AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
efaf53ba24
skia2/tests/PathOpsAngleTest.cpp

486 lines
20 KiB
C++
Raw Normal View History

path ops work in progress standardize tests use SK_ARRAY_COUNT everywhere debug why x87 differs from SIMD 64 various platform specific fixes git-svn-id: http://skia.googlecode.com/svn/trunk@8689 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-04-15 19:13:59 +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.
*/
harden and speed up path op unit tests PathOps tests internal routines direcctly. Check to make sure that test points, lines, quads, curves, triangles, and bounds read from arrays are valid (i.e., don't contain NaN) before calling the test function. Repurpose the test flags. - make 'v' verbose test region output against path output - make 'z' single threaded (before it made it multithreaded) The latter change speeds up tests run by the buildbot by 2x to 3x. BUG= Review URL: https://codereview.chromium.org/19374003 git-svn-id: http://skia.googlecode.com/svn/trunk@10107 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-07-16 16:11:16 +00:00
#include "PathOpsTestCommon.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 "SkIntersections.h"
path ops work in progress standardize tests use SK_ARRAY_COUNT everywhere debug why x87 differs from SIMD 64 various platform specific fixes git-svn-id: http://skia.googlecode.com/svn/trunk@8689 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-04-15 19:13:59 +00:00
#include "SkOpSegment.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 "SkPathOpsTriangle.h"
#include "SkRandom.h"
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
#include "SkTArray.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 "SkTSort.h"
path ops work in progress standardize tests use SK_ARRAY_COUNT everywhere debug why x87 differs from SIMD 64 various platform specific fixes git-svn-id: http://skia.googlecode.com/svn/trunk@8689 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-04-15 19:13:59 +00:00
#include "Test.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
static bool gDisableAngleTests = true;
static float next(float f)
{
int fBits = SkFloatAs2sCompliment(f);
++fBits;
float fNext = Sk2sComplimentAsFloat(fBits);
return fNext;
path ops work in progress standardize tests use SK_ARRAY_COUNT everywhere debug why x87 differs from SIMD 64 various platform specific fixes git-svn-id: http://skia.googlecode.com/svn/trunk@8689 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-04-15 19:13:59 +00:00
}
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 float prev(float f)
{
int fBits = SkFloatAs2sCompliment(f);
--fBits;
float fNext = Sk2sComplimentAsFloat(fBits);
return fNext;
}
DEF_TEST(PathOpsAngleFindCrossEpsilon, reporter) {
if (gDisableAngleTests) {
path ops work in progress BUG= Review URL: https://codereview.chromium.org/18058007 git-svn-id: http://skia.googlecode.com/svn/trunk@9908 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-07-08 17:17:02 +00:00
return;
}
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
SkRandom ran;
int maxEpsilon = 0;
for (int index = 0; index < 10000000; ++index) {
SkDLine line = {{{0, 0}, {ran.nextRangeF(0.0001f, 1000), ran.nextRangeF(0.0001f, 1000)}}};
for (int inner = 0; inner < 10; ++inner) {
float t = ran.nextRangeF(0.0001f, 1);
SkDPoint dPt = line.ptAtT(t);
SkPoint pt = dPt.asSkPoint();
