[] = {quad[2] - quad[0], quad[1] - quad[0], quad[2] - quad[1]};
double longest = SkTMax(v[0].length(), SkTMax(v[1].length(), v[2].length()));
return longest / dist;
}
static bool checkParallel(skiatest::Reporter* reporter, const SkDQuad& quad1, const SkDQuad& quad2) {
SkDVector sweep[2], tweep[2];
setQuadHullSweep(quad1, sweep);
setQuadHullSweep(quad2, tweep);
// if the ctrl tangents are not nearly parallel, use them
// solve for opposite direction displacement scale factor == m
// initial dir = v1.cross(v2) == v2.x * v1.y - v2.y * v1.x
// displacement of q1[1] : dq1 = { -m * v1.y, m * v1.x } + q1[1]
// straight angle when : v2.x * (dq1.y - q1[0].y) == v2.y * (dq1.x - q1[0].x)
// v2.x * (m * v1.x + v1.y) == v2.y * (-m * v1.y + v1.x)
// - m * (v2.x * v1.x + v2.y * v1.y) == v2.x * v1.y - v2.y * v1.x
// m = (v2.y * v1.x - v2.x * v1.y) / (v2.x * v1.x + v2.y * v1.y)
// m = v1.cross(v2) / v1.dot(v2)
double s0dt0 = sweep[0].dot(tweep[0]);
REPORTER_ASSERT(reporter, s0dt0 != 0);
double s0xt0 = sweep[0].crossCheck(tweep[0]);
double m = s0xt0 / s0dt0;
double sDist = sweep[0].length() * m;
double tDist = tweep[0].length() * m;
bool useS = fabs(sDist) < fabs(tDist);
double mFactor = fabs(useS ? distEndRatio(sDist, quad1) : distEndRatio(tDist, quad2));
if (mFactor < 5000) { // empirically found limit
return s0xt0 < 0;
}
SkDVector m0 = quad1.ptAtT(0.5) - quad1[0];
SkDVector m1 = quad2.ptAtT(0.5) - quad2[0];
return m0.crossCheck(m1) < 0;
}
/* returns
-1 if overlaps
0 if no overlap cw
1 if no overlap ccw
*/
static int quadHullsOverlap(skiatest::Reporter* reporter, const SkDQuad& quad1,
const SkDQuad& quad2) {
SkDVector sweep[2], tweep[2];
setQuadHullSweep(quad1, sweep);
setQuadHullSweep(quad2, tweep);
double s0xs1 = sweep[0].crossCheck(sweep[1]);
double s0xt0 = sweep[0].crossCheck(tweep[0]);
double s1xt0 = sweep[1].crossCheck(tweep[0]);
bool tBetweenS = s0xs1 > 0 ? s0xt0 > 0 && s1xt0 < 0 : s0xt0 < 0 && s1xt0 > 0;
double s0xt1 = sweep[0].crossCheck(tweep[1]);
double s1xt1 = sweep[1].crossCheck(tweep[1]);
tBetweenS |= s0xs1 > 0 ? s0xt1 > 0 && s1xt1 < 0 : s0xt1 < 0 && s1xt1 > 0;
double t0xt1 = tweep[0].crossCheck(tweep[1]);
if (tBetweenS) {
return -1;
}
if ((s0xt0 == 0 && s1xt1 == 0) || (s1xt0 == 0 && s0xt1 == 0)) { // s0 to s1 equals t0 to t1
return -1;
}
bool sBetweenT = t0xt1 > 0 ? s0xt0 < 0 && s0xt1 > 0 : s0xt0 > 0 && s0xt1 < 0;
sBetweenT |= t0xt1 > 0 ? s1xt0 < 0 && s1xt1 > 0 : s1xt0 > 0 && s1xt1 < 0;
if (sBetweenT) {
return -1;
}
// if all of the sweeps are in the same half plane, then the order of any pair is enough
if (s0xt0 >= 0 && s0xt1 >= 0 && s1xt0 >= 0 && s1xt1 >= 0) {
return 0;
}
if (s0xt0 <= 0 && s0xt1 <= 0 && s1xt0 <= 0 && s1xt1 <= 0) {
return 1;
}
// if the outside sweeps are greater than 180 degress:
// first assume the inital tangents are the ordering
// if the midpoint direction matches the inital order, that is enough
SkDVector m0 = quad1.ptAtT(0.5) - quad1[0];
SkDVector m1 = quad2.ptAtT(0.5) - quad2[0];
double m0xm1 = m0.crossCheck(m1);
if (s0xt0 > 0 && m0xm1 > 0) {
return 0;
}
if (s0xt0 < 0 && m0xm1 < 0) {
return 1;
}
REPORTER_ASSERT(reporter, s0xt0 != 0);
return checkParallel(reporter, quad1, quad2);
}
static double radianSweep(double start, double end) {
double sweep = end - start;
if (sweep > SK_ScalarPI) {
sweep -= 2 * SK_ScalarPI;
} else if (sweep < -SK_ScalarPI) {
sweep += 2 * SK_ScalarPI;
}
return sweep;
}
static bool radianBetween(double start, double test, double end) {
double startToEnd = radianSweep(start, end);
double startToTest = radianSweep(start, test);
double testToEnd = radianSweep(test, end);
return (startToTest <= 0 && testToEnd <= 0 && startToTest >= startToEnd) ||
(startToTest >= 0 && testToEnd >= 0 && startToTest <= startToEnd);
}
static bool orderTRange(skiatest::Reporter* reporter, const SkDQuad& quad1, const SkDQuad& quad2,
double r, TRange* result) {
SkTArray<double, false> t1Array, t2Array;
orderQuads(reporter, quad1, r, &t1Array);
orderQuads(reporter,quad2, r, &t2Array);
if (!t1Array.count() || !t2Array.count()) {
return false;
}
SkTQSort<double>(t1Array.begin(), t1Array.end() - 1);
SkTQSort<double>(t2Array.begin(), t2Array.end() - 1);
double t1 = result->tMin1 = t1Array[0];
double t2 = result->tMin2 = t2Array[0];
double a1 = quadAngle(reporter,quad1, t1);
double a2 = quadAngle(reporter,quad2, t2);
if (approximately_equal(a1, a2)) {
return false;
}
bool refCCW = angleDirection(a1, a2);
result->t1 = t1;
