Appendix A: Handling of Special Cases

Some mathematical functions implemented in Intel IPP are not defined for all possible argument values. This appendix describes how the corresponding Intel IPP image processing functions handle situations when their input arguments fall outside the range of function definition or may lead to ambiguously determined output results.

Table Special Cases for Intel IPP Image Processing Functions below summarizes these special cases for different functions and lists result values together with status codes returned by the functions. The status codes ending with Err (except for the ippStsNoErr status) indicate an error. When an error occurs, the function execution is interrupted. All other status codes indicate that the input argument is outside the range, but the function execution is continued with the corresponding result value.

Special Cases for Intel IPP Image Processing Functions
Function Base Name	Data Type	Case Description	Result Value	Status Code
ippiSqrt	16s 32f	Sqrt (x <0) Sqrt (x <0)	0 NAN_32F	ippStsSqrtNegArg ippStsSqrtNegArg
ippiDiv	8u 16s 32f 16sc 32sc 32fc	Div (0/0) Div (x/0) Div (0/0) Div (x/0), x>0 Div (x/0), x<0 Div (0/0) Div (x/0), x>0 Div (x/0), x<0 Div (0/0) Div (x/0), Div (0/0) Div (x/0), x>0 Div (x/0), x<0 Div (0/0) Div (x/0), x>0 Div (x/0), x<0	0 IPP_MAX_8U 0 IPP_MAX_16S IPP_MIN_16S NAN_32F INF_32F INF_NEG_32F 0 0 0 IPP_MAX_32S IPP_MIN_32S NAN_32F INF_32F INF_NEG_32F	ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero ippStsDivByZero
ippiDivC	all	Div(x/const), const=0	-	ippStsDivByZeroErr
ippiLn	8u 16s 32f	Ln (0) Ln (0) Ln (x<0) Ln (x<0) Ln(x<IPP_MINABS_32F)	0 IPP_MIN_16S IPP_MIN_16S NAN_32F INF_NEG_32F	ippStsLnZeroArg ippStsLnZeroArg ippStsLnNegArg ippStsLnNegArg ippStsLnZeroArg
ippiExp	8u 16s 32f	overflow overflow overflow	IPP_MAX_8U IPP_MAX_16S INF_32F	ippStsNoErr ippStsNoErr ippStsNoErr

Here x denotes an input value. For the definition of the constants used, see Image Data Types and Ranges in Intel® Integrated Performance Primitives Concepts.

Note that flavors of the same math function operating on different data types may produce diferent results for the equal argument values. However, for a given function and a fixed data type, handling of special cases is the same for all function flavors that have different descriptors in their names. For example, logarithm function ippiLn operating on 16s data treats zero argument values in the same way for all its flavors ippiLn_16s_C1RSfs, ippiLn_16s_C3RSfs, ippiLn_16s_C1IRSfs, and ippiLn_16s_C3IRSfs.