PATCH: VAX sltu and sgeu patterns
Jeffrey A Law
Mon Aug 21 07:25:00 GMT 2000
In message < 200008182011.QAA08290@hiauly1.hia.nrc.ca >you write:
> The patch fixes the reload problem arising from non offsettable memory.
> I havn't observed any input reloads arising from the failure of a pseudo
> to get a hard register. If this is a significant problem, then it appears
> other ports such as m68k are also affected.
Then they need to be fixed too.
> > > None of our other cc0 ports need this, that I recall.
> > That's what we used to do for conditional moves on the ia32 port until
> > we changed it not to use cc0 if I remember correctly.
> This would be a fairly major change. I would suggest that the patch be
> installed, and then after further testing, consideration be given as to
> whether further the above suggestion is implemented.
No, that's just papering over the bug. The cc0 setter and user must not
ever be separated. That means that a cc0 user must never, ever ever need
an input reload. If a pattern exists that can potentially need an input
reload on such a machine, then that pattern is wrong, plain and simple.
More information about the Gcc-patches