result->t2 = t2;
result->tMin = SkTMin(t1, t2);
result->a1 = a1;
result->a2 = a2;
result->ccw = refCCW;
return true;
}
static bool equalPoints(const SkDPoint& pt1, const SkDPoint& pt2, double max) {
return approximately_zero_when_compared_to(pt1.fX - pt2.fX, max)
&& approximately_zero_when_compared_to(pt1.fY - pt2.fY, max);
}
static double maxDist(const SkDQuad& quad) {
SkDRect bounds;
bounds.setBounds(quad);
SkDVector corner[4] = {
{ bounds.fLeft - quad[0].fX, bounds.fTop - quad[0].fY },
{ bounds.fRight - quad[0].fX, bounds.fTop - quad[0].fY },
{ bounds.fLeft - quad[0].fX, bounds.fBottom - quad[0].fY },
{ bounds.fRight - quad[0].fX, bounds.fBottom - quad[0].fY }
};
double max = 0;
for (unsigned index = 0; index < SK_ARRAY_COUNT(corner); ++index) {
max = SkTMax(max, corner[index].length());
}
return max;
}
static double maxQuad(const SkDQuad& quad) {
double max = 0;
for (int index = 0; index < 2; ++index) {
max = SkTMax(max, fabs(quad[index].fX));
max = SkTMax(max, fabs(quad[index].fY));
}
return max;
}
static bool bruteMinT(skiatest::Reporter* reporter, const SkDQuad& quad1, const SkDQuad& quad2,
TRange* lowerRange, TRange* upperRange) {
double maxRadius = SkTMin(maxDist(quad1), maxDist(quad2));
double maxQuads = SkTMax(maxQuad(quad1), maxQuad(quad2));
double r = maxRadius / 2;
double rStep = r / 2;
SkDPoint best1 = {SK_ScalarInfinity, SK_ScalarInfinity};
SkDPoint best2 = {SK_ScalarInfinity, SK_ScalarInfinity};
int bestCCW = -1;
double bestR = maxRadius;
upperRange->tMin = 0;
lowerRange->tMin = 1;
do {
do { // find upper bounds of single result
TRange tRange;
bool stepUp = orderTRange(reporter, quad1, quad2, r, &tRange);
if (stepUp) {
SkDPoint pt1 = quad1.ptAtT(tRange.t1);
if (equalPoints(pt1, best1, maxQuads)) {
break;
}
best1 = pt1;
SkDPoint pt2 = quad2.ptAtT(tRange.t2);
if (equalPoints(pt2, best2, maxQuads)) {
break;
}
best2 = pt2;
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("u bestCCW=%d ccw=%d bestMin=%1.9g:%1.9g r=%1.9g tMin=%1.9g\n",
bestCCW, tRange.ccw, lowerRange->tMin, upperRange->tMin, r,
tRange.tMin);
}
if (bestCCW >= 0 && bestCCW != (int) tRange.ccw) {
if (tRange.tMin < upperRange->tMin) {
upperRange->tMin = 0;
} else {
stepUp = false;
}
}
if (upperRange->tMin < tRange.tMin) {
bestCCW = tRange.ccw;
bestR = r;
*upperRange = tRange;
}
if (lowerRange->tMin > tRange.tMin) {
*lowerRange = tRange;
}
}
r += stepUp ? rStep : -rStep;
rStep /= 2;
} while (rStep > FLT_EPSILON);
if (bestCCW < 0) {
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
if (bestR >= maxRadius) {
SkDebugf("");
}
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
REPORTER_ASSERT(reporter, bestR < maxRadius);
return false;
}
double lastHighR = bestR;
r = bestR / 2;
rStep = r / 2;
do { // find lower bounds of single result
TRange tRange;
bool success = orderTRange(reporter, quad1, quad2, r, &tRange);
if (success) {
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("l bestCCW=%d ccw=%d bestMin=%1.9g:%1.9g r=%1.9g tMin=%1.9g\n",
bestCCW, tRange.ccw, lowerRange->tMin, upperRange->tMin, r,
tRange.tMin);
}
if (bestCCW != (int) tRange.ccw || upperRange->tMin < tRange.tMin) {
bestCCW = tRange.ccw;
*upperRange = tRange;
bestR = lastHighR;
break; // need to establish a new upper bounds
}
SkDPoint pt1 = quad1.ptAtT(tRange.t1);
SkDPoint pt2 = quad2.ptAtT(tRange.t2);
if (equalPoints(pt1, best1, maxQuads)) {
goto breakOut;
}
best1 = pt1;
if (equalPoints(pt2, best2, maxQuads)) {
goto breakOut;
}
best2 = pt2;
if (equalPoints(pt1, pt2, maxQuads)) {
success = false;
} else {
if (upperRange->tMin < tRange.tMin) {
*upperRange = tRange;
}
if (lowerRange->tMin > tRange.tMin) {
*lowerRange = tRange;
}
}
lastHighR = SkTMin(r, lastHighR);
}
r += success ? -rStep : rStep;
rStep /= 2;
} while (rStep > FLT_EPSILON);
} while (rStep > FLT_EPSILON);
breakOut:
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("l a2-a1==%1.9g\n", lowerRange->a2 - lowerRange->a1);
}
return true;
}
static void bruteForce(skiatest::Reporter* reporter, const SkDQuad& quad1, const SkDQuad& quad2,
bool ccw) {
if (!gPathOpsAngleIdeasEnableBruteCheck) {
return;
}
TRange lowerRange, upperRange;
bool result = bruteMinT(reporter, quad1, quad2, &lowerRange, &upperRange);
REPORTER_ASSERT(reporter, result);
double angle = fabs(lowerRange.a2 - lowerRange.a1);
REPORTER_ASSERT(reporter, angle > 3.998 || ccw == upperRange.ccw);
}
static bool bruteForceCheck(skiatest::Reporter* reporter, const SkDQuad& quad1,
