This is the mail archive of the gcc-patches@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]

Re: [PATCH/RFC] Start to use VIEW_CONVERT_EXPR more and start to fix PR 26069


> 2) Have the gimplifier swap which side of a MODIFY_EXPR, the
>    VIEW_CONVERT_EXPR is on.  This helps because otherwise a in
>    "VIEW_CONVERT_EXPR<int> a = b;" needs to be addressable.
> 3) Fix a bug in tree-ssa-operands which does not mark a in
>    "VIEW_CONVERT <int> a = b;" as addressable.

I agree that (2) is worthwhile since V_C_E on the LHS is peculiar.  However,
I don't understand the reference to being marked addressable.  There's no
reason to mark the operand of V_C_E as addressable.  In no way do we take
a pointer that can be used elsewhere.  And the only time we're going to be
forcing it into memory is in the BLKmode case, when it's already in memory.

> The one question I have about 1 is should we ignore the aliasing set and
> always do this transformation even though we know we turn undefined code
> into defined code?

I'd say yes, ignore it.


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