/*
* Copyright 2011 Google Inc.
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "SkMath.h"
#include "SkMatrix.h"
#include "SkMatrixUtils.h"
#include "SkRandom.h"
#include "Test.h"
static bool nearly_equal_scalar(SkScalar a, SkScalar b) {
const SkScalar tolerance = SK_Scalar1 / 200000;
return SkScalarAbs(a - b) <= tolerance;
}
static bool nearly_equal(const SkMatrix& a, const SkMatrix& b) {
for (int i = 0; i < 9; i++) {
if (!nearly_equal_scalar(a[i], b[i])) {
SkDebugf("not equal %g %g\n", (float)a[i], (float)b[i]);
return false;
}
}
return true;
}
static bool are_equal(skiatest::Reporter* reporter,
const SkMatrix& a,
const SkMatrix& b) {
bool equal = a == b;
bool cheapEqual = a.cheapEqualTo(b);
if (equal != cheapEqual) {
if (equal) {
bool foundZeroSignDiff = false;
for (int i = 0; i < 9; ++i) {
float aVal = a.get(i);
float bVal = b.get(i);
int aValI = *SkTCast<int*>(&aVal);
int bValI = *SkTCast<int*>(&bVal);
if (0 == aVal && 0 == bVal && aValI != bValI) {
foundZeroSignDiff = true;
} else {
REPORTER_ASSERT(reporter, aVal == bVal && aValI == aValI);
}
}
REPORTER_ASSERT(reporter, foundZeroSignDiff);
} else {
bool foundNaN = false;
for (int i = 0; i < 9; ++i) {
float aVal = a.get(i);
float bVal = b.get(i);
int aValI = *SkTCast<int*>(&aVal);
int bValI = *SkTCast<int*>(&bVal);
if (sk_float_isnan(aVal) && aValI == bValI) {
foundNaN = true;
} else {
REPORTER_ASSERT(reporter, aVal == bVal && aValI == bValI);
}
}
REPORTER_ASSERT(reporter, foundNaN);
}
}
return equal;
}
static bool is_identity(const SkMatrix& m) {
SkMatrix identity;
identity.reset();
return nearly_equal(m, identity);
}
static void test_matrix_recttorect(skiatest::Reporter* reporter) {
SkRect src, dst;
SkMatrix matrix;
src.set(0, 0, SK_Scalar1*10, SK_Scalar1*10);
dst = src;
matrix.setRectToRect(src, dst, SkMatrix::kFill_ScaleToFit);
REPORTER_ASSERT(reporter, SkMatrix::kIdentity_Mask == matrix.getType());
REPORTER_ASSERT(reporter, matrix.rectStaysRect());
dst.offset(SK_Scalar1, SK_Scalar1);
matrix.setRectToRect(src, dst, SkMatrix::kFill_ScaleToFit);
REPORTER_ASSERT(reporter, SkMatrix::kTranslate_Mask == matrix.getType());
REPORTER_ASSERT(reporter, matrix.rectStaysRect());
dst.fRight += SK_Scalar1;
matrix.setRectToRect(src, dst, SkMatrix::kFill_ScaleToFit);
REPORTER_ASSERT(reporter,
(SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask) == matrix.getType());
REPORTER_ASSERT(reporter, matrix.rectStaysRect());
dst = src;
dst.fRight = src.fRight * 2;
matrix.setRectToRect(src, dst, SkMatrix::kFill_ScaleToFit);
REPORTER_ASSERT(reporter, SkMatrix::kScale_Mask == matrix.getType());
REPORTER_ASSERT(reporter, matrix.rectStaysRect());
}
static void test_flatten(skiatest::Reporter* reporter, const SkMatrix& m) {
// add 100 in case we have a bug, I don't want to kill my stack in the test
static const size_t kBufferSize = SkMatrix::kMaxFlattenSize + 100;
char buffer[kBufferSize];
