Re: [PATCH, i386]: x87 fops can handle HImodes, too.

[Richard Henderson <rth at redhat dot com>]

On Wed, Sep 15, 2004 at 07:06:26AM +0200, Uros Bizjak wrote:
> Just to be sure, does it mean that
> 
> -         [(float:SF (match_operand:SI 1 "nonimmediate_operand" "m,?r"))
> +         [(float:SF (match_operand 1 "nonimmediate_operand" "m,?r"))
> 
> is not the same even if an extra constraint is added as
> +   && (GET_MODE (operands[1]) == SImode

Correct, because insn_data[x].operands[1].mode is different.

> If this is not the same, adding similar pattern as *fop_sf_2 with
>       ...
>       (float:SF (match_operand:HI 2 "nonimmediate_operand" "m,?r"))]))]
>       ...
> would satisfy reload?

Yes.  Although I'll warn you that such a patch, not being a bug fix,
will have to wait for stage1 to open for gcc 4.1.  But if you want
to work on this in the meantime and hold onto the patch, you might
look into using the new md file macro facilities so that you don't
have to duplicate code.


r~