testing_utilities Namespace Reference

A namespace for various testing related utility functions. Many of those functions will likely use the STL so this namespace should not be considered compatible with CUDA/HIP. More...

Classes
class	GlobalString
	Holds a single std::string that's intended to be read only and global. Use for storing the path of the root directory of Cholla. More...

Functions
int64_t	ulpsDistanceDbl (double const &a, double const &b)
	Compute the Units in the Last Place (ULP) difference between two doubles.
bool	nearlyEqualDbl (double const &a, double const &b, double &absoluteDiff, int64_t &ulpsDiff, double const &fixedEpsilon=1E-14, int64_t const &ulpsEpsilon=4)
	Check if two doubles are nearly equal.
void	wrapperEqual (int i, int j, int k, std::string const &dataSetName, double test_value, double fid_value, double fixedEpsilon=5.0E-12)
void	analyticConstant (system_test::SystemTestRunner testObject, std::string const &dataSetName, double value)
void	analyticSine (system_test::SystemTestRunner testObject, std::string const &dataSetName, double constant, double amplitude, double kx, double ky, double kz, double phase, double tolerance)
template<int checkType = 0>
void	Check_Results (double fiducialNumber, double testNumber, std::string const &outString, double fixedEpsilon=-999, int64_t ulpsEpsilon=-999)
	A simple function to compare two doubles with the nearlyEqualDbl function, perform a GTest assert on the result, and print out the values.
void	Check_Interface (reconstruction::InterfaceState const &test_data, reconstruction::InterfaceState const &fiducial_data, size_t const direction)
	Function for checking if every member in a reconstruction::InterfaceState struct matches the fiducial values.

Detailed Description

A namespace for various testing related utility functions. Many of those functions will likely use the STL so this namespace should not be considered compatible with CUDA/HIP.

Function Documentation

Check_Interface()

void testing_utilities::Check_Interface ( reconstruction::InterfaceState const &  test_data, reconstruction::InterfaceState const &  fiducial_data, size_t const  direction  )

inline

Function for checking if every member in a reconstruction::InterfaceState struct matches the fiducial values.

Parameters

[in]	test_data	The data to test
[in]	fiducial_data	The fiducial data
[in]	direction	What direction the test was run in.

Check_Results()

template<int checkType = 0>

void testing_utilities::Check_Results	(	double	fiducialNumber,
		double	testNumber,
		std::string const &	outString,
		double	fixedEpsilon = -999,
		int64_t	ulpsEpsilon = -999
	)

A simple function to compare two doubles with the nearlyEqualDbl function, perform a GTest assert on the result, and print out the values.

Template Parameters

checkType

The type of GTest assertion to use. "0" for and "EXPECT" and "1" for an "ASSERT"

Parameters

[in]	fiducialNumber	The fiducial number to test against
[in]	testNumber	The unverified number to test
[in]	outString	A string to be printed in the first line of the output message. Format will be "Difference in outString"
[in]	fixedEpsilon	The fixed epsilon to use in the comparison. Negative values are ignored and default behaviour is used
[in]	ulpsEpsilon	The ULP epsilon to use in the comparison. Negative values are ignored and default behaviour is used

nearlyEqualDbl()

bool testing_utilities::nearlyEqualDbl	(	double const &	a,
		double const &	b,
		double &	absoluteDiff,
		int64_t &	ulpsDiff,
		double const &	fixedEpsilon = 1E-14,
		int64_t const &	ulpsEpsilon = 4
	)

Check if two doubles are nearly equal.

This function checks if two doubles are "nearly equal" which is defined as either: A) the absolute difference between them is less than the fixedEpsilon argument or B) the units in the last place (ULP) difference is less than the ulpsEpsilon argument. Both of the epsilon arguments have default values which generally should not need to be changed.

Why does fixedEpsilon default to 1E-14? Running the Sod shock tube when Cholla was compiled with GCC 9.3.0 vs. XL 16.1.1-10 on Summit lead to absolute differences in the results up to 1.77636E-15. A priori we chose that a difference between two numbers that was less than one order of magnitude greater than the difference between compilers would be considered "equal". I.e. since the maximum absolute error between the GCC and XL compilers was ~1.7E-15 our allowed margin of error should be ~1E-14.

Why does ulpsEpsilon default to 4? Repeating the test above I computed the largest ULP difference that wasn't caught by the absolute difference requirement of 1E-14. It turns out that there were no uncaught differences at all so I kept ulpsEpsilon at 4 since that's the Googletest default for their floating point assertions

Parameters

[in]	a	The first double you wish to compare. Order doesn't matter.
[in]	b	The first double you wish to compare. Order doesn't matter.
[out]	absoluteDiff	The absolute difference between the numbers. Only returned if the numbers are not equal. If the numbers are equal then behaviour is undefined
[out]	ulpsDiff	The ULP difference between the numbers. Only returned if the numbers are not equal. If the numbers are equal then behaviour is undefined
[in]	fixedEpsilon	The allowed difference in real numbers. Defaults to 1E-14
[in]	ulpsEpsilon	The allowed difference of ULPs. Defaults to 4

Returns

bool Whether or not the numbers are equal

ulpsDistanceDbl()

int64_t testing_utilities::ulpsDistanceDbl	(	double const &	a,
		double const &	b
	)

Compute the Units in the Last Place (ULP) difference between two doubles.

This function is modified from Comparing Floating-Point Numbers Is Tricky by Matt Kline which is in turn based on Comparing Floating Point Numbers, 2012 Edition by Bruce Dawson. The latter seems to be the bible of floating point comparison and is the basis of Googletests ASSERT_DOUBLE_EQ assertion.

This particular function checks that the two numbers if the numbers are perfectly equal, +0, -0, Nan, inf, or differently signed then it computes the ULP difference between them are returns it

Parameters

[in]	a	The first double you wish to compare. Order doesn't matter.
[in]	b	The second double you wish to compare. Order doesn't matter.

Returns

int64_t The ULP distance between a and b.