------- Additional Comments From rth at gcc dot gnu dot org 2005-02-03 02:59
-------
(In reply to comment #13)
> [3] We could prohibit the simplification that extracts a symbol ref out of a
> jump table, so that an indexed load is always performed a run-time, and so
> changing the jump table is always safe.
I vote for this alternative. Nr 1 you'll have a hard time reliably detecting,
nr 2 requires too many changes to too many backends (not that we don't need a
new rtl multi-way branch representation, but we can't have it Now). I'm not
sure I believe in nr 4 or nr 6. Nr 5 would work, but you probably forgot to
put the label on the forced_labels list. Nr 7 is just gross.
My guess is that we should get the bulk of the optimization that we were
looking for from reading from the jump table from tree-level switch statement
optimizations. And if not... tough.
--
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed|2004-11-23 16:10:55 |2005-02-03 02:59:53
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18628
