opj_uint_adds() is questionable

[szukw000]

'opj_uintmath.h' contains the lines:

/**
Get the saturated sum of two unsigned integers
@return Returns saturated sum of a+b
*/
static INLINE OPJ_UINT32 opj_uint_adds(OPJ_UINT32 a, OPJ_UINT32 b) {
OPJ_UINT64 sum = (OPJ_UINT64)a + (OPJ_UINT64)b;
return -(OPJ_UINT32)(sum >> 32) | (OPJ_UINT32)sum;
}

'opj_uint_adds()' is always the first value of:

static INLINE OPJ_UINT32 opj_uint_min(OPJ_UINT32 a, OPJ_UINT32 b) {
return a < b ? a : b;
}

This means: the value returned MUST not be negative.

Should the '-' sign in 'opj_uint_adds()' be removed?

winfried

[mayeut]

Winfried,

Strictly C speaking, it's questionable. However, I don't know of any hardware that openjpeg will run on that does not use 2's complement arithmetic.

The high 32 bits part of sum is either 1 or 0 depending on overflow.
If an overflow occurs, -(OPJ_UINT32)(sum >> 32) == (OPJ_UINT32)-1 == UINT32_MAX, that's all ones so once ORed with the low part, it's still UINT32_MAX.
If no overflow occurs, -(OPJ_UINT32)(sum >> 32) == 0, overall result is low part of sum.

Matthieu