float xs[3] = { prev(pt.fX), pt.fX, next(pt.fX) };
float ys[3] = { prev(pt.fY), pt.fY, next(pt.fY) };
for (int xIdx = 0; xIdx < 3; ++xIdx) {
for (int yIdx = 0; yIdx < 3; ++yIdx) {
SkPoint test = { xs[xIdx], ys[yIdx] };
float p1 = SkDoubleToScalar(line[1].fX * test.fY);
float p2 = SkDoubleToScalar(line[1].fY * test.fX);
int p1Bits = SkFloatAs2sCompliment(p1);
int p2Bits = SkFloatAs2sCompliment(p2);
int epsilon = abs(p1Bits - p2Bits);
if (maxEpsilon < epsilon) {
SkDebugf("line={{0, 0}, {%1.7g, %1.7g}} t=%1.7g pt={%1.7g, %1.7g}"
" epsilon=%d\n",
line[1].fX, line[1].fY, t, test.fX, test.fY, epsilon);
maxEpsilon = epsilon;
}
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
}
}
}
path ops work in progress BUG= Review URL: https://codereview.chromium.org/18058007 git-svn-id: http://skia.googlecode.com/svn/trunk@9908 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-07-08 17:17:02 +00:00
}
}
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(PathOpsAngleFindQuadEpsilon, reporter) {
if (gDisableAngleTests) {
return;
}
SkRandom ran;
int maxEpsilon = 0;
double maxAngle = 0;
for (int index = 0; index < 100000; ++index) {
SkDLine line = {{{0, 0}, {ran.nextRangeF(0.0001f, 1000), ran.nextRangeF(0.0001f, 1000)}}};
float t = ran.nextRangeF(0.0001f, 1);
SkDPoint dPt = line.ptAtT(t);
float t2 = ran.nextRangeF(0.0001f, 1);
SkDPoint qPt = line.ptAtT(t2);
float t3 = ran.nextRangeF(0.0001f, 1);
SkDPoint qPt2 = line.ptAtT(t3);
qPt.fX += qPt2.fY;
qPt.fY -= qPt2.fX;
SkDQuad quad = {{line[0], dPt, qPt}};
// binary search for maximum movement of quad[1] towards test that still has 1 intersection
double moveT = 0.5f;
double deltaT = moveT / 2;
SkDPoint last;
do {
last = quad[1];
quad[1].fX = dPt.fX - line[1].fY * moveT;
quad[1].fY = dPt.fY + line[1].fX * moveT;
SkIntersections i;
i.intersect(quad, line);
REPORTER_ASSERT(reporter, i.used() > 0);
if (i.used() == 1) {
moveT += deltaT;
} else {
moveT -= deltaT;
}
deltaT /= 2;
} while (last.asSkPoint() != quad[1].asSkPoint());
float p1 = SkDoubleToScalar(line[1].fX * last.fY);
float p2 = SkDoubleToScalar(line[1].fY * last.fX);
int p1Bits = SkFloatAs2sCompliment(p1);
int p2Bits = SkFloatAs2sCompliment(p2);
int epsilon = abs(p1Bits - p2Bits);
if (maxEpsilon < epsilon) {
SkDebugf("line={{0, 0}, {%1.7g, %1.7g}} t=%1.7g/%1.7g/%1.7g moveT=%1.7g"
" pt={%1.7g, %1.7g} epsilon=%d\n",
line[1].fX, line[1].fY, t, t2, t3, moveT, last.fX, last.fY, epsilon);
maxEpsilon = epsilon;
}
double a1 = atan2(line[1].fY, line[1].fX);
double a2 = atan2(last.fY, last.fX);
double angle = fabs(a1 - a2);
if (maxAngle < angle) {
SkDebugf("line={{0, 0}, {%1.7g, %1.7g}} t=%1.7g/%1.7g/%1.7g moveT=%1.7g"
" pt={%1.7g, %1.7g} angle=%1.7g\n",
line[1].fX, line[1].fY, t, t2, t3, moveT, last.fX, last.fY, angle);
maxAngle = angle;
}
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
}
}
static int find_slop(double x, double y, double rx, double ry) {
int slopBits = 0;
bool less1, less2;
double absX = fabs(x);
double absY = fabs(y);
double length = absX < absY ? absX / 2 + absY : absX + absY / 2;
int exponent;
(void) frexp(length, &exponent);
double epsilon = ldexp(FLT_EPSILON, exponent);
do {
// get the length as the larger plus half the smaller (both same signs)
// find the ulps of the length
// compute the offsets from there
double xSlop = epsilon * slopBits;
double ySlop = x * y < 0 ? -xSlop : xSlop; // OPTIMIZATION: use copysign / _copysign ?
