In OpenCL C 2.0 Specification page174, table38, there is:

exp(x)                        ≤ 3 + floor(fabs(2 * x)) ulp for the full profile,
                              and ≤ 4 ulp for the embedded profile.

The ulp constraint of exp of embedded profile defined here(4) could be more restrict than full profile(3+floor(fabs(2x))) in most cases. Should it be 4 + floor(fabs(2 * x)) when testing embedded profile?
best wishes!