/* * Copyright (C) 2013 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #define _GNU_SOURCE 1 #include // This include (and the associated definition of __test_capture_signbit) // must be placed before any files that include (gtest.h in this case). // // is required to define generic macros signbit, isfinite and // several other such functions. // // is required to undef declarations of these macros in the global // namespace and make equivalent functions available in namespace std. Our // stlport implementation does this only for signbit, isfinite, isinf and // isnan. // // NOTE: We don't write our test using std::signbit because we want to be // sure that we're testing the bionic version of signbit. The C++ libraries // are free to reimplement signbit or delegate to compiler builtins if they // please. namespace { template inline int test_capture_signbit(const T in) { return signbit(in); } template inline int test_capture_isfinite(const T in) { return isfinite(in); } template inline int test_capture_isnan(const T in) { return isnan(in); } template inline int test_capture_isinf(const T in) { return isinf(in); } } #include "math_data_test.h" #include #include #include #include #include #include float float_subnormal() { union { float f; uint32_t i; } u; u.i = 0x007fffff; return u.f; } double double_subnormal() { union { double d; uint64_t i; } u; u.i = 0x000fffffffffffffLL; return u.d; } long double ldouble_subnormal() { union { long double e; unsigned char c[sizeof(long double)]; } u; // Subnormals must have a zero exponent and non zero significand. // On all supported representation the 17 bit (counting from either sides) // is part of the significand so it should be enough to set that. // It also applies for the case sizeof(double) = sizeof(long double) for (unsigned int i = 0; i < sizeof(long double); i++) { u.c[i] = 0x00; } u.c[sizeof(long double) - 3] = 0x80; u.c[2] = 0x80; return u.e; } TEST(math, fpclassify) { ASSERT_EQ(FP_INFINITE, fpclassify(INFINITY)); ASSERT_EQ(FP_INFINITE, fpclassify(HUGE_VALF)); ASSERT_EQ(FP_INFINITE, fpclassify(HUGE_VAL)); ASSERT_EQ(FP_INFINITE, fpclassify(HUGE_VALL)); ASSERT_EQ(FP_NAN, fpclassify(nanf(""))); ASSERT_EQ(FP_NAN, fpclassify(nan(""))); ASSERT_EQ(FP_NAN, fpclassify(nanl(""))); ASSERT_EQ(FP_NORMAL, fpclassify(1.0f)); ASSERT_EQ(FP_NORMAL, fpclassify(1.0)); ASSERT_EQ(FP_NORMAL, fpclassify(1.0L)); ASSERT_EQ(FP_SUBNORMAL, fpclassify(float_subnormal())); ASSERT_EQ(FP_SUBNORMAL, fpclassify(double_subnormal())); ASSERT_EQ(FP_SUBNORMAL, fpclassify(ldouble_subnormal())); ASSERT_EQ(FP_ZERO, fpclassify(0.0f)); ASSERT_EQ(FP_ZERO, fpclassify(0.0)); ASSERT_EQ(FP_ZERO, fpclassify(0.0L)); } TEST(math, isfinite) { ASSERT_TRUE(test_capture_isfinite(123.0f)); ASSERT_TRUE(test_capture_isfinite(123.0)); ASSERT_TRUE(test_capture_isfinite(123.0L)); ASSERT_FALSE(test_capture_isfinite(HUGE_VALF)); ASSERT_FALSE(test_capture_isfinite(HUGE_VAL)); ASSERT_FALSE(test_capture_isfinite(HUGE_VALL)); } TEST(math, isinf) { ASSERT_FALSE(test_capture_isinf(123.0f)); ASSERT_FALSE(test_capture_isinf(123.0)); ASSERT_FALSE(test_capture_isinf(123.0L)); ASSERT_TRUE(test_capture_isinf(HUGE_VALF)); ASSERT_TRUE(test_capture_isinf(HUGE_VAL)); ASSERT_TRUE(test_capture_isinf(HUGE_VALL)); } TEST(math, isnan) { ASSERT_FALSE(test_capture_isnan(123.0f)); ASSERT_FALSE(test_capture_isnan(123.0)); ASSERT_FALSE(test_capture_isnan(123.0L)); ASSERT_TRUE(test_capture_isnan(nanf(""))); ASSERT_TRUE(test_capture_isnan(nan(""))); ASSERT_TRUE(test_capture_isnan(nanl(""))); } TEST(math, isnormal) { ASSERT_TRUE(isnormal(123.0f)); ASSERT_TRUE(isnormal(123.0)); ASSERT_TRUE(isnormal(123.0L)); ASSERT_FALSE(isnormal(float_subnormal())); ASSERT_FALSE(isnormal(double_subnormal())); ASSERT_FALSE(isnormal(ldouble_subnormal())); } // TODO: isgreater, isgreaterequals, isless, islessequal, islessgreater, isunordered TEST(math, signbit) { ASSERT_EQ(0, test_capture_signbit(0.0f)); ASSERT_EQ(0, test_capture_signbit(0.0)); ASSERT_EQ(0, test_capture_signbit(0.0L)); ASSERT_EQ(0, test_capture_signbit(1.0f)); ASSERT_EQ(0, test_capture_signbit(1.0)); ASSERT_EQ(0, test_capture_signbit(1.0L)); ASSERT_NE(0, test_capture_signbit(-1.0f)); ASSERT_NE(0, test_capture_signbit(-1.0)); ASSERT_NE(0, test_capture_signbit(-1.0L)); } TEST(math, __fpclassifyd) { #if defined(__GLIBC__) #define __fpclassifyd __fpclassify #endif ASSERT_EQ(FP_INFINITE, __fpclassifyd(HUGE_VAL)); ASSERT_EQ(FP_NAN, __fpclassifyd(nan(""))); ASSERT_EQ(FP_NORMAL, __fpclassifyd(1.0)); ASSERT_EQ(FP_SUBNORMAL, __fpclassifyd(double_subnormal())); ASSERT_EQ(FP_ZERO, __fpclassifyd(0.0)); } TEST(math, __fpclassifyf) { ASSERT_EQ(FP_INFINITE, __fpclassifyf(HUGE_VALF)); ASSERT_EQ(FP_NAN, __fpclassifyf(nanf(""))); ASSERT_EQ(FP_NORMAL, __fpclassifyf(1.0f)); ASSERT_EQ(FP_SUBNORMAL, __fpclassifyf(float_subnormal())); ASSERT_EQ(FP_ZERO, __fpclassifyf(0.0f)); } TEST(math, __fpclassifyl) { EXPECT_EQ(FP_INFINITE, __fpclassifyl(HUGE_VALL)); EXPECT_EQ(FP_NAN, __fpclassifyl(nanl(""))); EXPECT_EQ(FP_NORMAL, __fpclassifyl(1.0L)); EXPECT_EQ(FP_SUBNORMAL, __fpclassifyl(ldouble_subnormal())); EXPECT_EQ(FP_ZERO, __fpclassifyl(0.0L)); } TEST(math, finitef) { ASSERT_TRUE(finitef(123.0f)); ASSERT_FALSE(finitef(HUGE_VALF)); } TEST(math, __isfinite) { #if defined(__GLIBC__) #define __isfinite __finite #endif ASSERT_TRUE(__isfinite(123.0)); ASSERT_FALSE(__isfinite(HUGE_VAL)); } TEST(math, __isfinitef) { #if