double x1 = x - xSlop;
double y1 = y + ySlop;
double x_ry1 = x1 * ry;
double rx_y1 = rx * y1;
less1 = x_ry1 < rx_y1;
double x2 = x + xSlop;
double y2 = y - ySlop;
double x_ry2 = x2 * ry;
double rx_y2 = rx * y2;
less2 = x_ry2 < rx_y2;
} while (less1 == less2 && ++slopBits);
return slopBits;
}
// from http://stackoverflow.com/questions/1427422/cheap-algorithm-to-find-measure-of-angle-between-vectors
static double diamond_angle(double y, double x)
{
if (y >= 0)
Sanitizing source files in Housekeeper-Nightly git-svn-id: http://skia.googlecode.com/svn/trunk@9440 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-05 07:01:06 +00:00
return (x >= 0 ? y/(x+y) : 1-x/(-x+y));
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
else
Sanitizing source files in Housekeeper-Nightly git-svn-id: http://skia.googlecode.com/svn/trunk@9440 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-05 07:01:06 +00:00
return (x < 0 ? 2-y/(-x-y) : 3+x/(x-y));
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
}
static const double slopTests[][4] = {
// x y rx ry
{-0.058554756452593892, -0.18804585843827226, -0.018568569646021160, -0.059615294434479438},
{-0.0013717412948608398, 0.0041152238845825195, -0.00045837944195925573, 0.0013753175735478074},
{-2.1033774145221198, -1.4046019261273715e-008, -0.70062688352066704, -1.2706324683777995e-008},
};
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(PathOpsAngleFindSlop, reporter) {
if (gDisableAngleTests) {
return;
}
for (int index = 0; index < (int) SK_ARRAY_COUNT(slopTests); ++index) {
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
const double* slopTest = slopTests[index];
double x = slopTest[0];
double y = slopTest[1];
double rx = slopTest[2];
double ry = slopTest[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
SkDebugf("%s xy %d=%d\n", __FUNCTION__, index, find_slop(x, y, rx, ry));
SkDebugf("%s rxy %d=%d\n", __FUNCTION__, index, find_slop(rx, ry, x, y));
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
double angle = diamond_angle(y, x);
double rAngle = diamond_angle(ry, rx);
double diff = fabs(angle - rAngle);
SkDebugf("%s diamond xy=%1.9g rxy=%1.9g diff=%1.9g factor=%d\n", __FUNCTION__,
angle, rAngle, diff, (int) (diff / FLT_EPSILON));
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
}
}
class PathOpsAngleTester {
public:
static int After(const SkOpAngle& lh, const SkOpAngle& rh) {
return lh.after(&rh);
}
static int ConvexHullOverlaps(const SkOpAngle& lh, const SkOpAngle& rh) {
return lh.convexHullOverlaps(rh);
}
static int Orderable(const SkOpAngle& lh, const SkOpAngle& rh) {
return lh.orderable(rh);
}
static int EndsIntersect(const SkOpAngle& lh, const SkOpAngle& rh) {
return lh.endsIntersect(rh);
}
static void SetNext(SkOpAngle& lh, SkOpAngle& rh) {
lh.fNext = &rh;
}
};
class PathOpsSegmentTester {
public:
static void ConstructCubic(SkOpSegment* segment, SkPoint shortCubic[4]) {
segment->debugConstructCubic(shortCubic);
}
static void ConstructLine(SkOpSegment* segment, SkPoint shortLine[2]) {
segment->debugConstructLine(shortLine);
}
