[PING] Several middle-end patches
Diego Novillo
dnovillo@google.com
Tue Apr 15 13:43:00 GMT 2008
On 4/15/08 7:48 AM, Richard Guenther wrote:
> Teach the propagator engine to not substitute into dead stmts (but DCE
> them)
> http://gcc.gnu.org/ml/gcc-patches/2008-03/msg01195.html
OK.
> Statistics infrastructure patch (sort of out-of-date again, but it
> seems there is zero interest in this!?)
s/interest/time/ Sorry.
> http://gcc.gnu.org/ml/gcc-patches/2008-03/msg00970.html
I like the idea. A few notes on the implementation:
Do passes_by_id and passes_by_id_size really need to be globals? Can't
we encapsulate them in passes.c?
What's the reason for statistics_add_fn and statistics_add? They just
call statistics_add_raw(). If anything, put the body of
statistics_add_raw in statistics_add_fn and make statistics_add be an
inline function that call statistics_add_fn(cfun).
Does it really make sense to index by string? Wouldn't it be quicker
and safer to index by a stats_id that we add to a common file? Much
like we do with timers and tree codes. It doesn't necessarily need to
be a .def file, we can just add to a global enum in tree-pass.h.
The string does give you the advantage of being both an index and a
label to show in the output, though. So, I'm don't feel to strongly
about it.
A single call satistics_add() may not suffice. We may be keeping track
of quantities that increment in steps >1. But I guess we can add to the
API as we need.
Diego.
More information about the Gcc-patches
mailing list