ARM PATCH: Fix fix patterns

[Richard Henderson]

On Mon, Mar 10, 2003 at 11:35:15AM +0000, Richard Earnshaw wrote:
> Richard, In reply to your comment in that PR, the documentation for FIX is 
> very confusingly written.  There are two entries for FIX in rtl.texi, one 
> for fix:MODE_FLOAT and one for fix:MODE_INT.  The first describes a 
> rounding operation and the second a conversion operation.  Maybe the 
> former should be given a different RTL code (ROUND?).

Well, really we need all of the round operations: crop to
zero, round to nearest, floor, ceil, round a-la the setting
in the fp control register.

We want all of this so that we can implement the full set
of c99 rounding functions as builtins.

As for (fix:INT (reg:FLOAT)) vs (fix:INT (fix:FLOAT (reg:FLOAT)))...
Well, sometimes I don't see the point of the argument.  We could
as well change the documentation such that this *isn't* undefined
for non-integral quantities, and save the creation of extra rtl.



r~