size_t size1 = m.writeToMemory(NULL);
size_t size2 = m.writeToMemory(buffer);
REPORTER_ASSERT(reporter, size1 == size2);
REPORTER_ASSERT(reporter, size1 <= SkMatrix::kMaxFlattenSize);
SkMatrix m2;
size_t size3 = m2.readFromMemory(buffer, kBufferSize);
REPORTER_ASSERT(reporter, size1 == size3);
REPORTER_ASSERT(reporter, are_equal(reporter, m, m2));
char buffer2[kBufferSize];
size3 = m2.writeToMemory(buffer2);
REPORTER_ASSERT(reporter, size1 == size3);
REPORTER_ASSERT(reporter, memcmp(buffer, buffer2, size1) == 0);
}
static void test_matrix_min_max_scale(skiatest::Reporter* reporter) {
SkScalar scales[2];
bool success;
SkMatrix identity;
identity.reset();
REPORTER_ASSERT(reporter, SK_Scalar1 == identity.getMinScale());
REPORTER_ASSERT(reporter, SK_Scalar1 == identity.getMaxScale());
success = identity.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success && SK_Scalar1 == scales[0] && SK_Scalar1 == scales[1]);
SkMatrix scale;
scale.setScale(SK_Scalar1 * 2, SK_Scalar1 * 4);
REPORTER_ASSERT(reporter, SK_Scalar1 * 2 == scale.getMinScale());
REPORTER_ASSERT(reporter, SK_Scalar1 * 4 == scale.getMaxScale());
success = scale.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success && SK_Scalar1 * 2 == scales[0] && SK_Scalar1 * 4 == scales[1]);
SkMatrix rot90Scale;
rot90Scale.setRotate(90 * SK_Scalar1);
rot90Scale.postScale(SK_Scalar1 / 4, SK_Scalar1 / 2);
REPORTER_ASSERT(reporter, SK_Scalar1 / 4 == rot90Scale.getMinScale());
REPORTER_ASSERT(reporter, SK_Scalar1 / 2 == rot90Scale.getMaxScale());
success = rot90Scale.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success && SK_Scalar1 / 4 == scales[0] && SK_Scalar1 / 2 == scales[1]);
SkMatrix rotate;
rotate.setRotate(128 * SK_Scalar1);
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(SK_Scalar1, rotate.getMinScale(), SK_ScalarNearlyZero));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(SK_Scalar1, rotate.getMaxScale(), SK_ScalarNearlyZero));
success = rotate.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success);
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(SK_Scalar1, scales[0], SK_ScalarNearlyZero));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(SK_Scalar1, scales[1], SK_ScalarNearlyZero));
SkMatrix translate;
translate.setTranslate(10 * SK_Scalar1, -5 * SK_Scalar1);
REPORTER_ASSERT(reporter, SK_Scalar1 == translate.getMinScale());
REPORTER_ASSERT(reporter, SK_Scalar1 == translate.getMaxScale());
success = translate.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success && SK_Scalar1 == scales[0] && SK_Scalar1 == scales[1]);
SkMatrix perspX;
perspX.reset();
perspX.setPerspX(SkScalarToPersp(SK_Scalar1 / 1000));
REPORTER_ASSERT(reporter, -SK_Scalar1 == perspX.getMinScale());
REPORTER_ASSERT(reporter, -SK_Scalar1 == perspX.getMaxScale());
// Verify that getMinMaxScales() doesn't update the scales array on failure.