const SkDQuad& quad2, bool ccw) {
TRange lowerRange, upperRange;
bool result = bruteMinT(reporter, quad1, quad2, &lowerRange, &upperRange);
REPORTER_ASSERT(reporter, result);
return ccw == upperRange.ccw;
}
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
static void makeSegment(SkOpContour* contour, const SkDQuad& quad, SkPoint shortQuad[3]) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
shortQuad[0] = quad[0].asSkPoint();
shortQuad[1] = quad[1].asSkPoint();
shortQuad[2] = quad[2].asSkPoint();
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
contour->addQuad(shortQuad);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
}
static void testQuadAngles(skiatest::Reporter* reporter, const SkDQuad& quad1, const SkDQuad& quad2,
Move from SkChunkAlloc to SkArenaAlloc for PathOps Attempt two. Remove ~SkOpContour because it is handled by the SkArenaAlloc. Change-Id: Id3049db97aebcc1009d403a031f2fac219f58f2f Reviewed-on: https://skia-review.googlesource.com/9381 Reviewed-by: Derek Sollenberger <djsollen@google.com> Commit-Queue: Herb Derby <herb@google.com>
2017-03-07 16:11:47 +00:00
int testNo, SkArenaAlloc* allocator) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
SkPoint shortQuads[2][3];
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
Path ops formerly found the topmost unprocessed edge and determined its angle sort order to initialize the winding. This never worked correctly with cubics and was flaky with paths consisting mostly of vertical edges. This replacement shoots axis-aligned rays through all intersecting edges to find the outermost one either horizontally or vertically. The resulting code is smaller and twice as fast. To support this, most of the horizontal / vertical intersection code was rewritten and standardized, and old code supporting the top-directed winding was deleted. Contours were pointed to by an SkTDArray. Instead, put them in a linked list, and designate the list head with its own class to ensure that methods that take lists of contours start at the top. This change removed a large percentage of memory allocations used by path ops. TBR=reed@google.com BUG=skia:3588 Review URL: https://codereview.chromium.org/1111333002
2015-05-11 14:21:27 +00:00
SkOpContourHead contour;
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
SkOpGlobalState state(&contour, allocator SkDEBUGPARAMS(false) SkDEBUGPARAMS(nullptr));
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
contour.init(&state, false, false);
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
makeSegment(&contour, quad1, shortQuads[0]);
makeSegment(&contour, quad1, shortQuads[1]);
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
SkOpSegment* seg1 = contour.first();
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
seg1->debugAddAngle(0, 1);
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
SkOpSegment* seg2 = seg1->next();
pathops coincidence and security rewrite Most changes stem from working on an examples bracketed by #if DEBUG_UNDER_DEVELOPMENT // tiger These exposed many problems with coincident curves, as well as errors throughout the code. Fixing these errors also fixed a number of fuzzer-inspired bug reports. * Line/Curve Intersections Check to see if the end of the line nearly intersects the curve. This was a FIXME in the old code. * Performance Use a central chunk allocator. Plumb the allocator into the global variable state so that it can be shared. (Note that 'SkGlobalState' is allocated on the stack and is visible to children functions but not other threads.) * Refactor Let SkOpAngle grow up from a structure to a class. Let SkCoincidentSpans grow up from a structure to a class. Rename enum Alias to AliasMatch. * Coincidence Rewrite Add more debugging to coincidence detection. Parallel debugging routines have read-only logic to report the current coincidence state so that steps through the logic can expose whether things got better or worse. More functions can error-out and cause the pathops engine to non-destructively exit. * Accuracy Remove code that adjusted point locations. Instead, offset the curve part so that sorted curves all use the same origin. Reduce the size (and influence) of magic numbers. * Testing The debug suite with verify and the full release suite ./out/Debug/pathops_unittest -v -V ./out/Release/pathops_unittest -v -V -x expose one error. That error is captured as cubics_d3. This error exists in the checked in code as well. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2128633003 Review-Url: https://codereview.chromium.org/2128633003
