i386.md floating point comparison with general register

Richard Henderson rth@cygnus.com
Wed Mar 31 23:46:00 GMT 1999


On Fri, Mar 26, 1999 at 10:12:29PM -0500, John Wehle wrote:
> Should this pattern perhaps be:
> 
>   [(set (cc0)
>         (match_operator 2 "VOIDmode_compare_op"
>                         [(match_operand:XF 0 "register_operand" "f")
>                          (float:XF
>                           (match_operand:SI 1 "memory_operand" "m"))]))

Sure, that gets you a closer match.  It does mean that the number
of matches will really drop -- reload can't force hard regs to memory
as a reload.

Of course, PPro/P2 greatly dislikes int->fp conversion combined with
another operation.  It's worth considering disabling that operation
entirely for newer cpus.


r~



More information about the Gcc mailing list