path ops -- rewrite angle sort This is a major change resulting from a minor tweak. In the old code, the intersection point of two curves was shared between them, but the intersection points and end points of sorted edges was computed directly from the intersection T value. In this CL, both intersection points and sorted points are the same, and intermediate control points are computed to preserve their slope. The sort itself has been completely rewritten to be more robust and remove 'magic' checks, conditions that empirically worked but couldn't be rationalized. This CL was triggered by errors generated computing the clips of SKP files. At this point, all 73M standard tests work and at least the first troublesome SKPs work. Review URL: https://codereview.chromium.org/15338003 git-svn-id: http://skia.googlecode.com/svn/trunk@9432 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-06-04 17:59:42 +00:00
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 ConstructQuad(SkOpSegment* segment, SkPoint shortQuad[3]) {
segment->debugConstructQuad(shortQuad);
path ops work in progress standardize tests use SK_ARRAY_COUNT everywhere debug why x87 differs from SIMD 64 various platform specific fixes git-svn-id: http://skia.googlecode.com/svn/trunk@8689 2bbb7eff-a529-9590-31e7-b0007b416f81
2013-04-15 19:13:59 +00:00
}
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 DebugReset(SkOpSegment* segment) {
segment->debugReset();
}
};
struct CircleData {
const SkDCubic fPts;
const int fPtCount;
SkPoint fShortPts[4];
};
static CircleData circleDataSet[] = {
{ {{{313.0155029296875, 207.90290832519531}, {320.05078125, 227.58743286132812}}}, 2, {} },
{ {{{313.0155029296875, 207.90290832519531}, {313.98246891063195, 219.33615203830394},
{320.05078125, 227.58743286132812}}}, 3, {} },
};
static const int circleDataSetSize = (int) SK_ARRAY_COUNT(circleDataSet);
DEF_TEST(PathOpsAngleCircle, reporter) {
SkOpSegment segment[2];
for (int index = 0; index < circleDataSetSize; ++index) {
CircleData& data = circleDataSet[index];
for (int idx2 = 0; idx2 < data.fPtCount; ++idx2) {
data.fShortPts[idx2] = data.fPts.fPts[idx2].asSkPoint();
}
switch (data.fPtCount) {
case 2:
PathOpsSegmentTester::ConstructLine(&segment[index], data.fShortPts);
break;
case 3:
PathOpsSegmentTester::ConstructQuad(&segment[index], data.fShortPts);
break;
case 4:
PathOpsSegmentTester::ConstructCubic(&segment[index], data.fShortPts);
break;
}
}
PathOpsAngleTester::Orderable(segment[0].angle(0), segment[1].angle(0));
}
struct IntersectData {
const SkDCubic fPts;
const int fPtCount;
double fTStart;
double fTEnd;
SkPoint fShortPts[4];
};
static IntersectData intersectDataSet1[] = {
{ {{{322.935669,231.030273}, {312.832214,220.393295}, {312.832214,203.454178}}}, 3,
0.865309956, 0.154740299, {} },
{ {{{322.12738,233.397751}, {295.718353,159.505829}}}, 2,
0.345028807, 0.0786326511, {} },
{ {{{322.935669,231.030273}, {312.832214,220.393295}, {312.832214,203.454178}}}, 3,
0.865309956, 1, {} },
{ {{{322.12738,233.397751}, {295.718353,159.505829}}}, 2,
0.345028807, 1, {} },
};
static IntersectData intersectDataSet2[] = {
{ {{{364.390686,157.898193}, {375.281769,136.674606}, {396.039917,136.674606}}}, 3,