2016-07-18 17:01:36 +00:00
seg2->debugAddAngle(0, 1);
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
int realOverlap = PathOpsAngleTester::ConvexHullOverlaps(*seg1->debugLastAngle(),
*seg2->debugLastAngle());
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
const SkDPoint& origin = quad1[0];
REPORTER_ASSERT(reporter, origin == quad2[0]);
double a1s = atan2(origin.fY - quad1[1].fY, quad1[1].fX - origin.fX);
double a1e = atan2(origin.fY - quad1[2].fY, quad1[2].fX - origin.fX);
double a2s = atan2(origin.fY - quad2[1].fY, quad2[1].fX - origin.fX);
double a2e = atan2(origin.fY - quad2[2].fY, quad2[2].fX - origin.fX);
bool oldSchoolOverlap = radianBetween(a1s, a2s, a1e)
|| radianBetween(a1s, a2e, a1e) || radianBetween(a2s, a1s, a2e)
|| radianBetween(a2s, a1e, a2e);
int overlap = quadHullsOverlap(reporter, quad1, quad2);
bool realMatchesOverlap = realOverlap == overlap || SK_ScalarPI - fabs(a2s - a1s) < 0.002;
if (realOverlap != overlap) {
SkDebugf("\nSK_ScalarPI - fabs(a2s - a1s) = %1.9g\n", SK_ScalarPI - fabs(a2s - a1s));
}
if (!realMatchesOverlap) {
DumpQ(quad1, quad2, testNo);
}
REPORTER_ASSERT(reporter, realMatchesOverlap);
if (oldSchoolOverlap != (overlap < 0)) {
overlap = quadHullsOverlap(reporter, quad1, quad2); // set a breakpoint and debug if assert fires
REPORTER_ASSERT(reporter, oldSchoolOverlap == (overlap < 0));
}
SkDVector v1s = quad1[1] - quad1[0];
SkDVector v1e = quad1[2] - quad1[0];
SkDVector v2s = quad2[1] - quad2[0];
SkDVector v2e = quad2[2] - quad2[0];
double vDir[2] = { v1s.cross(v1e), v2s.cross(v2e) };
bool ray1In2 = v1s.cross(v2s) * vDir[1] <= 0 && v1s.cross(v2e) * vDir[1] >= 0;
bool ray2In1 = v2s.cross(v1s) * vDir[0] <= 0 && v2s.cross(v1e) * vDir[0] >= 0;
if (overlap >= 0) {
// verify that hulls really don't overlap
REPORTER_ASSERT(reporter, !ray1In2);
REPORTER_ASSERT(reporter, !ray2In1);
bool ctrl1In2 = v1e.cross(v2s) * vDir[1] <= 0 && v1e.cross(v2e) * vDir[1] >= 0;
REPORTER_ASSERT(reporter, !ctrl1In2);
bool ctrl2In1 = v2e.cross(v1s) * vDir[0] <= 0 && v2e.cross(v1e) * vDir[0] >= 0;
REPORTER_ASSERT(reporter, !ctrl2In1);
// check answer against reference
bruteForce(reporter, quad1, quad2, overlap > 0);
}
// continue end point rays and see if they intersect the opposite curve
SkDLine rays[] = {{{origin, quad2[2]}}, {{origin, quad1[2]}}};
const SkDQuad* quads[] = {&quad1, &quad2};
SkDVector midSpokes[2];
SkIntersections intersect[2];
double minX, minY, maxX, maxY;
minX = minY = SK_ScalarInfinity;
maxX = maxY = -SK_ScalarInfinity;
double maxWidth = 0;
bool useIntersect = false;
double smallestTs[] = {1, 1};
for (unsigned index = 0; index < SK_ARRAY_COUNT(quads); ++index) {
const SkDQuad& q = *quads[index];
midSpokes[index] = q.ptAtT(0.5) - origin;
minX = SkTMin(SkTMin(SkTMin(minX, origin.fX), q[1].fX), q[2].fX);
minY = SkTMin(SkTMin(SkTMin(minY, origin.fY), q[1].fY), q[2].fY);
maxX = SkTMax(SkTMax(SkTMax(maxX, origin.fX), q[1].fX), q[2].fX);
maxY = SkTMax(SkTMax(SkTMax(maxY, origin.fY), q[1].fY), q[2].fY);
maxWidth = SkTMax(maxWidth, SkTMax(maxX - minX, maxY - minY));
intersect[index].intersectRay(q, rays[index]);
const SkIntersections& i = intersect[index];
REPORTER_ASSERT(reporter, i.used() >= 1);
bool foundZero = false;
double smallT = 1;
for (int idx2 = 0; idx2 < i.used(); ++idx2) {
double t = i[0][idx2];
if (t == 0) {
foundZero = true;
continue;
}
if (smallT > t) {
smallT = t;
}
}
REPORTER_ASSERT(reporter, foundZero == true);
if (smallT == 1) {
continue;
}
SkDVector ray = q.ptAtT(smallT) - origin;
SkDVector end = rays[index][1] - origin;
if (ray.fX * end.fX < 0 || ray.fY * end.fY < 0) {
continue;
}
double rayDist = ray.length();
double endDist = end.length();
double delta = fabs(rayDist - endDist) / maxWidth;
if (delta > 1e-4) {
useIntersect ^= true;
}
smallestTs[index] = smallT;
}
bool firstInside;
if (useIntersect) {
int sIndex = (int) (smallestTs[1] < 1);
REPORTER_ASSERT(reporter, smallestTs[sIndex ^ 1] == 1);
double t = smallestTs[sIndex];
const SkDQuad& q = *quads[sIndex];
SkDVector ray = q.ptAtT(t) - origin;
SkDVector end = rays[sIndex][1] - origin;
double rayDist = ray.length();
double endDist = end.length();
SkDVector mid = q.ptAtT(t / 2) - origin;
double midXray = mid.crossCheck(ray);