scales[0] = -5;
scales[1] = -5;
success = perspX.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, !success && -5 * SK_Scalar1 == scales[0] && -5 * SK_Scalar1 == scales[1]);
SkMatrix perspY;
perspY.reset();
perspY.setPerspY(SkScalarToPersp(-SK_Scalar1 / 500));
REPORTER_ASSERT(reporter, -SK_Scalar1 == perspY.getMinScale());
REPORTER_ASSERT(reporter, -SK_Scalar1 == perspY.getMaxScale());
scales[0] = -5;
scales[1] = -5;
success = perspY.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, !success && -5 * SK_Scalar1 == scales[0] && -5 * SK_Scalar1 == scales[1]);
SkMatrix baseMats[] = {scale, rot90Scale, rotate,
translate, perspX, perspY};
SkMatrix mats[2*SK_ARRAY_COUNT(baseMats)];
for (size_t i = 0; i < SK_ARRAY_COUNT(baseMats); ++i) {
mats[i] = baseMats[i];
bool invertable = mats[i].invert(&mats[i + SK_ARRAY_COUNT(baseMats)]);
REPORTER_ASSERT(reporter, invertable);
}
SkRandom rand;
for (int m = 0; m < 1000; ++m) {
SkMatrix mat;
mat.reset();
for (int i = 0; i < 4; ++i) {
int x = rand.nextU() % SK_ARRAY_COUNT(mats);
mat.postConcat(mats[x]);
}
SkScalar minScale = mat.getMinScale();
SkScalar maxScale = mat.getMaxScale();
REPORTER_ASSERT(reporter, (minScale < 0) == (maxScale < 0));
REPORTER_ASSERT(reporter, (maxScale < 0) == mat.hasPerspective());
SkScalar scales[2];
bool success = mat.getMinMaxScales(scales);
REPORTER_ASSERT(reporter, success == !mat.hasPerspective());
REPORTER_ASSERT(reporter, !success || (scales[0] == minScale && scales[1] == maxScale));
if (mat.hasPerspective()) {
m -= 1; // try another non-persp matrix
continue;
}
// test a bunch of vectors. All should be scaled by between minScale and maxScale
// (modulo some error) and we should find a vector that is scaled by almost each.
static const SkScalar gVectorScaleTol = (105 * SK_Scalar1) / 100;
static const SkScalar gCloseScaleTol = (97 * SK_Scalar1) / 100;
SkScalar max = 0, min = SK_ScalarMax;
SkVector vectors[1000];
for (size_t i = 0; i < SK_ARRAY_COUNT(vectors); ++i) {
vectors[i].fX = rand.nextSScalar1();
vectors[i].fY = rand.nextSScalar1();
if (!vectors[i].normalize()) {
i -= 1;
continue;
}
}
mat.mapVectors(vectors, SK_ARRAY_COUNT(vectors));
for (size_t i = 0; i < SK_ARRAY_COUNT(vectors); ++i) {
SkScalar d = vectors[i].length();
REPORTER_ASSERT(reporter, SkScalarDiv(d, maxScale) < gVectorScaleTol);
REPORTER_ASSERT(reporter, SkScalarDiv(minScale, d) < gVectorScaleTol);
if (max < d) {
max = d;
}
if (min > d) {
min = d;
}
}
REPORTER_ASSERT(reporter, SkScalarDiv(max, maxScale) >= gCloseScaleTol);
REPORTER_ASSERT(reporter, SkScalarDiv(minScale, min) >= gCloseScaleTol);
}
}
static void test_matrix_is_similarity(skiatest::Reporter* reporter) {
SkMatrix mat;
// identity
mat.setIdentity();
REPORTER_ASSERT(reporter, mat.isSimilarity());
// translation only
mat.reset();
mat.setTranslate(SkIntToScalar(100), SkIntToScalar(100));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// scale with same size
mat.reset();
mat.setScale(SkIntToScalar(15), SkIntToScalar(15));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// scale with one negative
mat.reset();