0.578520747, 1, {} },
{ {{{364.390686,157.898193}, {375.281769,136.674606}, {396.039917,136.674606}}}, 3,
0.578520747, 0.536512973, {} },
{ {{{366.608826,151.196014}, {378.803101,136.674606}, {398.164948,136.674606}}}, 3,
0.490456543, 1, {} },
};
static IntersectData intersectDataSet3[] = {
{ {{{2.000000,0.000000}, {1.33333333,0.66666667}}}, 2, 1, 0, {} },
{ {{{1.33333333,0.66666667}, {0.000000,2.000000}}}, 2, 0, 0.25, {} },
{ {{{2.000000,2.000000}, {1.33333333,0.66666667}}}, 2, 1, 0, {} },
};
static IntersectData intersectDataSet4[] = {
{ {{{1.3333333,0.6666667}, {0.000,2.000}}}, 2, 0.250000006, 0, {} },
{ {{{1.000,0.000}, {1.000,1.000}}}, 2, 1, 0, {} },
{ {{{1.000,1.000}, {0.000,0.000}}}, 2, 0, 1, {} },
};
static IntersectData intersectDataSet5[] = {
{ {{{0.000,0.000}, {1.000,0.000}, {1.000,1.000}}}, 3, 1, 0.666666667, {} },
{ {{{0.000,0.000}, {2.000,1.000}, {0.000,2.000}}}, 3, 0.5, 1, {} },
{ {{{0.000,0.000}, {2.000,1.000}, {0.000,2.000}}}, 3, 0.5, 0, {} },
};
static IntersectData intersectDataSet6[] = { // pathops_visualizer.htm:3658
{ {{{0.000,1.000}, {3.000,4.000}, {1.000,0.000}, {3.000,0.000}}}, 4, 0.0925339054, 0, {} }, // pathops_visualizer.htm:3616
{ {{{0.000,1.000}, {0.000,3.000}, {1.000,0.000}, {4.000,3.000}}}, 4, 0.453872386, 0, {} }, // pathops_visualizer.htm:3616
{ {{{0.000,1.000}, {3.000,4.000}, {1.000,0.000}, {3.000,0.000}}}, 4, 0.0925339054, 0.417096368, {} }, // pathops_visualizer.htm:3616
};
static IntersectData intersectDataSet7[] = { // pathops_visualizer.htm:3748
{ {{{2.000,1.000}, {0.000,1.000}}}, 2, 0.5, 0, {} }, // pathops_visualizer.htm:3706
{ {{{2.000,0.000}, {0.000,2.000}}}, 2, 0.5, 1, {} }, // pathops_visualizer.htm:3706
{ {{{0.000,1.000}, {0.000,2.000}, {2.000,0.000}, {2.000,1.000}}}, 4, 0.5, 1, {} }, // pathops_visualizer.htm:3706
}; //
static IntersectData intersectDataSet8[] = { // pathops_visualizer.htm:4194
{ {{{0.000,1.000}, {2.000,3.000}, {5.000,1.000}, {4.000,3.000}}}, 4, 0.311007457, 0.285714286, {} }, // pathops_visualizer.htm:4152
{ {{{1.000,5.000}, {3.000,4.000}, {1.000,0.000}, {3.000,2.000}}}, 4, 0.589885081, 0.999982974, {} }, // pathops_visualizer.htm:4152
{ {{{1.000,5.000}, {3.000,4.000}, {1.000,0.000}, {3.000,2.000}}}, 4, 0.589885081, 0.576935809, {} }, // pathops_visualizer.htm:4152
}; //
static IntersectData intersectDataSet9[] = { // pathops_visualizer.htm:4142
{ {{{0.000,1.000}, {2.000,3.000}, {5.000,1.000}, {4.000,3.000}}}, 4, 0.476627072, 0.311007457, {} }, // pathops_visualizer.htm:4100
{ {{{1.000,5.000}, {3.000,4.000}, {1.000,0.000}, {3.000,2.000}}}, 4, 0.999982974, 1, {} }, // pathops_visualizer.htm:4100
{ {{{0.000,1.000}, {2.000,3.000}, {5.000,1.000}, {4.000,3.000}}}, 4, 0.476627072, 1, {} }, // pathops_visualizer.htm:4100
}; //
static IntersectData intersectDataSet10[] = { // pathops_visualizer.htm:4186
{ {{{0.000,1.000}, {1.000,6.000}, {1.000,0.000}, {1.000,0.000}}}, 4, 0.788195121, 0.726275769, {} }, // pathops_visualizer.htm:4144
{ {{{0.000,1.000}, {0.000,1.000}, {1.000,0.000}, {6.000,1.000}}}, 4, 0.473378977, 1, {} }, // pathops_visualizer.htm:4144
