CLASS_CANNOT_CHANGE_MODE

Richard Henderson rth@redhat.com
Tue Feb 13 17:14:00 GMT 2001


On Wed, Feb 14, 2001 at 12:05:43PM +1100, Alan Modra wrote:
> Is this code in regclass.c:record_operand_costs in error?
[...]
> What seems to be missing is any sort of test on the register class before
> calling CLASS_CANNOT_CHANGE_MODE_P, which is wrong according to tm.texi

No, because we're not dealing with hard registers at all, 
but pseudos.  We're recording that for some subreg there
is a mode change that would be illegal if that pseudo
were to be allocated to the CLASS_CANNOT_CHANGE_MODE class.



r~



More information about the Gcc-bugs mailing list