/* * Copyright 2017 Google Inc. * * Use of this source code is governed by a BSD-style license that can be * found in the LICENSE file. */ #include "SkGaussFilter.h" #include <cmath> #include <tuple> #include <vector> #include "Test.h" // one part in a million static constexpr double kEpsilon = 0.000001; static double careful_add(int n, double* gauss) { // Sum smallest to largest to retain precision. double sum = 0; for (int i = n - 1; i >= 1; i--) { sum += 2.0 * gauss[i]; } sum += gauss[0]; return sum; } DEF_TEST(SkGaussFilterCommon, r) { using Test = std::tuple<double, std::vector<double>>; auto golden_check = [&](const Test& test) { double sigma; std::vector<double> golden; std::tie(sigma, golden) = test; SkGaussFilter filter{sigma}; double result[SkGaussFilter::kGaussArrayMax]; int n = 0; for (auto d : filter) { result[n++] = d; } REPORTER_ASSERT(r, static_cast<size_t>(n) == golden.size()); double sum = careful_add(n, result); REPORTER_ASSERT(r, sum == 1.0); for (size_t i = 0; i < golden.size(); i++) { REPORTER_ASSERT(r, std::abs(golden[i] - result[i]) < kEpsilon); } }; // The following two sigmas account for about 85% of all sigmas used for masks. // Golden values generated using Mathematica. auto tests = { // GaussianMatrix[{{Automatic}, {.788675}}] Test{0.788675, {0.593605, 0.176225, 0.0269721}}, // GaussianMatrix[{{4}, {1.07735}}, Method -> "Bessel"] Test{1.07735, {0.429537, 0.214955, 0.059143, 0.0111337}}, }; for (auto& test : tests) { golden_check(test); } } DEF_TEST(SkGaussFilterSweep, r) { // The double just before 2.0. const double maxSigma = nextafter(2.0, 0.0); auto check = [&](double sigma) { SkGaussFilter filter{sigma}; double result[SkGaussFilter::kGaussArrayMax]; int n = 0; for (auto d : filter) { result[n++] = d; } REPORTER_ASSERT(r, n <= SkGaussFilter::kGaussArrayMax); double sum = careful_add(n, result); REPORTER_ASSERT(r, sum == 1.0); }; for (double sigma = 0.0; sigma < 2.0; sigma += 0.1) { check(sigma); } check(maxSigma); }