current egcs miscompiles glibc-2.1's inline math
John Wehle
john@feith.com
Tue Apr 13 22:21:00 GMT 1999
>> This applies to any FP operation? What about conditional moves? What
>> about loads & stores?
>
> You could get away with cmov and memory loads. But I wouldn't
> recommend breaking the rule -- someone will just get confused.
I was reviewing my patches which have been committed and see that I also
split the integer to floating point conversion patterns, however I believe
that these are safe.
My impression is that most FP splits which occur before reg-stack
runs are safe so it's possible to add a split for a FP store as long
as there is also a define_insn which doesn't require a split.
-- John
-------------------------------------------------------------------------
| Feith Systems | Voice: 1-215-646-8000 | Email: john@feith.com |
| John Wehle | Fax: 1-215-540-5495 | |
-------------------------------------------------------------------------
More information about the Gcc-bugs
mailing list