mat.setScale(SkIntToScalar(-15), SkIntToScalar(15));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// scale with different size
mat.reset();
mat.setScale(SkIntToScalar(15), SkIntToScalar(20));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// scale with same size at a pivot point
mat.reset();
mat.setScale(SkIntToScalar(15), SkIntToScalar(15),
SkIntToScalar(2), SkIntToScalar(2));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// scale with different size at a pivot point
mat.reset();
mat.setScale(SkIntToScalar(15), SkIntToScalar(20),
SkIntToScalar(2), SkIntToScalar(2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// skew with same size
mat.reset();
mat.setSkew(SkIntToScalar(15), SkIntToScalar(15));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// skew with different size
mat.reset();
mat.setSkew(SkIntToScalar(15), SkIntToScalar(20));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// skew with same size at a pivot point
mat.reset();
mat.setSkew(SkIntToScalar(15), SkIntToScalar(15),
SkIntToScalar(2), SkIntToScalar(2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// skew with different size at a pivot point
mat.reset();
mat.setSkew(SkIntToScalar(15), SkIntToScalar(20),
SkIntToScalar(2), SkIntToScalar(2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// perspective x
mat.reset();
mat.setPerspX(SkScalarToPersp(SK_Scalar1 / 2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// perspective y
mat.reset();
mat.setPerspY(SkScalarToPersp(SK_Scalar1 / 2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// rotate
for (int angle = 0; angle < 360; ++angle) {
mat.reset();
mat.setRotate(SkIntToScalar(angle));
#ifndef SK_CPU_ARM64
REPORTER_ASSERT(reporter, mat.isSimilarity());
#else
// 64-bit ARM devices built with -O2 and -ffp-contract=fast have a loss
// of precision and require that we have a higher tolerance
REPORTER_ASSERT(reporter, mat.isSimilarity(SK_ScalarNearlyZero + 0.00010113f));
#endif
}
// see if there are any accumulated precision issues
mat.reset();
for (int i = 1; i < 360; i++) {
mat.postRotate(SkIntToScalar(1));
}
REPORTER_ASSERT(reporter, mat.isSimilarity());
// rotate + translate
mat.reset();
mat.setRotate(SkIntToScalar(30));
mat.postTranslate(SkIntToScalar(10), SkIntToScalar(20));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// rotate + uniform scale
mat.reset();
mat.setRotate(SkIntToScalar(30));
mat.postScale(SkIntToScalar(2), SkIntToScalar(2));
REPORTER_ASSERT(reporter, mat.isSimilarity());
// rotate + non-uniform scale
mat.reset();
mat.setRotate(SkIntToScalar(30));
mat.postScale(SkIntToScalar(3), SkIntToScalar(2));
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// all zero
mat.setAll(0, 0, 0, 0, 0, 0, 0, 0, 0);
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// all zero except perspective
mat.setAll(0, 0, 0, 0, 0, 0, 0, 0, SK_Scalar1);
REPORTER_ASSERT(reporter, !mat.isSimilarity());
// scales zero, only skews
mat.setAll(0, SK_Scalar1, 0,
SK_Scalar1, 0, 0,
0, 0, SkMatrix::I()[8]);
REPORTER_ASSERT(reporter, mat.isSimilarity());
}
// For test_matrix_decomposition, below.
