On October 18, 2015 6:06:51 PM GMT+02:00, Jan Hubicka <hubi...@ucw.cz> wrote:
>Hello,
>> > I was only tracking one isse I hit: Fortran/C interoperability nees
>LTO to
>> > produce same TYPE_CANONICAl for signed and unsigned version of
>size_t.
>> > Doing so broke useless_type_conversion because it used
>TYPE_CANONICAL. We
>> > discussed the topic on the GNU Cauldron and decided that it is
>cleaner to
>> > drop TYPE_CANONICAL from useless_type_conversion because it does
>not really
>> > belong there.
>>
>> OK, thanks for the explanation.
>>
>> > That is only change I plan into the area. The decision to drop
>comparsion of
>> > TYPE_MODE from the aggregate path was decision of the discussion
>about this
>> > particular patch and I do not really insist on it.
>> >
>> > Having fewer VCE expressions in the code is not a bad thing, but I
>do not
>> > really see it as an important change. I am sorry for the breakage
>in move
>> > expansion that I hoped to not be as important. I am willing to
>continue
>> > fixing the fallout (and be more cureful about it - obviously I
>originally
>> > underestimated the issue). I am also happy with simply adding back
>the mode
>> > checking and drop the changes we did to expr.c so far.
>>
>> I agree on the fewer VCE expressions goal (and I have an upcoming
>gigi change
>> to that effect) but some of them are essentially mandated by the RTL
>level
>> and, since GENERIC & GIMPLE are ultimately lowered to RTL, they need
>to take
>> that into account IMO. So, if the mode change is not really
>necessary for the
>> rest of the work, I'd restore the mode check (and this only affects
>Ada in
>> practice since apparently only the Ada compiler fiddles with the type
>mode).
>
>Why is Ada fliddling with the modes? Is it only for packed structures?
>
>I was wondering how to produce VCE convesions of aggregates with C
>frontend at
>all (that is getting them synthetized by the middle-end) to get non-ada
>testcases. Storing through union is never folded to one and I don't
>see any
>other obvious way of getting them. Perhaps it may be possible to get
>them via
>inliner on incompatible parameter and LTO, but that seems to be the
>only case
>I can think of right now.
>
>I am testing the change to compare modes and revert the two expr.c
>changes.
>Lets see what is Richard's opinion. The whole concept of modes on
>aggregate
>types is bit funny post-tree-ssa days when we do SRA. I suppose they
>may be
>tied to calling conventions, but should no longer be needed for code
>quality?
Adding back the mode check is fine if all types with the same TYPE_CANONICAL
have the same mode. Otherwise we'd regress here. I thought we do for
Struct x { int i; };
Typedef y x __attribute__((packed));
And then doing
X x;
Y y;
X = y;
Richard.
>Honza
>>
>> --
>> Eric Botcazou