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("rayDist>endDist:%d sIndex==0:%d vDir[sIndex]<0:%d midXray<0:%d\n",
rayDist > endDist, sIndex == 0, vDir[sIndex] < 0, midXray < 0);
}
SkASSERT(SkScalarSignAsInt(SkDoubleToScalar(midXray))
== SkScalarSignAsInt(SkDoubleToScalar(vDir[sIndex])));
firstInside = (rayDist > endDist) ^ (sIndex == 0) ^ (vDir[sIndex] < 0);
} else if (overlap >= 0) {
return; // answer has already been determined
} else {
firstInside = checkParallel(reporter, quad1, quad2);
}
if (overlap < 0) {
SkDEBUGCODE(int realEnds =)
cumulative pathops patch Replace the implicit curve intersection with a geometric curve intersection. The implicit intersection proved mathematically unstable and took a long time to zero in on an answer. Use pointers instead of indices to refer to parts of curves. Indices required awkward renumbering. Unify t and point values so that small intervals can be eliminated in one pass. Break cubics up front to eliminate loops and cusps. Make the Simplify and Op code more regular and eliminate arbitrary differences. Add a builder that takes an array of paths and operators. Delete unused code. BUG=skia:3588 R=reed@google.com Review URL: https://codereview.chromium.org/1037573004
2015-03-26 14:52:43 +00:00
PathOpsAngleTester::EndsIntersect(*seg1->debugLastAngle(),
*seg2->debugLastAngle());
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
SkASSERT(realEnds == (firstInside ? 1 : 0));
}
bruteForce(reporter, quad1, quad2, firstInside);
}
DEF_TEST(PathOpsAngleOverlapHullsOne, reporter) {
SkSTArenaAlloc Syntactic sugar, gets rid of some boilerplate. Change-Id: Ibdb28b7a8f1d5e4a4e18c12d423b987d7194e340 Reviewed-on: https://skia-review.googlesource.com/17837 Commit-Queue: Florin Malita <fmalita@chromium.org> Reviewed-by: Herb Derby <herb@google.com>
2017-05-24 18:53:44 +00:00
SkSTArenaAlloc<4096> allocator;
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
// gPathOpsAngleIdeasVerbose = true;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
const QuadPts quads[] = {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{{{939.4808349609375, 914.355224609375}, {-357.7921142578125, 590.842529296875}, {736.8936767578125, -350.717529296875}}},
{{{939.4808349609375, 914.355224609375}, {-182.85418701171875, 634.4552001953125}, {-509.62615966796875, 576.1182861328125}}}
};
for (int index = 0; index < (int) SK_ARRAY_COUNT(quads); index += 2) {
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
SkDQuad quad0, quad1;
quad0.debugSet(quads[index].fPts);
quad1.debugSet(quads[index + 1].fPts);
testQuadAngles(reporter, quad0, quad1, 0, &allocator);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
}
}
DEF_TEST(PathOpsAngleOverlapHulls, reporter) {
SkSTArenaAlloc Syntactic sugar, gets rid of some boilerplate. Change-Id: Ibdb28b7a8f1d5e4a4e18c12d423b987d7194e340 Reviewed-on: https://skia-review.googlesource.com/17837 Commit-Queue: Florin Malita <fmalita@chromium.org> Reviewed-by: Herb Derby <herb@google.com>
2017-05-24 18:53:44 +00:00
SkSTArenaAlloc<4096> allocator;
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
if (!gPathOpsAngleIdeasVerbose) { // takes a while to run -- so exclude it by default
return;
}
SkRandom ran;
for (int index = 0; index < 100000; ++index) {
if (index % 1000 == 999) SkDebugf(".");
SkDPoint origin = {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
QuadPts quad1 = {{origin, {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)},
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)}}};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
if (quad1.fPts[0] == quad1.fPts[2]) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
continue;
}
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
QuadPts quad2 = {{origin, {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)},
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)}}};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
if (quad2.fPts[0] == quad2.fPts[2]) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
continue;
}
SkIntersections i;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
SkDQuad q1, q2;
q1.debugSet(quad1.fPts);
q2.debugSet(quad2.fPts);
i.intersect(q1, q2);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
REPORTER_ASSERT(reporter, i.used() >= 1);
if (i.used() > 1) {
continue;
}
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
testQuadAngles(reporter, q1, q2, index, &allocator);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