static bool scalar_nearly_equal_relative(SkScalar a, SkScalar b,
SkScalar tolerance = SK_ScalarNearlyZero) {
// from Bruce Dawson
// absolute check
SkScalar diff = SkScalarAbs(a - b);
if (diff < tolerance) {
return true;
}
// relative check
a = SkScalarAbs(a);
b = SkScalarAbs(b);
SkScalar largest = (b > a) ? b : a;
if (diff <= largest*tolerance) {
return true;
}
return false;
}
static bool check_matrix_recomposition(const SkMatrix& mat,
const SkPoint& rotation1,
const SkPoint& scale,
const SkPoint& rotation2) {
SkScalar c1 = rotation1.fX;
SkScalar s1 = rotation1.fY;
SkScalar scaleX = scale.fX;
SkScalar scaleY = scale.fY;
SkScalar c2 = rotation2.fX;
SkScalar s2 = rotation2.fY;
// We do a relative check here because large scale factors cause problems with an absolute check
bool result = scalar_nearly_equal_relative(mat[SkMatrix::kMScaleX],
scaleX*c1*c2 - scaleY*s1*s2) &&
scalar_nearly_equal_relative(mat[SkMatrix::kMSkewX],
-scaleX*s1*c2 - scaleY*c1*s2) &&
scalar_nearly_equal_relative(mat[SkMatrix::kMSkewY],
scaleX*c1*s2 + scaleY*s1*c2) &&
scalar_nearly_equal_relative(mat[SkMatrix::kMScaleY],
-scaleX*s1*s2 + scaleY*c1*c2);
return result;
}
static void test_matrix_decomposition(skiatest::Reporter* reporter) {
SkMatrix mat;
SkPoint rotation1, scale, rotation2;
const float kRotation0 = 15.5f;
const float kRotation1 = -50.f;
const float kScale0 = 5000.f;
const float kScale1 = 0.001f;
// identity
mat.reset();
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// make sure it doesn't crash if we pass in NULLs
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, NULL, NULL, NULL));
// rotation only
mat.setRotate(kRotation0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// uniform scale only
mat.setScale(kScale0, kScale0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// anisotropic scale only
mat.setScale(kScale1, kScale0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation then uniform scale
mat.setRotate(kRotation1);
mat.postScale(kScale0, kScale0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// uniform scale then rotation
mat.setScale(kScale0, kScale0);
mat.postRotate(kRotation1);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation then uniform scale+reflection
mat.setRotate(kRotation0);
mat.postScale(kScale1, -kScale1);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// uniform scale+reflection, then rotate
mat.setScale(kScale0, -kScale0);
mat.postRotate(kRotation1);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation then anisotropic scale
mat.setRotate(kRotation1);
mat.postScale(kScale1, kScale0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation then anisotropic scale
mat.setRotate(90);
mat.postScale(kScale1, kScale0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// anisotropic scale then rotation
mat.setScale(kScale1, kScale0);
mat.postRotate(kRotation0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// anisotropic scale then rotation
mat.setScale(kScale1, kScale0);
mat.postRotate(90);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation, uniform scale, then different rotation
mat.setRotate(kRotation1);
mat.postScale(kScale0, kScale0);
mat.postRotate(kRotation0);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation, anisotropic scale, then different rotation
mat.setRotate(kRotation0);
mat.postScale(kScale1, kScale0);
mat.postRotate(kRotation1);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// rotation, anisotropic scale + reflection, then different rotation
mat.setRotate(kRotation0);
mat.postScale(-kScale1, kScale0);
mat.postRotate(kRotation1);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// try some random matrices
SkRandom rand;
for (int m = 0; m < 1000; ++m) {
SkScalar rot0 = rand.nextRangeF(-180, 180);
SkScalar sx = rand.nextRangeF(-3000.f, 3000.f);
SkScalar sy = rand.nextRangeF(-3000.f, 3000.f);
SkScalar rot1 = rand.nextRangeF(-180, 180);
mat.setRotate(rot0);
mat.postScale(sx, sy);
mat.postRotate(rot1);
if (SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2)) {
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
} else {
// if the matrix is degenerate, the basis vectors should be near-parallel or near-zero
SkScalar perpdot = mat[SkMatrix::kMScaleX]*mat[SkMatrix::kMScaleY] -
mat[SkMatrix::kMSkewX]*mat[SkMatrix::kMSkewY];
REPORTER_ASSERT(reporter, SkScalarNearlyZero(perpdot));
}
}
// translation shouldn't affect this
mat.postTranslate(-1000.f, 1000.f);
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// perspective shouldn't affect this
mat[SkMatrix::kMPersp0] = 12.f;
mat[SkMatrix::kMPersp1] = 4.f;
mat[SkMatrix::kMPersp2] = 1872.f;
REPORTER_ASSERT(reporter, SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
REPORTER_ASSERT(reporter, check_matrix_recomposition(mat, rotation1, scale, rotation2));
// degenerate matrices
// mostly zero entries
mat.reset();
mat[SkMatrix::kMScaleX] = 0.f;
REPORTER_ASSERT(reporter, !SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
mat.reset();
mat[SkMatrix::kMScaleY] = 0.f;
REPORTER_ASSERT(reporter, !SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
mat.reset();
// linearly dependent entries
mat[SkMatrix::kMScaleX] = 1.f;
mat[SkMatrix::kMSkewX] = 2.f;
mat[SkMatrix::kMSkewY] = 4.f;
mat[SkMatrix::kMScaleY] = 8.f;
REPORTER_ASSERT(reporter, !SkDecomposeUpper2x2(mat, &rotation1, &scale, &rotation2));
}
// For test_matrix_homogeneous, below.
