This is the mail archive of the gcc@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]

Re: More MEM tracking stuff


    what's the ultimate goal of these changes?  Improving alias analysis
    seems to be one point but what are you also trying to achieve?  Feel
    free to answer publically on the list since others might be interested
    - or where should I look for some explanation?

There are basically three things:

(1) Alias analysis can indeed be improved by knowing the underlying decl
for each MEM.

(2) With the recording of a decl for each MEM, the assembler output can
be annotated with the variable name for easier reading.  This has been
completed.

(3) By recording the alignment of a MEM inside the MEM, the kludge of
expand_expr_unaligned can be removed and we can also simplify the
alignment handling in tree expansion.  A major goal here is to get rid
of the copying we do in a lot of such cases.  This work is about half done.


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