}
}
DEF_TEST(PathOpsAngleBruteT, reporter) {
if (!gPathOpsAngleIdeasVerbose) { // takes a while to run -- so exclude it by default
return;
}
SkRandom ran;
double smaller = SK_Scalar1;
SkDQuad small[2];
SkDEBUGCODE(int smallIndex);
for (int index = 0; index < 100000; ++index) {
SkDPoint origin = {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
QuadPts quad1 = {{origin, {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)},
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)}}};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
if (quad1.fPts[0] == quad1.fPts[2]) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
continue;
}
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
QuadPts quad2 = {{origin, {ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)},
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{ran.nextRangeF(-1000, 1000), ran.nextRangeF(-1000, 1000)}}};
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
if (quad2.fPts[0] == quad2.fPts[2]) {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
continue;
}
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
SkDQuad q1, q2;
q1.debugSet(quad1.fPts);
q2.debugSet(quad2.fPts);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
SkIntersections i;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
i.intersect(q1, q2);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
REPORTER_ASSERT(reporter, i.used() >= 1);
if (i.used() > 1) {
continue;
}
TRange lowerRange, upperRange;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
bool result = bruteMinT(reporter, q1, q2, &lowerRange, &upperRange);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
REPORTER_ASSERT(reporter, result);
double min = SkTMin(upperRange.t1, upperRange.t2);
if (smaller > min) {
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
small[0] = q1;
small[1] = q2;
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
SkDEBUGCODE(smallIndex = index);
smaller = min;
}
}
#ifdef SK_DEBUG
DumpQ(small[0], small[1], smallIndex);
#endif
}
DEF_TEST(PathOpsAngleBruteTOne, reporter) {
// gPathOpsAngleIdeasVerbose = true;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
const QuadPts qPts[] = {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{{{-770.8492431640625, 948.2369384765625}, {-853.37066650390625, 972.0301513671875}, {-200.62042236328125, -26.7174072265625}}},
{{{-770.8492431640625, 948.2369384765625}, {513.602783203125, 578.8681640625}, {960.641357421875, -813.69757080078125}}},
{{{563.8267822265625, -107.4566650390625}, {-44.67724609375, -136.57452392578125}, {492.3856201171875, -268.79644775390625}}},
{{{563.8267822265625, -107.4566650390625}, {708.049072265625, -100.77789306640625}, {-48.88226318359375, 967.9022216796875}}},
{{{598.857421875, 846.345458984375}, {-644.095703125, -316.12921142578125}, {-97.64599609375, 20.6158447265625}}},
{{{598.857421875, 846.345458984375}, {715.7142333984375, 955.3599853515625}, {-919.9478759765625, 691.611328125}}},
};
TRange lowerRange, upperRange;
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
SkDQuad quads[SK_ARRAY_COUNT(qPts)];
for (int index = 0; index < (int) SK_ARRAY_COUNT(qPts); ++index) {
quads[index].debugSet(qPts[index].fPts);
}
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
bruteMinT(reporter, quads[0], quads[1], &lowerRange, &upperRange);
bruteMinT(reporter, quads[2], quads[3], &lowerRange, &upperRange);
bruteMinT(reporter, quads[4], quads[5], &lowerRange, &upperRange);
}
/*
The sorting problem happens when the inital tangents are not a true indicator of the curve direction
Nearly always, the initial tangents do give the right answer,
so the trick is to figure out when the initial tangent cannot be trusted.
If the convex hulls of both curves are in the same half plane, and not overlapping, sorting the
hulls is enough.
If the hulls overlap, and have the same general direction, then intersect the shorter end point ray
with the opposing curve, and see on which side of the shorter curve the opposing intersection lies.
Otherwise, if the control vector is extremely short, likely the point on curve must be computed
If moving the control point slightly can change the sign of the cross product, either answer could
be "right".
We need to determine how short is extremely short. Move the control point a set percentage of
the largest length to determine how stable the curve is vis-a-vis the initial tangent.
