[ICE] intrinsic call with allocatable array and scalar with -assume realloc_lhs

[Kacper_Kowalik]

Hi,
following code:

program ala
real, dimension(:), allocatable :: a
allocate(a(2))
a = 0.0
a = min(a, 1.0)
print *, a
deallocate(a)
end program ala

causes ICE when compiled with: ifort -assume realloc_lhs ala.f90
Compiler that I've used:
Intel(R) Fortran Intel(R) 64 Compiler XE for applications running on Intel(R) 64, Version 13.0.0.079 Build 20120731
Best regards,
Kacper Kowalik

[Heinz_B_Intel]

Hello Kacper I will investigate and will escalate to engineering, I will let you know about more details later. Heinz

[Heinz_B_Intel]

Hello Kacper I have escalated the problem to the compiler developers; case number is DPD200236655. I will let you know when a compiler with a fix will be available, Thanks for providing such a small test case Heinz

[Heinz_B_Intel]

Might be a good workaround for you: Swapping the arguments for MIN() prevents the bug; a = min(1.0, a) works as expected.

[Kacper_Kowalik]

HeinzB (Intel) wrote:

Might be a good workaround for you: Swapping the arguments for MIN() prevents the bug; a = min(1.0, a) works as expected.

That bug is still present in 13.0.1 and that workaround cannot be applied to all intrinsic funtion, like "ishft" where order of arguments matter. Better way to   avoid this is to explictly disable realloc_lhs in offending lines by adding (:), e.g. a(:) = min(a, 1.0)

Cheers,

Kacper

[Kacper_Kowalik]

Just noticed that new version of the compiler was released yesterday with the fix, sorry for the noise :-)

[Kacper_Kowalik]

Not fixed quite yet. Although it now works for original code with "min" it still ICEs for other functions (with ifort-13.1):

[fortran]

program ala
integer, dimension(:), allocatable :: a
allocate(a(2))
a = 0
a = ishft(a, -1)
deallocate(a)
end program ala

[/fortran]

[Steven_L_Intel1]

Sigh - I can reproduce this and will report it.  Thanks. Issue ID is DPD200240757.

[Steven_L_Intel1]

Kacper, the problem with ishft is expected to be fixed in Update 3. We have gone through all of the intrinsics where this could be a problem and implemented a common solution.

[Heinz_B_Intel]

The ICE has been fixed by 13.0 Update 1 already; sorry for the late update. I just tested Update 2 and it doesn't show the ICE either.

Heinz