Quad precision SPACING intrinsic

[Chris_S_]

I believe I have found a bug in the SPACING intrinsic when very small quad precision numbers are used. I am using version 10.1 of the linux 64 compiler, so if bug has already been fixed, please disregard this post.

It appears that when the true spacing between quad precision numbers falls below TINY, SPACING returns a nonsensical very large value. For example, SPACING(1Q-4898) = 3.362Q-4932 (equal to TINY), while SPACING(1Q-4899) = 1.487Q+4931 (notice the positive exponent). I even find SPACING(0Q0) = 4.583Q+4898.

A test program and output are attached.

[Steven_L_Intel1]

Your test program behaves correctly with the 12.0.3 compiler.

[mecej4]

> For example, SPACING(1Q-4898) = 3.362Q-4932 (equal to TINY)

This is correct, since SPACING(x) = TINY for | x | < TINY. In other words, SPACING gives only normalized numbers.

[Chris_S_]

Quoting mecej4

> For example, SPACING(1Q-4898) = 3.362Q-4932 (equal to TINY)

This is correct, since SPACING(x) = TINY for | x | < TINY. In other words, SPACING gives only normalized numbers.

mecej4: I understand that is what it should give, but SPACING(x) was not giving TINY for |x| < 2^110*TINY, it was instead giving something closer to HUGE. In any case, according to Steve Lionel's post, this issue appears to have been fixed in more recent releases (thanks, Steve!).