llvm/libclc/generic/lib/integer/rhadd.inc

//rhadd = (x+y+1)>>1
//This can be simplified to x>>1 + y>>1 + (1 if either x or y have the 1s bit set)
//This saves us having to do any checks for overflow in the addition sums
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE rhadd(__CLC_GENTYPE x, __CLC_GENTYPE y) {
    return (x>>(__CLC_GENTYPE)1)+(y>>(__CLC_GENTYPE)1)+((x&(__CLC_GENTYPE)1)|(y&(__CLC_GENTYPE)1));
}