static bool scalar_array_nearly_equal_relative(const SkScalar a[], const SkScalar b[], int count) {
for (int i = 0; i < count; ++i) {
if (!scalar_nearly_equal_relative(a[i], b[i])) {
return false;
}
}
return true;
}
// For test_matrix_homogeneous, below.
// Maps a single triple in src using m and compares results to those in dst
static bool naive_homogeneous_mapping(const SkMatrix& m, const SkScalar src[3],
const SkScalar dst[3]) {
SkScalar res[3];
SkScalar ms[9] = {m[0], m[1], m[2],
m[3], m[4], m[5],
m[6], m[7], m[8]};
res[0] = src[0] * ms[0] + src[1] * ms[1] + src[2] * ms[2];
res[1] = src[0] * ms[3] + src[1] * ms[4] + src[2] * ms[5];
res[2] = src[0] * ms[6] + src[1] * ms[7] + src[2] * ms[8];
return scalar_array_nearly_equal_relative(res, dst, 3);
}
static void test_matrix_homogeneous(skiatest::Reporter* reporter) {
SkMatrix mat;
const float kRotation0 = 15.5f;
const float kRotation1 = -50.f;
const float kScale0 = 5000.f;
const int kTripleCount = 1000;
const int kMatrixCount = 1000;
SkRandom rand;
SkScalar randTriples[3*kTripleCount];
for (int i = 0; i < 3*kTripleCount; ++i) {
randTriples[i] = rand.nextRangeF(-3000.f, 3000.f);
}
SkMatrix mats[kMatrixCount];
for (int i = 0; i < kMatrixCount; ++i) {
for (int j = 0; j < 9; ++j) {
mats[i].set(j, rand.nextRangeF(-3000.f, 3000.f));
}
}
// identity
{
mat.reset();
SkScalar dst[3*kTripleCount];
mat.mapHomogeneousPoints(dst, randTriples, kTripleCount);
REPORTER_ASSERT(reporter, scalar_array_nearly_equal_relative(randTriples, dst, kTripleCount*3));
}
// zero matrix
{
mat.setAll(0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f);
SkScalar dst[3*kTripleCount];
mat.mapHomogeneousPoints(dst, randTriples, kTripleCount);
SkScalar zeros[3] = {0.f, 0.f, 0.f};
for (int i = 0; i < kTripleCount; ++i) {
REPORTER_ASSERT(reporter, scalar_array_nearly_equal_relative(&dst[i*3], zeros, 3));
}
}
// zero point
{
SkScalar zeros[3] = {0.f, 0.f, 0.f};
for (int i = 0; i < kMatrixCount; ++i) {
SkScalar dst[3];
mats[i].mapHomogeneousPoints(dst, zeros, 1);
REPORTER_ASSERT(reporter, scalar_array_nearly_equal_relative(dst, zeros, 3));
}
}
// doesn't crash with null dst, src, count == 0
{
mats[0].mapHomogeneousPoints(NULL, NULL, 0);
}
// uniform scale of point
{
mat.setScale(kScale0, kScale0);
SkScalar dst[3];
SkScalar src[3] = {randTriples[0], randTriples[1], 1.f};
SkPoint pnt;
pnt.set(src[0], src[1]);
mat.mapHomogeneousPoints(dst, src, 1);
mat.mapPoints(&pnt, &pnt, 1);
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[0], pnt.fX));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[1], pnt.fY));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[2], SK_Scalar1));
}