{ {{{0.000,1.000}, {1.000,6.000}, {1.000,0.000}, {1.000,0.000}}}, 4, 0.788195121, 1, {} }, // pathops_visualizer.htm:4144
}; //
static IntersectData intersectDataSet11[] = { // pathops_visualizer.htm:4704
{ {{{979.305,561.000}, {1036.695,291.000}}}, 2, 0.888888874, 0.11111108, {} }, // pathops_visualizer.htm:4662
{ {{{1006.695,291.000}, {1023.264,291.000}, {1033.840,304.431}, {1030.318,321.000}}}, 4, 1, 0, {} }, // pathops_visualizer.htm:4662
{ {{{979.305,561.000}, {1036.695,291.000}}}, 2, 0.888888874, 1, {} }, // pathops_visualizer.htm:4662
}; //
static IntersectData intersectDataSet12[] = { // pathops_visualizer.htm:5481
{ {{{67.000,912.000}, {67.000,913.000}}}, 2, 1, 0, {} }, // pathops_visualizer.htm:5439
{ {{{67.000,913.000}, {67.000,917.389}, {67.224,921.726}, {67.662,926.000}}}, 4, 0, 1, {} }, // pathops_visualizer.htm:5439
{ {{{194.000,1041.000}, {123.860,1041.000}, {67.000,983.692}, {67.000,913.000}}}, 4, 1, 0, {} }, // pathops_visualizer.htm:5439
}; //
static IntersectData intersectDataSet13[] = { // pathops_visualizer.htm:5735
{ {{{6.000,0.000}, {0.000,4.000}}}, 2, 0.625, 0.25, {} }, // pathops_visualizer.htm:5693
{ {{{0.000,1.000}, {0.000,6.000}, {4.000,0.000}, {6.000,1.000}}}, 4, 0.5, 0.833333333, {} }, // pathops_visualizer.htm:5693
{ {{{0.000,1.000}, {0.000,6.000}, {4.000,0.000}, {6.000,1.000}}}, 4, 0.5, 0.379043969, {} }, // pathops_visualizer.htm:5693
}; //
static IntersectData intersectDataSet14[] = { // pathops_visualizer.htm:5875
{ {{{0.000,1.000}, {4.000,6.000}, {2.000,1.000}, {2.000,0.000}}}, 4, 0.0756502183, 0.0594570973, {} }, // pathops_visualizer.htm:5833
{ {{{1.000,2.000}, {0.000,2.000}, {1.000,0.000}, {6.000,4.000}}}, 4, 0.0756502184, 0, {} }, // pathops_visualizer.htm:5833
{ {{{0.000,1.000}, {4.000,6.000}, {2.000,1.000}, {2.000,0.000}}}, 4, 0.0756502183, 0.531917258, {} }, // pathops_visualizer.htm:5833
}; //
static IntersectData intersectDataSet15[] = { // pathops_visualizer.htm:6580
{ {{{490.435,879.407}, {405.593,909.436}}}, 2, 0.500554405, 1, {} }, // pathops_visualizer.htm:6538
{ {{{447.967,894.438}, {448.007,894.424}, {448.014,894.422}}}, 3, 0, 1, {} }, // pathops_visualizer.htm:6538
{ {{{490.435,879.407}, {405.593,909.436}}}, 2, 0.500554405, 0.500000273, {} }, // pathops_visualizer.htm:6538
Sanitizing source files in Housekeeper-Nightly git-svn-id: http://skia.googlecode.com/svn/trunk@14196 2bbb7eff-a529-9590-31e7-b0007b416f81
2014-04-15 03:04:18 +00:00
}; //
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 IntersectData intersectDataSet16[] = { // pathops_visualizer.htm:7419
{ {{{1.000,4.000}, {4.000,5.000}, {3.000,2.000}, {6.000,3.000}}}, 4, 0.5, 0, {} }, // pathops_visualizer.htm:7377
{ {{{2.000,3.000}, {3.000,6.000}, {4.000,1.000}, {5.000,4.000}}}, 4, 0.5, 0.112701665, {} }, // pathops_visualizer.htm:7377
{ {{{5.000,4.000}, {2.000,3.000}}}, 2, 0.5, 0, {} }, // pathops_visualizer.htm:7377
}; //
#define I(x) intersectDataSet##x
static IntersectData* intersectDataSets[] = {
I(1), I(2), I(3), I(4), I(5), I(6), I(7), I(8), I(9), I(10),
I(11), I(12), I(13), I(14), I(15), I(16),
};
#undef I
#define I(x) (int) SK_ARRAY_COUNT(intersectDataSet##x)