defined(__GLIBC__) #define __isfinitef __finitef #endif ASSERT_TRUE(__isfinitef(123.0f)); ASSERT_FALSE(__isfinitef(HUGE_VALF)); } TEST(math, __isfinitel) { #if defined(__GLIBC__) #define __isfinitel __finitel #endif ASSERT_TRUE(__isfinitel(123.0L)); ASSERT_FALSE(__isfinitel(HUGE_VALL)); } TEST(math, finite) { ASSERT_TRUE(finite(123.0)); ASSERT_FALSE(finite(HUGE_VAL)); } TEST(math, isinf_function) { // The isinf macro deals with all three types; the isinf function is for doubles. ASSERT_FALSE((isinf)(123.0)); ASSERT_TRUE((isinf)(HUGE_VAL)); } TEST(math, __isinff) { ASSERT_FALSE(__isinff(123.0f)); ASSERT_TRUE(__isinff(HUGE_VALF)); } TEST(math, __isinfl) { ASSERT_FALSE(__isinfl(123.0L)); ASSERT_TRUE(__isinfl(HUGE_VALL)); } TEST(math, isnan_function) { // The isnan macro deals with all three types; the isnan function is for doubles. ASSERT_FALSE((isnan)(123.0)); ASSERT_TRUE((isnan)(nan(""))); } TEST(math, __isnanf) { ASSERT_FALSE(__isnanf(123.0f)); ASSERT_TRUE(__isnanf(nanf(""))); } TEST(math, __isnanl) { ASSERT_FALSE(__isnanl(123.0L)); ASSERT_TRUE(__isnanl(nanl(""))); } TEST(math, isnanf) { ASSERT_FALSE(isnanf(123.0f)); ASSERT_TRUE(isnanf(nanf(""))); } TEST(math, __isnormal) { #if defined(__BIONIC__) ASSERT_TRUE(__isnormal(123.0)); ASSERT_FALSE(__isnormal(double_subnormal())); #else // __BIONIC__ GTEST_LOG_(INFO) << "glibc doesn't have __isnormal.\n"; #endif // __BIONIC__ } TEST(math, __isnormalf) { #if defined(__BIONIC__) ASSERT_TRUE(__isnormalf(123.0f)); ASSERT_FALSE(__isnormalf(float_subnormal())); #else // __BIONIC__ GTEST_LOG_(INFO) << "glibc doesn't have __isnormalf.\n"; #endif // __BIONIC__ } TEST(math, __isnormall) { #if defined(__BIONIC__) ASSERT_TRUE(__isnormall(123.0L)); ASSERT_FALSE(__isnormall(ldouble_subnormal())); #else // __BIONIC__ GTEST_LOG_(INFO) << "glibc doesn't have __isnormall.\n"; #endif // __BIONIC__ } TEST(math, __signbit) { ASSERT_EQ(0, __signbit(0.0)); ASSERT_EQ(0, __signbit(1.0)); ASSERT_NE(0, __signbit(-1.0)); } TEST(math, __signbitf) { ASSERT_EQ(0, __signbitf(0.0f)); ASSERT_EQ(0, __signbitf(1.0f)); ASSERT_NE(0, __signbitf(-1.0f)); } TEST(math, __signbitl) { ASSERT_EQ(0L, __signbitl(0.0L)); ASSERT_EQ(0L, __signbitl(1.0L)); ASSERT_NE(0L, __signbitl(-1.0L)); } TEST(math, acos) { ASSERT_DOUBLE_EQ(M_PI/2.0, acos(0.0)); } TEST(math, acosf) { ASSERT_FLOAT_EQ(static_cast(M_PI)/2.0f, acosf(0.0f)); } TEST(math, acosl) { ASSERT_DOUBLE_EQ(M_PI/2.0L, acosl(0.0L)); } TEST(math, asin) { ASSERT_DOUBLE_EQ(0.0, asin(0.0)); } TEST(math, asinf) { ASSERT_FLOAT_EQ(0.0f, asinf(0.0f)); } TEST(math, asinl) { ASSERT_DOUBLE_EQ(0.0L, asinl(0.0L)); } TEST(math, atan) { ASSERT_DOUBLE_EQ(0.0, atan(0.0)); } TEST(math, atanf) { ASSERT_FLOAT_EQ(0.0f, atanf(0.0f)); } TEST(math, atanl) { ASSERT_DOUBLE_EQ(0.0L, atanl(0.0L)); } TEST(math, atan2) { ASSERT_DOUBLE_EQ(0.0, atan2(0.0, 0.0)); } TEST(math, atan2f) { ASSERT_FLOAT_EQ(0.0f, atan2f(0.0f, 0.0f)); } TEST(math, atan2l) { ASSERT_DOUBLE_EQ(0.0L, atan2l(0.0L, 0.0L)); } TEST(math, cos) { ASSERT_DOUBLE_EQ(1.0, cos(0.0)); } TEST(math, cosf) { ASSERT_FLOAT_EQ(1.0f, cosf(0.0f)); } TEST(math, cosl) { ASSERT_DOUBLE_EQ(1.0L, cosl(0.0L)); } TEST(math, sin) { ASSERT_DOUBLE_EQ(0.0, sin(0.0)); } TEST(math, sinf) { ASSERT_FLOAT_EQ(0.0f, sinf(0.0f)); } TEST(math, sinl) { ASSERT_DOUBLE_EQ(0.0L, sinl(0.0L)); } TEST(math, sincos) { double s, c; sincos(0.0, &s, &c); ASSERT_DOUBLE_EQ(0.0, s); ASSERT_DOUBLE_EQ(1.0, c); } TEST(math, sincosf) { float s, c; sincosf(0.0f, &s, &c); ASSERT_FLOAT_EQ(0.0f, s); ASSERT_FLOAT_EQ(1.0f, c); } TEST(math, sincosl) { long double s, c; sincosl(0.0L, &s, &c); ASSERT_DOUBLE_EQ(0.0L, s); ASSERT_DOUBLE_EQ(1.0L, c); } TEST(math, tan) { ASSERT_DOUBLE_EQ(0.0, tan(0.0)); } TEST(math, tanf) { ASSERT_FLOAT_EQ(0.0f, tanf(0.0f)); } TEST(math, tanl) { ASSERT_DOUBLE_EQ(0.0L, tanl(0.0L)); } TEST(math, acosh) { ASSERT_DOUBLE_EQ(0.0, acosh(1.0)); } TEST(math, acoshf) { ASSERT_FLOAT_EQ(0.0f, acoshf(1.0f)); } TEST(math, acoshl) { ASSERT_DOUBLE_EQ(0.0L, acoshl(1.0L)); } TEST(math, asinh) { ASSERT_DOUBLE_EQ(0.0, asinh(0.0)); } TEST(math, asinhf) { ASSERT_FLOAT_EQ(0.0f, asinhf(0.0f)); } TEST(math, asinhl) { ASSERT_DOUBLE_EQ(0.0L, asinhl(0.0L)); } TEST(math, atanh) { ASSERT_DOUBLE_EQ(0.0, atanh(0.0)); } TEST(math, atanhf) { ASSERT_FLOAT_EQ(0.0f, atanhf(0.0f)); } TEST(math, atanhl) { ASSERT_DOUBLE_EQ(0.0L, atanhl(0.0L)); } TEST(math, cosh) { ASSERT_DOUBLE_EQ(1.0, cosh(0.0)); } TEST(math, coshf) { ASSERT_FLOAT_EQ(1.0f, coshf(0.0f)); } TEST(math, coshl) { ASSERT_DOUBLE_EQ(1.0L, coshl(0.0L)); } TEST(math, sinh) { ASSERT_DOUBLE_EQ(0.0, sinh(0.0)); } TEST(math, sinhf) { ASSERT_FLOAT_EQ(0.0f, sinhf(0.0f)); } TEST(math, sinhl) { ASSERT_DOUBLE_EQ(0.0L, sinhl(0.0L)); } TEST(math, tanh) { ASSERT_DOUBLE_EQ(0.0, tanh(0.0)); } TEST(math, tanhf) { ASSERT_FLOAT_EQ(0.0f, tanhf(0.0f)); } TEST(math, tanhl) { ASSERT_DOUBLE_EQ(0.0L, tanhl(0.0L)); } TEST(math, log) { ASSERT_DOUBLE_EQ(1.0, log(M_E)); } TEST(math, logf) { ASSERT_FLOAT_EQ(1.0f, logf(static_cast(M_E))); } TEST(math, logl) { ASSERT_DOUBLE_EQ(1.0L, logl(M_E)); } TEST(math, log2) { ASSERT_DOUBLE_EQ(12.0, log2(4096.0)); } TEST(math, log2f) { ASSERT_FLOAT_EQ(12.0f, log2f(4096.0f)); } TEST(math, log2l) { ASSERT_DOUBLE_EQ(12.0L, log2l(4096.0L)); } TEST(math, log10) { ASSERT_DOUBLE_EQ(3.0, log10(1000.0)); } TEST(math, log10f) { ASSERT_FLOAT_EQ(3.0f, log10f(1000.0f)); } TEST(math, log10l) { ASSERT_DOUBLE_EQ(3.0L, log10l(1