compilation invoke real*16 is too slow, is this normal?

[bsmile_2nd]

The command I use is ifort -O3 -xT -mp1 -o a.out ***.f where reals are defined as real*16.

If what is used for real is real*8, then it takes seconds to finish, but if using real*16, it takes minutes, is this normal? The source code of the files to be compiled is about 5000 lines.

[TimP]

If your real*8 code spends most of the time in vectorized add and multiply, real*16 would be expected to take around 10 times as long. Where real*16 spends much of the time in divide and sqrt(), the slowdown would bemuch greater. Only the Itanium CPUs were designed to support real*16 without such an increase in run time.

[bsmile_2nd]

Thanks, tim18!