[gfortran] Fixed cray_pointers_2.f90 aliasing issues.

[vzakhari]

The test started failing after llvm/llvm-project#140803
The issue is that the Cray pointee and the underlying storage
are both accessed in subroutines violating Fortran aliasing rules
for the dummy arguments.

GCC tried to "solve" that by using -fno-inline option, but
it should be better to make the aliasing explicit with proper TARGET
attribute placement.

[tarunprabhu]

Thanks Slava.

We don't allow modifying these tests. For now, this test should be disabled instead by adding it to the "skipped" tests in Fortran/gfortran/regression/DisabledFiles.cmake. Please add a comment explaining why this was disabled (the contents of your commit message would work well).

If this test is useful, you could consider adding it with your modifications to Fortran/UnitTests. You can use the assign-goto test as a template. I believe that that test is a modified version of one from the gfortran test suite.

[tarunprabhu]

Please disable the test instead.

[ashermancinelli]

Thanks for the explanation @tarunprabhu!

As you said Slava, this test appears to break the rules for cray pointers that the HPE docs lay out. Our own docs on aliasing explain this to our users and how they can get around the issues that come from their misuse of the feature. It may be a useful standalone test demonstrating that the workaround we recommend to our users really does allow their code to work as they intend. I don't like gfortran's approach of simply disabling some optimizations to get the test to pass. +1 for disabling the test and adding our own modified test.

[tarunprabhu]

These DejaGNU directives are not necessary. We may want to retain a comment with the command line options though. If any are ever supported in flang, we may need to update the flags with which this test is built. Perhaps something like

Suggested change

	! { dg-do run }
	! { dg-options "-O3 -fcray-pointer -fbounds-check -fno-inline" }
	! { dg-timeout-factor 4 }
	! The original gfortran test on which this is based is compiled with
	! -fcray-pointer -fbounds-check -fno-inline. Of these, we may want to compile
	! with -fbounds-check (or its equivalent) if it is ever supported in flang

[tarunprabhu]

Instead of this, could we have a comment indicating what the original source of this test was and what changes were made to it? It may be a useful reference in the future.

[tarunprabhu]

Thanks for the changes, Slava. LGTM.

[vzakhari]

Thank you for the review, Tarun!