000.0L)); } TEST(math, cbrt) { ASSERT_DOUBLE_EQ(3.0, cbrt(27.0)); } TEST(math, cbrtf) { ASSERT_FLOAT_EQ(3.0f, cbrtf(27.0f)); } TEST(math, cbrtl) { ASSERT_DOUBLE_EQ(3.0L, cbrtl(27.0L)); } TEST(math, sqrt) { ASSERT_DOUBLE_EQ(2.0, sqrt(4.0)); } TEST(math, sqrtf) { ASSERT_FLOAT_EQ(2.0f, sqrtf(4.0f)); } TEST(math, sqrtl) { ASSERT_DOUBLE_EQ(2.0L, sqrtl(4.0L)); } TEST(math, exp) { ASSERT_DOUBLE_EQ(1.0, exp(0.0)); ASSERT_DOUBLE_EQ(M_E, exp(1.0)); } TEST(math, expf) { ASSERT_FLOAT_EQ(1.0f, expf(0.0f)); ASSERT_FLOAT_EQ(static_cast(M_E), expf(1.0f)); } TEST(math, expl) { ASSERT_DOUBLE_EQ(1.0L, expl(0.0L)); ASSERT_DOUBLE_EQ(M_E, expl(1.0L)); } TEST(math, exp2) { ASSERT_DOUBLE_EQ(8.0, exp2(3.0)); } TEST(math, exp2f) { ASSERT_FLOAT_EQ(8.0f, exp2f(3.0f)); } TEST(math, exp2l) { ASSERT_DOUBLE_EQ(8.0L, exp2l(3.0L)); } TEST(math, expm1) { ASSERT_DOUBLE_EQ(M_E - 1.0, expm1(1.0)); } TEST(math, expm1f) { ASSERT_FLOAT_EQ(static_cast(M_E) - 1.0f, expm1f(1.0f)); } TEST(math, expm1l) { ASSERT_DOUBLE_EQ(M_E - 1.0L, expm1l(1.0L)); } TEST(math, pow) { ASSERT_TRUE(isnan(pow(nan(""), 3.0))); ASSERT_DOUBLE_EQ(1.0, (pow(1.0, nan("")))); ASSERT_TRUE(isnan(pow(2.0, nan("")))); ASSERT_DOUBLE_EQ(8.0, pow(2.0, 3.0)); } TEST(math, powf) { ASSERT_TRUE(isnanf(powf(nanf(""), 3.0f))); ASSERT_FLOAT_EQ(1.0f, (powf(1.0f, nanf("")))); ASSERT_TRUE(isnanf(powf(2.0f, nanf("")))); ASSERT_FLOAT_EQ(8.0f, powf(2.0f, 3.0f)); } TEST(math, powl) { ASSERT_TRUE(__isnanl(powl(nanl(""), 3.0L))); ASSERT_DOUBLE_EQ(1.0L, (powl(1.0L, nanl("")))); ASSERT_TRUE(__isnanl(powl(2.0L, nanl("")))); ASSERT_DOUBLE_EQ(8.0L, powl(2.0L, 3.0L)); } TEST(math, ceil) { ASSERT_DOUBLE_EQ(1.0, ceil(0.9)); } TEST(math, ceilf) { ASSERT_FLOAT_EQ(1.0f, ceilf(0.9f)); } TEST(math, ceill) { ASSERT_DOUBLE_EQ(1.0L, ceill(0.9L)); } TEST(math, floor) { ASSERT_DOUBLE_EQ(1.0, floor(1.1)); } TEST(math, floorf) { ASSERT_FLOAT_EQ(1.0f, floorf(1.1f)); } TEST(math, floorl) { ASSERT_DOUBLE_EQ(1.0L, floorl(1.1L)); } TEST(math, fabs) { ASSERT_DOUBLE_EQ(1.0, fabs(-1.0)); } TEST(math, fabsf) { ASSERT_FLOAT_EQ(1.0f, fabsf(-1.0f)); } TEST(math, fabsl) { ASSERT_DOUBLE_EQ(1.0L, fabsl(-1.0L)); } TEST(math, ldexp) { ASSERT_DOUBLE_EQ(16.0, ldexp(2.0, 3.0)); } TEST(math, ldexpf) { ASSERT_FLOAT_EQ(16.0f, ldexpf(2.0f, 3.0f)); } TEST(math, ldexpl) { ASSERT_DOUBLE_EQ(16.0L, ldexpl(2.0L, 3.0)); } TEST(math, fmod) { ASSERT_DOUBLE_EQ(2.0, fmod(12.0, 10.0)); } TEST(math, fmodf) { ASSERT_FLOAT_EQ(2.0f, fmodf(12.0f, 10.0f)); } TEST(math, fmodl) { ASSERT_DOUBLE_EQ(2.0L, fmodl(12.0L, 10.0L)); } TEST(math, remainder) { ASSERT_DOUBLE_EQ(2.0, remainder(12.0, 10.0)); } TEST(math, remainderf) { ASSERT_FLOAT_EQ(2.0f, remainderf(12.0f, 10.0f)); } TEST(math, remainderl) { ASSERT_DOUBLE_EQ(2.0L, remainderl(12.0L, 10.0L)); } TEST(math, drem) { ASSERT_DOUBLE_EQ(2.0, drem(12.0, 10.0)); } TEST(math, dremf) { ASSERT_FLOAT_EQ(2.0f, dremf(12.0f, 10.0f)); } TEST(math, fmax) { ASSERT_DOUBLE_EQ(12.0, fmax(12.0, 10.0)); ASSERT_DOUBLE_EQ(12.0, fmax(12.0, nan(""))); ASSERT_DOUBLE_EQ(12.0, fmax(nan(""), 12.0)); } TEST(math, fmaxf) { ASSERT_FLOAT_EQ(12.0f, fmaxf(12.0f, 10.0f)); ASSERT_FLOAT_EQ(12.0f, fmaxf(12.0f, nanf(""))); ASSERT_FLOAT_EQ(12.0f, fmaxf(nanf(""), 12.0f)); } TEST(math, fmaxl) { ASSERT_DOUBLE_EQ(12.0L, fmaxl(12.0L, 10.0L)); ASSERT_DOUBLE_EQ(12.0L, fmaxl(12.0L, nanl(""))); ASSERT_DOUBLE_EQ(12.0L, fmaxl(nanl(""), 12.0L)); } TEST(math, fmin) { ASSERT_DOUBLE_EQ(10.0, fmin(12.0, 10.0)); ASSERT_DOUBLE_EQ(12.0, fmin(12.0, nan(""))); ASSERT_DOUBLE_EQ(12.0, fmin(nan(""), 12.0)); } TEST(math, fminf) { ASSERT_FLOAT_EQ(10.0f, fminf(12.0f, 10.0f)); ASSERT_FLOAT_EQ(12.0f, fminf(12.0f, nanf(""))); ASSERT_FLOAT_EQ(12.0f, fminf(nanf(""), 12.0f)); } TEST(math, fminl) { ASSERT_DOUBLE_EQ(10.0L, fminl(12.0L, 10.0L)); ASSERT_DOUBLE_EQ(12.0L, fminl(12.0L, nanl(""))); ASSERT_DOUBLE_EQ(12.0L, fminl(nanl(""), 12.0L)); } TEST(math, fma) { ASSERT_DOUBLE_EQ(10.0, fma(2.0, 3.0, 4.0)); } TEST(math, fmaf) { ASSERT_FLOAT_EQ(10.0f, fmaf(2.0f, 3.0f, 4.0f)); } TEST(math, fmal) { ASSERT_DOUBLE_EQ(10.0L, fmal(2.0L, 3.0L, 4.0L)); } TEST(math, hypot) { ASSERT_DOUBLE_EQ(5.0, hypot(3.0, 4.0)); } TEST(math, hypotf) { ASSERT_FLOAT_EQ(5.0f, hypotf(3.0f, 4.0f)); } TEST(math, hypotl) { ASSERT_DOUBLE_EQ(5.0L, hypotl(3.0L, 4.0L)); } TEST(math, erf) { ASSERT_DOUBLE_EQ(0.84270079294971489, erf(1.0)); } TEST(math, erff) { ASSERT_FLOAT_EQ(0.84270078f, erff(1.0f)); } TEST(math, erfl) { ASSERT_DOUBLE_EQ(0.84270079294971489L, erfl(1.0L)); } TEST(math, erfc) { ASSERT_DOUBLE_EQ(0.15729920705028513, erfc(1.0)); } TEST(math, erfcf) { ASSERT_FLOAT_EQ(0.15729921f, erfcf(1.0f)); } TEST(math, erfcl) { ASSERT_DOUBLE_EQ(0.15729920705028513l, erfcl(1.0L)); } TEST(math, lrint) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // lrint/lrintf/lrintl obey the rounding mode. ASSERT_EQ(1235, lrint(1234.01)); ASSERT_EQ(1235, lrintf(1234.01f)); ASSERT_EQ(1235, lrintl(1234.01L)); fesetround(FE_TOWARDZERO); // lrint/lrintf/lrintl obey the rounding mode. ASSERT_EQ(1234, lrint(1234.01)); ASSERT_EQ(1234, lrintf(1234.01f)); ASSERT_EQ(1234, lrintl(1234.01L)); fesetround(FE_UPWARD); // llrint/llrintf/llrintl obey the rounding mode. ASSERT_EQ(1235L, llrint(1234.01)); ASSERT_EQ(1235L, llrintf(1234.01f)); ASSERT_EQ(1235L, llrintl(1234.01L)); fesetround(FE_TOWARDZERO); // llrint/llrintf/llrintl obey the rounding