This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug middle-end/57748] [4.8/4.9 Regression] ICE on ARM with -mfloat-abi=softfp -mfpu=neon


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57748

--- Comment #10 from Martin Jambor <jamborm at gcc dot gnu.org> ---
The problem is that the type of the record that contains the scalar
data we are accessing has non-BLK mode despite that we are not
accessing a part of it.  This is because it has a zero sized trailing
array:

typedef struct S { V a; V b[0]; } P __attribute__((aligned (1)));

in the x86_64-linux example from comment #9, and 

struct resolved_chain {
 u64 nr;
 struct resolved_ip ips[0];
};

in the original testcase.

Because it has non-BLK mode, it passes the condition in
expand-assignment that is there to handle accesses to small,
scalar-sized structures (or even arrays) which should be written to
memory through movmisalign_optab.  E.g. to access an element of a
vector.

However, in these testcases, the structure we are writing to is bigger
than a scalar.  I wonder why structures with trailing zero-sized
arrays have non-BLK mode.  I think that is the root of the problem,
actually.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]