automatic array stack size limit bug

automatic array stack size limit bug

Bild des Benutzers swise@math.uci.edu

Fortran users,

The problem in ifort 8.0, now inherited in ifort 9.0, that puts a severe limit on automatic array sizes should be considered a bug by Intel. Steve Lionel, with all due respect, keeps suggesting that we abstain from using large automatic arrays in favor of allocatable arrays, or that we increase the stack size.

But these are unacceptable fixes. Even when the stack size is unlimited, on most machines the sizes of the automatic arrays is limited so that 3D computation using them is not possible. Keep in mind that with ifc 7.1, NAG f95, and the free g95 compilers have nowhere even close to the size restriction. (I can give examples if this thread progresses; its not a machine problem that stack size can fix.) And I don't want to go through 1000's of lines of code to make my automatic arrays allocatable just to get around this bug. Plus there are efficiency issues with doing so.

So where did the problem come from? I suspect that it popped up during the merge of the DEC/Compac/HP and Intel compilers, becuase ifc 7.1 was fine in this respect.

I'm a non-commercial user, who was very happy with ifc 7.1. So why do stay with Intel through the development of ifort versions 8 and 9? Because, there is no doubt in my mind that ifort can be a great compiler, and I want to support its progress, especially with Mac computers coming into the family. (Intel, please support a free, non-commercial version for Mac OS X!) And who better to develop a compiler than the chip maker.

Commercial users, please complain about this bug. Don't change your arrays from automatic to allocatable just to get around this bug. Threaten to jump ship if they don't fix it. Commercial and non-comercial users alike, please add your names to this thread if you agree.

Dr. Steven Wise
Mathematics
UC, Irvine

1 Beitrag / 0 neu
Nähere Informationen zur Compiler-Optimierung finden Sie in unserem Optimierungshinweis.