mode. ASSERT_EQ(1234L, llrint(1234.01)); ASSERT_EQ(1234L, llrintf(1234.01f)); ASSERT_EQ(1234L, llrintl(1234.01L)); } TEST(math, rint) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // rint/rintf/rintl obey the rounding mode. feclearexcept(FE_ALL_EXCEPT); // rint/rintf/rintl do set the FE_INEXACT flag. ASSERT_EQ(1234.0, rint(1234.0)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0, rint(1234.01)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) != 0); feclearexcept(FE_ALL_EXCEPT); // rint/rintf/rintl do set the FE_INEXACT flag. ASSERT_EQ(1234.0f, rintf(1234.0f)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0f, rintf(1234.01f)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) != 0); feclearexcept(FE_ALL_EXCEPT); // rint/rintf/rintl do set the FE_INEXACT flag. ASSERT_EQ(1234.0, rintl(1234.0L)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0, rintl(1234.01L)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) != 0); fesetround(FE_TOWARDZERO); // rint/rintf obey the rounding mode. ASSERT_EQ(1234.0, rint(1234.01)); ASSERT_EQ(1234.0f, rintf(1234.01f)); ASSERT_EQ(1234.0, rintl(1234.01L)); } TEST(math, nearbyint) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // nearbyint/nearbyintf/nearbyintl obey the rounding mode. feclearexcept(FE_ALL_EXCEPT); // nearbyint/nearbyintf/nearbyintl don't set the FE_INEXACT flag. ASSERT_EQ(1234.0, nearbyint(1234.0)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0, nearbyint(1234.01)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); feclearexcept(FE_ALL_EXCEPT); ASSERT_EQ(1234.0f, nearbyintf(1234.0f)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0f, nearbyintf(1234.01f)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); feclearexcept(FE_ALL_EXCEPT); // nearbyint/nearbyintf/nearbyintl don't set the FE_INEXACT flag. ASSERT_EQ(1234.0, nearbyintl(1234.0L)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); ASSERT_EQ(1235.0, nearbyintl(1234.01L)); ASSERT_TRUE((fetestexcept(FE_ALL_EXCEPT) & FE_INEXACT) == 0); fesetround(FE_TOWARDZERO); // nearbyint/nearbyintf/nearbyintl obey the rounding mode. ASSERT_EQ(1234.0, nearbyint(1234.01)); ASSERT_EQ(1234.0f, nearbyintf(1234.01f)); ASSERT_EQ(1234.0, nearbyintl(1234.01L)); } TEST(math, lround) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // lround ignores the rounding mode. ASSERT_EQ(1234, lround(1234.01)); ASSERT_EQ(1234, lroundf(1234.01f)); ASSERT_EQ(1234, lroundl(1234.01L)); } TEST(math, llround) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // llround ignores the rounding mode. ASSERT_EQ(1234L, llround(1234.01)); ASSERT_EQ(1234L, llroundf(1234.01f)); ASSERT_EQ(1234L, llroundl(1234.01L)); } TEST(math, ilogb) { ASSERT_EQ(FP_ILOGB0, ilogb(0.0)); ASSERT_EQ(FP_ILOGBNAN, ilogb(nan(""))); ASSERT_EQ(INT_MAX, ilogb(HUGE_VAL)); ASSERT_EQ(0, ilogb(1.0)); ASSERT_EQ(3, ilogb(10.0)); } TEST(math, ilogbf) { ASSERT_EQ(FP_ILOGB0, ilogbf(0.0f)); ASSERT_EQ(FP_ILOGBNAN, ilogbf(nanf(""))); ASSERT_EQ(INT_MAX, ilogbf(HUGE_VALF)); ASSERT_EQ(0, ilogbf(1.0f)); ASSERT_EQ(3, ilogbf(10.0f)); } TEST(math, ilogbl) { ASSERT_EQ(FP_ILOGB0, ilogbl(0.0L)); ASSERT_EQ(FP_ILOGBNAN, ilogbl(nanl(""))); ASSERT_EQ(INT_MAX, ilogbl(HUGE_VALL)); ASSERT_EQ(0L, ilogbl(1.0L)); ASSERT_EQ(3L, ilogbl(10.0L)); } TEST(math, logb) { ASSERT_EQ(-HUGE_VAL, logb(0.0)); ASSERT_TRUE(isnan(logb(nan("")))); ASSERT_TRUE(isinf(logb(HUGE_VAL))); ASSERT_EQ(0.0, logb(1.0)); ASSERT_EQ(3.0, logb(10.0)); } TEST(math, logbf) { ASSERT_EQ(-HUGE_VALF, logbf(0.0f)); ASSERT_TRUE(isnanf(logbf(nanf("")))); ASSERT_TRUE(__isinff(logbf(HUGE_VALF))); ASSERT_EQ(0.0f, logbf(1.0f)); ASSERT_EQ(3.0f, logbf(10.0f)); } TEST(math, logbl) { ASSERT_EQ(-HUGE_VAL, logbl(0.0L)); ASSERT_TRUE(isnan(logbl(nanl("")))); ASSERT_TRUE(isinf(logbl(HUGE_VALL))); ASSERT_EQ(0.0L, logbl(1.0L)); ASSERT_EQ(3.0L, logbl(10.0L)); } TEST(math, log1p) { ASSERT_EQ(-HUGE_VAL, log1p(-1.0)); ASSERT_TRUE(isnan(log1p(nan("")))); ASSERT_TRUE(isinf(log1p(HUGE_VAL))); ASSERT_DOUBLE_EQ(1.0, log1p(M_E - 1.0)); } TEST(math, log1pf) { ASSERT_EQ(-HUGE_VALF, log1pf(-1.0f)); ASSERT_TRUE(isnanf(log1pf(nanf("")))); ASSERT_TRUE(__isinff(log1pf(HUGE_VALF))); ASSERT_FLOAT_EQ(1.0f, log1pf(static_cast(M_E) - 1.0f)); } TEST(math, log1pl) { ASSERT_EQ(-HUGE_VALL, log1pl(-1.0L)); ASSERT_TRUE(isnan(log1pl(nanl("")))); ASSERT_TRUE(isinf(log1pl(HUGE_VALL))); ASSERT_DOUBLE_EQ(1.0L, log1pl(M_E - 1.0L)); } TEST(math, fdim) { ASSERT_DOUBLE_EQ(0.0, fdim(1.0, 1.0)); ASSERT_DOUBLE_EQ(1.0, fdim(2.0, 1.0)); ASSERT_DOUBLE_EQ(0.0, fdim(1.0, 2.0)); } TEST(math, fdimf) { ASSERT_FLOAT_EQ(0.0f, fdimf(1.0f, 1.0f)); ASSERT_FLOAT_EQ(1.0f, fdimf(2.0f, 1.0f)); ASSERT_FLOAT_EQ(0.0f, fdimf(1.0f, 2.0f)); } TEST(math, fdiml) { ASSERT_DOUBLE_EQ(0.0L, fdiml(1.0L, 1.0L)); ASSERT_DOUBLE_EQ(1.0L, fdiml(2.0L, 1.0L)); ASSERT_DOUBLE_EQ(0.0L, fdiml(1.0L, 2.0L)); } TEST(math, round) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_TOWARDZERO); // round ignores the rounding mode and always rounds away from zero. ASSERT_DOUBLE_EQ(1.0, round(0.5)); ASSERT_DOUBLE_EQ(-1.0, round(-0.5)); ASSERT_DOUBLE_EQ(0.0, round(0.0)); ASSERT_DOUBLE_EQ(-0.0, round(-0.0)); ASSERT_TRUE(isnan(round(nan("")))); ASSERT_DOUBLE_EQ(HUGE_VAL, round(HUGE_VAL)); } TEST(math, roundf) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_TOWARDZERO); // roundf ignores the rounding mode and always rounds away from