// rotation of point
{
mat.setRotate(kRotation0);
SkScalar dst[3];
SkScalar src[3] = {randTriples[0], randTriples[1], 1.f};
SkPoint pnt;
pnt.set(src[0], src[1]);
mat.mapHomogeneousPoints(dst, src, 1);
mat.mapPoints(&pnt, &pnt, 1);
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[0], pnt.fX));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[1], pnt.fY));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[2], SK_Scalar1));
}
// rotation, scale, rotation of point
{
mat.setRotate(kRotation1);
mat.postScale(kScale0, kScale0);
mat.postRotate(kRotation0);
SkScalar dst[3];
SkScalar src[3] = {randTriples[0], randTriples[1], 1.f};
SkPoint pnt;
pnt.set(src[0], src[1]);
mat.mapHomogeneousPoints(dst, src, 1);
mat.mapPoints(&pnt, &pnt, 1);
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[0], pnt.fX));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[1], pnt.fY));
REPORTER_ASSERT(reporter, SkScalarNearlyEqual(dst[2], SK_Scalar1));
}
// compare with naive approach
{
for (int i = 0; i < kMatrixCount; ++i) {
for (int j = 0; j < kTripleCount; ++j) {
SkScalar dst[3];
mats[i].mapHomogeneousPoints(dst, &randTriples[j*3], 1);
REPORTER_ASSERT(reporter, naive_homogeneous_mapping(mats[i], &randTriples[j*3], dst));
}
}
}
}
DEF_TEST(Matrix, reporter) {
SkMatrix mat, inverse, iden1, iden2;
mat.reset();
mat.setTranslate(SK_Scalar1, SK_Scalar1);
REPORTER_ASSERT(reporter, mat.invert(&inverse));
iden1.setConcat(mat, inverse);
REPORTER_ASSERT(reporter, is_identity(iden1));
mat.setScale(SkIntToScalar(2), SkIntToScalar(4));
REPORTER_ASSERT(reporter, mat.invert(&inverse));
iden1.setConcat(mat, inverse);
REPORTER_ASSERT(reporter, is_identity(iden1));
test_flatten(reporter, mat);
mat.setScale(SK_Scalar1/2, SkIntToScalar(2));
REPORTER_ASSERT(reporter, mat.invert(&inverse));
iden1.setConcat(mat, inverse);
REPORTER_ASSERT(reporter, is_identity(iden1));
test_flatten(reporter, mat);
mat.setScale(SkIntToScalar(3), SkIntToScalar(5), SkIntToScalar(20), 0);
mat.postRotate(SkIntToScalar(25));
REPORTER_ASSERT(reporter, mat.invert(NULL));
REPORTER_ASSERT(reporter, mat.invert(&inverse));
iden1.setConcat(mat, inverse);
REPORTER_ASSERT(reporter, is_identity(iden1));
iden2.setConcat(inverse, mat);
REPORTER_ASSERT(reporter, is_identity(iden2));
test_flatten(reporter, mat);
test_flatten(reporter, iden2);
mat.setScale(0, SK_Scalar1);
REPORTER_ASSERT(reporter, !mat.invert(NULL));
REPORTER_ASSERT(reporter, !mat.invert(&inverse));
mat.setScale(SK_Scalar1, 0);
REPORTER_ASSERT(reporter, !mat.invert(NULL));
REPORTER_ASSERT(reporter, !mat.invert(&inverse));
// rectStaysRect test
{
static const struct {
SkScalar m00, m01, m10, m11;
bool mStaysRect;
}