*/
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
static const QuadPts extremeTests[][2] = {
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
{
{{{-708.0077926931004,-154.61669472244046},
{-707.9234268635319,-154.30459999551294},
{505.58447265625,-504.9130859375}}},
{{{-708.0077926931004,-154.61669472244046},
{-711.127526325141,-163.9446090624656},
{-32.39227294921875,-906.3277587890625}}},
}, {
{{{-708.0077926931004,-154.61669472244046},
{-708.2875337527566,-154.36676458635623},
{505.58447265625,-504.9130859375}}},
{{{-708.0077926931004,-154.61669472244046},
{-708.4111557216864,-154.5366642875255},
{-32.39227294921875,-906.3277587890625}}},
}, {
{{{-609.0230951752058,-267.5435593490574},
{-594.1120809906336,-136.08492475411555},
{505.58447265625,-504.9130859375}}},
{{{-609.0230951752058,-267.5435593490574},
{-693.7467719138988,-341.3259237831895},
{-32.39227294921875,-906.3277587890625}}}
}, {
{{{-708.0077926931004,-154.61669472244046},
{-707.9994640658723,-154.58588461064852},
{505.58447265625,-504.9130859375}}},
{{{-708.0077926931004,-154.61669472244046},
{-708.0239418990758,-154.6403553507124},
{-32.39227294921875,-906.3277587890625}}}
}, {
{{{-708.0077926931004,-154.61669472244046},
{-707.9993222215099,-154.55999389855003},
{68.88981098017803,296.9273945411635}}},
{{{-708.0077926931004,-154.61669472244046},
{-708.0509091919608,-154.64675214697067},
{-777.4871194247767,-995.1470120113145}}}
}, {
{{{-708.0077926931004,-154.61669472244046},
{-708.0060491116379,-154.60889321524968},
{229.97088707895057,-430.0569357467175}}},
{{{-708.0077926931004,-154.61669472244046},
{-708.013911296257,-154.6219143988058},
{138.13162892614037,-573.3689311737394}}}
}, {
{{{-543.2570545751013,-237.29243831075053},
{-452.4119186056987,-143.47223056267802},
{229.97088707895057,-430.0569357467175}}},
{{{-543.2570545751013,-237.29243831075053},
{-660.5330371214436,-362.0016148388},
{138.13162892614037,-573.3689311737394}}},
},
};
static double endCtrlRatio(const SkDQuad quad) {
SkDVector longEdge = quad[2] - quad[0];
double longLen = longEdge.length();
SkDVector shortEdge = quad[1] - quad[0];
double shortLen = shortEdge.length();
return longLen / shortLen;
}
static void computeMV(const SkDQuad& quad, const SkDVector& v, double m, SkDVector mV[2]) {
SkDPoint mPta = {quad[1].fX - m * v.fY, quad[1].fY + m * v.fX};
SkDPoint mPtb = {quad[1].fX + m * v.fY, quad[1].fY - m * v.fX};
mV[0] = mPta - quad[0];
mV[1] = mPtb - quad[0];
}
static double mDistance(skiatest::Reporter* reporter, bool agrees, const SkDQuad& q1,
const SkDQuad& q2) {
if (1 && agrees) {
return SK_ScalarMax;
}
// how close is the angle from inflecting in the opposite direction?
SkDVector v1 = q1[1] - q1[0];
SkDVector v2 = q2[1] - q2[0];
double dir = v1.crossCheck(v2);
REPORTER_ASSERT(reporter, dir != 0);
// solve for opposite direction displacement scale factor == m
// initial dir = v1.cross(v2) == v2.x * v1.y - v2.y * v1.x
// displacement of q1[1] : dq1 = { -m * v1.y, m * v1.x } + q1[1]
// straight angle when : v2.x * (dq1.y - q1[0].y) == v2.y * (dq1.x - q1[0].x)
// v2.x * (m * v1.x + v1.y) == v2.y * (-m * v1.y + v1.x)
// - m * (v2.x * v1.x + v2.y * v1.y) == v2.x * v1.y - v2.y * v1.x
// m = (v2.y * v1.x - v2.x * v1.y) / (v2.x * v1.x + v2.y * v1.y)
// m = v1.cross(v2) / v1.dot(v2)
double div = v1.dot(v2);
REPORTER_ASSERT(reporter, div != 0);
double m = dir / div;
SkDVector mV1[2], mV2[2];
computeMV(q1, v1, m, mV1);
computeMV(q2, v2, m, mV2);
double dist1 = v1.length() * m;
double dist2 = v2.length() * m;
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("%c r1=%1.9g r2=%1.9g m=%1.9g dist1=%1.9g dist2=%1.9g "
" dir%c 1a=%1.9g 1b=%1.9g 2a=%1.9g 2b=%1.9g\n", agrees ? 'T' : 'F',
endCtrlRatio(q1), endCtrlRatio(q2), m, dist1, dist2, dir > 0 ? '+' : '-',
mV1[0].crossCheck(v2), mV1[1].crossCheck(v2),
mV2[0].crossCheck(v1), mV2[1].crossCheck(v1));
}
if (1) {
bool use1 = fabs(dist1) < fabs(dist2);
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("%c dist=%1.9g r=%1.9g\n", agrees ? 'T' : 'F', use1 ? dist1 : dist2,
use1 ? distEndRatio(dist1, q1) : distEndRatio(dist2, q2));
}
return fabs(use1 ? distEndRatio(dist1, q1) : distEndRatio(dist2, q2));
}
return SK_ScalarMax;
}
static void midPointAgrees(skiatest::Reporter* reporter, const SkDQuad& q1, const SkDQuad& q2,
bool ccw) {
SkDPoint mid1 = q1.ptAtT(0.5);
SkDVector m1 = mid1 - q1[0];
SkDPoint mid2 = q2.ptAtT(0.5);