zero. ASSERT_FLOAT_EQ(1.0f, roundf(0.5f)); ASSERT_FLOAT_EQ(-1.0f, roundf(-0.5f)); ASSERT_FLOAT_EQ(0.0f, roundf(0.0f)); ASSERT_FLOAT_EQ(-0.0f, roundf(-0.0f)); ASSERT_TRUE(isnanf(roundf(nanf("")))); ASSERT_FLOAT_EQ(HUGE_VALF, roundf(HUGE_VALF)); } TEST(math, roundl) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_TOWARDZERO); // roundl ignores the rounding mode and always rounds away from zero. ASSERT_DOUBLE_EQ(1.0L, roundl(0.5L)); ASSERT_DOUBLE_EQ(-1.0L, roundl(-0.5L)); ASSERT_DOUBLE_EQ(0.0L, roundl(0.0L)); ASSERT_DOUBLE_EQ(-0.0L, roundl(-0.0L)); ASSERT_TRUE(isnan(roundl(nanl("")))); ASSERT_DOUBLE_EQ(HUGE_VALL, roundl(HUGE_VALL)); } TEST(math, trunc) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // trunc ignores the rounding mode and always rounds toward zero. ASSERT_DOUBLE_EQ(1.0, trunc(1.5)); ASSERT_DOUBLE_EQ(-1.0, trunc(-1.5)); ASSERT_DOUBLE_EQ(0.0, trunc(0.0)); ASSERT_DOUBLE_EQ(-0.0, trunc(-0.0)); ASSERT_TRUE(isnan(trunc(nan("")))); ASSERT_DOUBLE_EQ(HUGE_VAL, trunc(HUGE_VAL)); } TEST(math, truncf) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // truncf ignores the rounding mode and always rounds toward zero. ASSERT_FLOAT_EQ(1.0f, truncf(1.5f)); ASSERT_FLOAT_EQ(-1.0f, truncf(-1.5f)); ASSERT_FLOAT_EQ(0.0f, truncf(0.0f)); ASSERT_FLOAT_EQ(-0.0f, truncf(-0.0f)); ASSERT_TRUE(isnan(truncf(nanf("")))); ASSERT_FLOAT_EQ(HUGE_VALF, truncf(HUGE_VALF)); } TEST(math, truncl) { auto guard = make_scope_guard([]() { fesetenv(FE_DFL_ENV); }); fesetround(FE_UPWARD); // truncl ignores the rounding mode and always rounds toward zero. ASSERT_DOUBLE_EQ(1.0L, truncl(1.5L)); ASSERT_DOUBLE_EQ(-1.0L, truncl(-1.5L)); ASSERT_DOUBLE_EQ(0.0L, truncl(0.0L)); ASSERT_DOUBLE_EQ(-0.0L, truncl(-0.0L)); ASSERT_TRUE(isnan(truncl(nan("")))); ASSERT_DOUBLE_EQ(HUGE_VALL, truncl(HUGE_VALL)); } TEST(math, nextafter) { ASSERT_DOUBLE_EQ(0.0, nextafter(0.0, 0.0)); ASSERT_DOUBLE_EQ(4.9406564584124654e-324, nextafter(0.0, 1.0)); ASSERT_DOUBLE_EQ(-4.9406564584124654e-324, nextafter(0.0, -1.0)); } TEST(math, nextafterf) { ASSERT_FLOAT_EQ(0.0f, nextafterf(0.0f, 0.0f)); ASSERT_FLOAT_EQ(1.4012985e-45f, nextafterf(0.0f, 1.0f)); ASSERT_FLOAT_EQ(-1.4012985e-45f, nextafterf(0.0f, -1.0f)); } TEST(math, nextafterl) { ASSERT_DOUBLE_EQ(0.0L, nextafterl(0.0L, 0.0L)); // Use a runtime value to accomodate the case when // sizeof(double) == sizeof(long double) long double smallest_positive = ldexpl(1.0L, LDBL_MIN_EXP - LDBL_MANT_DIG); ASSERT_DOUBLE_EQ(smallest_positive, nextafterl(0.0L, 1.0L)); ASSERT_DOUBLE_EQ(-smallest_positive, nextafterl(0.0L, -1.0L)); } TEST(math, nexttoward) { ASSERT_DOUBLE_EQ(0.0, nexttoward(0.0, 0.0L)); ASSERT_DOUBLE_EQ(4.9406564584124654e-324, nexttoward(0.0, 1.0L)); ASSERT_DOUBLE_EQ(-4.9406564584124654e-324, nexttoward(0.0, -1.0L)); } TEST(math, nexttowardf) { ASSERT_FLOAT_EQ(0.0f, nexttowardf(0.0f, 0.0L)); ASSERT_FLOAT_EQ(1.4012985e-45f, nexttowardf(0.0f, 1.0L)); ASSERT_FLOAT_EQ(-1.4012985e-45f, nexttowardf(0.0f, -1.0L)); } TEST(math, nexttowardl) { ASSERT_DOUBLE_EQ(0.0L, nexttowardl(0.0L, 0.0L)); // Use a runtime value to accomodate the case when // sizeof(double) == sizeof(long double) long double smallest_positive = ldexpl(1.0L, LDBL_MIN_EXP - LDBL_MANT_DIG); ASSERT_DOUBLE_EQ(smallest_positive, nexttowardl(0.0L, 1.0L)); ASSERT_DOUBLE_EQ(-smallest_positive, nexttowardl(0.0L, -1.0L)); } TEST(math, copysign) { ASSERT_DOUBLE_EQ(0.0, copysign(0.0, 1.0)); ASSERT_DOUBLE_EQ(-0.0, copysign(0.0, -1.0)); ASSERT_DOUBLE_EQ(2.0, copysign(2.0, 1.0)); ASSERT_DOUBLE_EQ(-2.0, copysign(2.0, -1.0)); } TEST(math, copysignf) { ASSERT_FLOAT_EQ(0.0f, copysignf(0.0f, 1.0f)); ASSERT_FLOAT_EQ(-0.0f, copysignf(0.0f, -1.0f)); ASSERT_FLOAT_EQ(2.0f, copysignf(2.0f, 1.0f)); ASSERT_FLOAT_EQ(-2.0f, copysignf(2.0f, -1.0f)); } TEST(math, copysignl) { ASSERT_DOUBLE_EQ(0.0L, copysignl(0.0L, 1.0L)); ASSERT_DOUBLE_EQ(-0.0L, copysignl(0.0L, -1.0L)); ASSERT_DOUBLE_EQ(2.0L, copysignl(2.0L, 1.0L)); ASSERT_DOUBLE_EQ(-2.0L, copysignl(2.0L, -1.0L)); } TEST(math, significand) { ASSERT_DOUBLE_EQ(0.0, significand(0.0)); ASSERT_DOUBLE_EQ(1.2, significand(1.2)); ASSERT_DOUBLE_EQ(1.53125, significand(12.25)); } TEST(math, significandf) { ASSERT_FLOAT_EQ(0.0f, significandf(0.0f)); ASSERT_FLOAT_EQ(1.2f, significandf(1.2f)); ASSERT_FLOAT_EQ(1.53125f, significandf(12.25f)); } TEST(math, significandl) { ASSERT_DOUBLE_EQ(0.0L, significandl(0.0L)); ASSERT_DOUBLE_EQ(1.2L, significandl(1.2L)); ASSERT_DOUBLE_EQ(1.53125L, significandl(12.25L)); } TEST(math, scalb) { ASSERT_DOUBLE_EQ(12.0, scalb(3.0, 2.0)); } TEST(math, scalbf) { ASSERT_FLOAT_EQ(12.0f, scalbf(3.0f, 2.0f)); } TEST(math, scalbln) { ASSERT_DOUBLE_EQ(12.0, scalbln(3.0, 2L)); } TEST(math, scalblnf) { ASSERT_FLOAT_EQ(12.0f, scalblnf(3.0f, 2L)); } TEST(math, scalblnl) { ASSERT_DOUBLE_EQ(12.0L, scalblnl(3.0L, 2L)); } TEST(math, scalbn) { ASSERT_DOUBLE_EQ(12.0, scalbn(3.0, 2)); } TEST(math, scalbnf) { ASSERT_FLOAT_EQ(12.0f, scalbnf(3.0f, 2)); } TEST(math, scalbnl) { ASSERT_DOUBLE_EQ(12.0L, scalbnl(3.0L, 2)); } TEST(math, gamma) { ASSERT_DOUBLE_EQ(log(24.0), gamma(5.0)); } TEST(math, gammaf) { ASSERT_FLOAT_EQ(logf(24.0f), gammaf(5.0f)); } TEST(math, gamma_r) { #if defined(__BIONIC__) int sign; ASSERT_DOUBLE_EQ(log(24.0), gamma_r(5.0, &sign)); ASSERT_EQ(1, sign); #else // __BIONIC__ GTEST_LOG_(INFO) << "glibc doesn't have gamma_r.\n"; #endif // __BIONIC__ } TEST(math, gammaf_r) { #if defined(__BIONIC__) int sign; ASSERT_FLOAT_EQ(logf(24.0f), gammaf_r(5.0f, &sign)); ASSERT_EQ(1, sign); #else // __BIONIC__ GTEST_LOG_(INFO) << "glibc doesn't have gammaf_r.