[PATCH] Dispose of non-gimplifier users of is_gimple_tmp_var
Nathanael Nerode
neroden@fastmail.fm
Mon Aug 16 13:54:00 GMT 2004
>On Mon, 2004-08-16 at 07:48, Nathanael Nerode wrote:
>> Both of these uses appear to be an attempt to retain better debugging
>> information. Accordingly, they are checking the wrong thing.
>>
>What does this fix?
Nothing ATM. Once is_gimple_tmp_var is fixed, it leads to
In tree-ssa-live.c:
* a preference to use names of actual user variables rather than
front-end temporaries when combining two partitions (it unlikely that
a user variable and a front-end temporary would be combined, but
it's possible)
* removes a preference to use names of front-end temporaries rather than
gimple temporaries when combining two partitions
In tree-ssa-copyrename.c:
* Allows front-end temporaries to be (potentially) coalesced with user
variables and other front-end temporaries
(It may allow artificial static and external variables
with SSA_NAMEs to be coalesced when they shouldn't be; not sure, but I
doubt it. If so this is a mistake, but can be fixed trivially. And
also I'd expect some form of breakage during bootstrap & regression
testing if it did, and this would be detected right now.)
I believe these are basically improvements.
The motivation here is that we're going to have to change the semantics
of is_gimple_tmp_var and possibly break it into two or more separate
functions. Getting rid of totally spurious uses simplifies this process.
--
This space intentionally left blank.
More information about the Gcc-patches
mailing list