static const int intersectDataSetSizes[] = {
I(1), I(2), I(3), I(4), I(5), I(6), I(7), I(8), I(9), I(10),
I(11), I(12), I(13), I(14), I(15), I(16),
};
#undef I
static const int intersectDataSetsSize = (int) SK_ARRAY_COUNT(intersectDataSetSizes);
DEF_TEST(PathOpsAngleAfter, reporter) {
for (int index = intersectDataSetsSize - 1; index >= 0; --index) {
IntersectData* dataArray = intersectDataSets[index];
const int dataSize = intersectDataSetSizes[index];
SkOpSegment segment[3];
for (int index2 = 0; index2 < dataSize - 2; ++index2) {
for (int temp = 0; temp < (int) SK_ARRAY_COUNT(segment); ++temp) {
PathOpsSegmentTester::DebugReset(&segment[temp]);
}
for (int index3 = 0; index3 < (int) SK_ARRAY_COUNT(segment); ++index3) {
IntersectData& data = dataArray[index2 + index3];
SkPoint temp[4];
for (int idx2 = 0; idx2 < data.fPtCount; ++idx2) {
temp[idx2] = data.fPts.fPts[idx2].asSkPoint();
}
switch (data.fPtCount) {
case 2: {
SkDLine seg = SkDLine::SubDivide(temp, data.fTStart,
data.fTStart < data.fTEnd ? 1 : 0);
data.fShortPts[0] = seg[0].asSkPoint();
data.fShortPts[1] = seg[1].asSkPoint();
PathOpsSegmentTester::ConstructLine(&segment[index3], data.fShortPts);
} break;
case 3: {
SkDQuad seg = SkDQuad::SubDivide(temp, data.fTStart, data.fTEnd);
data.fShortPts[0] = seg[0].asSkPoint();
data.fShortPts[1] = seg[1].asSkPoint();
data.fShortPts[2] = seg[2].asSkPoint();
PathOpsSegmentTester::ConstructQuad(&segment[index3], data.fShortPts);
} break;
case 4: {
SkDCubic seg = SkDCubic::SubDivide(temp, data.fTStart, data.fTEnd);
data.fShortPts[0] = seg[0].asSkPoint();
data.fShortPts[1] = seg[1].asSkPoint();
data.fShortPts[2] = seg[2].asSkPoint();
data.fShortPts[3] = seg[3].asSkPoint();
PathOpsSegmentTester::ConstructCubic(&segment[index3], data.fShortPts);
} break;
}
}
SkOpAngle& angle1 = const_cast<SkOpAngle&>(segment[0].angle(0));
SkOpAngle& angle2 = const_cast<SkOpAngle&>(segment[1].angle(0));
SkOpAngle& angle3 = const_cast<SkOpAngle&>(segment[2].angle(0));
PathOpsAngleTester::SetNext(angle1, angle3);
// These data sets are seeded when the set itself fails, so likely the dataset does not
// match the expected result. The tests above return 1 when first added, but
// return 0 after the bug is fixed.
SkDEBUGCODE(int result =) PathOpsAngleTester::After(angle2, angle1);
SkASSERT(result == 0 || result == 1);
}
}
}
void SkOpSegment::debugConstruct() {
addStartSpan(1);
addEndSpan(1);
debugAddAngle(0, 1);
}
void SkOpSegment::debugAddAngle(int start, int end) {
SkASSERT(start != end);
SkOpAngle& angle = fAngles.push_back();
angle.set(this, start, end);
}
void SkOpSegment::debugConstructCubic(SkPoint shortQuad[4]) {
addCubic(shortQuad, false, false);
addT(NULL, shortQuad[0], 0);
addT(NULL, shortQuad[3], 1);
debugConstruct();
}
void SkOpSegment::debugConstructLine(SkPoint shortQuad[2]) {
addLine(shortQuad, false, false);
addT(NULL, shortQuad[0], 0);
addT(NULL, shortQuad[1], 1);
debugConstruct();
}
void SkOpSegment::debugConstructQuad(SkPoint shortQuad[3]) {
addQuad(shortQuad, false, false);
addT(NULL, shortQuad[0], 0);
addT(NULL, shortQuad[2], 1);
debugConstruct();
}
Reference in New Issue Copy Permalink