\n"; #endif // __BIONIC__ } TEST(math, lgamma) { ASSERT_DOUBLE_EQ(log(24.0), lgamma(5.0)); } TEST(math, lgammaf) { ASSERT_FLOAT_EQ(logf(24.0f), lgammaf(5.0f)); } TEST(math, lgammal) { ASSERT_DOUBLE_EQ(logl(24.0L), lgammal(5.0L)); } TEST(math, lgamma_r) { int sign; ASSERT_DOUBLE_EQ(log(24.0), lgamma_r(5.0, &sign)); ASSERT_EQ(1, sign); } TEST(math, lgamma_r_17471883) { int sign; sign = 0; ASSERT_DOUBLE_EQ(HUGE_VAL, lgamma_r(0.0, &sign)); ASSERT_EQ(1, sign); sign = 0; ASSERT_DOUBLE_EQ(HUGE_VAL, lgamma_r(-0.0, &sign)); ASSERT_EQ(-1, sign); } TEST(math, lgammaf_r) { int sign; ASSERT_FLOAT_EQ(logf(24.0f), lgammaf_r(5.0f, &sign)); ASSERT_EQ(1, sign); } TEST(math, lgammaf_r_17471883) { int sign; sign = 0; ASSERT_FLOAT_EQ(HUGE_VALF, lgammaf_r(0.0f, &sign)); ASSERT_EQ(1, sign); sign = 0; ASSERT_FLOAT_EQ(HUGE_VALF, lgammaf_r(-0.0f, &sign)); ASSERT_EQ(-1, sign); } TEST(math, lgammal_r) { int sign; ASSERT_DOUBLE_EQ(log(24.0L), lgamma_r(5.0L, &sign)); ASSERT_EQ(1, sign); } TEST(math, lgammal_r_17471883) { int sign; sign = 0; ASSERT_DOUBLE_EQ(HUGE_VAL, lgammal_r(0.0L, &sign)); ASSERT_EQ(1, sign); sign = 0; ASSERT_DOUBLE_EQ(HUGE_VAL, lgammal_r(-0.0L, &sign)); ASSERT_EQ(-1, sign); } TEST(math, tgamma) { ASSERT_DOUBLE_EQ(24.0, tgamma(5.0)); } TEST(math, tgammaf) { ASSERT_FLOAT_EQ(24.0f, tgammaf(5.0f)); } TEST(math, tgammal) { ASSERT_DOUBLE_EQ(24.0L, tgammal(5.0L)); } TEST(math, j0) { ASSERT_DOUBLE_EQ(1.0, j0(0.0)); ASSERT_DOUBLE_EQ(0.76519768655796661, j0(1.0)); } TEST(math, j0f) { ASSERT_FLOAT_EQ(1.0f, j0f(0.0f)); ASSERT_FLOAT_EQ(0.76519769f, j0f(1.0f)); } TEST(math, j1) { ASSERT_DOUBLE_EQ(0.0, j1(0.0)); ASSERT_DOUBLE_EQ(0.44005058574493355, j1(1.0)); } TEST(math, j1f) { ASSERT_FLOAT_EQ(0.0f, j1f(0.0f)); ASSERT_FLOAT_EQ(0.44005057f, j1f(1.0f)); } TEST(math, jn) { ASSERT_DOUBLE_EQ(0.0, jn(4, 0.0)); ASSERT_DOUBLE_EQ(0.0024766389641099553, jn(4, 1.0)); } TEST(math, jnf) { ASSERT_FLOAT_EQ(0.0f, jnf(4, 0.0f)); ASSERT_FLOAT_EQ(0.0024766389f, jnf(4, 1.0f)); } TEST(math, y0) { ASSERT_DOUBLE_EQ(-HUGE_VAL, y0(0.0)); ASSERT_DOUBLE_EQ(0.08825696421567697, y0(1.0)); } TEST(math, y0f) { ASSERT_FLOAT_EQ(-HUGE_VALF, y0f(0.0f)); ASSERT_FLOAT_EQ(0.088256963f, y0f(1.0f)); } TEST(math, y1) { ASSERT_DOUBLE_EQ(-HUGE_VAL, y1(0.0)); ASSERT_DOUBLE_EQ(-0.78121282130028868, y1(1.0)); } TEST(math, y1f) { ASSERT_FLOAT_EQ(-HUGE_VALF, y1f(0.0f)); ASSERT_FLOAT_EQ(-0.78121281f, y1f(1.0f)); } TEST(math, yn) { ASSERT_DOUBLE_EQ(-HUGE_VAL, yn(4, 0.0)); ASSERT_DOUBLE_EQ(-33.278423028972114, yn(4, 1.0)); } TEST(math, ynf) { ASSERT_FLOAT_EQ(-HUGE_VALF, ynf(4, 0.0f)); ASSERT_FLOAT_EQ(-33.278423f, ynf(4, 1.0f)); } TEST(math, frexp) { int exp; double dr = frexp(1024.0, &exp); ASSERT_DOUBLE_EQ(1024.0, scalbn(dr, exp)); } TEST(math, frexpf) { int exp; float fr = frexpf(1024.0f, &exp); ASSERT_FLOAT_EQ(1024.0f, scalbnf(fr, exp)); } TEST(math, frexpl) { int exp; long double ldr = frexpl(1024.0L, &exp); ASSERT_DOUBLE_EQ(1024.0L, scalbnl(ldr, exp)); } TEST(math, modf) { double di; double df = modf(123.75, &di); ASSERT_DOUBLE_EQ(123.0, di); ASSERT_DOUBLE_EQ(0.75, df); } TEST(math, modff) { float fi; float ff = modff(123.75f, &fi); ASSERT_FLOAT_EQ(123.0f, fi); ASSERT_FLOAT_EQ(0.75f, ff); } TEST(math, modfl) { long double ldi; long double ldf = modfl(123.75L, &ldi); ASSERT_DOUBLE_EQ(123.0L, ldi); ASSERT_DOUBLE_EQ(0.75L, ldf); } TEST(math, remquo) { int q; double d = remquo(13.0, 4.0, &q); ASSERT_EQ(3, q); ASSERT_DOUBLE_EQ(1.0, d); } TEST(math, remquof) { int q; float f = remquof(13.0f, 4.0f, &q); ASSERT_EQ(3, q); ASSERT_FLOAT_EQ(1.0, f); } TEST(math, remquol) { int q; long double ld = remquol(13.0L, 4.0L, &q); ASSERT_DOUBLE_EQ(3L, q); ASSERT_DOUBLE_EQ(1.0L, ld); } // https://code.google.com/p/android/issues/detail?id=6697 TEST(math, frexpf_public_bug_6697) { int exp; float fr = frexpf(14.1f, &exp); ASSERT_FLOAT_EQ(14.1f, scalbnf(fr, exp)); } TEST(math, exp2_STRICT_ALIGN_OpenBSD_bug) { // OpenBSD/x86's libm had a bug here, but it was already fixed in FreeBSD: // http://svnweb.FreeBSD.org/base/head/lib/msun/src/math_private.h?revision=240827&view=markup ASSERT_DOUBLE_EQ(5.0, exp2(log2(5))); ASSERT_FLOAT_EQ(5.0f, exp2f(log2f(5))); ASSERT_DOUBLE_EQ(5.0L, exp2l(log2l(5))); } TEST(math, nextafterl_OpenBSD_bug) { // OpenBSD/x86's libm had a bug here. ASSERT_TRUE(nextafter(1.0, 0.0) - 1.0 < 0.0); ASSERT_TRUE(nextafterf(1.0f, 0.0f) - 1.0f < 0.0f); ASSERT_TRUE(nextafterl(1.0L, 0.0L) - 1.0L < 0.0L); } #include "math_data/acos_intel_data.h" TEST(math, acos_intel) { DoMathDataTest<1>(g_acos_intel_data, acos); } #include "math_data/acosf_intel_data.h" TEST(math, acosf_intel) { DoMathDataTest<1>(g_acosf_intel_data, acosf); } #include "math_data/acosh_intel_data.h" TEST(math, acosh_intel) { DoMathDataTest<2>(g_acosh_intel_data, acosh); } #include "math_data/acoshf_intel_data.h" TEST(math, acoshf_intel) { DoMathDataTest<2>(g_acoshf_intel_data, acoshf); } #include "math_data/asin_intel_data.h" TEST(math, asin_intel) { DoMathDataTest<1>(g_asin_intel_data, asin); } #include "math_data/asinf_intel_data.h" TEST(math, asinf_intel) { DoMathDataTest<1>(g_asinf_intel_data, asinf); } #include "math_data/asinh_intel_data.h" TEST(math, asinh_intel) { DoMathDataTest<2>(g_asinh_intel_data, asinh); } #include "math_data/asinhf_intel_data.h" TEST(math, asinhf_intel) { DoMathDataTest<2>(g_asinhf_intel_data, asinhf); } #include "math_data/atan2_intel_data.h" TEST(math, atan2_intel) { DoMathDataTest<2>(g_atan2_intel_data, atan2); } #include "math_data/atan2f_intel_data.h" TEST(math, atan2f_intel) { DoMathDataTest<2>(g_atan2f_intel_data, atan2f); } #include "math_data/atan_intel_data.h" TEST(math, atan_intel) { DoMathDataTest<1>(g_atan_intel_data, atan); } #include "math_data/atanf_intel_data.h" TEST(math, atanf_intel) { DoMathDataTest<1>(g_atanf_intel_data, atanf); } #include "math_data/atanh_intel_data.h" TEST(math, atanh_intel) { DoMathDataTest<2>(g_atanh_intel_data, atanh); } #include "math_data/atanhf_intel_data.h" TEST(math, atanhf_intel) { DoMathDataTest<2>(g_atanhf_intel_data, atanhf); } #include "math_data/cbrt_intel_data.h" TEST(math, cbrt_intel) { DoMathDataTest<1>(g_cbrt_intel_data, cbrt); } #include "math_data/cbrtf_intel_data.h" TEST(math, cbrtf_intel) { DoMathDataTest<1>(g_cbrtf_intel_data, cbrtf); } #include "math_data/ceil_intel_data.h" TEST(math, ceil_intel) { DoMathDataTest<1>(g_ceil_intel_data, ceil); } #include "math_data/ceilf_intel_data.h" TEST(math, ceilf_intel) { DoMathDataTest<1>(g_ceilf_intel_data, ceilf); } #include "math_data/copysign_intel_data.h" TEST(math, copysign_intel) { DoMathDataTest<1>(g_copysign_intel_data, copysign); } #include "math_data/copysignf_intel_data.h" TEST(math, copysignf_intel) { DoMathDataTest<1>(g_copysignf_intel_data, copysignf); } #include "math_data/cos_intel_data.h" TEST(math, cos_intel) { DoMathDataTest<1>(g_cos_intel_data, cos); } #include "math_data/cosf_intel_data.h" TEST(math, cosf_intel) { DoMathDataTest<1>(g_cosf_intel_data, cosf); } #include "math_data/cosh_intel_data.h" TEST(math, cosh_intel) { DoMathDataTest<2>(g_cosh_intel_data, cosh); } #include "math_data/coshf_intel_data.h" TEST(math, coshf_intel) { DoMathDataTest<2>(g_coshf_intel_data, coshf); } #include "math_data/exp_intel_data.h" TEST(math, exp_intel) { DoMathDataTest<1>(g_exp_intel_data, exp); } #include "math_data/expf_intel_data.h" TEST(math, expf_intel) { DoMathDataTest<1>(g_expf_intel_data, expf); } #include "math_data/exp2_intel_data.h" TEST(math, exp2_intel) { DoMathDataTest<1>(g_exp2_intel_data, exp2); } #include "math_data/exp2f_intel_data.h" TEST(math, exp2f_intel) { DoMathDataTest<1>(g_exp2f_intel_data, exp2f); } #include "math_data/expm1_intel_data.h" TEST(math, expm1_intel) { DoMathDataTest<1>(g_expm1_intel_data, expm1); } #include "math_data/expm1f_intel_data.h" TEST(math, expm1f_intel) { DoMathDataTest<1>(g_expm1f_intel_data, expm1f); } #include "math_data/fabs_intel_data.h" TEST(math, fabs_intel) { DoMathDataTest<1>(g_fabs_intel_data, fabs); } #include "math_data/fabsf_intel_data.h" TEST(math, fabsf_intel) { DoMathDataTest<1>(g_fabsf_intel_data, fabsf); } #include "math_data/fdim_intel_data.h" TEST(math, fdim_intel) { DoMathDataTest<1>(g_fdim_intel_data, fdim); } #include "math_data/fdimf_intel_data.h" TEST(math, fdimf_intel) { DoMathDataTest<1>(g_fdimf_intel_data, fdimf); } #include "math_data/floor_intel_data.h" TEST(math, floor_intel) { DoMathDataTest<1>(g_floor_intel_data, floor); } #include "math_data/floorf_intel_data.h" TEST(math, floorf_intel) { DoMathDataTest<1>(g_floorf_intel_data, floorf); } #include "math_data/fma_intel_data.h" TEST(math, fma_intel) { DoMathDataTest<1>(g_fma_intel_data, fma); } #include "math_data/fmaf_intel_data.h" TEST(math, fmaf_intel) { DoMathDataTest<1>(g_fmaf_intel_data, fmaf); } #include "math_data/fmax_intel_data.h" TEST(math, fmax_intel) { DoMathDataTest<1>(g_fmax_intel_data, fmax); } #include "math_data/fmaxf_intel_data.h" TEST(math, fmaxf_intel) { DoMathDataTest<1>(g_fmaxf_intel_data, fmaxf); } #include "math_data/fmin_intel_data.h" TEST(math, fmin_intel) { DoMathDataTest<1>(g_fmin_intel_data, fmin); } #include "math_data/fminf_intel_data.h" TEST(math, fminf_intel) { DoMathDataTest<1>(g_fminf_intel_data, fminf); } #include "math_data/fmod_intel_data.h" TEST(math, fmod_intel) { DoMathDataTest<1>(g_fmod_intel_data, fmod); } #include "math_data/fmodf_intel_data.h" TEST(math, fmodf_intel) { DoMathDataTest<1>(g_fmodf_intel_data, fmodf); } #include "math_data/frexp_intel_data.h" TEST(math, frexp_intel) { DoMathDataTest<1>(g_frexp_intel_data, frexp); } #include "math_data/frexpf_intel_data.h" TEST(math, frexpf_intel) { DoMathDataTest<1>(g_frexpf_intel_data, frexpf); } #include "math_data/hypot_intel_data.h" TEST(math, hypot_intel) { DoMathDataTest<1>(g_hypot_intel_data, hypot); } #include "math_data/hypotf_intel_data.h" TEST(math, hypotf_intel) { DoMathDataTest<1>(g_hypotf_intel_data, hypotf); } #include "math_data/ilogb_intel_data.h" TEST(math, ilogb_intel) { DoMathDataTest<1>(g_ilogb_intel_data, ilogb); } #include "math_data/ilogbf_intel_data.h" TEST(math, ilogbf_intel) { DoMathDataTest<1>(g_ilogbf_intel_data, ilogbf); } #include "math_data/ldexp_intel_data.h" TEST(math, ldexp_intel) { DoMathDataTest<1>(g_ldexp_intel_data, ldexp); } #include "math_data/ldexpf_intel_data.h" TEST(math, ldexpf_intel) { DoMathDataTest<1>(g_ldexpf_intel_data, ldexpf); } #include "math_data/llrint_intel_data.h" TEST(math, llrint_intel) { DoMathDataTest<1>(g_llrint_intel_data, llrint); } #include "math_data/llrintf_intel_data.h" TEST(math, llrintf_intel) { DoMathDataTest<1>(g_llrintf_intel_data, llrintf); } #include "math_data/log_intel_data.h" TEST(math, log_intel) { DoMathDataTest<1>(g_log_intel_data, log); } #include "math_data/logf_intel_data.h" TEST(math, logf_intel) { DoMathDataTest<1>(g_logf_intel_data, logf); } #include "math_data/log10_intel_data.h" TEST(math, log10_intel) { DoMathDataTest<1>(g_log10_intel_data, log10); } #include "math_data/log10f_intel_data.h" TEST(math, log10f_intel) { DoMathDataTest<1>(g_log10f_intel_data, log10f); } #include "math_data/log1p_intel_data.h" TEST(math, log1p_intel) { DoMathDataTest<1>(g_log1p_intel_data, log1p); } #include "math_data/log1pf_intel_data.h" TEST(math, log1pf_intel) { DoMathDataTest<1>(g_log1pf_intel_data, log1pf); } #include "math_data/log2_intel_data.h" TEST(math, log2_intel) { DoMathDataTest<1>(g_log2_intel_data, log2); } #include "math_data/log2f_intel_data.h" TEST(math, log2f_intel) { DoMathDataTest<1>(g_log2f_intel_data, log2f); } #include "math_data/logb_intel_data.h" TEST(math, logb_intel) { DoMathDataTest<1>(g_logb_intel_data, logb); } #include "math_data/logbf_intel_data.h" TEST(math, logbf_intel) { DoMathDataTest<1>(g_logbf_intel_data, logbf); } #include "math_data/lrint_intel_data.h" TEST(math, lrint_intel) { DoMathDataTest<1>(g_lrint_intel_data, lrint); } #include "math_data/lrintf_intel_data.h" TEST(math, lrintf_intel) { DoMathDataTest<1>(g_lrintf_intel_data, lrintf); } #include "math_data/modf_intel_data.h" TEST(math, modf_intel) { DoMathDataTest<1>(g_modf_intel_data, modf); } #include "math_data/modff_intel_data.h" TEST(math, modff_intel) { DoMathDataTest<1>(g_modff_intel_data, modff); } #include "math_data/nearbyint_intel_data.h" TEST(math, nearbyint_intel) { DoMathDataTest<1>(g_nearbyint_intel_data, nearbyint); } #include "math_data/nearbyintf_intel_data.h" TEST(math, nearbyintf_intel) { DoMathDataTest<1>(g_nearbyintf_intel_data, nearbyintf); } #include "math_data/nextafter_intel_data.h" TEST(math, nextafter_intel) { DoMathDataTest<1>(g_nextafter_intel_data, nextafter); } #include "math_data/nextafterf_intel_data.h" TEST(math, nextafterf_intel) { DoMathDataTest<1>(g_nextafterf_intel_data, nextafterf); } #include "math_data/pow_intel_data.h" TEST(math, pow_intel) { DoMathDataTest<1>(g_pow_intel_data, pow); } #include "math_data/powf_intel_data.h" TEST(math, powf_intel) { DoMathDataTest<1>(g_powf_intel_data, powf); } #include "math_data/remainder_intel_data.h" TEST(math, remainder_intel) { DoMathDataTest<1>(g_remainder_intel_data, remainder); } #include "math_data/remainderf_intel_data.h" TEST(math, remainderf_intel) { DoMathDataTest<1>(g_remainderf_intel_data, remainderf); } #include "math_data/remquo_intel_data.h" TEST(math, remquo_intel) { DoMathDataTest<1>(g_remquo_intel_data, remquo); } #include "math_data/remquof_intel_data.h" TEST(math, remquof_intel) { DoMathDataTest<1>(g_remquof_intel_data, remquof); } #include "math_data/rint_intel_data.h" TEST(math, rint_intel) { DoMathDataTest<1>(g_rint_intel_data, rint); } #include "math_data/rintf_intel_data.h" TEST(math, rintf_intel) { DoMathDataTest<1>(g_rintf_intel_data, rintf); } #include "math_data/round_intel_data.h" TEST(math, round_intel) { DoMathDataTest<1>(g_round_intel_data, round); } #include "math_data/roundf_intel_data.h" TEST(math, roundf_intel) { DoMathDataTest<1>(g_roundf_intel_data, roundf); } #include "math_data/scalb_intel_data.h" TEST(math, scalb_intel) { DoMathDataTest<1>(g_scalb_intel_data, scalb); } #include "math_data/scalbf_intel_data.h" TEST(math, scalbf_intel) { DoMathDataTest<1>(g_scalbf_intel_data, scalbf); } #include "math_data/scalbn_intel_data.h" TEST(math, scalbn_intel) { DoMathDataTest<1>(g_scalbn_intel_data, scalbn); } #include "math_data/scalbnf_intel_data.h" TEST(math, scalbnf_intel) { DoMathDataTest<1>(g_scalbnf_intel_data, scalbnf); } #include "math_data/significand_intel_data.h" TEST(math, significand_intel) { DoMathDataTest<1>(g_significand_intel_data, significand); } #include "math_data/significandf_intel_data.h" TEST(math, significandf_intel) { DoMathDataTest<1>(g_significandf_intel_data, significandf); } #include "math_data/sin_intel_data.h" TEST(math, sin_intel) { DoMathDataTest<1>(g_sin_intel_data, sin); } #include "math_data/sinf_intel_data.h" TEST(math, sinf_intel) { DoMathDataTest<1>(g_sinf_intel_data, sinf); } #include "math_data/sinh_intel_data.h" TEST(math, sinh_intel) { DoMathDataTest<2>(g_sinh_intel_data, sinh); } #include "math_data/sinhf_intel_data.h" TEST(math, sinhf_intel) { DoMathDataTest<2>(g_sinhf_intel_data, sinhf); } #include "math_data/sincos_intel_data.h" TEST(math, sincos_intel) { DoMathDataTest<1>(g_sincos_intel_data, sincos); } #include "math_data/sincosf_intel_data.h" TEST(math, sincosf_intel) { DoMathDataTest<1>(g_sincosf_intel_data, sincosf); } #include "math_data/sqrt_intel_data.h" TEST(math, sqrt_intel) { DoMathDataTest<1>(g_sqrt_intel_data, sqrt); } #include "math_data/sqrtf_intel_data.h" TEST(math, sqrtf_intel) { DoMathDataTest<1>(g_sqrtf_intel_data, sqrtf); } #include "math_data/tan_intel_data.h" TEST(math, tan_intel) { DoMathDataTest<1>(g_tan_intel_data, tan); } #include "math_data/tanf_intel_data.h" TEST(math, tanf_intel) { DoMathDataTest<1>(g_tanf_intel_data, tanf); } #include "math_data/tanh_intel_data.h" TEST(math, tanh_intel) { DoMathDataTest<2>(g_tanh_intel_data, tanh); } #include "math_data/tanhf_intel_data.h" TEST(math, tanhf_intel) { DoMathDataTest<2>(g_tanhf_intel_data, tanhf); } #include "math_data/trunc_intel_data.h" TEST(math, trunc_intel) { DoMathDataTest<1>(g_trunc_intel_data, trunc); } #include "math_data/truncf_intel_data.h" TEST(math, truncf_intel) { DoMathDataTest<1>(g_truncf_intel_data, truncf); }