Optimization bug with 4.8.1

[Ignacio Fernández Galván]

--------------------------------------------
On Fri, 25/10/13, Steve Kargl <sgk@troutmask.apl.washington.edu> wrote:

>> So far, I've determined the problem is not the MAX statement, but
>> rather the calculation/use of ib,ic and I get the  correct value if
>> I add some write statement in the loop, but I've been  unsuccessful
>> in fixing this without a write.
>
> The above suggests that you are indexing an array with an 
> out-of-bounds value.
> 
> Compile the code with -fcheck=all -Wall -Werror  -fmax-errors=1.

Unfortunately I can't do that. This program uses its own memory management with C code, I believe, and the iWork array is declared with a very small size (8 or so), even though much larger indices are referenced.

I forgot to say that this code works (with -O2 and -O3) with other compilers, like gfortran 4.6, portland, intel... so the problem is only introduced with -O2 in 4.8.1, and in particular it can be avoided with -fno-aggressive-loop-optimizations.

To debug a bit further:

      n=0
      DO i=1,m
         WRITE(6,*) is(3,i),is(2,i)
      END DO
      DO i=1,m
         ib = is(3,i)
         ic = is(2,i)
         n=MAX(n,ib*ic)
      END DO
      WRITE(6,*) n

with -O0 I get:

1                    3
3                    2
1                    2
1                    2
6

but with -O2:

1                    3
3                    2
1                    2
1                    2
3

This looks like a compiler bug to me. What I want to know is if it's known (and fixed in later versions), and if there's an easy way to work around it... If it's not known, I'll try to create a self-contained test case.

Ignacio