gRectStaysRectSamples[] = {
{ 0, 0, 0, 0, false },
{ 0, 0, 0, SK_Scalar1, false },
{ 0, 0, SK_Scalar1, 0, false },
{ 0, 0, SK_Scalar1, SK_Scalar1, false },
{ 0, SK_Scalar1, 0, 0, false },
{ 0, SK_Scalar1, 0, SK_Scalar1, false },
{ 0, SK_Scalar1, SK_Scalar1, 0, true },
{ 0, SK_Scalar1, SK_Scalar1, SK_Scalar1, false },
{ SK_Scalar1, 0, 0, 0, false },
{ SK_Scalar1, 0, 0, SK_Scalar1, true },
{ SK_Scalar1, 0, SK_Scalar1, 0, false },
{ SK_Scalar1, 0, SK_Scalar1, SK_Scalar1, false },
{ SK_Scalar1, SK_Scalar1, 0, 0, false },
{ SK_Scalar1, SK_Scalar1, 0, SK_Scalar1, false },
{ SK_Scalar1, SK_Scalar1, SK_Scalar1, 0, false },
{ SK_Scalar1, SK_Scalar1, SK_Scalar1, SK_Scalar1, false }
};
for (size_t i = 0; i < SK_ARRAY_COUNT(gRectStaysRectSamples); i++) {
SkMatrix m;
m.reset();
m.set(SkMatrix::kMScaleX, gRectStaysRectSamples[i].m00);
m.set(SkMatrix::kMSkewX, gRectStaysRectSamples[i].m01);
m.set(SkMatrix::kMSkewY, gRectStaysRectSamples[i].m10);
m.set(SkMatrix::kMScaleY, gRectStaysRectSamples[i].m11);
REPORTER_ASSERT(reporter,
m.rectStaysRect() == gRectStaysRectSamples[i].mStaysRect);
}
}
mat.reset();
mat.set(SkMatrix::kMScaleX, SkIntToScalar(1));
mat.set(SkMatrix::kMSkewX, SkIntToScalar(2));
mat.set(SkMatrix::kMTransX, SkIntToScalar(3));
mat.set(SkMatrix::kMSkewY, SkIntToScalar(4));
mat.set(SkMatrix::kMScaleY, SkIntToScalar(5));
mat.set(SkMatrix::kMTransY, SkIntToScalar(6));
SkScalar affine[6];
REPORTER_ASSERT(reporter, mat.asAffine(affine));
#define affineEqual(e) affine[SkMatrix::kA##e] == mat.get(SkMatrix::kM##e)
REPORTER_ASSERT(reporter, affineEqual(ScaleX));
REPORTER_ASSERT(reporter, affineEqual(SkewY));
REPORTER_ASSERT(reporter, affineEqual(SkewX));
REPORTER_ASSERT(reporter, affineEqual(ScaleY));
REPORTER_ASSERT(reporter, affineEqual(TransX));
REPORTER_ASSERT(reporter, affineEqual(TransY));
#undef affineEqual
mat.set(SkMatrix::kMPersp1, SkScalarToPersp(SK_Scalar1 / 2));
REPORTER_ASSERT(reporter, !mat.asAffine(affine));
SkMatrix mat2;
mat2.reset();
mat.reset();
SkScalar zero = 0;
mat.set(SkMatrix::kMSkewX, -zero);
REPORTER_ASSERT(reporter, are_equal(reporter, mat, mat2));
mat2.reset();
mat.reset();
mat.set(SkMatrix::kMSkewX, SK_ScalarNaN);
mat2.set(SkMatrix::kMSkewX, SK_ScalarNaN);
REPORTER_ASSERT(reporter, !are_equal(reporter, mat, mat2));
test_matrix_min_max_scale(reporter);
test_matrix_is_similarity(reporter);
test_matrix_recttorect(reporter);
test_matrix_decomposition(reporter);
test_matrix_homogeneous(reporter);
}
DEF_TEST(Matrix_Concat, r) {
SkMatrix a;
a.setTranslate(10, 20);
SkMatrix b;
b.setScale(3, 5);
SkMatrix expected;
expected.setConcat(a,b);
REPORTER_ASSERT(r, expected == SkMatrix::Concat(a, b));
}