SkDVector m2 = mid2 - q2[0];
REPORTER_ASSERT(reporter, ccw ? m1.crossCheck(m2) < 0 : m1.crossCheck(m2) > 0);
}
DEF_TEST(PathOpsAngleExtreme, reporter) {
if (!gPathOpsAngleIdeasVerbose) { // takes a while to run -- so exclude it by default
return;
}
double maxR = SK_ScalarMax;
for (int index = 0; index < (int) SK_ARRAY_COUNT(extremeTests); ++index) {
fix fuzzers Many old pathops-related fuzz failures have built up while the codebase was under a state a flux. Now that the code is stable, address these failures. Most of the CL plumbs the debug global state to downstream routines so that, if the data is not trusted (ala fuzzed) the function can safely exit without asserting. TBR=reed@google.com GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2426173002 Review-Url: https://chromiumcodereview.appspot.com/2426173002
2016-10-20 15:32:18 +00:00
const QuadPts& qu1 = extremeTests[index][0];
const QuadPts& qu2 = extremeTests[index][1];
SkDQuad quad1, quad2;
quad1.debugSet(qu1.fPts);
quad2.debugSet(qu2.fPts);
Mike R: please sanity check SkPostConfig.h Mike K: please sanity check Test.cpp and skia_test.cpp Feel free to look at the rest, but I don't expect any in depth review of path ops innards. Path Ops first iteration used QuickSort to order segments radiating from an intersection to compute the winding rule. This revision uses a circular sort instead. Breaking out the circular sort into its own long-lived structure (SkOpAngle) allows doing less work and provides a home for caching additional sorting data. The circle sort is more stable than the former sort, has a robust ordering and fewer exceptions. It finds unsortable ordering less often. It is less reliant on the initial curve tangent, using convex hulls instead whenever it can. Additional debug validation makes sure that the computed structures are self-consistent. A new visualization tool helps verify that the angle ordering is correct. The 70+M tests pass with this change on Windows, Mac, Linux 32 and Linux 64 in debug and release. R=mtklein@google.com, reed@google.com Author: caryclark@google.com Review URL: https://codereview.chromium.org/131103009 git-svn-id: http://skia.googlecode.com/svn/trunk@14183 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-14 17:08:59 +00:00
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("%s %d\n", __FUNCTION__, index);
}
REPORTER_ASSERT(reporter, quad1[0] == quad2[0]);
SkIntersections i;
i.intersect(quad1, quad2);
REPORTER_ASSERT(reporter, i.used() == 1);
REPORTER_ASSERT(reporter, i.pt(0) == quad1[0]);
int overlap = quadHullsOverlap(reporter, quad1, quad2);
REPORTER_ASSERT(reporter, overlap >= 0);
SkDVector sweep[2], tweep[2];
setQuadHullSweep(quad1, sweep);
setQuadHullSweep(quad2, tweep);
double s0xt0 = sweep[0].crossCheck(tweep[0]);
REPORTER_ASSERT(reporter, s0xt0 != 0);
bool ccw = s0xt0 < 0;
bool agrees = bruteForceCheck(reporter, quad1, quad2, ccw);
maxR = SkTMin(maxR, mDistance(reporter, agrees, quad1, quad2));
if (agrees) {
continue;
}
midPointAgrees(reporter, quad1, quad2, !ccw);
SkDQuad q1 = quad1;
SkDQuad q2 = quad2;
double loFail = 1;
double hiPass = 2;
// double vectors until t passes
do {
q1[1].fX = quad1[0].fX * (1 - hiPass) + quad1[1].fX * hiPass;
q1[1].fY = quad1[0].fY * (1 - hiPass) + quad1[1].fY * hiPass;
q2[1].fX = quad2[0].fX * (1 - hiPass) + quad2[1].fX * hiPass;
q2[1].fY = quad2[0].fY * (1 - hiPass) + quad2[1].fY * hiPass;
agrees = bruteForceCheck(reporter, q1, q2, ccw);
maxR = SkTMin(maxR, mDistance(reporter, agrees, q1, q2));
if (agrees) {
break;
}
midPointAgrees(reporter, quad1, quad2, !ccw);
loFail = hiPass;
hiPass *= 2;
} while (true);
// binary search to find minimum pass
double midTest = (loFail + hiPass) / 2;
double step = (hiPass - loFail) / 4;
while (step > FLT_EPSILON) {
q1[1].fX = quad1[0].fX * (1 - midTest) + quad1[1].fX * midTest;
q1[1].fY = quad1[0].fY * (1 - midTest) + quad1[1].fY * midTest;
q2[1].fX = quad2[0].fX * (1 - midTest) + quad2[1].fX * midTest;
q2[1].fY = quad2[0].fY * (1 - midTest) + quad2[1].fY * midTest;
agrees = bruteForceCheck(reporter, q1, q2, ccw);
maxR = SkTMin(maxR, mDistance(reporter, agrees, q1, q2));
if (!agrees) {
midPointAgrees(reporter, quad1, quad2, !ccw);
}
midTest += agrees ? -step : step;
step /= 2;
}
#ifdef SK_DEBUG
// DumpQ(q1, q2, 999);
#endif
}
if (gPathOpsAngleIdeasVerbose) {
SkDebugf("maxR=%1.9g\n", maxR);
}
}
Reference in New Issue Copy Permalink