A libm-test-ulps file is used to define maximal error constraints in Units of Least Precision (ULP). If a function's error exceeds these values, the test fails.
Syntax and Keywords
Specify the function name followed by the allowed error for each floating-point type. Supported keywords are float, double, ldouble, and float128.
Function "yn":
float: 2
double: 6
Testing Different Rounding Modes
By default, tests use FE_TONEAREST. To test other rounding modes, prepend the function name with an underscore and the mode name:
_downward (FE_DOWNWARD)_towardzero (FE_TOWARDZERO)_upward (FE_UPWARD)
Example for FE_DOWNWARD:
Function "yn_downward":
float: 3
double: 7