ice with do i = 1,n

The following code produces an ICE when compiled with 32-bit or 64-bit when n >= 2**31

10. do i = 1,n ! works with = one,n *** Error 637: Internal compiler error - floating point exception

However it works with do I = one,n or if n < 2^31

This occurs for Ver 8.20 and earlier versions that I have tested.

   integer*4, parameter :: million = 1000000
   integer*8, parameter :: billion = 1000*million
   integer*8, parameter :: n = 3*billion ! works with 1*billion
   integer*4 :: one = 1
   integer*8 :: i
   real*8    :: s, e, d

   d = 0.001d0
   s = 0
   do i = 1,n   ! works with = one,n
     s = s + d
   end do

   e = n*d
   write (*,*) 'n=',  n
   write (*,*) 's,e=',s, e
   write (*,*) 's-e=',s-e
   end

Thanks John. I have made a note of this.

integer*4, parameter :: one = 1 ... do i = one,n ! this fails

while integer*4 :: one = 1 ... do i = one,n ! this is ok

Hi John I am very much astonished to believe that such a compilation errors exist. But, good that it has been identified. This really helps, to improvise our FTN95. Thank you

This has now been fixed for the next release of FTN95.

The following variant of John's example involves no REAL variables of any KIND, but the 8.10 compilers (32 and 64 bit targets) abort compilation with a startling error message.

The modified source code:

   program fbug
   implicit none
   integer*8, parameter :: n = 3_4*1000_4*1000_4*1000_4 ! 3 billion
   integer*4 :: one = 1 
   integer*8 :: i , s, d

   print *,' n = ',n
   d = 2
   s = 0

   do i = 1_4,n   ! works with = one,n 
     s = s + d 
   end do 

   write (*,*) 'n, s =',  n , s
   end 

The compiler's error message:

s:\FTN95>ftn95 fbug.f90 /64 /lgo
[FTN95/x64 Ver. 8.10.0 Copyright (c) Silverfrost Ltd 1993-2017]
0011)    do i = 1_4,n   ! works with = one,n
*** Internal compiler error - floating point exception

The current fix also works for this variant.

Quoted from PaulLaidler The current fix also works for this variant.

Thanks, Paul.

Impressive: A fix was provided to a problem just six hours after the problem was reported!

Paul,

Thanks for the prompt response to this and other fixes. Could a Ver 8.21 be made available ?

John

John

That is an option that is being considered.