ZYF

When Steve Lionel answered on comp.lang.fortran about “how to rename an array”, He told in Fortran 2003 there’s a MOVE_ALLOC intrinsic subroutine/function which has been implemented in Intel Visual Fortran 9.1.028 and later.

Just quote as follows:

This lets you move the “allocation” of one array (with its data) to another, the original array now being unallocated. With this, expanding an array requires one copy and two arrays allocated for a while, but it saves a second copy that would be required without it.

Let’s say you have an array A that you want to double its size while preserving the original data.

Without MOVE_ALLOC you’d do this:

OLD_SIZE = SIZE(A)
ALLOCATE (TEMP_ARRAY(OLD_SIZE)
TEMP_ARRAY = A ! First copy
DEALLOCATE (A)
ALLOCATE (A(2*OLD_SIZE))
A(1:OLD_SIZE) = TEMP_ARRAY ! Second copy
DEALLOCATE (TEMP_ARRAY)

With MOVE_ALLOC you can do this:

ALLOCATE (TEMP_ARRAY(2*SIZE(A))
TEMP_ARRAY(1:SIZE(A)) = A
CALL MOVE_ALLOC (TEMP_ARRAY, A)
! TEMP_ARRAYis now unallocated

You say you are using “Visual Fortran” but don’t provide further details. MOVE_ALLOC is supported in Intel Visual Fortran 9.1.028 and higher, not in Compaq or Digital Visual Fortran. For Intel Visual Fortran, you’ll currently find it described only in the Release Notes and not the regular manuals.

Steve

从Greg Wilson 的blog上看到这样一篇文章 Sign Error => Five Papers Retracted，提到最近Scripps Instititue撤掉了他们在各种顶级期刊上发表的5篇文章，仅仅因为他们的计算机程序中的一个符号错误，导致计算结果存在问题。他们很勇敢的承认了自己研究的失误，为自己文章中的错误结果对其他研究人员的误导表示了歉意。

现在的研究习惯基本上是这样的一种思路，你的研究结果必须得到验证才可以被认可，Scripps的研究结果因为发表在顶级期刊上，其权威性更强，其所承担的验证算例的权重更大。如果你的研究结果与其不相符合，就导致结果无法被认可，而更可怕的是一些错误的结果可能碰巧与其结论相符，导致更大的错误发生，其造成的损失是可想而知的。

但Scripps Institute 能够在发现错误之后，勇敢的站出来，还是相当值得称赞的，虽然这意味着他们数年的研究结果全部报销，而研究的权威性也有可能受到质疑。

如果我们的研究结果出了问题，我们是否有勇气站出来，告诉所有人“我们错了，对不起”？